MCMC results in stepped likelihoods and stops sampling some parameters

[evolucionario]

What is the current observed behaviour?

MCMC runs look normal up to a certain point in which there is a significan jump in the likelihood, and some parameters, typically the clock rate is not sampled anymore (stays at a constant value).

How may we reproduce this bug?

The bug appears with even simple examples like the this one. Some times it runs ok but other times it results in the described behavior.

#NEXUS

[created by ips on Tue Feb 1 11:34:20 2022]

begin data; dimensions ntax=10 nchar=500; format datatype=dna missing=N gap=-; matrix t2 taccatacgaggtagcgcgcgaaaggtgattttggtcgtttaccacagcttccacctggtgttgcacgaacacgggataagtggctaacggactagacgtaacaccgatggcgctgaaggcaacgtcacgagggcttaggaacttcgcgtgtacatgtggctcgttgccctttcaggtaacccgcttgcggttgaatctccctaagattcggcaatctccacgaatataagtctgcccgatagaactaatgcaagtgaaagtgttgcgcccacctccatgcccccttgacggtaaagcaggcagccgaccaagagctaggattcgcatttggtcagccgtgcatttttgaccgaccgaatgcagaagctaaggcgctcacgtgtcgcaggagttctttagatctaagaatgattcacaggtattgacatagaactctcaaaccaggtcgtgtaattctcgtgcaaacttatggcccacgcgcaacggacgggtgagtagt t3 tcccatccgaggcagcgcttgaaaggtgattttagtcgtttacgacagcttccaccgggtgttgcacgaatacgggataagtgcctaacggactagaggtaagaccgctggcgatgaaagcaacgtcacgaggccttaggaacttcgcgtgtacagctgtctcgttgccctgtcacgccccccgcttgcggtcgaatctccctaagattcggcaatctccaccaatataagtctgctcgatagaactattgcaagtgaaagtgttgcgcccacctccatgcccccttgacggtaaatcaggaagccgaccaggagctaggattcccagttcatcagccgtgcattttcgaccgaccgaacgcagaagctaaggcggtcacgtgtcgctggagtcctttagaacaaataattattcccaggtattgacatggaactctcaaaacaagtcgtgtaattctcgtgcaaacttatggcccacgtgcaccgaacgggtgagtagt t6 tcccttccgaggcagcgcttgaaaggtgattttagtcgtttacgacagcttccaccgggtgttgcacgaatacggggtaagtgcctaacggaatagaggtaagaccggtggcgatgaaagcaacgtcacgaggccttaggaacttcgcgtgtactgctggctcgttgccctgtcacgccccccgcttgcggtcgaatctccctaagattcggcaatctccaccaatataagtctgctagatagaactattgcaagtgaaagtgttgcgcccacctccatgcccccttgacggtaaatcaggaagccgtccaggagctaggattcccagttgatcagccgtgcattttcgaccgaccgaacacagaagctaaggcggtcacgtgtcgctggagtcctttagaacaaataattattcccaggtattgacatggaactctcaaaacaagtcgtgtaattctcgtgcatacttatggcccacgtgcaacgaacgggtgagtagt t4 tcccttccgaggcagcgcttgaaaggtgattttagtcgtttacgacaccttccaccgggtgttgcacgaatacggggtaagtgcctaacggaatagaggtaagaccggtggcgatgaaagcaacgtcacgaggccttaggaacttcgcgtgtacagctggctcgttgccctgtcacgccccccgcttgcggtcgaatctccctaagattcggcaatctccaccaatataagtctgctcgatagaactattgcaagtgaaagtgttgcgcccacctccatgcccccttgacggtaaatcaggaagccgtccagcagctaggattcccagttgatcagccgtgcattttcgaccgaccgaacacagaagctaaggcggtcacgtgtcgctggagtcctttagaacaaataattattcccaggtattgacatggaactctcaaaacaagtcgtgtaattctcgtgcatacttatggcccacgtgcaacgaacgggtgagtagt t9 taccctatgatgcccccggcgtgaagtgaatgtgggccgttcactcatcacccagataatgtagggcgaagctgctaactttgcctaacataccatgcttcagtccgctagagacgttagccaggtggtgacggcctaaaaactgcgaatgtacaagtggcgatttgcccccaaacggctgcggccccgcgtgccgatgccttctgacgacgccatctatacttatattcctctgcgtagtacacctgctacaagcgagtgctctacgtaccactgcactgcctcgtctccgcaagtcagtcggcctactcggagccgcgaggcacaggatgtgagccctggattcgcgagcgaccccaagatgtagaacaggggttcaggtgacctggaaggccttgaaatatccgcctgcgtcttgaacagtagaatactactgtgcagtacaggctactagttctcttatagactgactctcctcgcttgtcctactgctgaatagg t1 taccataccctgcagcacgcgtcaagtgactttgaacggttcgcccatctctcacttaatctcgcaaaaacatggtcataatgaataacagaatttgcgcaatagagcgggcgactatagcctcgtcataagtgcctataggctgcgcgtctacgaggggctacttggcgatccacggttgtcgggtcccgttgaaaatcgctctgagtcagcatcctggactaatattagtctacgaaatacaacgaatacaagccagcgctgtgattccttcgtcacgccccgttcgccacaaaggattcagaggtccggtaactccaacacacagtttgaaagcgctgacttcctagcggtccccaagcggtagctgaaccgtaaaggtgtccaattaggccggcaaatctaacccgtcttgacggacacggaaataggactgtctaaacgatccctgtacttcccgtgcttgctggtggcgcaggtctgaagtacggcactgtatt t8 tcccattccatttagcccgcgttaaatcaatttgtgcagttagctcatcaaccacccaatgttaaccggccatggtcataatgcataacacactatgcggcagtaacctggcgacgatagaaacgtcggaagacccttagaactgcgcggggagatggagctacacgcccttggacggcttccactcccagtgggagatccctcgtactcgcgacactgtacgaatgttactctgctatattcaaccagtaaaagcgggaattttgtggcctccggcacgcctgattcacggcgggtgagtgcgctaaccggaatctccgacacactggaggtaagagctgagaccggaaccaaccacataaagtatcttatatgttgaatttggcaattagccccccaaatctactccttcttgagagcaaatagaatagtcattacaaaacggtacctttgctgctctttctaacatgcgacacacgtgtgactaactgcagtctaat t10 tgccataccaatgagccggcgtccagtcaatttgtgcggctaccacatctaccacccgatgttaaacgaacatggtcattacgcaaaacagactataccgaagtaagctgccgacaattcaaacgtgggaggaccgtaagaactgagcggacagatggcgctacacacccttcgatcgcttccgctcccggtgggagatccctcgtactcgccactctggaccaatggtacaccgatatatccaaccagtacaagcgggaactttgtgtcctcctgcacgcatcatacacggcatgtctgtgcgctgagcgggatctcagacatactcgatgtaagagcagaatgctggaccgacctcataaggtagcttacccgttcaatttcgcacttagcccgccaattgtacgcctgcttgacagacaatagaatcgacattttacaacggtacctatactcgtctttctaaatggcgatacacgtgtaacaaagtgaagtctggt t7 tgccattccaaggagccggcgtcaagtcgatttgtgctggtaccacagataccacctaatgttaaccgaccatcgtcattatgaataacggactatgccgaagtcagcagacgacaattcaaacgttggcagaccgtaaggacttagcccggagaaggggccacacgaccttcgatggcttccgctcccagtgggagaatgctggtactcaactctctggaccaatggtacaccactatatgcaaccagtacaagcgggagttttgaggcctccggtacgccgcagacacggcaagtctgtgcgctaaccgggatctctcacacactggatgtaagagctgaatgctagaccgacctcataaggtagcttatacgttgtatttctcacttagccctcccaatgtacgcattcttgacagacaatagactagtcatttcacaagggcacctgtactcctctttctaaatggcgacacacgtgtgacaaactgcagtcaagt t5 tgacatcccaaggagccggcgtcaagtcgatttgtgctggtacctaagataccacctaatgttaaccgaccatgttcattatgagtaacagactatgccgaagtcagcagacgacaattcaaacgttggcagaccgaaaggacttagcccggagatggggccacacgacctccgatcgcttccgctcccattgggagaatcctggtactcgacactctggaccaatggtacaccgctatacgcaaccagtcgaagcggcagttgtgtggcctccggtacgcctcagacacggcaagtctgtgcgctaaccgggatctatgactcactggatgcaagagctgaatgctggaccgacctcataaggtagcttatacgttgaatttcgcactttgccctcctaatctacgcattcttgacagacaataggctagtcattacacaacggcacctgtactcctctttctaaatggcgacacacgtgtgacaaactgcagtctagt ; end;

