Observation of B+ ->K+ eta gamma

We report measurements of radiative B decays with K eta gamma final states, using a data sample of 253 /fb recorded at the Upsilon(4S) resonance with the Belle detector at the KEKB e+e- storage ring. We observe B+ ->K+ eta gamma for the first time with a branching fraction of (8.4 +- 1.5(stat) +1.2 -0.9(syst)) X 10^{-6} for M(Keta)<2.4 GeV/c^2, and find evidence of B0 ->K0 eta gamma. We also search for B ->K3*(1780) gamma.

Radiative B decays, which proceed mainly through the b → sγ process 2 , have played an important role in a search for physics beyond the Standard Model (SM).Although the inclusive branching fraction has been measured to be (3.3±0.4)×10−4 [1], we know little about its constituents.So far, measured exclusive final states such as K * (892)γ [2,3], K * 2 (1430)γ [2,4], Kππγ [4] and Kφγ [5] only explain one third of the inclusive rate.Detailed knowledge of exclusive final states reduces the theoretical uncertainty in the measurement of the inclusive B → X s γ branching fraction using the pseudo-reconstruction technique, as well as in the measurement of B → X s ℓ + ℓ − [6].In this analysis, the decay mode B → Kηγ is studied for the first time.In addition to improving the understanding of b → sγ final states, B 0 → K 0 S ηγ can be used to study time-dependent CP asymmetry [7], which is sensitive to physics beyond the SM.The mode B → Kηγ can also be used to search for radiative B decays through possible Kη resonances, e.g., K * 3 (1780) observed by the LASS experiment [8].
The analysis is based on 253 fb −1 of data taken at the Υ(4S) resonance (on-resonance) and 28 fb −1 at an energy 60 MeV below the resonance (offresonance), which were recorded by the Belle detector [9] at the KEKB asymmetric e + e − collider (3.5 GeV on 8 GeV) [10].The on-resonance data corresponds to 275 million BB events.The Belle detector is comprised of a silicon vertex detector, a 50-layer central drift chamber (CDC), an array of aerogel Cherenkov counters (ACC), time-of-flight scintillation counters (TOF) and an electromagnetic calorimeter of CsI(Tl) crystals (ECL) located inside a superconducting solenoid coil that provides a 1.5 T magnetic field.An instrumented iron flux-return for K 0 L /µ detection is located outside the coil.Two different inner detector configurations were used.For the first sample of 152 million BB pairs, a 2.0 cm radius beampipe and a 3-layer silicon vertex detector were used; for the latter 123 million BB pairs, a 1.5 cm radius beampipe, a 4-layer silicon detector and a small-cell inner drift chamber were used [11].
We reconstruct B + → K + ηγ and B 0 → K 0 S ηγ via η → γγ and η → π + π − π 0 .All charged tracks used in the reconstruction (except charged pions from K 0 S ) are required to have a center-of-mass (CM) momentum greater than 100 MeV/c and to have an impact parameter within ±5 cm of the interaction point along the positron beam axis and within 0.5 cm in the transverse plane.
In order to identify kaon and pion candidates, we use a likelihood ratio based on the light yield in the ACC, TOF information and specific ionization measurements in the CDC.For the requirement applied on the likelihood ratio, we obtain an efficiency (pion misidentification probability) of 90% (10%) for charged kaon candidates, and an efficiency (kaon misidentification probability) of 98% (10%) for charged pion candidates.Tracks identified as electrons are excluded.
K 0 S candidates are formed from π + π − combinations with invariant mass within 8 MeV/c 2 (∼ 2σ) of the nominal K 0 S mass.The two pions are required to have a common vertex displaced from the interaction point.The K 0 S momentum direction is required to be consistent with the K 0 S flight direction.Neutral pion candidates are formed from pairs of photons that have an invariant mass within 16 MeV/c 2 (∼ 3σ) of the nominal π 0 mass and a momentum greater than 100 MeV/c in the CM frame.Each photon is required to have an energy greater than 50 MeV in the laboratory frame.A mass-constrained fit is then performed to obtain the π 0 momentum.
For η → γγ reconstruction, we require that the invariant mass of the two photons satisfy 0.515 GeV/c 2 < M γγ < 0.570 GeV/c 2 and that each photon have an energy greater than 50 MeV in the laboratory frame.We also require | cos θ η hel | < 0.9, where θ η hel is the angle between the photon momentum and η boost direction from the laboratory frame in the η rest frame.A massconstrained fit is then performed to obtain the η momentum.For η → π + π − π 0 , we apply a requirement on the three-pion invariant mass, 0.532 GeV/c 2 < M π + π − π 0 < 0.562 GeV/c 2 .
