Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I'm looking for the solution $x$ of

$$x^n+nx-n=0.$$

Thoughts: From graphing it for several $n$ it seems there is always a solution in the interval $[\tfrac{1}{2},1)$. For $n=1$, the solution is the fraction $\tfrac{1}{2}$ and for higher $n$, the solution shifts to the right.

I then saw that the equation reads $$F_n(x)=x$$ with

$$F:[0,1]\to[0,1],\ \ \ \ \ \ F_n(x):=1-\frac{x^n}{n}.$$

I think I got all conditions together for making the iteration of $F_n$'s the general soluton for the equation. I computed $$x_5=F_5(F_5(F_5(F_5(F_5(F_5(F_5(F_5(x_S)))))))),$$

with starting value $x_S=\tfrac{1}{2}$ and it seems to be the solution of the equation for $n=5$.

Related wikipedia links: Fixed point theorem, Banach fixed-point theorem, Fixed point iteration;

I wonder:

Have I found the soluton and can one evaluate the iteration to a closed form?

Might be that it involves polylogs. edit: At least Wolfram Alpha claims to know $$n(x)=W\left(\frac{-\log(x)}{x-1}\right)/\log(x),$$ even if $n(x)$ isn't too interesting.

In my case, how is the relation between the function $x(n)$ and (I think) the fixed point combinator for the iteration?

Does it matter what I choose for $x_S$ here? Can one relate it's value and the number for necessary iteration for a good agreement with the real value?

(Also, are there any results on which polynomial equations this technique works? The property $F:[0,1]\to[0,1]$ seemed accidental to me.)

share|improve this question
    
A related problem. –  Mhenni Benghorbal Feb 6 '13 at 13:07
4  
@Nick Kidman: Not a solution, but a remark: the fact that there is always a solution on $(1/2,1)$ for $n>1$ follows directly from the fact that $f_n(1/2)=1/2^n+n/2-n=1/2^n-n/2<0$ and $f_n(1)=1+n-n=1>0$. So, by the intermediate value theorem, there exists $c\in(1/2,1)$ such that $f_n(c)=0$. –  Dennis Gulko Feb 6 '13 at 13:07
    
It seems like you want to do something like Newton Raphson (or root approximation). Note that for $n=1$, you will not converge towards a solution if you use any starting value that is not $x_S = \frac {1}{2}$. –  Calvin Lin Feb 6 '13 at 13:25

2 Answers 2

You can get a good approximation of the solution as $n \to \infty$ by supposing that $x$ can be written as an asymptotic series in powers of $1/n$, say

$$ x \sim 1 + \sum_{k=1}^{\infty} \frac{a_k}{n^k}, $$

then substituting this into the given equation and calculating the coefficients recursively. For example we can calculate $a_1$ and $a_2$ by writing

$$ \begin{align*} 0 &\approx \left(1 + \frac{a_1}{n} + \frac{a_2}{n^2}\right)^n + n\left(1+\frac{a_1}{n} + \frac{a_2}{n^2}\right) - n \\ &= \left(1 + \frac{a_1}{n} + \frac{a_2}{n^2}\right)^n + a_1 + \frac{a_2}{n} \\ &= a_1 + e^{a_1} + \frac{2a_2 (1+e^{a_1}) - a_1^2 e^{a_1}}{2n} + O\left(\frac{1}{n^2}\right). \end{align*} $$

By sending $n \to \infty$ we get that $a_1 + e^{a_1} = 0$ and so

$$ a_1 = -W(1), $$

where $W$ is the Lambert W function.

Then, setting the coefficient of $1/n$ to $0$ and substituting the above value of $a_1$ we find that

$$ a_2 = \frac{W(1)^3}{2(1+W(1))}. $$

Thus we have

$$ x \approx 1 - W(1) n^{-1} + \frac{W(1)^3}{2(1+W(1))} n^{-2}. $$

This approximation seems to be pretty good. Below is a plot which compares the numerical roots with the asymptotic formula for $1 \leq n \leq 10$.

enter image description here

This series might actually converge for large enough $n$ but I don't see how to prove it. If the implicit function theorem could be employed then you would be set.

share|improve this answer

Isn't this simply the result of the Banach fixed-point theorem? In you case one must prove that there exists $q$ such that: $$d(F_n(x),F_n(y))=\frac{|y^n-x^n|}{n} \le q |x-y|$$ Which is true for all $n>1, x>0$, with say, $q=0.9$, So that it doesn't matter what $x_S$ you choose.

Regarding a closed form - convergence of an iterative series says nothing about the existence of such a closed form solution.

share|improve this answer
    
Yes, I obtained the solution that way, I haven't used that before though and so I wasn't sure, especially since I solved all $n$ equations simultaneously and I couldn't find an explicit reference regarding the value of starting values $x_S$. I added some references in the quation now too. / "Convergence of an iterative series says nothing about the existence of such a closed form solution." - well yes, that's why I'm asking for this specifc example :) Wolfram Alpha computes $n(x)$ explicitly in terms of polylogs but $n$ in terms of $x$ is some kind of useless and I don't know hot to invert it. –  NikolajK Feb 6 '13 at 13:51
    
@NickKidman - if you knew that, what makes you think that your iterative series has anything to do with a closed form solution? the question should be re-titled "does a closed form solution for equation X exist". –  nbubis Feb 6 '13 at 13:55
    
I now added a remark why I think one might exist in the comment above. That's also why I wrote about polylogs in the question, I though it was long enough already. –  NikolajK Feb 6 '13 at 13:56

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

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