PHYSICAL REVIEW D 97, 114025 (2018)
Editors' Suggestion Featured in Physics
Muon − 2andαð 2 Þ: A new data-based analysis g MZ † ‡ Alexander Keshavarzi,1,* Daisuke Nomura,2,3, and Thomas Teubner1, 1Department of Mathematical Sciences, University of Liverpool, Liverpool L69 3BX, United Kingdom 2KEK Theory Center, Tsukuba, Ibaraki 305-0801, Japan 3Yukawa Institute for Theoretical Physics, Kyoto University, Kyoto 606-8502, Japan
(Received 6 April 2018; published 25 June 2018)
This work presents a complete reevaluation of the hadronic vacuum polarization contributions to the had;VP anomalous magnetic moment of the muon, aμ , and the hadronic contributions to the effective QED Δα ð 2 Þ þ − → coupling at the mass of the Z boson, had MZ , from the combination of e e hadrons cross section data. Focus has been placed on the development of a new data combination method, which fully incorporates all correlated statistical and systematic uncertainties in a bias free approach. All available þ − → e e hadrons cross section data have been analyzed and included, where the new datap compilationffiffiffi has yielded the full hadronic R-ratio and its covariance matrix in the energy range mπ ≤ s ≤ 11.2 GeV. Using these combined data and perturbative QCD above that range results in estimates of the hadronic had;LO VP −10 vacuum polarization contributions to g − 2 of the muon of aμ ¼ð693.26 2.46Þ × 10 and had;NLO VP −10 aμ ¼ð−9.82 0.04Þ × 10 . The new estimate for the Standard Model prediction is found to be SM −10 aμ ¼ð11659182.04 3.56Þ × 10 , which is 3.7σ below the current experimental measurement. The prediction for the five-flavor hadronic contribution to the QED coupling at the Z boson mass is Δαð5Þ ð 2 Þ¼ð276 11 1 11Þ 10−4 α−1ð 2 Þ¼128 946 0 015 had MZ . . × , resulting in MZ . . . Detailed comparisons with results from similar related works are given.
DOI: 10.1103/PhysRevD.97.114025
I. INTRODUCTION prediction is also improved to determine whether the − 2 ¼ g discrepancy is well established. The anomalous magnetic moment of the muon, aμ SM ð − 2Þ 2 The uncertainty of aμ is completely dominated by the g μ= , stands as an enduring test of the Standard Model had hadronic contributions, aμ , attributed to the contributions (SM), where the ∼3.5σ (or higher) discrepancy between the exp SM from the nonperturbative, low energy region of hadronic experimental measurement aμ and the SM prediction aμ resonances. The hadronic contributions are divided into the could be an indication of the existence of new physics exp hadronic vacuum polarization (VP) and hadronic light-by- beyond the SM. For aμ , the value is dominated by the light (LbL) contributions, which are summed to give measurements made at the Brookhaven National Laboratory – (BNL) [1 3], resulting in a world average of [4] had had;VP had;LbL aμ ¼ aμ þ aμ : ð1:2Þ
exp −10 aμ ¼ 11659209.1ð5.4Þð3.3Þ × 10 : ð1:1Þ This analysis, KNT18, is a complete reevaluation, in line with previous works [7–9], of the hadronic vacuum had;VP Efforts to improve the experimental estimate at Fermilab polarization contributions, aμ . The hadronic vacuum (FNAL) [5] and at J-PARC [6] aim to reduce the exper- polarization contributions can be separated into the leading- imental uncertainty by a factor of 4 compared to the BNL order (LO) and higher-order contributions, where the LO measurement. It is therefore imperative that the SM and next-to-leading order (NLO) contributions are calcu- lated in this work.1 These are calculated utilizing dispersion *[email protected] † integrals and the experimentally measured cross section, [email protected] ‡ [email protected] σ0 ð Þ ≡ σ0ð þ − → γ → þ γÞ ð Þ had;γ s e e hadrons ; 1:3 Published by the American Physical Society under the terms of the Creative Commons Attribution 4.0 International license. Further distribution of this work must maintain attribution to 1The next-to-next-to-leading order (NNLO) contributions have SM the author(s) and the published article’s title, journal citation, recently been determined in [10] and are included as part of aμ and DOI. Funded by SCOAP3. below.
