Seg Site difference with 'ghost' unsampled pops compared to ms

[stsmall]

Hi @paulstaab,
I was attempting to understand the affect of admixture on the SFS, when I came across an inconsistency (likely my misunderstanding) between ms and scrm. If I run the following code:

scrm 20 100 -t 142 -r 195 10001 -I 3 0 20 0 -n 2 0.5 -n 3 0.25 -es 0.08 2 .2 -en 0.08 4 .025 -ej 0.08 4 1 -ej 0.08 2 3 -ej 1 1 3
avg segsites scrm: 176

ms 20 100 -t 142 -r 195 10001 -I 3 0 20 0 -n 2 0.5 -n 3 0.25 -es 0.08 2 .2 -en 0.08 4 .025 -ej 0.08 4 1 -ej 0.08 2 3 -ej 1 1 3
avg segsites ms: 466

If I change the code to -I 3 20 20 20, the results are the similar: ms=930, scrm=896

Is this something I am misunderstanding?
thank you,
@stsmall

[paulstaab]

Thanks for the report and sorry that scrm does not seem to work correctly here.

It looks like it is sorting the events at time 0.08 in the wrong order:

scrm 20 100 -t 142 -r 195 10001 -I 3 0 20 0 -n 2 0.5 -n 3 0.25 -es 0.08 2 .2 -ej 0.08 2 3 -ej 0.08 4 1 -ej 1 1 3 --print-model
[...]
At Time 3200:
 Population Sizes:      10000      5000      2500     10000
 Growth Rates:              0         0         0         0
 Migration Matrix: 
         0         0         0         0
         0         0         0         0
         0         0         0         0
         0         0         0         0
 100% of pop 2 move to pop 3
 80% of pop 2 move to pop 4
 100% of pop 4 move to pop 1
[...]

so that no lines are moved to pop 4. I confirmed this by removing the -ej 0.08 4 1. Then, scrm continues to work, which it shouldn't because lines in pop 3 & 4 can not merge anymore if any lines are assigned to pop 4.

I will look into why this happens, but this may take a while.

I the meanwhile, you can use the -eps argument of scrm. The command

scrm 20 100 -t 142 -r 195 10001 -I 3 0 20 0 -n 2 0.5 -n 3 0.25 -eps 0.08 2 1 .2 -ej 0.08 2 3 -ej 1 1 3 --print-model

produces

At Time 3200:
 Population Sizes:      10000      5000      2500
 Growth Rates:              0         0         0
 Migration Matrix: 
         0         0         0
         0         0         0
         0         0         0
 80% of pop 2 move to pop 1
 100% of pop 2 move to pop 3

and numbers of segsites that look comparable with ms.

Hope this helps!

Best wishes,
Paul

[paulstaab]

Note for myself: Model::addSingleMigrationEvent stores all events in a matrix for each time point, and thereby causes the events to be executed in basically random order. This needs to be reworked completely.