The atoms must stay in the light
A clock needs a reliably repeating beat. An optical clock chooses two internal energy states of an atom. Laser light can move an atom from the lower state to the higher one when its frequency matches the energy difference. Sweep the laser frequency and plot the excited fraction: a resonance line appears.
Closely spaced frequencies require a long interrogation. An atom that flies out of the beam interrupts that interrogation. Opposing lattice beams create many small traps that hold cold atoms in place. Thousands of atoms can then answer the same question, reducing the fluctuations associated with an individual atom.
The trapping light also changes atomic energies. At a magic wavelength, the two clock states experience nearly equal light shifts, reducing the transition frequency’s sensitivity to lattice intensity. The lattice light holds the atoms; a separate clock laser probes the transition.
Why ask on both sides?
Exactly at the top of the line, absorption is largest. Moving either way reduces it. A single low reading cannot tell the controller which side it occupies. Two-side interrogation supplies that missing direction. A low-frequency probe is followed by a newly prepared ensemble and a high-frequency probe.
When the commanded centre matches the resonance, equally spaced probes give equal mean excitation fractions. Shift that centre upward and the high-side probe moves farther from resonance as the low-side probe approaches it. Their difference becomes a signed error signal. Independently known offsets calibrate how much that difference changes per hertz.
The apparatus must also read the atomic states. Fluorescence can help distinguish atoms in the two states, from which their populations are estimated. A finite ensemble fluctuates from trial to trial, and state assignment can make mistakes. Even a centred pair gives observed differences that fluctuate around zero.
Sources: [2]
Feedback between interrogations
After each pair, the controller moves the laser frequency opposite the estimated error. Correcting a fraction of it can give a smooth approach. An aggressive correction can amplify counting fluctuations and the effects of delay. Nicholson and colleagues used differences between excitation fractions on opposite sides of the line to lock their strontium clock laser; their 2015 Figure 1 connects spectroscopy with stability measurements.
The atoms cannot supply that information continuously. Cooling and preparation take time, state readout takes time, and there may be a wait before the next ensemble is ready. The laser must retain its own short-term stability between probes. A complete cycle record shows when the atomic information was collected and which command it supported.
One way to describe frequency fluctuations is to compare averages over adjacent intervals. Repeat the calculation at different averaging times to obtain Allan deviation. When the input is a sequence of feedback commands, the statistic describes those commands. Independent frequency or phase observations are needed to characterize the actual output between interrogations. A constant offset cancels in the difference: a quiet trace can remain displaced.
A whole resonance can move
Imagine a perfect lock while the surrounding enclosure warms. Thermal radiation changes the atomic energy levels. If the two states shift differently, the transition frequency moves. Feedback faithfully follows the shifted line and the error remains small. Finding that displacement requires a separate measurement of temperature and atomic response.
In a static approximation the thermal shift scales with the fourth power of absolute temperature. Comparing the frequency against an independent reference at several known temperatures calibrates the coefficient. A withheld temperature tests the prediction. An operating temperature can then support a conditional correction. A nonuniform enclosure can invalidate the assumption that one measured temperature represents the radiation seen by the atoms.
The actual evaluation by Nicholson and colleagues included the dynamic blackbody contribution, lattice light, magnetic fields and other effects, with specially calibrated vacuum thermometers. The teaching model retains one independently calibrated static thermal shift. Establishing a clock frequency also requires the remaining shift measurements and a connection from the output to independent clock standards.
Sources: [2]