2470-0010=2018=97(11)=114025(27) 114025-1 Published by the American Physical Society KESHAVARZI, NOMURA, and TEUBNER PHYS. REV. D 97, 114025 (2018) where the superscript 0 denotes the bare cross section The structure of this paper is as follows. Section II (undressed of all vacuum polarization effects) and the describes all changes to the treatment and combination of subscript γ indicates the inclusion of effects from final hadronic cross section data since [9]. Section III gives had;LO VP state photon radiation. At LO, the dispersion relation reads details concerning the contributions to aμ and Z Δα ðM2 Þ from the different hadronic final states and α2 ∞ had Z had;LO VP ds aμ ¼ RðsÞKðsÞ; ð1:4Þ energy regions, culminating in updated estimates of both 2 2 3π s s had;LO VP had;NLO VP Δα ð Þ th aμ , aμ , and had MZ . These results are 2 then compared with other works and the new prediction of where α ¼ αð0Þ, s ¼ mπ, RðsÞ is the hadronic R-ratio had;VP th aμ is combined with all other SM contributions to given by SM determine the SM prediction aμ , with details given σ0 ðsÞ σ0 ðsÞ concerning the consequences of this on the existing ð Þ¼ had;γ ≡ had;γ ð Þ g − 2 discrepancy and the new result for αðM2 Þ. Finally, R s σ ð Þ 4πα2 ð3 Þ ; 1:5 Z pt s = s the conclusions of this work, along with discussions of the prospects for g − 2 in the future and potential improve- and KðsÞ is a well known kernel function [11,12],given SM ments for aμ , are presented in Sec. IV. also in Eq. (45) of [7], but differing by a normalization 2 factor of mμ=ð3sÞ. This kernel function [which behaves as 2 II. CHANGES SINCE THE LAST ANALYSIS KðsÞ ∼ mμ=ð3sÞ at low energies], coupled with the factor of (HLMNT11) 1=s in the integrand of Eq. (1.4), causes the hadronic vacuum polarization contributions to be dominated by the A. Radiative corrections low energy domain. At NLO, the data input is identical, with corresponding dispersion integrals and kernel func- 1. Vacuum polarization corrections tions [13] (see also [7]). In the previous analysis [9] Equations (1.4) and (1.8) require the experimental cross (denoted as HLMNT11), the LO hadronic vacuum polari- section to be undressed of all VP effects as VP (running zation contributions were found to be coupling) corrections to the hadronic cross section are had;VP counted as part of the higher order contributions to aμ . had;LO VP −10 aμ ðHLMNT11Þ¼ð694.91 4.27Þ × 10 ; ð1:6Þ Any new and old data that have not been corrected for VP effects require undressing. However, recent data are more which resulted in the total SM prediction of commonly undressed in the experimental analyses already, removing the need to apply a correction to these data sets. SMð 11Þ¼ð11659182 8 4 9Þ 10−10 ð Þ aμ HLMNT . . × : 1:7 This benefits the data combination as new, more precise data undressed of VP effects are dominating the fit for This, compared with the experimental measurement, gave a many channels which, in turn, reduces the impact of the g − 2 discrepancy of 3.3σ. had;VP extra radiative correction uncertainty which is applied to In addition to calculating aμ , the combination of each channel. hadronic cross section data is also used to calculate the Following the previous analyses, the “bare” cross section αð 2Þ 0 running (momentum dependent) QED coupling, q . σ in Eq. (1.5) is determined to be This, in particular, is then used to determine the effective had QED coupling at the Z boson mass, αðM2 Þ, which is the σ0 ¼ σ ð Þ Z had Cvp had; 2:1 least precisely known of the three fundamental electroweak (EW) parameters of the SM [the Fermi constant GF, MZ, where C is a multiplicative VP correction. For this αð 2 Þ vp and MZ ] and hinders the accuracy of EW precision fits. purpose, the self-consistent KNT18 vacuum polarization had;VP Using an identical data input as that used for aμ , the routine, vp_knt_v3_0, has been updated via an iteration hadronic contributions to the effective QED coupling are of the new data input.2 An in-depth discussion of the determined from the dispersion relation determination of the vacuum polarization and the corre- Z sponding routine will be given in [14]. 2 ∞ ð5Þ 2 αq RðsÞ The decisions to undress all data sets previously cor- Δα ðq Þ¼− P ds ; ð1:8Þ had 3π ð − 2Þ rected in [9] are unchanged, with the exception of two sth s s q measurements of the inclusive hadronic R-ratio by the where the superscript (5) indicates the contributions Crystal Ball Collaboration [15,16]. A reexamination of the from all quark flavors except the top quark, which is experimental analyses of these two measurements has added separately. Together with the leptonic contributions, 2 αð Þ¼ 2 this is used to determine the running coupling q This routine is available for use by contacting the authors α ð1 − Δα ð 2Þ − Δα ð 2ÞÞ = had q lep q . directly.
