Medium Energy Starting Events (MESE)
- MESE is a neutrino event selection method that identifies contained-vertex interactions down to 1 TeV, extending IceCube’s HESE program.
- Its multi-stage veto architecture—including outer-layer, downgoing track, and fiducial cuts—reduces atmospheric muon backgrounds by up to ten orders of magnitude.
- The approach leverages deep neural network classification to differentiate event topologies, enabling precise measurements of the diffuse astrophysical neutrino spectrum and flavor composition.
Medium Energy Starting Events (MESE) is an IceCube contained-vertex event selection for neutrino interactions whose vertices lie inside the detector and whose energies extend down to about . It is a veto-based, all-flavor, all-sky starting-event sample developed as a lower-threshold extension of High-Energy Starting Events (HESE), designed to reject penetrating atmospheric muons while retaining sensitivity to diffuse astrophysical neutrinos, spectral structure from the TeV to PeV scale, and flavor composition through topology-separated fits (Basu et al., 2023, Basu et al., 8 Jul 2025, V. et al., 9 Jul 2025).
1. Definition and role within IceCube event samples
In IceCube usage, a starting event is an interaction for which the neutrino interacts inside the detector volume, so that the interaction vertex is contained and the outer detector can be used as an active veto against entering atmospheric muons. MESE applies this starting-event logic in the medium-energy regime. The 2023 MESE methods paper presents it explicitly as an extension of HESE from a threshold of order down to about , while the 2025 diffuse-spectrum proceedings describes MESE as IceCube’s low-threshold, all-flavor, all-sky contained-vertex sample for measuring the diffuse astrophysical neutrino flux from roughly to (Basu et al., 2023, Basu et al., 8 Jul 2025).
The sample is intended to preserve the principal advantages of starting-event analyses—full-sky acceptance, direct visibility of the initial interaction, and sensitivity to all neutrino flavors—while operating in an energy range where atmospheric backgrounds are much larger than in HESE. The 2025 flavor-composition analysis defines MESE as a selection of events with interaction vertices inside the detector and energies of at least , and states that the flavor fit assumes the flavor ratio is constant across the energy scales considered there, namely to . In that analysis, the MESE sample contains a total of 9888 events over 11.4 years of IceCube data (V. et al., 9 Jul 2025).
MESE is therefore not a single one-off dataset but a methodological class of IceCube starting-event analyses. The 2023 sensitivity paper stresses that all HESE-tagged events are retained in the final MESE sample, and that MESE then adds lower-energy starting events that survive a more elaborate veto chain (Basu et al., 2023). A plausible implication is that MESE functions as the TeV-scale continuation of the contained-event program rather than as a separate detector concept.
2. Veto logic, containment, and fiducialization
The defining technical problem of MESE is the rejection of cosmic-ray-induced atmospheric muons that enter the detector from outside and can mimic internal neutrino vertices after stochastic energy loss. To address this, the 2023 methods paper describes a multi-stage veto architecture. The outer veto region includes all outer strings, a top layer of thickness 90 m, a 120 m thick layer around the dust-dominated region close to the center of the detector, and a single layer of DOMs at the bottom (Basu et al., 2023).
The first stage is an outer-layer veto. For bright HESE-like events with deposited charge photoelectrons, up to 3 veto-layer hits are allowed. For dimmer events with charge photoelectrons, the event must have zero hits in the veto region. The paper states that this outer-layer veto rejects more than 99.99\% of bright entering muons (Basu et al., 2023).
A second stage targets lower-energy muons that evade the boundary veto and then produce a large stochastic loss in the interior. After the outer veto, the analysis considers hits associated with randomly sampled track hypotheses. If an event has more than 2 PE of incoming charge associated with such a track hypothesis, it is rejected. The 2025 spectrum proceedings describes the same logic more generically as a Downgoing Track Veto that searches for isolated DOM photon hits associated with multiple downgoing track hypotheses passing through the reconstructed vertex region (Basu et al., 2023, Basu et al., 8 Jul 2025).
