Jump to content


ETA shrinkage

  • Please log in to reply
2 replies to this topic

#1 Go-Wun Choi

Go-Wun Choi


  • Members
  • Pip
  • 1 posts

Posted 12 July 2018 - 06:33 AM

Dear Members of Certara Forum


I have an issue during the development of popPK modeling. I tried to fix the parameters that have large ETA shrinkage to zero. However when I fixed it to zero, the model excution always failed with following error message : out of range final -LL value : 1.0E+70

In NONMEM, I can fix the ETA like this example code :


How can I handle this issue?


And, is there the way to check ETA bar and its p-value in NLME? (I have checked it in manual)

I have an issue during the development of popPK modeling. I tried to fix the parateres that have large ETA shrinkage to zero. However when I fixed it to zero, the model excution always failed with following error message like this : out of range final -LL value : 1.0E+70

In NONMEM, I can fix the ETA like this example code :

  • smouksassi1 likes this

#2 Simon Davis

Simon Davis

    Advanced Member

  • Administrators
  • 1,111 posts

Posted 12 July 2018 - 09:26 AM

Dear Go-Wun, Instead of trying to fix to zero,  I wold first try to remove random effects from the parameter with the worst shrinkage, as I have done below for CL2




alternatively if I understand you correctly and you still want to freeze a parameter it's quite simpe either in the UI or the code; below I froze (fixed) V2




fixef(tvV = c(, 5.35391, ))
    fixef(tvV2(freeze) = c(, 10.9549, ))   # commenting that v2 is frozen
    fixef(tvCl = c(, 2.249, ))
    fixef(tvCl2 = c(, 2.56732, ))



Attached Files

#3 smouksassi1


    Advanced Member

  • Members
  • PipPipPip
  • 169 posts
  • LocationMontreal

Posted 12 July 2018 - 05:32 PM

Dear Go Wun Choi


Can you provide a reproducible example on our end that reproduces  this -eLL error ?

There is ways to fix a random effect to zero or enable disable it.


In Phoenix PML each parameter has a name 
so you can just comment out the parameter that you don't need without afecting the other part of the model there is no sequential numbering of OMEGA so you will not have to edit your other parts of code
CL =tvCL #*exp (nCL)


some alrogrithms like SAEM or QRPEM do not like 0 variance fixing you might need to fix to a small value instead


if you have all of your n on one block or line you might need to edit the ranef

ranef( nCL,nV)






0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users