114025-2 − 2 αð 2 Þ 97, MUON g AND MZ : A NEW DATA-BASED ANALYSIS PHYS. REV. D 114025 (2018) shown that not only do they include some treatment of the ð0Þ ð0Þ α σ ðsÞ¼σ ðsÞ 1 þ ηðsÞ ; ð2:4Þ VP, but that they also both account for this correction with had;ðγÞ had π sizable systematic uncertainties. Applying an extra VP correction here would be an overestimate of the correction has been used.3 Here, the function ηðsÞ is given e.g. in [22] and its corresponding uncertainty. Therefore, undressing of and the subscript γ indicates the one photon inclusive cross these data is not applied here. In addition, it should be noted section. However, experimental cross section measure- that in a separate work [17], two measurements by the ments, by nature, include all virtual and soft real radiation KLOE Collaboration [18,19] in the πþπ− channel are now effects. Therefore, to estimate possibly missing photon undressed of VP effects using an updated routine [20] FSR, ideally only the effects from (hard) real radiation compared to the one used previously [21]. above/within resolution/cut parameters which are specific The undressing of narrow resonances in the cc¯ and bb¯ for a given experiment or analysis have to be estimated. regions requires special attention. Importantly, the elec- Whereas in the calculation of the inclusive correction a tronic width of an individual resonance, Γee, must be regularization of the virtual and real soft contributions is undressed of vacuum polarization effects using a para- required to obtain the infrared finite result η, the hard real metrization of the VP where the correction excludes the radiation, ηhard;real, can be estimated numerically from contribution of that resonance, such that Z pffiffi s−2 sΛ ηhard;realð Þ¼ 0ρ ð 0Þ ð Þ 2 2 s ds fin s; s ; 2:5 ðα=α ðM ÞÞ 4 2 Γ0 ¼ no res res Γ : ð2:2Þ m ee 1 þ 3α=ð4πÞ ee where m is the mass of the (scalar) particle, Λ is a finite α infrared cutoff parameter on the invariant mass of the Here, no res is the effective QED coupling neglecting the ρ contribution of the resonance itself and is given by emitted photon, and fin is the radiator function (see Appendix B of [23]). α In the case of the KþK− channel, by far the largest α ðsÞ ≡ ; ð : Þ no res 1 − Δα ð Þ 2 3 contribution to aμ and Δα (and their errors) comes from the no res s energy region of the ϕ peak, where the phase space for real Δα ð Þ radiation is severely restricted. While the calculation of real where no res s is determined from Eq. (1.8) such that the input RðsÞ does not include the resonance that is being radiation accounting for all experimental cuts would be very corrected. To include the resonance would lead to an complicated and beyond the scope of this work, an estimate Λ inconsistent definition of the narrow resonance. can easily be made based on Eq. (2.5), by relating to the cuts in the photon accolinearity given in individual exper- imental analyses. In Fig. 1, the result for the fully inclusive 2. Final state radiative corrections correction ηðsÞ (left panel) is compared to the estimates for Final state radiation (FSR) cannot be separated in an the real hard radiation, where ηhard;realðsÞ (right panel) now unambiguous way in the measured hadronic cross sections. depends on the accolinearity cuts as given by the two Therefore, while formally of higher order in α, FSR experimental analyses depicted. Clearly, at and around photons have to be taken into account in the definition the ϕ peak, phase space restrictions strongly suppress any of the one particle irreducible hadronic blobs and will hard real radiation and using the inclusive correction would already be included as part of the leading order hadronic lead to an overestimate of the possible effects. Given the VP contributions. However, depending on the experimental small size of both the possible correction in the ϕ region and þ − analyses, some amount of real photon FSR may have been the contribution of the K K channel above the ϕ to both removed during the event selection. Adding back these mean value and error of aμ and Δα, no correction or missed contributions is model dependent and not feasible additional error estimate due to FSR is now applied in the þ − for general hadronic final states. It is therefore necessary to K K channel. For the neutral kaon channel, hard photon estimate the possible effects and their impact on the radiation (which would resolve the quark charges) is accuracy of the data compilations. Similar to the case of similarly suppressed and no FSR correction or additional VP corrections, an extra radiative correction uncertainty error is applied in this channel also. due to FSR effects is then estimated channel by channel. The