I was wondering how the convergence of the series $$\sum_{k\geq1}\frac{(k+1)^y-k^y}{k^x},$$ depends on $x$ and $y$, where $x$ and $y$ are real numbers. I've been able to show that it converges if and only if $y\leq x$, by distinguishing nine cases; whether $x$ and $x-y$ are positive, negative or zero. If desired I can include a sketch of my proof. My question is whether there is a proof that works for all real $x$ and $y$? Or at least a proof that seems less artificial?
- 95,224
- 7
- 108
- 199
- 63,261
- 7
- 75
- 163
-
3Your proof must be invalid: if $x=y=1$, the series diverges. – TonyK Nov 13 '16 at 15:14
-
1$(k+1)^y -k^y = y\int_k^{k+1} t^{y-1} dt = y k^{y-1} +y\int_k^{k+1} (t^{y-1}-k^{y-1}) dt $ $ = y k^{y-1} +y(y-1)\int_k^{k+1} \int_k^t u^{y-2} du dt = y k^{y-1}+\mathcal{O}(y (y-1)k^{y-2})$ – reuns Nov 13 '16 at 18:55
-
Anyone care to explain the downvote? – Servaes Mar 10 '24 at 18:46
2 Answers
Use the Limit Comparison test with $$\sum_{k=1}^\infty \frac{1}{k^{x-y+1}}$$
$$\lim_k \frac{\frac{(k+1)^y-k^y}{k^x}}{\frac{1}{k^{x-y+1}}}=\lim_k \frac{(k+1)^y-k^y}{k^x}\frac{k^{x-y+1}}{1}=\lim_k \frac{(k+1)^y-k^y}{k^{y-1}}\\ =\lim_k \frac{(1+\frac{1}{k})^y-1}{\frac{1}{k}}$$
Now, $$\lim_{x\to 0} \frac{(1+x)^y-1}{x}= \frac{d}{dx}(1+x)^y|_{x=0}=y$$
Therefore $$\lim_k \frac{(1+\frac{1}{k})^y-1}{\frac{1}{k}}=y$$
Conclusion
If $y \neq 0$ , by the limit comparison test, the series is convergent if $x-y-1>1$ and divergent if $x-y-1 \leq 1$.
If $y=0$ your series is $\sum_{k\geq 0} 0$ is always convergent.
- 132,525
-
You don't really need limit comparison: and observe case $n=1$ as most significant to convergence.$$\begin{align}\sum_{k\ge1}\frac{(k+1)^y-k^y}{k^x}&=\sum_{k\ge1}\sum_{n\ge1}\frac{y(y-1)(y-2)\dots(y-n)}{n!}\frac{k^{y-n}}{k^x}\&=\sum_{n\ge1}\frac{y(y-1)(y-2)\dots(y-n)}{n!}\sum_{k\ge1}k^{y-n-x}\end{align}$$ – Simply Beautiful Art Nov 13 '16 at 16:30
-
@Servaes The sum over $k$ is the zeta function, which approaches $1$. So, asymptotically, the sum over $n$ converges if the sum over $k$ converges. – Simply Beautiful Art Nov 14 '16 at 14:18
If $y\ne0$, $$(k+1)^y-k^y=k^y\left[(1+\frac1k)^y-1\right]\sim yk^{y-1}$$ so the term of the sum is equivalent to $yk^{y-x-1}$, just apply Riemann criterion (and separate special cases if any) : the series converges if and only if $y-x-1<-1$, so $x>y$.
When $y=0$, $\frac{(k+1)^y-k^y}{k^x}=0$, so the series is convergent whater $x$ is.
Is that OK for you, @TonyK ?
- 4,846
-
-
@TonyK : which is why I added "and separate special cases" :-) – Nicolas FRANCOIS Nov 13 '16 at 16:02
-
@Servaes : asymptotic equivalent : $u\sim v\iff \frac{u}{v}\longrightarrow1$. – Nicolas FRANCOIS Nov 13 '16 at 16:03
-
No, your final statement is unambiguously wrong. When you mention special cases like that, it just implies that they must be proven differently, not that your final statement doesn't apply to them. – TonyK Nov 13 '16 at 17:14
-
-
3