NLO QCD corrections to Higgs boson plus jet production with full top-quark mass dependence
We present the next-to-leading order QCD corrections to the production of a Higgs boson in association with one jet at the LHC including the full top-quark mass dependence. The mass of the bottom quark is neglected. The two-loop integrals appearing in the virtual contribution are calculated numerically using the method of Sector Decomposition. We study the Higgs boson transverse momentum distribution, focusing on the high region, where the top-quark loop is resolved. We find that the next-to-leading order QCD corrections are large but that the ratio of the next-to-leading order to leading order result is similar to that obtained by computing in the limit of large top-quark mass.
During Run I and early Run II of the LHC great progress has been made in establishing many of the properties of the Higgs boson particle discovered in 2012. Already the spin and CP properties are well constrained and its couplings to the Standard Model (SM) weak vector bosons and heavier fermions (top quarks and tau leptons) have been measured Aad et al. (2016). So far the measured properties are consistent with the predictions of the SM.
Among the different channels for the production of a SM Higgs boson, gluon fusion, which we consider here, is the mechanism yielding the largest contribution. At lowest order in perturbation theory this process is mediated by a closed loop of heavy quarks and Next-to-Leading Order (NLO) QCD corrections therefore require the computation of two-loop contributions. A Higgs Effective Field Theory (HEFT) was derived long ago Wilczek (1977), in which the heavy quarks are integrated out and the Higgs boson couples directly to the gluons. This allows to simplify considerably the computations.
One interesting regime to consider is that of Higgs boson production with a transverse momentum of the order of the top-quark mass, , or larger. Here the top quark loop is resolved and it becomes possible to disentangle the SM contribution from effects of New Physics. However, in this regime finite top-quark mass effects are not negligible and the effective theory approximation becomes increasingly poor. In other words, events in which the Higgs boson is recoiling against one or more jets acquiring a large transverse momentum do not fall into the validity range of the effective field theory description. It is thus important to go beyond the HEFT approximation and include finite top-quark mass effects to obtain reliable predictions in this kinematical range.
Within the HEFT approximation corrections to inclusive Higgs boson production are known to NLO QCD accuracy Anastasiou et al. (2016), whilst the fully differential corrections for + 1 jet production are known to NNLO QCD accuracy Boughezal et al. (2013); Chen et al. (2015); Boughezal et al. (2015a, b). Finite quark mass corrections to + 1 jet production have been known at LO for a long time Ellis et al. (1988); Baur and Glover (1990) and LO results are also known for the higher multiplicity processes + 2 jets Del Duca et al. (2001a, b) and recently + 3 jets Campanario and Kubocz (2013); Greiner et al. (2017). The HEFT results for + 1 jet have also been supplemented by an expansion in at NLO QCD accuracy Harlander et al. (2012); Neumann and Wiesemann (2014) and combined with the exact Born and real corrections Neumann and Williams (2017). They were also included in multi-purpose Monte Carlo generators to produce merged samples matched to parton showers Buschmann et al. (2015); Hamilton et al. (2015); Frederix et al. (2016) and used to improve the Higgs NNLO QCD transverse momentum distributions in the HEFT above the top-quark mass threshold Chen et al. (2016).
One of the first major steps towards the computation of the full two-loop NLO QCD virtual corrections was made in Ref. Bonciani et al. (2016), where the planar master integrals were computed analytically in the Euclidean region and shown to contain elliptic integrals. At the same time an expansion valid in the limit of small bottom-quark mass allowed to gain insight into the NLO QCD effects due to nearly massless quarks Mueller and Öztürk (2016); Melnikov et al. (2016, 2017); Lindert et al. (2017). Very recently a NLO QCD result expanded in the regime where the Higgs boson and top-quark masses are small, relevant for the description of the Higgs boson transverse momentum distribution at large , was also studied Kudashkin et al. (2017); Lindert et al. (2018).
On the experimental side, recently the CMS collaboration has considered events where the Higgs boson transverse momentum is larger than 450 Sirunyan et al. (2017); a feat made possible through the use of boosted techniques Butterworth et al. (2008), which allow the Higgs to be identified through its decay to bottom quarks.
In this letter we present the first NLO QCD computation of Higgs boson production in association with one jet retaining the full top-quark mass dependence. In the following sections we present the computational setup used for this calculation and selected phenomenological results.
