A Search for Time Variation of the Fine Structure Constant

John K. Webb1 , Victor V. Flambaum1 , Christopher W. Churchill2 , Michael J. Drinkwater1 , John D. Barrow3

arXiv:astro-ph/9803165v4 16 Dec 1998

School of Physics, University of New South Wales, Sydney, NSW 2052, Australia
2
Department of Astronomy & Astrophysics, Pennsylvania
State University, University Park, PA, 16802, USA
3
Astronomy Centre, University of Sussex, Brighton, BN1 9QJ, UK
(Accepted 1998 December 11 for publication in Physical Review Letters)

A method offering an order of magnitude sensitivity gain is described for using quasar spectra to
investigate possible time or space variation in the fine structure constant . Applying the technique
to a sample of 30 absorption systems, spanning redshifts 0.5 < z < 1.6, obtained with the Keck I
telescope, we derive limits on variations in over a wide range of epochs. For the whole sample
/ = 1.1 0.4 105 . This deviation is dominated by measurements at z > 1, where / =
1.9 0.5 105 . For z < 1, / = 0.2 0.4 105 , consistent with other known constraints.
Whilst these results are consistent with a time-varying , further work is required to explore possible
systematic errors in the data, although careful searches have so far not revealed any.
95.30.-k, 95.30.Sf, 95.30.Dr, 98.80.Es

There are several theoretical motivations to search for

space-time variations in the fine structure constant, .
Theories which attempt to unify gravity and other fundamental forces may require the existence of additional
compact space dimensions. Any cosmological evolution
in the mean scale factor of these additional dimensions
will manifest itself as a timevariation of our bare threedimensional coupling constants, [1]. Alternatively, theories have been considered which introduce new scalar
fields whose couplings with the Maxwell scalar Fab F ab
allow timevarying [2]. The measurement of any variation in would clearly have profound implications for
our understanding of fundamental physics.
Spectroscopic observations of gas clouds seen in absorption against background quasars can be used to
search for time variation of . Analyses involving optical spectroscopy of quasar absorbers have concentrated
on the relativistic fine-structure splitting of alkali-type
doublets; the separation between lines in one multiplet is
proportional to 2 , so small variations in the separation
are directly proportional to , to a good approximation.
Whilst the simplicity of that method is appealing, the
relativistic effect causing the fine splitting is small, restricting the potential accuracy. We demonstrate below
how a substantial sensitivity gain is achieved by comparing the wavelengths of lines from different species, and
develop a new procedure, simultaneously analysing the
MgII 2796/2803 doublet and up to five FeII transitions,
(FeII 2344, 2374, 2383, 2587, 2600
A), from three different multiplets. These particular transitions are chosen
because: (i) they are commonly seen in quasar absorption systems, (ii) they fall into and span a suitable restwavelength range, (iii) an excellent database was available [4], (iv) extremely precise laboratory wavelengths
have been measured, (v) the large Fe and Mg nuclear
charge difference yields a considerable sensitivity gain.
We describe the details of the theoretical develop-

ments in a separate paper [5], here summarizing the main

points. The energy equation for a transition from the
ground state within a particular multiplet, observed at
some redshift z, is given by

 
2
z

1
+
Ez = Ec + Q1 Z 2
0
(1)
 2
 4
2 4 z
z
K1 (LS)Z 2
+
K
(LS)
Z
2
0
0
where Z is the nuclear charge, L and S are the electron
total orbital angular momentum and total spin respectively, and Ec is the energy of the configuration centre.
The term in the coefficient Q1 describes a relativistic correction to Ec for a given change in , 0 is the zero
redshift value, and z is the value at some redshift z.
Re-arranged, this gives

