CASPEr@night

A search for dark matter in the noise of the world's idle NMR spectrometers.

The model is SETI@home — a search that runs on instruments the world already owns, during the hours nobody else wants, with the analysis centralized so that participating costs a facility almost nothing.

How it works

Every NMR receiver's noise floor conceals a measurement — of the spins, of the probe itself, and of any dark-matter field that couples to nuclear spins. Contributing takes one short session on an idle instrument.

Fill a tube with water. A standard 5 mm tube of plain water (~550 μL). Tap or distilled is fine — the software asks exactly what it is. A 90/10 H₂O/D₂O mix is better still: the field-stepped sweep steers the field through the deuterium lock, so a lockable sample covers a band of adjacent masses in one night instead of a single point.
Run one command. Type xpy spin_noise_run in TopSpin, answer six dialogs, and walk away. The default run is about 45 minutes wall clock — you are needed for roughly the first five. The field-stepped sweep is the same script with one extra word (xpy spin_noise_run sweep). Nothing pulses your sample at high power, and the script never changes instrument state silently.
Upload one zip. The run bundles data, calibrations, and metadata into a single file. You upload it with one command and keep the local copy — a failed upload never loses data. Consoles on isolated networks are fine: carry the zip out on a USB stick and upload from any machine.

What your facility gets back

A per-facility sensitivity report — your probe's measured spin-noise feature and its calibrated distance from the fundamental sensitivity ceiling of inductive detection. This is the number no vendor datasheet provides, and the number that settles whether a cryoprobe upgrade, a new preamplifier, or nothing at all is what your sensitivity budget actually wants.
Co-authorship on the network publications your data enter — both the fleet-scale spin-noise metrology papers and the dark-matter exclusion papers.
Participation in a fundamental-physics first — a coordinated, multi-site nuclear-spin search in an axion mass band no nuclear-spin experiment has reached, where each distinct magnet is a new mass point.

Short runs count

We know the queue is full — the best instruments book out weeks ahead, and not every facility has idle time. The physics is forgiving: sensitivity grows as the fourth root of measurement time, so a 15-minute wedge between queued experiments already delivers 40% of the reach of a full 10-hour overnight run, and a single hour delivers 56%. A full night is the ideal — a gap between runs is enough to participate.

The 2020 pilot result

In May 2020, seven 101-second receiver-open noise records were taken on a 600 MHz cryoprobe spectrometer to settle a disagreement about the fundamental sensitivity of inductively detected NMR. Our re-analysis found the ¹H spin-noise line at 2.30 ± 0.04 times the system noise floor — a 94σ combined detection. Reinterpreted as a resonant search for the axion wind, the same 12 minutes of noise yield a worst-case 90%-confidence exclusion on the axion–proton gradient coupling, g_ap < 5.1 × 10⁻⁵ GeV⁻¹ at m_a = 2.48 μeV — the first laboratory dark-matter constraint on this coupling at this mass, a 15× mass extension over the previous highest-mass nuclear-spin search. We state this plainly: the data are archival and opportunistic, and the limit is a deliberately worst-case construction — the full calibration envelope is taken against us and the entire observed line power is attributed to signal. Unpolarized thermal runs will not beat astrophysical bounds. The honest value is the unexplored mass coverage, the fleet-scale sensitivity metrology, and the multi-site coincidence infrastructure — with a hyperpolarization path toward far deeper reach.

Mass coverage

Each spectrometer's field strength corresponds to one axion mass point. The world's fleet of 300–1200 MHz instruments spans 1.24–4.96 μeV — coverage grows with every distinct magnet that joins, which is the respect in which your 750 is not redundant with someone else's 800. The same arithmetic means the network has no minimum size of institution: a 400 at a primarily-undergraduate department covers mass points a national lab literally cannot, with the same co-authorship.

One magnet, one mass point: m_a[μeV] ≈ 4.14 × 10⁻³ × ν_L[MHz]. Every point shown sits above the previous highest-mass nuclear-spin search at 0.166 μeV.

Where the network is

Each dot is a participating facility — shown only with the facility's consent, and only at city level. The map grows as the network does.

