Prima che apra il pdf... c'è forse del ming men di mezzo, in esso?
Non mi pare..."""
arXiv:1008.3907v1 [astro-ph.CO] 23 Aug 2010
Evidence for spatial variation of the fine structure constant
J. K. Webb1 , J. A. King1 , M. T. Murphy2 , V. V. Flambaum1 , R. F. Carswell3 , and M. B. Bainbridge1
1
School of Physics, University of New South Wales, Sydney, NSW 2052, Australia
2
Centre for Astrophysics and Supercomputing, Swinburne University of Technology,
Mail H39, PO Box 218, Victoria 3122, Australia and
3
Institute of Astronomy, Madingley Road, Cambridge, CB3 0HA, England.
(Dated: August 25, 2010)
We previously reported observations of quasar spectra from the Keck telescope suggesting a
smaller value of the fine structure constant, α, at high redshift. A new sample of 153 measurements
from the ESO Very Large Telescope (VLT), probing a different direction in the universe, also depends
on redshift, but in the opposite sense, that is, α appears on average to be larger in the past. The
combined dataset is well represented by a spatial dipole, significant at the 4.1σ level, in the direction
right ascension 17.3 ± 0.6 hours, declination −61 ± 9 degrees. A detailed analysis for systematics,
using observations duplicated at both telescopes, reveals none which are likely to emulate this result.
PACS numbers: 06.20.Jr, 95.30.Dr, 95.30.Sf, 98.62.Ra, 98.80.-k, 98.80.Es, 98.80.Jk
more extensive sky coverage. The ESO midas pipeline
Quasar spectroscopy as a test of fundamental
was used for the first data reduction step, including wave-
physics.— The vast light-travel times to distant quasars
allows us to probe physics at high redshift. The rela- length calibration, although enhancements were made to
tive positions of wavenumbers, ωz , of atomic transitions derive a more robust and accurate wavelength solution
detected at redshift z = λobs /λlab − 1, can be com- from an improved selection of thorium-argon calibration
lamp emission lines [10]. Echelle spectral orders from sev-
pared with laboratory values, ω0 , via the relationship
2 2
2
ωz = ω0 + Q αz − α0 /α0 and the coefficient Q mea- eral exposures of a given quasar were combined using
sures the sensitivity of a given transition to a change in uves popler [11]. A total of 60 quasar spectra from the
α. The variation in both magnitude and sign of Q for dif- VLT have been used for the present work, yielding 153
absorption systems. Absorption systems were identified
ferent transitions is a significant advantage of the Many
Multiplet method [1, 2], helping to combat potential sys- via a careful visual search of each spectrum, using rd-
tematics. gen [12], scanning for commonly detected transitions at
the same redshift, hence aligned in velocity coordinates.
The first application of this method, 30 measurements
Several transition matches were required for acceptance
of ∆α/α = (αz − α0 ) /α0 , signalled a smaller α at high
and, given the high spectral resolution, chance matches
redshift at the 3σ significance level. By 2004 we had made
were eliminated.
143 measurements of α covering a wide redshift range,
using further data from the Keck telescope obtained by Absorption system modelling.— As in our previous
3 separate groups, supporting our earlier findings, that studies, vpfit was used to model the profiles in each ab-
towards that general direction in the universe at least, α sorption system [13] with some enhancements, described
may have been smaller at high redshift, at the 5σ level [3– in [8]. A comprehensive list of the transitions used, their
5]. The constant factor at that point was (undesirably) laboratory wavelengths, oscillator strengths, and Q coef-
the telescope/instrument. ficients are compiled in [4, 8].
Subsequently, only 1 further independent statistical The following general procedures were adhered to: (i)
study has been completed [6], but difficulties with the For each absorption system, physically related parame-
analysis methods mean those results do not add to or ters (redshifts and b-parameters) are tied, in order to
provide a check on our earlier results [7]. A small num- minimise the required number of free parameters and
ber of individual α measurements have been made, but derive the strongest possible constraints on line posi-
provide no general conclusions since the systematic com- tions, and hence ∆α/α. (ii) Parameters were tied only
ponent of the error on ∆α/α ∼ 10−5 . for species with similar ionisation potentials, to min-
imise possible introduction of random effects on α, mim-
New data from the VLT.— We have now analysed
a large dataset from a different observatory, the VLT. icked by spatial (and hence velocity) segregation effects;
Full details and searches for systematic errors will be (iii) Line broadening is typically dominated by turbulent
given elsewhere[8, 9]. Here we summarize the evidence rather than thermal motion. Both limiting-case models
for spatial variation in α emerging from the combined were applied and ∆α/α determined for each. The final
Keck+VLT samples. Quasar spectra, obtained from the ∆α/α was derived from a likelihood-weighted average.
