S.A. Levshakov, I.I. Agafonova, P. Molaro, and D. Reimers
Aims. To specify factors affecting the accuracy of Δα/α measurements and to find the cause of discrepant Δα/α values reported by different groups.
Methods. Statistical tests for normality, homogeneity and independence of data points as well as robust inference to assess the accuracy of the Δα/α values derived from large-size datasets. The SIDAM method and correlation analysis to evaluate the errors of individual Δα/α measurements.
Results.
(1) The sample of Murphy et al.(2003) shows the presence of
uncontrolled errors which were overlooked in previous studies. The data do
not allow to claim that the fine-structure constant changes at the level
of Δα/α ∼ 10-6.
(2) The VLT/UVES archive QSO spectra used
by Chand et al.(2004) suffer from calibration errors which mimic the
value of Δα/α at the level of 10-5. Since statistical
properties of these errors are unknown, the archive spectra
cannot provide the estimate of Δα/α at the level of
10-6.
(3) The error of Δα/α at zabs = 1.15 reported in
Levshakov et al.(2006) was underestimated because it did not account for
correlations between individual measurements. Here we correct our result
to Δα/α = (-0.1+/-1.8)10-6. At present this value together
with Δα/α = (5.4+/-2.5)10-6 at zabs = 1.84 (Levshakov et
al.2007) are the most accurate estimates of Δα/α, which
are, however, discrepant at the 2sigma level. The only known uncertainty
which can affect these values is a relative shift of the wavelength scales
between the blue and red arms of the UVES. This problem requires a special
study.
(4) The limiting accuracy of the Δα/α measurements is
determined by the data points correlations introduced by systematic
errors. We propose a procedure to evaluate the limiting accuracy.
Page created on April 18, 2007.