Precision Measurement and Metrology

Precision measurement uses a physical system to estimate a quantity with a stated uncertainty. Metrology adds the discipline needed to make that estimate comparable across instruments, laboratories, places, and times: an explicit measurand, a calibrated measurement model, traceability, uncertainty propagation, stability analysis, and tests for unrecognized bias.

Atomic, molecular, and optical systems are unusually powerful metrological resources because quantum mechanics supplies reproducible energy differences. For two stationary states,

$$h\nu_0 = E_e-E_g.$$

The transition frequency $\nu_0$ can be interrogated repeatedly, compared with an oscillator, and connected by frequency division or a frequency comb to electronic counters. The same logic extends beyond clocks. A magnetic field produces a Zeeman shift, an acceleration produces an interferometer phase, an electric dipole moment produces an energy shift, and a change in a dimensionless constant changes a frequency ratio.

The quantum transition is not an incorruptible number displayed by nature. The measured resonance is shifted by fields, motion, collisions, probing, gravity, and servo dynamics. It is broadened by finite interrogation, decoherence, oscillator noise, and inhomogeneity. Detection adds projection noise and technical noise. Precision AMO metrology is the art of exploiting the reproducibility of quantum structure while measuring every relevant way the apparatus perturbs it.

Helpful background. Ramsey Interferometry explains phase-to-population readout, Precision Spectroscopy separates line centers from linewidths, and Fisher Information quantifies how an outcome distribution changes with a parameter.

Enter this chapter

Measurement goal	Lessons
Realize and compare frequency references	Atomic Clocks, Optical Clocks, Frequency Standards
Measure motion or magnetic fields	Atom-Interferometric Sensors, Magnetometry
Infer constants or test a physical model	Fundamental Constants, Variation of Constants Searches, Tests of Fundamental Symmetries
Use molecular structure to enhance a signal	Precision Molecular Measurements

Metrological Language

Measurand and measurement model

The measurand is the quantity intended to be measured. It must be specified closely enough that different implementations refer to the same quantity. “The transition frequency” is often incomplete. A clock measurand might instead be:

the unperturbed frequency of a specified transition in a specified isotope, for an atom at rest, in a declared reference frame, with stated field and environmental conventions.

A measurement model relates the output quantity $Y$ to input quantities:

$$Y = f \left( X_1,X_2,\ldots,X_n \right).$$

The inputs can include raw observations, calibration constants, environmental parameters, theoretical coefficients, and corrections. A reported result is therefore not merely an average of detector readings. It is an inference through $f$.

Indication, correction, and result

For a frequency measurement, write a representative model as

$$\nu_{\mathrm{obs}} = \nu_0 + \sum_i \Delta\nu_i + \epsilon_{\mathrm{stat}},$$

where $\nu_0$ is the target unperturbed frequency, $\Delta\nu_i$ are systematic shifts, and $\epsilon_{\mathrm{stat}}$ represents zero-mean statistical fluctuation in the stated averaging protocol. The corrected estimate is

$$\widehat{\nu}_0 = \overline{\nu}_{\mathrm{obs}} - \sum_i \widehat{\Delta\nu}_i.$$

Some laboratories tabulate corrections to add rather than shifts to subtract. Both conventions are valid, but the sign convention must be stated and checked with a synthetic example.

Precision, stability, and uncertainty

These words answer different questions:

Term	Question	Typical evaluation
Resolution	What is the smallest display increment or distinguishable feature?	digitization, linewidth, response curve
Repeatability	How closely do repeated results agree under fixed conditions?	repeated trials
Precision	How dispersed are repeated estimates?	standard deviation or interval
Stability	How do frequency or phase fluctuations depend on averaging time?	Allan-family statistics, spectra
Bias	How far is an estimator displaced in expectation?	calibration, reversal, comparison
Measurement uncertainty	How dispersed are the values attributed to the measurand using the available information?	standard uncertainty, covariance propagation, or a specified coverage interval
Traceability	Through what documented calibration chain is the result related to a reference?	comparison records and uncertainty chain

A standard uncertainty is a nonnegative dispersion parameter, commonly expressed as a standard deviation. A coverage interval additionally needs a probability or coverage convention; uncertainty is not simply the interval itself. These distinctions follow the International Vocabulary of Metrology.

Stability does not establish that a frequency is correct. A quiet oscillator can have a fixed offset. Conversely, a well-evaluated reference can have a large short-term instability and still yield an unbiased long-term result.

Accuracy is not a number by itself

In formal metrology, measurement accuracy is a qualitative closeness concept rather than a quantity with its own numerical value. Experimental papers often use “accuracy” informally for small systematic uncertainty. Clearer reporting separates:

$$\text{measured offset}, \qquad \text{statistical uncertainty}, \qquad \text{systematic uncertainty}, \qquad \text{total uncertainty}.$$

Turning a physical quantity into a frequency shift

Suppose a quantity $x$ changes a measured frequency. Near an operating point,

$$\delta\nu \simeq \left. \frac{\partial\nu}{\partial x} \right|_{x_0} \delta x.$$

Define the dimensional transduction coefficient

$$k_x = \frac{\partial\nu}{\partial x}.$$

Then a frequency-estimation uncertainty $u(\nu)$ corresponds, in the linear regime, to

$$u(x) \simeq \frac{u(\nu)}{|k_x|}.$$

Large $|k_x|$ is useful only if nuisance quantities do not produce indistinguishable shifts. Selectivity and independent controls matter as much as raw response.

