PHYSICAL REVIEW A 89, 014101 (2014)

Analysis of atom-interferometer clocks

Steven Peil* and Christopher R. Ekstrom
United States Naval Observatory, Washington, DC 20392, USA
(Received 20 November 2013; published 21 January 2014)
We analyze the nature and performance of clocks formed by stabilizing an oscillator to the phase difference
between two paths of an atom interferometer. The phase evolution has been modeled as being driven by the
proper-time difference between the two paths, although it has an ambiguous origin in the nonrelativistic limit
and it requires a full quantum-field-theory treatment in the general case. We present conditions for identifying
deviations from the nonrelativistic limit as a way of testing the proper-time-driven phase evolution model.
We show that the system performance belies the premise that an atom-interferometer clock is referenced to
a divided-down Compton oscillation, and we suggest that this implies there is no physical oscillation at the
Compton frequency.
DOI: 10.1103/PhysRevA.89.014101

PACS number(s): 03.65.w, 06.30.Ft, 03.75.Gg

I. INTRODUCTION

In a recent paper, Lan et al. [1] claim to have created

a Compton clock, an oscillator with frequency stabilized
to a particles mass via a purported intrinsic oscillation at
the Compton frequency, 0 = mc2 /. The ultimate limits
of such a clock would far surpass any current or proposed
traditional atomic clock; the experiment used 133 Cs, for which
0 /2 = 3 1025 Hz, about 10 orders of magnitude higher
than the frequency of optical atomic clocks, the current state
of the art.
Whether 0 corresponds to a physical oscillation is a matter
of debate. De Broglie first associated a massive particle with
an oscillation at frequency 0 in developing his wave theory of
matter, and it was integral to his pilot-wave theory of quantum
mechanics [2]. Despite the success of the theory of matter
waves, the idea of an internal clock for a massive particle has
been mostly ignored [3]. But in recent years it has appeared
in the literature with claims that the redshift of the Compton
oscillation frequency is measured in an atomic gravimeter [4],
as well as with a search for evidence of an internal oscillation
in the electron in channeling experiments [5]. In a recent book,
Penrose claims that a stable, massive particle behaves as a very
precise quantum clock, oscillating at 0 /(2 ) [6].
Yet the overall phase of a quantum state, to which the
Compton clock would fundamentally correspond, is widely
accepted to be unobservable, while it is the relative phase
between two states that corresponds to a physical oscillation.
This is manifest in the squaring of the state vector characterizing the system to generate the probability distribution for
measurement outcome, which at the same time does away
with the overall phase. In recent literature it has been claimed
that 0 does not correspond to a physical oscillation, with
the points being made that it is not Doppler shifted to an
observer in motion [7] and that an atom interferometer is a
single accelerometer and not two clocks [8].
What seems to be accepted is that, in nonrelativistic
quantum mechanics, the Compton frequency is the rate of
accumulation of the quantum phase with respect to proper time
for a system. In an atom-interferometer clock, the difference

[email protected]

1050-2947/2014/89(1)/014101(4)

014101-1

of quantum phases of two motional states of an atom is used as

a frequency reference [1]. Here we analyze the source of phase
evolution and frequency stability in this system, including the
role of the Compton frequency and purported oscillation.
II. PHASE EVOLUTION IN AN ATOM INTERFEROMETER

Over the past two decades atom interferometers have

developed into high-precision tools for inertial measurements
[9,10], tests of the foundations of general relativity [11],
and determination of the ratio h/m of Plancks constant to
atomic mass [12,13]. Laser-atom interactions used for the
beam-splitting process produce superpositions of atomic states
with different velocities that travel along two physically displaced paths and are subsequently recombined. The resulting
interference reveals the difference in phase accumulated along
each path.
The free-propagation phase factor for an atom with velocity
v can be written as
ei = ei0 = ei0 t/ ,