situation is different in the two pion channel. A For the important πþπ− and KþK− channels, detailed study similar to the two kaon channel showed that in studies have been performed for this analysis. Here, and principle larger contributions from real radiation of the especially in the limited energy range below 2 GeV, it has order of the inclusive correction can arise. However, these been shown that scalar QED provides a good description of contributions are strongly dependent on the cut applied in photon FSR; see e.g. [22–24]. In the past, to estimate þ − þ − possible FSR effects in π π and K K production, the 3Here, the term “fully inclusive” means inclusive of effects fully inclusive, order α correction to the cross section, from virtual and real (both soft and hard) one-photon emission.
114025-3 KESHAVARZI, NOMURA, and TEUBNER PHYS. REV. D 97, 114025 (2018)
1.00025 K+K- CFSR CMD-2 (95) 1.05 2000 2000 σ0(e+e- → K+K-) SND (00) 1.0002 σ0(e+e- → K+K-) π 1500 / 1500
1.045 α ) [nb] ) [nb] - - π / K (s)
1.00015 K α + + K K (s) η → →
1000 1000 -
1.04 - hard,real e e 1.0001 η + 1 + + (e (e 0 0 1 + σ σ 500 500 1.035 1.00005
1.03 0 1 0 1.01 1.015 1.02 1.025 1.03 1.035 1.04 1.045 1.05 1.01 1.015 1.02 1.025 1.03 1.035 1.04 1.045 1.05 √s [GeV] √s [GeV]
FIG. 1. The effect of final state radiation in the KþK− channel in the ϕ resonance region. Left panel: the fully inclusive FSR correction ηðsÞ. Right panel: hard real radiation ηhard;realðsÞ, estimated with accolinearity cuts used in the two analyses [25,26]. The eþe− → KþK− cross section is also plotted for reference.
1400 experimental analysis, they are (at least to some extent) π+π- 1.012 CFSR estimated by Monte Carlo (MC) simulation and contribute to σ0(e+e- → π+π-) 1200 the systematic uncertainties.4 However, for many data sets, it 1.01 is far from clear to which extent FSR effects are included in 1000 the systematic errors. Therefore, possible effects are ) [nb] π 1.008 - / π
α accounted for by applying an additional uncertainty deter- 800 + π
(s)
η mined as a fraction of the respective contribution. →
1.006 600 - e 1 + + (e 1.004 0 3. Estimating extra uncertainty due to 400 σ radiative corrections 1.002 200 As in [7–9], extra uncertainties are estimated whenever
1 0 additional radiative corrections are applied. This is done, 0.4 0.6 0.8 1 1.2 1.4 first and foremost, to account for any under- or over- √s [GeV] correction that may occur due to a lack of information ηð Þ concerning the treatment of radiative corrections in the FIG. 2. The behavior of the inclusive FSR correction, s , for experimental analyses. However, these radiative correction the process eþe− → πþπ−. The eþe− → πþπ− cross section is also plotted for reference. uncertainties also account for any possibly incorrect treat- ment in the analyses, e.g. missed FSR or inconsistent subtraction of VP contributions. This is especially true for Eq. (2.5) and would require a more detailed, measurement- older data, where there is very little or even no information by-measurement analysis, which is beyond the scope of this at all regarding how the data have been treated. work. In addition, for many of the data sets the required vp In each channel, the difference Δaμ between the detailed information is not available. Therefore, in sets estimates of aμ with and without additional VP corrections which are understood to not include the full FSR correc- is determined. For the uncertainty due to VP, one-third of tions, the fully inclusive correction (as shown in Fig. 2)is þ − the shift applied to the respective measurements. In the π π data combination, recent sets from radiative return, where 1 δ vp ¼ Δ vp ð Þ additional photons are part of the leading order cross aμ 3 aμ 2:6 section and are studied in detail as part of the analyses, have now become dominant. Therefore, the impact of the is taken. This is reduced from one-half in [9], safe in the fully inclusive FSR correction to older sets is suppressed knowledge that the KNT18 VP routine [14], which is for both mean value and error in comparison to previous determined iteratively in a self-constistent way, is accurate analyses. The procedure to determine the corresponding to the level of a permille when correcting the cross section error estimate is discussed in the next section, while resulting numbers are given in Sec. III A, which contains 4In the recent work [27], a version of the MC program the detailed analysis of the πþπ− channel. CARLOMAT [28] is discussed, which includes real photon radi- For the subleading, multihadron channels, there are, at ation in multihadron channels. This could be used to further study present, no equivalent FSR calculations. Depending on the the FSR effects.