For a dimensionless parameter $X$, the logarithmic sensitivity coefficient

$$K_X = \frac{\partial\ln\nu}{\partial\ln X}$$

gives

$$\frac{\delta\nu}{\nu} \simeq K_X \frac{\delta X}{X}.$$

Frequency-ratio comparisons involve differences of sensitivity coefficients and are dimensionless; a practical comparison still has transfer, reference, and calibration uncertainties.

Calibration is an experiment

If the shift model is

$$\Delta\nu_B = \beta B^2,$$

then neither $\beta$ nor $B$ should be treated as an unexamined label. $\beta$ may come from theory, an auxiliary measurement, or an in situ field scan. $B$ may be inferred from a magnetically sensitive transition rather than a remote probe. The calibration procedure has its own transfer function, gradients, time dependence, and uncertainty.

Why Quantum Systems Make References

Identical particles and reproducible structure

Atoms of one isotope are not manufactured artifacts with individually different dimensions. Within a specified internal state and environment, they realize the same quantum Hamiltonian. This supports reproducibility across apparatus and laboratories.

The statement has limits. Isotope composition, hyperfine state, motional state, local fields, collisions, trap light, and relativistic potential all matter. Identical bare particles do not imply identical perturbed frequencies.

Narrow transitions and quality factor

A transition with ordinary frequency $\nu_0$ and observed linewidth $\Delta\nu$ has quality factor

$$Q = \frac{\nu_0}{\Delta\nu}.$$

For a fixed signal-to-noise ratio, larger $Q$ usually permits finer fractional discrimination. Optical frequencies also provide a larger phase accumulation per unit fractional detuning than microwave frequencies:

$$\phi = 2\pi \int_0^T \delta\nu(t)\,dt.$$

A narrow natural linewidth is not enough. The local oscillator, probe time, coherence, state preparation, and systematic shifts determine the usable line.

Long coherence and controlled isolation

Useful references suppress or characterize environmental coupling while retaining controlled preparation and readout. Examples include:

• hyperfine “clock” states with canceled first-order Zeeman sensitivity;
• forbidden optical transitions with long excited-state lifetimes;
• ions confined near the Lamb–Dicke regime;
• neutral atoms trapped at a magic wavelength;
• molecular states with enhanced sensitivity to symmetry-violating interactions; and
• spin ensembles with long transverse coherence.

Isolation and sensitivity compete. A sensor must couple strongly enough to the desired quantity while rejecting or separately measuring nuisance couplings.

Ratios are especially powerful

An absolute frequency measurement compares against a realization of the unit hertz. A ratio

$$R = \frac{\nu_A}{\nu_B}$$

is dimensionless. Optical combs can compare distant optical frequencies coherently, and many oscillator fluctuations can cancel in synchronous or transfer-oscillator schemes. Repeated ratio measurements are central to clock validation and searches for changes in dimensionless constants.

Frequency as a Measurable Quantity

Phase is the immediate observable

Frequency is inferred from accumulated phase. For an oscillator

$$V(t) = V_0 \cos \left[ 2\pi\nu_0t+\phi(t) \right],$$

the instantaneous frequency deviation is

$$\delta\nu(t) = \frac{1}{2\pi} \frac{d\phi}{dt}.$$

No finite-duration experiment measures an instantaneous mathematical frequency. It estimates an average phase slope over a specified interval and bandwidth.

Fractional frequency

For a nominal reference frequency $\nu_0$, define

$$y(t) = \frac{\nu(t)-\nu_0}{\nu_0}.$$

Fractional frequency makes microwave and optical systems comparable and connects directly to relativistic and constant-variation signals. An average over an interval $\tau$ is

$$\overline y_k(\tau) = \frac{1}{\tau} \int_{t_k}^{t_k+\tau} y(t)\,dt.$$

The associated accumulated time or phase error depends on the integral of $y(t)$, not just its pointwise value.

Passive and active references

In a passive atomic standard, an external local oscillator interrogates the atoms. The atoms provide a discriminator; a servo corrects the oscillator. Most primary and optical clocks use this architecture.

In an active standard, stimulated emission from the atomic medium directly sustains an oscillator. Hydrogen masers are a familiar example. The distinction concerns how the output is generated, not whether quantum states are involved.

Passive-reference loop

Let $\nu_{\mathrm{LO}}$ be the local-oscillator frequency and $\nu_a$ the perturbed atomic resonance. An interrogation produces an error signal

$$e_k \simeq D \left( \nu_{\mathrm{LO},k}-\nu_{a,k} \right) + n_k$$

near the lock point, where $D$ is the discriminator slope and $n_k$ is measurement noise. A simple digital integrator updates

$$\nu_{\mathrm{LO},k+1} = \nu_{\mathrm{LO},k} - g e_k.$$

The loop tracks the perturbed atomic resonance, not automatically the unperturbed measurand $\nu_0$. Corrections for Zeeman, Stark, collision, motion, gravity, probe, and servo effects remain necessary.

