TITLE: A theorem by Harald Cramér?
QUESTION [7 upvotes]: In the paper “On the order of magnitude of the difference between consecutive prime numbers” by Harald Cramér there is the following statement:

Suppose $\{X_n\}_{n=2}^\infty$ is a sequence of independent random variables, such that $X_n \sim Bern(\frac{1}{\ln(n)})$.
  Then $\lim_{n \to \infty} \sup |\frac{\sqrt{\ln(n)}(\Sigma_{i=2}^n X_i - li(n))}{\sqrt{2n \ln(\ln(n))}}| = 1$

However, he does not prove this result there, but rather states, that it is proved in his paper “Prime numbers and probability” (which I could not find)
My question is:
How can this statement be proved?
Probably, it has something to do with the Law of Iterated Logarithm, but I do not know for sure ...

REPLY [11 votes]: This is a special case of a general law of the iterated logarithm for non-iid random variables (r.v.'s), which states the following: 

Suppose that $Y_1,Y_2,\dots$ are independent zero-mean r.v.'s, $S_n:=\sum_1^n Y_i$, $B_n:=Var\, S_n\to\infty$, $|Y_n|\le M_n\in(0,\infty)$, and $M_n=o((B_n/\ln\ln B_n)^{1/2})$. Then 
  $$\limsup_n\frac{S_n}{\sqrt{2B_n\ln\ln B_n}}=1
$$
  almost surely. 

See e.g. V. Petrov, Ch. X, Theorem 1. This theorem is due to Kolmogorov (1929). 
(Just in case, here is a reference to Cramér's paper: 
Cramér, H. 1935 Prime numbers and probability. Skand. Mat.-Kongr. 8, 107--115. 
I found it in Granville's paper.)