Let $\mathcal{T}$ the set of sequences $t\mapsto at^\alpha$, with $t\in\mathbb{N}$, $\alpha>0$ and $a\in\mathbb{R}$. If necessary, it is possible to add boundedness assumptions, such that $\alpha\geq \alpha_{\min}>0$ and $0<a_{\min}\leq |a|\leq a_{\max}$.

I'm looking for a metric $d$ on $\mathcal{T}$ such that $$\forall\varepsilon>0,\, \exists \alpha(\varepsilon)>0,\, M(\varepsilon)>0,\, t(\varepsilon)\in\mathbb{N} : \left(d(u,v)>\varepsilon\right) \implies \left(\forall t\geq t(\varepsilon),\, |u(t)-v(t)|\geq M(\varepsilon)t^{\alpha(\varepsilon)}\right).$$

In other words, as soon as $d(u,v)>\varepsilon$ the rate of divergence of $u$ and $v$ is bounded from below by some function of $\varepsilon$ only.

I tried to use the metric $d(at^\alpha,bt^\beta) = \max(|a-b|,|\alpha- \beta)$, but it fails, as for example $d\left((1+2\varepsilon)t,t^{1+\frac{1}{n}}\right)>\varepsilon$, yet these two functions don't satisfy the desired property, as their rate of divergence can be made as small as possible by letting $n$ go to $\infty$.

So I'd like to either find a metric that works, or at least to prove that it does or does not exist.

  • $\begingroup$ There is another maybe useful metric on the space of monomials $at^\alpha$, $a>0$, $\alpha\in\mathbb R$. Put $d(at^\alpha,bt^\beta)=\max(|\ln a-\ln b|,|\alpha-\beta|)$. Then $d(at^\alpha,bt^\beta)=\mu$ implies $\frac{at^\alpha}{b t^\beta}\leq e^\mu t^\mu$ for $t>1$. Maybe taking quotients of the monomials makes more sense than taking differences. $\endgroup$ – Helmut Mar 27 '18 at 14:11

I'll write your $t(\epsilon)$ as $t_0(\epsilon)$ to avoid ambiguity. The condition implies that if $u(t)=v(t)$ for some $u,v,t,$ then $t<t_0(\tfrac12 d(u,v)).$ In particular $d(nt,t^2)\to 0$ as $n\to\infty,$ because the two functions are equal at $t=n.$ (To clarify: I mean $d(u_n,v)\to 0$ where $u_n$ is the monomial $nt^1$ and $v$ is the monomial $t^2.$) This means $nt\to t^2,$ but by the same logic $nt\to t^3,$ which means $d(t^2,t^3)=0,$ so $d$ is not a metric.

  • $\begingroup$ Could you elaborate on the triangle inequality? Did you mean "is implied by both"? Also, would it be possible to take $M(\varepsilon)=1$? $\endgroup$ – Augustin Mar 26 '18 at 13:48
  • $\begingroup$ @Augustin: that answer was wrong (the triangle inequality didn't work, I'd made a mistake when checking). The true answer seems to be that there is no such metric. $\endgroup$ – Dap Mar 26 '18 at 18:51
  • $\begingroup$ That seems correct to me. It means I spent a significant amount of time looking for something that doesn't exist... Anyway, thanks! $\endgroup$ – Augustin Mar 27 '18 at 9:59

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.