Mysterious CFUniform error

[tgvaughan]

There have been a number (2 to date) of analyses which have been cut short by CFUniform throwing an IllegalStateException due to the proposed ARG being invalid.

This may not actually be CFUniform's fault, as CFUniform is one of the only operators that still explicitly checks for validity of the proposal. (It may have been invalid prior to CFUniform making the state changes.)

Update: this may be related/caused by to https://github.com/CompEvol/beast2/issues/846.

[fifthguy]

Hi,

I suppose this makes 3:

"Fatal exception: Something is broken: CFUniform proposed illegal state!"

Can you think of a workaround, or where the issue might be?

Thank you!

Best, T

[tgvaughan]

Sigh, I hoped this had been accidentally fixed by a subsequent commit somehow. Are you able to reproduce this issue with any regularity? The reason this hasn't been fixed is that I haven't actually seen it for years. A minimal example that produces the error would hugely help, so if you have stubled on that I'd be incredibly grateful if you could attach it here!

[noemiebk]

Hi, I don´t know if you are still looking for an example for this error but I just ran into it. I have attached the .xml file (renamed .xml.txt as GitHub would not let me upload the .xml directly) and the nohup file showing the error message (nohup_errorCFUniformIllegal.txt) Two other runs with the same output file ended with another error (nohup_errornull.txt). I would be very grateful if you have any hint how I could solve this problem. best, Noemie 8treemodel.xml.txt

nohup_errorCFUniformIllegal.txt nohup_errornull.txt

[tgvaughan]

Thanks Noemie. Do you happen to have the state file for one of the runs that ended prematurely?

I actually encountered a the same error recently in someone else's analysis, and it actually turned out to be an upstream problem in beast: numerical errors creep into th exponential growth population model when the growth rate of an exponential population becomes tiny (i.e. < 1e-10 or so) but non-zero. Since bacter uses the population model (whichever one you choose) in proposing some of the state changes, these numerical errors can break things. I suspect this is what is happening in your case too. We're working on a solution in beast itself (see https://github.com/CompEvol/beast2/issues/846) , but in the meantime you might be able to avoid the problem by setting a small lower bound on the growth rate (say 1e-5).