Identifying the origin of the 146-GeV excess at the LHC

You should know two things about this paper. First, the core idea is actually new: nobody has proposed using Hγ, Hj, and Hjj to separate the mixing and leptophilic interpretations of the CMS 146 GeV eµ excess, and the quantitative HL-LHC projections are concrete. Second, the headline 7.1/7.6σ numbers rest on a leptophilic benchmark point whose vacuum stability and perturbativity the authors explicitly say they do not check. That is the load-bearing soft spot, and they know it.

What the paper does well: the statistical treatment is internally consistent—the yields in Table 1 reproduce the quoted Z values under the Poisson likelihood ratio, and the quadrature combination gives 7.6. The ISR/FSR photon discrimination via the two signal regions (SR A on m_eµ, SR B on m_eµγ with pT(eµγ)<10 GeV) is a sensible and clean handle. The paper also engages seriously with low-energy constraints—MEG II, LEP dimuon, muonium-antimuonium—and explains how λ7 can create a blind spot for µ→eγ. The mixing benchmark is at least partially checked (λ3 is said to be dictated by vacuum stability), and the Hj channel is honestly reported as weak (Z=0.6).

The soft spots. The leptophilic benchmark (Y_eµ=0.7/√2, Y_ττ=−0.4/√2, λ3=10, λ7=5.2, m_A=m_H+=600 GeV) involves O(1) Yukawas and a large quartic. The authors state in Section 2.2 that 'a consistency check with vacuum stability and perturbativity requires a more involved analysis, which is beyond the scope of this paper.' For a paper whose final significance number depends on that point, this is a genuine gap, not a formality. If that point is excluded, the leptophilic yields in the Hγ signal regions (22.0 and 27.6 events at 3000 fb⁻¹) could change substantially. Also, the analysis uses Delphes with flat 20% systematic uncertainties, and no code or UFO files are provided; reproducibility is limited. The good news is that the authors already show the combined significance drops to 5.4 under a 20% systematic, so the qualitative conclusion—that the channels can distinguish the two scenarios—is not extremely fragile.

Who is this for: readers working on LFV Higgs phenomenology and HL-LHC projections. They will find the discrimination strategy and the signal-region definitions useful. But they should not treat 7.6 as a prediction; it is a conditional projection at an unvalidated benchmark.

My recommendation: send it to peer review. A good referee should require the authors to either demonstrate that the leptophilic benchmark satisfies vacuum stability and perturbativity or rerun the analysis with a benchmark that does. They should also be asked to release the simulation setup or a simplified version. The idea is worth the referee time, even if the current numbers need to be reworked.

The central 7.1/7.6 sigma claim rests on the leptophilic benchmark specified in Section 2.2 (Y_eµ=0.7/√2, Y_ττ=−0.4/√2, λ3=10, λ7=5.2, m_A=m_H+=600 GeV). The paper explicitly states: 'A consistency check with vacuum stability and perturbativity requires a more involved analysis, which is beyond the scope of this paper.' The Hγ yields (22.0 and 27.6 events at 3000 fb^-1) and the other leptophilic yields in Table 1 are computed at this unvalidated point. If this point violates bounded-from-below or unitarity constraints, the leptophilic scenario as constructed is not viable, and the quoted significance is not tied to a consistent model. The large λ7=5.2 needed to suppress two-loop µ→eγ and the O(1) Yukawas make this a genuine risk rather than a formality. The mixing benchmark, by contrast, is at least partially checked (λ3 is said to be 'dictated by vacuum stability').

The paper's quantitative claim is a projection based on two model benchmarks. The mixing benchmark is checked against vacuum stability in Section 2.1, but the leptophilic benchmark in Section 2.2 explicitly defers the required theoretical consistency check. Since all leptophilic yields in Table 1 derive from this benchmark, any invalidity would directly change the predicted event counts and the LLR significances of Eq. (16). Other concerns (20% systematics, lepton PDF uncertainties, fast simulation) are acknowledged in the text and would soften but not invalidate the qualitative conclusion. Therefore the unvalidated theoretical consistency of the leptophilic benchmark is the most load-bearing uncertainty. A dedicated scan of the scalar potential would either retire the concern or force a re-evaluation of the projected sensitivity.

If correct, the paper would demonstrate a concrete experimental program for identifying the production mechanism of a 146 GeV resonance at the HL-LHC, with a strong statistical separation in the Hγ channel. The analysis is internally consistent: the yields in Table 1 reproduce the quoted Z values under Eq. (16), and the quadrature combination of Z_A and Z_B gives about 7.1. The use of explicit benchmark points and public event-generation tools is a strength, as is the clear targeting of physically different photon/jet topologies. However, the headline significance hinges on a single leptophilic benchmark whose theoretical consistency is explicitly deferred, and on the orthogonality of the two Hγ signal regions.

