Clock Atom Interferometry: Principles & Applications

• Clock atom interferometry is a matter-wave technique that uses superpositions of atomic clock states to measure inertial effects and proper-time differences.
• Experimental implementations range from free-space Mach-Zehnder setups to state-dependent traps, enabling gravimetry, gravitational-wave detection, and dark matter searches.
• Advances in large-momentum-transfer schemes, laser stabilization, and quantum optimal control are key to enhancing sensitivity and coherence in these precision experiments.

Searching arXiv for recent and foundational papers on clock atom interferometry to ground the article in published work. Clock atom interferometry denotes a class of matter-wave interferometric methods in which atomic wave packets are manipulated on clock transitions or with internal clock states, so that the interferometer phase encodes inertial effects, differential proper time, or both. In this literature, clock interferometry refers to the coherent splitting of a clock into two different paths and recombining in a way that reveals the proper time difference between them, while a genuine clock is a quantum system in a superposition of non-degenerate energy eigenstates with observable beat frequency $\nu_{\rm clock}=(E_2-E_1)/h$ (Meltzer et al., 2024, Sinha et al., 2011). Experimental realizations span single-photon interferometers on strontium clock transitions, trapped and guided interferometers with microwave or radio-frequency control, self-interfering clocks based on internal-state superpositions, and proposals aimed at tests of gravitational time dilation, local position invariance, gravitational-wave detection, and dark-matter searches (Hu et al., 2017, Margalit et al., 2015).

1. Conceptual basis

A central distinction in clock atom interferometry is between mass interferometry and clock interferometry. For a genuine clock, the atom must be prepared in a superposition of two energy eigenstates,

$$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$

so that the observable ticking is set by the beat frequency $\nu_{\rm clock}=(E_2-E_1)/h$ (Sinha et al., 2011). By contrast, an atom in a stationary state does not provide a periodic observable signal. This distinction underlies the statement that standard atom interferometers with atoms in stationary states do not test the gravitational redshift at the Compton frequency; in the formulation of Wolf and collaborators, “an atom is NOT a clock ticking at the Compton frequency” (Sinha et al., 2011).

This interpretive point is reinforced by analyses of atom-interferometer clocks that model phase evolution as being driven by proper-time differences between interferometer arms. In the non-relativistic limit, the observable phase evolution rate reduces to $mv^2/(2\hbar)$, while the overall Compton term is common to both arms and cancels from physical observables (Peil et al., 2014). The resulting conclusion is that atom-interferometer clock performance does not behave as if it were referenced to a divided-down physical oscillation at $\omega_0=mc^2/\hbar$; instead, the output is set by the relative phase evolution between interferometer arms (Peil et al., 2014).

Within genuine clock interferometry, the proper-time sensitivity is carried by the internal energy splitting. In the freely falling-clock framework, the clock phase is written as

$\delta\phi = -\frac{\Delta E}{\hbar}\Delta\tau,$

with $\Delta\tau$ the total proper-time difference including velocity and gravitational-potential contributions (Roura, 2024). In self-interfering-clock experiments and related proposals, the internal clock state can also carry which-path information. The corresponding fringe visibility is governed by the overlap of the internal states carried by the two arms; for the experimentally demonstrated self-interfering clock, the visibility follows $V=|\cos(\phi/2)|$ when a relative clock phase $\phi$ is imprinted between the arms (Margalit et al., 2015).

2. Experimental architectures

The experimental landscape includes free-space, guided, trapped, and internal-only architectures. A useful classification is given in the table below.

Platform	Representative implementation	Characteristic feature
Single-photon Mach-Zehnder on an optical clock transition	$^{88}$Sr on the $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$0 transition at $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$1 nm	Gravimeter and gravity gradiometer; no reduction of contrast up to $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$2 ms with fiber noise cancellation (Hu et al., 2017)
Large-momentum-transfer single-photon interferometer	$$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$3Sr on the $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$4 transition at $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$5 nm	Momentum separation up to $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$6 and gradiometers up to $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$7 (Rudolph et al., 2019)
State-dependent trapped clock-type interferometer	$$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$8Rb bi-chromatic adiabatic shell traps	Two independently controllable shell traps and coherence times up to $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$9 ms (Mas et al., 2019)
Fully confined or guided clock interferometer	RF Sagnac ring and magic-wavelength optical tweezers	State-dependent transport or tweezer splitting with Ramsey readout (Stevenson et al., 2015, Meltzer et al., 2024)

