Text from the interactive explanation
A nuclear clock synchronized to 229Th
Beichen Huang, Gaowei Yan, Qi Xiao, Wenhao Bu, Chengchun Zhao, Zhen Zhang, Chao Yan, Zhi-Ang Chen, Peixiong Zhang, Gleb Penyazkov, Zhenhai Zhan, Lingfeng Yan, Yuefei Wang, Lin Li, Shanming Li, Dapeng Jiang, Xiaobo Qian, Xuegang Liu, Qiange He, Taoxiang Sun, Haochen Tian, Bingkun Lu, Ningyuan Ma, Juxian Li, Yanzhang Wu, Qiaorui Gong, Yuxiang Li, Haoyu Shi, Xiangliang Li, Longsheng Ma, Shining Zhu, Yuxiang Mo, Jun Lin, Li You, Yige Lin, Xibo Zhang, Yin Hang, Liangbi Su & Shiqian Ding
Original paperNature · 2026 · Huang et al.
The first thorium-229 nuclear clock steers a 148.4-nm ultraviolet laser onto a resonance of the thorium nucleus itself, checking and correcting it about every 12 seconds, forever. Here you run that loop.
Or drag the needle in the scene. With feedback on, the loop pulls it back.
Demo: the laser is kicked every few seconds until you take over.
Illustrative crystal, beam and nuclei (glow exaggerated: in reality only ~4 × 10−3 of the nuclei at the entrance face are excited, and on resonance the crystal absorbs only ~3 × 10−4 of the light). Model line shape and error signal from the paper’s equations; Measured dots: Fig. 1e (δ = 12 kHz). The 0.888-kHz wiggle is slowed for display and 1 clock cycle ≈ 0.6 s on screen (11.8 s in the lab).
Simulated Each dot = one clock cycle. Zoomed 70× relative to the scope; kicks shoot off-scale (magenta markers). Per-cycle noise is calibrated to the measured instability of the optimized clock, 5 × 10−13/√(τ/s) at an 11.8-s cycle (≈ 0.29 kHz per cycle).
01The discriminator
How the nucleus says “too red” or “too blue”
On resonance, the thorium in the crystal absorbs only a few parts in ten thousand of the light. A bare dip that small cannot tell you which side of the line you are on. So the team wiggles the laser frequency back and forth and lets a lock-in amplifier listen for the part of the transmitted light that flickers at the wiggle rate.
Line: first-harmonic response of a 27-kHz Lorentzian (Methods, eq. 9) at the current δ. Dots: the measured optimized discriminator (Fig. 1e, TS1 + phototube, δ = 12 kHz, 10 s per point, error bars 1σ). Drag across the plot to move the laser.
Too red (below the line): swinging up moves the laser toward the nucleus, so more light is absorbed: the transmitted light dips when the frequency rises. The lock-in output is negative.
Too blue (above the line): the light brightens when the frequency rises, flickering in step with the wiggle: positive output.
Dead centre : both swings move away from the peak equally, so the light dips twice per wiggle, at twice the rate, and the first-harmonic output is zero. That zero crossing is the clock’s lock point; the slope beside it converts a signal into a detuning.
The slope of the error signal at the centre sets how precisely one noisy measurement pins down the laser. For a Lorentzian line of width Δ = 27 kHz it peaks at a wiggle amplitude of δ = Δ/(2√2) ≈ 9.5 kHz.
The team chose δ = 12 kHz : it keeps 96.4 % of the maximum slope while widening the monotonic range between the two lobes by about 13 % , so the clock tolerates larger frequency excursions during feedback.
Try it in the clock above: at δ = 40 kHz the slope drops to a quarter, and because the detector noise stays the same, every cycle’s estimate of the laser’s position gets about four times noisier. You can see it in the frequency record.
02The result
Upgrade the clock, watch it settle
The same loop ran in four configurations. Each step gave the nuclear signal more detected light or more absorbing nuclei, and the clock’s fractional instability fell from 1 × 10−11/√(τ/s) to 5 × 10−13/√(τ/s), averaging down to the 10−15 level.
