After my obvious mistake in parametrization I ran several tests with Alder. Unfortunately Alder has same problems than Rolloff and trying to estimate regional ancestries and admixture dates is problematic for two main reasons:
- both programs assume only one migration. If there has been several migrations between populations under test, as it often regionally is, the spectrum of IBD segments is large and the Admixture-induced Linkage Disequilibrium curve becomes too vague to determine admixture dates.
- in case of obvious one migration the problem is that in some point the results reaches a point where too many segments are "constant" for a long time and those segments don't follow the average recombination rate assumed by Alder. This leads to a too recent admixture event. For example, calculating admixture dates for ancient Saami and Nganasan admixture events in Finland gives almost exactly similar curves, pointing out to a fixation of certain IBD segments.
What can I do? It is possible to pick remote populations and use them as proxies for the assumed admixture event. For example using Danes instead of Swedes to show North Germanic admixture in Finland. This is however possible only in cases we already know the prehistory well enough to make this initial setting relevant.