6

I have the recurrence relation:

$$g(k, 0, x) = k,$$

$$g(k, n, x) = \dfrac{1}{2} \log_{k}{\left(\dfrac{k^{g(k, n - 1, x)}x}{g(k, n - 1, x)}\right)},$$

and I would like to solve it, if it is possible.

By the way, $\lim_{n \to \infty}{g(k, n, x)} = f^{-1}(k, x), f(k, x) = k^{x}x$.

Taylor
  • 1,128
  • Maybe I'm asking a stupid question, but how is $f^{-1}(k, x)$ defined? – tpb261 Jul 09 '15 at 09:53
  • @tpb261 I would define it as the inverse function of $f(k, x)$. There's no such thing as a stupid question. :) – Taylor Jul 09 '15 at 09:54
  • so, f(2, 3) = 24. then $f^{-1}(2, 24) = 3$. Am I right? Or are we trying to do this: $f^{-1}(3, 24) = 2$. A bit confused. Notation says the first one. Jus confirming. – tpb261 Jul 09 '15 at 09:57
  • @tpb261 Yes, the first is correct! :) – Taylor Jul 09 '15 at 09:58
  • This does not look like a linear algebra question... By the way, have you tried using properties of the log on the first expressiong of $g$? It looks like you get something similar to what you have in your update, but simpler. I don't think you can get much more, since you are shuffling exponentials and rational functions... – bartgol Jul 09 '15 at 18:08
  • @bartgol Trust me, I have tried almost everything, but my skill set is limited by the fact that I am only a high school student (well, I left in June), but yes, I have tried using the properties of logarithms. – Taylor Jul 09 '15 at 19:04
  • Why the W hate? – anon Jul 10 '15 at 12:29
  • $\log_k$ means what? Logarithm to the base $k$? $k$-times iterated logarithm? Something else? – Gerry Myerson Jul 10 '15 at 13:04
  • "Please, do not even bother putting an answer involving the Lambert W function, or I will not accept your answer": if you leave this comment, I will not even bother to look at your question. –  Jul 10 '15 at 13:37
  • @YvesDaoust I am sorry, but it is just that, on previous questions, people have likened this recurrence relation to the Lambert W function, which, I guess is the same, but the Lambert W function does not yield any values, but this does. I will remove this from my question, and I am sorry for making you feel this way. – Taylor Jul 10 '15 at 14:09
  • @YvesDaoust I have removed this from the post, and, again, I apologise for any offence caused. I would appreciate it if you would take a look at the question, but I understand if you do not want to. – Taylor Jul 10 '15 at 14:10
  • @GerryMyerson $\log_k$ means logarithm with base-$k.$ – Taylor Jul 10 '15 at 14:11
  • @anon I have removed this from my post, and, likewise to YvesDaoust, I am sorry. – Taylor Jul 10 '15 at 14:11
  • @Taylor: it is much more acceptable when you explain the reason :-) –  Jul 10 '15 at 16:01
  • 1
    A meager result: when $x=k^{k+1}$, $g(k,n,x)=k$, for all $n$. –  Jul 10 '15 at 16:18
  • @YvesDaoust Hmm, nice! – Taylor Jul 10 '15 at 16:19
  • This recurrence relation is not fully specified. You should add information about the domain of $n,k,x$ and a complete set of initial conditions. At the time it's not even clear if you are looking for a family of functions with parameter $n$ or for a family of functions iterated over $n$ and $k$. – Markus Scheuer Jul 11 '15 at 22:03
  • How did you know or decide that $g(k,n,x)$ equals $k$ at $n=0$ and $k^xx$ at $n=\infty$? – tyobrien Jul 13 '15 at 00:28
  • @TyO'Brien It is clear by the graph that it traces. – Taylor Jul 13 '15 at 00:31
  • So the function you graphed was the exact recursive formula as shown above? – tyobrien Jul 13 '15 at 00:37

1 Answers1

3

I understand simplify the $n$-th term not solve the recurrence relation.

