No, for two reasons:
1) x1 and x2 are not equal and have no apparent relationship
2) matrix transformations are non-associative
The only thing I can think of is working backwards from the output of the first chunk, removing the rotation, removing the scale, then adding the new rotation and scale back.