We determine the strong coupling as from a next-to-leading order analysis of processes used for the NNPDF2.1 parton determination, which includes data from neutral and charged current deep-inelastic scattering, Drell-Yan and inclusive jet production. We find alpha(S)(M-Z) = 0.1191 +/- 0.0006(exp), where the uncertainty includes all statistical and systematic experimental uncertainties, but not purely theoretical uncertainties, which are expected to be rather larger. We study the dependence of the results on the dataset, by providing further determinations based respectively on deep-inelastic data only, and on HERA data only. The deep-inelastic fit gives the consistent result alpha(S)(M-Z) = 0.1177 +/- 0.0009(exp), but the result of the HERA-only fit is only marginally consistent. We provide evidence that individual data subsets can have runaway directions due to poorly determined PDFs, thus suggesting that a global dataset is necessary for a reliable determination. (C) 2011 Elsevier B.V. All rights reserved.
Bibliographical note13 pages, 9 figures; final version to be published in Phys. Lett. B. Tolerance analysis for individual experiments (figs.6, 7 and tab.3) and fits with increased numbers of replicas (tab 1, 2) added. Many further small corrections