Line-b error signal, S1 + PMT (Fig. 2c): 80 s per point, error bars 1σ.
In the shot-noise limit the 1-s signal-to-noise ratio scales as SNR ∝ A·√Ṅe: fractional absorption A times the square root of the detected photoelectron rate.
Measured points with 68.3 % confidence bars and dashed τ−1/2 fits (Fig. 3). Simulated amber: a run of the clock loop with noise calibrated to the measured instability, time-accelerated (≈ 1 simulated hour per 25 s); it is not the measured record.
03Spectroscopy
First, find the nucleus
Before locking, the line has to be found. Sweep the laser across 420 MHz of the vacuum-ultraviolet spectrum near 2,020.4 THz and watch the thorium in TS1 answer with fluorescence. The crystal’s electric-field gradient splits the nuclear transition into five narrow lines plus a broad unresolved structure .
Line b (mg = ±5/2 → mis = ±3/2) is the clock line: narrow (27 kHz) and only weakly temperature-sensitive. Line c is ±1/2 → ±1/2 (39 kHz). Lines a, d and e are the other quadrupole-resolved components; QU is a broad quadrupole-unresolved structure about 0.9 MHz wide.
After the laser excites the nuclei, they keep glowing: the 148-nm fluorescence decays with a single-exponential lifetime of 631(2) s, about 10.5 minutes , measured in TS1 after excitation on QU.
That long memory shapes the measurement. For each fluorescence point the crystal is illuminated for 400 s, then the photomultiplier counts for 200 s, and leftover glow from the previous point is subtracted. The faster readout used for the clock does not wait for the glow at all: it watches the laser light that makes it through the crystal.
04Reproducibility
Two crystals, one frequency
A clock reference is only useful if another copy ticks at the same rate. The team grew two crystals by different methods in different labs and measured line b in each, against a hydrogen-maser-referenced frequency chain.
S1 Bridgman–Stockbarger · SIOM
Photo Fig. 2g, on 5-mm grid paper.
TS1 temperature gradient · SICCAS + Tsinghua
Photo Fig. 2h. Crystallized by slowly lowering the heater temperature (2 °C h−1, < 10−3 Pa) with the crucible kept still; the scarce isotope left no room for repeat attempts.
Span shown: 10 THz
Thick bars: statistical 1σ; thin whiskers: combined with the 0.185-kHz systematic uncertainty from the crystal-temperature calibration (for JILA, the combined uncertainty). Uncertain digits blur and re-roll as you zoom.
Line b in TS1 sits 5.6(13) × 102 Hz above line b in S1, a fractional difference of 2.8(6) × 10−13 : the two independently grown crystals agree at the 10−13 level.
Against a value inferred from earlier JILA measurements on other crystals (a quadratic temperature fit evaluated at 301.4 K), the offsets are −0.50(122) kHz for S1 and +0.06(122) kHz for TS1, both consistent with zero.
Why it holds still. The crystal is held at 301.4(1) K; its 3-mK rms temperature wobble shifts line b by only about 6 Hz (3 × 10−15). Multilayer μ-metal shielding, a bifilar heater winding and twisted leads leave about 1 mG of residual field, an estimated Zeeman contribution below 1 Hz. Both are below the present instability.
05The laser
Making 148.4-nm light, continuously
No laser crystal emits at 148.4 nm. The team builds the light by four-wave mixing in hot cadmium vapour : two ultraviolet photons and one red photon combine into a single vacuum-ultraviolet photon.
The 375-nm field is tuned to the cadmium 51S0 → 61S0 two-photon resonance , which enhances the nonlinear response; a 710-nm photon completes the sum.
Oven at 600 °C ; 5 W at 375 nm (frequency-doubled 750-nm fibre laser in a bow-tie cavity) plus 10 W at 710 nm give 10 μW of continuous 148.4-nm light.
A Brewster-angle MgF2 prism separates the VUV; about 5 μW reaches the crystal in a beam about 1 mm across. Its linewidth is estimated to be sub-hertz , far narrower than the nuclear lines.
The frequency chain
1,397-nm laser on a cryogenic silicon cavity, stable at the 10−17 level.
