Nature 536, 326–328 (2016); doi:10.1038/nature18594

Some of the analyses (samples X1–2, X1–3, X1–4 and X2–4) reported in this Letter’s Extended Data Table 1 and in the Supplementary Data spreadsheet were previously published in refs 28 and 30. The other analyses are new measurements, as are all measurements of aluminium concentration in the metal we studied. Owing to a typographical error, an erroneous value for the magnesium composition of sample X2–4 was reported in refs 28 and 30 (which has no impact on the conclusions of those papers because magnesium concentration was not used). Our recent re-evaluation of the raw microprobe data confirms that we have here reported the correct value (0.94 mol% in Extended Data Table 1, corresponding to 0.59 wt% in the Supplementary Data spreadsheet) for the magnesium composition of sample X2–4. (Therefore, no correction to this Letter is required.)