 
2
z

1
+
Ez = Ez=0 + [Q1 + K1 (LS)] Z 2
0
 

(2)
4
z
K2 (LS)2 Z 4

1
0
Eq.(2) is an extremely convenient formulation, the second and third terms contributing only if deviates from
the laboratory value. Accurate values for the relativistic coefficients, Q1 , K1 and K2 , have been computed using relativistic many-body calculations and experimental
data. The coefficients and laboratory rest wavelengths
are given in eq.(3). For FeII, the relativistic coefficients,
(Q1 ), are at least one order of magnitude larger than the
spin-orbit coefficients, (K1 ). The variation of the FeII
transition frequencies with is thus completely dominated by the Q1 term. In MgII, the relativistic corrections are small due to the smaller nuclear charge Z (see
eq.(2)), so whilst a change in induces a relatively large
change in the observed wavelengths of the FeII transitions, the change is small for MgII. The relative shifts
1

are substantially greater than those for a single alkali

doublet alone (such as MgII), so MgII acts as an anchor against which the larger FeII shifts are measured.
A comparison of the observed wavelengths of light and
heavy atoms thus provides a dramatic increase in sensitivity compared to analyses of alkali doublets alone.
Since it is already clear from previous observational
constraints that any change in will be very small [6],
[7], [8], it is vital that Ez=0 is known accurately enough.
Indeed, the change in the frequency interval between
MgII 2796 and FeII 2383 induced by a fractional change
/ = 105 is, using eq.(3), 0.03 cm1 . Thus, independent of the quality of the observations, the limiting accuracy in a determination of / is 105 for an uncertainty in the laboratory frequency of 0.03 cm1 . This
highlights the advantage of comparing light and heavy
atoms. Previous analyses of alkali doublets, [6], [7], have
used frequencies of about the accuracy above, but have
been restricted to placing limits of / 104 .
Very precise laboratory spectra of the MgII 2796 and
MgII 2803 lines have recently been obtained [13], in excellent agreement with previous accurate measurements of
MgII 2796 alone [9], [10]. Similarly precise FeII hollow
cathode spectra exist [12]. Inserting these laboratory
wavelengths and our Q and K coefficients into eq.(2),
we obtain the dependence of frequency on for MgII
(top two equations in eq.(3) below) and FeII:
2

P J = 1/2 :
J = 3/2 :
6
D J = 9/2 :
J = 7/2 :
6
F J = 11/2 :
J = 9/2 :
6
P J = 7/2 :

parameters: (i) the redshifts of the corresponding FeII

and MgII components are tied; (ii) the column densities
N(FeII), N(MgII) for individual components can vary independently of each other, but the
velocity dispersion
parameters b(FeII), b(MgII), (b = 2),
p in corresponding components, are constrained by (24/56)b(MgII)
b(FeII) b(MgII). We carry out 150 separate fits to
each absorption system, varying slightly each time, using eq.(3) to compute the input rest-frame wavelengths.
That procedure is performed 3 times; twice where the
b parameters are related according to the two extremes
above (ie. thermal or turbulent line-broadening) and a
third time where all b parameters vary independently.
The fitting procedure returns a value of 2min which was
computed as a function of /. We used the standard
statistical procedure of estimating 1 errors on /
from 2min 1.
Several consistency checks are imposed before accepting a result. First, the 2min for each individual spectral
region being fitted must be statistically acceptable (ie.
its reduced 2 is 1). It follows that 2min for the fit
as a whole is statistically consistent with the number of
degrees of freedom for that fit. Second, we require statistical consistency between the three separate analyses
(for the three different b constraints). If any of the three
/ differ by more than 1 from either of the other
two, the system is rejected. These criteria lead to the
rejection of only about 1/10 of the sample, reflecting the
good general statistical robustness of the procedure. The
final / accepted corresponded to the lowest of the 3
values of 2min .
The best-fit values of /, and the 1 error bars,
are plotted against redshift in Fig. 1. For z < 1 there
is no departure from the present-day value. The scatter
in the data is consistent with the individual error bars,
ie. there is no evidence for any space or timevariations
in . However, at z > 1 the situation is less clear, with
14 points giving 2 = 34.9, 3 falling above zero and the
rest below. Fig. 1 shows there is a dip in a relatively
narrow redshift interval 0.9 < z < 1.2, where 9 out of the
10 points lie below zero, and it is this dip which cause
the overall z > 1 2 to be high. Assuming that this is a
statistical fluctuation and that we have somehow underestimated the errors, we can increase the error bars on
each point by a constant amount, S, where S 0 such
that the reduced 2 = 1, ie. the error on the ith point
becomes i = i + S (where i is the value illustrated in
Fig. 1). The results of this procedure, for the sample as
a whole, and in two redshift ranges are:

