2
$\begingroup$

I am realy stuck in solving the following limit problem. Can you find any function $g(x)$ by which $$\lim_{a\rightarrow \infty} \frac{a^N}{\log a} \int_{0}^\infty \frac{e^{-x}}{(1+ag(x))^N}dx = c$$ where $c$ is a nonzero constant.

The solution to this problem may contain some general properties on $g(x)$. But I can't even find a specific $g(x)$ for a specific $N$, say $N=2$.

$\endgroup$
  • $\begingroup$ Also posted here. $\endgroup$ – Lucian Apr 28 '14 at 9:58
  • 2
    $\begingroup$ I guess you need/want $g(x)$ to be independent of $a$, right? (otherwise trivially pick $g(x)=cst=1/(\log a)^N$). If so, this is impossible unless $g$ oscillates or vanishes (for finite or infinite points): for example if $0<c_1\leq g(x)\leq c_2$ for two fixed constants then the integral is of order $\mathcal{O}(1/(a^N))$, so the log kills everything. $\endgroup$ – leo monsaingeon Apr 28 '14 at 10:24
2
$\begingroup$

I propose you take $$g(x):=\frac{N}{c}\exp\frac{-x}{N}.$$ I too doubt that you can find an answer independent on $N$.

The method I used to determine $g$ might lead to an answer which is more intreseting, so I provide it here. Define $$\phi(a):=\int_0^\infty\frac{\exp(-x)}{(1+ag(x))^N} dx$$ so that $$\phi'(a)=\int_0^\infty u'(x)v(x) dx $$ with $$u(x)=\frac{1}{a(1+ag(x))^N} \\ v(x)=\frac{g(x)\exp(-x)}{g'(x)}.$$Integrating by parts gives $$a\phi'(a)=\alpha\phi(a)-\frac{g(0)}{g'(0)(1+ag(0))^N} $$ if we have the relation $$v'(x)=-\alpha\exp(-x) , ~~~~\alpha\in\mathbb C$$ which holds whenever $$ g(g'+g'')=(\alpha+1)(g')^2 .$$ Assume for now that this equation is satisfied. Then $$ \phi(a)=-a^\alpha \int_0^a \frac{g(0)}{g'(0)(1+tg(0))^Nt^{1+\alpha}} dt $$ which is of the required growth-type only if $\alpha=-N$, in which case $$\phi(a)\sim_\infty -\frac{\ln a}{a^Ng'(0)},$$ so that $$c=-\frac{1}{g'(0)}.$$

Now we solve for $g$, which satisfies the equation $$1+\frac{g''}{g'}=(1+\alpha)\frac{g'}{g}$$. This equation is integrable and there exists two constants $\delta, \gamma$ such that $$g(x)=(\gamma+\delta\exp(-x))^{-1/\alpha}.$$ The constant $\gamma$ is useless, while $\alpha$ and $\delta$ are completely determined.

$\endgroup$
1
$\begingroup$

As has been pointed out, if $g(x)>\epsilon$ for some $\epsilon>0$, the limit will be zero. On the other hand, if $g(x)=0$ on a set of positive measure, the limit is $\infty$, so it all depends on how fast $g$ tends to zero. The function $g(x)=x^{1/N}$ actually is a solution with limit $N$, as one sees by substituting $y=a^nx$, using de l'Hospital and substituting back. However, this solution depends on $N$. But the initial thought implies that, if there is a solution that depends on $N$, there probably is no solution independent of $N$.

$\endgroup$

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.