In free-space implementations, the optical field acts directly as beam splitter and mirror on a single-photon transition. The sequence is typically Mach-Zehnder type, $\nu_{\rm clock}=(E_2-E_1)/h$0–$\nu_{\rm clock}=(E_2-E_1)/h$1–$\nu_{\rm clock}=(E_2-E_1)/h$2, with the laser phase transferred at each pulse and the phase read out in the population imbalance of the two internal states (Hu et al., 2017). In confined implementations, by contrast, atoms can remain trapped at all times: state-dependent traps move the two internal states along distinct trajectories, or a trapped atom is split and recombined by adiabatic control of the confining potential (Stevenson et al., 2015, Meltzer et al., 2024).

A distinct but related line uses purely internal interference. The proposed internal atomic clock interferometer prepares a single atom in a superposition of two clock states and one ground state, producing Ramsey interference with a visibility modulation interpreted as the beating of two clock oscillations (Zhou, 2023). Since no splitting or recombining is involved, the coherence time can be as long as the trap lifetime or the clock state lifetime (Zhou, 2023).

3. Optical-clock-transition implementations

The first realization of a matter-wave interferometer based on single-photon interaction on the ultra-narrow optical clock transition of strontium used bosonic $\nu_{\rm clock}=(E_2-E_1)/h$3Sr on the magnetically induced $\nu_{\rm clock}=(E_2-E_1)/h$4 transition at $\nu_{\rm clock}=(E_2-E_1)/h$5 nm (Hu et al., 2017). The experiment employed a two-stage MOT yielding $\nu_{\rm clock}=(E_2-E_1)/h$6 atoms cooled to $\nu_{\rm clock}=(E_2-E_1)/h$7, vertical launch in a fountain using frequency-ramped optical lattices at $\nu_{\rm clock}=(E_2-E_1)/h$8 nm, and a Mach-Zehnder pulse sequence on the clock transition (Hu et al., 2017). With active fiber noise cancellation of the $\nu_{\rm clock}=(E_2-E_1)/h$9 m optical fiber, no reduction of interferometric contrast was observed up to $mv^2/(2\hbar)$0 ms, limited by geometric constraints of the apparatus, and the observed contrast reached up to $mv^2/(2\hbar)$1 for $mv^2/(2\hbar)$2 ms (Hu et al., 2017). In the gradiometric configuration, the sensitivity approached the shot noise limit, and the Allan deviation at $mv^2/(2\hbar)$3 ms gave a relative phase sensitivity at $mv^2/(2\hbar)$4 s averaging of $mv^2/(2\hbar)$5 mrad (Hu et al., 2017).

A subsequent characterization of a gravimeter and gravity gradiometer based on the same $mv^2/(2\hbar)$6–$mv^2/(2\hbar)$7 transition reported a relative sensitivity of $mv^2/(2\hbar)$8 for the gravitational acceleration and a differential phase sensitivity of $mv^2/(2\hbar)$9 in a gravity gradiometer configuration at an artificially introduced differential phase of $\omega_0=mc^2/\hbar$0 rad (Hu et al., 2019). The total interferometry time reached $\omega_0=mc^2/\hbar$1 ms, longer than previously reported for such interferometers, and the primary sensitivity limitation was identified as the intrinsic noise of the interferometry laser itself (Hu et al., 2019). The corresponding Mach-Zehnder phase expression was written as

$\omega_0=mc^2/\hbar$2

making explicit the coupling of the clock transition frequency, gravity, chirp rate, pulse duration, and laser phases (Hu et al., 2019).

Large-momentum-transfer clock atom interferometry was first realized on the $\omega_0=mc^2/\hbar$3 nm $\omega_0=mc^2/\hbar$4 intercombination line of strontium (Rudolph et al., 2019). Using single-photon interactions, the experiment demonstrated Mach-Zehnder interferometers with momentum separation up to $\omega_0=mc^2/\hbar$5 and gradiometers of up to $\omega_0=mc^2/\hbar$6 (Rudolph et al., 2019). It also circumvented excited-state decay limitations and extended the gradiometer duration to $\omega_0=mc^2/\hbar$7 times the excited state lifetime by storing the atomic population in the ground state during the waiting time through velocity-selective pulses (Rudolph et al., 2019). Because the interferometry pulses had broad velocity acceptance, all experiments were performed with laser-cooled atoms at a temperature of $\omega_0=mc^2/\hbar$8 rather than with ultracold or quantum-degenerate gases (Rudolph et al., 2019).