= 35669.286(2) + 119.6x
= 35760.835(2) + 211.2x
= 38458.9871(20) + 1394x + 38y
= 38660.0494(20) + 1632x + 0y (3)
= 41968.0642(20) + 1622x + 3y
= 42114.8329(20) + 1772x + 0y
= 42658.2404(20) + 1398x 13y

z 4
where x = [( z0 )2 1] and y = [(
0 ) 1].
The astronomical data used for this analysis was obtained using the HIRES echelle spectrograph [14] on
the Keck I 10m telescope during three observing runs
in 1994-1996. High-quality spectra of 25 quasars were
obtained, in which intervening absorption systems at
low/intermediate redshift have been identified containing
FeII, MgII, and other species. Full observational details
are given in ref. [4].
We now determine the relative positions of the FeII and
MgII lines and estimate / for each absorption system in the sample. Measuring each line (ie. MgII 2796,
MgII 2803 and 5 FeII lines) independently is not optimal, because the number of fitting parameters is not
minimised, as discussed below. The procedure used is iterative, where all available lines are fitted simultaneously
with Voigt profiles, using VPFIT, a non-linear leastsquares programme designed specifically for analysing
quasar absorption spectra [15]. We minimise the total
number of free parameters by linking physically related

= 1.09 0.36 105

(0.6 < z < 1.6)

(1.0 < z < 1.6)

= 0.17 0.39 105

= 1.88 0.53 10

(0.6 < z < 1.0)

(4)

The values of S are, in order, S = 0.46, 0.06, 0.53. The

whole sample departs from zero at the 3.0 level towards
2

smaller values of . However, the z < 1 points alone show

no significant trend, suggesting that there are no significant errors in the adopted laboratory FeII and MgII
wavelengths, or in the general procedure. However, at
z > 1, there is a 3.5 deviation. We have also experimented with other statistical methods which also give
more conservative error estimates than a simple weighted
mean (eg. a Bayesian method), and obtain results consistent with the above (to be reported elsewhere).
What systematic (non-physical) effects could mimic
such an effect (either a general trend towards negative
/ for z > 1, or, the curious cusp in the range
0.9 < z < 1.2)? The quality of the spectroscopic data
does not deteriorate with redshift (this would be revealed
by larger error bars at higher redshift) but in fact improves [4]. The unmodified error bars reflect uncertainties due to signal-to-noise, spectral resolution, velocity
structure, the number of absorption lines fitted, and assume we have properly de-convolved the absorption features into their correct number of individual components.
However this assumption may be wrong. There may be
weak blended unresolved lines, even with data of the high
quality we have here. For a specific absorption system,
an interloper could indeed mimic a shift, eg. if it were
blended with one of the MgII 2796 or 2803 anchors.
Nevertheless it is unlikely that this occurs preferentially
for z > 1 and not for z < 1. Nor should random blending create a negative / trend; it should merely increase the scatter about a mean of zero. An extensive
search for a systematic blend (ie. a weak transition from
some other species at the same redshift as the absorbing
cloud) falling close to one of the Mg or Fe absorption
lines, proved negative (details to be reported elsewhere).
Uncertainties in the Fe or Mg laboratory wavelengths
cannot be responsible, since any errors would be an order of magnitude below the observed effect, and in any
case, the z < 1 points reveal no offset. We also checked
to see whether uncertainty in the instrumental resolution
could introduce errors, and found the results insensitive
to this. It is conceivable that certain absorption redshifts
could be less reliable, due to the positioning of the FeII
and MgII lines with respect to the echelle order edges,
where wavelength calibration may be worse. At higher
wavelengths, the HIRES optical format is such that there
are gaps in the wavelength coverage. Also, the ThAr calibration spectrum has fewer lines per unit wavelength
at higher wavelengths. These effects could mean that
wavelength calibration is less reliable for higher redshift
points than the lower redshift ones. Alternatively, perhaps there is a subtle bug in the wavelength calibration
software. The z > 1 effect we have found is very small:
a gradual drift of 2.5 times the wavelength calibration
residuals, over the range corresponding to the observed
FeII and MgII lines, could produce an effect of the significance we have found. Previous analyses of similar quality data have not demanded such high precision, so such

