5
$\begingroup$

I am wondering if the following assertion is true:

Let $f,g:\mathbb{R}_+\rightarrow [0,1]$ be completely monotone functions on $\mathbb{R}_+^*$, that is, $(-1)^n f^{(n)}(x)\geq 0$ and $(-1)^n g^{(n)}(x)\geq 0$ for any $x>0$ and any $n\in \mathbb{N}$. Assume that $f(0)=g(0)$ and $\lim_{x\rightarrow \infty} f(x) = \lim_{x\rightarrow \infty} g(x)=0$, and that there exist $0<a<A$ such that $f\leq g$ on $[0,a]$ and $f\leq g$ on $[A,\infty)$.

Then $f\leq g$ on $\mathbb{R}_+$ (?).

If not, does anyone have a counter-example?

N.B.: 1) one can think of $f$ and $g$ as two Laplace transforms of positive measures.

2) Removing any of the assumptions seems to lead to a counter-example.

Thank you for your help!

$\endgroup$
4
  • $\begingroup$ Thanks Christian, do they cross? Btw I should add that $lim_{x\rightarrow \infty} f(x)=lim_{x\rightarrow \infty} g(x)=0$, $f$ and $g$ being truly Laplace transforms. $\endgroup$
    – Alphonse
    Sep 26, 2015 at 16:23
  • $\begingroup$ Your conditions are contradictive: there are no such functions, if $f(0)=0$ and $\lim_\infty f=0$, how can $f$ be monotone? $\endgroup$ Sep 26, 2015 at 17:11
  • $\begingroup$ Hi, i did not say that $f(0)=0$, just that $f(0)=g(0)$. $\endgroup$
    – Alphonse
    Sep 26, 2015 at 17:15
  • $\begingroup$ yes, unless I'm missing something, this solution is $t=1$, which corresponds to $e^{-x/2}=1$, hence $x=0$. But anyway, I originally forgot to mention that $f$ and $g$ must have the same final value (which I edited), for which I apologize. $\endgroup$
    – Alphonse
    Sep 26, 2015 at 18:02

1 Answer 1

6
$\begingroup$

No. Let's take $g(x)=\epsilon e^{-\epsilon x}+(1-\epsilon)e^{-\alpha x}$, $$ f(x) = \int_{\epsilon}^{1+\epsilon} e^{-tx}\, dt = \frac{1}{x}e^{-\epsilon x}(1-e^{-x}) . $$ Then clearly $f\le g$ near infinity, and near zero, $f(x)\simeq 1-(1/2+\epsilon)x$, $g(x)\simeq 1-(\epsilon^2+(1-\epsilon)\alpha)x$. I also want $f\le g$ near zero, and this gives a condition on $\alpha$ (given $\epsilon$). For small $\epsilon$, I can take $\alpha\approx 1/2$.

On the other hand, as $\epsilon\to 0$, $\alpha\to 1/2$, we have that $f(1)\to 1-e^{-1}$, $g(1)\to e^{-1/2}$, and now a calculator will tell us that $f(1)>g(1)$ for sufficiently small $\epsilon$. (There might be easier counterexamples.)

$\endgroup$
4
  • $\begingroup$ Thanks Christian, seems like a nice counterexample. Do you think it would still fail if I assume further that $f(x) = \int_0^\infty e^{-tx} \phi_f(t) dt$ and $g(x) = \int_0^\infty e^{-tx} \phi_g(t) dt$, where $\phi_f,\ \phi_g >0$ on $\mathbb{R}_+$? $\endgroup$
    – Alphonse
    Sep 27, 2015 at 8:26
  • $\begingroup$ @Alphonse: I think essentially the same counterexample will still work, if we just take $\phi_{f,g}$ that approximate my choices $\phi_f=\chi_{(\epsilon, 1+\epsilon)}$, $\phi_g=\epsilon\delta_{\epsilon} + (1-\epsilon)\delta_{\alpha}$. $\endgroup$ Sep 27, 2015 at 13:12
  • $\begingroup$ Yes that is what I thought at first, but then I realized that you really needed a compact support for your example to work. If you make the support of $\phi_f$ 'bigger and bigger', then you lose the $1/x$ part which makes it work. $\endgroup$
    – Alphonse
    Sep 27, 2015 at 13:20
  • $\begingroup$ @Alphonse: I can always add the same positive function to both $\phi_f$, $\phi_g$, so a compactly supported counterexample would suffice. (There is the issue of not being allowed singular measures now, but I don't think this can be a serious problem.) $\endgroup$ Sep 27, 2015 at 13:27

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.