Clock interferometric principles also appear in compact optical-clock development based on Ramsey-Bordé atom interferometry with a thermal strontium beam (Fartmann et al., 2024). In that system, the narrow $\omega_0=mc^2/\hbar$9 intercombination line at $\delta\phi = -\frac{\Delta E}{\hbar}\Delta\tau,$0 nm yielded a $\delta\phi = -\frac{\Delta E}{\hbar}\Delta\tau,$1 kHz broad spectral feature, and analysis of the Ramsey fringe slope together with the fluorescence detection noise yielded an estimated short-term stability of $\delta\phi = -\frac{\Delta E}{\hbar}\Delta\tau,$2 (Fartmann et al., 2024). This implementation was presented as a ground testbed for future clock systems in mobile and space applications (Fartmann et al., 2024).

4. Trapped, guided, and state-dependent clock interferometers

Clock atom interferometry is not restricted to free-fall light-pulse geometries. In bi-chromatic adiabatic magnetic shell traps, two strong RF fields dress the $\delta\phi = -\frac{\Delta E}{\hbar}\Delta\tau,$3 and $\delta\phi = -\frac{\Delta E}{\hbar}\Delta\tau,$4 states of rubidium Bose-Einstein condensates, creating two independently controllable shell traps (Mas et al., 2019). Matching of the shell positions and axial frequencies allows microwave pulses near the hyperfine splitting to act as a “clock” beam splitter between the dressed states, thereby creating a state-dependent clock-type interferometer in a fully trapped geometry (Mas et al., 2019). Under optimal matching, the microwave transition linewidth reached $\delta\phi = -\frac{\Delta E}{\hbar}\Delta\tau,$5 Hz for a $\delta\phi = -\frac{\Delta E}{\hbar}\Delta\tau,$6 ms pulse, coherence times reached $\delta\phi = -\frac{\Delta E}{\hbar}\Delta\tau,$7 ms, and the radial trap frequency could be as low as $\delta\phi = -\frac{\Delta E}{\hbar}\Delta\tau,$8–$\delta\phi = -\frac{\Delta E}{\hbar}\Delta\tau,$9 Hz, enabling the atoms to expand into a $\Delta\tau$0D sheet for direct imaging interferometry (Mas et al., 2019). The same work emphasizes robustness to homogeneous DC magnetic-field fluctuations and operation without magnetic shielding (Mas et al., 2019).

A different fully confined architecture is the Sagnac interferometer based on superpositions of fully confined atoms and state-dependent transport along a closed path (Stevenson et al., 2015). In this protocol, atoms in two internal clock states are held in separate traps and transported in opposite directions around a ring, with Ramsey pulses mapping the accumulated Sagnac phase into the resulting population imbalance between the two internal states (Stevenson et al., 2015). The Sagnac phase is written as

$\Delta\tau$1

and the signal takes the form $\Delta\tau$2 (Stevenson et al., 2015). The analysis quantifies limitations arising from motional excitation and finite temperature, and discusses an implementation with adiabatic radio-frequency potentials that is inherently robust against common-mode noise and phase noise from the reference oscillator (Stevenson et al., 2015).

Optical trapping offers another route. An optically trapped Ramsey-type atom interferometer with $\Delta\tau$3 Bose-condensed $\Delta\tau$4Rb atoms used the $\Delta\tau$5 microwave clock transition, with interference fringes of contrast approaching $\Delta\tau$6 observed for short evolution times (Altin et al., 2010). With smaller samples of $\Delta\tau$7 atoms, a coherence time of $\Delta\tau$8 s was observed (Altin et al., 2010). Numerical simulations of the Gross-Pitaevskii equations showed that dephasing due to spatial dynamics driven by interparticle interactions accounts for much of the observed decay in fringe visibility at long interrogation times (Altin et al., 2010).