Current SI second

As of 2026-10-14, the BIPM definition of the second fixes the unperturbed ground-state hyperfine transition frequency of $^{133}\mathrm{Cs}$ to exactly

$$\Delta\nu_{\mathrm{Cs}} = 9\,192\,631\,770\ \mathrm{Hz}.$$

Optical standards are being developed toward a possible future redefinition. The BIPM roadmap and current FAQ describe a possible 2030 decision, contingent on agreement and the required criteria. They do not replace the current cesium definition.

Stability

Allan variance

For adjacent fractional-frequency averages of duration $\tau$, the two-sample Allan variance is

$$\sigma_y^2(\tau) = \frac{1}{2} \left\langle \left[ \overline y_{k+1}(\tau) - \overline y_k(\tau) \right]^2 \right\rangle.$$

The Allan deviation is

$$\sigma_y(\tau) = \sqrt{\sigma_y^2(\tau)}.$$

It is a function of averaging time. A statement such as “the stability is $10^{-16}$” is incomplete without $\tau$, the statistic variant, sampling protocol, uncertainty, and treatment of drift or dead time.

For $M$ adjacent samples,

$$\widehat{\sigma}_y^2(\tau) = \frac{1}{ 2(M-1) } \sum_{k=1}^{M-1} \left( \overline y_{k+1} - \overline y_k \right)^2.$$

Overlapping Allan deviation uses more of a phase record and has different degrees of freedom. Modified Allan, Hadamard, and total deviations answer related but distinct questions. The chosen statistic should match the noise and data structure.

White-frequency scaling

For uncorrelated cycle-to-cycle frequency estimates,

$$\sigma_y(\tau) \propto \tau^{-1/2}.$$

Flicker noise, random walk, environmental cycles, drift, and servo dynamics produce different slopes or floors. A fitted power law over a short range does not identify a unique microscopic noise source.

Dead time and oscillator aliasing

If atoms are interrogated only during part of each cycle, local-oscillator noise during dead time is not observed continuously. Periodic sampling can alias oscillator noise into the measurement band, producing the Dick effect. Increasing atom number does not suppress this classical oscillator contribution. Higher duty cycle, synchronous comparison, better local oscillators, and zero-dead-time architectures address it.

Comparison is the observable

A single oscillator cannot reveal its own absolute instability. Stability is inferred from a comparison:

$$y_{AB}(t) = y_A(t)-y_B(t).$$

Separating individual contributions requires assumptions, a third reference, cross-correlation, or a known noise hierarchy. Common-mode environment and shared oscillators can make a comparison look quieter than either independent system.

Systematic Shifts

Combining frequency corrections

For a clock or frequency sensor, write

$$\frac{\widehat{\nu}_0-\nu_{\mathrm{ref}}}{\nu_0} = y_{\mathrm{obs}} + \sum_i c_i,$$

where $c_i$ are fractional corrections to add under this convention. Each entry should include:

• physical mechanism;
• mathematical model;
• measured operating parameter;
• correction and sign;
• standard uncertainty;
• sensitivity to analysis choices;
• covariance with other entries; and
• validation or reversal test.

The uncertainty of an uncertain correction remains part of the final result, even if the correction itself is small.

Representative mechanisms

Mechanism	Typical dependence	Diagnostic
Linear Zeeman	$m_F B$	reverse $m_F$, field scan, magnetically sensitive line
Quadratic Zeeman	$\beta B^2$	vary field magnitude, independent field monitor
DC Stark	quadratic scalar and tensor shifts for a nondegenerate state without a permanent dipole	vary field magnitude and orientation; reverse an applied field to isolate its cross term with a stray field
Blackbody radiation	dynamic polarizability and thermal spectrum	temperature map, emissivity and view-factor model
Probe light	intensity, detuning, pulse area, line shape	interleaved intensity and duration scan
Trap or lattice light	differential polarizability, multipolar terms	wavelength and depth scan
Motion	first- and second-order Doppler, recoil, time dilation	sideband thermometry, velocity reversal
Collisions	density and state correlations	atom-number or density extrapolation
Line pulling	neighboring components and asymmetric response	resolve components, vary preparation and fit model
Servo error	discriminator offset, drift, loop delay	reverse modulation, alter gain and cycle
Gravity	gravitational potential	surveyed height and potential model

The relevant list depends on the platform. A correction budget copied from a different species or apparatus is not an evaluation.

Covariance-aware propagation

For

$$Y = f(\mathbf X),$$

the first-order combined variance is

$$u_c^2(Y) = \sum_{i,j} \frac{\partial f}{\partial X_i} \frac{\partial f}{\partial X_j} \operatorname{Cov} \left( X_i,X_j \right).$$

Equivalently,

$$u_c^2(Y) = \mathbf g^{\mathsf T} \mathbf C \mathbf g,$$

where $g_i=\partial f/\partial X_i$ and $\mathbf C$ is the covariance matrix. Adding standard uncertainties in quadrature assumes the relevant covariances vanish.

