For every positive integer $k$, write $$\ln_k(x)=\log\log\cdots\log(x),$$ where the logarithm has been taken $k$ times. For instance, $$\ln_1(x) =\log x,\ln_2(x) =\log\log x,\cdots.$$ Define a function $f:\mathbb{N}\rightarrow \mathbb{R}$ by $$f(n)=n\prod_{m=1}^{k[n]}\ln_m(n),$$ where $k[n]$ is the largest integer, s.t. $\ln_{k[n]}(n)≥ 1$. Does the series $\sum_{n=3}^{\infty}\frac{1}{f(n)}$ converge?
I tried the condensation test (since 1/f(n) presents n in the denominator), so I considered the sum of (2^n)1/f(2^n): here the isolated 2^n are cancelled out. Then the denominator(from a certain n) presents the term loglog(2^n)=nlog(2*((log(2))^1/n)), so I have again an isolated n in the denominator: I do the condensation test again, considering (2^n)*(2^(2^n))*f(2^(2^n)). So, from a certain n, in the denominator I have logloglog(2^(2^n)), from which I can take out n. I proceed this way doing condensation test infinite time, and I end up with an infinite sum that I don't know wether it converges or diverges, nor if I can write in a compact way, and I don't know if I am on the right way to solve the problem.