There remain issues to resolve here. Firstly, recent admixture generates large-scale LD which may cause difficulties in a dense dataset as the allele distributions are not independent. These effects may be hard to alleviate with our simple LD correction described below. STRUCTURE [10] allows careful modeling. Secondly, more ancient admixture, especially if the admixed population is genetically now homogeneous, may lead to a causal eigenvalue not very different from the values generated by the sampling noise. Suppose, for example, in our simulation above, we let population C mate panmictically for another 20 generations. Then we will get three clusters for A, B, C that are nearly collinear, but not exactly because of the recent 20-generation divergence, which is reflecting genetic drift unique to that population.