errors could have gone unnoticed in other spectroscopic

analyses.
Can any of these effects produce the apparent trend
we find? We have carried out extensive numerical experiments aimed specifically at testing for these potential systematic problems. These involved carrying out an
identical analysis on the wavelength calibration spectra
(ThAr) as for the QSO spectra, measuring the emission
line wavelengths and inferring a change in from shifts
betwen the measured and laboratory wavelengths (details to be reported elsewhere). The results were unambiguous, in that any errors in the wavelength calibration
across all spectral orders are so small that they could not
contribute significantly to the apparent offset at z > 1,
unless the literature values for the calibration spectral
lines (ThAr) are substantially in error, with a gradual
non-linear shift in the ThAr wavelengths.
We consider all of the above potential errors to be
rather unlikely. Since the effect is dominated by the
0.9 < z < 1.2 points, and because a genuine physical
effect confined to one specific epoch in the history of the
universe does not seem at present to be well motivated
by theoretical expectations, we presume the explanation
is that there are additional undiscovered velocity components in the absorbing gas for those particular absorption
sytems, even though all points have been subjected to the
same analysis. Further observations of different species
in these clouds could answer this. Our results should
thus be regarded as stringent upper limits on any possible time-variation rather than a positive detection of a
change.
We may compare our results with other recent values.
Observations at z 3 [7] have yielded an upper limit
|/| < 3.5 104 . Our analysis is at lower redshift,
so a comparison requires choosing some (arbitrary) functional form of the evolution. At lower redshifts, a recent
analysis of radio wavelength spectra of atomic hydrogen
and molecular gas [8] gives limits |/| < 3 106 at
two redshifts, z = 0.25, 0.68. An additional recent constraint of 3.5 5.5 106 [7] comes from a comparison of
hyperfine and optical redshifts. These limits are consistent with the results given in eq.(4). The strongest terrestrial constraint on the time evolution of comes from
the Oklo natural nuclear reactor [16]. The Oklo event
is estimated to have taken place around 1.8 billion years
ago (corresponding to z 0.1). We adopt a cosmological
model with q0 = 1/2, = 0 and take the age of the universe to be 13 109 years. The Oklo data has recently
been re-examined [17], and upper limits on a change in
are 0.9 107 < (Oklo 0 )/0 < 1.2 107 , assuming the weak and strong couplings are unchanged.
These bounds are clearly consistent with our results for
z < 1. For the cosmological parameters quoted, if we
adopt z = 0.8, the mid-point of our lower redshift range,
eq.(4) implies
3

hi

= 2.2 5.1 1016 yr1

(0.6 < z < 1.0)

(5)

C. Lineweaver, J. Magueijo, D. Morton, M. Murphy, J.

Pickering, O. Sushkov, A. Thorne, A. Vidal-Madjar, S.
Vogt and the HIRES team, D. Wineland for various important contributions. We also thank A. Dryer and SUN
Microsytems Australia Pty Ltd for computers and L.
Evans for computing assistance. JDB is supported by
PPARC.