begin mrbayes; lset nst=1; constraint clade1 = t2 t3 t6 t4; constraint clade2 = t9 t1 t8 t10 t7 t5; calibrate clade1 = offsetlognormal(3.209, 4.754, 4.031) clade2 = offsetlognormal(17.66, 18.37, 1.485); prset statefreqpr=fixed(equal) nodeagepr=calibrated brlenspr=clock:uniform treeagepr=uniform(0, 100) Clockvarpr=strict clockratepr=exponential(1) topologypr=constraints(clade1, clade2); mcmc ngen=100000 samplefreq=100 nruns=1 nchains=3 starttree=parsimony; sumt filename=SimCladeAge15.nex burnin=250 contype=halfcompat; end;

Would you be able to compile and run MrBayes to test fixes to this bug?

• [ ] No, sorry.

What is the environment that you run MrBayes in?

• Operating system (including variant and release): macOS Monterrey Version 12.0.1
• Version of MrBayes: MrBayes 3.2.7a x86_64

---

Version

MrBayes 3.2.7a

Features: SSE AVX Beagle MPI Host type: x86_64-apple-darwin21.1.0 (CPU: x86_64) Compiler: clang 13.0.0

Other information that may be of use to us in resolving this issue

Screenshots of the likelihood and clock rate from Tracer v1.71

[evolucionario]

The problem persists, even with other datasets and other versions of OS X such as MacOS Ventura with Apple M1 chips.