A more recent proposal uses optical tweezers at the magic wavelength to implement clock interferometry with an alkaline-earth-like atom held in an optical trap (Meltzer et al., 2024). By combining adiabatic splitting and recombining schemes with a modified Ramsey sequence on the clock states, the proposed interferometer achieves a linear sensitivity to the gravitational time dilation and is insensitive to relative fluctuations in the intensity of the tweezer beams (Meltzer et al., 2024). For the ground-state output probability, the paper gives

$\Delta\tau$9

with $V=|\cos(\phi/2)|$0, so that for small $V=|\cos(\phi/2)|$1 the visibility scales linearly with the gravitational redshift (Meltzer et al., 2024).

5. Relativistic time, complementarity, and equivalence-principle tests

The relativistic motivation for clock atom interferometry is that internal-state phase accumulation follows proper time. For weak gravitational fields and nonrelativistic velocities, one proposed freely falling clock scheme writes the proper-time increment as

$V=|\cos(\phi/2)|$2

and derives a measurable phase in a light-pulse interferometer based on single-photon clock transitions (Roura, 2024). The proposed measurement is implementable with no additional requirements in Fermilab’s MAGIS-100 experiment or in the corresponding $V=|\cos(\phi/2)|$3-m prototypes, and for MAGIS-100 standard parameters the relativistic phase is stated to be approximately $V=|\cos(\phi/2)|$4 rad, measurable to the $V=|\cos(\phi/2)|$5 level in about $V=|\cos(\phi/2)|$6 shots (Roura, 2024).

Tests of local position invariance and the universality of clock rates motivate specific clock-interferometric geometries. One proposal extends such tests to atom interferometry generating delocalized quantum clocks and finds a favorable $V=|\cos(\phi/2)|$7 scaling (Pumpo et al., 2022). In the recoilless case the interferometer phase is

$V=|\cos(\phi/2)|$8

where $V=|\cos(\phi/2)|$9 parameterizes a violation of local position invariance (Pumpo et al., 2022). The same work emphasizes that the proposed test is robust against initial conditions and recoil effects and enables optical frequencies, with projected sensitivity exceeding that of state-of-the-art localized clocks (Pumpo et al., 2022).

Internal-only schemes probe the same conceptual territory from a different angle. The proposed internal atomic clock interferometer uses two clock states and a shared ground state, producing a Ramsey population

$\phi$0

with a visibility envelope $\phi$1 (Zhou, 2023). The visibility modulation is interpreted as the beating of the individual clock oscillations and as a direct consequence of complementarity; under a different gravitational potential, local position invariance predicts that the modulation changes accordingly (Zhou, 2023).

The complementarity aspect has already been demonstrated experimentally in the self-interfering-clock experiment with $\phi$2Rb (Margalit et al., 2015). There, a clock implemented as a superposition of two spin states was split into two spatially separated wave packets and then made to tick at different rates, simulating a proper-time lag (Margalit et al., 2015). The entanglement between the clock’s time and its path yielded which-path information and a controllable reduction of visibility, with complete suppression when the clock states became orthogonal (Margalit et al., 2015). In this sense, time becomes an internal witness of path distinguishability rather than merely a phase parameter.

6. Large momentum transfer, control, and technical limits

Large-momentum-transfer enhancement is a major theme because many proposed long-baseline experiments require momentum separations beyond $\phi$3 (Chiarotti et al., 2022). A practical analysis of sequential $\phi$4 pulses on single-photon clock transitions showed that laser frequency noise places stringent requirements on the interferometry laser linewidth, even for an atom at rest interacting with resonant light (Chiarotti et al., 2022). In the presence of white frequency noise, the operational fidelity is written as

$\phi$5

and for $\phi$6 with $\phi$7 the linewidth must be well below $\phi$8 Hz to keep the sequence fidelity significantly above $\phi$9 (Chiarotti et al., 2022). The same study presented high-power, frequency-stabilized laser sources for the $^{88}$0–$^{88}$1 clock transitions of cadmium and strontium, with the strontium source delivering about $^{88}$2 W at $^{88}$3 nm and the cadmium source up to about $^{88}$4 W at $^{88}$5 nm (Chiarotti et al., 2022).

