While finding the Taylor Series of a function, when are you allowed to substitute? And why?

For example:

Around $x=0$ for $e^{2x}$ I apparently am allowed to substitute $u=2x$ and then use the known series for $e^u$. But for $e^{x+1}$ I am not allowed to substitute $u=x+1$.

I know the technique for finding the Taylor Series of $e^{x+1}$ around $x=0$ by taking $e^{x+1}=e\times e^x$. However, I am looking for understanding and intuition for when and why it is allowed to apply substitution.

Note: there are several question that are similar to this one, but I have found none that actually answers the question "why"; or that shows a complete proof.

EDIT: Thanks to the answer of Markus Scheuer I should refine the question to cases where the series is finite, for example $n\to3$

  • $\begingroup$ The real questioni is: why on Earth shall you substitute $u = 2x$?? Leave $e^{2x}$ and use the Taylor series for that! $\endgroup$ – Mycroft Nov 12 '16 at 10:19
  • $\begingroup$ @AlanTuring The point is that I try to understand why I can do that for $e^{2x}$ but not for $e^{x+1}$ $\endgroup$ – GambitSquared Nov 12 '16 at 10:20

A function $f(x)$ analytic at $x=0$ can be represented as power series within an open disc with radius of convergence $R$. \begin{align*} f(x)=\sum_{n=0}^\infty a_nx^n\qquad\qquad \qquad |x|<R \end{align*}

Any substitution $x=g(u)$ is admissible as long as we respect the radius of convergence. \begin{align*} f(g(u))=\sum_{n=0}^\infty a_n \left(g(u)\right)^n\qquad\qquad\quad |g(u)|<R \end{align*}

We know $f(u)=e^u$ can be represented as Taylor series convergent for all $u\in\mathbb{R}$, i.e. the radius of convergence $R=\infty$. \begin{align*} f(u)=e^u=\sum_{n=0}^\infty \frac{u^n}{n!}\qquad\qquad\qquad u\in \mathbb{R} \end{align*}

Substitution $u=2x$

We consider

\begin{align*} f(2x)=e^{2x}=\sum_{n=0}^\infty \frac{(2x)^n}{n!}\qquad\qquad\qquad 2x\in \mathbb{R} \end{align*}

This substitution is admissible for all $x \in \mathbb{R}$ since $$2x\in\mathbb{R}\qquad\Longleftrightarrow\qquad x\in\mathbb{R}$$ So, the radius of convergence of the Taylor series of $f(2x)=e^{2x}$ is $\infty$.

We obtain \begin{align*} f(2x)=e^{2x}=\sum_{n=0}^\infty \frac{(2x)^n}{n!}\qquad\qquad\qquad x\in \mathbb{R} \end{align*}

Substitution $u=x+1$

We consider

\begin{align*} f(x+1)=e^{x+1}=\sum_{n=0}^\infty \frac{(x+1)^n}{n!}\qquad\qquad\qquad x+1\in \mathbb{R} \end{align*}

This substitution is admissible for all $x \in \mathbb{R}$ since $$x+1\in\mathbb{R}\qquad\Longleftrightarrow\qquad x\in\mathbb{R}$$ So, the radius of convergence of the Taylor series of $f(x+1)=e^{x+1}$ is $\infty$.

We obtain \begin{align*} f(x+1)=e^{x+1}=\sum_{n=0}^\infty \frac{(x+1)^n}{n!}\qquad\qquad\qquad x\in \mathbb{R} \end{align*}

We also obtain \begin{align*} e\cdot e^x&=\left(\sum_{k=0}^\infty \frac{1}{k!}\right)\left(\sum_{l=0}^\infty \frac{x^l}{l!}\right)\\ &=\sum_{n=0}^\infty \left(\sum_{{k+l=n}\atop{k,l\geq 0}}\frac{1}{k!}\cdot\frac{x^l}{l!}\right)\\ &=\sum_{n=0}^\infty \left(\sum_{l=0}^n\frac{1}{(n-l)!}\cdot\frac{x^l}{l!}\right)\\ &=\sum_{n=0}^\infty\left(\sum_{l=0}^n\binom{n}{l}x^l\right)\frac{1}{n!}\\ &=\sum_{n=0}^\infty\frac{(x+1)^n}{n!}\\ &=e^{x+1} \end{align*}

Conclusion: We can use any substitution for convenience as long as we respect the radius of convergence.

  • $\begingroup$ I understand now that one is allowed to substitute in this case if $n\to\infty$. Is it correct to say that the series of $e^{x+1}$ that is the result of substitution is centered around $x=-1$, while the series that is calulated directly is centered around $x=0$? And doesn't this make a difference if I want to give an approximation of $e^{x+1}$ around $x=0$? I mean if I want to approximate I usually don't let $n\to\infty$, but rather to a fixed number, say $n\to3$. Would that make the case different? $\endgroup$ – GambitSquared Nov 14 '16 at 15:07
  • $\begingroup$ @ImreVégh: Yes, it is correct that the representation of $e^{x+1}$ above is centered around $x=-1$. Note, that if we have a power series centered around some value $x=x_0$, we can select any value $x_1$ within the disc of convergence $x_1\in d(x_0,R)$ of $f(x)=\sum_{n=0}^\infty a_n(x-x_0)^n$ as new center by expanding $\sum_{n=0}^\infty a_n ((x-x_1)-(x_0-x_1))^n$ at powers of $x-x_1$. The new radius of convergence $R_1$ is at least as big, that $d(x_1,R_1)\subset d(x_0,R)$. $\endgroup$ – Markus Scheuer Nov 14 '16 at 15:23
  • $\begingroup$ @ImreVégh: If we want to expand the special case $e^{x+1}$ around $x=0$ we can go the easy way (as you already did) and write $e^{x+1}=e\cdot e^x=\sum_{n=0}^\infty \frac{e}{n!} x^n$ with coefficients $\frac{e}{n!}$ of $x^n$. The somewhat more cumbersome way would be to expand $(x+1)^n$ in $\sum_{n=0}^\infty \frac{(x+1)^n}{n!}$ and arrange the terms according to increasing powers of $x$. $\endgroup$ – Markus Scheuer Nov 14 '16 at 15:27

If $ f(x)=P_n(x)+x^n\epsilon(x)$ then


with $\lim_{x\to 0}\epsilon(x)=0$.

thus we need that

$\lim_{x\to 0}\epsilon(u(x))=0$

so, we must have

$$\lim_{x\to 0}u(x)=0$$

  • $\begingroup$ Hmm, I understand. And why is substitution allowed in the first place? When $\lim_{x\to 0}\epsilon(u(x))=0$? $\endgroup$ – GambitSquared Nov 12 '16 at 9:31

The quantity $2x$ is a product and as $x\to 0$ it remains a small number.

The quantity $x+n$ for $n\neq 0$ is not a little quantity anymore, and so you are not anymore around zero but you're around $n$.


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.