Consider the complex function $$f(z) = \sqrt{z^2+1}.$$

Obviously, $f(z)$ has branch points at $z = \pm i$. One way of defining a branch cut would be to exclude the points on the imaginary axis with $|z| \geq 1$. Another way of defining a branch cut appears to be to exclude the (finite) region of the imaginary axis with $|z| \leq 1$.

If we define $f(z)$ as $$f(z) = e^{1/2\log(z^2+1)},$$ the first branch cut can be arrived at by taking the principal branch of $\log(z)$ with the branch cut $(-\infty,0]$. At first I thought the second branch cut could be arrived at by taking the branch cut $[0,\infty)$ for $\log(z)$. Indeed, this would exclude imaginary $z$ with $|z| \leq 1$, but it would of course also exclude all real $z$, thus constituting a different branch cut for $f(z)$.

I think it would be possible to arrive at the second branch cut for $f(z)$ differently, by defining $$f(z) = \sqrt{r_1r_2}e^{i(\theta_1+\theta_2)/2}, $$ where $r_1 = |z-i|, r_2 = |z+i|$, $\theta_1 = \arg(z-i), \theta_2 = \arg(z+i).$ However, I don't really like this approach, since I think $$f(z) = e^{1/2\log(z^2+1)}$$ is the proper way to define $f$.

Any comments on this? Is there a branch cut for $\log(z)$ which gives the correct branch cut for $f$?


Consider the function $$ h(z) = \exp\left\{ \int_{i}^{z}\frac{w}{1+w^2}dw \right\} = \exp\left\{\frac{1}{2}\int_{i}^{z}\frac{1}{w+i}+\frac{1}{w-i}\,dw\right\}, $$ where the integral from $i$ to $z$ is taken along any simple path from $i$ to $z$ in the open region $\Omega$ obtained by subtracting from $C$ the closed segment from $-i$ to $i$. It doesn't matter which path is chosen from $i$ to $z$ because the difference of exponent integrals along two different paths $\gamma_1$, $\gamma_2$ will be a closed path integral with the same winding number around $i$ as around $-i$, resulting in a difference of integrals that is an integer multiple of $2\pi i$. Hence $h(z)$ does not depend on the path chosen from $i$ to $z$, so long as it does not cross the segment from $-i$ to $i$. So $h$ is holomorphic in $\mathbb{C}\setminus [-i,i]$, and $$ h'(z) = \exp\left\{\int_{i}^{z}\frac{w}{1+w^2}dw\right\}\frac{z}{1+z^2}=h(z)\frac{z}{1+z^2} \\ \implies \left(\frac{h(z)^2}{1+z^2}\right)'=0. $$ After multiplying by an appropriate constant, the resulting function $g(z)=Ch(z)$ will satisfy $g(z)^2=1+z^2$ everywhere in $\mathbb{C}\setminus[-i,i]$.

So even thought there isn't a logarithm for $1+z^2$ in $\mathbb{C}\setminus[-i,i]$, everything works out because there is a "logarithm modulo integer multiples of $2\pi i$." I'll leave you to make sense of that phrase in whatever way you want.

There's no problem defining an actual holomorphic logarithm in $\mathbb{C}\setminus\{ (\infty,-i]\cup[i,\infty)\}$ because this is a simply connected region where $w/(1+w^2)$ is holomorphic, which gives an antiderivative of $w/(1+w^2)$ in this region, and hence a holomorphic square root function $\sqrt{1+w^2}$ in this region.

  • 1
    $\begingroup$ @ÉtienneBézout : I clarified the last paragraph. A function which is holomorphic on an open simply-connected region has a holomorphic antiderivative, and that's how you always get a nice root $\sqrt{1+w^2}$ when the region is simply connected and does not contain $\pm i$. $\endgroup$ Oct 3 '16 at 21:48
  • 1
    $\begingroup$ Sorry for the late reply. Thanks! Regarding my other question: I've been taught that we always define $f(z)^\alpha$ as $\exp(\alpha \log f(z))$. While it works, I guess it might not be true that we require this definition, if we want another branch cut. $\endgroup$ Oct 5 '16 at 9:43
  • 1
    $\begingroup$ @ÉtienneBézout : That is correct. For example, if you want $\sqrt{(z-1)(z-2)(z-3)}$, you can specify a region so that any closed path has to circle none or two roots. So you could remove the line from $1$ to $2$ and a line from $3$ to $\infty$ that doesn't pass through $1$ or $2$, and that would be a valid domain. Or you could create a simply-connected domain by removing segments from $1$ to $-\infty$, from $2$ to $i\infty$ and from $3$ to $+\infty$. The first would not technically give you a log in the plane (only on a Riemann surface,) while the second would be standard. $\endgroup$ Oct 5 '16 at 19:44
  • 1
    $\begingroup$ @ÉtienneBézout : Note also that you can have branch cuts along curves, too, and not just straight lines. There's nothing wrong with that. $\endgroup$ Oct 5 '16 at 19:50
  • 1
    $\begingroup$ @TrialAndError : what I don't get is why $ \int_p^z \frac{w}{1+w^2} dw$ is analytic (I am assuming you meant $p$ a point in the domain, and not $i$). In fact, if $\gamma_1$ and $\gamma_2$ are two paths, then it can be seen that $\int_{\gamma_1} \frac{dw}{w+i} - \int_{\gamma_2} \frac{dw}{w+i} = 2 \pi i \mathrm{Ind}_{-i}(\gamma_1 - \gamma_2)$ and $\int_{\gamma_1} \frac{dw}{w-i} - \int_{\gamma_2} \frac{dw}{w-i} = 2 \pi i \mathrm{Ind}_i(\gamma_1 - \gamma_2)$. But from this, can does it follow that the integral is well-defined? $\endgroup$
    – Ruben
    Jan 12 '17 at 0:17

I think the OP might have been originally looking for a construction similar to this:

Consider the following Möbius transformation:


It is not hard to verify that the image of the requested region under this transformation is: $$\phi(\mathbb{C}\setminus [-i,i])=\mathbb{C}\setminus(\infty, 0]$$ Where by $[a,b]$ I mean the straight finite line in $\mathbb{C}$ connecting $a$ and $b$.

In this region, the principal branch of $Log(z)$ is well defined, and so we can use it to take a square root $$g(z)=exp(\frac{1}{2}Log(\frac{z+i}{z-i})) \implies g(z)^2=\frac{z+i}{z-i}$$

This is almost what we need. In this case, we can receive the desired result by multiplying by a simple function:

$$f(z)=(z-i)g(z)\implies f(z)^2=(z-i)(z+i)=z^2+1$$

And $f$ is indeed continuous in the specified domain.


We will let $\Omega_1=\mathbb{C}\setminus\{ir\mid r\in\mathbb R,\left\vert r\right\vert\geq1\}$, and let $\Omega_2=\mathbb{C}\setminus\{ir\mid r\in\mathbb R,\left\vert r\right\vert\leq1\}$. Geometrically, $\Omega_1$ and $\Omega_2$ "have the same shape" on the Riemann sphere: they are related by the inversion $z\mapsto 1/z$. So it is not surprising that the inversion will come in handy. Indeed, if $f$ is a branch of $\sqrt{z^2+1}$ on $\Omega_1$, then $z\cdot f(z^{-1})$ is a branch on $\Omega_2$.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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