(1)


where is the proper time for the atom and = 1/ 1 v 2 /c2
is the Lorentz factor. This can be seen from the path-integral
formulation of quantum mechanics [1] or by substituting relativistic momentum and energy expressions into the phase factor for a massive-particle wave packet [7]. The assumption in
each case is a well-defined (classical) trajectory characterizing
the particles motion; a full quantum-field-theoretic treatment
is required when the system is sensitive to relativistic effects
as a result of sufficiently high velocity and/or precision. In
this semiclassical limit, the free-propagation phase difference
between the two paths in the interferometer evolves at a rate
given by
= 0 ( 1 1)

(2)

for the case where the second path corresponds to zero velocity
in the laboratory frame. This can be treated as the frequency
of a two-state system formed by the motional states in the
interferometer.
When v  c, the Lorentz factor can be expanded as 1
1 (v 2 /2c2 ) (v 4 /8c4 ) + . To lowest order in v/c, the
Published by the American Physical Society

Form Approved
OMB No. 0704-0188

Report Documentation Page

Public reporting burden for the collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and
maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information,
including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington
VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to a penalty for failing to comply with a collection of information if it
does not display a currently valid OMB control number.

1. REPORT DATE

3. DATES COVERED
2. REPORT TYPE

JAN 2014

00-00-2014 to 00-00-2014

4. TITLE AND SUBTITLE

5a. CONTRACT NUMBER

Analysis of atom-interferometer clocks

5b. GRANT NUMBER

5c. PROGRAM ELEMENT NUMBER

6. AUTHOR(S)

5d. PROJECT NUMBER

5e. TASK NUMBER
5f. WORK UNIT NUMBER

7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES)

8. PERFORMING ORGANIZATION
REPORT NUMBER

United States Naval Observatory,Washington,DC,20392

9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES)

10. SPONSOR/MONITORS ACRONYM(S)

11. SPONSOR/MONITORS REPORT
NUMBER(S)

12. DISTRIBUTION/AVAILABILITY STATEMENT

Approved for public release; distribution unlimited

13. SUPPLEMENTARY NOTES

PHYSICAL REVIEW A 89, 014101 (2014)

14. ABSTRACT

15. SUBJECT TERMS

16. SECURITY CLASSIFICATION OF:
a. REPORT

b. ABSTRACT

c. THIS PAGE

unclassified

unclassified

unclassified

17. LIMITATION OF
ABSTRACT

18. NUMBER
OF PAGES

Same as
Report (SAR)

19a. NAME OF
RESPONSIBLE PERSON

Standard Form 298 (Rev. 8-98)

Prescribed by ANSI Std Z39-18

BRIEF REPORTS

PHYSICAL REVIEW A 89, 014101 (2014)

above phase factor reduces to

ei = ei0 t/ ei0 t eimv

t/2

(3)

In this case the Compton frequency comes in only through

an overall phase common to each path, which cancels when
phase differences are considered. For the remaining factor, the
phase evolves at a rate determined by the atoms kinetic energy,
and the rate at which the relative phase in the interferometer
evolves is just mv 2 /(2). There is no longer a factor of 0 in
the phase evolution of the interferometer, and no relativistic
proper time is involved.
Of course, this phase evolution rate can be written as
0 v 2 /(2c2 ), which has been called the time-dilated Compton
frequency [1]. But the same term can also be written as dB /2,
where dB is the (angular) de Broglie wave frequency, the
frequency associated with the physical oscillation of a matter
wave with momentum mv. In the low-velocity limit, then,
the phase evolution cannot unambiguously be attributed to
the proper-time difference since the rate is indistinguishable
from that driven by kinetic energy. These two origins of phase
evolution give differing rates for arbitrary velocities, where
formally a full quantum-field-theoretic formalism is required,
with Eq. (2) quantifying the rate for the proper-time-driven
model and the expression for a kinetic-energy-driven phase
evolution differing from this by the sign of the exponent of
[14].
If an interferometer were sensitive enough to resolve the
next term in the expansion of 1 , the form of the phase
evolution rate could be empirically tested. This term, (v 4 /8c4 ),
gives a contribution to the rate of phase evolution of
= 0 (v 4 /8c4 ),