The paper asks whether the LHC can determine the production mechanism of a possible new particle at 146 GeV that decays to an electron-muon pair, if the excess hinted by LHC data persists. Two scenarios are on the table: the resonance is produced like the 125 GeV Higgs boson, via mixing with it (hadronic initial states), or it is produced directly from electron-muon collisions involving sea leptons inside the proton (leptophilic initial states). Using a two-Higgs-doublet model as a concrete realization, the authors simulate three final states, $H\gamma$, $Hj$, and $Hjj$, at the high-luminosity LHC with $3000~\text{fb}^{-1}$. They find that the $H\gamma$ channel alone separates the two origins at $7.1\sigma$, and all three channels combined reach $7.6\sigma$ (about $5\sigma$ after a rough 20% systematic allowance). The result matters because it turns an ambiguous excess into a testable diagnosis of how the resonance was born.

What carries the argument. The mechanism that carries the argument is the contrast between initial-state and final-state photon radiation in $H\gamma$ production, translated into two complementary signal regions. SR A selects events where the $e\mu$ invariant mass peaks at 146 GeV, tagging photons radiated from initial-state leptons; SR B requires the full $e\mu\gamma$ system to reconstruct 146 GeV and $p_T(e\mu\gamma)<10~\text{GeV}$, tagging final-state radiation while suppressing mixing-scenario events that recoil against jets. The $Hj$ and $Hjj$ channels use the same logic for jets: hadronic production radiates hard jets, leptophilic production does not. The quantitative discriminator is the Poisson likelihood-ratio test statistic $Z=\sqrt{2[\mu_1\ln(\mu_1/\mu_0)-(\mu_1-\mu_0)]}$, with $\mu_0$ the leptophilic-plus-background hypothesis and $\mu_1$ the mixing-plus-background hypothesis; using the leptophilic case as the null makes the quoted significance conservative.

Core claim. The central discovery claim is that the origin of the 146 GeV $e\mu$ resonance can be identified by the kinematics of its associated radiation. In the mixing scenario the resonance is produced from gluon fusion, so a hard associated photon is mostly final-state radiation off a lepton, and the $e\mu\gamma$ system recoils against jets. In the leptophilic scenario the resonance comes from $e\mu$ annihilation via lepton parton densities, so the photon is often initial-state radiation, and the $e\mu$ pair reconstructs the mass on its own. The authors define two signal regions in $H\gamma$: SR A on $|m_{e\mu}-146~\text{GeV}|<2.5~\text{GeV}$ and SR B on $|m_{e\mu\gamma}-146~\text{GeV}|<2.5~\text{GeV}$ with $p_T(e\mu\gamma)<10~\text{GeV}$. At $3000~\text{fb}^{-1}$ the leptophilic scenario predicts 22.0 and 27.6 events in these regions, while mixing predicts 0.0 and 4.5, over small backgrounds; the combined separation is $Z=7.1$. Adding monojet-like $Hj$ ($Z=0.6$) and VBF-like $Hjj$ ($Z=2.8$) yields $Z=7.6$.

The premise it rests on. The load-bearing premise is that the chosen leptophilic benchmark point, with $Y_{e\mu}=0.7/\sqrt{2}$, $Y_{\tau\tau}=-0.4/\sqrt{2}$, $\lambda_3=10$, $\lambda_7=5.2$, and $m_A=m_{H^\pm}=600~\text{GeV}$, is a physically viable scalar potential; the paper does not perform the vacuum-stability and perturbativity check, so if that point is invalid the predicted leptophilic yields and the claimed $7.6\sigma$ separation would change.

At $3000~\text{fb}^{-1}$, the $H\gamma$ channel alone can separate the two production scenarios at $7.1\sigma$ by combining SR A and SR B.

Adding the $Hj$ and $Hjj$ channels raises the combined separation to $7.6\sigma$ under statistical uncertainties only.

If a flat 20% systematic uncertainty is assigned to expected yields, the combined separation falls to about $5\sigma$.

In the $H\gamma$ signal regions, the mixing scenario leaves almost no events (0.0 and 4.5), while the leptophilic scenario gives 22.0 and 27.6, so one measurement with $3000~\text{fb}^{-1}$ would visibly pick a side.

The $p_T(e\mu\gamma)<10~\text{GeV}$ cut is a transferable handle: any narrow dilepton resonance whose production could be hadronic or leptonic can be classified by asking whether the dilepton system reconstructs the mass before or after adding a hard photon.

The quoted yields rely on one leptophilic benchmark whose vacuum stability and perturbativity are not checked in the paper; if that point is excluded, the 22.0 and 27.6 predictions and hence the $7.6\sigma$ number would move.

A full treatment of lepton-PDF uncertainties and detector systematics would bracket the yield tables; the paper's 20% profiling is a first estimate, not a complete error budget.

If the excess persists but the observed $H\gamma$ yields land between the two predictions, the simple two-scenario dichotomy would need a third option, such as a mixed production mechanism or modified couplings.

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review