[R-sig-ME] differences between lmer and glmmTMB - follow up

[Don Cohen]

[Summary: two models that differ only in one added fixed effect
 the deviance of the one with extra fixed effect coming out higher]

I think this is a related problem:

> mfull <- lmer(data = strict, LN_GCS ~ poly(hour,2) + poly(group_size,2) + poly(age,2) + poly(dom_rank,2) + Pregnancy + takeover + (1|ID) + (1|groupid) + (1|AssayNum))

> summary(mfull)
Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: LN_GCS ~ poly(hour, 2) + poly(group_size, 2) + poly(age, 2) +
    poly(dom_rank, 2) + Pregnancy + takeover + (1 | ID) + (1 |
    groupid) + (1 | AssayNum)
   Data: strict
REML criterion at convergence: 2088.6
...
> m <- lmer(data = strict, LN_GCS ~ poly(hour,2) + poly(group_size,2) + poly(age,2) + poly(dom_rank,2) + Pregnancy + (1|ID) + (1|groupid) + (1|AssayNum))

> summary(m)
Linear mixed model fit by REML. t-tests use Satterthwaite's method [
lmerModLmerTest]
Formula: LN_GCS ~ poly(hour, 2) + poly(group_size, 2) + poly(age, 2) +
    poly(dom_rank, 2) + Pregnancy + (1 | ID) + (1 | groupid) +
    (1 | AssayNum)
   Data: strict
REML criterion at convergence: 2086.4
...

The REML criterion is supposeed to be -2 x loglik (deviance), right?
How can it be lower for m than for mfull when mfull contains all the same 
effects (plus one more) ?  

> AICctab(mfull, m, weights = T)
      dAICc df weight
m      0.0  15 0.89
mfull  4.2  16 0.11

If I use glmmTMB I get the expected results - same loglik, AIC
different by 2