SUBMET: J-PARC Millicharged Particle Search

• SUBMET is a dedicated experimental program that searches for millicharged particles (χ) using proton fixed-target techniques at J-PARC.
• The experiment evolved from a four-layer to a two-layer coincidence design, improving sensitivity and reducing background through beam timing and optimized photoelectron yield.
• Key design features include precise mechanical integration, rigorous afterpulse modeling, and scalable detector modules that target charges below 10⁻³ e for sub-GeV masses.

The SUB-Millicharge ExperimenT (SUBMET) is a dedicated experimental program for the search for millicharged particles, $\chi$, at J-PARC, centered on the detection of extremely small ionization signals produced by particles with electric charge $Q=\epsilon e$ and sub-GeV to GeV-scale masses. In its J-PARC realization, SUBMET is a proton fixed-target experiment installed in the Neutrino Monitor building about $280$ m downstream of the graphite target, with detector concepts and projections developed for $N_{\rm POT}=10^{22}$ and an intended reach below $Q=10^{-3}e$ for $m_\chi<1.6~{\rm GeV}/c^2$ (Kim et al., 2021, Campagnari et al., 25 Jul 2025).

1. Development of the program

SUBMET has been documented through a sequence of design stages rather than a single immutable hardware configuration. The 2020 Letter of Intent proposed a four-layer plastic-scintillator detector at the B2 location of the Neutrino Monitor building, about $280$ m from the target and about $30$ m underground, with projected $95\%$ CL sensitivity down to $\epsilon \simeq 3\times10^{-4}$ for $Q=\epsilon e$0 and $Q=\epsilon e$1 for $Q=\epsilon e$2 at $Q=\epsilon e$3 (Choi et al., 2020). A 2021 feasibility study then reformulated SUBMET as a two-layer coincidence detector, explicitly using beam timing from the J-PARC bunch structure to suppress dark-count backgrounds, and quoted improved sensitivity to $Q=\epsilon e$4 for $Q=\epsilon e$5 and $Q=\epsilon e$6 for $Q=\epsilon e$7 (Kim et al., 2021). A later detector-design paper describes the mechanically integrated apparatus, reports that data taking began in June 2024, and states that the experiment is optimized for particles with $Q=\epsilon e$8 and $Q=\epsilon e$9 (Campagnari et al., 25 Jul 2025).

Stage	Detector concept	Quoted reach or status
2020 LOI (Choi et al., 2020)	Four layers of BC-408 bars	$280$0 below $280$1; $280$2 below $280$3
2021 feasibility study (Kim et al., 2021)	Two-layer stacked scintillator bars with beam timing	$280$4 below $280$5; $280$6 below $280$7
2025 mechanical integration note (Campagnari et al., 25 Jul 2025)	Built two-layer EJ-200 detector with 160 modules	Data taking began June 2024

The differing layer counts, bar dimensions, and sensitivity numbers are therefore features of successive published design stages. A common misunderstanding is to treat all quoted SUBMET configurations as simultaneous descriptions of one finalized detector. The publications instead document optimization from the LOI to the feasibility study and then to an installed instrument.

2. Production channels and detection principle

In the J-PARC fixed-target program, SUBMET searches for millicharged particles produced by a $280$8 GeV proton beam on graphite. The early sensitivity studies used neutral-meson and charmonium production as the baseline source term. The 2020 Letter of Intent treated $280$9, $N_{\rm POT}=10^{22}$0, and $N_{\rm POT}=10^{22}$1 channels, with yields parameterized as

$N_{\rm POT}=10^{22}$2

where $N_{\rm POT}=10^{22}$3 vanishes as $N_{\rm POT}=10^{22}$4 (Choi et al., 2020). The 2021 feasibility study retained the meson-dominated production picture and used PYTHIA8 inputs $N_{\rm POT}=10^{22}$5, $N_{\rm POT}=10^{22}$6, and $N_{\rm POT}=10^{22}$7 per POT (Kim et al., 2021). The later detector-design paper states the broader motivation in terms of proton-target production of $N_{\rm POT}=10^{22}$8, $N_{\rm POT}=10^{22}$9, $Q=10^{-3}e$0, $Q=10^{-3}e$1, $Q=10^{-3}e$2, and $Q=10^{-3}e$3, with the resulting $Q=10^{-3}e$4 flux forward-biased along the beam axis (Campagnari et al., 25 Jul 2025).