4 facilities · 5 distinct fields · updated 2026-09-21

Where the project stands

Status — September 2026

The network has its first complete Bruker sessions and its second per-site sensitivity figure. On 14–16 September three sessions were acquired on a room-temperature 400 MHz Agilent instrument at Southern Illinois University Carbondale and passed the pipeline end to end: the water spin-noise feature stands 67% above the noise floor after 30 minutes (89σ), a bracketed gain ladder measured the receiver's compression, and the three sessions combine into a deliberately worst-case exclusion on the axion–proton coupling at 1.65 μeV — unpublished and far above astrophysical bounds, but the first laboratory nuclear-spin number at that mass. On 25 September the first live Bruker session, on a 400 MHz Avance Neo at the University of Torino, resolved the water spin-noise feature as a dip of 41% of the noise floor. On 5–6 October the University of Oulu's 500 MHz Avance III HD ran the first complete Bruker sessions — 30 minutes, 3 hours and 9.5 hours overnight, with the reference blocks that turn a spin-noise spectrum into a sensitivity figure — and all three show a 29% dip, 25–28 Hz wide, giving a second, nearly three-times-tighter exclusion at 2.07 μeV, stable across the three sessions as the construction predicts. Getting there took five console faults found by the Torino and Oulu operators in two weeks, each fixed within a day, and two analysis defects that only real Bruker reference blocks could reveal. Version 0.8.0 of the software is released: the facility report now takes the dark-matter wind at its actual direction for the site and the time of each row rather than assuming it along the field, which moves the Oulu figure from 8.3 to 6.5 × 10⁻⁵ GeV⁻¹ and the Carbondale figure from 2.5 to 1.8 × 10⁻⁴, and it fits the per-row line powers against the known sidereal swing of the signal — on Oulu's overnight run that bounds a modulated excess five times tighter than the plain construction on the same hours, consistent with zero. Version 0.8.1 (16 October) is released after the first live tests of the optional modes at Oulu: the probe-tuning scan found a setting 250 kHz from the pulse optimum at which the spin-noise dip deepens from 29% to 36% of the floor, the field-stepped sweep's dialogs no longer block the console's lock controls, acquisition commands bypass TopSpin's command spooler where a facility has it on, and the analysis now ignores a reference block that carries no line — the fix that recovered Torino's 1,791-row overnight run, in which the dip is present in every sampled row and the per-row fit against the sidereal swing bounds a modulated excess nearly eight times tighter than the plain construction. Version 0.8.2 (19 October) is released: every session now keeps its pulse calibration as data, a ten-step nutation curve at the calibrated 90° power from which the report fits the 90° pulse and checks the small-flip amplitude the transduction constant rests on, and measures its receiver-gain curve at two levels per octave, each visited twice, so that compression and drift separate and the gain bridge is measured rather than assumed; the headline construction is unchanged, and neither block has yet run on hardware. The script brings its own parameter files and pulse programs, acquires in digital mode, probes one row before committing a night, reads every setting back, and records what it did; the report reads TopSpin 3 and 4 bundles alike. Agilent/Varian data are read natively, every facility report carries a per-site axion-coupling exclusion that tightens as a site adds sessions, and the analysis streams long runs. The v0.6 additions — spectral tiling across ±25 ppm of field, the fluorine-19 channel, the persistent line catalog — remain.

Four facilities — in the United States, Finland, Italy, and Germany — have registered, spanning five distinct fields and therefore five distinct axion-mass bands; more are in conversation. The Agilent path has now run on a real instrument; the Bruker/TopSpin acquisition script — the primary target, since most of the fleet runs TopSpin — has passed its desk tests on a real console (TopSpin 4.4.0, Torino) and its first live sessions begin this week; until one has run end to end we call the Bruker side a good start rather than a working system. JEOL and Magritek Spinsolve benchtop paths are in development in the public repository.

Join the network

One tube of water, one command, one upload — and your spectrometer spends a night doing physics it was never asked to do.

Not ready to commit a spectrometer? Join the low-volume announcements list. Or write directly: [email protected] — for the software bundle, a pilot-run slot, or questions about the physics.