A final stage imposes a charge-dependent and zenith-dependent fiducial-volume scaling. This cut is applied separately to track-classified and cascade-classified events, is more permissive for cascades than for tracks, and rejects more from the Southern sky because atmospheric muons are downgoing. In the 2025 spectrum paper this is summarized as a Fiducial Volume Cut whose effective fiducial volume shrinks for dimmer events and for directions with more severe muon contamination (Basu et al., 2023, Basu et al., 8 Jul 2025).
The aggregate effect is large. The 2023 methods paper states that after all cuts, the atmospheric muon background is suppressed by ten orders of magnitude relative to earlier levels in the selection (Basu et al., 2023). The 2025 flavor paper gives a shorter description, stating that the selection is based on a series of veto criteria to reject muon events and obtain a neutrino-rich dataset (V. et al., 9 Jul 2025).
3. Morphology, reconstruction, and topology-sensitive observables
MESE is an all-flavor sample, so it contains multiple event morphologies. The 2025 flavor analysis states that the sample consists of electron, muon, and tau neutrinos from the whole sky and that events are categorized, based on the light distribution in the detector, as cascades, tracks, and double cascades. The flavor-topology mapping used there is explicit: cascades arise from 0 charged-current interactions and neutral-current interactions of all flavors; tracks arise from 1 charged-current interactions and from 2 charged-current interactions in which the tau decays to a muon; double cascades arise from other 3 charged-current interactions (V. et al., 9 Jul 2025).
Initial track-versus-cascade classification in that analysis is performed during selection using a deep neural network. Above 4, the quoted performance is about 5 efficiency for true cascades and about 6 efficiency for true tracks (V. et al., 9 Jul 2025). The 2025 diffuse-spectrum paper also states that MESE uses a DNN classifier to separate events into track-like and cascade-like morphologies (Basu et al., 8 Jul 2025).
Because many true 7 events are initially absorbed into the single-cascade category, MESE flavor analyses add a dedicated double-cascade stage. The final-level MESE events are passed through a reconstruction that maximizes the likelihood under a double-cascade hypothesis using the spatial and timing information of the deposited charge in the DOMs. The reconstructed observables include the energy of each cascade, the sum of the two energies, the sum of the energies deposited within 8 of each cascade vertex compared to the total energy, the relative energy asymmetry between the two cascades, and the tau decay length obtained from the separation between the two cascades (V. et al., 9 Jul 2025).
The 2023 MESE methods paper defines two of these variables explicitly,
9
and notes that 0 only for ideal double cascades. Tau candidates are selected with 1, 2, each cascade energy at least 1 TeV, and final tau-candidate event energy 3 (Basu et al., 2023). The 2025 flavor analysis states that the final double-cascade classification requires reconstructed total energy 4 and reconstructed length 5, leading to an expected 7 events with 70\% purity in the double-cascade class (V. et al., 9 Jul 2025).
The sparsity of double-cascade events is a detector-resolution issue as much as a flux issue. The 2025 flavor analysis notes that most tau neutrinos in MESE are at TeV-scale energies, where tau decay lengths are only a few meters to a few tens of meters, while the string spacing in IceCube is about 125 m, making the two cascades hard to resolve (V. et al., 9 Jul 2025).
Topology enters the inference directly. For cascades and tracks, the flavor fit uses 2D histograms of reconstructed energy versus reconstructed 6, while for double cascades it uses 3D histograms of reconstructed energy versus reconstructed 7 versus reconstructed length (V. et al., 9 Jul 2025).
4. Background components, atmospheric self-veto, and statistical framework
MESE analyses fit several physical components simultaneously. The 2023 methods paper lists four: astrophysical neutrinos, conventional atmospheric neutrinos, prompt atmospheric neutrinos, and atmospheric muons (Basu et al., 2023). The 2025 flavor paper uses the same set, and states that in the best-fit distributions shown there the prompt atmospheric flux fits to zero, so it is not visible in the figure, and that no atmospheric muons pass the double-cascade classification (V. et al., 9 Jul 2025).
