2
$\begingroup$

I'm consider $e^{-x}$ with $x >> 0$. I want to express it in a series of the form

$$ e^{-x} = \sum_{i=1}^{\infty} \frac{a_i}{x^k}$$

But analysis seems to resist it:

Motivation:

$ \frac{1}{x^k}$ rapidly decays to 0, as does $e^{-x}$ so I'm thinking a sum of such terms should intuitively yield a series.

Work So Far:

My first idea was that substitution $ u = \frac{1}{x}$ yield that

$$ e^{-x} = e^{-\frac{1}{u}}$$

Now I could look to express this as a Taylor series around $u = \gamma$, observing that

$$ \frac{d}{du} e^{ - \frac{1}{u}} = \frac{1}{u^2} e^{-\frac{1}{u}} $$

Deriving again we have

$$ \frac{d^2}{du^2} e^{ - \frac{1}{u}} = \left(- \frac{2}{u^3} + \frac{1}{u^4} \right)e^{-\frac{1}{u}} $$

Again:

$$ \frac{d^3}{du^3} e^{ - \frac{1}{u}} = \left( \frac{2\times 3}{u^4} - \frac{4}{u^5} + \frac{1}{u^6}\right)e^{-\frac{1}{u}} $$

And In General

$$ \frac{d^k}{du^k} e^{ - \frac{1}{u}}= $$

$$ \left((-1)^{k-1} \frac{2 \times 3 \times ... k}{u^{k+1}} + (-1)^{k-2} \frac{4 \times 5 \times 6 ... \times (k+1)}{u^{k+2}} + (-1)^{k-3} \frac{6 \times 7 \times ... (k+2)}{u^{k+3}}... +\frac{1}{u^{2k}} \right) e^{-\frac{1}{u}} $$

$$ = e^{- \frac{1}{u}} \sum_{i = 0}^{k-1} \frac{(2k-1-i)!}{(2k -1 -2i)!} \frac{(-1)^{-i}}{u^{2k-i}} = \frac{e^{-\frac{1}{u}}}{u^{2k}}\sum_{i = 0}^{k-1} \frac{(2k-1-i)!}{(2k -1 -2i)!} (-1)^{i} u^{i} $$

Then it follows that we can define $e^{-\frac{1}{u}}$ around a point $u = \gamma$ as

$$ e^{- \frac{1}{u}} = e^{-\frac{1}{\gamma}} + \sum_{k=1}^{\infty} \left(\frac{e^{-\frac{1}{\gamma}}}{k!\gamma^{2k}}\sum_{i = 0}^{k-1} \frac{(2k-1-i)!}{(2k -1 -2i)!} (-1)^{i} \gamma^{i} \right) ( u - \gamma)^{k} $$

And therefore:

$$ e^{-x} = e^{-\frac{1}{\gamma}} + \sum_{k=1}^{\infty} \left(\frac{e^{-\frac{1}{\gamma}}}{k!\gamma^{2k}}\sum_{i = 0}^{k-1} \frac{(2k-1-i)!}{(2k -1 -2i)!} (-1)^{i} \gamma^{i} \right) \left( \frac{1}{x} - \gamma \right)^{k} $$

But this explodes if $\gamma = 0$, so the mechanics of building THAT converging series are going to be very different than this (namely NO convergence around $x < 1$ and then something nice...)

Error:

There is an error in my derivation, namely:

$$ \frac{d^3}{du^3} e^{ - \frac{1}{u}} = \left( \frac{2\times 3}{u^4} - \frac{6}{u^5} + \frac{1}{u^6}\right)e^{-\frac{1}{u}} $$.

The remaining series derivation is also incorrect for the same mistake not caught.

$\endgroup$
3
  • 1
    $\begingroup$ I am not sure, but I think the laurent-series will not converge fast enough towards zero to represent the function $e^{-x}$ $\endgroup$
    – Peter
    Mar 31 '16 at 9:11
  • 2
    $\begingroup$ Such a series development cannot exist because the (complex) exponential function has an essential singularity at infinity. $\endgroup$
    – Martin R
    Mar 31 '16 at 9:20
  • $\begingroup$ If you want a practical solution, apart from using the FPU-provided function, find $n$ with $2^n>|x|$ and compute as $(e^{-2^{-n}x})^{2^n}$ where the outer power is the result of $n$ squarings. $\endgroup$ Mar 31 '16 at 11:55
7
$\begingroup$

$$\frac{\partial}{\partial x} \exp(-x)=-\exp(-x)\\ \frac{\partial}{\partial x}\sum_{i=1}^{\infty} a_i x^{-i}=\sum_{i=1}^{\infty} \frac{a_i}{-i} x^{-i-1}$$ We would like $$\sum_{i=1}^{\infty} a_i x^{-i}=\sum_{i=1}^{\infty} \frac{a_i}{i} x^{-i-1}$$ If the coefficients in $x^{-1}$ are equal, then $a_1$ must be 0, because the rhs. has no term in $x^{-1}$. By induction we see that all $a_i$ must be 0. Thus, there is no such series that approximates $\exp(-x)$.

You may be interested in Pade approximations.

$\endgroup$
1
$\begingroup$

You wish to have an asymptotic expansion of $e^{-x}$ as $x \to \infty$. Allowing $x$ to be complex, and writing $z = 1/x$, we have $e^{-1/z}$ for which we want a Taylor series expansion around $z = 0$. Like Wouter has pointed, this is impossible because $e^{-1/z}$ has an essential singularity at $z = 0$. You may write a Laurent series, but I expect this to be identical to the Taylor series of $e^{-x}$.

This also shows that $e^{-x}$ is a term that is "beyond all orders" when it comes to asymptotics. It decays faster than every possible power law.

$\endgroup$

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.