# best and worst case number of key comparisons of an algorithm

Consider the following algorithm:

function randomQueries(A, n <- A.size)
if n > 1 then
p <- n - 1;
x <- partition(A, p);
randomQueries(A[0,1,...,x-1], x)
randomQueries(A[x+1,...,n-1], n-1-x);
for y = 1 to x do
for z = x + 1 to n - 1 do
query(A, y, z);
else if n = 1 then
query(A,0,0);


partition is Hoare's partition-routine (so after the partition, all keys in A[0,...,x-1] are <= A[x] and all keys in A[x+1,...,n-1] are >= A[x]. But I think it only really matters that partition performs n key comparisons. Also, query(A, y, z) uses one key comparison.

Deduce with justification, asymptotically tight bounds for the worst and best case number of key comparisons.

I think the best case is $$\Theta(n^2)$$; choosing $$x=n/2$$ each time (which is possible if you arrange the array so that the last element of each subarray visited by the randomQueries function is always the median of the subarray) shows that the best case number $$T^{best}(n) \leq n + n/2(n/2-1) + T^{best}(n/2) + T^{best}(n/2-1) \leq n^2 + 2T^{best}(n/2)$$ for n large enough, which simplifies to $$O(n^2).$$ Also, each call takes $$n$$ key comparisons and there are $$\Omega(n)$$ calls because as long as $$n > 1,$$ there will be a recursive call and each call "removes" one index (x) each time.

Similarly, if one always chooses the last element ($$x=\frac{n}2$$), the worst case runtime should be at least $$n + T^{worst}(n-1)$$, which evaluates to $$\Omega(n^2).$$ I'm not sure how to find an upper bound for the worst case.

• What are the possible values of $x$? Oct 11, 2021 at 10:53
• What is randomQueries ? Mar 10 at 13:02

Denote by $$T(n)$$ the number of comparisons on a list of length $$n$$. Then $$T(0) = 0$$, $$T(1) = 1$$, and $$T(n) = n + T(x) + T(n-1-x) + x(n-1-x)$$ for some value of $$x$$ that could depend on the array.

Let us prove by induction that $$T(n) = n(n+1)/2$$, whatever $$x$$ is. This holds when $$n = 0$$ and $$n = 1$$. For $$n > 1$$, we have $$T(n) = n + \frac{x(x+1)}{2} + \frac{(n-1-x)(n-x)}{2} + x(n-1-x) = \frac{n(n+1)}{2}.$$ You can also prove this combinatorially: every two elements get compared, including an element and itself.