1
$\begingroup$

I don't think the series:

$$\sum \frac{k^{-\alpha}}{1+\alpha^{-k}}$$

converges for $\alpha \leq 0$ since the terms don't even converge to zero for those values of $\alpha$.

I have tried to do a ratio test where I get that the series converges if:

$$0 < \lim_{k \to \infty}\frac{1+\alpha^{-k}}{1+\alpha^{-(k+1)}}<1$$

But how can I find the values of $\alpha$ for which this is true? It appears that if $0 < \alpha < 1$ then the limit is $\alpha$ but if $\alpha \geq 1$ then it is $1$.

How can this be proven?

$\endgroup$
  • $\begingroup$ The series also converges for $-1 < \alpha < 0$. Write the term as $\frac{1}{k^{\alpha}} \cdot \frac{\alpha^k}{1+\alpha^k}$, and look how each of the factors behaves for different values of $\alpha$. $\endgroup$ – Daniel Fischer Jan 3 '16 at 19:17
2
$\begingroup$

Let $u_k = \frac{1}{1 + \alpha^{-k}}$ and $v_k = k^{-\alpha}$. Consider the behaviour of $(u_k)$ and $(v_k)$ for different values of $\alpha$ separately.

For $\lvert\alpha\rvert > 1$, we have $\alpha^{-k}\to 0$ and hence $u_k \to 1$. Therefore the series $\sum_k u_k\cdot v_k$ converges if and only if $\sum_k v_k$ converges for these $\alpha$. It is well-known that the series $\sum_k k^{-\alpha}$ is convergent if and only if $\alpha > 1$, so the series $\sum_k \frac{k^{-\alpha}}{1 + \alpha^{-k}}$ converges for $\alpha > 1$ and diverges for $\alpha < -1$.

For $\alpha = 1$, we have $u_k = \frac{1}{2}$ for all $k$, so the series is half the harmonic series, which diverges. For $\alpha = -1$, we have $1 + \alpha^{-k} = 0$ for odd $k$, so the terms of the series aren't even all defined then, thus we cannot properly speak of the convergence or divergence of the series $\sum_k \frac{k^{-\alpha}}{1+\alpha^{-k}}$ for $\alpha = -1$.

For $\alpha = 0$, the terms of the series as written aren't defined as well, but if we rewrite $u_k$ as $\frac{\alpha^k}{1 + \alpha^k}$, or by abuse of notation interpret $\alpha^{-k}$ as $\infty$ and then $\frac{k^{-\alpha}}{1 + \infty}$ as $0$, we get a series all of whose terms are $0$, which is clearly convergent.

For $0 < \lvert\alpha\rvert < 1$, we have $u_k \sim \alpha^k$, i.e. $u_k$ converges to $0$ exponentially, and $v_k$ grows at most polynomially, so $u_k\cdot v_k$ also converges to $0$ exponentially and the series converges. In this case, we can use the ratio or root tests to quantify the informal summary:

$$\biggl\lvert\frac{u_{k+1}v_{k+1}}{u_kv_k}\biggr\rvert = \biggl\lvert\frac{(k+1)^{-\alpha}\alpha^{k+1}}{1+\alpha^{k+1}}\cdot \frac{1 + \alpha^k}{k^{-\alpha}\alpha^k}\biggr\rvert = \lvert\alpha\rvert\cdot \biggl(\frac{k}{k+1}\biggr)^{\alpha}\cdot \frac{1+\alpha^k}{1+\alpha^{k+1}} \to \lvert\alpha\rvert,$$

so by the ratio test the series is (absolutely) convergent for $0 < \lvert\alpha\rvert < 1$.

Summing up, the series converges for $\alpha > 1$ and for $0 < \lvert\alpha\rvert < 1$, in the "right" interpretation also for $\alpha = 0$. The series diverges for $\alpha = 1$ and for $\alpha < -1$, it is not well-defined for $\alpha = -1$ (and for $\alpha = 0$ in the literal interpretation).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.