We reconstruct B meson candidates from an η, a charged or neutral kaon and the highest energy photon within the acceptance of the barrel ECL (33 • < θ γ < 128 • , where θ γ is the polar angle of the photon with respect to the electron beam in the laboratory frame).Here, the invariant mass of the Kη system is required to be less than 2.4 GeV/c 2 .This selection corresponds to γ is the photon energy in the B rest frame, and includes 84% of events from the b → sγ process.The highest energy photon candidate is required to be consistent with an isolated electromagnetic shower, i.e., 95% of the energy in an array of 5 × 5 crystals should be concentrated in an array of 3 × 3 crystals and no charged tracks should be associated with it.In order to reduce the background from decays of π 0 and η mesons, we combine the photon candidate with each of the other photons that have CM energy greater than 30 MeV (200 MeV) in the event and reject the event if the invariant mass of any pair is within 18 MeV/c 2 (32 MeV/c 2 ) of the nominal π 0 (η) mass.This condition is referred to as the π 0 /η veto.
We use two independent kinematic variables for the B reconstruction: the beam-energy constrained mass , where E * beam is the beam energy, and p * γ , E * γ , p * Kη , E * Kη are the momenta and energies of the photon and the Kη system, respectively, calculated in the CM frame.In the M bc calculation, the photon momentum is rescaled so that | p * γ | = (E * beam − E * Kη )/c is satisfied.We require M bc > 5.2 GeV/c 2 and −150 MeV < ∆E < 80 MeV.We define the B signal region to be M bc > 5.27 GeV/c 2 .In the case that multiple candidates are found in the same event, we take the candidate that has the η mass closest to the nominal mass3 after applying the background suppression described later.
The largest source of background originates from continuum e + e − → qq (q = u, d, s, c) production including contributions from initial state radiation (e + e − → qqγ).In order to suppress this background, we use the likelihood ratio (LR) described in Ref. [4], which utilizes the information from a Fisher discriminant [12] formed from six modified Fox-Wolfram moments [13] and the cosine of the angle between the B meson flight direction and the beam axis.The LR requirement retains 44% of the signal, while rejecting 98% of the continuum background.
In order to extract the signal yield, we perform a binned likelihood fit to the M bc distribution.The M bc distribution of the signal component is modeled by a Crystal Ball line shape [14], with the parameters determined from the signal Monte Carlo (MC) and calibrated using control samples of The M bc distribution of the continuum background is modeled by an ARGUS function [15] whose shape is determined from the off-resonance data.Here, the LR requirement is not applied to the off-resonance data in order to compensate for the limited amount of data in that sample.The possible bias due to this is taken as systematic error on the fitted yield.Background from hadronic B decays is divided into two components, which we refer to as generic BB background and rare B background in this paper.The former comprises B decays through b → c transitions including color-suppressed B decays such as B 0 → D 0 π 0 , and the latter covers charmless B decays.Each of them is modeled by another ARGUS function.The shape of these distributions is determined using corresponding MC samples.In order to study the contamination from other b → sγ decays, we examine a B → K * (892)γ MC sample and an inclusive b → sγ MC sample that is modeled as an equal mixture of sd and su quark pairs and is hadronized using JETSET [16], where the X s mass spectrum is fitted to the model of Kagan and Neubert [17].We find that feed-down from other b → sγ decays is small, but not negligible, and model its M bc distribution with an ARGUS function.
Figure 1 shows the M bc distributions for B + → K + ηγ and B 0 → K 0 S ηγ, respectively.These distributions, as well as the distribution for the combined mode, are fitted to the sum of signal, continuum, generic BB, rare B background and b → sγ feed-down components.In the fit, the normalization of generic BB, rare B and b → sγ are fixed according to the luminosity and b → sγ branching fraction, while the normalization of the continuum component is allowed to float.We find signal yields of 81 ± 14, 20.9 +7.3 −6.5 and 102 ± 16 events with statistical significances of 7.1σ, 3.7σ and 8.1σ, for the charged, neutral and combined modes, respectively.Here, the significance is defined as −2 ln(L(0)/L max ), where L max and L(0) are the maximum values of the likelihood when the signal yield is left free or fixed to zero, respectively.
Figure 2 shows the γγ and π + π − π 0 invariant mass distributions for events inside the B signal region.Here, we do not apply the best candidate selection.We observe clear peaks at the nominal η mass.The Kη invariant mass distribution for events inside the B signal region is shown in Fig. 3. Here, the background distributions are obtained from the off-resonance data without the LR requirement or from the corresponding MC samples, and are normalized using the fit result.We find that the signal is concentrated between 1.3 GeV/c 2 and 1.9 GeV/c 2 and is falling above 1.9 GeV/c 2 .Therefore, our requirement M Kη < 2.4 GeV/c 2 is expected to include most of the B → Kηγ signal.We do not see any clear resonant structure in the M Kη distribution.
The systematic error on the signal yield due to the fitting procedure is estimated by varying the value of each fixed parameter by ±1σ and extracting the new signal yield for each case.The difference between the background shape for the continuum MC with and without the LR requirement is taken as an additional error to the continuum background shape.We set the normalization of either the generic BB or rare B backgrounds to zero and to twice its nominal value to account for its uncertainty.The changes of the yields for each procedure are added in quadrature, and are regarded as the systematic error on the signal yield.We also calculate a statistical significance for each case, and regard the smallest value as the significance including the systematic error.The result is listed in Table 1.
