When splitting a node of a regression tree into two subsets, how are the variable and its split value determined?

The variable that has a split value with the smallest regularization penalty is selected.

At each step, the selected variable and split value result in the most balanced split of the observations in the parent node into two children nodes.

At each step, the variable and its split value that are selected result in the largest decrease in the variance of the outcome variable.

children regions.