A later analysis revisited the same question for alternating-direction LMT sequences and found a different cumulative scaling (Jiang et al., 14 Aug 2025). In that treatment, the population error from $^{88}$6 pulses scales linearly with $^{88}$7, not as $^{88}$8, because errors left behind after each pulse are Doppler-detuned from the next pulse and do not continue to accumulate coherently (Jiang et al., 14 Aug 2025). The paper states that for a realistic laser-noise spectrum with RMS noise amplitude $^{88}$9 Hz and $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$00 kHz, the fidelity after $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$01 LMT pulses remains above $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$02, and that parasitic-path contributions are negligible for any loss mechanism (Jiang et al., 14 Aug 2025). Taken together, these two analyses indicate an active refinement of the laser-noise limit rather than a settled single-number bound.

Beyond linewidth reduction, the control problem has been attacked with quantum optimal control. For strontium clock interferometry, gradient-based optimized pulse waveforms were compared with primitive square pulses and composite pulses in full interferometer simulations with up to $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$03 pulses (Chen et al., 2022). The optimized pulses showed order-of-magnitude lower infidelity across polarization, detuning, amplitude, and magnetic-field errors, maintained transfer efficiency above $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$04, and reduced phase errors to the mrad-per-pulse scale under realistic noise assumptions (Chen et al., 2022). The same study defined the interferometric contrast as $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$05 for the ensemble-averaged output (Chen et al., 2022).

Laser engineering for clock atom interferometry has therefore become a subject in its own right. A dedicated strontium laser system coherently combined the power of two Titanium:Sapphire lasers, demonstrated chirps of $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$06 MHz in $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$07 ms while phase-locked to an optical reference, and delivered $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$08 W pulsed beams to the atoms via a mode-cleaning optical fiber with active noise cancellation (DeRose et al., 2022). The development directly targeted single-photon interferometry on the $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$09 nm clock transition in ultracold strontium (DeRose et al., 2022).

7. Applications and emerging directions

The immediate applications of clock atom interferometry are precision inertial sensing and metrology. Experiments have already demonstrated operation as gravimeters and gravity gradiometers on optical clock transitions (Hu et al., 2017, Hu et al., 2019). The bi-chromatic shell-trap platform is described as sensitive to spatially varying electric or magnetic fields, which could be DC, AC, RF fields or microwaves, or even local variations in gravity, and the low horizontal confinement enables direct imaging interferometry in a $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$10D sheet geometry (Mas et al., 2019). Compact Ramsey-Bordé implementations with thermal strontium beams target higher stability than optical vapour cell clocks while remaining less complex than cold atomic clocks, with explicit motivation from mobile and space applications (Fartmann et al., 2024).

Long-baseline fundamental-physics applications recur throughout the literature. Single-photon interferometers on optical clock transitions are presented as a new class of high-precision sensors for gravitational-wave detection in so far unexplored frequency ranges and for ultralight dark-matter searches (Hu et al., 2017). Large-momentum-transfer clock interferometry is treated as a key ingredient in proposals for gravitational-wave detection and dark-matter searches, and the move to narrower transitions such as $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$11 is explicitly motivated by the prospect of many thousands of pulses with negligible lifetime loss (Rudolph et al., 2019). The freely falling-clock proposal and the $$|\psi(t)\rangle = e^{-i\frac{E_1 t}{\hbar}}|\psi_1\rangle + e^{-i\frac{E_2 t}{\hbar}}|\psi_2\rangle,$$12 local-position-invariance proposal both frame clock atom interferometry as a route to direct measurements of gravitational time dilation with quantum systems (Roura, 2024, Pumpo et al., 2022).

Proposed extensions now reach beyond single instruments. A programmable quantum sensing network based on entangled atomic ensembles uses optical clock qubits to emulate mass superpositions in atom and atom-clock interferometry, with a non-local Ramsey interferometer as the basic sensing primitive (Fromonteil et al., 23 Sep 2025). In that architecture, Bell-type seed states distributed via photonic channels are amplified by collective operations into non-local and local superpositions, and gravitationally induced phase shifts appear in network-based interference patterns (Fromonteil et al., 23 Sep 2025). This suggests that clock atom interferometry is evolving from single-device precision measurement toward distributed architectures in which clock states, entanglement, and spacetime sensitivity are co-designed.