Error: Log-likelihood not converged | Complex Trait Genetics Forum

abi
New Member

Posts: 1

Error: Log-likelihood not converged Apr 18, 2017 4:13:01 GMT

Quote

Post by abi on Apr 18, 2017 4:13:01 GMT

Hi
I am new to GCTA. I am getting the following error constantly

Error: Log-likelihood not converged (stop after 100 iteractions).

You can specify the option --reml-maxit to allow for more iterations.

However when i use --reml-maxit 1000 it doesnot show any output:

Analysis started: Mon Apr 17 23:10:12 2017

Options:

--reml-maxit 1000

--grm /home/agr75/data/analysis/2016_Shoepeg/Genotypes/Heritability_MAF/2016_shoepeg_MAF_10

--pheno 2016_PlantHt_phenos.phen.txt

--out /home/agr75/data/analysis/2016_Shoepeg/Genotypes/Heritability_MAF/2016_shoepeg_MAF_10_PlantHt

Analysis finished: Mon Apr 17 23:10:12 2017

Computational time: 0:0:0

I am getting this error only for one particular phenotype (other phenotype ran perfectly).

jxb
New Member

Posts: 1

Error: Log-likelihood not converged Jun 30, 2021 21:54:11 GMT

Quote

Post by jxb on Jun 30, 2021 21:54:11 GMT

Hi, I wonder if you have any solution for this issue. I also go into the same issue "Log-likelihood not converged (stop after 100 iteractions)". Do you know what I need to do if I want it to output a hsq file no matter it converges or not?

redrice
New Member

Posts: 2

Error: Log-likelihood not converged Aug 25, 2021 17:53:00 GMT

Quote

Post by redrice on Aug 25, 2021 17:53:00 GMT

I have the same issue with a binary trait. In the .log file, I can see the log-likelihood became fairly stable after iteration# 16. After that, it jumps among a few values differ only by 2 (the log-likelihood itself is more than 40000). I wonder if there is a way to set the converge threshold so I can force it to converge. I really need the .hsq output file because, though I can find the variance explained by each components in the .log file, and I can calculate the heritability from that, it seems no simple way to do the transformation to the underlying scale if a prevalence is given.

redrice
New Member

Posts: 2

Error: Log-likelihood not converged Aug 30, 2021 19:03:11 GMT

Quote

Post by redrice on Aug 30, 2021 19:03:11 GMT

An update: I tried to use --reml-no-constrain, and it did converge. Some components variances are negative, though usually small. Not sure how to compare this with the results I obtained without the option in other (related) traits.