Detection relies on the fact that the ionization signal is quadratically suppressed by the fractional charge. The detector-design paper writes

$Q=10^{-3}e$5

which is the central reason SUBMET is optimized for single- to few-photoelectron response (Campagnari et al., 25 Jul 2025). In the 2021 feasibility study, a GEANT4 optical simulation for a through-going unit-charge MIP gave

$Q=10^{-3}e$6

per layer, and the two-layer detection probability was modeled as

$Q=10^{-3}e$7

In the sub-millicharge regime $Q=10^{-3}e$8, this implies $Q=10^{-3}e$9 and, because the accepted $m_\chi<1.6~{\rm GeV}/c^2$0 flux also scales as $m_\chi<1.6~{\rm GeV}/c^2$1, the signal yield scales as $m_\chi<1.6~{\rm GeV}/c^2$2 (Kim et al., 2021). This strong scaling is why SUBMET optimization studies consistently identify photoelectron yield as the dominant experimental lever.

3. Installed detector, geometry, and mechanical integration

The mechanically integrated SUBMET detector described in 2025 is a two-layer system of long plastic scintillator bars coupled to photomultiplier tubes, positioned $m_\chi<1.6~{\rm GeV}/c^2$3 m downstream from the target and about $m_\chi<1.6~{\rm GeV}/c^2$4 below the beam axis. The bars are tilted upward by $m_\chi<1.6~{\rm GeV}/c^2$5 relative to the floor to align with the expected $m_\chi<1.6~{\rm GeV}/c^2$6 arrival direction, and the coincidence signature is a left-to-right traversal through both layers within a narrow time window (Campagnari et al., 25 Jul 2025).

Each layer contains $m_\chi<1.6~{\rm GeV}/c^2$7 bars, for $m_\chi<1.6~{\rm GeV}/c^2$8 modules per layer and $m_\chi<1.6~{\rm GeV}/c^2$9 modules in total. The bars are EJ-200 plastic scintillators with dimensions $280$0, and the reported material properties are scintillation efficiency $280$1 photons/MeV for a $280$2 MeV electron, emission peak $280$3 nm, attenuation length $280$4 mm, rise time $280$5 ns, and decay time $280$6 ns (Campagnari et al., 25 Jul 2025). Optical wrapping consists of two layers of $280$7m Teflon, one layer of $280$8m aluminum foil, and one layer of $280$9m insulating tape, increasing the thickness by about $30$0 mm. One Hamamatsu R7725 PMT is attached to one end of each bar using optical grease, and an LED circuit is housed at the opposite end for testing and calibration (Campagnari et al., 25 Jul 2025).

The mechanical structure is hierarchical. Modules are grouped into $30$1 supermodules, supermodules are mounted into a cage holding twenty such units, and two aluminum-extrusion tables support the two detector layers. The cage with supermodules has dimensions $30$2 (Campagnari et al., 25 Jul 2025). Finite-element analyses quoted in the design paper give a maximum von Mises stress of about $30$3 MPa, a safety factor of about $30$4, static displacement not exceeding $30$5 mm for Table 1, and worst-case seismic displacements of about $30$6 mm for Table 1 and $30$7 mm for Table 2 (Campagnari et al., 25 Jul 2025). These figures are presented as evidence that bar-to-bar registration and PMT coupling remain stable under static and seismic loads.

Magnetic shielding is implemented with two layers of $30$8 mm mu-metal around the four PMTs in each supermodule. The reported attenuation of perpendicular external fields exceeds $30$9, PMT gain stability was confirmed up to $95\%$0 mT, and cosmic-muon pulse height was unaffected in that field range (Campagnari et al., 25 Jul 2025). Before shipment, all $95\%$1 modules were validated with $95\%$2 reconstructed cosmic-muon tracks; the system was then disassembled, transported, and reassembled at the J-PARC Neutrino Monitor building (Campagnari et al., 25 Jul 2025).