Ii Computational setup
We compute using conventional dimensional regularization (CDR) with . The top-quark mass is renormalized in the on-shell scheme and the QCD coupling and gluon wave-function in the scheme with light quarks, with the top quark loops subtracted at zero momentum. The top-quark mass renormalization is performed by inserting the mass counterterm into the heavy quark propagators. Alternatively, the mass renormalization can be calculated by taking the derivative of the one-loop amplitude with respect to . We have used both methods as a cross-check.
The sampling of the phase-space generator has been adapted to generate a nearly uniform distribution of points in .
ii.1 Born and Real radiation
The computation of everything but the virtual amplitudes is performed within the POWHEG-BOX-V2 framework Alioli et al. (2010), taking advantage of the existing HJ generator Campbell et al. (2012) for HEFT, in which the Born and real radiation amplitudes are computed using Madgraph4 Stelzer and Long (1994); Alwall et al. (2007) and the virtual amplitudes are taken from MCFM Campbell et al. (2010). The subtraction of the infra-red divergences is performed using FKS Frixione et al. (1996).
We supplemented this code with the analytical Born amplitudes with full top-quark mass dependence from Ref. Baur and Glover (1990), whereas the one-loop real radiation contribution was generated with GoSam Cullen et al. (2012, 2014) using the BLHA Binoth et al. (2010); Alioli et al. (2014) interface developed in Ref. Luisoni et al. (2013). For the purpose of this computation GoSam has been improved such that it now automatically switches to quadruple precision in regions where the amplitude becomes unstable due to one of the final state partons becoming soft or collinear to another parton. The amplitudes generated by GoSam are computed at run time with Ninja Mastrolia et al. (2012); van Deurzen et al. (2014); Peraro (2014) using the quadninja feature. The scalar one-loop integrals are computed with the OneLoop van Hameren (2011) integral library. As a consistency check the virtual amplitudes in HEFT were cross checked with GoSam, whereas the real radiation amplitudes in the full theory were compared against OpenLoops Cascioli et al. (2012).
ii.2 The virtual amplitude
The Lorentz structure of the and partonic amplitudes can be decomposed, after imposing parity conservation, transversality of the gluon polarization vectors and the Ward identity, in terms of 2 and 4 tensor structures respectively, see for example Ref. Gehrmann et al. (2012). This decomposition is not unique. For the amplitude we follow Ref. Boggia et al. (2017) and choose to decompose it such that three of the form factors (which multiply the tensor structures) are related by cyclic permutations of the external gluon momenta whilst the fourth is itself invariant under such permutations. For the amplitude our decomposition is chosen such that the form factors are related by interchanging the external quark and anti-quark momenta. We separately compute all form factors and use these symmetries as a cross-check of our result.
In order to compute the amplitudes we closely follow the method of Refs. Borowka et al. (2016a, b). We construct projection operators for each of the form factors and contract them with the amplitude omitting external spinors and polarization vectors. This procedure allows us to write the amplitude in terms of scalar integrals.
The Feynman diagrams contributing to the two-loop virtual amplitude are generated using Qgraf Nogueira (1993) and further processed using Reduze von Manteuffel and Studerus (2012), Ginac Bauer et al. (2000), Fermat Lewis (), and Mathematica. We cross checked the amplitudes with expressions obtained from a two-loop extension of GoSam, which uses Qgraf and form Vermaseren (2000); Kuipers et al. (2013). The integrals appearing in the amplitude are reduced to master integrals using a customized version of the program Reduze. To simplify the numerical evaluation we choose a quasi-finite basis of master integrals von Manteuffel et al. (2015). The resulting integrals are calculated numerically using SecDec Borowka et al. (2015, 2018). Neglecting crossings we evaluate a total of 102 planar and 18 non-planar two-loop integrals.
The Higgs boson mass is set to and the top-quark mass is chosen such that , which means that . Fixing the ratio of the Higgs boson to top-quark mass allows us to reduce by one the number of independent scales appearing in the two-loop virtual amplitudes. This simplifies the integral reduction and the form of the resulting reduced amplitude.
We subtract the infra-red and collinear poles of the virtual amplitude to obtain the finite part of the virtual amplitude as required in the POWHEG-BOX-V2 framework Heinrich et al. (2017). The IR subtraction procedure requires the one-loop amplitudes up to order , which we compute numerically using the same procedure as for the two-loop amplitudes.