A central ingredient is the atmospheric neutrino self-veto. In the Southern sky, an atmospheric neutrino can be accompanied by a muon from the same air shower; if that muon reaches the detector and triggers the veto, the atmospheric neutrino is effectively suppressed. The 2023 MESE paper states that the 10-year analysis uses an improved understanding of atmospheric backgrounds via more accurate modeling of the detector self-veto. Its detector-specific procedure injects muons of varying energies into the final-level neutrino sample, estimates how many events are vetoed by the accompanying muon, and determines veto probabilities as a function of neutrino energy, zenith, and entry depth in the detector (Basu et al., 2023). This North/South asymmetry is prominent in the quoted expected rates per year: astrophysical 8, 9; atmospheric 0, 1; atmospheric 2, 3, with atmospheric neutrinos split very unevenly between North and South, 4 per year (Basu et al., 2023).
The nuisance model is correspondingly large. The 2023 paper includes normalization parameters for conventional atmospheric neutrinos, prompt atmospheric neutrinos, and atmospheric muons; atmospheric-neutrino production uncertainties using the Barr et al. parameterization; interpolation between the Gaisser-H4a and GST primary cosmic-ray models; detector systematics via SnowStorm for bulk-ice absorption, bulk-ice scattering, ice anisotropy, hole ice, and DOM optical efficiency; and a dedicated self-veto nuisance parameter (Basu et al., 2023). The 2025 diffuse-spectrum proceedings describes the fit more compactly as a forward-folding binned likelihood analysis implemented with NNMFit using Aesara, with model parameters reweighting simulation and significances quoted through Wilks’ theorem using
5
Its observables are reconstructed track energy, reconstructed cascade energy, and reconstructed zenith, shown as 6 (Basu et al., 8 Jul 2025).
5. Diffuse-spectrum measurements and the question of spectral structure
MESE was developed in part to clarify whether the diffuse astrophysical neutrino spectrum remains a single power law when the contained-event threshold is pushed down to the TeV scale. The 2023 methods paper framed this explicitly as a test of the earlier 7 excess seen in the 2-year MESE dataset, but it presented sensitivities rather than final 10-year fitted parameters (Basu et al., 2023).
The 2025 proceedings on the diffuse astrophysical spectrum gives the first explicit MESE-based spectral measurement over the full range above a TeV. Its baseline single power law fit finds a per-flavor normalization at 8
9
with
0
where
1
The best fit overall is a broken power law with
2
3
and
4
corresponding to a break near 5 (Basu et al., 8 Jul 2025).
The model comparison is central. For the broken power law relative to the single power law, the proceedings reports
6
It also reports a log-parabola fit with
7
and
8
The abstract emphasizes the latter figure as “strong evidence for structure in the spectrum beyond a single power law,” while the detailed comparison table shows that the broken power law performs even better (Basu et al., 8 Jul 2025).
The same proceedings identifies two features as driving the preference for curvature relative to a single power law: an excess around 9 and a deficit at a few hundred TeV. By contrast, the SPL + AGN and SPL + BL Lac template normalizations fit to zero, and in the best-fit broken-power-law result the prompt normalization is also a free parameter that fits to zero (Basu et al., 8 Jul 2025).
Earlier MESE work is relevant because it framed this spectral question as a low-energy excess problem. The 2016 dark-matter paper, using the 2-year MESE dataset with exposure 0 days, described an excess in the 10–100 TeV interval. It reported a maximum local significance of about 1 when using the MESE best-fit single-power-law index 2, increasing to 3 for 4 and 5 for 6. In a two-component astrophysical-plus-dark-matter interpretation, its likelihood-ratio analysis found a significance up to 7, explicitly local and model dependent (Chianese et al., 2016). The later 10-year MESE spectrum measurement is therefore best understood as a broader diffuse-spectrum analysis of the same energy region rather than as a narrow excess search.
6. Flavor composition and the all-flavor physics program
MESE’s all-flavor design makes it useful not only for spectral measurement but also for flavor inference. The 2025 three-flavor composition analysis uses 11.4 years of MESE data and states that the analysis sample contains 9888 events with interaction vertices inside the detector and energies of 8 and above. The event counts are
9
cascades above 0,
1
tracks above 2, and
3
double cascades above 4 (V. et al., 9 Jul 2025).
