Picture a formula as a machine with an input dial. Most testing happens with the dial in the middle: x=3, x=10, a list of 20 numbers. Edge cases sit at the ends of the dial and at its odd notches: x=0, x=1, x=−1, very large x, the empty list, a triangle squashed flat, the endpoint of an interval. These are the inputs where some step quietly stops making sense, such as dividing by a count that is now 0 or taking the square root of something that just went negative.
Some rules survive their edges and some do not. A polygon with n sides has n−3 diagonals from each vertex: a hexagon gives 6−3=3, and the edge case n=3 gives 0, which is right, because a triangle has no diagonals. The identity xx2=x, by contrast, looks fine for every number you try until x=0, where the left side becomes 00 and has no value. The finding is an exclusion: the identity holds for every x=0.
An edge case is not the same thing as a counterexample, even though the same number can play either part. Testing x=0 in x1 inspects a rule at a risky spot; the rule may well survive, or need a stated exception. A counterexample is an input that makes a 'for all' claim false, such as x=21 against 'every number is less than its square.' The question to ask is whether you are probing behavior or trying to break a claim.
Think of it like this. Testing edge cases is like testing a bridge with the heaviest truck, an empty road and a gale, not just ordinary weekday traffic. A design that only ever saw average loads tells you nothing about the moments that bring bridges down. Unlike a bridge, a formula can also fail at a tiny input, such as zero.
Core idea. Edge cases are the extreme or special inputs — zero, infinity, empty, the boundary — where a formula or argument can behave differently.