For different distributions a different number of points is expected to lie on the hull. One of the simplest cases to analyze is that of points uniformly distributed in a unit square.
Theorem: If n points are sampled from a uniform distribution
in a unit square, then the expected number of points on the convex
hull is at most O(log n).
Proof: First, we will break the hull into
four parts, the upper left, upper right, lower left, and lower right
hulls. This is done by breaking the hull at its leftmost, rightmost,
topmost and bottommost points. We will show that each has O(log n)
points. By symmetry, we may consider one, say the upper right hull.
Rather than bounding the number of points on the hull, we will bound a
larger quantity.
A point p(i) = (x(i), y(i) ) is said to be dominated
by another point p(j) if x(j) >= x(i) and y(j) >= y(i) . The maxima
of a point set P are the points that are not dominated by any other
point in P . Observe that each vertex of the convex hull is a maxima
(but not vice versa), and therefore there are at least as many maxima
points as convex hull vertices. So it suffices to prove that the
expected number of maxima is O(log n).
Suppose that the points are sorted in decreasing order of x coordinate. We will assume that all points have distinct coordinates to simplify things. Clearly p(i) is in the maxima if and only if p(i) has the largest y coordinate among the subset {p(1), p(2), ... p(i)}. What is the probability that this is true? The y coordinates of these points are independent and identically distributed. Therefore each of these points is equally likely to be the maximum, and so the probability that p(i) has the largest y coordinate is just 1/i. Thus, each point p(i) is a maxima with probability 1/i.

The sum is the harmonic series and it is well known that it is very
close to ln n . From the comments made earlier it follows that the
expected number of points on all the hulls is at most four times this
quantity, which is O(log n).
Lower Bound on Convex Hulls.
We can also prove a lower bound on the complexity of computing a convex hull. The simplest approach to this is to show how to reduce "sorting" to convex hulls.
Sorting, under the decision tree model of computation, is known to take Omega(n log n) time. (Because there are n! possible orders of the numbers, and you need to evaluate enough binary decisions to figure out which of the orders you are in, and lg(n!) is about n lg n). Suppose we had a black box algorithm to compute the convex hull of n points. We could then create a sorting algorithm as follows:Sort(x1, x2, x3, ..., xn)The run time of this algorithm is:forall i=1..n Let yi=xi*xi Create pi = (xi,yi) end compute convex hull (p1, p2, ... pn), which returns points in convex hull in order q1, q2, ... qn. Let qk point has smallest x-coordinate. return x-coordinates of qk, qk+1, ... qn, q1, q2, ... qk-1.
O(n) to create 2D points ??? to compute convex hull 0(n) to find the right starting place to read off the convex hull points.if computing the convex hull took less than O(n log n) time, this would be a faster than O(n log n) algorithm for sorting (which is known to be impossible). Therefore, the best possible running time for convex hull is O(n log n).