4. Timing logic, backgrounds, and afterpulse modeling

The dominant background identified in the 2021 feasibility study is random coincidence between two layers due to PMT dark counts. The J-PARC beam structure is therefore integral to the SUBMET strategy. The beam has spills at $95\%$3 s repetition, with eight bunches separated by $95\%$4 ns, and site timing signals are used to gate the trigger and suppress random coincidences by a factor of order $95\%$5 (Kim et al., 2021). In that study, a conservative total background budget of $95\%$6 over three years was adopted, and the main qualitative conclusion was that beam timing can render dark-count coincidences negligible at the analysis level (Kim et al., 2021).

The 2025 afterpulse study adds a more detailed treatment of PMT-intrinsic delayed backgrounds. It considers the installed detector as $95\%$7 EJ-200 plus R7725 modules and reports that afterpulses follow large PMT pulses, especially primaries with peak height greater than $95\%$8 mV. Because the J-PARC bunch spacing is about $95\%$9 ns, afterpulses generated by one bunch can fall into later bunch windows and mimic single-photoelectron signals (Campagnari et al., 22 Oct 2025). The waveform readout window is $\epsilon \simeq 3\times10^{-4}$0s, and dedicated June 2024 data were taken with the trigger delay shifted so that the $\epsilon \simeq 3\times10^{-4}$1s region after the last bunch was recorded. The analysis used $\epsilon \simeq 3\times10^{-4}$2k events, divided equally into parameter extraction and validation samples (Campagnari et al., 22 Oct 2025).

The modeling window is defined by

$\epsilon \simeq 3\times10^{-4}$3

excluding the first $\epsilon \simeq 3\times10^{-4}$4 ns because of strong baseline fluctuations immediately after a large pulse. The instantaneous afterpulse rate is modeled as

$\epsilon \simeq 3\times10^{-4}$5

where $\epsilon \simeq 3\times10^{-4}$6 is the primary pulse area, $\epsilon \simeq 3\times10^{-4}$7 is a per-module time constant extracted from the $\epsilon \simeq 3\times10^{-4}$8 distribution, and $\epsilon \simeq 3\times10^{-4}$9 is fixed by the measured afterpulse yield in $Q=\epsilon e$00 (Campagnari et al., 22 Oct 2025). Two functional forms for the $Q=\epsilon e$01-dependence were tested,

$Q=\epsilon e$02

and

$Q=\epsilon e$03

The resulting prediction for an arbitrary time interval $Q=\epsilon e$04 with $Q=\epsilon e$05 ns is obtained by integrating the exponential model over that window (Campagnari et al., 22 Oct 2025).

The paper reports that the model reproduces observed afterpulse rates with approximately $Q=\epsilon e$06 precision and that both the linear and exponential-in-area parameterizations perform comparably, with the linear model adopted for simplicity (Campagnari et al., 22 Oct 2025). Beam-off data showed dark current and radiation-induced rates of order $Q=\epsilon e$07 Hz, negligible compared with the order-$Q=\epsilon e$08 Hz afterpulse rates in the analysis windows. An operationally important result is that more than $Q=\epsilon e$09 of collision events contain at least one large pulse; the afterpulse model therefore allows SUBMET to retain a substantial fraction of beam-on data that would otherwise be vetoed (Campagnari et al., 22 Oct 2025).

5. Sensitivity, optimization, and experimental context

The quoted SUBMET reach depends on which design stage is considered. The 2020 LOI projected $Q=\epsilon e$10 CL sensitivity to $Q=\epsilon e$11 for $Q=\epsilon e$12 and $Q=\epsilon e$13 for $Q=\epsilon e$14 with $Q=\epsilon e$15 (Choi et al., 2020). The 2021 feasibility study, using a two-layer coincidence design and explicit beam timing, improved these numbers to $Q=\epsilon e$16 for $Q=\epsilon e$17 and $Q=\epsilon e$18 for $Q=\epsilon e$19 (Kim et al., 2021). The 2025 detector-design paper repeats the same associated-JHEP sensitivity figures for a three-year run with $Q=\epsilon e$20 and frames them as the design target of the installed apparatus (Campagnari et al., 25 Jul 2025).

