Optical Atomic Clocks
Clocks that count light, not microwaves, now reach an uncertainty of 5.5 parts in 10^19 (NIST, 2025), about 100 times better than caesium; a decision on redefining the second is planned for 2030.
Open in the interactive tree →An optical clock locks a laser to a visible-light transition of trapped atoms or ions and counts the light's cycles with a femtosecond frequency comb (Nobel Prize 2005 for Hall and Hänsch). Katori built the first optical lattice clock in Tokyo in 2003; a JILA strontium clock was evaluated at 2.1 x 10^-18 in 2015, and an aluminium-ion clock at NIST reached 5.5 x 10^-19 in 2025. Because light oscillates tens of thousands of times faster than the caesium microwave, such clocks resolve far smaller frequency shifts. In 2024 a laser excited the thorium-229 nucleus for the first time, the basis of a proposed nuclear clock.
As of now
In October 2026 the second is still defined by caesium-133. The 28th General Conference on Weights and Measures meets on 13-15 October 2026 in Versailles and reviews progress; the plan is for the time committee to shortlist one proposal in 2028-2029 and for the 2030 conference to decide, either on a single optical transition or on a weighted mix of several, with an expected accuracy gain of 10 to 100 times. Mandatory criteria include continuity with caesium and regular contributions of optical clocks to international atomic time. In July 2025 NIST published a single-ion aluminium clock with 5.5 x 10^-19 systematic uncertainty (Physical Review Letters 135, 033201), and in April 2024 a PTB/TU Wien team excited the thorium-229 nuclear transition (about 8.36 eV, 148.4 nm) with a laser.
Where today's AI could help
Low AI leverage. Clock accuracy is limited by physics and hardware (lasers, traps, shifts from the environment); machine learning mainly speeds up tuning, locking and timescale prediction.
- Bayesian and reinforcement-learning control of laser cooling and of the clock lock
- Predicting and steering clock noise inside a timescale such as UTC
- Automated search for the largest systematic shifts in a new clock species
Shown so far
- In 2024 a Bayesian frequency-estimation lock improved the fractional frequency stability of a cold-atom microwave clock by 5.1 dB over conventional PID locking (Physical Review Applied 22, 044058). source
- In 2023 a reinforcement-learning controller trained in simulation cooled atoms in a magneto-optical trap in a real experiment, the first step of most cold-atom clocks. source
Prerequisites
- Atomic Clock1955Same principle as the caesium clock, but counting light instead of microwaves
- Laser1960Ultra-stable lasers probe the atoms
- Bose-Einstein Condensate & Cold Atoms1995Laser-cooled atoms and ions held in optical lattices or traps