Carried over a 4-km phase-stabilized fibre to stabilize an Er-fibre frequency comb ; its offset is referenced to a hydrogen maser , which also counts the comb for traceability.
The 750-nm and 710-nm lasers are phase-locked to comb modes, so the VUV inherits the cavity’s stability.
The clock’s correction each cycle is written through an acousto-optic modulator between the 1,397-nm laser and the comb, steering the VUV onto the nucleus.
06Context
Why put a clock in a nucleus?
One of a kind
Nuclear transitions usually sit far above anything a laser can reach. Thorium-229’s isomer is the singular exception: the only known nuclear transition that current lasers can address directly, at 148.4 nm.
~1017 per cm3
A transparent crystal holds a macroscopic number of nuclei (TS1: 2.2 × 1017 cm−3), opening a route to compact, robust and field-deployable frequency references.
Beyond timekeeping
The authors point to compact nuclear clocks, nuclear quantum sensors and precision tests of fundamental physics. A near-simultaneous independent report (Toscani De Col et al., Nature 2026) also demonstrated 229Th clock operation, with a different VUV source and detector.
The clock is already within a factor of two of its photon-shot-noise limit, and the authors note that the remaining 15 % dead time in each cycle, an overhead of the present digital feedback, could be reduced substantially with improved feedback control.
07Sources & credits
Sources
Data
Huang, B., Yan, G., Xiao, Q. & Bu, W. Data supporting “A nuclear clock synchronized to 229Th”. figshare, doi:10.6084/m9.figshare.33386191 (CC BY 4.0).
Reused material and changes
Measured Fig. 1e, Fig. 2a–f and Fig. 3 data are replotted from the figshare source data; values rounded, long series downsampled, Fig. 2a frequencies stored as offsets from 2,020.407 THz. Fig. 3 points are shown with their 68.3 % confidence intervals.
Calculated The depth profile used to shade the nuclei is the paper’s calculated steady-state excitation fraction for TS1 (Fig. 2j).
Model Line shape, error signal and slope follow Methods eqs. 9–10 and 22–23 with Δ = 27 kHz; the error-signal amplitude is scaled to the measured Fig. 1e trace.
Simulated The clock loop corrects the full estimated detuning each cycle, with Gaussian per-cycle noise σν = A·ν0/√Tc calibrated to each configuration’s measured fitted instability A and cycle time Tc (Extended Data Table 1). Simulated records and Allan deviations are illustrations of the reported behaviour, not the measured data; time is slowed or accelerated for display.
Illustrative The crystal, beam, detector, photon rain and cadmium ladder drawings are schematic; nuclear glow is greatly exaggerated.
Photographs of S1 and TS1 (Fig. 2g,h) and Fig. 1 are reproduced from the paper and figshare under CC BY 4.0, resized.
Typefaces: Instrument Serif, Inter Tight and IBM Plex Mono (SIL Open Font License).
Topics in this iPaper
Physics
Here we demonstrate a 229Th nuclear clock by stabilizing a continuous-wave, narrow-linewidth 148.4 nm vacuum-ultraviolet (VUV) laser
Abstract, page 1
nuclear clock
A nuclear clock transfers the frequency reference from an electronic to a nuclear transition
Abstract, page 1
laser spectroscopy
We perform continuous-wave laser spectroscopy of the 229Th isomeric transition in home-grown 229Th:CaF2 crystals
Conclusion, page 5
frequency metrology
The overlapping Allan deviation is statistically consistent with τ−1/2 scaling over the measured interval
Nuclear clock operation and instability, page 5
frequency modulation spectroscopy
we modulate the VUV frequency sinusoidally and demodulate the transmitted VUV signal to extract the absorption-induced response
Fig. 1 caption and Absorption spectroscopy, pages 3-4
vacuum ultraviolet lasers
We generate continuous-wave 148.4-nm radiation by resonance-enhanced four-wave mixing in cadmium vapour
VUV laser for nuclear clock interrogation, page 2
crystal growth
TS1 is fabricated using the temperature-gradient technique
Fabrication of 229Th:CaF2 crystals, page 2