Optimization studies in both early and later design stages converge on the same conclusion: photoelectron yield is the main driver. In the 2021 feasibility study, doubling $Q=\epsilon e$21 in acceptance improved $Q=\epsilon e$22 by less than $Q=\epsilon e$23, while doubling $Q=\epsilon e$24 improved $Q=\epsilon e$25 by about $Q=\epsilon e$26 (Kim et al., 2021). The 2020 LOI reached a similar conclusion with a different baseline detector: doubling $Q=\epsilon e$27 improved $Q=\epsilon e$28 by about $Q=\epsilon e$29, increasing the background by $Q=\epsilon e$30 worsened $Q=\epsilon e$31 by about $Q=\epsilon e$32, and increasing $Q=\epsilon e$33 by factors of $Q=\epsilon e$34 and $Q=\epsilon e$35 improved $Q=\epsilon e$36 by about $Q=\epsilon e$37 and $Q=\epsilon e$38, respectively (Choi et al., 2020). These comparisons are design-specific, but they all point to the same scaling logic in the sub-millicharge regime.

Within the broader experimental landscape, SUBMET occupies a proton fixed-target scintillation niche. The NA64 program at CERN is based on missing energy or missing momentum in $Q=\epsilon e$39 GeV electron and muon beams and projects $Q=\epsilon e$40 in muon mode and $Q=\epsilon e$41 in electron mode around $Q=\epsilon e$42 MeV (Gninenko et al., 2018). The SENSEI search in the NuMI beam used a low-threshold silicon detector in the MINOS cavern and set world-leading constraints for masses between $Q=\epsilon e$43 MeV and $Q=\epsilon e$44 MeV, improving previous bounds by as much as a factor of $Q=\epsilon e$45 in the $Q=\epsilon e$46–$Q=\epsilon e$47 MeV interval (Barak et al., 2023). A later Jefferson Lab Hall A proposal with a $Q=\epsilon e$48 Skipper-CCD array reports that such a configuration can exceed the sensitivity of all existing searches below $Q=\epsilon e$49 GeV and is competitive with SUBMET across the overlapping mass range (Essig et al., 2024). SUBMET is therefore best understood as one member of a wider mCP search ecosystem that includes proton fixed-target scintillation detectors, missing-energy experiments, and ultralow-threshold silicon beam-dump searches.

6. Broader uses of the SUBMET label

Later literature uses the SUBMET label in a broader conceptual sense than the specific J-PARC scintillator apparatus. One strand is the terrestrial-accumulation program for Earth-bound millicharged particles. In that setting, underground electrostatic accelerators are proposed as devices that directly accelerate ambient thermalized mCPs by

$Q=\epsilon e$50

with MV-scale facilities such as LUNA, CASPAR, and JUNA cited as examples (Pospelov et al., 2020). For a model beamline with $Q=\epsilon e$51 m, radius $Q=\epsilon e$52 mm, and depth of about $Q=\epsilon e$53 km, the flux of accelerated particles above threshold is written

$Q=\epsilon e$54

and the study states that, for top-down accumulation with $Q=\epsilon e$55 GeV, detectable rates can probe $Q=\epsilon e$56 down to about $Q=\epsilon e$57 with realistic backgrounds of order $Q=\epsilon e$58 Hz (Pospelov et al., 2020). This is not the same configuration as the J-PARC detector, but the acronym is explicitly used there for an underground accelerator-based search strategy.

A second strand appears in the reinterpretation of Cavendish tests of Coulomb’s law as both accumulators and detectors for room-temperature mCP populations. That proposal specifies an outer accumulator shell with

$Q=\epsilon e$59

and an inner Cavendish shell with

$Q=\epsilon e$60

together with a solid inner sphere of radius $Q=\epsilon e$61 m and low vacuum $Q=\epsilon e$62 atm (Berlin et al., 29 Oct 2025). The signal is an oscillating interior potential difference,

$Q=\epsilon e$63

and the paper argues that an accumulator-enhanced Cavendish configuration could reach ambient densities as low as $Q=\epsilon e$64 and may detect the irreducible cosmic-ray mCP population for sub-GeV masses (Berlin et al., 29 Oct 2025).

These later uses of the acronym show that “SUBMET” can denote either the specific J-PARC scintillator experiment or a broader family of terrestrial mCP search concepts. In the experimental literature, however, the unqualified name most commonly refers to the J-PARC program documented in the Letter of Intent, the feasibility study, the mechanical integration note, and the afterpulse-background analysis (Choi et al., 2020, Kim et al., 2021, Campagnari et al., 25 Jul 2025, Campagnari et al., 22 Oct 2025).