In this section we present results for + 1 jet production at the LHC at a center-of-mass energy of 13 . Jets were clustered using the anti-kt jet algorithm implemented in FastJet Cacciari and Salam (2006); Cacciari et al. (2008, 2012) with a radial distance of and requiring a minimum transverse momentum of . We used the PDF4LHC15_nlo_30_pdfas Butterworth et al. (2016); Dulat et al. (2016); Harland-Lang et al. (2015); Ball et al. (2015) set interfaced through LHAPDF Buckley et al. (2015) for both LO and NLO predictions, and fixed the default value of factorization and renormalization scales and to , defined as
where the sum runs over all final state partons . This scale is known to give a good convergence of the perturbative expansion and stable differential -factors (ratio of NLO to LO predictions) in the effective theory Greiner et al. (2016). To estimate the theoretical uncertainty we vary independently and by factors of and , and exclude the opposite variations. The total uncertainty is taken to be the envelope of this 7-point variation.
To better highlight the differences arising from the two-loop massive contributions, we compare the new results with full top-quark mass dependence, which we label as “full theory result” or simply “full” in the following, to two different approximations. In addition to predictions in the effective theory, which are referred to as HEFT in the following, we show results in which everything but the virtual amplitudes is computed with full top-quark mass dependence. In this latter case only the virtual contribution is computed in the effective field theory and reweighted by the full theory Born amplitude for each phase space point. Following Ref. Maltoni et al. (2014) we call this prediction “approximated full theory” and label it as FT from now on.
We start by presenting the total cross sections, which are reported in Table 1. For comparison we present results also for the HEFT and FT approximations.
|Theory||LO [pb]||NLO [pb]|
Together with the prediction obtained with the central scale defined according to Eq. (1) we show the upper and lower values obtained by varying the scales. While at LO the top-quark mass effects lead to an increase of , at NLO this increase is of the order of compared to the HEFT approximation, and there is an increase of about in the total NLO cross section when comparing the FT result with the full theory one. It is important to keep in mind that when taking into account massive bottom-quark loop contributions, the interference effects are sizable and cancel to a large extent the increase in the total cross section observed here between the HEFT and the full theory results (see e.g. the results in Ref. Greiner et al. (2017)). Note, however, that the bottom-quark mass effects at LO are of the order of or smaller above the top quark threshold.
Considering more differential observables, it is well known that very significant effects due to resolving the top-quark loop are displayed by the Higgs boson transverse momentum distribution, which is softened for larger values of by the full top-quark mass dependence. By considering the high energy limit of a point-like gluon-gluon Higgs interaction and one mediated via a quark loop it is possible to derive the scaling of the squared transverse momentum distribution Forte and Muselli (2016); Caola et al. (2016), which drops as in the effective theory, and goes instead as in the full theory. This fact was shown to hold numerically at LO for up to three jets in Ref. Greiner et al. (2017). It is interesting to verify this also after NLO QCD corrections are applied. To do so, in Figure 1 we show the transverse momentum spectrum of the Higgs boson at LO and NLO in the HEFT approximation and with the full top-quark mass dependence.
In the upper panel we display each differential distribution with the theory uncertainty band originating from scale variation. To highlight the different scaling in , in the middle panel we normalize all the distributions to the LO curve in the effective theory. It is thus possible to see that for low transverse momenta the full theory predictions overshoot slightly the effective theory ones. For the two predictions start deviating more substantially. At LO the two uncertainty bands do not overlap any more above , whereas at NLO this happens already around due to reduction of the uncertainty at this order. The logarithmic scale also allows to see that the relative scaling behavior within the two theory descriptions is preserved between LO and NLO. The curves in the lowest panel of Figure 1 show the differential -factor in HEFT and in the full theory. In both cases above they become very stable and amount to about and respectively. Thus the NLO corrections are large also in the full theory. This broadly agrees with the conclusions of Ref. Lindert et al. (2018), where the expanded result showed a similar enhancement of the -factor by about in the tail compared to the HEFT.
