the distribution of a sum by convolution

Add up two independent random quantities — two dice, two waiting times, two measurement errors — and you get a new random total. Its distribution is rarely just the average of the two; the spread changes shape in a specific way. Convolution is the operation that gives the distribution of the sum S = X + Y of two independent variables.

The idea is to account for every way the total can land on a given value. For the sum to equal s, whatever X turns out to be, Y must be exactly s - X. So you sweep X over all its values, and for each one weight by the chance Y fills the gap, then add. For independent discrete variables, p_S(s) = sum over all k of p_X(k) times p_Y(s - k). For independent continuous variables, the sum turns into an integral: f_S(s) = integral over all x of f_X(x) times f_Y(s - x) dx. That sliding-and-multiplying-and-adding operation is exactly what mathematicians call a convolution.

Convolution explains why sums smooth out and bunch toward the middle: extreme totals need both parts to be extreme at once, which is unlikely, so the middle fills in. Add many independent pieces and this smoothing drives the bell shape of the central limit theorem. Caution: the formula here requires independence; for dependent variables you must use the full joint distribution instead.