Interactive · Pedagogical tool

How do we hunt for dark matter?

A working dark-matter experiment is a delicate balance — a lower energy threshold here, a quieter background there, a deeper lab, a longer run. Slide the controls and watch your sensitivity curve move against the world-leading limits from DarkSide-50, LZ, and the projected reach of DarkSide-20k.

This interactive page needs JavaScript. The sensitivity plots and slider-driven calculations all run live in your browser — turn JavaScript on (or open the page in a JS-enabled browser) to use it.

In a direct-detection experiment, the signal is fearfully small — a few keV of recoil energy in a tonne of liquid argon or xenon, deep underground, far from any disturbance. Every choice we make pulls the sensitivity in a different direction. Here you can play with those choices: drop the energy threshold, lengthen the run, swap argon for xenon, descend below a kilometre of rock, and watch the limit you'd set move against today's best results.

Swipe: Low mass (sub-GeV → 5 GeV) · High mass (1 GeV → 10 TeV)

What does this plot show?

Low-mass sensitivity. Spin-independent WIMP–nucleon cross-section that your experiment can exclude (90% C.L.) versus WIMP mass, from 40 MeV to 5 GeV. Below your "your experiment" curve is the cross-section space your detector can rule out; above it is still allowed.

The reference curves track the current leading-experiment envelope. CRESST-III (cryogenic calorimeters with sub-30 eV thresholds) leads below ~200 MeV/c²; DarkSide-50 S2+Migdal (argon, ionisation-only) leads from ~200 MeV to ~2.5 GeV; PandaX-4T (xenon, S2-only) takes over from ~2.5 to 5 GeV. DarkSide-20k S2 projections show the next generation. The shaded region beneath is the neutrino fog: cross-sections below it can only be reached at very high cost, since coherent ν scattering looks like a WIMP signal and you can't subtract it.

Target Argon (A=40)

Heavier nuclei get a stronger A² coherent enhancement for spin-independent scattering, but lose recoil energy to lighter WIMPs less efficiently.

“WIMP” is used here as shorthand. The slider tracks the sub-GeV regime where the canonical thermal-relic WIMP scenario no longer strictly applies — below ~1 GeV the page is best read as exploring a generic spin-independent contact-interaction dark-matter particle.

How this control works

Switching target changes everything: the A² coherent boost (xenon's A=131 vs argon's A=40), the Helm form-factor falloff at high recoil energy, the kinematic maximum recoil for a given WIMP mass, the realistic background environment (³⁹Ar, ⁸⁵Kr, ²²²Rn lines differ), and the neutrino-fog floor. The two reference curves on the plot follow each target's strengths: argon owns the low-mass S2+Migdal regime; xenon dominates the canonical high-mass WIMP region.

The internal calibrations (CAL_AR, CAL_XE) are anchored separately on the published DS-50 minimum and LZ minimum respectively — switching target uses the appropriate one.

S1 threshold 30 keV

Scintillation (S1) detection threshold in nuclear-recoil keV. The S1+S2 analysis needs both channels, so its effective threshold is the higher of S1 and S2. S1+S2 gets pulse-shape discrimination — much lower BG, but a higher floor in mass. DS-50 S1+S2 ≈ 30 keV_nr ‖ DS-20k ≈ 6 keV_nr (≈ 2 keV_ee) ‖ LZ ≈ 5 keV_nr

How this control works

The S1+S2 (scintillation + ionisation) coincidence analysis runs in parallel with an S2-only analysis. The S1+S2 channel's effective threshold is the larger of the two slider values — both signals must be present — and in return you get pulse-shape discrimination that drops backgrounds dramatically. That low-BG advantage is paid back at low mass: the higher threshold rules out sub-10-GeV WIMPs whose recoils are too soft to make an S1.

Pushing this slider down extends high-mass reach by integrating more of the rising signal tail; it also makes the model's recovered limit visibly track the LZ reference at high mass.

