Traditional touch-typing suggests that each finger should handle a column of keys, but of course, the standard keyboard doesn't align columns, and so that's spawned a number of models that arrange keys in a
square grid. Fine - it's a nice idea, and it seems to enhance the implementation of traditional typing technique.
After analyzing my own technique, I found out that it
a) is not traditional
b) doesn't work as well on a square grid layout
The attached picture shows what I think is actually happening - the keys are acting as a
hexagonal grid. What's the important difference? In square grids, the vertically nearby keys are either n or sqrt(2)*n distance away. In hexagonal grids, the vertically nearby keys are both (approximately) sqrt(3)/2 distance away. In practical typing, many users seem to dynamically adjust which fingers hit which keys based on desired digraphs to strike, etc. Example: for the digraph "ER", I obviously am going to use my middle and index fingers. But for "TR", I might also be inclined to use those two fingers, rather than strike both with the index! Normally I press O with the ring finger, but for "PO" I use ring+middle.
The need for such techniques is reduced with optimized layouts like Colemak, etc., but I wonder if a dynamic layout scheme that threw the one-to-one finger-to-key mapping concept out the window might come up with something even better.