Text from the interactive explanation
A thorium-229 optical nuclear clock with feedback loop
L. Toscani De Col, T. Riebner, I. Morawetz, F. Schneider, N. Sempelmann, J. Schlachet-Lépinay, F. Schaden, M. Bartokos, G. A. Kazakov, K. Beeks, B. Gerstenecker, M. Pimon, S. Lahs, A. Hellerschmied, T. Lercher, H. Denker, J. Premper, A. Niessner, M. Matus, M. Čížek, O. Číp, V. Lal, G. Zitzer, V. Petrov, J. Tiedau, M. V. Okhapkin, E. Peik & T. Schumm
Original paperThorium-229 nuclei inside a millimetre-sized crystal now steer the very laser that probes them. Closing that feedback loop turned them into the first stand-alone optical nuclear clock, which ran for a full day.
What you are watching. A simulated run of the clock loop, using the paper's numbers. The detector counts photons at two frequencies either side of the nuclear line, and their imbalance tells the loop which way to nudge the laser. Drag the amber laser off the line, or knock it, and watch the nuclei pull it back.
An 8.4-electronvolt coincidence
Atomic clocks tick with electrons. Nuclear excited states usually sit thousands to millions of times higher in energy, out of a laser's reach. Thorium-229 is the exception: its nucleus has an excited "isomer" state only 8.4 electronvolts up, at 148 nm in the vacuum ultraviolet.
That low energy is a near-cancellation. The electromagnetic (Coulomb) and strong-nuclear contributions to the energies of the two nuclear states are each at the mega-electronvolt scale, and they almost exactly cancel. If the strength of electromagnetism, set by the fine-structure constant α, or the strength of the strong force changed even slightly, the small remainder would change by a much larger fraction. The nuclear transition works like a magnifier for changes in nature's constants.
Nuclei also couple only weakly to stray fields, so many of them can sit in a room-temperature crystal and still form a clean reference.
Nudge α
One clock cycle, by hand
Now you are the feedback loop. The curve is the measured 5/2 → 3/2 line of thorium-229 in calcium fluoride: a Lorentzian 112 kHz wide at 148 nm that absorbs 0.75% of the light at its centre. Place the laser, count photons, then read the correction off the error signal.
The detector's verdict
Try the probe spacing: the uncertainty is lowest at Δf = FWHM/√3 ≈ 65 kHz, where the Lorentzian is steepest. That is the spacing the paper uses.
Photon budget (paper). 65 pW of 148-nm light reaches a detector with 10% efficiency: about 4.9 million counts per second. With 0.75% absorption that gives a signal-to-noise ratio of about 17 after one second, and a centre estimate good to about 1.3 kHz after 20 s.
If the laser starts more than about 43 kHz from the centre, beyond the peaks of the error signal, one step cannot bring it all the way back; the loop needs a few cycles. Far out on the wings, the imbalance drowns in shot noise.
How long should the clock listen?
Count briefly and the random arrival of photons (shot noise) makes every correction jittery. Count for a long time and the laser's reference cavity wanders between corrections. The paper ran both: T = 20 s and T = 30 min.
Allan deviation σy(τ) is the standard yardstick of clock stability: the typical fractional change in frequency between consecutive averages of length τ. Lower is better.
Simulated The amber curve is a two-day Monte Carlo run of a teaching model of the loop, using the paper's photon numbers and line shape plus a cavity random walk tuned to the measured data. It falls on the measured red curve at T = 20 s and within about a factor of two of the blue curve at T = 30 min. Measured curves: Fig. 2 data.
For τ longer than T, both measured curves follow the same shot-noise line, 3 × 10⁻¹² / √(τ/s): from then on, the nucleus sets the stability. With T = 30 min, the stability over one second is that of the cavity, about 10⁻¹⁵. With T = 20 s, the laser follows the nucleus quickly, which the dark-matter search below needs. The clock ran for a day without intervention and reached the low 10⁻¹⁴ range within a run. It was also unaffected by the crystal's ≈0.1-K temperature drifts (the line moves by 1.8 kHz per kelvin).
Where the beam crosses the crystal
From day to day, the clock agreed with itself only to about 5 × 10⁻¹³. Each run, the realigned 0.5-mm beam crossed a slightly different part of the 3.1-mm crystal. So the team moved the crystal on purpose.
Scan 1 · centre
Measured Fig. 3: line centres from T = 20 s clock operation over 20 min at each position, relative to the mean of all five.
The line centres differ by up to 1.7 kHz, far more than the 0.13 kHz expected from counting statistics, yet returning to the same spot (scans 2 and 4) gives the same value. The authors conjecture that local strain in the crystal shifts the nuclear line. A reproducible beam path, more uniform doping, crystals grown with less strain, or hosts such as ThF₄, where thorium is part of the lattice rather than a dopant, are expected to help.
Listening for dark matter
Some theories propose that dark matter is an ultralight field that oscillates everywhere at a frequency set by its particle's mass. If it couples to photons, gluons or quarks, it would make α, the strong-force scale or quark masses oscillate, and the thorium frequency with them. Hide such a wave in the clock's real 23-hour record and see whether the paper's analysis would catch it.
