[R-sig-ME] Obscure convergence warning in optwrap() - convergence code 3 from bobyqa

[Michael Cone]

Using lmer (1.1.7, compiled today from the github master branch), I got 
the following warning:

In optwrap(object at optinfo$optimizer, ff, x0, lower = lower, control = 
control$optCtrl,  :
   convergence code 3 from bobyqa: bobyqa -- a trust region step failed 
to reduce q

These warnings are thrown while calculating bootstrap confidence 
intervals with confint() for a simple repeated-measures model. The 
warnings seem to depend on the numerical value of the dependent variable 
(see code below).

Likelihood profiling also throws warnings:

1: In optwrap(optimizer, devfun, x at theta, lower = x at lower, calc.derivs 
= TRUE) :
   convergence code 3 from bobyqa: bobyqa -- a trust region step failed 
to reduce q
2: In optwrap(optimizer, par = start, fn = function(x) dd(mkpar(npar1,  
:
   convergence code 3 from bobyqa: bobyqa -- a trust region step failed 
to reduce q
3: In optwrap(optimizer, par = thopt, fn = mkdevfun(rho, 0L), lower = 
fitted at lower) :
   convergence code 3 from bobyqa: bobyqa -- a trust region step failed 
to reduce q

Someone encountered the same warnings in 2013 and, on Ben Bolker's 
suggestion, tracked down the Fortran code responsible, but apparently 
couldn’t identify the problem.
http://stats.stackexchange.com/questions/89945/meaning-of-a-convergence-warning-in-glmer

Here's an example:

set.seed(12345)
## 30 subjects, 1000 repeated measures on each
subj.true <- rep(rnorm(30, 16000, 3500), each=1000)
df <- data.frame(subj = as.factor(rep(LETTERS[1:30], each=1000)),
                  subj.true = subj.true,
                  meas = subj.true + rnorm(30000, 0, 1200))
## this works without errors
fm1 <- lmer(meas ~ 1 + (1|subj), data=df)
boot.pp <- confint(fm1, method="boot", nsim=200)
## this throws 12 identical 'convergence code 3' warnings
df$pct <- 100*(df$meas - df$subj.true)/df$subj.true
fm2 <- lmer(pct ~ 1 + (1|subj), data=df)
boot.pp2 <- confint(fm2, method="boot", nsim=200)
## likelihood profiling gives 3 warnings
lr.pp2 <- confint(fm2)

Is this something obvious/known? I apologize if so, but I didn't find 
much regarding this on the web.

> sessionInfo()
R version 3.0.2 (2013-09-25)
Platform: x86_64-pc-linux-gnu (64-bit)

locale:
  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C
  [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8
  [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8
  [7] LC_PAPER=en_US.UTF-8       LC_NAME=C
  [9] LC_ADDRESS=C               LC_TELEPHONE=C
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base

other attached packages:
[1] lme4_1.1-7       Rcpp_0.11.2      Matrix_1.1-2     data.table_1.9.2

loaded via a namespace (and not attached):
  [1] boot_1.3-9      compiler_3.0.2  grid_3.0.2      lattice_0.20-24
  [5] MASS_7.3-29     minqa_1.2.3     nlme_3.1-113    nloptr_1.0.4
  [9] plyr_1.8.1      reshape2_1.4    splines_3.0.2   stringr_0.6.2
[13] tools_3.0.2


Michael