Worked example: correlated corrections

Suppose a fractional comparison gives

$$y_{\mathrm{obs}} = 7.40\times10^{-16}.$$

Under the correction-to-add convention, let

$$\begin{array}{c|c|c} \text{effect} & c_i/(10^{-16}) & u_i/(10^{-16}) \\ \hline \text{blackbody} & -5.20 & 0.40 \\ \text{quadratic Zeeman} & +0.60 & 0.20 \\ \text{density} & -0.30 & 0.30 \end{array}$$

Then

$$\widehat y = \left( 7.40-5.20+0.60-0.30 \right) \times10^{-16} = 2.50\times10^{-16}.$$

If the blackbody and density corrections have correlation coefficient $\rho=0.50$, their covariance contribution is

$$2\rho u_{\mathrm{BBR}}u_{\mathrm{den}} = 0.12\times10^{-32}.$$

The systematic standard uncertainty is

$$\begin{aligned} u_{\mathrm{sys}} &= \sqrt{ 0.40^2+0.20^2+0.30^2 + 2(0.50)(0.40)(0.30) } \times10^{-16} \\ &= 0.640\times10^{-16}. \end{aligned}$$

With an independent statistical uncertainty $u_{\mathrm{stat}}=0.50\times10^{-16}$,

$$u_{\mathrm{tot}} = \sqrt{ u_{\mathrm{sys}}^2 + u_{\mathrm{stat}}^2 } = 0.812\times10^{-16}.$$

Ignoring the covariance would give $u_{\mathrm{sys}}=0.539\times10^{-16}$ and understate the declared uncertainty.

An illustrative uncertainty model, not a universal clock-noise curve. For independent contributions, take $u_{\mathrm{stat}}=3u_0\sqrt{\tau_0/\tau}$ and a fixed calibration standard uncertainty $u_{\mathrm{cal}}=u_0$. Then $u_c=\sqrt{u_{\mathrm{stat}}^2+u_{\mathrm{cal}}^2}$. Repeating the same measurement reduces the statistical part but does not improve a calibration held fixed. This plot is not an Allan deviation: a constant frequency bias cancels from adjacent differences and can remain invisible in a stability plot.

Nonlinearity and model uncertainty

Linear covariance propagation can fail when:

• the response is strongly nonlinear over the input uncertainty;
• distributions are asymmetric or bounded;
• a correction depends on a fitted model with several solutions;
• the measurand lies near a cancellation point; or
• model discrepancy dominates parameter uncertainty.

Monte Carlo propagation through the measurement model can capture nonlinearity, but it does not repair an omitted physical mechanism. Alternative models, stress tests, and independent comparison are needed for model uncertainty.

Quantum Projection Noise

Binomial population fluctuations

Prepare $N$ independent two-level atoms with excited-state probability $p$. If $K$ atoms are detected in the excited state,

$$K \sim \operatorname{Binomial}(N,p),$$

so

$$\mathbb E[K] = Np, \qquad \operatorname{Var}(K) = Np(1-p).$$

For the measured fraction

$$\widehat p = \frac{K}{N},$$ $$\operatorname{Var}(\widehat p) = \frac{p(1-p)}{N}.$$

At $p=1/2$,

$$\Delta\widehat p = \frac{1}{2\sqrt N}.$$

This is quantum projection noise for an ideal fixed-$N$, independent-atom population measurement. Atom-number fluctuations, detection noise, preparation noise, and technical correlations add other terms.

From population noise to phase noise

For a Ramsey fringe

$$p(\phi) = \frac12 \left[ 1+C\cos\phi \right],$$

operate at a midfringe point with $p=1/2$ and

$$\left| \frac{dp}{d\phi} \right| = \frac{C}{2}.$$

Linear error propagation gives

$$\Delta\phi_{\mathrm{QPN}} = \frac{ \Delta\widehat p }{ |dp/d\phi| } = \frac{1}{ C\sqrt N }.$$

The $1/\sqrt N$ scaling follows from independent trials. Contrast loss reduces the signal slope and worsens phase sensitivity.

Ideal clock-stability estimate

If the Ramsey phase is

$$\phi = 2\pi\delta\nu T,$$

then one cycle has fractional-frequency uncertainty

$$\Delta y_{\mathrm{QPN}} = \frac{1}{ 2\pi\nu_0TC\sqrt N }.$$

For independent cycles of duration $T_c$ and total averaging time $\tau$,

$$\sigma_{y,\mathrm{QPN}}(\tau) \simeq \frac{1}{ 2\pi\nu_0TC\sqrt N } \sqrt{ \frac{T_c}{\tau} }.$$

This formula assumes midfringe operation, white independent projection noise, no atom-number uncertainty, no oscillator aliasing, and no systematic floor.