(4)

or as a fractional frequency,
/ = v 4 /[8c4 ( 1 1)],

(5)

which reduces to v 2 /4c2 to lowest order in v/c. Resolving this

term in an interferometer would be evidence of quantum phase
evolution due to relativistic proper time.
III. COMPARISON TO ATOMIC CLOCKS

In order to assess the precision with which an atom

interferometer can measure the evolution of the relative phase
between the two paths, we develop an analogy to conventional
atomic clocks.
Conventional passive atomic clocks operate by stabilizing
the frequency of a local oscillator to an atomic resonance.
Atomic states with well-defined internal energies make up a
two-state system that serves as the frequency reference for the
local oscillator, and the achievable signal-to-noise ratio and
quality factor Q of the resonance dictate clock performance.
The signal-to-noise ratio is determined by the statistics of
measuring the quantum states of the number of atoms in
the system Na , and Q = f/f , the ratio of frequency to
linewidth, which is often measurement-time limited. In the
standard quantum limit (SQL), the precision of the frequency
reference improves with averaging time as

( )1 = Q Na ,
(6)

FIG. 1. (Color online) Demonstration of conditions for an atom

interferometer to be sensitive to relativistic effects. For the parameters
discussed in the text, the plot shows the fractional frequency /,
representing the first-order correction to the low-velocity limit for the
phase evolution rate according to the proper-time model (blue curve,
right axis), and the averaging time required to resolve this frequency
(red curve, left axis) as a function of = v/c. For the system in
Ref. [1], = 100 1012 .

where is the Allan deviation used for characterizing clocks

and oscillators [15]. This gives the averaging time f required
for a clock to resolve a fractional frequency f/f as

f = [Q Na (f/f )]2 .
(7)
In an atom interferometer, the velocity states make up
the two-state system serving as the frequency reference. The
quality factor is Q = 0 ( 1 1)/, where the linewidth is
determined by the interaction (Ramsey) time T , /(2 )
1/2T
. As an example, we consider a signal-to-noise ratio
Na of 1000, corresponding to a sample size of the order
of 106 atoms, and a linewidth /(2 ) = 1.5 Hz resulting
from a 0.32-s measurement time. For these values, we plot in
Fig. 1 the time required to resolve the frequency difference
/, from Eq. (5), as a function of = v/c, assuming
the SQL for all averaging times. From the plot we see that
at 870 1012 (v = 0.26 m/s), phase evolution due to
relativistic proper time can be resolved with about 1 hour
of averaging. For cesium, this velocity corresponds to about
70 photon recoils [16], which can be reached with state-ofthe-art beam-splitting techniques, so testing phase evolution
due to relativistic proper time beyond the nonrelativistic limit
should be possible with current technology [17].
The expected SQL for the system in Ref. [1] can be
determined from the experimental parameters. The velocity
difference between atoms in the two paths of the interferometer
is due to the recoil from 10 photons, giving v 0.035 m/s
and /(2 ) 200 kHz. The 320-ms measurement time and
106 atoms contributing to the interferometer signal correspond
to /(2 ) 1.5 Hz and a signal-to-noise ratio of 1000.
So the expected performance for this system is characterized
by an Allan deviation versus averaging time of ( ) =
2.5 109 / . This is consistent with the standard-quantumlimited performance referred to in [1], and it corresponds to
an averaging time of more than 7 1014 s, or 2 107 years,
to test proper-time-driven phase evolution, as can be seen from
Fig. 1. The performance actually demonstrated was two orders
of magnitude worse than the SQL, making the averaging
time required 7 1018 s, or 2 1011 years (more than
10 times the age of the universe). So that interferometer was

