2
$\begingroup$

Let $f:(0, \infty) \longrightarrow (0, \infty)$ be a monotonously increasing function (in fact, a step function) and let $P$ be a polynomial of degree $N$. Suppose I know that for some $k$, the limit $$A:= \lim_{\varepsilon\rightarrow 0} \int_0^\infty \frac{e^{-\varepsilon s}}{s^k}(f(s) - P(s))\, \mathrm{d}s$$ exists. Can I conplude that $f$ has an asymptotic expansion which is given by $P$? More precisely, can I conclude that $$\left| f(s) -P(s)\right| \leq C s^{k-1}$$ for some constant $C>0$?

$\endgroup$
2
$\begingroup$

No. The limit can exist due to cancellations, without $|f-P|$ being as small pointwise as you hoped for.

Let's take $k=0$, $P(s)=s$, and $f(s)=n+1/2$ on $n<s<n+1$. Then $$ \int_n^{n+1}e^{-\epsilon s} (f(s)-P(s))\, ds =-e^{-\epsilon(n+1/2)}\int_{-1/2}^{1/2} te^{-\epsilon t}\, dt = \frac{\epsilon}{12}e^{-\epsilon(n+1/2)}+O(\epsilon^2)e^{-\epsilon n} , $$ so the sum over $n$ of these terms equals $$ \frac{\epsilon}{12(1-e^{-\epsilon})} + O(\epsilon) . $$ Thus the limit exists ($A=1/12$), but $|f(s)-P(s)|\ge c>0$ for arbitrarily large $s$.

| cite | improve this answer | |
$\endgroup$

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.