I'm currently trying to prove the power rule using the epsilon-delta definition of a derivative. I've already done it for the basic limit definition, but I thought it might be a helpful exercise to test my understanding by doing it this way. However, I'm struggling and would really appreciate any help/hints.

My work so far:

The epsilon-delta definition says a function is differentiable if, for every $\epsilon > 0$, there exists a $\delta > 0$ such that $|x-x_0| < \delta$ implies $\frac{f(x)-f(x_0)}{x-x_0} - f'(x_0) < \epsilon$. Thus, I need to find a delta, as function of epsilon, and possibly $x_0$, such that whenever the first inequality is true the second one is as well.

So, fix $x_o \in R$, and let $|x-x_0| < \delta$, then $|\frac{x^n - x_{0}^{n}}{x-x_0} - nx_{0}^{n-1}| = |\frac{x^n - nxx_{0}^{n-1} + (n-1)x_{0}^{n}}{x-x_0}|$. Upper-bounding this with $x < \delta + x_0$ and simplifying, we get $|\frac{(\delta + x_0)^n - n\delta x_{0}^{n-1}-x_{0}^{n}}{\delta}| = |\frac{(\delta + x_0)^n}{\delta} -nx_{0}^{n-1} - x_{0}^{n}| < \epsilon$.

Is that right so far, and if so, any advice on how to push through this last step and solve for $\delta$? I'm stumped, to be honest.

-Thank you in advance for your help!

  • $\begingroup$ How denominator changed with $\delta$ to have upper bound? $\endgroup$ – Atbey Nov 22 '17 at 6:31
  • $\begingroup$ Do a binomial expansion on $(\delta - x_0)^n$ the first two terms cancel with what is there, leaving $\delta |{n\choose 2} x_0^{n-2} +{n\choose 3} x_0^{n-3}\delta + \cdots + \delta^{n-2}| < \epsilon$ and show that everything inside the absolute value is bounded. $\endgroup$ – Doug M Nov 22 '17 at 6:58
  • $\begingroup$ Your last line is incorrect. The middle term should be $|\frac {(\delta +x_0)^n -x_0^n}{\delta}-nx_0^{n-1}|$ $\endgroup$ – DanielWainfleet Nov 22 '17 at 10:53


$$(x^n - x_0^n) = (x - x_0)\sum_{k=0}^{n-1} x^k x_0^{n-1-k}$$

One can write

$$ \frac{x^n - x_0^n}{x - x_0} - n x_0^{n-1} = \sum_{k=0}^{n-1} (x^k x_0^{n-1-k} - x_0^{n-1}) $$ Define $M = |x_0| + 1$ and suppose that $|x-x_0|\le 1$, then $|x|\le M$ and $|x_0|\le M$ and $$ |x^k-x_0^k| \le |x - x_0|\sum_{p=0}^{k-1} |x|^p |x_0|^{k-1-p}\le |x - x_0|\sum_{p=0}^{k-1} M^{k-1}\le |x-x_0|k M^{k-1} $$

We get $$ \left|\frac{x^n - x_0^n}{x - x_0} - n x_0^{n-1}\right| \le \sum_{k=1}^{n-1} |x-x_0|k M^{k-1} |x_0|^{n-1-k}\le |x - x_0|M^{n-2}\sum_{k=1}^{n-1}k \le |x - x_0|M^{n-2}\frac{n (n-1)}{2} $$

For $n>1$ we can take $$\delta = \min\left(1, \frac{2\varepsilon}{n(n-1)(|x_0|+1)^{n-2}}\right)$$ For $n=1$ there is no condition on $\delta$ because $f(x) = f(x_0) + (x-x_0)$.

  • $\begingroup$ Using this hint at the beginning, I get $|\frac{x^n - x_{0}^{n}}{x-x_0} - nx_{0}^{n-1}| = |(1-n)x_{0}^{n-1} + xx_{0}^{n-2} + ... + x_0x^{n-2} + x^{n-1}| < \epsilon$. Is this what you had in mind? I don't understand how this helps, since there doesn't seem to be an easy way to cancel all of the x terms floating around. $\endgroup$ – P. Reinecke Nov 22 '17 at 6:57
  • $\begingroup$ See my edit above. $\endgroup$ – Gribouillis Nov 22 '17 at 7:15

For $x\ne 0$ we have $$(x+d)^n-x^n=-x^n+\sum_{j=0}^n x^{n-j}d^j\binom {n}{j}=nx^{n-1}d +d^2 P(d)$$ where $P(d)$ is a polynomial. So there exists $M\in \Bbb R^+$ such that $\forall d\; (|d|\leq 1\implies |P(d)|<M).$ Therefore for $0<|d|\leq 1$ we have $$|\frac {(x+d)^n-x^n}{d}-nx^{n-1}| =$$ $$= |\frac {nx^{n-1}d+d^2P(d)}{d}-nx^{n=1}| =$$ $$= |dP(x)|\leq |d|M.$$ And $|d|M\to 0$ as $d\to 0.$

To prove that $M$ exists : If $P(d)=\sum_{j=0}^ma_jd^j$ then $|d|\leq 1\implies |P(d)|\leq$ $\sum_{j=0}^m|a_j|\cdot |d|^j\leq $ $\leq \sum_{j=0}^m|a_j|=M.$

For $x=0\ne d$ we have $\frac {(x+d)^n-x^n}{d}=\frac {(0+d)^n-0^n}{d}=d^{n-1}.$

  • $\begingroup$ Note that the degree of $P(d)$ and the values of the co-efficients of $P(d)$ are determined by $x$ and $ n$ but are independent of $d.$ $\endgroup$ – DanielWainfleet Nov 22 '17 at 11:40

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.