Convert Ramsey phase into a frequency estimate

Start with atoms in a known energy state. A resonant half-pi pulse creates an equatorial superposition with phase set by the reference oscillator. During free evolution, a constant ordinary-frequency offset $\delta\nu$ produces phase $\phi=2\pi\delta\nu t$. A second half-pi pulse with a controlled analysis phase rotates a chosen equatorial component onto the detected population difference. Choosing the analysis convention to operate at quadrature gives a sine fringe with maximal small-offset slope. Another pulse convention can reverse its sign without changing the information.

Free-evolution duration and pulse duration are distinct. The elementary formula assumes pulses short enough that their additional phase accumulation and detuning response can be neglected or separately corrected. Contrast summarizes a specified reduction of fringe amplitude from imperfect preparation, loss of coherence, or readout. It is not automatically a microscopic decoherence rate. The populations after a final pulse depend on that pulse's reference phase, so detecting energy populations during free evolution does not by itself implement the complete Ramsey measurement.

For N independent atoms in each of m independent cycles, the combined success fraction has midpoint standard deviation $1/(2\sqrt{Nm})$. Dividing by the frequency-response slope gives the local frequency uncertainty. The expression concerns independent projection noise and a calibrated contrast. Collective squeezing changes the relevant population variance and response, while correlated oscillator fluctuations require a joint noise model. More atoms cannot average away a common unknown reference error as though each atom possessed its own independently fluctuating reference.

A simple detector model independently flips each true binary label with probability $\epsilon$. The observed probability is $\epsilon+(1-2\epsilon)p_1$, so the observed contrast becomes $C_{\mathrm{eff}}=(1-2\epsilon)C$ for $0\le\epsilon<1/2$. At one-half error probability the readout carries no information. Unknown asymmetric errors can also shift the midpoint and must be calibrated. Subtracting their average effect cannot remove the extra uncertainty in a finite recorded sample or uncertainty in the error calibration itself.

Longer free evolution increases phase response but also narrows the unambiguous frequency range of a single fringe. It may reduce contrast and increases cycle time. Dead time consumes the observation budget without accumulating the desired free-evolution phase. A realistic comparison evaluates information per complete cycle time, using the actual contrast at that interrogation duration. If oscillator noise evolves during gaps, independent-cycle projection noise can cease to be the dominant limit; a clock stability analysis then requires its temporal spectrum and sampling schedule.

Sensitivity per square root time summarizes how uncertainty averages in a regime where independent-cycle assumptions hold. It is not a frequency response bandwidth. A sensor that interrogates for a long interval averages over temporal variations and may report one estimate only after each complete cycle. It can have excellent sensitivity to a constant offset while missing fast changes. State the target signal's temporal behavior and the range needed to avoid fringe ambiguity alongside the scalar sensitivity. Better short-term sensitivity need not imply better accuracy of the frequency reference.