014101-2

BRIEF REPORTS

PHYSICAL REVIEW A 89, 014101 (2014)

not sensitive to relativistic effects, and the phase evolution

was indistinguishable from that driven by the difference in
kinetic energy due to absorption and emission of photons in
the beam-splitting process.
We can also assess the potential performance of interferometer clocks in comparison to conventional atomic
clocks. In most state-of-the-art atomic clocks the linewidth
is measurement time limited, so comparing Qs comes down
to comparing frequencies. For microwave (fountain) clocks,
the number of atoms will be comparable to an interferometer
system, so the difference in quantum-limited performance is
due to only the two frequencies. For an interferometer clock
to achieve stability comparable to a microwave clock with
angular frequency , it is required that 0 ( 1 1) .
For cesium, 0 /(2 ) = 3 1025 Hz and /(2 ) = 9.2
109 Hz; it is necessary for ( 1 1) v 2 /(2c2 ) to be on the
order of 3 1016 , giving v/c 2.5 108 and v 7.4 m/s.
Beam splitting for atom interferometers can produce relative
velocities of roughly 100 photon-recoil velocities, or about
0.3 m/s for cesium, and it may be possible to extend this
another order of magnitude [17]. This potential improvement
could make the quantum-limited performance similar to
microwave clocks. In order to achieve atom-interferometer
performance comparable to optical clocks, another several orders of magnitude in frequency would be required. Continuing
to use the nonrelativistic approximation, it would be necessary
to generate velocities v 2 103 m/s to achieve a frequency
of 5 1014 Hz. This is well beyond current technology; in addition, for a reasonably sized interaction region, such high velocities limit the measurement time and reduce the effective Q.
IV. ATOM-INTERFEROMETER CLOCK
AND COMPTON OSCILLATION

In Ref. [1], the velocity states in an atom interferometer

were used as a two-state frequency reference for a clock. If
the resulting clock were referenced to the purported Compton
oscillation, it would suggest the promise of a reference with a
frequency of 1025 Hz and correspondingly high Q.
The total phase determining the interferometer signal is
the combination of the free propagation phase, discussed
above, and the laser-atom phase from the beam-splitting
interactions. The interferometer in Ref. [1] was turned into
a clock by feeding back to the laser frequencies used in
the experiment to adjust the laser-atom interaction and keep
the total phase zero. Adjusting the laser frequencies would
normally modify the reference frequency via the effect on the
atoms velocity, which is entirely due to photon recoil. An
arrangement was used to decouple the atoms velocity from
the raw laser frequency L by imposing L = N m , where
m = + is the difference in frequencies of the beams
used to impart velocity to the atoms and N is not necessarily
an integer. The atomic velocity then depends not on L but
on N, which is determined by a set of frequency multipliers.
The physical output of the clock comes from adjusting an
oscillator that determines m (and L ) to keep the total phase
of the interferometer constant (equal to zero). According to
Ref. [1], when total = 0, m = 0 /(2nN 2 ).
In the previous section we analyzed the performance of an
atom-interferometer clock based on the SQL for the two-state

system comprised of different motional states separated in

phase evolution rate by . The SQL for a true Compton clock
would be governed by the frequency of the purported Compton
oscillation, 1020 times higher. This same performance applies
even to a system referenced to a divided-down Compton
oscillation, as claimed for the interferometer in Ref. [1].
This dividing down is reminiscent of optical clocks, where
the optical frequency is divided down to an electronically
countable microwave frequency. The performance of the
system is still determined by the optical frequency; the high stability associated with the optical signal is transferred without
degradation to the lower-frequency oscillation. This is simply a
consequence of the propagation of uncertainties, which shows
that dividing down the frequency of an oscillator does not
change the relative uncertainty, i.e., stability, of the signal [18].
If the frequency of the interferometer clock were divided
down from the frequency of a physical oscillation at 0 in the
same way, the stability of the clock should be the same as that
oscillation [19],
/ = 0 /0 .

