2

I am trying to find the roots of equation $z = 1 + \sqrt{z}$ using Lagrange's expansion given on this book page no 15.

I expanded $z = 1 + \zeta z^p$ as $$z = 1 + \zeta + \frac{2p}{2!}\zeta^2 + \frac{3p(3p-1)}{3!}\zeta^3+\frac{4p(4p-1)(4p-2)}{4!}\zeta^4 + \dots$$ putting $\zeta = 1$ and $p = 1/2$ I obtained the series $$z = 1 + 1 + \frac{1}{2} + \frac18 + \frac{-1}{128} + 0 + \dots$$

The series seems to have formula $\displaystyle 2 + \sum_{n=2}^\infty \frac 1{n!}\prod_{k=0}^{n-2} (n \cdot \frac12 - k)$ which I found it to be $2 + \frac12(1 + \sqrt 5)$ from Mathematica while the root of $z = 1 + \sqrt z$ is $1 + \frac12(1 + \sqrt 5)$. Can anyone help me where it went wrong? Also can anyone help me to evaluate that series with hand?

hasExams
  • 2,285

1 Answers1

3

You can solve the equation explicitly by squaring both sides then solving the resulting quadratic. Compare the root you find with sum of the first terms of the series you gave.

In more detail.

We have

$$ z - 1 = \sqrt{z} \quad \Longrightarrow \quad z^2 - 2z + 1 = z, \tag{1} $$

so that, by the quadratic formula, the two possible roots are

$$ z = \frac{3 \pm \sqrt{5}}{2}. $$

The smaller candidate root is not a solution to the original equation since

$$ \frac{3 - \sqrt{5}}{2} < \frac{3 - \sqrt{4}}{2} = \frac{1}{2}, $$

which would make the left-hand side negative and the right-hand side positive in the first equation in $(1)$. Thus we conclude that the only possible root is

$$ z = \frac{3 + \sqrt{5}}{2} \approx 2.61803. $$

The sum of the first terms of the series you found is approximately

$$ 1 + 1 + \frac{1}{2} + \frac18 - \frac{1}{128} \approx 2.61719, $$

so it looks like you've got everything right.

Evaluating the series directly.

If you'd like, you can get the closed form for your series the hard way by noticing that

$$ \prod_{k=0}^{n-2} (np - k) = \frac{(np)!}{(np-n+1)!}, $$

for $n \geq 2$, so that the general term in your sum is

$$ \frac{(np)!}{n!(np-n+1)!} \zeta^n = \frac{1}{n} \binom{np}{n-1} \zeta^n \tag{2} $$

Let's take $p=1/2$ now and consider only the odd terms, so that $n = 2m+1$. The expression in $(2)$ becomes

$$ \frac{1}{2m+1} \binom{m+1/2}{2m} \zeta^{2m+1}. $$

Thanks to Sasha's answer here this is equal to

$$ \frac{1}{4^m} \binom{1/2}{m} \zeta^{2m+1} = \zeta \binom{1/2}{m} \left(\frac{\zeta^2}{4}\right)^m, $$

so that the sum of the odd terms in your series is

$$ \zeta \sum_{m=0}^{\infty} \binom{1/2}{m} \left(\frac{\zeta^2}{4}\right)^m = \zeta \sqrt{1+\zeta^2/4} = \frac{\zeta}{2}\sqrt{4+\zeta^2}. \tag{3} $$

Next we consider the even terms. By taking $n = 2m$ in $(2)$ we see that the general even term is

$$ \frac{1}{2m} \binom{m}{2m-1} \zeta^{2m}. $$

For all $m > 1$ the quantity $2m-1$ is a negative integer, and in that case $\binom{m}{2m-1} = 0$. Thus the sum of the even terms is just

$$ 1 + \frac{1}{2} \binom{1}{1} \zeta^2 = 1 + \frac{1}{2} \zeta^2. \tag{4} $$

Summing $(3)$ and $(4)$ we find that

$$ \begin{align} z &= 1 + \zeta + \frac{2p}{2!}\zeta^2 + \frac{3p(3p-1)}{3!}\zeta^3+\frac{4p(4p-1)(4p-2)}{4!}\zeta^4 + \dots \\ &= 1 + \frac{1}{2} \zeta^2 + \frac{\zeta}{2}\sqrt{4+\zeta^2}. \end{align} $$

By taking $\zeta = 1$ we see that this agrees with the result we found earlier, that $z = (3+\sqrt{5})/2$.

  • could you help me evaluate the series ... please – hasExams May 07 '13 at 15:58
  • @hasExams Did you do what I suggested? What you will find is precisely the closed form of the series. – Antonio Vargas May 07 '13 at 16:52
  • yes ... the latter one from series is $1$ more than the value of root from quadratic. I don't know what went wrong. this is my first time handling Lagrange expansion. – hasExams May 07 '13 at 17:12
  • also I had put 2 + Sum[Product[n 1/2 - k, {k, 0, n - 2}]/n!, {n, 2, Infinity}] on Mathematica. – hasExams May 07 '13 at 17:13
  • @hasExams I think you may have made a typo somewhere in checking your result. I've updated my answer with some details. – Antonio Vargas May 07 '13 at 18:07
  • Could you give me some hints to evaluate that series manually?? is it binomial series or ... something. I find it very interesting. – hasExams May 07 '13 at 18:12
  • @hasExams I hit a snag trying to evaluate the series and asked a question about it here. – Antonio Vargas May 07 '13 at 18:24
  • thank you very much @Antonio Vargas :) – hasExams May 07 '13 at 18:26
  • @hasExams I've updated the answer. As you can see, summing the series directly is a lot of work which is useless when you can already find the exact value of the sum using a different approach. Usually you won't be able to sum the series given by Lagrange inversion explicitly; the point is that it gives you a good enough numerical approximation if you use enough terms of the series. – Antonio Vargas May 07 '13 at 19:44
  • thank you very much for update ... I am seeing it. the original question was to show the equivalence. also this is very interesting!! – hasExams May 07 '13 at 19:45
  • @hasExams, You're welcome :). If you'd like to see another interesting application of the Lagrange formula you may like to read my answer this question from a while back. – Antonio Vargas May 07 '13 at 20:01
  • @hasExams, Check this one out too. – Antonio Vargas May 07 '13 at 21:12