For

$$\begin{aligned} \nu_0 &= 4.29\times10^{14}\ \mathrm{Hz}, & T &= 0.500\ \mathrm{s}, \\ N &= 1000, & C &= 0.80, & T_c &= 1.00\ \mathrm{s}, \end{aligned}$$

the ideal estimate is

$$\sigma_{y,\mathrm{QPN}}(\tau) \simeq 2.93\times10^{-17} \sqrt{ \frac{1\ \mathrm{s}}{\tau} }.$$

At $\tau=100\ \mathrm{s}$ this is

$$\sigma_{y,\mathrm{QPN}} \simeq 2.93\times10^{-18}.$$

An experiment above this line is not necessarily poorly designed: local-oscillator noise, dead time, detection, collisions, or deliberately small atom number may dominate.

Projection noise is not detector noise

Projection noise remains even with a perfect detector because each measurement samples a quantum probability distribution. Detector noise can be estimated by repeatedly measuring a known eigenstate or by independent calibration. Subtracting detector variance to infer projection noise should be disclosed, and uncertainty in the subtraction must be propagated.

Standard Quantum Limit and Squeezing Preview

Independent-particle scaling

For $N$ independent probes repeated $m$ times, a phase uncertainty typically scales as

$$\Delta\phi \propto \frac{1}{\sqrt{mN}}.$$

This is often called the standard quantum limit or shot-noise limit in the specified protocol. The phrase is not universal: resource counting, allowed states, loss, interrogation time, and estimator assumptions must be declared.

Standard Quantum Limit derives this independent-probe benchmark from Fisher-information additivity and explains how to compare protocols with the same resources.

An ideal maximally correlated state can produce a phase slope scaling with $N$ and, in a noiseless local-estimation setting, approach

$$\Delta\phi \propto \frac{1}{N\sqrt m}.$$

This Heisenberg scaling is not an automatic experimental advantage. State preparation, reduced dynamic range, decoherence, readout, and prior phase knowledge can remove the gain.

Collective spin

For $N$ two-level atoms, define collective operators

$$\mathbf J = \frac12 \sum_{i=1}^{N} \boldsymbol{\sigma}^{(i)}.$$

A coherent spin state polarized along $x$ has

$$\langle J_x\rangle = \frac{N}{2},$$

and transverse variances

$$(\Delta J_y)^2 = (\Delta J_z)^2 = \frac{N}{4}.$$

The uncertainty disk is isotropic in the tangent plane. Interactions or measurement backaction can redistribute fluctuations and create a squeezed state.

Wineland metrological parameter

For phase estimation using a measured transverse component, a common metrological squeezing parameter is

$$\xi_R^2 = \frac{ N(\Delta J_\perp)^2 }{ |\langle\mathbf J\rangle|^2 }.$$

A coherent spin state has $\xi_R^2=1$. Under the assumptions of the interferometric protocol,

$$\xi_R^2<1$$

indicates reduced local phase variance relative to a coherent spin state. For a fixed number of spin-1/2 particles with collective linear phase encoding, it also excludes a fully separable particle state. It does not by itself prove genuine entanglement among all $N$ particles.

Define the noise-variance ratio in decibels by

$$R_{\mathrm{dB}} = 10\log_{10}\xi_R^2.$$

Negative $R_{\mathrm{dB}}$ means reduced noise. The positive gain convention would use $-R_{\mathrm{dB}}$ instead. For $R_{\mathrm{dB}}=-6.0\ \mathrm{dB}$,

$$\xi_R^2 = 10^{-0.6} \simeq 0.251.$$

The phase standard deviation improves by

$$\sqrt{\xi_R^2} \simeq 0.501,$$

For equal cycle times and independent repetitions, the averaging time needed to reach the same variance ideally falls to $0.251$ of the unsqueezed value. State-preparation or readout overhead can reduce that advantage.

Noise can erase asymptotic scaling

Independent dephasing, loss, and correlated local-oscillator noise can change the best scaling and optimal state. Fragile highly entangled states may perform worse than moderately squeezed states. The relevant question is finite-resource performance under the measured noise model, not the noiseless asymptote.

AMO Sensor Map

Clocks and frequency standards

A clock compares an oscillator with a transition whose unperturbed frequency defines or realizes a reference. Microwave standards use hyperfine transitions; optical standards use electronic transitions with much larger carrier frequencies. The platform must provide:

• state preparation;
• a coherent local oscillator;
• an interrogation discriminator;
• state-selective readout;
• a servo;
• frequency distribution or division; and
• an evaluated systematic budget.

Clock stability and clock systematic uncertainty are separate performance axes.

Magnetic-field sensing

For a spin with gyromagnetic ratio $\gamma$,

$$\omega_L = \gamma B.$$

Optical pumping creates spin polarization, the field drives Larmor precession, and optical rotation or absorption reads it out. Sensitivity depends on spin number, coherence time, measurement bandwidth, optical depth, pumping, collisions, and magnetic shielding. Zero-field, finite-field, scalar, vector, and radio-frequency magnetometers answer different measurement questions.

Electric fields and polarizability

For a nondegenerate state without a permanent laboratory-frame dipole, a weak static field often produces a quadratic Stark shift,

$$\Delta E = -\frac12 \alpha(0)E^2.$$

AC fields use dynamic polarizability and can include scalar, vector, and tensor components. Electric-field sensing must distinguish the desired field from trap, patch, blackbody, and probe fields.

