Simple argument to explain why the series for $\zeta(s)$ diverges for Re(s)=0

81 Views Asked by At

Question

I am trying to find a simple argument students not trained to university level maths that the series

$$\zeta(s) = \sum \frac{1}{n^s}$$

diverges for $\sigma=0$, where $s=\sigma+it$.

Here "simple" means avoiding Dirichlet series theory taught at university.

--

My attempt

If we look at the terms we see that $$\left|\frac{1}{n^{s}}\right|=\left|\frac{1}{n^{\sigma}}\frac{1}{n^{it}}\right|=\left|\frac{1}{n^{\sigma}}\right|$$

Writing $n^{it}$ as $e^{it\ln(n)}$ makes clear it has a magnitude of 1.

When $\sigma<0$ the magnitude of the terms grow larger than 1, making it easy to conclude the series diverges for $s<0$.

However, for $s=0$, this isn't so easy. Each term is

$$\frac{1}{n^{\sigma}} \frac{1}{n^{it}} = e^{-it\ln(n)}$$

Although the magnitude of such terms is still 1, the rotation might mean they do or don't cancel.

One case of cancelling is

$$ \sum_n e^{-i \pi n} = 1 -1 +1 -1 + \ldots $$

which oscillates, definitely doesn't converge, and many would say doesn't diverge.

Q1: Can we say that because $t\ln(n)$ grows slower than $\pi n$, the cancelling will be insufficient to prevent divergence?

I then think of another case of cancelling

$$ \sum_n e^{-i\pi \frac{n}{2}} = 1 + i -1 -i +1 +i -1 -i + \ldots $$

Q2: Does the argument change to say that because $t\ln(n)$ grows slower than $\pi n/R$, for any real $R$, the cancelling will still be insufficient to prevent divergence?