Details and justification are given in [8]; (iv) Where ap-
ESO Science Archive, were selected, prioritising primar-
ily by expected signal to noise but with some preference propriate (and where available), isotopic structures are
given to higher redshift objects and to objects giving included in the fitting procedure (for Mgi, Mgii, Aliii,
2
Siii, and Feii); (v) Velocity structures were determined and hence, on average, observe different directions on the
initially choosing the strongest unsaturated transitions sky. We are thus motivated to explore a simple spatial
in each system. Normalised residuals across each tran- dependence using the combined dataset.
sition fitted were examined and the fit progressively re- The Keck sample we use is as presented in [5] with a
fined with the introduction of each additional transition minor modification: 3 points were removed. 2 had been
to the fit; (vi) Transitions falling in spectral regions con- included erroneously (from a spectrum known to have
taminated by telluric features or atmospheric absorption calibration problems) and 1 further point was clipped,
were discarded. Any data regions contaminated by cos- having a residual greater than 3σ against a modified LTS
mic rays, faulty CCD pixels, or any other unidentified fit to the Keck data.
noise effects, were also discarded; (vii) A few gravita- Initially the 3 datasets (i.e. Keck, VLT and combined)
tional lenses were identified by being difficult or impos- are fitted using a simple possible dipole+monopole
sible to model successfuly. The non-point source quasar model, represented here by ∆α/α = A cos Θ+m, where m
image and the resultant complex line-of-sight geometry is a constant allowing an offset from the terrestrial value,
can significantly alter apparent relative line strengths. Θ is the angle on the sky between quasar sightline and
These systems were discarded; (viii) In all cases we de- best-fit dipole position, and A is the dipole amplitude.
rived the final model without solving for ∆α/α. The To examine the probability of the observed dipole
introduction of ∆α/α as an additional free parameter model arising by chance, we bootstrap the sample, re-
was only done once the profile velocity structure had peatedly randomising the association between ∆α/α and
been finalised, eliminating any possible bias towards a quasar sightline, fitting ∆α/α = A cos Θ + m at each re-
‘preferred’ ∆α/α. One potential consequence of this ap- alisation. We then numerically determine the probability
of obtaining a value of χ2 less than or equal to the actual
proach might conceivably be a small bias on ∆α/α to-
value by comparing with the χ2 probability distribution
wards zero, should some ‘fitting-away’ of ∆α/α occur, by
column density adjustments or velocity structure deci- from the bootstrap process.
sions. The reverse is not true, i.e. it cannot bias towards Figure 1 illustrates the best-fit dipole equatorial co-
a non-zero ∆α/α. ordinates on an all-sky map, with approximate 1σ error
vpfit [13] minimises χ2 simultaneously over all species. contours derived from the covariance matrix. Figure 2
Whilst the strongest components may appear in all illustrates the ∆α/α binned data and the best-fit dipole
species, weaker components can sometimes fall below the model. Best-fit parameters are given in the captions.
detection threshold and hence are excluded, such that As a second trial model, allowing for a spatial gradient
a component which appears in MgII, for example, does in α, we assign a distance to each ∆α/α measurement
not appear in FeII. There is no solution to this (known) of r(z) = ct(z) where c is the speed of light and t(z)
problem but its effect merely adds an additional random is the look-back time at redshift z. The model is then
scatter on ∆α/α for an ensemble of observations. ∆α/α = Br(z) cos Θ + m. Figure 3 illustrates ∆α/α
Spatially dependent α.— An initial inspection of ∆α/α vs look-back time distance projected onto the dipole
vs redshift for the new VLT dataset reveals a redshift axis, r cos Θ, using the best-fit dipole parameters for this
trend, opposite in sign compared to the earlier Keck model. This model seems to represent the data reason-
data. Splitting each sample at z = 1.8, our 2004 Keck ably well and the data show a strong correlation, signifi-
sample [5] gave ∆α/α z<1.8 = −0.54 ± 0.12 × 10−5 and cant at the 4.1σ level.