S2 threshold 0.5 keV

Ionization (S2) detection threshold in nuclear-recoil keV. The S2-only analysis runs in parallel — no PSD (~5× higher BG, narrower ROI), but reaches the lowest masses and picks up the Migdal channel below ~1.5 GeV. Best of the two analyses is plotted. DS-50 S2-only ≈ 0.6 keV ‖ DS-20k ≈ 0.3 keV

How this control works

S2 (ionisation-only) sees the secondary scintillation light from drifted ionisation electrons — orders of magnitude amplified, so the energy threshold can drop into the few-hundred-eV range. The price is losing the S1 coincidence and therefore pulse-shape discrimination: you face a much larger electron-recoil background, modelled here as ~5× the S1+S2 channel's BG.

S2-only is what opens up the sub-GeV WIMP regime, especially via the Migdal effect (a soft electron ejected during the nuclear recoil). The model adds a phenomenological Migdal contribution below ~1.5 GeV to track DS-50's published S2 + Migdal result qualitatively.

Exposure 0.05 t·yr

Detector mass × live time. Doubles the signal — and the background, so the win is only √2 once background-limited. DS-50 ≈ 0.05 t·yr ‖ LZ ≈ 4.2 t·yr ‖ DS-20k ≈ 200 t·yr

How this control works

In the background-free regime, sensitivity scales as 1/exposure: every extra tonne-year proportionally tightens the limit. Once you're background-limited it drops to 1/√exposure: doubling exposure doubles both signal and background, and the limit only improves by √2. The model captures this crossover via the Feldman–Cousins-like n₉₀ formula, smoothly interpolating between 2.3 (BG-free) and 1.282·√N_BG (BG-dominated).

An ν-induced background term is added so that, with very large exposure on a clean detector, the limit asymptotes properly into the ν-fog floor rather than improving without bound.

Intrinsic background 0.01 /(keV·t·yr)

Radio-impurities, surface contamination, and irreducible neutrino scattering. Below ~10⁻⁴ you are touching the neutrino fog.

How this control works

Every milli-becquerel of ²³⁸U, ²³²Th, ⁸⁵Kr or ³⁹Ar in or around the detector is a candidate fake-signal source. Cleaning is hard: underground-sourced argon, electroformed copper, screened steels, clean rooms cleaner than chip fabs. The S2-only channel sees ~5× the intrinsic BG because it lacks pulse-shape discrimination — the model handles this automatically.

Below ~10⁻⁴ /(keV·t·yr) the irreducible coherent-neutrino contribution dominates and the limit improvement plateaus regardless of how much cleaner you make the detector — this is the ν-fog floor in BG-rate form.

Depth 3800 m.w.e.

Overburden shields the detector from cosmic-ray muons. Below ~3000 m.w.e. the muon-induced background falls below intrinsic. Boulby ≈ 2805 ‖ Gran Sasso ≈ 3800 ‖ SNOLAB ≈ 6010 ‖ Jinping ≈ 6720

How this control works

Cosmic-ray muons spall neutrons out of surrounding rock, and those neutrons mimic a WIMP recoil exactly. The muon flux falls by roughly a factor of 10 per ~500 m.w.e. (the model uses this attenuation), so by Gran Sasso (≈3800 m.w.e.) muon-induced BG sits at the level of typical intrinsic; by SNOLAB / Jinping it's negligible.

Above ~3000 m.w.e. the depth slider stops mattering much — intrinsic and ν-induced backgrounds take over. Below ~1000 m.w.e. cosmogenic dominates everything and the limit gets dramatically worse.

Your experiment now

Loading…

What each control changes

A glance-table connecting the sliders to which part of the limit they move. Exposure and background interact via the BG-free → BG-limited crossover; thresholds set what mass range you can even reach.

Control	What it changes
Target (Ar / Xe)	Where the curve sits at every mass: A² boost, kinematic reach, form-factor falloff, background environment, ν-fog floor — all retargeted.
S1 threshold	The high-mass arm of the limit. Sets the effective threshold for the low-BG S1+S2 analysis; doesn't directly affect the sub-GeV S2-only regime.
S2 threshold	The low-mass end of the limit. Opens up sub-GeV WIMPs via the S2-only / Migdal channel. Best-of-two-analyses is plotted.
Exposure	Vertical scale at every mass. 1/exposure when BG-free; 1/√exposure once BG-limited. Hits the ν-fog floor eventually.
Background	Where the crossover happens. Cleaner detectors stay in the 1/exposure regime longer. Doesn't itself change kinematic reach.
Depth	Cosmogenic floor only. Mostly below ~3000 m.w.e.; above that, intrinsic and ν take over and depth stops mattering.