To conclude this section we compare the new predictions for the Higgs boson transverse momentum with the one in FT. At LO the two predictions are identical by construction, it is however interesting to check how good FT can reproduce the full theory results. In the main panel of Figure 2 we plot the three curves. To highlight better the differences among the two predictions, in the middle panel we normalize the distributions to the LO prediction. This allows to compare the two differential -factors, which behave very similarly over the full kinematical range. As already observed in the case of double Higgs boson production Borowka et al. (2016b), the scale uncertainty band of the full theory predictions is slightly reduced compared to the one in FT. In order to quantify the difference between the two predictions, in the lower panel we display the ratio of the Full NLO curve to the FT NLO curve. This allows to see that the full top-quark mass virtual contribution enhances the predictions obtained by reweighting the HEFT virtual by an almost constant factor of about .
Iv Conclusions and Outlook
In this letter we have presented for the first time NLO QCD corrections to Higgs boson plus jet production retaining the full top-quark mass dependence. We observe that the size of the NLO corrections is large but, for our choice of the renormalization and factorization scale, the -factor is approximately constant above the top-quark threshold. Compared to FT predictions, the full two-loop contribution enhances the NLO predictions by about at the level of the total cross section and by about at the level of the differential transverse momentum distribution for . Despite a completely different scaling, the -factors in the HEFT and in the full theory behave in a very similar way above .
The result removes the theoretical uncertainty on differential + 1 jet distributions due to the unknown top mass corrections at NLO in QCD. Besides the transverse momentum distribution, shown here, this calculation enables accurate predictions to be made also for other observables where the top-quark mass effects may play a significant role.
As the experimental precision at the LHC improves in the coming years, this result aids the study of the Higgs boson properties also in boosted regimes. Providing a more accurate theoretical description of the Higgs boson production at large transverse momentum allows not only to unravel the details of the electroweak symmetry breaking mechanism, but also to search for indirect signs of New Physics.
We would like to thank Gudrun Heinrich for encouraging us to undertake this computation and also for numerous discussions and comments on the manuscript. We thank Nigel Glover and Hjalte Frellesvig for interesting discussions and for their insightful comments on the Lorentz structure of the amplitude. We thank Tiziano Peraro for helping us in the upgrade of the Ninja interface. This research was supported by the Munich Institute for Astro- and Particle Physics (MIAPP) of the DFG cluster of excellence “Origin and Structure of the Universe”. SPJ was supported by the Research Executive Agency (REA) of the European Union under the Grant Agreement PITN-GA2012316704 (HiggsTools) during part of this work. We gratefully acknowledge support and resources provided by the Max Planck Computing and Data Facility (MPCDF).
- preprint: MPP-2018-7
- G. Aad et al. (ATLAS, CMS), JHEP 08, 045 (2016), eprint 1606.02266.
- F. Wilczek, Phys. Rev. Lett. 39, 1304 (1977).
- C. Anastasiou, C. Duhr, F. Dulat, E. Furlan, T. Gehrmann, F. Herzog, A. Lazopoulos, and B. Mistlberger, JHEP 05, 058 (2016), eprint 1602.00695.
- R. Boughezal, F. Caola, K. Melnikov, F. Petriello, and M. Schulze, JHEP 06, 072 (2013), eprint 1302.6216.
- X. Chen, T. Gehrmann, E. W. N. Glover, and M. Jaquier, Phys. Lett. B740, 147 (2015), eprint 1408.5325.
- R. Boughezal, F. Caola, K. Melnikov, F. Petriello, and M. Schulze, Phys. Rev. Lett. 115, 082003 (2015a), eprint 1504.07922.
- R. Boughezal, C. Focke, W. Giele, X. Liu, and F. Petriello, Phys. Lett. B748, 5 (2015b), eprint 1505.03893.
- R. K. Ellis, I. Hinchliffe, M. Soldate, and J. J. van der Bij, Nucl. Phys. B297, 221 (1988).
- U. Baur and E. W. N. Glover, Nucl. Phys. B339, 38 (1990).
- V. Del Duca, W. Kilgore, C. Oleari, C. Schmidt, and D. Zeppenfeld, Phys. Rev. Lett. 87, 122001 (2001a), eprint hep-ph/0105129.
- V. Del Duca, W. Kilgore, C. Oleari, C. Schmidt, and D. Zeppenfeld, Nucl. Phys. B616, 367 (2001b), eprint hep-ph/0108030.
- F. Campanario and M. Kubocz, Phys. Rev. D88, 054021 (2013), eprint 1306.1830.