∆α/α z>1.8 = −0.74 ± 0.17 × 10−5 . The present 2010 Given the relatively low statistical significance of the
VLT sample, which will be discussed in detail in [8] gives monopole term m for both models above (see captions,
∆α/α z<1.8 = −0.06 ± 0.16 × 10−5 and ∆α/α z>1.8 = Figures 2 and 3), and because the theoretical interpre-
+0.61 ± 0.20 × 10−5 . Errors here and throughout this tation of a monopole term is unclear, a third model was
fitted, ∆α/α = Br cos Θ, giving B = 1.10 ± 0.25 × 10−6
paper are 1σ estimates.
GLyr−1 with a significance of 4.2σ and giving parame-
Errors on individual ∆α/α values for the VLT sample
2 2 2 2
ters right ascension 17.4 ± 0.6 hours, declination −58 ± 6
are σtot = σstat +σsys , where σsys was derived empirically
using a modification of the Least Trimmed Squares (LTS) degrees.
method, where only 85% of data, those points with the An alternative to empirically increasing the ∆α/α er-
smallest squared residuals, are fitted. σsys was assumed ror bars to incorporate a systematic component is to as-
2 2
constant for all VLT absorbers and was found to be ≈ sume σtot = σstat and to iteratively trim the data during
0.88 × 10−5 , showing that the scatter in the VLT ∆α/α model fitting. This provides a further stringent test of
is greater than expected on the basis of statistical-errors whether the apparent gradient in α is dominated by a
alone. Errors on ∆α/α for the Keck sample are discussed subset of the data, perhaps more prone to some unknown
2 2
in [4], although we derive a new estimate of σsys for the systematic than the remainder. Adopting σtot = σstat
Keck points using the LTS method. will clearly result in higher significance levels. In Figure
The Keck (Mauna Kea, Hawaii) and VLT (Paranal, 4 we plot the statistical significance of the dipole in units
Chile) locations on Earth are separated by 45◦ in latitude of σ and find that over 40% of the data must be discarded
3
FIG. 1. All-sky plot showing the independent Keck (green)
and VLT (blue) best-fit dipoles, and the combined sample
(red), in equatorial co-ordinates. Approximate 1σ confidence
FIG. 3. ∆α/α vs Br cos Θ for the model ∆α/α = Br cos Θ +
contours are from the covariance matrix. A bootstrap anal-
m showing the gradient in α along the best-fit dipole. The
ysis gives the chance-probability of getting the observed (or
best-fit direction is at right ascension 17.4±0.6 hours, declina-
better) alignment between the independent Keck and VLT
tion −62±6 degrees, for which B = (1.1±0.2)×10−6 GLyr−1
dipoles is only 4%. The cosmic microwave background dipole
and m = (−1.9 ± 0.
× 10−6 . This dipole+monopole model
and antipole are illustrated for comparison.
is statistically preferred over a monopole-only model also at
the 4.1σ level. A cosmology with parameters (H0 , ΩM , ΩΛ ) =
(70.5, 0.2736, 0.726) was used [14].
FIG. 4. As an alternative to increasing ∆α/α error bars, to
account for the additional scatter in the data as described in
2 2
the text, we instead use σtot = σstat and iteratively clip the
FIG. 2. ∆α/α for the combined Keck and VLT data vs angle
Θ from best-fit dipole, ∆α/α = A cos Θ + m, A = (0.97 ± most deviant ∆α/α value, fitting ∆α/α = A cos Θ + m. The
0.21) × 10−5 and m = (−0.18 ± 0.08) × 10−5 . Dashed lines vertical dashed line illustrates where the dotted curve χ2 = 1,
ν
when ∼ 8% of the data has been trimmed. Almost 50% of the
illustrate ±1σ errors on the dipole fit. The best-fit dipole is at
right ascension 17.3 ± 0.6 hours, declination −61 ± 9 degrees data must be discarded before the significance drops below 4σ
showing that the dipole signal is generally present in entire
and is statistically preferred over a monopole model at the
dataset.