Jump to a configuration

What each control does

01Threshold & the S2-only trick

A WIMP of mass M_χ can only deposit E_R^max ≈ 2μ²v²/M_N in a collision. For light WIMPs that maximum sits below the detector threshold and the experiment goes blind. Dropping the threshold opens up sub-10 GeV WIMPs — and DarkSide-50 did exactly this by reading only the ionization (S2) signal, sacrificing the pulse-shape discrimination that suppresses electron-recoil backgrounds in exchange for an effective threshold down at ~0.6 keVnr. Push the threshold slider down and watch your line march toward the dotted S2-only reference.

02Exposure

Mass × time. In the background-free regime sensitivity scales as 1/exposure; with background it slows to 1/√exposure. This is why making detectors larger is only the easy half of the problem — the other half is keeping them clean enough to stay background-free for as long as possible.

03Background

Every milli-becquerel of ²³⁸U, ²³²Th, ⁸⁵Kr or ³⁹Ar in the detector is a fake signal candidate. We chase them down by underground-sourced argon, electroformed copper, screened steels, and clean rooms cleaner than chip fabs. Below ~10⁻⁴ /(keV·t·yr) we hit an irreducible floor of coherent neutrino scattering.

04Depth

Cosmic-ray muons spall neutrons out of surrounding rock, and those neutrons mimic a WIMP recoil exactly. The muon flux falls roughly exponentially with depth — by Gran Sasso (≈3800 m.w.e.) it is a million times below the surface, enough that muon-induced backgrounds drop below the intrinsic floor.

05Target

Coherent SI scattering enhances rates by A², so xenon (A=131) gains a factor ~10 over argon (A=40). But argon's lower mass gives more recoil energy per WIMP collision at fixed mass, and pulse-shape discrimination in liquid argon offers a powerful electron-recoil rejection lever xenon cannot match. Both technologies are needed: target complementarity breaks model degeneracies.

06The neutrino fog

No detector — however large, however quiet, however deep — can escape the ⁸B, atmospheric, and diffuse-supernova neutrinos that scatter coherently off nuclei and look exactly like WIMPs. The shaded region is where that fog begins to limit sensitivity gain. Beyond it, only directional detection or annual modulation can keep cutting.

Model & assumptions

Physics model used

Sensitivity calculations use the Standard Halo Model (ρ₀ = 0.3 GeV/cm³, v₀ = 220 km/s, v_esc = 544 km/s), Helm form factors, and a Feldman–Cousins-like 90% C.L. upper limit (smooth interpolation between background-free n₉₀ = 2.3 and BG-dominated n₉₀ = 1.282·√N_BG).

Two analysis channels run in parallel: a discriminated S1+S2 channel (effective threshold = max of the two slider values, low BG, wide ROI ≈ 200 keV) and an S2-only channel that loses pulse-shape discrimination (~5× higher BG, narrow ROI ≈ 5 keV) but reaches lower threshold and picks up a phenomenological Migdal contribution below ~1.5 GeV. The best limit of the two is plotted at every mass. Each channel's signal rate is integrated over its actual energy window — this is what makes the S1 slider visibly affect high-mass reach while leaving the low-mass S2-only regime untouched.

A coherent-ν background term is included so exposure asymptotes properly into the ν-fog floor rather than scaling linearly forever; cosmogenic BG attenuates by a factor of 10 per ~500 m.w.e. of depth.

Normalisation: semi-empirical, not first-principles

The shape of the limit (mass dependence, threshold cliff, A², exposure scaling) is computed from standard recoil kinematics. The absolute normalisation is two calibrated constants — CAL_AR tuned so DS-50's high-mass limit reproduces its published 1.1×10⁻⁴⁴ cm² minimum at 100 GeV, and CAL_XE tuned so the LZ preset recovers 2.2×10⁻⁴⁸ cm² at 40 GeV (arXiv:2410.17036, 4.2 t·yr).

