1
$\begingroup$

I am currently trying to proof the following exercise:

Let $n\in \mathbb N$ and $g:\mathbb R \to \mathbb R$ a $n$-times continuously differentiable function with $g^{(j)}(0) = 0$ for all $j\in \{0, 1, ... ,n-1\}$ where $g^{(j)}$ denotes the $j$-th derivative. Proof that $g(x) = O(\lvert x\rvert^n)$ for $x\to 0$.

I tried to proof this via induction over $n$:

$n=1$: I have to show that there are $C>0, \delta > 0$ with $\lvert x\rvert < \delta \Rightarrow \lvert g(x) \lvert \leq C\lvert x \rvert$. Since $g$ is continously differentiable on $\mathbb R$ so it is on (without loss of generality) $[0;a]$. Then $g'$ is continuous on $[0;a]$ and in particular $C:=\sup g' \in [0;a]$ with $C<\infty$.
By the mean value theorem there is $\xi \in ]0,a[$ with $$g'(\xi) =\frac{g(0)-g(a)}{0-a} = \frac{g(a)}{a}$$ which implies, choosing $\delta := a$ $$|g(x)| \leq C|x|$$for $|x| < \delta$. Now I don't know if this is correct so far, but assuming it is, I don't know how to proceed with $n \to n+1$. Any hints appreciated!

$\endgroup$
3
  • $\begingroup$ You want $\sup |g'|$ $\endgroup$ – zhw. Jan 22 '17 at 18:00
  • $\begingroup$ Note that you don't need the MVT for the $n=1$ case; nor do you need $g\in C^1.$ Suppose we only know $g(0)=0$ and $g'(0)$ exists. Then $$\lim_{x\to 0}\frac{g(x)}{x}=\lim_{x\to 0} \frac{g(x)-g(0)}{x-0} = g'(0).$$ This implies $g(x)/x$ is bounded near $0$ for nonzero $x,$ giving $g(x)= O(x).$ $\endgroup$ – zhw. Jan 22 '17 at 18:10
  • $\begingroup$ That is true. Didn't think about that. Thanks! $\endgroup$ – Staki42 Jan 22 '17 at 18:21
2
$\begingroup$

Here's a proof that uses only the MVT. Fix $x >0$. Then there exists $\xi_1 \in (0,x)$ such that $$ g(x) = g(x) - g(0) = g'(\xi_1)(x-0) = g'(\xi_1) x. $$ Now we apply the MVT again to get $\xi_2 \in (0,\xi_1)$ such that $$ g'(\xi_1) = g'(\xi_1)-g'(0) = g''(\xi_2)(\xi_1-0) = g''(\xi_2)\xi_1. $$ Iterating, we find $0 <\xi_n<\xi_{n-1} < \cdots < xi_1 < x$ such that $$ g^{(k)}(\xi_k) = g^{(k+1)}(\xi_{k+1}) \xi_k. $$

Consequently $$ g(x) = g'(\xi_1)x = g''(\xi_2) \xi_1 x = \cdots = g^{(n)}(\xi_n) x \prod_{i=1}^{n-1} \xi_i $$ and so $$ |g(x)| \le |g^{(n)}(\xi_n)| x^n. $$ The continuity of $g^{(n)}$ at $0$ then shows that $$ |g(x)| = O(x^n). $$ A similar argument works for $x < 0$.

$\endgroup$
1
  • $\begingroup$ Note that the MVT is Taylor's theorem with the Lagrange form of the remainder: $f(x) = f(a) + f'(\xi)(x-a).$ $\endgroup$ – zhw. Jan 22 '17 at 17:58
1
$\begingroup$

Using Taylor's Theorem with Lagrange form of the remainder,

$$g(x)=\sum_{k=0}^{n-1}g^{(k)}(0)\frac{x^k}{k!} + g^{(n)}(\xi_x)\frac{x^n}{n!}$$ where $|\xi_x|<|x|$.

With the given assumptions, $\displaystyle g(x)=g^{(n)}(\xi_x)\frac{x^n}{n!}$, hence $$\left|\frac{g(x)}{x^n}\right|=\left|\frac{g^{(n)}(\xi_x)}{n!}\right| $$

Since $g^{(n)}$ is continuous, it is bounded on a neighborhood of $0$ by some $M$.

Hence $$\left|\frac{g(x)}{x^n}\right|\leq \left|\frac{M}{n!}\right| $$ for sufficiently small $x$.

$\endgroup$
2
  • $\begingroup$ The problem here is that this exercise should be done without knowledge of Taylo's Theorem as it is rather in preparation for the proof of Taylor's Theorem. :/ $\endgroup$ – Staki42 Jan 22 '17 at 17:19
  • $\begingroup$ See my other answer. $\endgroup$ – Gabriel Romon Jan 22 '17 at 17:32
1
$\begingroup$

Without Taylor's Theorem, via induction.

OP wrote the base case.

Suppose the result holds for some $n\geq 1$. Let $g$ be a $n+1$-times continuously differentiable function with $g^{(j)}(0) = 0$ for all $j\in \{0, 1, ... ,n\}$.

Note that $h:=g'$ is a $n$-times continuously differentiable function with $h^{(j)}(0) = 0$ for all $j\in \{0, 1, ... ,n-1\}$. The induction hypothesis yields some $M\geq 0$ and a neighborhood of $0$, say $[-a,a]$ such that $\displaystyle |x|\leq a \implies |h(x)|\leq M |x^n|$.

Therefore, with $|x|\leq a$, $$\displaystyle |g(x)|=|g(x)-g(0)|=\left|\int_0^x g'(t) dt \right| = \left|\int_0^x h(t) dt \right| \leq M\frac{|x|^{n+1}}{(n+1)!}$$

and the result is proved.

$\endgroup$
1
  • $\begingroup$ Thanks, that helped a lot! $\endgroup$ – Staki42 Jan 22 '17 at 17:38
1
$\begingroup$

We can get the result assuming only that $0= g(0)=g'(0)= \cdots = g^{(n-1)}(0)$ and that $g^{(n)}(0)$ exists. Proof sketch: Apply the MVT $n-1$ times, as in the answer of @Glitch, to see

$$ \tag 1|g(x)|\le |g^{(n-1)}(c_x)||x^{n-1}|$$

for some $c_x\in (0,x).$ We can write the right side of $(1)$ as

$$|g^{(n-1)}(c_x)-g^{(n-1)}(0)||x^{n-1}| = \left |\frac{g^{(n-1)}(c_x)-g^{(n-1)}(0)}{c_x}\right ||c_xx^{n-1}|$$ $$ \le \left |\frac{g^{(n-1)}(c_x)-g^{(n-1)}(0)}{c_x}\right ||x^{n}|.$$

As $x \to 0,$ $c_x \to 0,$ so the fraction in the last line $\to g^{(n)}(0).$ This gives the desired $O(x^n)$ result.

$\endgroup$

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.