The astrophysical flavor fractions at Earth are parameterized by 5, 6, and 7, constrained through
8
Under the baseline broken power law flux assumption, the best-fit astrophysical flavor ratio at Earth is
9
The paper states that the 68\% CL contour closes for the first time in an IceCube flavor-ratio measurement; that each flavor fraction is constrained to be 0 with more than 1 CL; that zero electron flavor is rejected at 2; that zero tau flavor is rejected at 3; and that a neutron-decay-dominated source scenario is rejected at 4 (V. et al., 9 Jul 2025).
The baseline astrophysical spectrum for that flavor analysis is itself a MESE result from a companion diffuse-spectrum study: 5
6
7
As a cross-check, the paper also fits a single power law with
8
It states that the best fits and contour shapes are comparable under the two assumptions, but that the 95\% CL contour closes under the single-power-law assumption and does not close under the broken-power-law assumption. The explanation given is that the harder SPL index predicts more high-energy neutrinos and therefore more longer tau decay lengths and more identifiable double cascades, artificially tightening flavor constraints if the spectrum is modeled incorrectly (V. et al., 9 Jul 2025).
The physical interpretation is conventional source phenomenology after standard oscillations. The source scenarios explicitly discussed are pion decay,
9
muon-damped pion decay,
0
and neutron decay,
1
The paper states that the MESE best fit lies on the line connecting these scenarios and is closest to the standard pion-decay scenario, while the muon-damped scenario remains within the 68\% CL contour (V. et al., 9 Jul 2025).
7. Relation to HESE, starting-track variants, and broader uses
MESE belongs to the broader IceCube family of starting-event analyses. Relative to HESE, it lowers the contained-event threshold from about 60 TeV to about 1 TeV, preserving the same containment-and-veto logic but requiring tighter control of atmospheric backgrounds and self-veto effects (Basu et al., 2023, Basu et al., 8 Jul 2025). The 2025 HESE flavor proceedings makes the relationship explicit in outlook form, stating that “Additional sensitivity to the flavor composition is expected from cascades at lower energies, as shown by Medium Energy Starting Event Sample” (Lad et al., 9 Jul 2025).
A closely related but distinct branch is the starting-track program. The 2019 ESTES proceedings describes the Enhanced Starting Track Event Selection as similar to MESE above 2 but specialized to starting 3 tracks rather than a broad all-topology sample; it states directly that “ESTES is similar to MESE for the neutrino energies of interest, above 1 TeV, however MESE is designed to focus on cascade like events” (Silva et al., 2019). The later 10.3-year starting-track diffuse measurement is equally MESE-like in philosophy while remaining topology-specific: it uses veto-selected starting tracks, measures the astrophysical flux over a 90\% sensitive energy range of 3–500 TeV or 3–550 TeV, and finds no preference for a broken power law in that channel (Silva et al., 2023).
MESE has also served as a public contained-event dataset for problems beyond the baseline diffuse spectrum and flavor ratio. The 2020 “cosmic magnetometers” study used 2 years of MESE together with 6 years of HESE to search for synchrotron-cooling imprints in the diffuse neutrino spectrum and flavor composition. Because atmospheric contamination dominates at low energy, it restricted MESE events to 4, analyzed MESE and HESE separately, and found no evidence for the signal, deriving an upper limit of 5 kG–6 MG (95\% C.L.) on the average magnetic field strength of the sources (Bustamante et al., 2020).
Taken together, these studies establish MESE as IceCube’s principal all-flavor contained-event framework in the TeV-to-PeV regime. Its distinctive role is to connect veto-selected, morphology-aware event reconstruction with the energy range in which atmospheric and astrophysical components overlap most strongly. That combination makes it central to present IceCube work on diffuse-spectrum curvature, flavor composition, and medium-energy starting-event methodology (Basu et al., 8 Jul 2025, V. et al., 9 Jul 2025).
