Search in leptonic channels for heavy resonances decaying to long-lived neutral particles

This is the first CMS search for a heavy resonance decaying to two long-lived neutral particles that each decay to a displaced dilepton pair, on 4.1 (5.1) fb⁻¹ at 7 TeV. The result is a null one: 4 candidates in electrons against 1.4 expected, 0 in muons against 0.02, and 95% CL limits on σB of roughly 0.7–10 fb for 0.1 < cτ < 200 cm, 10–100 fb at M_H0 = 125 GeV. That is a real, new constraint on hidden-valley and split-SUSY-style benchmarks, and it is stated without inflation. The selection is sensible for the topology — high-purity displaced tracks, a common vertex fit with a χ² cut, a collinearity requirement, a hollow isolation cone — and the background work is better than it needed to be: two-exponential fits to Lxy/σxy, a data-driven mass shape with the lifetime cuts removed, and three independent cross-checks in Section 6.4 that all agree. The systematic table is enumerated rather than hand-waved, and the authors flag their own coverage gaps (no electron limits below M_H0 = 200 GeV, no exact limits for unmodelled masses, monotonic interpolation only). I'd take those acknowledgements at face value; they are not evasions. The soft spot is the one the stress test names, and it mostly lands. The entire cτ dependence of Figs. 5–7 enters through ε1, obtained by reweighting a single sample generated at a mean transverse decay length of ≈20 cm out to ≈600 cm, and neither Table 1 nor the figures carry any MC-statistical term for that reweighting. I'd temper the concern, though: at long cτ the efficiency is dominated by the geometric probability that the X decays inside the tracker at all, not by the tail of the generated sample, so the shape is more robust than a naive e⁻⁵ tail argument suggests. The missing uncertainty is a precision gap, not a soundness one, and it matters least where the limits are already weakest. The 20% tracking systematic, transferred from cosmic muons to electrons by assumption, is large but conservative and openly discussed. Who should read it: anyone working on long-lived particle searches or setting hidden-valley benchmarks, and anyone who wants a clean template for a displaced-vertex sideband analysis. It has been superseded by Run 2, so cite it as the channel's opening shot rather than for current numbers. Send it to referees — the analysis deserves the scrutiny and will survive it.

The paper's headline is the limit on σB as a function of the X lifetime, cτ (Figs. 5–7 and the Summary's '0.7–10 fb for 0.1 < cτ < 200 cm'). The entire cτ dependence enters through the signal efficiency ε1, which Section 4.1 obtains by reweighting a single simulated sample generated at a mean transverse decay length of ≈20 cm. The paper states the reweighting range (mean transverse decay lengths ≈0.7–600 cm) but never quotes the statistical uncertainty on ε1 after reweighting, and Table 1's systematic budget (tracking 20%, trigger 2.6%/11%, pileup, PDF, scales, NLO, luminosity) contains no term for it. At the extremes of the quoted range the effective number of MC events is exponentially suppressed: for a 20 cm sample, the fraction of events with true decay length beyond ~100 cm is ~e^-5, and these are the events that dominate the reweighted efficiency at cτ ≈ 200 cm. If the bootstrap uncertainty on ε1(cτ) is, say, 30–50% at the endpoints, the limits there should be widened by the same factor, and the smooth curves in Figs. 5–7 overstate the precision of the cτ dependence. This is a precision concern, not a soundness one: the qualitative claim (limits in the few-to-tens-of-fb range) survives a factor-2 shift, but the paper's quantitative statement about the cτ shape does not currently carry an uncertainty that reflects its dominant input.

The reader's weakest_assumption (background shape/normalisation extrapolated from lifetime-loosened sidebands, Section 5.2) is a real soft spot, but the paper bounds its impact: the background normalisation affects the limits only near the Z (Section 7), the quoted X masses (20, 50, 150, 350 GeV) are not near the Z, and three independent cross-checks (Section 6.4) give consistent results. So I agree only partially that this is the most load-bearing concern for the headline numbers. The more load-bearing input for the headline is ε1(cτ), because the limit is essentially NX_limit/(2Lε1) and NX_limit ≈ 3.0 is nearly cτ-independent; the entire shape of the limit curve versus cτ is carried by the reweighted efficiency. The paper is otherwise careful: the efficiency formula (Eq. 1) is correct, the trigger and tracking systematics are estimated from data (tag-and-probe, cosmic muons, embedded tracks), the background is cross-checked three ways, and the limitations (no electron limits below MH0 = 200 GeV, no exact limits for unmodelled masses) are stated. The reweighting-statistics gap is a precision issue at the endpoints of the quoted cτ range, not a threat to the qualitative claim, so the verdict remains ACCEPT; I would ask the authors to add the bootstrap band to Figs. 5–7 and to Table 1.

