Since $a_n=\sum_{k=1}^{n}\frac{1}{k^2}$ is obviously an increasing sequence (I prefer to avoid the letter $i$ as a summation index, since it can be easily mistaken with the imaginary unit) it is enough to show that $\zeta(2)=\sum_{n\geq 1}\frac{1}{n^2}$ is less than two. Fourier theory ensures that $$\frac{\pi-x}{2}=\sum_{n\geq 1}\frac{\sin(nx)}{n}$$
for any $x\in(0,2\pi)$, with such identity holding uniformly over compact subsets of $(0,2\pi)$. Since
$$ \int_{0}^{2\pi}\sin(nx)\sin(mx)\,dx = \pi \delta(m,n)$$
we have
$$ \int_{0}^{2\pi}\left(\frac{\pi-x}{2}\right)^2\,dx = \pi \zeta(2)$$
and the given claim is equivalent to
$$ \int_{0}^{2\pi}(\pi -x)^2\,dx = 2\int_{0}^{\pi}(\pi -x)^2\,dx = 2\int_{0}^{\pi}x^2\,dx < 8\pi $$
or to
$$ \pi < 2\sqrt{3} $$
which is equivalent to the perimeter of the regular hexagon circumscribed to a unit circle is greater than the perimeter of the unit circle. This follows from the fact that if $A,B$ are bounded convex sets in $\mathbb{R}^2$ with $B\subsetneq A$, then $\mu(\partial B)<\mu(\partial A)$.
Obviously you cannot do any induction if you were using $\leq 2$ as the induction step fails.
I tried coming up with why $\frac1n$ could be a reasonable choice, but had no luck.
– b00n heT Oct 04 '18 at 12:00