114025-4 − 2 αð 2 Þ 97, MUON g AND MZ : A NEW DATA-BASED ANALYSIS PHYS. REV. D 114025 (2018) and that newer data sets are commonly undressed of VP 2. Fitting data effects by the experiment with a modern routine. The previous analyses [7–9] employed a nonlinear χ2- For the extra uncertainties due to FSR, there are now no þ − 0 0 minimization utilizing fitted renormalization factors as contributions from the K K and KSKL channels (see the nuisance parameters that represented the energy indepen- πþπ− discussion above). For the channel, the full difference dent systematic uncertainties. Although a powerful method, πþπ− between the estimates of aμ with and without additional recent literature [30] (see also [31]) has highlighted the FSR corrections is taken as the FSR uncertainty. For all possibility that an incorrect treatment of multiplicative other channels, including the inclusive data combination normalization uncertainties in a χ2 minimization can incur above 1.937 GeV, a fraction of 1% of the respective cross a systematic bias (see Chap. 4 of [30]). In addition, section is applied as the uncertainty. although the nonlinear χ2-minimization used in [9] was The numerical estimates of all additional radiative adjusted to include covariance matrices, the method’s correction uncertainties are given in the respective sections reliance on fitting energy independent renormalization for the individual channels. The same procedures are factors prevented the use of correlated uncertainties to Δα applied in the calculation of the contributions to had. their full capacity. As recent precise data, specifically radiative return measurements in the πþπ− and KþK− B. Data combination channels, have been released with energy dependent uncertainties and nontrivial bin-to-bin correlations for both The combination of σ0ðeþe− → hadronsÞ data and its the statistical and the systematic uncertainties, the fit corresponding uncertainty has undergone a great deal of procedure had to be modified to allow the full use of all scrutiny since [9]. This has been due to the ever increasing available correlations in a bias-free approach. Therefore, amount of data that have become available and due to a instead of fitting renormalization factors (nuisance param- better understanding of the treatment of correlated uncer- eters), an iterative fit procedure as advocated in [30] has tainties that are now a prominent feature of many cross been adopted, which reinitializes the full covariance matri- section measurements. In [9], covariance matrices in the πþπ− ces at each iteration step and, consequently, avoids bias. channel from BABAR [29] and KLOE [18,19] for Previously, in [7–9], a constant cross section had been statistical and systematic uncertainties had already been assumed across the width of each cluster. In this work, included based on the method of fitted renormalization the fitted cross section values at the cluster centers are factors. In the following, a new method for the combination obtained from the iterative χ2-minimization where the cross σ0 of had data is introduced, which will allow one to fully take section is taken to be linear between adjacent cluster into account all correlated uncertainties. centers. This allows for a more stable fit and is consistent had;VP with the trapezoidal rule integration utilized for the aμ pffiffiffi 1. Clustering data points Δα ¼ ðmÞ and had integrals. If data points at energies s Ei Within each hadronic channel, data points from different are combined into cluster m, then the weighted average of 5 experiments are assigned to clusters (see [7–9] for details). the cross