6

Suppose I have a closed line segment (inside the punctured closed unit disk) joining two fixed points of the unit circle of the complex plane. What is the image of this segment on the Riemann surface of the square root through the map $f(z)=z^2$? How can I visualize this image?

et $z_1, z_2$ be the fixed points with $|z_1|=|z_2|=1.$ Then a point on the line segment can be written as $z_t=tz_1+(1-t)z_2$ for $0\le t\le 1.$ Then $$f(z_t)=t^2 (z_1-z_2)^2+2z_2t(z_1-z_2)+z_2^2.$$ Since $f$ is a bijection, the image of the segment should be a simple curve inside the closed unit disk (of the Riemann surface of the square root) joining two points on the boundary. Is there any way that I can visualize this?

Bumblebee
  • 18,974
  • 1
    $f(z) = z^2$ is not a bijection. And why do you mention “square root”? – Martin R May 10 '25 at 09:01
  • @MartinR: My understanding is that the square root function is a bijection from its Riemann surface to the punctured complex plane. Isn't squaring the inverse function? – Bumblebee May 10 '25 at 09:11
  • 1
    Well, a line segment in the punctured unit disk (that wasn't mentioned before) is contained in a rotated halfplane, where $z^2$ is injective. – Martin R May 10 '25 at 09:27

1 Answers1

9

The image of a segment by $z^2$ is a quadratic Bézier curve

The image of a segment by $z^2$ is a quadratic Bézier curve (https://en.wikipedia.org/wiki/B%C3%A9zier_curve) in the plane $\mathbb{C} \simeq \mathbb{R}^2$.

A quadratic Bézier curve is either an arc of a parabola or a straight line segment. Convert quadratic bezier curve to parabola

Let $z_1, z_2 \in \mathbb{C}$ be two distinct points inside the unit disk. The line segment joining them can be parametrized as $$ z(t) = (1 - t) z_1 + t z_2, \quad t \in [0,1]. $$

Applying the squaring map $z \mapsto z^2$, we obtain the curve $$ w(t) = z(t)^2 = \left[(1 - t) z_1 + t z_2\right]^2. $$

Expanding the square: $$ w(t) = (1 - t)^2 z_1^2 + 2(1 - t)t z_1 z_2 + t^2 z_2^2. $$

This is exactly the form of a quadratic Bézier curve in the complex plane, with control points: $$ P_0 = z_1^2, \quad P_1 = z_1 z_2, \quad P_2 = z_2^2. $$

Therefore, the image of the segment under the map $z \mapsto z^2$ is the curve $$ w(t) = (1 - t)^2 P_0 + 2(1 - t)t P_1 + t^2 P_2, \ with \ t\in [0,1]. $$

Why is a quadratic Bézier curve a segment of a parabola?

We will consider only the case when $|z_1| = |z_2| = 1$. For the general case, see the reference above. If you visualize a few examples as shown in the images below, you will see that they clearly look like arcs of parabolas.

This is especially evident in the case $|z_1| = |z_2| = 1$, where the Bézier curve has an obvious line of symmetry passing through the origin, aligned with the direction of the midpoint $(z_1^2 + z_2^2)/2$.

Indeed, in this case we have $||P_0|| = |z_1|^2 = 1$ and $||P_2|| = |z_2|^2 = 1$. We assume that $z_1 \neq \pm z_2$ and, to avoid a special case that would require a slightly different approach, we also assume $z_1 \neq \pm i z_2$. We will deal with the case $z_1=\pm i z_2$ later.

Under this assumption, the vectors $z_2^2 - z_1^2$ and $z_2^2 + z_1^2$ are nonzero and orthogonal, since $$ \langle P_2 + P_0, P_2 - P_0 \rangle = ||P_2||^2 - ||P_0||^2 = 0. $$ So they form an orthogonal basis of $\mathbb{R}^2$. In particular, there exist real numbers $C$ and $D$ such that $$ z_1 z_2 = C(z_2^2 + z_1^2) + D(z_2^2 - z_1^2). $$

But in fact $D = 0$, because $$ \frac{z_2^2 + z_1^2}{z_1 z_2} = \frac{z_2}{z_1} + \frac{z_1}{z_2} = u + \overline{u} = 2\,\text{Re}(u) \in \mathbb{R}, $$ where $u = z_2/z_1$ satisfies $|u| = 1$, so $z_1/z_2=u^{-1} = \overline{u}$. Because $z_1 \neq \pm z_2$, we have $C \neq \pm \frac{1}{2}$.

Now reparametrize the curve by replacing $t \in [0,1]$ with a new parameter $\alpha \in [-1/2, 1/2]$ defined by $\alpha = t - 1/2$. Then,

$$ w(t) = \left(\frac{1}{4} - \alpha + \alpha^2\right) z_1^2 + 2\left(\frac{1}{4}-\alpha^2 \right) z_1 z_2 + \left(\frac{1}{4} + \alpha + \alpha^2\right) z_2^2, \tag{1} $$ which simplifies to $$ w(t) = \alpha (z_2^2 - z_1^2) + \left( \frac{1}{4}(1 + 2C) + \alpha^2(1 - 2C) \right)(z_2^2 + z_1^2). $$

So, in the basis $(z_2^2 - z_1^2, z_2^2 + z_1^2)$, the image of the line segment is the arc of a parabola parametrized by $$ \alpha \mapsto \left( \alpha, \frac{1}{4}(1 + 2C) + \alpha^2(1 - 2C) \right), \ with \ \alpha \in [-1/2,1/2]. $$

This is indeed a parabola that does not pass through the origin because $C \neq \pm \frac{1}{2}$.