- N. Greiner, S. Höche, G. Luisoni, M. Schönherr, and J.-C. Winter, JHEP 01, 091 (2017), eprint 1608.01195.
- R. V. Harlander, T. Neumann, K. J. Ozeren, and M. Wiesemann, JHEP 08, 139 (2012), eprint 1206.0157.
- T. Neumann and M. Wiesemann, JHEP 11, 150 (2014), eprint 1408.6836.
- T. Neumann and C. Williams, Phys. Rev. D95, 014004 (2017), eprint 1609.00367.
- M. Buschmann, D. Goncalves, S. Kuttimalai, M. Schonherr, F. Krauss, and T. Plehn, JHEP 02, 038 (2015), eprint 1410.5806.
- K. Hamilton, P. Nason, and G. Zanderighi, JHEP 05, 140 (2015), eprint 1501.04637.
- R. Frederix, S. Frixione, E. Vryonidou, and M. Wiesemann, JHEP 08, 006 (2016), eprint 1604.03017.
- X. Chen, J. Cruz-Martinez, T. Gehrmann, E. W. N. Glover, and M. Jaquier, JHEP 10, 066 (2016), eprint 1607.08817.
- R. Bonciani, V. Del Duca, H. Frellesvig, J. M. Henn, F. Moriello, and V. A. Smirnov, JHEP 12, 096 (2016), eprint 1609.06685.
- R. Mueller and D. G. Öztürk, JHEP 08, 055 (2016), eprint 1512.08570.
- K. Melnikov, L. Tancredi, and C. Wever, JHEP 11, 104 (2016), eprint 1610.03747.
- K. Melnikov, L. Tancredi, and C. Wever, Phys. Rev. D95, 054012 (2017), eprint 1702.00426.
- J. M. Lindert, K. Melnikov, L. Tancredi, and C. Wever, Phys. Rev. Lett. 118, 252002 (2017), eprint 1703.03886.
- K. Kudashkin, K. Melnikov, and C. Wever (2017), eprint 1712.06549.
- J. M. Lindert, K. Kudashkin, K. Melnikov, and C. Wever (2018), eprint 1801.08226.
- A. M. Sirunyan et al. (CMS) (2017), eprint 1709.05543.
- J. M. Butterworth, A. R. Davison, M. Rubin, and G. P. Salam, Phys. Rev. Lett. 100, 242001 (2008), eprint 0802.2470.
- S. Alioli, P. Nason, C. Oleari, and E. Re, JHEP 06, 043 (2010), eprint 1002.2581.
- J. M. Campbell, R. K. Ellis, R. Frederix, P. Nason, C. Oleari, and C. Williams, JHEP 07, 092 (2012), eprint 1202.5475.
- T. Stelzer and W. F. Long, Comput. Phys. Commun. 81, 357 (1994), eprint hep-ph/9401258.
- J. Alwall et al., JHEP 09, 028 (2007), eprint 0706.2334.
- J. M. Campbell, R. K. Ellis, and C. Williams, Phys. Rev. D81, 074023 (2010), eprint 1001.4495.
- S. Frixione, Z. Kunszt, and A. Signer, Nucl. Phys. B467, 399 (1996), eprint hep-ph/9512328.
- G. Cullen, N. Greiner, G. Heinrich, G. Luisoni, P. Mastrolia, G. Ossola, T. Reiter, and F. Tramontano, Eur. Phys. J. C72, 1889 (2012), eprint 1111.2034.
- G. Cullen et al., Eur. Phys. J. C74, 3001 (2014), eprint 1404.7096.
- T. Binoth et al., Comput. Phys. Commun. 181, 1612 (2010), eprint 1001.1307.
- S. Alioli et al., Comput. Phys. Commun. 185, 560 (2014), eprint 1308.3462.
- G. Luisoni, P. Nason, C. Oleari, and F. Tramontano, JHEP 10, 083 (2013), eprint 1306.2542.
- P. Mastrolia, E. Mirabella, and T. Peraro, JHEP 06, 095 (2012), [Erratum: JHEP11,128(2012)], eprint 1203.0291.
- H. van Deurzen, G. Luisoni, P. Mastrolia, E. Mirabella, G. Ossola, and T. Peraro, JHEP 03, 115 (2014), eprint 1312.6678.
