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.

As a specific example of this question and a follow up to this one does anyone know a nice way to calculate the germs at $z=1$ of

$$f(z)=\sqrt{1+\sqrt{z}}$$

My attempts have been messy at best, and I'd rather avoid trying to wade through Taylor series if I can! Any ideas would be most welcome!

share|improve this question
    
What exactly do you mean by "calculating the germ"? Do you want the power series of $f$ around $z= 1$ or are you asking for something else? –  mrf Jun 1 '12 at 16:09
1  
Of course, the point is that there's more than one power series, because this is a multi-valued "function"... –  Zhen Lin Jun 1 '12 at 16:43
add comment

3 Answers

up vote 7 down vote accepted
+150

I'm not sure what you're looking for since you asked two distinct questions in the other posts, so I'll answer both.

Power series of the principal branch

For all $|t|<1$: $$\left(\sqrt{1+t}+\sqrt{1-t}\right)^2=2\left(1+\sqrt{1-t^2}\right)$$ $$\frac{\sqrt{1+t}+\sqrt{1-t}}{2}=\frac{1}{\sqrt 2}\sqrt{1+\sqrt{1-t^2}}$$ The left hand side is simply the even part of $\sqrt{1-t}$, thus if $\sqrt{1-t}=\sum_{n=0}^\infty a_n t^n$ we have: $$\frac{1}{\sqrt 2}\sqrt{1+\sqrt{1-t^2}}=\sum_{n=0}^\infty a_{2n} t^{2n}$$ $$\sqrt{1+\sqrt z}=\sum_{n=0}^\infty \sqrt 2 a_{2n} (1-z)^n$$ (Note: $a_0=1$ and $a_n=-1/2 \frac{(2n-2)!}{n!(n-1)!4^{n-1}}$ for $n>0$.)

Call this germ $g_0$.

Other germs

If we interpret the square root as a multi-valued function, there are in principle 3 other branches obtained by flipping the sign of one or both radicals. But $\sqrt{1-\sqrt{z}}~\sim \sqrt{(1-z)/2}$ is not analytic around $z=1$, so around $z=1$ there are only two (distinct) germs $g_0$ and $-g_0$.

share|improve this answer
    
Thank you - this is just what I was looking for! Am I right in thinking that when $z \neq 1$ we do get $4$ germs? If I'm right, what's the best way to rigorously show it? Once I understand that I shall certainly award the bounty! Many thanks again! –  Edward Hughes Jun 2 '12 at 15:52
1  
Around $z=z_0\not\in\{0,1\}$, $1\pm\sqrt z$ is analytic and non-zero in a neighborhood of $z_0$, so $\pm\sqrt{1\pm\sqrt z}$ is analytic around $z_0$. You can trivially prove these 4 germs are distinct by looking at the value at $z=z_0$. –  Generic Human Jun 3 '12 at 9:10
    
Ah excellent - I understand now! Many thanks! –  Edward Hughes Jun 3 '12 at 10:03
add comment

We have $$ \sqrt{1+\sqrt{z}} = \sqrt{2}\left(1+\sum_{n=1}^\infty \frac{(-1)^{n-1}}{16^n n}\binom{4n - 2}{2n-1} (z-1)^n\right) $$

Here is a very general method, in the particular case of an algebraic function, to prove the identity.

I — Notations

Let $f$ be the function $z \mapsto \sqrt{1+\sqrt{1+z}}/\sqrt 2$, which is holomorphic in a neighbourhood of zero. We take a determination of the square root holomorphic around 1 and such that $\sqrt 1 = 1$. Note that in order to simply the computation, I shifted the variable and normalized the value at zero.

II — Algebraic equation

The function $f$ obviously satisfies the algebraic equation $$ (f(z)^2 - 1/2)^2 = \frac{1+z}{4}, $$ or, equivalently, $P(f(z), z) = 0$, where $$ P(Y, z) = 4Y^4 - 4Y^2 -z. $$

III — Differential equation

The function $f$ satisfies the following linear differential equation : $$ 16 z (z+1) f''(z) + 8(1+2z) f'(z) - f(z) = 0 $$

This is a general fact that an algebraic function satisfies a linear differential equation with polynomial coefficient.