This is the first CMS search for displaced dilepton vertices from long-lived neutral particles, and it opens a kinematic regime that the earlier D0 searches did not reach and that the contemporaneous ATLAS searches addressed through the muon spectrometer rather than the tracker. The model-independent limits on the mean number of X bosons in Fig. 4 are a useful, reusable product independent of the benchmark H0 -> XX hypothesis, and the paper is explicit about how the limits translate to alternative production modes (Z' spin-1 resonances, H0 -> XY, all-SM Higgs production), which is the kind of falsifiable mapping that makes a search result durable. The analysis is careful in its background treatment: three independent cross-checks of the normalisation are described in Section 6.4, the mass shape is derived from data rather than simulation, and the systematic budget in Table 1 is itemised. The stated limitations (no electron-channel limits below M_H0 = 200 GeV/c^2, the assumption of uncorrelated single-lepton trigger efficiencies) are acknowledged rather than hidden. If the result holds, the benchmark long-lived-particle scenario is excluded above the quoted rates across a wide mass,

The CMS collaboration asks whether a heavy resonance could decay into two massive neutral particles that each travel a measurable distance before turning into a pair of opposite-charge leptons. Ordinary standard-model processes almost never produce that signature: two tracks meeting at a vertex displaced from the proton collision point, with the pair's invariant mass forming a narrow peak. Using 4.1 fb$^{-1}$ of 7 TeV electron data and 5.1 fb$^{-1}$ of muon data, the search finds four electron-channel candidates and none in the muon channel, both consistent with the small expected background. What the reader gets is a quantified exclusion: 95% confidence upper limits on $pp$ production of a resonance decaying to two long-lived particles, each to $e^+e^-$ or $mu^+mu^-$, typically 0.7-10 fb for lifetimes $0.1 < c tau < 200$ cm, rising to 10-100 fb for a 125 GeV/$c^2$ resonance.

What carries the argument. The load-bearing mechanism is a displaced-vertex selection built to keep leptons that standard prompt-lepton identification would throw away. Tracks must be high-purity, have $p_T > 41$ GeV/$c$ (electrons) or $33$ GeV/$c$ (muons), lie at $|eta| < 2$, and carry a transverse impact-parameter significance $|d_0/sigma_d| > 3$ (2). Pairs of opposite-charge candidates are fitted to a common vertex with $chi^2/ ext{dof} < 5$ and required to sit more than 8 (5) standard deviations from the primary vertex in the transverse plane, with no track hit closer to the beam line than the vertex itself. Isolation in a hollow cone $0.03 < Delta R < 0.3$, a collinearity requirement between the candidate momentum and the primary-to-secondary-vertex direction, a cosmic-ray opening-angle cut, and a dilepton mass above 15 GeV/$c^2$ complete the selection. The background is modelled in two pieces: the transverse decay-length significance distribution is fitted with two falling exponentials and integrated into the signal region to normalise the background, while the mass shape is taken from data with the lifetime-related cuts removed and fitted as a Breit-Wigner times a turn-on function plus a falling exponential. Limits come from an unbinned likelihood fit to the dilepton mass spectrum with the CL$_s$ method, with each systematic uncertainty entering as a log-normal nuisance parameter.

Core claim. The paper's claim is a null result with teeth. No significant excess of displaced dilepton vertices is observed over standard-model expectations, and that absence is converted into 95% confidence upper limits on the production cross section of a heavy resonance times the branching fraction of the long-lived neutral particle $X$ to leptons, as a function of the $X$ lifetime. For the benchmark model, a gluon-fused $H^0$ decaying to $XX$ with each $X to ell^+ell^-$, the limits are typically 0.7-10 fb over $X$ lifetimes $0.1 < c tau < 200$ cm for $H^0$ masses of 200-1000 GeV/$c^2$ and $X$ masses of 20-350 GeV/$c^2$, and 10-100 fb for $M_{H^0} = 125$ GeV/$c^2$. The authors state that these are the most stringent limits in these channels to date. The observed event counts are four electron-channel candidates, two of them above the $Z$ mass, against an expected background of $1.4^{+1.8}_{-1.2}$, and zero muon-channel candidates against $0.02^{+0.09}_{-0.02}$.

The premise it rests on. The background under the signal peak is extrapolated from data with the lifetime-related cuts removed, on the assumption that dilepton mass and lifetime-related variables are only weakly correlated in background events, so if that correlation is stronger than the simulation indicates the quoted limits shift, especially for $X$ masses near the $Z$.

The benchmark Higgs-to-hidden-sector scenario, $H^0 to XX$ with $X to ell^+ell^-$, is excluded above roughly 0.7-10 fb across the quoted mass and lifetime window, so any model predicting a larger rate in this topology is ruled out by these data.

Because the limits are quoted on the mean number of $X$ bosons passing the selection, they can be re-used for any production model once its acceptance is known; the paper explicitly checks that a spin-1 $Z'$ resonance changes the acceptance by less than 3% at 1000 GeV/$c^2$ and less than 25% at 200 GeV/$c^2$ in the muon channel.

For a 125 GeV/$c^2$ resonance the limits are 10-100 fb, roughly an order of magnitude weaker than at high mass, because the high lepton $p_T$ thresholds accept only a few percent of the signal there.

The displaced-vertex technique demonstrated here extends the reach of prompt-lepton searches to lifetimes of order metres, a region where conventional analyses are blind.

The same event topology is a natural target for a lifetime-dependent scan rather than a fixed set of mass points; the paper's own admission that no simulated samples exist for intermediate masses and that limits must be inferred by monotonicity suggests a gap a dedicated grid would close.

Because the search is nearly background-free, its reach scales roughly linearly with luminosity until systematic uncertainties on track-finding dominate, so the 20% tracking-efficiency uncertainty, not statistics, is likely to be the binding constraint in any high-luminosity repetition.

The method transfers directly to other long-lived candidates that decay to lepton pairs, such as long-lived sleptons or heavy neutral leptons, provided the vertex and isolation requirements are retuned to the softer lepton spectrum those would produce.

A future observation of even a handful of displaced vertices clustered at one dilepton mass would be far more informative than a rate measurement, since the background estimate under the peak is essentially zero.

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