This is therefore a semi-empirical sensitivity engine, not an independent absolute-rate calculation. With those two anchors in place, the model reproduces the DS-50 minimum to within ~5%, the LZ minimum to within ~5%, and the DS-20k TDR projection to within ~1.7× (the residual is profile-likelihood vs counting — see below).

Reference curves and sources

The DS-50 S1+S2 baseline is the original DarkSide-50 high-mass result; DS-50 S2+Migdal is from arXiv:2207.11967 (PRL 130, 101001, 12.3 t·d), approximated from the published figure.

LZ is the 4.2 t·yr WS2024+WS2022 combined limit from arXiv:2410.17036 (minimum 2.2×10⁻⁴⁸ cm² at 40 GeV/c²). LZ light-DM (dashed) extends this to 3–9 GeV/c² using the 5.7 t·yr WS2025 analysis in arXiv:2512.08065, which also reports the first 4.5σ evidence of ⁸B solar-neutrino CEνNS in xenon. DS-20k S1+S2 (nominal Fid 10 y, 200 t·yr fiducial) is read from the DarkSide-20k Technical Design Report (DARKSIDE-CSN2-TDR-2112, Figure 2 top panel, 90% C.L. exclusion). Minimum ≈ 1×10⁻⁴⁸ cm² at ~150 GeV.

CRESST-III is a coarse digitisation of the envelope of two cryogenic CRESST-III analyses: the SOS detector with 6.7 eV threshold (arXiv:2405.06527, PRD 110, 083038), which sets the leading SI limit in 74–202 MeV/c²; and Detector A, a 23.6 g CaWO₄ crystal with 30.1 eV threshold (arXiv:1904.00498, PRD 100, 102002), which reaches ~10⁻³⁸ cm² near 1 GeV and ~10⁻⁴² cm² at 10 GeV per the CRESST decade roadmap (arXiv:2505.01183). Pedagogical — not publication-grade.

PandaX-4T is the low-DM / ionisation-only S2 search from arXiv:2507.11930 (PRL 135, 211001), 259-day data, 1.04 t·yr effective US2 exposure — the leading SI constraint in the 2.5–5 GeV/c² window, with σ(3 GeV/c²) = 1.1×10⁻⁴³ cm². The paper notes mild upward fluctuations above ~3.5 GeV/c² from a slight event excess in the S2 (5–8 electron) region — reflected in the curve shape.

DS-20k S2+Migdal projections (1-year, and 10-year by 1/√t scaling) come from arXiv:2407.05813 Fig 4a (2N_e ultimate fit, data via Zenodo 13911875). The liquid-argon neutrino fog (n=2 boundary) is from O'Hare 2021 (PRL 127, 251802), read directly from the cajohare/NeutrinoFog Ar_SI data file. Below 100 MeV the fog values are extrapolated since O'Hare's calculation stops at the kinematic edge.

What this does not include

Published Migdal analyses use shape-likelihood methods that distinguish the Migdal electron spectrum from the ³⁹Ar background — this tool's counting analysis cannot reproduce that. At sub-GeV mass your S2 curve will be significantly more conservative than the published projection. The qualitative shape and dependences on threshold, exposure, depth and target are what's intended to be explored.

Spin-dependent operators, light mediators, momentum- or velocity-suppressed interactions are out of scope here — the model assumes spin-independent contact scattering. Annual modulation and directional detection (the natural next steps once ν-fog is in reach) are also outside scope.

Why real analyses differ from this counting model

Real published limits use profile-likelihood fits that exploit the predicted dR/dE versus a background model with its own spectral shape — that distinguishes signal from background per event, not just by counting. At the DS-20k preset the user curve sits ~1.7× above the TDR minimum because of this; honest residual.

If you ever want a closer match: the gap is a mass-dependent boost factor that could in principle be folded in — noted in the site TODO list as an optional future enhancement.