IV — Recurrence

Write $f(z) = \sum_{n\geqslant 0} u_n z^n$. Thus $$ 16 z (z+1) f''(z) + 8(1+2z) f'(z) - f(z) = 16n(n-1)u_n + 16 (n+1)n u_{n+1} + 8 (n+1) u_{n+1} + 16 n u_n - u_n $$ which implies that $$ (4 n - 1)(4n+1)u_n + 8(n+1)(2n+1)u_{n+1} = 0. $$

V — Resolution

We check easily that the sequence defined by $$v_n = \frac{(-1)^{n-1}}{16^n n}\binom{4n - 2}{2n-1}$$ if $n>0$ and $v_0 = 1$ satisfies the first order recurrence above. Since $u_0 = 1 = v_0$, we can conclude that $$u_n = v_n$$

VI — Automation

Here is a Maple session showing how to automate the proof of steps II to V.

> with(gfun):
> f := sqrt(1+sqrt(1+z))/sqrt(2);
                                                              1/2 1/2  1/2
                                                  (1 + (1 + z)   )    2
                                             f := ------------------------
                                                             2

> holexprtodiffeq(f, y(z));      
                                                                         / 2      \
                                              /d      \        2         |d       |
                          {-y(z) + (8 + 16 z) |-- y(z)| + (16 z  + 16 z) |--- y(z)|, y(0) = 1}
                                              \dz     /                  |  2     |
                                                                         \dz      /

> diffeqtorec(%, y(z), u(n));    
                                         2                         2
                              {(-1 + 16 n ) u(n) + (24 n + 8 + 16 n ) u(n + 1), u(0) = 1}
> rsolve(%, u(n));
                                                                        n
                                                   GAMMA(2 n - 1/2) (-1)
                                              -1/2 ----------------------
                                                      1/2
                                                    Pi    GAMMA(2 n + 1)

You have to use the duplication formula for $\Gamma$ to retrieve the binomial coefficient.

share|improve this answer
    
Great answer! I think there's a small mistake in your final expression since you get $1/8^n$ instead of $1/16^n$. –  Generic Human Jun 2 '12 at 8:39
    
You're right, thanks ! –  Lierre Jun 2 '12 at 8:45
add comment

Let $w = \sqrt{1 + \sqrt{z}}$. Since taking square roots of complex numbers is unpleasant, let's get rid of them to get a nicer equation: $$(w^2 - 1)^2 = z$$ Since we want to expand around $z = 1$, we should change variables: set $u = z - 1$. Our problem now is to find $a_0, a_1, a_2, \ldots$ so that $w = a_0 + a_1 u + a_2 u^2 + \cdots$ satisfies $$(w^2 - 1)^2 = 1 + u$$

First, by neglecting terms of order $u$ or higher, we find that $({a_0}^2 - 1)^2 = 1$, so $$a_0 \in \{ \pm \sqrt{2}, 0 \}$$ (Officially, what we are doing is working in the power series ring $\mathbb{C} [[ u ]]$ modulo $u$.) Now, we divide into cases.

  1. We ignore the case $a_0 = 0$, because this corresponds to a ramification point in the associated Riemann surface and so $w$ has no power series here. (If you try to do the computation by brute force, you will find that all the coefficients are $0$!)

  2. Take $a_0 = \sqrt{2}$. Neglecting terms of order $u^2$ or higher, we get $$(w^2 - 1)^2 = ((\sqrt{2} + a_1 u)^2 - 1)^2 = (1 + 2 \sqrt{2} a_1 u)^2 = 1 + 4 \sqrt{2} a_1 u = 1 + u \pmod{u^2}$$ so $a_1 = \frac{1}{8} \sqrt{2}$. Now, neglecting terms of order $u^3$ or higher, we get $$(w^2 - 1)^2 = 1 + u + \left( \frac{5}{16} + 4 \sqrt{2} a_2 \right) u^2 = 1 + u \pmod{u^3}$$ so $a_2 = - \frac{5}{128} \sqrt{2}$. And so on.

  3. Take $a_0 = -\sqrt{2}$. Observe that nothing changes if we substitute $-w$ for $w$, so the coefficients for this expansion must be the same as the ones computed in (2), except with the opposite sign.

share|improve this answer
add comment

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.