The whole idea is that we’re doing a digitby digit sort. We are going to do it from least significant to mostsignificant.
Let's think of here are n numbers and the numbers are binary, each integer is b bits long, so in other words the range of a number is 0 to 2b-1. We split each integer into b/r digits, that means my number as being in base 2r. I get a bunch of digits in base 2r. This b/r is the number of rounds and 2r is the maximum value of each round. If we choose counting sort algorithm for each round’s sorting. The time complexity is O((b/r)*(n+2r)).We choose a right r to ensure the function minimum. Differentiate with respect to r and set to 0, we can get r <= lgn, then the function changed to O(bn/lgn),and then we make 2b = nd, such that b = d*lgn, so at last if d = O(1), then we get a linear time sorting algorithm.