(8)

The observation that is derived from the difference of two

frequencies, the phase evolution rates of the arms of the
interferometer, each of which could be considered a divideddown Compton frequency, does not change this. Because the
two rates are both fractions of the same higher frequency, the
correlations introduce a nonzero covariance, with uncertainty
propagation again showing the final stability to be the same
as that of the oscillation at 0 [20]. On the other hand, if the
covariance between the two frequencies is zero, indicating no
correlation, the derived stability agrees with that calculated in
the previous section.
The only qualification to this argument is that the factor
in multiplying (dividing down) the Compton frequency,
= 1 1, has an associated uncertainty that needs to be
included in the error propagation and that would limit the
clock performance in this model. This technical-noise limit
should be independent of any standard quantum limit, so the
predictions of clock performance from a Compton-oscillation
model can still be differentiated from a clock governed
by the properties of the two-velocity-state system, and the
empirical stability provides a means of testing the physical
reality of the Compton oscillation. We can try to assess the
technical-noise limit in Ref. [1] from experimental values
provided. In the low-velocity limit, uncertainty propagation
from the Compton oscillation to the clock output includes
the term / = 2v/v = 4N/N. An uncertainty on N is
not stated, but from the number of significant digits it can be
inferred that N/N 1012 [21]. The Compton oscillation
model therefore predicts a technical-noise-limited stability no
worse than 1012 , significantly better than that predicted by
the two-velocity-state SQL and better than that demonstrated.
So the system in [1], despite being designed to operate
as a divided-down Compton clock, empirically operates as a
clock with no connection to the Q of the purported Compton
oscillation. The Compton oscillations for an atom in the two
paths of the interferometer behave as if they are completely
uncorrelated, even though they are affiliated with the same
particle. This evidence strongly suggests that the frequency
0 does not correspond to a physical oscillation. In the

014101-3

BRIEF REPORTS

PHYSICAL REVIEW A 89, 014101 (2014)

absence of a physical oscillation, what remains is the quantum

phase, which is just a product of the systems proper time
and the parameter 0 . The phase still accumulates at a high
rate (compared to frequencies that we are accustomed to for
atomic clocks) due to the large value of 0 , but this does
not contribute to making a clock until a relative phase is
generated. In that case, for common velocities, the rest-mass
contribution mostly cancels, and the resulting frequency is not
exceptional [22].
Finally, we point out that, if these conclusions stand, there
may be an impact on past analysis of an atom-interferometric
measurement of the gravitational redshift and a corresponding
test of local position invariance (LPI). In Ref. [4], a measurement of g was reinterpreted as a measurement of the
gravitational redshift of the Compton frequency and a test
of the dependence of the redshift on gravitational potential U .
If the notion of a physical Compton oscillation is abandoned,
the implications for these results is unclear. The nature of the
measurement changes from a measurement of the redshift (of
the frequency of a physical oscillation) to a measurement of
time dilation (due to the geometry of space-time). In particular,
it is not obvious that the anomalous redshift parameter used

in [4] would enter in the same way. This may be a topic for
future consideration.

[1] S. Y. Lan, P. C. Kuan, B. Estey, D. English, J. M. Brown, M. A.