4.1σ level.
to push the result below 4σ, implying a remarkable in- ferent wavelength settings. This may apply to either or
ternal consistency within the data. both Keck and VLT spectra. Since spectrograph slit il-
luminations are different for quasar (point source) and
Empirical test for systematics.— One potential effec-
ThAr calibration lamp (uniform illumination), the sub-
tive relative distortion might be due to slight mechanical
sequent combination of individual exposures to form a
mis-alignments of the spectrograph slits for the 2 arms,
1-dimensional spectrum may then contain relative veloc-
red and blue, of the UVES spectrograph on the VLT.
ity shifts between spectral segments coming from differ-
However, this specific effect appears to be substantially
smaller than required to explain values of ∆α/α∼ 10−5 ent exposures. This effect will exist in our data at some
level and it is clearly important to know the impact on
seen in the present work [15].
an ensemble of measurements of α.
A more subtle but related effect may be slight off-
centre placement of the quasar image in the spectrograph Fortunately, 6 quasars in our sample have both Keck
slit, by different amounts for different exposures, at dif- and VLT spectra, allowing a direct and empirical check
4
on the effect above, and indeed any other systematics significant amounts of extra scatter into the data above
which produce relative velocity shifts along the spectrum. what is already observed, implying that this is an over-
To do this we selected small spectral segments, each a few estimate of a systematic effect of this type. Additionally,
̊ wide, flanked by unabsorbed continuum flux from the
A the trend of δv(λobs )i against wavelength is different in
quasar, and fitted Voigt profiles using vpfit, but adding magnitude and sign for each quasar pair, implying that
an additional free parameter allowing a velocity shift be- these effects are likely to average out for an ensemble of
tween the Keck and VLT segments, δv(λobs )i , where λobs observations. Thus, application of the effect as described
is the observed wavelength and i refers to the ith quasar. above should be regarded as extreme in terms of impact
All available absorption lines in the 6 spectra were used, on estimating ∆α/α.
including both Lyman-α forest lines and heavy element Conclusions.— Quasar spectra obtained using 2 sepa-
lines but excluding telluric features. In this way we can rate observatories reveal a spatial dependence of the fine
map any effective relative distortions in the calibrations stucture constant at a significance of 4.1σ, estimated con-
between each pair of spectra. A total of 694 measure- servatively, taking into account both statistical and sys-
ments were used from the 6 pairs of spectra over the tematic errors. Two independent datasets reveal a strik-
observed wavelength range 3506 < λ < 8945 ̊. A ing internal consistency and the directions of the indepen-
dently derived spatial dipoles agree well, with a chance
We formed a composite function δv(λobs ) after first
probability of 4%. The apparent symmetry in magnitude
normalising δv(λobs )i = 0 for each i to remove any po-
of the ∆α/α variation between northern and southern
tential small constant velocity offsets from each spectrum
hemisphere quasar data is also striking. A subset of the
(expected from off-centering of the quasar in the spectro-
quasar spectra observed at both observatories permits a
graph slit), which cannot influence α.
direct test for systematics. None are found which are
Finally we fit the composite δvλobs with a linear
likely to emulate the apparent cosmological dipole in α
function f (δv) = aλobs + b where a = (−7 ± 14) ×
we detect. To explain our results in terms of systematics
−1
10−5 km s−1 ̊ , b = 0.38 ± 0.71 km s−1 . The final f (δv)
A
will require at least 2 different and finely tuned effects.
thus shows a weak (but statistically insignificant) velocity
Future similar measurements targeting the apparent pole
drift, and provides an empirical transformation between
and anti-pole directions will maximise detection sensitiv-
the Keck and VLT wavelength scales. For each quasar
ity, and further observations duplicated on 2 indepen-
absorption system, we modify the input laboratory wave-
dent telescopes will better constrain systematics. Above
lengths used in the Voigt profile fitting procedure λlab to
all, an independent technique is required to check these
λlab = λlab + ∆λlab where ∆λlab = λlab δv (λobs )/c, and
results. Qualitatively, our results suggest a violation of
finally use the λlab to re-compute ∆α/α for the entire
the Einstein Equivalence Principle, and could infer a very
sample. Since we do not know whether the Keck or the
large or infinite universe, within which our ‘local’ Hubble
VLT observation causes the non-zero values of the param-
volume represents a tiny fraction, with correspondingly
eters a, b above, we applied the transformation separately
small variations in the physical constants.
to both and examine the impact in each case.