Inertial and gravitational sensing

For an ideal three-pulse light-pulse atom interferometer,

$$\Phi_a = \mathbf k_{\mathrm{eff}} \mathbin{\cdot} \mathbf a\,T^2.$$

The phase can estimate acceleration, gravity, a gravity gradient, or rotation depending on geometry and differencing. Laser phase, mirror motion, wavefronts, Coriolis effects, gravity gradients, and wave-packet closure enter the same measurement chain.

Relativistic geodesy

For a small height difference $h$ in a nearly uniform gravitational field,

$$\frac{\Delta\nu}{\nu} \simeq \frac{gh}{c^2}.$$

Numerically,

$$\frac{g}{c^2} \simeq 1.09\times10^{-16}\ \mathrm{m}^{-1}.$$

A height difference of $1\ \mathrm{cm}$ therefore corresponds to a fractional shift near

$$1.09\times10^{-18}.$$

At this level, one needs a gravitational-potential model rather than only a tape-measured geometric height.

Forces, gradients, and imaging

Position-dependent energy shifts and phases allow measurement of:

• gravity and magnetic gradients;
• surface forces and Casimir–Polder interactions;
• electric patch fields near traps;
• inertial forces in guided systems;
• material magnetic fields;
• pressure and collision shifts; and
• temperature through occupation or transition ratios.

Spatial resolution, field sensitivity, bandwidth, and probe backaction trade against one another.

Fundamental Tests

Constants from spectra

Atomic and molecular energies depend on dimensionless combinations such as the fine-structure constant $\alpha$, mass ratios, nuclear moments, and strong-interaction parameters. A calculated transition can be written schematically as

$$\nu = \nu_{\mathrm{scale}} F \left( \alpha,\mu, g_I,\ldots \right),$$

where $\mu=m_e/m_p$ or another declared mass ratio. Extracting a constant requires both measurement and theory, including their covariance and possible shared input data.

Searches for variation

For a ratio $R=\nu_A/\nu_B$,

$$\delta\ln R = \left( K_{\alpha,A}-K_{\alpha,B} \right) \delta\ln\alpha + \left( K_{\mu,A}-K_{\mu,B} \right) \delta\ln\mu + \cdots.$$

Only dimensionless ratios are operationally meaningful in a constant-variation claim. Different clocks or molecular transitions are chosen to provide linearly independent sensitivity vectors.

Time series can search for:

• secular drift;
• annual or gravitational-potential modulation;
• oscillations from a coherently varying field;
• transient defects or encounters; and
• correlations across a sensor network.

The signal model and look-elsewhere treatment must be declared before interpreting a spectral peak.

Symmetry-violating observables

AMO systems amplify or isolate small effects through long coherence, large internal fields, near-degeneracies, and state reversals. Major targets include:

• permanent electric dipole moments;
• atomic parity violation;
• nuclear anapole moments;
• Lorentz and CPT symmetry tests;
• equivalence-principle tests; and
• spin-dependent or composition-dependent new forces.

A null result is a constraint on a specified effective interaction, not a proof that a symmetry is exact at every scale.

Molecules as enhancement platforms

Molecules offer rotational structure, parity doublets, large polarizability, and strong internal effective electric fields. These can enhance sensitivity and enable powerful reversals. They also introduce dense spectra, state-preparation complexity, tensor shifts, and molecular structure theory. Enhancement factors require independent validation.

Testing shifts and comparing instruments

Blinding and analysis choices

When analyst choices can move the final answer, a hidden offset or blinded signal can reduce confirmation bias. A useful blind preserves the ability to debug stability, reversals, and uncertainty budgets while hiding the quantity that would reveal the expected result. Unblinding criteria should be fixed in advance.

Reversals

A reversal changes the sign or functional form of a target signal while leaving many nuisance effects unchanged. Examples include:

• electric-field reversal for an EDM;
• magnetic sublevel reversal for a Zeeman contribution;
• wavevector reversal in atom interferometry;
• polarization or orientation reversal;
• isotope or species comparison; and
• alternating high and low control parameters to measure a shift slope.

No reversal is perfect. Correlated changes in field magnitude, leakage, temperature, geometry, or detector response can create a false odd signal.

Interleaved evaluation

To estimate a shift coefficient, alternate two operating points faster than the relevant drift:

$$\widehat k_x = \frac{ \overline\nu(x_2) - \overline\nu(x_1) }{ x_2-x_1 }.$$

Randomized or balanced ordering can reduce bias from monotonic drift. Nonlinearity requires more than two points. Extrapolation to an operating point far outside the calibration range is risky.

Independent comparison

Agreement between two systems is strongest when they differ in:

• species or transition;
• trap and interrogation architecture;
• environmental sensitivities;
• frequency-transfer path;
• analysis code;
• personnel; and
• calibration history.

Two nominally independent clocks sharing the same laser, comb, temperature model, or gravitational-potential input can share hidden covariance.