What other physical phenomena, other than time variability of , could give rise to the observational effect we
report? The spacing of the MgII and FeII isotopes is such
that a significant change in the isotopic ratios could explain the observations. However, the change would need
to be substantial; for example, this would require most of
the Mg in the universe at z > 1 to be in 26 Mg (the present
epoch abundance is 10%), and a physical mechanism
found to convert almost all the 26 Mg into 24 Mg by the
present epoch. If large scale magnetic fields exist, and
the quasar light is polarised, these could potentially give
rise to correlated apparent shifts in absorbers in neighbouring regions of the universe. However, for magnetic
fields to be responsible for the global effect in for z > 1
(but not for z < 1), a sharp variation at z 1 or some
form of oscillatory variation would be required, both of
which are hard to motivate.
Some authors have suggested more exotic forms of evolution of the constants, including oscillations [1]. These
could arise from new light bosons with mass m, producing periodic variations in the frequency of the radiation emitted at high z, with a modulation frequency
m1 [3]. The creation of timevarying by approximate global symmetry ( [2]) allows oscillatory variations
introduced by the decaying mean oscillations of a scalar
field coupling to Fab F ab . We note that as we move
above z = 1.25, for a critical-density non-inflationary
universe, we can encounter causally disconnected regions
of the universe. The observational sensitivity achieved
by these observations exceeds that of current microwave
background observations and a larger dataset than ours
may contain important new cosmological information.
The work we have presented here demonstrates the
possibilites for extending this type of study to incorporate different species, other than Mg and Fe, spanning a
wider redshift baseline, so equally impressive constraints
should be obtainable at higher redshifts. Future analyses of other species will be hampered by the lack of accurate laboratory wavelengths. We hope that this paper
will provide an impetus for new high-precision laboratory
measurements.

[1] W. Marciano, Phys. Rev. Lett., 52, 489 (1984); J.D. Barrow, Phys. Rev. D 35, 1805 (1987); T. Damour & A. M.
Polyakov, Nucl. Phys. B 423, 532 (1994).
[2] S.M. Carroll, astro-ph/9806099 (1998).
[3] C.T. Hill and G.G. Ross, Nucl. Phys. B 311, 253 (1988);
Phys. Lett. B 203, 125 (1988); C.T. Hill, P.J. Steinhardt,
and M.S. Turner, Phys. Lett. B 252, 343 (1980).
[4] C. W. Churchill, The Low Ionization Gaseous Content of
Intermediate Redshift Galaxies, PhD. thesis, UC Santa
Cruz (1997); C. W. C. Churchill, S. S. Vogt, J. Charlton,
Ap. J. Suppl. in prep.
[5] V. A. Dzuba, V. V. Flambaum, & J. K. Webb. Submitted
to Phys. Rev. Lett.
[6] D. A. Varshalovich & A. Y. Potekhin, Space Sci. Revs.,
74, 259 (1995); D. A. Varshalovich, V. E. Panchuk & A.
V. Ivanchik, Astron. Lett., 22, 6 (1996).
[7] L. L. Cowie & A. Songalia, Ap.J., 453, 596 (1995).
[8] M. J. Drinkwater, J. K. Webb, J. D. Barrow and V. V.
Flambaum, Mon. Not. R. astron. Soc., 298, 457 (1998).
[9] R. E. Drullinger, D. J. Wineland, and J.C. Bergquist,
Appl. Phys., 22, 365 (1980).
[10] W. Nagourney & H.G. Dehmelt, Bull. Am. Phys. Soc.,
26, 805 (1981).
[11] D. C. Morton, Ap. J. Suppl., 77, 119 (1981).
[12] G. Nave, R. C. M. Learner, A. P. Thorne, and C. J. Harris, J. Opt. Soc. Am. B, 8, 2028 (1991).
[13] J. C. Pickering, A. P. Thorne, and J. K. Webb, Mon.
Not. R. astron. Soc., accepted.
[14] S. S. Vogt et al. In Instrumentation in Astronomy VIII,
eds. D. L. Crawford & E. R. Craine, SPIE, 2198, 362
(1994).
[15] J. K. Webb, PhD thesis, University of Cambridge (1987).
[16] A. I. Shylakhter, Nature, 264, 340 (1976); ATOMKI Report A/l (1983).
[17] T. Damour & F. Dyson, Nucl. Phys. B, 480, 37 (1996).

We are very grateful to R. Carswell, J. Charlton, V.

Dzuba, A. Fernandez-Soto, J. Garcia-Bellido, R. Learner,

FIG. 1. Plot of / vs z for 30 FeII/MgII absorption

systems. All the transitions given in equation 3 were used,
where available. Plotted error bars were determined from
2min 1 (but statistical results estimated using larger errors,
as discussed in the text).