java.lang.Error when running discrete phylogeographic analyses with asymmetric rate matrices

[Rebecca Duncan]

Hi BEAST2 users,

I am using BEAST2.7.7 to do phylogeographical modeling of poliovirus transmission. When I started, I used symmetric rates and never had an issue with my runs, but I switched to using asymmetric rate matrices because it is more realistic for my data, and because I wanted to be able to calculate Bayes Factors for transitions from each location to each other location as a measure of confidence in a particular path/direction. The problem is that with nearly all of my runs, I am receiving the error message in my stderr log:

0 NaN
1 NaN
2 NaN
3 NaN
4 NaN
5 NaN
6 NaN
7 NaN
8 NaN
9 NaN
10 NaN
java.lang.Error: randomChoiceUnnormalized falls through -- negative components in input distribution?
Please report error to Marc

I found a post in the BEAST2 github issues about this error (https://github.com/CompEvol/beast2/issues/1135), but it does not resolve my issue. I get the same error repeatedly in some runs, and eventually the runs fail without reaching the MCMC I requested. Also, analyses with the same xml file don't always fail at the same point in the chain, sometimes they run for shorter or longer before failing.

Has anyone else had this issue and resolved it? At this point I am running analyses several times until I get one with decent traces and ESS values, but I don't know if that is a valid solution.

Thanks in advance for your help!

Rebecca