1

Let $\{x_n\}_{n=1}^\infty$ be a sequence satisfying the recurrence relation:

$$ x_n = a\left(1- \sum_{k=0}^{n-1}x_k\right) $$

Where $ x_0 = 1 $, and $a \in [0,1]$ is chosen so that $$ \sum_{k=1}^{\infty} x_k = 1$$

Given a positive integer $d$, how do I generate a sequence $\{y_n\}$ such that

$$ \sum_{k=p}^{p+(d-1)} y_k = x_{\frac {p+(d-1)}d}$$

for example, if $a = 0.5$ and $d = 4$,

$$ \sum_{k=1}^{4} y_k = x_1 = 0.5 $$

and

$$ \displaystyle \sum_{k=5}^{8} y_k = x_2 = 0.25 $$

I originally thought that this would be related to compound interest, but doing arithmetic by hand, have not found this to be the case.

I have limited mathematical knowledge, so if the answer requires anything beyond algebra, please explain or cite references to the form you are using. If the title of the question can be made clearer, please feel free to edit.

My use case is a computer application that will calculate $ g(x) $ from $x = 0$, so iterative solutions work for me.

Ben Grossmann
  • 225,327
  • Your notation is a bit baffling. I'm going to edit your question so that it says what I think you mean. However, if something I change is wrong, feel free to update my changes appropriately, or click "rollback" to bring the question back to its former state. – Ben Grossmann Aug 27 '14 at 04:02
  • Sorry, I realized after posting that I had made a couple of major mistakes in how I stated my problem. Hopefully, it's more readable now. I may still have some of the formatting wrong, esp. with regard to how I refer to the series that I have in mind. – SailsMan63 Aug 27 '14 at 04:32
  • No problem. Can I fix the formatting, then? – Ben Grossmann Aug 27 '14 at 04:35
  • Sure. I've simply been trying to google formatting conventions, and I'm not always sure what to ask – SailsMan63 Aug 27 '14 at 04:36
  • Please, changing so much the post is not fair ! Start another question and, if you want, delete the first one. Currently, looking at the new version of the post, I bet that my naswer looks perfectly ridiculous. – Claude Leibovici Aug 27 '14 at 04:36
  • Just noticed one of the edits this morning, but feel hesitant about rolling back. (Don't want to add more to the confusion.) $d$ is not guaranteed to be integer - it could be any positive real. – SailsMan63 Aug 27 '14 at 16:12

2 Answers2

0

Note that so long as $-1 < a < 1$, we have $$ \sum_{x = 1}^\infty x a^x = \left(\sum_{x = 0}^\infty (x+1) a^x\right) - \left(\sum_{x=0}^\infty a^x\right) \\= \frac{1}{(1-a)^2} - \frac{1}{1-a} = \frac{a}{(1-a)^2} $$ So, in order to have $\sum_{x=1}^\infty f(x) = 1$, we would need to have $a$ such that $\frac a{1-a} = 1 \implies a = \frac {3 - \sqrt 5}2$.

As for your function $g$: it seems that for a fixed positive integer $d$, you require $$ \sum_{x = p}^{p+d-1} g(x) = f\left(\frac{p + d - 1}{d}\right) $$ In the case of $d = 4$, this precisely means that $$ g(1) + \cdots + g(4) = f(1)\\ g(5) + \cdots + g(8) = f(2)\\ \vdots \\ g(4n + 1) + \cdots + g(4n+4) = f(n+1) $$ One definition of $g$ that would work is $$ g(x) = \frac 14 f\left( 1 + \lfloor (x-1)/4 \rfloor \right) $$ Where $\lfloor x \rfloor$ is the "floor" or "greatest integer" function. It seems that there are infinitely many suitable solutions, and this is one of them.


As for the latest iteration of your question: let $x[k]$ denote $x_k$, because subscripts are annoying. If we are given a sequence $\{x_k\}$ satisfying those properties, then if you define $$ y[k] = \frac 1d x\left[ 1 + \lfloor (k-1)/d \rfloor \right] $$ then the resulting sequence $\{y_n\}_{n=1}^\infty$ should do the job. This works regardless of the conditions on $\{x_n\}.$

Ben Grossmann
  • 225,327
0

For the beginning, rewrite $$f(x) = xa^x=a x a^{x-1}=a\frac{d(a^x)}{da}$$ So, $$\displaystyle \sum_{x=1}^{\infty} f(x) = a \sum_{x=1}^{\infty}\frac{d(a^x)}{da}=a\frac{d}{da}\sum_{x=1}^{\infty}a^x=a\frac{d}{da}\Big(\frac{a}{1-a}\Big)=\frac{a}{(1-a)^2}$$ So, if $$\displaystyle \sum_{x=1}^{\infty} f(x) =1$$ solving $$\frac{a}{(1-a)^2}=1$$ corresponds to the quadratic $a^2-3a+1=0$ the roots of which being $$a_{\pm}=\frac{1}{2} \left(3\pm\sqrt{5}\right)$$ bur since $a [0,1]$ the only solution is $$a_{\pm}=\frac{1}{2} \left(3-\sqrt{5}\right)$$