How would you describe the runtime complexity if I have an algorithm that at each step, the size of the array reduced by an exponentially-increasing amount?
For example, for each step in the algorithm, it is actually processing n - 2^k items, where k is the number of steps it has run so far. So for the first step, we process n - 1 items, the second n - 2, the third n - 4, the fourth n - 8, etc. until the subtracted amount exceeds n and the algorithm is done.
Suppose for processing $n$ items of your array, you needs $\mathcal{O}(n)$ time. Let $T_k(n)$ be running time of the algorithm:
$$T_k(n) = \begin{cases} T_{k-1}(n-2^k) + Θ(n-2^k) & \text{ if } n \geq 2^k, \\ 1 & \text{ Otherwise } . \end{cases}$$
Note that $k=\log n$ so the number of steps of recursion is at most $\log n$.
It will be: $$T(n,k)=\sum_{i=1}^k n-2^i=kn-\sum_{i=1}^k2^i=kn-(2^{k+1}-1)=kn-2^{k+1}+1$$
If you want something without $k$, notice that $k\le \log(n)$ since otherwise you will process negative amount of items in some iteration. Then, this means that $T(n,k)=kn-2^{k+1}+1\le kn \le n\log(n)$ and hence $T(n,k)=O(n\log(n))$.
This is also tight, assuming that $k$ is large enough. Indeed, if we assume that $k=\log(n)$ then $2^{k+1}=2n$, and hence $T(n,k)=n\log(n)-2n+1=\Theta(n\log(n))$
