If $f(x)=e^x$ and $f_n(x)$ is the $n^{th}$ order Taylor Polynomial of this function, then we know that the Taylor Theorem states there exists $t\in [x_0, x]$ such that the remainder term of is given by,


as $f^{(n+1)}(t)=f(t)=e^t$. Now we also know,


$$\Rightarrow e^x = f_n(x) + \frac{e^t}{(n+1)!}(x-x_0)^{n+1}$$

On the right hand side, $f_n$ is a polynomial and so is the expression for the remainder term. However $e^x$ is not a polynomial. How is it possible to write a non polynomial function in terms of a polynomial? We can use a similar procedure and arrive at this contradiction for other non polynomial functions as well. So what is the gap in my understanding?

  • 1
    $\begingroup$ The second term on the RHS (the remainder term) is not a polynomial, because $t$ is a function of $x$, not a constant. $\endgroup$
    – user169852
    Mar 15, 2021 at 5:23
  • $\begingroup$ We aren't saying that the functions are equal, just that they take the same value at the particular number $x$. $\endgroup$
    – saulspatz
    Mar 15, 2021 at 5:30

1 Answer 1


$t$ is just a value between $x_0$ and $x$ and $e^t$ is how to evaluate this constant. It does depend on $x$, but nobody said that Taylor polynomial is expressing the function in full. It has an error in its approximation.

The term with $t$ is that error term, so it talks about an error in Taylor polynomial approximation. It is not part of polynomial and of course that it depends on $x$, because typically more you are away from $x_0$ worse the error.

Just because it has $(x-x_0)^{n+1}$ in it does not make it a part of the polynomial, there are other expressions for the error term that look less like an additional polynomial term.

So, yes, besides $(x-x_0)^{n+1}$ the error term may have one additional dependency on $x$, or it may not, we do not know. All we know is that $t$ is between the two. In general it is movable regardless if this move is expressible as a function on $x$ or not.

Typically this $t$ is used to say that max error cannot be worse than, in your case, the maximum value of $e^t$ where $t$ is between $x$ and $x_0$ together with other values in that term. Some functions will have this max value fixed for quite some possible range between $x$ and $x_0$, some will not. But anyway $t$ is not really useful beside deciding this max error, because we know nothing else about $t$ except that it is between $x$ and $x_0$.

If $t$ would be a fixed value then the error term would be in form of $k(x-x_0)^{n+1}$, meaning every function would be expressible as a polynomial of $n+1$ degree, and unless the function is such a polynomial that is not the case.

  • 1
    $\begingroup$ -1 vote and accepted answer. Funny. $\endgroup$
    – Alex Peter
    Mar 16, 2021 at 19:22

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .