3

Suppose that $f$ defined on $(a,\infty)$ is bounded on each finite interval $(a,b),a>b$. For a non-negative integer $k$, it is to be shown that $\lim_{x\to \infty} \frac{f(x+1)-f(x)}{x^k}=l\implies \lim_{x\to \infty}\frac{f(x)}{x^{k+1}}=\frac l{k+1}$

Given an $\displaystyle \epsilon >0,$there exists an integer $\displaystyle N$ such that

$$\displaystyle x \geq N \Longrightarrow \ \left| \frac{f( x+1) -f( x)}{x^{k}} -l \right| < \epsilon $$

Noting that \begin{align*} f( x) & =\sum _{i=1}^{[ x]} f( x-[ x] +i) -f( x-[ x] +i-1) +\color{red}{f( x-[ x])} \end{align*} it follows that \begin{align*} f( x) & =\sum _{i=1}^{N}\overbrace{( f( x-[ x] +i) -f( x-[ x] +i-1))}^{h( i)} +\sum _{i=N+1}^{[ x]}( f( x-[ x] +i) -f( x-[ x] +i-1)) +\color{red}{( \ )} \end{align*} It follows that

\begin{align*} |\frac{f( x)}{x^{k+1}} -\frac{l}{k+1} | & \leq \sum _{i=1}^{N} |\frac{h( i)}{x^{k+1}} -\frac{l}{[ x]( k+1)} |+\sum _{i=N+1}^{[ x]} |\frac{h( i)}{x^{k+1}} -\frac{l}{[x](k+1)} |+\color{red}{( )x^{-k-1}}\\ & \leq {\textstyle \frac{\overbrace{M}^{\sup _{1\leq i\leq N} h( i)}}{x^{k+1}} +\frac{N|l|}{[ x]( k+1)} +\frac{\epsilon ([ x] -N)}{x} +|\frac{l}{x} -\frac{l}{[ x]( k+1)} |([ x] -N) +\color{red}{\overbrace{\color{red}{\sup _{t\in [ 0,1)}f(t)}}^{M'} x^{-k-1}}}\\ & =\frac{M+M'}{x^{k+1}} +{\textstyle \frac{N|l|}{[ x]( k+1)} +\color{purple}{|\frac{l[ x]}{x} -\frac{l}{( k+1)} |} -N|\frac{l}{x} -\frac{l}{[ x]( k+1)} |+\frac{\epsilon ([ x] -N)}{x}} \tag 1 \end{align*}

In $(1)$, except the purple term every other term can be made arbitrarily small. How do I take care of the purple term so that I can conlude the desired result by limit definition.

Koro
  • 11,766
  • For $k=0$ see https://math.stackexchange.com/q/192963/42969. – Martin R Feb 15 '22 at 16:23
  • @MartinR: yes, I can prove this for $k=0$. Thanks. – Koro Feb 15 '22 at 16:24
  • @Koro: do you wanna try another Idea? for example polynomials? – Khosrotash Feb 15 '22 at 16:52
  • @Khosrotash: yeah, I would like to know that as well. I also want to say why the method used in the post is the way it is: this is how I recall proving Cauchy's theorem (if $\lim x_n=l$ then $\frac{x_1+x_2+...+x_n}n=l$) and this worked for $k=0$ case. Another way I tried was to do some manipulations in the expressions like setting $g(x)=f(x)x^{-k}$ and then trying to use the fact that the given statement is true for $k=0$ but neither seemed to work. – Koro Feb 15 '22 at 16:56
  • The usual proof for Cesaro-Stolz works here to prove the more general theorem which I mentioned in comments to Martin's answer. I think that approach/proof is simpler. – Paramanand Singh Feb 16 '22 at 08:28

3 Answers3

3

First we prove the statement for the case $l=0$, i.e. $ \lim_{n \to \infty}\frac{f(x+1)-f(x)}{x^k} = 0 $.

Given $\epsilon > 0$ there is a $y > a$ such that for all $x \ge y$: $$ \left | \frac{f(x+1)-f(x)}{x^k} \right | < \epsilon \, . $$ Similarly as in If $\lim_{x\to+\infty}[f(x+1)-f(x)]= \ell,$ then $\lim\limits_{x\to+\infty}\frac{f(x)}x=\ell$. we write $$ \frac{f(x)}{x^{k+1}} = \frac{1}{x}\left(\sum_{i=1}^{\lfloor x-y\rfloor}\frac{f(x-i+1)-f(x-i)}{(x-i)^k} \cdot \left( 1-\frac ix\right)^k \right)+\frac{f(x-\lfloor x-y\rfloor)}{x^{k+1}} $$ which implies $$ \left| \frac{f(x)}{x^{k+1}}\right| \le \frac{\lfloor x-y\rfloor}{x} \epsilon + \frac{M}{x^{k+1}} $$ with $M = \sup \{ |f(z)| : y \le z < y+1\}$. It follows that $$ \limsup_{n \to \infty }\left| \frac{f(x)}{x^{k+1}}\right| \le \epsilon \, . $$ This holds for all $\epsilon > 0$, and therefore $$ \lim_{n \to \infty} \frac{f(x)}{x^{k+1}} = 0 \, . $$


For the general case with $\lim_{x\to \infty} \frac{f(x+1)-f(x)}{x^k}=l$ we set $$ g(x) = f(x) - \frac{x^{k+1}}{k+1} l \, . $$ $g$ is bounded on every finite interval $(a, b)$ and satisfies $$ \frac{g(x+1)-g(x)}{x^k} = \frac{f(x+1)-f(x)}{x^k} - l \frac{(x+1)^{k+1}-x^{k+1}}{(k+1)x^k} \\ = \frac{f(x+1)-f(x)}{x^k} - l \frac{(1+1/x)^{k+1}-1}{(k+1)/x} \to l - l = 0 $$ and then $$ \frac{f(x)}{x^{k+1}} = \frac{g(x)}{x^{k+1}} + \frac{l}{k+1} \to 0 + \frac{l}{k+1} = \frac{l}{k+1} \, . $$

Martin R
  • 128,226
  • Thanks a lot for the answer. It solves my question. As I stated earlier in the comments, I managed to prove the result for $k=0$. Then, I tried to come up with some function that reduces to $k=0$ case (like g in this answer) but couldn't come up with any. This is an exercise problem in a book and right before this exercise lies the exercise for $k=0$ case. I want to know how to get rid of purple term in OP. This serves the purpose that one doesn't need a prerequisite of knowing the result for $k=0$ case. So can you please suggest something to get rid of the purple term also. Thanks. :) – Koro Feb 16 '22 at 04:27
  • @Koro: Your term $|\frac{l[ x]}{x} -\frac{l}{( k+1)} | =l \cdot |\frac{[ x]}{x} -\frac{1}{( k+1)} |$ converges to $l \cdot |1- \frac{1}{k+1}| = l \cdot \frac{k}{k+1}$, it does not become small. – Martin R Feb 16 '22 at 04:36
  • Yeah, I noted that too as mentioned in the post. But there must be some way to get past this. The ideas in my (incomplete) proof in OP are inspired by the way we prove Cauchy's theorem on limits ( if $\lim x_n=L$ then $\frac {x_1+...+x_n}n\to L$). I think I am not breaking $\frac l{k+1}$ in the way that will serve the purpose. – Koro Feb 16 '22 at 04:39
  • 1
    @Koro: I tried to generalize the proof in https://math.stackexchange.com/a/192972/42969 to arbitrary $k$. Unless I am mistaken, that is also what you did. (Apparently you assume that $f$ is defined on $(0, \infty)$ and not on $(a, \infty)$, but that can perhaps be salvaged.) – Anyway, my observation was that the transfer from $k=0$ to arbitrary $k$ is much easier in the case $l=0$, therefore I reduced the problem to that case. – Martin R Feb 16 '22 at 05:04
  • 1
    +1 I think it should be possible to prove the analog of Cesaro Stolz in the same manner : If $g(x) $ is strictly increasing and tends to $\infty$ and $f$ is bounded on finite intervals then $(f(x+1)-f(x))/(g(x+1)-g(x))\to L\implies f(x) /g(x) \to L$. – Paramanand Singh Feb 16 '22 at 06:28
  • @MartinR Thanks a lot for the help. I'll think about it too meanwhile. :) And yes, I ended up assuming $f$ defined on (0,$\infty$) unknowingly. – Koro Feb 16 '22 at 06:28
  • In my previous comment I think we also need to ensure that $g$ is bounded on finite intervals as well. – Paramanand Singh Feb 16 '22 at 06:46
  • 2
    @Koro: I think that I have found the flaw in your proof (and I had made the same error). One can only use that $$ \left| \frac{f(x-i+1)-f(x-i)}{(x-i)^k} - l\right| < \epsilon $$ and not $$ \left| \frac{f(x-i+1)-f(x-i)}{x^k} - l\right| < \epsilon $$ In the case $l=0$ this can be fixed (see my edit), but in the general case your estimate is wrong. – Martin R Feb 16 '22 at 08:20
  • My question is answered now. Thank you so much :). +1 – Koro Feb 16 '22 at 08:56
3

The following is an alternative approach and does not really try to fix the flaw in yours. I am not sure if that can be fixed easily.


As mentioned in comments to Martin's answer we can prove more generally the analogue of Cesaro-Stolz :

Theorem: Let $f, g$ be functions defined on interval $(a, \infty) $ such that $f$ is bounded on any bounded subinterval of $(a, \infty) $ and let $g$ be strictly increasing with $g(x) \to\infty $ as $x\to\infty $. Then we have $$\lim_{x\to \infty} \frac{f(x+1)-f(x)}{g(x+1)-g(x)}\to L\implies \lim_{x\to\infty} \frac{f(x)} {g(x)} \to L.$$

Let $\epsilon >0$ and then we have a number $y>a$ such that $$L-\epsilon <\frac{f(x+1)-f(x)}{g(x+1)-g(x)}<L+\epsilon $$ and $g(x) >0$ for all $x\geq y$. Since $g$ is strictly increasing the above leads us to $$(L-\epsilon) (g(x+1)-g(x))<f(x+1)-f(x)<(L+\epsilon) (g(x+1)-g(x))\tag{1}$$ Let $m=\lfloor x-y\rfloor$ so that $m$ is a non-negative integer depending on $x$ (if one wants to be explicit one can write $m_x$ instead of $m$) and $$y\leq x-m< y+1$$ Then we have for each value of $i=0,1,2,\dots,m$ $$(L-\epsilon) (g(x+1-i)-g(x-i))<f(x+1-i)-f(x-i)<(L+\epsilon) (g(x+1-i)-g(x-i))$$ This is simply obtained by replacing $x$ with $x-i$ in $(1)$. Adding the above $m+1$ inequalities we get $$(L-\epsilon) (g(x+1)-g(x-m))<f(x+1)-f(x-m)<(L+\epsilon)(g(x+1)-g(x-m))$$ Dividing the above by $g(x+1)$ we get $$(L-\epsilon) \left(1-\frac{g(x-m)}{g(x+1)}\right)<\frac{f(x+1)}{g(x+1)}-\frac{f(x-m)}{g(x+1)}<(L+\epsilon) \left(1-\frac{g(x-m)}{g(x+1)}\right)$$ Taking limits as $x\to\infty $ and noting that $x-m\in[y, y+1]$ and $f, g$ being bounded in $[y, y+1]$ ($g$ is bounded because of monotone nature) we get $$L-\epsilon \leq \liminf_{x\to\infty} \frac{f(x+1)}{g(x+1)}\leq \limsup_{x\to\infty} \frac{f(x+1)}{g(x+1)}\leq L+\epsilon $$ Since $\epsilon >0$ is arbitrary it follows that $f(x) /g(x) \to L$ as $x\to\infty $.

The proof can be adapted to deal with a strictly decreasing $g$ which tends to $-\infty $. Also the argument $x+1$ can be replaced by $x+c$ for any non-zero $c$. And we have a similar result for $x\to-\infty $.

The current question is handled by setting $g(x) =x^{k+1}/(k+1)$.

  • 2
    I had just figured this out (motivated by your comments) :) – It suffices to require that $f$ is bounded on any subinterval, for $g$ this follows from the monotony. Also the generalized inequality chain $$ \liminf_{x \to \infty} \frac{f(x+1)-f(x)}{g(x+1)-f(x)} \le \liminf_{x \to \infty} \frac{f(x)}{g(x)} \le \limsup_{x \to \infty} \frac{f(x)}{g(x)}\le \limsup_{x \to \infty} \frac{f(x+1)-f(x)}{g(x+1)-f(x)} $$ holds as in the discrete Stolz–Cesàro theorem. – Martin R Feb 16 '22 at 08:58
  • @MartinR: thanks. I will remove the condition on $g$. – Paramanand Singh Feb 16 '22 at 09:09
  • This is also very nice. Thank you so much :). +1 – Koro Feb 16 '22 at 09:10
  • One confusion: Shouldn't it have been stated that 'for all $x\ge y$' instead of $x>y$ after Theorem? – Koro Feb 17 '22 at 04:15
  • 1
    Thanks @Koro for catching typo. Will fix. However remember that limit definition works with both $>$ and $\geq $. The argument here needs $\geq $. – Paramanand Singh Feb 17 '22 at 05:57
  • Yes, I understand that about the limit definition. Thanks. – Koro Feb 17 '22 at 06:50
0

take $f(x)=ax^{k+1}+bx^{k}+cx^{k-1}+...$ $$degree(f(x))=k+1$$and degree of $$f(x+1)-f(x)=a(x+1)^{k+1}+b(x+1)^k+c(x+1)^{k-1}+...\\-(ax^{k+1}+bx^{k}+cx^{k-1}+...)\\=(a-a)\not{x^{k+1}}+a\times{^{k+1}}C_1 x^{k}+...\\=a(k+1)x^k+...$$so $$\lim_{x\to \infty} \frac{f(x+1)-f(x)}{x^k}=l\\ \lim_{x\to \infty} \frac{a(k+1)x^k+...}{x^k}=l\\a(k+1)=l\\a=\frac{l}{k+1}$$so $$\lim_{x\to \infty} \frac{f(x)}{x^{k+1}}=\lim_{x\to \infty} \frac{ax^{k+1}+bx^{k}+cx^{k-1}+...}{x^{k+1}}=a=\frac{l}{k+1}\\$$

Khosrotash
  • 25,772