Note also that we indeed proved that the image of the line segment between $z_1$ and $z_2$ is the entire parabola, since the argument did not rely on the restriction $t \in [0,1]$.

For the case $z_1=\pm i z_2$, that correspond to case of angle $\pi/2$ between $z_1$ and $z_2$, we will have $z_1^2=-z_2^2$ and $z_1z_2=iz_2^2$ and we can consider the simpler orthonormal basis $(z_2^2,iz_2^2)$ instead. From (1) we get, replacing $\alpha\in [-1/2,1/2]$ by $\beta=2\alpha\in [-1,1]$,

$$w(t)=\beta z_2^2 + \left(\frac{1}{2}-\frac{1}{2}\beta^2\right)iz_2^2,$$

So in this orthonormal coordinate system, the parabola is given by

$$\beta \mapsto \left(\beta, \frac{1}{2} - \frac{1}{2} \beta^2\right), $$

which is convenient because we can directly apply the standard formula for the focus of a parabola of the form $y = a + b x^2$ in this case.

$$F=\left(0,a+\frac{1}{4b}\right)=\left(0,0 \right).$$

The focus of the parabola is the origin.

So the focus in the case $z_1 = \pm i z_2$ is the origin. Since the examples below and Eric in the comments suggest that this is always the case, we will now prove it. When $z_1 \neq \pm i z_2$, the basis $(z_2^2 - z_1^2, z_2^2 + z_1^2)$ may not be orthonormal, so we cannot use the classical formula for the focus unless we perform boring and lengthy calculations to normalize the basis. Instead, I will use a different approach.

If the angle between $z_1$ and $z_2$ is smaller than $\pi/2$, it is easy to see that one can find a radius $r > 1$ such that the line between $z_1$ and $z_2$ intersects the circle $|z| = r$ at two points $w_1$ and $w_2$ such that the angle between them is $\pi/2$.

Analogously, if the angle between $z_1$ and $z_2$ is larger than $\pi/2$, we can find a radius $r < 1$ such that the line between $z_1$ and $z_2$ intersects the circle $|z| = r$ at two points $w_1$ and $w_2$ such that the angle between them is again $\pi/2$.

enter image description here

Since the line between $z_1$ and $z_2$ has the same image under $z \mapsto z^2$ as the line between $w_1$ and $w_2$, its image under $z^2$ is the same parabola.

Now, the points $w_1/r$ and $w_2/r$ are in the unit circle and they have an angle of $\pi/2$ between them. We already proved that the image of the segment $[w_1/r, w_2/r]$ is an arc of a parabola with focus at the origin. This implies that the image of $[w_1, w_2]$ under $z^2$ is also a parabola with focus at the origin. I am using two facts here: for $r \neq 0$ and $S \subset \mathbb{C}$, define $rS = \{rs \colon s \in S\}$. Then:

A. For every $r \neq 0$, the transformation $z \mapsto rz$ takes a parabola $P$ with focus $F$ to a parabola $rP$ with focus at $rF$. This follows from the geometric definition of a parabola.

B. Let $\phi(z) = z^2$. If $S$ is a subset of $\mathbb{C}$, then $\phi(rS) = r^2 \phi(S)$. This is a simple calculation.

Examples

Examples if the extremes lie in the unit circle $|z_1|=|z_2|=1$

enter image description here

enter image description here

enter image description here

enter image description here

Examples if the extremes do not lie on the unit circle

Bezier curve

enter image description here

enter image description here

enter image description here

enter image description here

  • 1
    Wow. I mean the form is clear but I didn't think to relate it to Bézier curves. Cool. – Jackozee Hakkiuz May 11 '25 at 01:11
  • 1
    Note that Bézier curves are just parts of parabolas (which have the origin as a focus). – Eric May 11 '25 at 01:25
  • @Eric Indeed. I am trying to incorporate this in the answer. The case $|z_1|=|z_2|$ is easier to do it, but you can find the complete think here https://math.stackexchange.com/questions/1257576/convert-quadratic-bezier-curve-to-parabola – Daniel Smania May 11 '25 at 01:36
  • 1
    Fwiw, you could also prove the parabola bit without explicitly calculating the orthonormal base by noting that $z=1+ix/2 -> 1-x^2/4+ix$ is clearly a parabola with origin focus and that all lines not through the origin are just that line rotated and scaled. – Eric May 12 '25 at 19:33
  • 1
    @Eric That is exactly what I did. But you are right, It could be simpler – Daniel Smania May 12 '25 at 19:34
  • Ahh, misunderstood your third part (though the casework based on rotation and getting the angle to be pi/2 seems unnecessary to me). I agree you mostly did this. – Eric May 12 '25 at 19:40