Noting $u_n=g(k,n,x)$ one has $$2u_n = \log_{k}{\left(\dfrac{k^{g(k, n - 1, x)}x}{g(k, n - 1, x)}\right)}$$ i.e.$$2u_n=\log_{k}\frac {k^{u_{n-1}}x}{u_{n-1}}\iff k^{2u_n}=\frac {k^{u_{n-1}}x}{u_{n-1}}$$ Hence $$u_{n-1}k^{(2u_n-u_{n-1})}=x$$

►"Answer" (Bis):Please allow me a note about this question I can not entirely understood before. A recurrence relation is well defined when initial conditions are given. In this case, it appears $u_0=k$ to be sufficient. Anyway, one has $$ 2u_n=\log_{k}\frac {k^{u_{n-1}}x}{u_{n-1}} ; u_0=k \Rightarrow x=k^{(u_1+1-k)}$$ So, the variable $x$ must be constant? The answer to this is, simply, here the recurrence relation is not of numbers but of functions, only the first term $u_0$ is constant. (I did not see this and I was looking for numbers). What follows is my "answer" as I found it, however I beg you read the REMARK below. We have$$ u_{n-1}k^{(2u_n-u_{n-1})}=x \Rightarrow \frac {u_n}{u_{n-1}}=k^{(3u_n-u_{n-1}-2u_{n+1})} $$ Note both telescopic (1) and (2):

(1) $$\frac {u_1}{u_0}\cdot\frac {u_2}{u_1}\cdot\frac {u_3}{u_2}\cdot.......\cdot\frac{u_n}{u_{n-1}}=\frac{u_n}{u_0}$$ (2)$$\sum_{k=0}^{k=n}(3u_k-u_{k-1}-2u_{k+1})=-2u_{n+1} +u_n+2u_1-u_0$$ Hence $$\frac {u_n}{u_0}=k^{(-2u_{n+1} +u_n+2u_1-u_0)}$$ But $$2u_1=\log_{k}\frac{k^{u_0}x}{u_0}\Rightarrow 2u_1-u_0=\log_{k}\frac {x}{u_n}$$ i.e. $$u_n=k^{(-2u_{n+1}+u_n+\log_{k}x)}$$ Thus $$\boxed{ 2u_{n+1}=u_n+\log_{k}\frac {x}{u_n}}$$ Let us calculate three terms (the first one is the given initial condition). $$u_0=k$$ $$u_1=\frac {k+\log_{k}\frac {x}{k}}{2}$$ $$2u_2=\frac {k+\log_{k}\frac {x}{k}}{2}+\log_{k}\frac {2x}{k+\log_{k}\frac {x}{k}}$$ The terms become more and more complicated as n increases.

REMARK.- See this and related, please. All the above can be shortened considerably from the given by @Taylor formulation and the "simplification" given before. If the terms of the recurrence are regarded as what they are, functions and not numbers, then the "answer" is easy and short. Actually this "answer" (Bis) is exactly the formulation given by @Taylor!!! which is easy to verify.

Piquito
  • 29,594
  • I am sorry, but I did not mean for you to solve the recurrence relation for $x.$ My question may have been misleading, but I am, in fact, looking for an explicit formula for the recurrence relation.

    Again, I am sorry for misleading you.

    However, I will upvote your answer, but, I am afraid that I cannot award you with the bounty, as this answer is not what I was looking for.

    – Taylor Jul 10 '15 at 14:08
  • Anyway you maybe have something useful: $u_{n-1}k^{(2u_n-u_{n-1})}$ is the identity function for all$n$. Regards. – Piquito Jul 10 '15 at 14:52
  • Yes, I know that, and I found that yesterday, also. Thank you for help. – Taylor Jul 10 '15 at 14:54
  • 1
    Good luck. I have leaved this problem because studying another one (not of today). – Piquito Jul 10 '15 at 16:18
  • So, Taylor, it's not at all clear then what you are looking for. Ataulfo's form of the recurrence looks very neat --- do you have some reason to believe it's possible to do better? or some concrete idea of what "better" means? – Gerry Myerson Jul 11 '15 at 04:43
  • @GerryMyerson I wanted to find an explicit formula for the recurrence relations. I am sorry, as this was unclear in my question. I thought that solving a recurrence relation was the same as finding its explicit formula. Also, as I have already said, I did find this form as well, unfortunately, as it is a rather easy manipulation of the above formulae. So, to be clear, I want to find an explicit formula. I do not know if this is possible. If it is not, I will be more than happy to award Ataulfo with the bounty, unless another more detailed / helpful answer arises. – Taylor Jul 11 '15 at 10:48
  • Still not clear, Taylor. Do you, or don't you, want a solution for the recurrence? that is, a formula of the form $u_n=\dots$ where the right side depends only on $n$, not on any of $u_{n-1},u_{n-2},\dots$? If you want a solution, please edit your question so it says you want a solution. – Gerry Myerson Jul 11 '15 at 12:43
  • @GerryMyerson Yes, that is what I want. – Taylor Jul 11 '15 at 13:27
  • Good. Then please edit that into the body of your question. People shouldn't have to go through a long list of comments to find out that when you write "simplify", you actually mean "solve". Also, note that most recurrence relations don't have solutions in closed form. Is there any reason to believe that your recurrence has such a solution? – Gerry Myerson Jul 11 '15 at 23:54