1
$\begingroup$

I have a function $f(x)$ which is defined as the solution to a certain differential equation. The boundary conditions are that in the $x\rightarrow \infty$ limit, it should be asymptotically equivalent to $\exp(-x)$, i.e. $$\lim_{x\rightarrow \infty} \frac{f(x)}{\exp(-x)}=c_{\infty}$$ where $c_{\infty}$ is some finite constant (not $\infty$, not 0).

In the $x\rightarrow -\infty$ limit, it should be asymptotically equivalent to a polynomial, say $x^n$: $$\lim_{x\rightarrow -\infty} \frac{f(x)}{x^n}=c_{-\infty}$$ where $c_{-\infty}$ is some finite constant (not $\infty$, not 0).

I hoped to approximate this function with elementary functions, or at least with fairly well-known special functions. So, I thought, which functions behave exponential-like in one limit but polynomial-like in the other? The sigmoid function $\frac{1}{1+\exp(-x)}$ is an example, it approaches $\exp(x)$ in one limit and $1$ in the other limit. By repeated integration of the sigmoid function, we can find a function that behaves like $\exp(x)$ in one limit, and like $x^n$ in the other.

This works fine, but, alas, repeatedly integrating $\frac{1}{1+\exp(-x)}$ gives rise to Polylogarithms, which are still somewhat more complicated than I would have hoped.

So: are there functions that can be expressed in terms of elementary functions, that behave like $\exp(-x)$ for very large positive $x$ and like $x^n$ for very large negative $x$?

Edit: Repeated integration of any sigmoid-like function (like tanh, arctan or the error function) would work, though I would be surprised if elementary sigmoid-like functions existed whose integrals are simpler than those of $\frac{1}{1+\exp(-x)}$. Multiplying a sigmoid with a polynomial does not work - doing that makes at least one of the limits diverge.

$\endgroup$
2
$\begingroup$

If you want $c_{\infty}, c_{-\infty}$ both in $(0,\infty)$, I suggest $$ f(x) = \frac{x^n}{a+x^ne^x} $$ with $a$ chosen large enough to avoid dividing by zero in case of odd $n$.

Then for $x\to\infty$ we have $$ \frac{f(x)}{e^{-x}} = \frac{x^n}{ae^{-x}+x^n} \to 1 $$ and for $x\to\infty$ $$ \frac{f(x)}{x^n} = \frac{1}{a+x^ne^x} \to \frac1a $$ since $x^ne^x\to 0$ for $x\to-\infty$.

$\endgroup$
  • $\begingroup$ Thanks, but perhaps I failed to be clear enough. The functions should "asymptotically behave like" an exponential in one limit and a polynomial in the other. They should be asymptotically equivalent. $c_{\infty}=$ and $c_{-\infty}=$ should be neither infinity nor zero. $\endgroup$ – Wouter Apr 7 '16 at 15:39
  • $\begingroup$ @Wouter: Edited with a new suggestion, then. $\endgroup$ – Henning Makholm Apr 7 '16 at 15:42

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.