Problematic newicks when both -c and -en in use

[standard-aaron]

I'm working on simulating selection conditioned on present-day freq (-c mode) and changing demography (-en)

I get the following extremely odd errors in tree mode:

./discoal 2 100 1 -c 0.99 -wn 0 -i 1 -T | grep '\['
[1](1:1.821753,0:1.821753);
[1](1:0.110785,0:0.110785);
[1](1:0.177698,0:0.177698);
[1](0:0.335563,1:0.335563);
[1](0:0.148914,1:0.148914);
[1](0:0.632295,1:0.632295);

(i.e., these trees look fine) However, the following ought to be equivalent, but I get completely different trees:

$ ./discoal 2 100 1 -c 0.99 -wn 0 -en 0.1 0 1 -i 1 -T | grep '\['
[1](0:0.100000,1:0.100000);
[1](0:0.100000,1:0.100000);
[1](0:0.076275,1:0.076275);
[1](1:0.100000,0:0.100000);
[1](0:0.100000,1:0.100000);

I find that making a call to -en at a particular time, it causes the vast majority of trees to coalescent at that exact time. In practice, this happens when i do e.g. "-en 0.1 0 0.55" but i'm just trying to make a poitn that "-en 0.1 0 1" ought to have the same behavior as no -en call whatsoever. Am I missing something?

[andrewkern]

looks like a bug to me... i'll start digging.

[standard-aaron]

Some more sleuthing:

when I make calls of this form:

./discoal 2 <N> 1 -c <f> -wn 0 -en <t> 0 1 -i 1 -T

I notice that approximately N * (f^2 + (1-f)^2) of the trees have a TMRCA of t. So, I think there's something going on with how it treats the case where the two samples are derived/derived or ancestral/ancestral.

Additionally, I find that when I simulate trees with a larger number of samples (i.e. n > 2), this reduces the probability that there is at least 1 coalescence at exactly this time somewhere in the tree

[andrewkern]

okay I found a bug that I think was responsible for this. Using the -wn flag with two alleles forced some simulations into a place where they were incorrectly coalescing at the population size change. This should now be fixed. Can you pull the latest version and let me know if the results look ok? thanks as always for your careful use of discoal!!

[standard-aaron]

My trajjies and trees are looking nicer now. And you're the one I should thank -- discoal is awesome

[standard-aaron]

Motion to make 'trajjy' a thing

[standard-aaron]

@andrewkern : couple of followup q's:

would this bug have affected -wn simulations where there are no -en calls (i.e. constant popsize?)

is there any difference between -ws <t> -a 0 and -wn <t>?

[andrewkern]

unlikely that the bug should have affected simulations without -en, but the patch may have led to new issues.... codewise there is a difference between -ws <t> -a 0 and -wn. in practice they may yield the same results, although I haven't benchmarked that specifically. If you want drift trajectories I'd stick with -wn

[andrewkern]

@35ajstern everything working okay now? if so i'll close out this issue

[standard-aaron]

Yes, seems to be working okay now. Thanks Andy