Contamination in the MACHO dataset and the puzzle of LMC Microlensing

In a recent series of three papers, Belokurov, Evans, and Le Du, and Evans and Belokurov, reanalysed the MACHO collaboration data and gave alternative sets of microlensing events and an alternative optical depth to microlensing toward the Large Magellanic Cloud (LMC). Even though they examined less than 0.2% of the data they claimed that by using a neural net program they had reliably selected a better (and smaller) set of microlensing candidates. Estimating the optical depth from this smaller set, they claim that the MACHO collaboration overestimated the optical depth by a significant factor and that the MACHO microlensing experiment is consistent with lensing by known stars in the Milky Way and LMC. As we show below, the analysis by these authors contains several errors which render their conclusions meaningless. Their efficiency analysis is clearly in error, and since they did not search through the entire MACHO dataset, they do not know how many microlensing events their neural net would find in the data or what optical depth their method would give. Examination of their selected events suggests that their method misses low S/N events and thus would have lower efficiency than the MACHO selection criteria. In addition, their method is likely to give many more false positives (non-lensing events identified as lensing). Both effects would increase their estimated optical depth. Finally, we note that the EROS discovery that LMC event-23 is a variable star reduces the MACHO collaboration estimates of optical depth and Macho halo fraction by around 8%, and does open the question of additional contamination.