CMS has run the first collider search for inelastic dark matter that decays into a pair of low-momentum electrons far from the collision point, and it found no excess over Standard Model expectations. That is a null result, but a useful one: it closes a gap that earlier searches left open. The CMS collaboration described the analysis on September 23, 2026, in connection with the ICHEP 2026 conference.

The idea behind the search is easy to state and hard to execute. This article works through it as a set of questions: what inelastic dark matter is, why electrons that carry barely any momentum are so hard to catch, what the detector did about it, and what the empty result does and does not rule out.

What is inelastic dark matter?

Most dark matter searches imagine a single particle that scatters off ordinary matter. Inelastic dark matter adds a twist. The dark sector contains two nearly degenerate states, usually called χ1 and χ2, and the main interaction links one to the other rather than each to itself.

The CERN Courier’s 2023 summary of an earlier CMS analysis describes how this arises: a small Majorana mass component combines with a Dirac fermion field, producing two dark matter states with mostly off-diagonal couplings. A dark photon, written A′, serves as the portal to the visible world.

At the LHC, the production chain runs as follows. A proton-proton collision makes a dark photon, which turns into a χ2 and a χ1. The χ1 escapes the detector unseen and shows up as missing transverse momentum. The heavier χ2 travels some distance and decays back into χ1 plus a pair of charged leptons.

A theory paper on inelastic dark matter at the LHC lifetime frontier argued that if the mediator is heavier than about 10 GeV, production at low-energy accelerators is kinematically suppressed. The LHC’s collision energy could, however, produce GeV-scale dark matter from dark photon decays in a range of roughly 1 to 100 GeV.

Why look for electrons that barely move?

The reason is the mass gap. When the χ2 and χ1 are close in mass, the decay products share a small energy budget. The two electrons leave with little momentum each, and they emerge close together.

That clashes with how experiments are built. Standard electron reconstruction at CMS leans on the electromagnetic calorimeter, which measures energy deposits, and it becomes efficient only above about 5 GeV of transverse momentum. A 1 GeV electron looks like noise, and when it starts several centimetres or more from the beam line it also looks nothing like a typical electron.

Feature Standard electron reconstruction Low-momentum electron algorithm used here
Main ingredient Calorimeter energy deposits plus tracks Tracks with dedicated machine learning
Efficient down to about 5 GeV transverse momentum about 1 GeV transverse momentum
Origin General-purpose Developed for studying B-hadron decays

The CMS write-up says the algorithm was originally built for B-hadron decays, where low-momentum electrons are also common. Here it is applied at TeV-scale collision energies and to electrons that appear at a measurable distance from the primary vertex.

How CMS separates a signal from the rest

Finding the electrons is only half the job. The other half is making sure that what is found is not a pileup of ordinary processes. The analysis looks for two electrons that are close together, originate from a displaced point and come with missing transverse momentum. Machine-learning classifiers use that topology to separate signal from Standard Model backgrounds, and very few background events are expected to pass the selection, according to CMS.

A displaced vertex is a strong filter. Nearly everything that happens in a proton collision comes from the primary interaction point or from short-lived particles with well-known decay lengths. Two soft electrons that point back to nothing at the beam line are unusual, and background processes that can mimic them are rare.

  • Pair of electrons: close together, soft, and from a common displaced vertex.
  • Missing transverse momentum: the escaping χ1 particles carry momentum away.
  • Collision energy: 13 TeV, as stated in the title of the CMS Physics Analysis Summary, EXO-24-036.

What came out, and what did not

CMS saw no significant excess over expectations. The collaboration describes this as the first collider limit on inelastic dark matter in the electron channel. It constrains scenarios with small mass differences and measurable displacements, a region that other approaches reach poorly.

The public news item does not give numerical exclusion limits, an integrated luminosity or a mass range, so this article does not either. Those details are in the analysis summary, which is a preliminary document and not yet a peer-reviewed paper. Readers who want the figures should take them from there.

How does this compare with the earlier search? In 2023, CMS published a search for inelastic dark matter in events with two displaced muons, using 138 fb-1 of 13 TeV proton-proton data from 2016 to 2018 and also seeing no significant excess. That analysis, published in Physical Review Letters, was described as the first dedicated collider search for the model. The electron version is its natural sibling.

The CERN Courier summary of the muon search states that couplings above roughly 10-7 to 10-9 were excluded for masses of 3 to 80 GeV, under stated assumptions, including a 10 percent mass splitting and a dark photon three times as heavy as χ1. Those figures belong to the muon analysis and should not be assumed for the electron result.

Why muons were not enough

Muons and electrons see the same dark sector but the detector treats them very differently. A muon is heavy and penetrating. An electron of 1 GeV is light and easy to scatter. If the mass splitting is small, an electron pair may be the only visible decay, or the muon pair may be kinematically forbidden.

There is also a pure coverage argument. Two channels give two independent chances to see the same underlying process, and an excess in one with a null in the other would say something about how the dark sector couples to leptons. A null in both says the model, at those parameters, is not there.

Direct detection has a related blind spot, though the sources used here do not spell it out. The theory literature notes that the calculation of dark matter scattering off nucleons and electrons matters for estimating direct detection rates, and in inelastic models the scattering is suppressed when the two states differ by more than the energy available in the collision. That general feature is why collider searches are valuable here. For a view of how underground searches are doing, see our article on the LZ single-event result.

What comes next

The CMS article says the larger Run 3 dataset will allow future searches to probe lower masses and longer displacements. That is a forward-looking statement and not a result. Run 3 collected more data at a similar energy, and the hardest part of the program is keeping triggers efficient for soft, displaced particles that the standard triggers were not designed for.

The wider point is about method. A signal that is soft and delayed hides in plain sight unless the reconstruction is built to look for it. The low-momentum electron tool shows that existing detectors can reach new territory by rewriting software, not hardware.

Two practical details matter for what follows. First, a search like this depends on the trigger, the part of the experiment that decides in microseconds which collisions are worth recording. Soft electrons from a displaced decay are exactly the kind of signal a trigger can discard, so analyses of this kind lean on other features of the event, such as the missing transverse momentum, to get recorded at all. Second, a preliminary result can still change. The analysis summary is not a journal paper, and the final numbers may shift after review.

How to read a null result like this

An empty search is not an absence of physics. It is a statement with edges. The result excludes a certain combination of mass, mass splitting and decay distance, and leaves everything outside that combination open. For a displaced search the decay distance is the awkward axis: if the χ2 decays too quickly, the electrons look prompt and fall into the enormous ordinary background. If it decays too far out, the particle leaves the detector before anything happens.

So the limit plot has a sweet spot, and the value of a new technique is that it widens it. Pushing the electron threshold from about 5 GeV down to about 1 GeV widens it on the side of small mass differences. When the full analysis is read, three things deserve attention before trusting any headline: the assumed mass splitting, the range of decay lengths covered, and the integrated luminosity behind the limit.

It also helps to separate what the theory says from what the data says. The model is a hypothesis with a handful of free parameters. The search is a measurement. Nothing in a null result makes the dark sector less likely to exist; it only shrinks the places it could be hiding.

Sources