This work is supported by the Australian Research
There was one complicating aspect of this effect ex- Council. RFC is grateful to the Leverhulme Trust for an
cluded from the discussion above, arising from a 7th spec- Emeritus grant. We thank Steve Curran, Elliott Koch,
tral pair. The δv(λobs )7 showed a more significant non- Julian Berengut, John Barrow and Paul Davies for dis-
zero slope than the other 6, suggesting a small but signif- cussions throughout this work.
icant calibration problem with that particular spectrum.
We therefore applied a slightly more complicated trans-
formation to the data to allow for this, using a Monte
Carlo simulation to estimate the potential impact on our
full combined Keck and VLT sample of both the previ- [1] J. K. Webb et al., Phys. Rev. Lett. 82, 884 (Feb. 1999),
arXiv:astro-ph/9803165.
ous effect measured in 6 quasars plus the effect derived
[2] V. A. Dzuba, V. V. Flambaum, and J. K.
from the 7th quasar simultaneously, applied in appropri-
Webb, Phys. Rev. Lett. 82, 888 (Feb. 1999),
ate proportions. The full details of this analysis will be
arXiv:physics/9802029.
discussed separately in [9]. [3] J. K. Webb et al., Phys. Rev. Lett. 87, 091301 (Aug.
A systematic of the same magnitude as that from the 2001), arXiv:astro-ph/0012539.
7th pair cannot be present in any large fraction of our [4] M. T. Murphy, J. K. Webb, and V. V. Flambaum,
Mon. Not. Roy. Astron. Soc. 345, 609 (Oct. 2003),
data, otherwise it would generate large numbers of no-
arXiv:astro-ph/0306483.
ticeable outliers. If we apply f (δv) from the 6 quasar
[5] M. T. Murphy, V. V. Flambaum, and J. K. Webb, in
pairs, the significance of the dipole is reduced to 3.1σ. Astrophysics, Clocks and Fundamental Constants, Lec-
Blindly including the effect of the 7th pair under a Monte ture Notes in Physics, Berlin Springer Verlag, Vol. 648,
Carlo method reduces the significance to a most likely edited by S. G. Karshenboim & E. Peik (2004) pp. 131–
value of 2.2σ. However, in this circumstance we add 150, arXiv:astro-ph/0310318.
5
[6] H. Chand et al., Astron. Astrophys. 417, 853 (Apr. 2004),
arXiv:astro-ph/0401094.
[7] M. T. Murphy, J. K. Webb, and V. V. Flambaum,
Mon. Not. Roy. Astron. Soc. 384, 1053 (Mar. 2008),
arXiv:astro-ph/0612407.
[8] J. A. King et al.(2010), in preparation; to be submitted
to Mon. Not. Roy. Astron. Soc.
[9] F. E. Koch et al.(2010), in preparation; to be submitted
to Mon. Not. Roy. Astron. Soc.
[10] M. T. Murphy et al., Mon. Not. Roy. Astron. Soc. 378,
221 (Jun. 2007), arXiv:astro-ph/0703623.
[11] M. T. Murphy, “uves popler,” (2010), http://
astronomy.swin.edu.au/~mmurphy/UVES_popler.
[12] R. F. Carswell, “rdgen,” (2004),
http://www.ast.cam.
ac.uk/~rfc/rdgen.html.
[13] R. F. Carswell and J. K. Webb, “vpfit - Voigt profile
fitting program. Version 9.5,” (2010),
http://www.ast.
cam.ac.uk/~rfc/vpfit.html.
[14] G. Hinshaw et al., Astrophys. J. Supp. 180, 225 (Feb.
2009), arXiv:0803.0732.
[15] P. Molaro et al., Astron. Astrophys. 481, 559 (Apr.
2008), arXiv:0712.3345.
FIG. 5. Supplementary figure. All-sky illustration of the combined Keck and VLT ∆α/α measurements. Squares are VLT
points. Circles are Keck points. Triangles are quasars observed at both Keck and VLT. Symbol size indicates deviation of
∆α/α from the monopole value m in ∆α/α = A cos Θ + m (Figures 2 and 3). The grey shaded area represents the Galactic
plane with the Galactic centre indicated as a bulge. The blue dashed line shows the equatorial region of the α-dipole. More
and larger blue squares are seen south of the equatorial region and more and larger red circles are seen north of it.
"""