Closure tests

For three frequency ratios,

$$R_{AB}R_{BC}R_{CA} = 1$$

ideally. In logarithmic form,

$$\ln R_{AB} + \ln R_{BC} + \ln R_{CA} = 0.$$

Closure tests expose inconsistency but do not alone identify which link is wrong. Covariance from shared oscillators and combs must be propagated.

Exercises

1. Specify the measurand

An experiment reports “the frequency of the $^{87}\mathrm{Sr}$ clock transition.”

1. List at least six conditions or conventions needed to turn this phrase into a usable measurand definition.
2. Distinguish the raw observed line center from the unperturbed frequency.
3. Explain why the local gravitational potential belongs in the specification at fractional uncertainties near $10^{-18}$.

Solution

A useful definition should specify at least:

1. the isotope and electronic transition;
2. hyperfine and magnetic sublevels, or the averaging procedure over them;
3. the atom’s motional and confinement convention;
4. the electromagnetic-field convention, including the unperturbed limit;
5. treatment of collisions and density;
6. the probe-light extrapolation or operational point;
7. the reference frame and relativistic convention;
8. the gravitational potential to which the frequency is referred; and
9. the line-center and averaging definition.

The observed line center includes shifts from lattice light, blackbody radiation, magnetic fields, motion, collisions, the probe, gravity, and the servo. The unperturbed estimate is obtained through a stated measurement equation and corrections.

Near Earth,

$$\frac{\Delta\nu}{\nu} \simeq \frac{gh}{c^2} \simeq 1.09\times10^{-16} \left( \frac{h}{1\ \mathrm m} \right).$$

Thus $10^{-18}$ corresponds to roughly a centimetre of height under the uniform-$g$ approximation. A result at that level must state the potential reference rather than treating laboratory location as irrelevant.

2. Compute an Allan deviation

Five adjacent fractional-frequency averages at a fixed averaging time $\tau$ are

$$\overline{\mathbf y} = \left( 1,\, 2,\, 0,\, -1,\, 1 \right) \times10^{-13}.$$

Compute the non-overlapping sample Allan deviation using

$$\widehat{\sigma}_y^2 = \frac{1}{2(M-1)} \sum_{k=1}^{M-1} \left( \overline y_{k+1}-\overline y_k \right)^2.$$

Why does this number not establish the absolute frequency offset?

Solution

The adjacent differences are

$$\left( 1,\, -2,\, -1,\, 2 \right) \times10^{-13}.$$

Their squared sum is

$$\left( 1+4+1+4 \right) \times10^{-26} = 10\times10^{-26}.$$

With $M=5$,

$$\widehat{\sigma}_y^2 = \frac{10\times10^{-26}}{2(4)} = 1.25\times10^{-26}.$$

Therefore

$$\widehat{\sigma}_y = 1.12\times10^{-13}.$$

Allan deviation uses adjacent differences and is insensitive to a constant frequency offset. The entire data set could be shifted by $10^{-9}$ without changing this result.

3. Interpret squeezing in decibels

An ensemble reports $10\log_{10}\xi_R^2=-6.0\ \mathrm{dB}$.

1. Convert this to $\xi_R^2$.
2. Find the phase-standard-deviation ratio relative to a coherent spin state.
3. Find the ideal ratio of averaging times needed to reach the same phase variance.
4. List four checks needed before accepting a metrological-advantage claim.

Solution

By definition,

$$-6.0 = 10\log_{10}\xi_R^2,$$

so

$$\xi_R^2 = 10^{-0.6} \simeq 0.251.$$

The standard deviation scales as

$$\frac{ \Delta\phi_{\mathrm{sq}} }{ \Delta\phi_{\mathrm{CSS}} } = \sqrt{\xi_R^2} \simeq 0.501.$$

For white independent noise, variance falls inversely with averaging time. The squeezed protocol therefore needs ideally

$$\frac{ \tau_{\mathrm{sq}} }{ \tau_{\mathrm{CSS}} } = \xi_R^2 \simeq 0.251$$

of the averaging time.

Checks include contrast and signal slope, atom-number equality, detection noise and subtraction, state-preparation overhead, coherence lifetime, phase dynamic range, equal total cycle time, loss, and direct task-level performance on independent data.

4. Design a fundamental-test measurement

Choose one target: an electron electric dipole moment, a variation of $\alpha$, a spin-dependent new force, or a violation of the equivalence principle.

Design a measurement plan that identifies:

1. the observable and transduction coefficient;
2. the quantum system and interrogation;
3. at least three signal reversals or comparison channels;
4. at least six systematic effects;
5. the primary statistical model;
6. a blinding or holdout policy; and
7. how the result maps to a physical constraint without overclaiming.

Solution

There is no unique solution. For an electron-EDM search in a polar molecule, the target interaction can be written schematically

$$H_d = -d_e \mathcal E_{\mathrm{eff}} \Sigma,$$

where $\mathcal E_{\mathrm{eff}}$ is a molecular-structure enhancement and $\Sigma$ is a spin projection. The observable is a phase or frequency difference that is odd under reversal of the effective internal electric field.