section value Rm and energy Em for the cluster In this work, the clustering algorithm has been improved. It center are is universal for all different channels and only differs in the " #" # ð Þ ð Þ −1 δ δ XN m ðmÞ XN m assigned maximum cluster size (and res, a maximum R 1 R ¼ i ð2:7Þ cluster size applicable at individual narrow resonances). m ð ˜ ðmÞÞ2 ð ˜ ðmÞÞ2 δ δ i¼1 dRi i¼1 dRi A scan over (and res if applicable) is performed to determine a suitable clustering configuration which must and avoid both over- and underclustering. Too wide or over- " #" # populated clusters would effectively lead to a rebinning of ðmÞ ð Þ ðmÞ −1 XN E m XN 1 data points from individual experiments and risk loss of E ¼ i ; ð2:8Þ information, while a too narrow clustering would result, in m ð ˜ ðmÞÞ2 ð ˜ ðmÞÞ2 i¼1 dRi i¼1 dRi the extreme, in an erratic point-to-point representation of the cross section and no gain in the accuracy. The preferred ðmÞ 2 where Ri is the cross section value of data point i χ = : : : ð Þ configuration is then chosen based on the global min d o f contributing to cluster m, N m is the total number of data had;VP and the uncertainty on aμ , combined with checks by points contributing to cluster m, and eye of the resulting spectral function. qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ˜ ðmÞ ¼ ð ðmÞ Þ2 þð ðmÞ Þ2 ð Þ dRi dRi;stat dRi;sys : 2:9 5 χ2 The effect that different clusterings have on the min=d:o:f. of ðmÞ ðmÞ had;VP the fit, the resulting aμ value, and its error has been discussed dRi;stat and dRi;sys denote the absolute statistical and in detail in [9]. systematic uncertainties, respectively. With a linear cross
114025-5 KESHAVARZI, NOMURA, and TEUBNER PHYS. REV. D 97, 114025 (2018) section now assumed, if data point i belongs to cluster m where ð Þ and E m >E , then its interpolant cross section value Ri i m m ðmÞ ðnÞ is given by XN XN −1ð ðiÞ ðjÞÞ¼ C−1ð ðmÞ ðnÞÞ ð Þ VI m ;n I i ;j : 2:15 ð Þ i¼1 j¼1 ðE m − E Þ Riþ ¼ þ i m ð − Þ ð Þ m Rm ð − Þ Rmþ1 Rm ; 2:10 Emþ1 Em The solution to this yields the cluster centers Rm and the covariance matrix VIðm; nÞ which describes the correlation þ ðmÞ where the indicates that Ei >Em. If, on the other hand, between the errors dRm and dRn. As in Eqs. (2.10) and ðmÞ Rj Rjþ ðnÞ Ei En, j− ðnÞ ðmÞ or Rn ,ifE
114025-6 − 2 αð 2 Þ 97, MUON g AND MZ : A NEW DATA-BASED ANALYSIS PHYS. REV. D 114025 (2018)
1400 Linear Linear 2000 1200 Quadratic Quadratic Cubic Cubic 1000 1500 ) [nb] ) [nb] − − π Κ + + 800 π Κ
→ →
1000 −
600 − e e + + (e (e 0 400 0 σ σ 500 200
0 0 0.6 0.65 0.7 0.75 0.8 0.85 0.9 1.01 1.015 1.02 1.025 1.03 √s [GeV] √s [GeV] (a) (b)
FIG. 3. The differences observed using linear, quadratic, and cubic integration routines in prominent resonance regions in the πþπ− and KþK− channels.
had;LO VP ¼ had;LO VP ðδ had;LO VPÞ ðδ had;LO VPÞ resonances if data are sparse. However, with the current aμ aμ aμ stat aμ sys density of cross section measurements, especially in the ðδ had;LO VPÞ ðδ had;LO VPÞ dominant hadronic channels, the differences between trap- aμ vp aμ fsr ezoidal rule integration and any higher order polynomial ¼ had;LO VP ðδ had;LO VPÞ ð Þ aμ aμ tot: 3:1 approximation are consequently small and of no concern. −10 This can be seen in the plots in Fig. 3. All results for aμ are given in units 10 . For the had;VP Δα The calculation of the uncertainty of aμ and had Δα ð 2 Þ contributions to had MZ , only the mean value with has been modified to improve the determination of the error the corresponding total uncertainty, contribution at the integral boundaries. Should the upper or lower integral boundary, Ea, with energy Em