- T. Peraro, Comput. Phys. Commun. 185, 2771 (2014), eprint 1403.1229.
- A. van Hameren, Comput.Phys.Commun. 182, 2427 (2011), eprint 1007.4716.
- F. Cascioli, P. Maierhofer, and S. Pozzorini, Phys. Rev. Lett. 108, 111601 (2012), eprint 1111.5206.
- T. Gehrmann, M. Jaquier, E. W. N. Glover, and A. Koukoutsakis, JHEP 02, 056 (2012), eprint 1112.3554.
- M. Boggia et al. (2017), eprint 1711.09875.
- S. Borowka, N. Greiner, G. Heinrich, S. Jones, M. Kerner, J. Schlenk, U. Schubert, and T. Zirke, Phys. Rev. Lett. 117, 012001 (2016a), [Erratum: Phys. Rev. Lett.117,no.7,079901(2016)], eprint 1604.06447.
- S. Borowka, N. Greiner, G. Heinrich, S. P. Jones, M. Kerner, J. Schlenk, and T. Zirke, JHEP 10, 107 (2016b), eprint 1608.04798.
- P. Nogueira, J.Comput.Phys. 105, 279 (1993).
- A. von Manteuffel and C. Studerus (2012), eprint 1201.4330.
- C. W. Bauer, A. Frink, and R. Kreckel, J. Symb. Comput. 33, 1 (2000), eprint cs/0004015.
- R. Lewis, Computer Algebra System Fermat, URL http://home.bway.net/lewis/.
- J. Vermaseren (2000), eprint math-ph/0010025.
- J. Kuipers, T. Ueda, J. Vermaseren, and J. Vollinga, Comput.Phys.Commun. 184, 1453 (2013), eprint 1203.6543.
- A. von Manteuffel, E. Panzer, and R. M. Schabinger, JHEP 02, 120 (2015), eprint 1411.7392.
- S. Borowka, G. Heinrich, S. P. Jones, M. Kerner, J. Schlenk, and T. Zirke, Comput. Phys. Commun. 196, 470 (2015), eprint 1502.06595.
- S. Borowka, G. Heinrich, S. Jahn, S. P. Jones, M. Kerner, J. Schlenk, and T. Zirke, Comput. Phys. Commun. 222, 313 (2018), eprint 1703.09692.
- G. Heinrich, S. P. Jones, M. Kerner, G. Luisoni, and E. Vryonidou, JHEP 08, 088 (2017), eprint 1703.09252.
- M. Cacciari and G. P. Salam, Phys. Lett. B641, 57 (2006), eprint hep-ph/0512210.
- M. Cacciari, G. P. Salam, and G. Soyez, JHEP 04, 063 (2008), eprint 0802.1189.
- M. Cacciari, G. P. Salam, and G. Soyez, Eur. Phys.J. C72, 1896 (2012), eprint 1111.6097.
- J. Butterworth et al., J. Phys. G43, 023001 (2016), eprint 1510.03865.
- S. Dulat, T.-J. Hou, J. Gao, M. Guzzi, J. Huston, P. Nadolsky, J. Pumplin, C. Schmidt, D. Stump, and C. P. Yuan, Phys. Rev. D93, 033006 (2016), eprint 1506.07443.
- L. A. Harland-Lang, A. D. Martin, P. Motylinski, and R. S. Thorne, Eur. Phys. J. C75, 204 (2015), eprint 1412.3989.
- R. D. Ball et al. (NNPDF), JHEP 04, 040 (2015), eprint 1410.8849.
- A. Buckley, J. Ferrando, S. Lloyd, K. Nordstroem, B. Page, M. Ruefenacht, M. Schoenherr, and G. Watt, Eur. Phys. J. C75, 132 (2015), eprint 1412.7420.
- N. Greiner, S. Höche, G. Luisoni, M. Schönherr, J.-C. Winter, and V. Yundin, JHEP 01, 169 (2016), eprint 1506.01016.
- F. Maltoni, E. Vryonidou, and M. Zaro, JHEP 11, 079 (2014), eprint 1408.6542.
- S. Forte and C. Muselli, JHEP 03, 122 (2016), eprint 1511.05561.
- F. Caola, S. Forte, S. Marzani, C. Muselli, and G. Vita, JHEP 08, 150 (2016), eprint 1606.04100.