Measured record, amplitudes, threshold and confidence level from the paper's data. Hypothetical The injected wave is yours: its amplitude is A = d · k · 2.40 × 10⁻³¹ eV / mφ, the paper's relation, for a field carrying the local dark-matter density.
The team searched the T = 20 s run (about 23 hours) with a Lomb–Scargle periodogram, a spectrum suited to unevenly spaced data. One thousand Monte Carlo simulations of the clock loop set a detection threshold: a peak above it would have less than a 5% chance of being a noise fluke. No peak crossed it. A 95% confidence level then turns that silence into upper limits on the couplings.
For coupling to photons, the limits match the best comparable atomic-clock comparisons. For the strong force and quark masses, which atomic clocks barely feel, the thorium clock reaches 10 to 1,000 times deeper into the parameter space than earlier atomic-clock experiments. Because the absorption loop corrects every 20 seconds, it also reaches about ten times further in frequency than an earlier analysis of the thorium line over 10 months, probing masses that before could only be constrained from the line's shape. The frequency window runs from 1/(total time) to 1/T: longer runs reach lighter bosons, faster cycles heavier ones.
A slow-drift search found a slope of (2 ± 4) × 10⁻¹⁴ per day in the thorium/ytterbium ratio, consistent with zero.
From 10−12 towards 10−16
For τ beyond T, the clock is limited by shot noise: σy(τ) ≈ (Γ/f₀) · (1/SNR) · (τ/s)−1/2. Three levers remain: more laser power, more absorption, and a narrower line Γ.
Projection Calculated with the paper's shot-noise scaling; the absorption cross-section grows as the line narrows. The paper projects ≈10⁻¹⁵/√τ for 1 µW and a 10× longer path, and ≈10⁻¹⁶/√τ for a 1-kHz line with 100 pW. Neither has been demonstrated yet.
More power could come from an enhancement cavity around the frequency-doubling crystal (about ten times), commercial 3-W ultraviolet lasers (hundreds of nanowatts), four-wave mixing in cadmium vapour (300 nW already reported) or periodically poled crystals (at least 1 µW expected). Even then, absorption stays far from saturation: about 3 × 10⁸ photons are absorbed in 10 minutes, while some 10¹⁶ nuclei sit in the beam.
A narrower line needs a better host. In ThF₄, thorium is part of the crystal rather than a dopant, which should reduce broadening; spinless solids could remove the 0.1–1 kHz broadening from neighbouring fluorine nuclei. At ≈10⁻¹⁶/√τ, a nuclear clock that keeps the simplicity and small form factor of a solid would match today's best optical atomic clocks and be four orders of magnitude more sensitive to changes in fundamental constants than the clock in this paper.
The paper
L. Toscani De Col, T. Riebner, I. Morawetz, F. Schneider, N. Sempelmann, J. Schlachet-Lépinay, F. Schaden, M. Bartokos, G. A. Kazakov, K. Beeks, B. Gerstenecker, M. Pimon, S. Lahs, A. Hellerschmied, T. Lercher, H. Denker, J. Premper, A. Niessner, M. Matus, M. Čížek, O. Číp, V. Lal, G. Zitzer, V. Petrov, J. Tiedau, M. V. Okhapkin, E. Peik & T. Schumm. A thorium-229 optical nuclear clock with feedback loop. Nature 658, 897–902 (2026). doi:10.1038/s41586-026-11084-4 · Open access, CC BY 4.0.
Data. Measured beat-frequency records, Allan deviations, Lomb–Scargle amplitudes, Monte Carlo detection threshold and 95% confidence level: S. Lahs, Zenodo, doi:10.5281/zenodo.22657608 (CC BY 4.0). Re-encoded and re-plotted here; the exclusion curves are recomputed from the confidence level with the paper's relation and match the published bands.
Figures. Fig. 1 and Fig. 4 reproduced from the article under CC BY 4.0, resized. Line-centre values in section 4 are from Fig. 3.
Models. The clock-loop scene, the single-cycle instrument and the Allan simulation are teaching simulations using the paper's line shape, photon numbers and loop timing, plus a cavity noise model tuned to the measured curves. The projections in section 6 use the paper's shot-noise formula. The dark-matter panel adds hypothetical waves to the measured record; the schematic in section 1 uses toy numbers.
Topics in this iPaper
Physics
Here we realize a thorium-229 nuclear clock by stabilizing a continuous-wave laser to the 148-nm nuclear transition with rapid feedback based on absorption spectroscopy.
Abstract, page 1 (Nature 658, 897)
nuclear clock
Our system is a nuclear clock that operates as a stand-alone device.
Main text, page 2 (Nature 658, 898), final paragraph before 'Experimental set-up'
frequency metrology
The standard metric for characterizing the stability of a clock is the Allan deviation
Clock performance, page 3 (Nature 658, 899)
absorption spectroscopy
has enabled the measurement of the nuclear transition with absorption spectroscopy
Main text, page 2 (Nature 658, 898)
ultralight dark matter
We use the nuclear clock to constrain models of ultralight dark matter by searching for periodic fluctuations and slow drifts in the nuclear transition energy
Abstract, page 1 (Nature 658, 897)
variation of fundamental constants
The variation of α(t) relates to the variation of the ratio of the 229Th and ytterbium clock frequencies
Methods, 'Variation of the fundamental constants', page 7