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

[Don Cohen]

I wonder whether I'm missing something fundamental.
Should these two calls produce very similar results?

> mbase <- lmer(data = strict, LN_GCS ~ hour + group_size + age + dom_rank + Pregnancy + (1|ID) + (1|groupid) + (1|AssayNum))
> TMBmbase <- glmmTMB(data = strict, LN_GCS ~ hour + group_size + age + dom_rank + Pregnancy + (1|ID) + (1|groupid) + (1|AssayNum))

And similarly these two?

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))

TMBmfull <- glmmTMB(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))

The results do look very similar with one glaring exception:

summary(mfull) says
REML criterion at convergence: 2088.6
summary(mbase) says
REML criterion at convergence: 2141.6

AICctab shows a difference in AIC of 42.7 (df 16 and 11)

whereas

summary(TMBmfull)
     AIC      BIC   logLik deviance df.resid
  2133.0   2206.7  -1050.5   2101.0      727

summary(TMBmbase)
     AIC      BIC   logLik deviance df.resid
  2135.4   2186.1  -1056.7   2113.4      732

for a difference in AIC of 2.4 !
(though AICctab shows a difference of 2.0 ?)

Is there some explanation?  Should I believe one and not the other?