Let the probability of an event \(A\) occurring in a Bernoulli experiment be \(P:=p(A)\). We define members of a real sequence \((F_n)_{n\in\mathbb N}\) as follows:
Then it is almost certain that the sequence members \(F_n\) will approximate the probability \(P\) with virtually any accuracy, if \(n\) is large enough. Formally,
\[\lim_{n\to\infty}p(|F_n(A)-P|\le \epsilon)=1\]
for arbitrarily small (but fixed) real number \(\epsilon > 0\). We can also say that the relative frequencies of an event in a Bernoulli experiment (if repeated a large number of times) approximate the probability of that event:
\[F_n(A)\approx p(A).\]
Proofs: 1