Hohensee, and H. Muller, Science 339, 554 (2013).
[2] L. de Broglie as translated by J. W. Haslett, Am. J. Phys. 40, 1315
(1972).
[3] D. J. Bohm and B. J. Hiley, Found. Phys. 12, 1001 (1982).
[4] H. Muller, A. Peters, and S. Chu, Nature (London) 463, 926
(2010).
[5] P. Catillon, N. Cue, M. J. Gaillard, R. Genre, M. Gouan`ere, R. G.
Kirsch, J. C. Poizat, J. Remillieux, L. Roussel, and M. Spighel,
Found. Phys. 38, 659 (2008).
[6] R. Penrose, Cycles of Time: An Extraordinary View of the
Universe (Knopf, New York, 2011).
[7] D. M. Greenberger, W. P. Schleich, and E. M. Rasel, Phys. Rev.
A 86, 063622 (2012).
[8] S. Sinha and J. Samuel, Classical Quantum Gravity 28, 145017
(2011).
[9] T. L. Gustavson, P. Bouyer, and M. A. Kasevich, Phys. Rev. Lett.
78, 2046 (1997).
[10] M. J. Snadden, J. M. McGuirk, P. Bouyer, K. G. Haritos, and
M. A. Kasevich, Phys. Rev. Lett. 81, 971 (1998).
[11] S. Fray, C. A. Diez, T. W. Hansch, and M. Weitz, Phys. Rev.
Lett. 93, 240404 (2004).
[12] D. S. Weiss, B. C. Young, and S. Chu, Phys. Rev. Lett. 70, 2706
(1993).
[13] R. Bouchendira, P. Clade, S. Guellati-Khelifa, F. Nez, and F.
Biraben, Phys. Rev. Lett. 106, 080801 (2011).
[14] Considering the difference in these two phase evolution rates
was instrumental in the original development of de Broglies
matter-wave theory.
[15] D. A. Howe, D. W. Allan, and J. A. Barnes, Proceedings of
the 35th Annual Symposium on Frequency Control (Electronic
Industries Association, Washington, DC, 1981), pp. A1A44.
[16] A photon recoil is the recoil velocity of an atom from
absorbing or emitting a single photon.

[17] S. W. Chiow, T. Kovachy, H.-C. Chien, and M. A. Kasevich,

Phys. Rev. Lett. 107, 130403 (2011).
[18] If is the lower microwave frequency signal, opt is the optical
frequency that is stabilized, and M is a large integer (usually on
the order of 105 ) relating the two, = opt /M, then =
opt /M, and / = opt /(M ) = opt /opt .
[19] corresponds to the uncertainty associated with the clock
output, as opposed to the linewidth, .
[20] From Eq. (2) we have = 1 0 0 . Uncertainty propagation for the absolute uncertainties when subtracting two values
gives ()2 = ( 1 0 )2 + (0 )2 2 1 0,0 , where 0,0 is the
covariance of 0 with itself, i.e., (0 )2 . The expression for
the absolute uncertainty reduces to = 0 ( 1 1), and the
relative uncertainty is then / = 0 ( 1 1)/[0 ( 1
1)] = 0 /0 . Propagation of uncertainties shows that not only
does dividing down the frequency of an oscillator not affect the
stability but differencing two frequencies divided down from the
same oscillator does not affect the stability; in each case, the final
signal should exhibit the same stability as the high-frequency
oscillation.
[21] This is using the lower precision included in the Supplemental
Material; the higher precision quoted in [1] gives a more
stringent N/N 1014 .
[22] At first glance, it seems that the system could exhibit a
performance consistent with the notion of a Compton clock
for large enough velocities to make 0 , which requires
v c. But, in addition to being an unrealistic limit, the
nonrelativistic formalism used to derive the phase evolution
in the interferometer is not valid for relativistic velocities
and must be replaced with a full quantum-field-theoretic
treatment in that case [23], as we have emphasized all
along.
[23] R. Feynman and A. Hibbs, Quantum Mechanics and Path
Integrals (McGraw-Hill, New York, 1965).

V. CONCLUSIONS

We have analyzed the nature and performance of clocks

based on atom interferometers. We have presented the conditions required to test proper-time-driven phase evolution for
general conditions beyond the nonrelativistic limit. We have
shown that atom-interferometer clocks are not competitive
with state-of-the-art atomic clocks; in particular, performance
comparable to optical clocks seems to be out of reach. This
modest performance is despite claims that the system is a
Compton clock, referenced to a purported oscillation at 0 ,
which suggests possible performance 10 orders of magnitude
better than optical clocks. This discrepancy, illustrated using
the propagation of uncertainties from the presumed Compton
oscillation to the clock output, strongly suggests that there is
no physical oscillation at 0 .
ACKNOWLEDGMENTS

We acknowledge useful discussions with J. Hanssen.

014101-4