Fermat said:
I thought about showing the sequence of terms does not tend to 0. Can you help with that?
I think that's a good idea. Notice that $\displaystyle \begin{align*} \frac{a^n}{n^2\,3^n} = \left( \frac{1}{n^2} \right) \left( \frac{a}{3} \right) ^n \end{align*}$. Notice that $\displaystyle \begin{align*} \frac{1}{n^2} \to 0 \end{align*}$ as $\displaystyle \begin{align*} n \to \infty \end{align*}$.
If $\displaystyle \begin{align*} \left| a \right| < 3 \end{align*}$ then $\displaystyle \begin{align*} \left| \frac{a}{3} \right| < 1 \end{align*}$ and $\displaystyle \begin{align*} \left( \frac{a}{3} \right) ^n \to 0 \end{align*}$ as $\displaystyle \begin{align*} n \to \infty \end{align*}$. Thus $\displaystyle \begin{align*} \left( \frac{1}{n^2} \right) \left( \frac{a}{3} \right) ^n \to 0 \cdot 0 = 0 \end{align*}$ as $\displaystyle \begin{align*} n \to \infty \end{align*}$ and so may converge.
If $\displaystyle \begin{align*} a = 3 \end{align*}$ then $\displaystyle \begin{align*} \frac{a}{3} = 1 \end{align*}$ and so $\displaystyle \begin{align*} \left( \frac{a}{3} \right) ^n = 1^n \to 1 \end{align*}$ as $\displaystyle \begin{align*} n \to \infty \end{align*}$, and thus $\displaystyle \begin{align*} \left( \frac{1}{n^2} \right) \left( \frac{a}{3} \right) ^n \to 0 \cdot 1 = 0 \end{align*}$ as $\displaystyle \begin{align*} n \to \infty \end{align*}$ and so may converge.
If $\displaystyle \begin{align*} a = -3 \end{align*}$ then $\displaystyle \begin{align*} \frac{a}{3} = -1 \end{align*}$ and so will oscillate between -1 and 1. Since the product of a bounded function (such as this one) and a function that goes to 0 has a limiting value of 0, that means $\displaystyle \begin{align*} \left( \frac{1}{n^2}\right) \left( \frac{a}{3} \right) ^n \to 0 \end{align*}$ as $\displaystyle \begin{align*} n \to \infty \end{align*}$ and so may converge.
Now finally, if $\displaystyle \begin{align*} \left| a \right| > 3 \end{align*}$, then $\displaystyle \begin{align*} \left| \frac{a}{3} \right| > 1 \end{align*}$ and thus $\displaystyle \begin{align*} \left( \frac{a}{3} \right) ^n \to \pm \infty \end{align*}$ as $\displaystyle \begin{align*} n \to \infty \end{align*}$ (depending on the sign of $\displaystyle \begin{align*} a \end{align*}$).
That means $\displaystyle \begin{align*} \left( \frac{1}{n^2} \right) \left( \frac{a}{3} \right) ^n \end{align*}$ gives an indeterminate form $\displaystyle \begin{align*} 0 \times \infty \end{align*}$ and so has to be transformed.
Rewrite it as $\displaystyle \begin{align*} \left( \frac{1}{n^2} \right) \left( \frac{a}{3} \right) ^n = \frac{\left( \frac{a}{3} \right) ^n}{n^2} \end{align*}$ which is now an indeterminate form $\displaystyle \begin{align*} \frac{\infty}{\infty} \end{align*}$ and so L'Hospital's Rule can be used.
$\displaystyle \begin{align*} \lim_{n \to \infty} \frac{\left( \frac{a}{3} \right) ^n}{n^2} &= \lim_{n \to \infty} \frac{\left( \frac{a}{3} \right) ^n \ln{ \left( \frac{a}{3} \right) }}{2n} \textrm{ by L'Hospital's Rule} \\ &= \lim_{n \to \infty} \frac{\left( \frac{a}{3} \right) ^n \left[ \ln{ \left( \frac{a}{3} \right) } \right] ^2 }{2} \textrm{ by L'Hospital's Rule again} \\ &\to \infty \textrm{ as } n \to \infty \end{align*}$
Since this limit is not 0 when $\displaystyle \begin{align*} |a| > 3 \end{align*}$, the series can not possibly converge when $\displaystyle \begin{align*} |a| > 3 \end{align*}$. Thus the series is divergent if $\displaystyle \begin{align*} |a| > 3 \end{align*}$.