The signal reconstruction efficiency is estimated using the MC simulation and is corrected for discrepancies between data and MC using control samples.The signal MC has uniform Kη invariant mass and cos θ hel distributions, where θ hel is the decay helicity angle between the kaon momentum and opposite to B momentum in the Kη rest frame.We find that the efficiency is almost independent of the Kη invariant mass and cos θ hel .Table 1 shows the signal efficiencies and the branching fractions for each B → Kηγ mode.Here, we assume an equal production rate for B 0 B 0 and B + B − .The error on the branching frac- tion includes the following systematic uncertainties: photon detection (2.8%), tracking (1.0% to 1.2% per track), kaon identification (0.8%), pion identification (0.5% per pion), K 0 S detection (4.5%), π 0 detection (1.5%), η detection in η → γγ mode (2.0%), π 0 /η veto and LR (5.9% and 4.4% for charged and neutral modes, respectively), possible Kη mass dependence of the efficiency (2.1% and 4.4% for charged and neutral modes, respectively), possible cos θ hel dependence of the efficiency (2.5% and 3.4% for charged and neutral modes, respectively), uncertainty in the η branching fraction (0.7% for η → γγ and 1.8% for η → π + π − π 0 ), and uncertainty in the number of BB events (1.1%).The systematic errors from the π 0 /η veto and LR requirement are estimated using control samples of −3.2 events for the charged, neutral and combined modes, respectively.Here and in the following, we quote statistical and systematic errors in the first and second position.The M bc distribution and fit result for the combined mode is shown in Fig. 4. We provide only upper limits due to our inability to distinguish B → K * 3 (1780)γ from nonresonant decays.The 90% confidence level upper limit N is calculated from the relation N 0 L(n)dn = 0.9 ∞ 0 L(n)dn, where L(n) is the maximum likelihood in the M bc fit with the signal yield fixed at n.In order to include the systematic errors from the fitting procedure in the upper limit for the yield, the positive systematic error is added to N. The obtained yield upper limits, efficiencies and products of branching fractions B(B → K * 3 (1780)γ)×B(K * 3 (1780) → Kη) are listed in Table 2. Here, the number of BB events and the reconstruction efficiency are lowered by 1σ when we calculate the upper limit for the branching fractions.If we assume B(K * 3 (1780) → Kη) = (11 +5 −4 )% [18], the 90% confidence level limits correspond to B → K * 3 (1780)γ branching fractions of 3.9 × 10 −5 , 8.3 × 10 −5 and 3.7 × 10 −5 , respectively for charged, neutral and combined modes, which substantially improve the limits set by the ARGUS collaboration [19].Some extensions of the SM predict a large CP asymmetry in the b → sγ process [20].We measure the partial rate asymmetry where N ∓ is the signal yield for B ∓ → K ∓ ηγ and w is the probability that a signal event is reconstructed with the wrong kaon (and hence B) charge.This probability is found to be less than 1% in our signal MC sample, and hence we ignore its negligible effect on A CP .N ∓ is obtained by fitting separately the M bc distributions for the negatively and positively charged modes shown in Fig. 5.We find N − = 34.0+9.8 −9.0 and N + = 46.7 +10.5 −9.8 .The systematic error on A CP consists of the following contributions.The error from the fitting procedure is estimated to be 0.045 by varying each fixed parameter one by one, and extracting A CP for each procedure, in the same way as before.Here, we assume no asymmetry for the generic BB background, but allow 100% asymmetry for the rare B and 6% asymmetry for b → sγ [21].The error from the overall detector bias is studied with the B 0 → D − (K − π + π 0 )π + control sample and is found to be 0.035.By adding these errors and the possible asymmetry in kaon identification (0.014) in quadrature, we obtain A CP = −0.16± 0.09 ± 0.06.
In conclusion, we observe the decay mode B + → K + ηγ and find the first evidence of B 0 → K 0 ηγ.The branching fraction and partial rate asymmetry of B + → K + ηγ are measured to be (8.4±1.5 +1.2 −0.9 )×10 −6 and −0.16±0.09±0.06 for M Kη < 2.4 GeV/c 2 .The branching fraction of B 0 → K 0 ηγ is measured to be (8.7 +3.1 −2.7 +1.9 −1.6 ) × 10 −6 .We also search for B → K * 3 (1780)γ, but find no evidence.Although the signal yield for B 0 → K 0 S ηγ is small, this mode can be used in the near future to study time-dependent CP asymmetries in radiative B decays and to search for new physics.

Table 1
Measured signal yields, efficiencies, branching fractions (B) and significances including systematic error (S) for B → Kηγ.The first and second errors are statistical and systematic, respectively.Efficiencies include the sub-decay branching fractions.