Useful channels include laboratory electric-field reversal, spin projection reversal, molecular orientation reversal, magnetic-field reversal, and states with different EDM sensitivity. Systematic effects include magnetic fields correlated with electric-field switching, leakage current, geometric phase, motional magnetic field, imperfect state preparation, differential Stark shifts, pulse phase, detector asymmetry, field gradients, and analysis-window choices.

The primary likelihood may model state-resolved counts as binomial or multinomial outcomes with nuisance parameters for contrast, offset, and readout. A hidden synthetic EDM offset can be added before the final combination and removed only after cuts, reversals, systematic models, and unblinding criteria are frozen.

The final result is an estimate or interval for the measured EDM-like coefficient. Mapping it to $d_e$ uses the calculated $\mathcal E_{\mathrm{eff}}$ and its uncertainty. Mapping further to a particle-physics mass scale requires an explicit effective-operator model; the experiment does not exclude every source of CP violation or every beyond-standard-model theory.

References

• D. W. Allan, “Statistics of atomic frequency standards,” Proceedings of the IEEE 54, 221–230 (1966), doi:10.1109/PROC.1966.4634.
• D. Budker and M. Romalis, “Optical magnetometry,” Nature Physics 3, 227–234 (2007), doi:10.1038/nphys566.
• Bureau International des Poids et Mesures, The International System of Units (SI Brochure), 9th ed., 2026 update, doi:10.59161/AUEZ1291.
• Bureau International des Poids et Mesures, “Roadmap to the redefinition of the second,” accessed 2026-10-14, BIPM redefinition portal.
• A. D. Cronin, J. Schmiedmayer, and D. E. Pritchard, “Optics and interferometry with atoms and molecules,” Reviews of Modern Physics 81, 1051–1129 (2009), doi:10.1103/RevModPhys.81.1051.
• N. Dimarcq et al., “Roadmap towards the redefinition of the second,” Metrologia 61, 012001 (2024), doi:10.1088/1681-7575/ad17d2.
• W. M. Itano et al., “Quantum projection noise: population fluctuations in two-level systems,” Physical Review A 47, 3554–3570 (1993), doi:10.1103/PhysRevA.47.3554.
• Joint Committee for Guides in Metrology, Evaluation of measurement data—Guide to the expression of uncertainty in measurement, JCGM 100:2008, doi:10.59161/JCGM100-2008E.
• Joint Committee for Guides in Metrology, Guide to the expression of uncertainty in measurement—Part 1: Introduction, JCGM GUM-1:2023, doi:10.59161/JCGMGUM-1-2023.
• Joint Committee for Guides in Metrology, International Vocabulary of Metrology—Basic and General Concepts and Associated Terms, 3rd ed., JCGM 200:2012; definition 2.26, measurement uncertainty, accessed 2026-10-14.
• H. Katori, M. Takamoto, V. G. Pal’chikov, and V. D. Ovsiannikov, “Ultrastable optical clock with neutral atoms in an engineered light shift trap,” Physical Review Letters 91, 173005 (2003), doi:10.1103/PhysRevLett.91.173005.
• A. D. Ludlow, M. M. Boyd, J. Ye, E. Peik, and P. O. Schmidt, “Optical atomic clocks,” Reviews of Modern Physics 87, 637–701 (2015), doi:10.1103/RevModPhys.87.637.
• T. L. Nicholson et al., “Systematic evaluation of an atomic clock at $2\times10^{-18}$ total uncertainty,” Nature Communications 6, 6896 (2015), doi:10.1038/ncomms7896.
• L. Pezzè, A. Smerzi, M. K. Oberthaler, R. Schmied, and P. Treutlein, “Quantum metrology with nonclassical states of atomic ensembles,” Reviews of Modern Physics 90, 035005 (2018), doi:10.1103/RevModPhys.90.035005.
• N. F. Ramsey, “A molecular beam resonance method with separated oscillating fields,” Physical Review 78, 695–699 (1950), doi:10.1103/PhysRev.78.695.
• F. Riehle, Frequency Standards: Basics and Applications (Wiley-VCH, 2004), doi:10.1002/3527605991.
• W. J. Riley, Handbook of Frequency Stability Analysis, NIST Special Publication 1065 (2008), doi:10.6028/NIST.SP.1065.
• M. S. Safronova, D. Budker, D. DeMille, D. F. Jackson Kimball, A. Derevianko, and C. W. Clark, “Search for new physics with atoms and molecules,” Reviews of Modern Physics 90, 025008 (2018), doi:10.1103/RevModPhys.90.025008.
• G. Santarelli et al., “Frequency stability degradation of an oscillator slaved to a periodically interrogated atomic resonator,” IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control 45, 887–894 (1998), doi:10.1109/58.710548.
• D. J. Wineland, J. J. Bollinger, W. M. Itano, F. L. Moore, and D. J. Heinzen, “Spin squeezing and reduced quantum noise in spectroscopy,” Physical Review A 46, R6797–R6800 (1992), doi:10.1103/PhysRevA.46.R6797.
• D. J. Wineland, J. J. Bollinger, W. M. Itano, and D. J. Heinzen, “Squeezed atomic states and projection noise in spectroscopy,” Physical Review A 50, 67–88 (1994), doi:10.1103/PhysRevA.50.67.