Lmer failue to converge in random slope linear mixed model I am fitting a random slope liner mixed model with rank as the DV and season and position as fixed effects. I also have a random intercept for season and a random slope for season*round.
Model1.0<-lmer(Rank~Season+Position +(1|Season)+(1+Season|Round),     
data=mydata, REML=T)

The model without the random slope worked fine however, I get some warning messages when the model is run with the season*round random slope component.
Warning messages:
1: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
  Model failed to converge with max|grad| = 1.53894 (tol = 0.002,component1)
2: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
  Model is nearly unidentifiable: very large eigenvalue
 - Rescale variables?;Model is nearly unidentifiable: large eigenvalue ratio
 - Rescale variables?

I attempted to rescale the variables in mydata with the following code (I found it online after searching these warning messages)
numcols<-grep("^c\\.",names(mydata))
mydatas<-mydata
mydatas[,numcols]<-scale(mydatas[,numcols])
Model1.0s<-update(Model1.0,data=mydatas)

After running the model again with this updated data, R still returned the same warning messages. The rescaling appeared to work so I'm not sure why R is still requesting that I rescale my data. Any suggestions?
In regards to the first warning message, after searching this and other forums I tried different optimizers ("bobyqa" and "Nelder Mead") but neither seemed to work. Are there any other suggestions for getting around the failure to converge warning?
I am using the lmer package and have 104982 rows of data. There are 13 seasons in the analysis and approx. 286 rounds (~22 per season). 
 A: (Season|Round) specifies that the effects of Season vary across levels of Round which seems to be a bit over-parametrized for this particular problem given the numerical issues reported. A potential remedy is to lower the number of estimated parameters to: (1|Season/Round). This would be equivalent to (1|Season) + (1|Season:Round) and correspond to a nested design. As an alternative one might also consider using: (1|Season)+(0+Season|Round) where in this case the model will not estimate correlations between the slopes and the intercepts. 
Finally as the error message suggests you might want to $z$-transform your original features to unit variance and mean zero. That can be easily done using scale so your new Position for example would be mydata$PositionZ <- scale(mydata$Position). Notice that $z$-transforming your variable will change the interpretation of your $\beta$ coefficients. One unit higher of the $z$-transformed variable will correspond to one standard deviation higher in the original domain of the variable. Finally notice that $z$-transformation sense only for continuous variables; you can always $z$-transform discrete variables but the scaling will be meaningless. :D
