Normal random variables
Cramér's theorem is the result that if X and Y are independent real-valued random variables whose sum X + Y is a normal random variable, then both X and Y must be normal as well. By induction, if any finite sum of independent real-valued random variables is normal, then the summands must all be normal.
Thus, while the normal distribution is infinitely divisible, it can only be decomposed into normal distributions (if the summands are independent).
Contrast with the central limit theorem, which states that the average of independent identically distributed random variables with finite mean and variance is asymptotically normal. Cramér's theorem shows that a finite average is not normal, unless the original variables were normal.
Cramér's theorem may also refer to another result of the same mathematician concerning the partial sums of a sequence of independent, identically distributed random variables, say X1, X2, X3, …. It is well known, by the law of large numbers, that in this case the sequence
converges in probability to the mean of the probability distribution of Xk. Cramér's theorem in this sense states that the probabilities of "large deviations" away from the mean in this sequence decay exponentially with the rate given by the Cramér function, which is the Legendre transform of the cumulant-generating function of Xk.
- Asymptotic equipartition property
- Cochran's theorem, on decomposing sum of squares of normal distributions
- Indecomposable distribution, on decomposability
- Raikov's theorem, on the decomposition of Poisson distributions
- Infinite divisibility (probability)
- Lukacs, Eugen: Characteristic functions. Griffin, London 1960 (2. Edition 1970), ISBN 0-85264-170-2.