1 Introduction

Let \({\mathcal {F}}\) denote the class of all functions \(f\) which are univalent in \(\Delta \equiv \{\zeta \in {\mathbb {C}}\,:\,\left| \zeta \right| <1\}\), convex in the direction of the real axis and normalized by \(f(0)={f^{\prime }}(0)-1=0\). Recall that an analytic function \(f\) is said to be convex in the direction of the real axis if the intersection of \(f(\Delta )\) with each horizontal line is either a connected set or empty.

For a given subclass \(A\) of \({\mathcal {F}}\), the following sets: \(\bigcap _{f\in A} f(\Delta )\) and \(\bigcup _{f\in A} f(\Delta )\) are called the Koebe set for \(A\) and the covering set for \(A\). We denote them by \(K_A\) and \(L_A\), respectively. The radius of the largest disk with center at the origin contained in \(K_A\) is called the Koebe constant for \(A\). Analogously, the radius of the smallest disk with center at the origin that contains \(L_A\) is called the covering constant for \(A\).

In the class \({\mathcal {F}}\), we consider functions which satisfy the property of \(n\)-fold symmetry:

$$\begin{aligned} f(\varepsilon z)=\varepsilon f(z)\quad \text {for all}\quad z\in \Delta \ , \end{aligned}$$

where \(\varepsilon =e^{2\pi i/n}\). The subclass of \({\mathcal {F}}\) consisting of \(n\)-fold symmetric functions is denoted by \({\mathcal {F}}^{(n)}\). By the definition, for every \(f\in {\mathcal {F}}^{(n)}\) a set \(f(\Delta )\) is \(n\)-fold symmetric, which means that \(f(\Delta )=\varepsilon f(\Delta )\). In other words, \(f(\Delta )\) may be obtained as the union of rotations about a multiple of \(2\pi /n\) from a set \(f(\Delta )\cap \left\{ w: \arg w \in \left[ 0,2\pi /n\right] \right\} \). From this reason, the following notation is useful:

$$\begin{aligned} \Lambda _0=\left\{ w: \arg w \in \left[ 0,\frac{2\pi }{n}\right] \right\} ,\quad \Lambda _j=\varepsilon ^j \Lambda _0\ ,\ j=1,2,\ldots ,n-1 \end{aligned}$$

and

$$\begin{aligned} \Lambda ^*=\left\{ w: \arg w \in \left[ \frac{\pi }{2}-\frac{\pi }{n},\frac{\pi }{2}\right] \right\} . \end{aligned}$$

The main aim of the paper is to find the Koebe set and the covering set for the class \({\mathcal {F}}^{(n)}\) when \(n\) is an odd positive integer. Similar problems in related classes were discussed, for instance, in [1, 2, 5] and in the papers of the authors [3, 4].

At the beginning, let us consider the general properties of the Koebe sets and the covering sets for \({\mathcal {F}}^{(n)}\).

In [4], we proved that

Theorem 1

The sets \(K_{{\mathcal {F}}^{(n)}}\) and \(L_{{\mathcal {F}}^{(n)}}\), for \(n\in {\mathbb {N}}\), are symmetric with respect to both axes of the coordinate system.

Theorem 2

The sets \(K_{{\mathcal {F}}^{(n)}}\) and \(L_{{\mathcal {F}}^{(n)}}\), for \(n\in {\mathbb {N}}\), are \(n\)-fold symmetric.

To prove both the theorems, it is enough to consider functions

$$\begin{aligned} g(z)=\overline{f(\overline{z})} \end{aligned}$$
(1)

and

$$\begin{aligned} h(z)=-f(-z). \end{aligned}$$
(2)

Obviously,

$$\begin{aligned} f\in {\mathcal {F}}^{(n)} \Leftrightarrow g,h\in {\mathcal {F}}^{(n)}. \end{aligned}$$
(3)

Moreover, if \(D=f(\Delta )\) then \(g(\Delta )=\overline{D}\), \(h(\Delta )=-D\).

Taking (3) into account, it is clear that the coordinate axes are the lines of symmetry for both the sets \(K_{{\mathcal {F}}^{(n)}}\) and \(L_{{\mathcal {F}}^{(n)}}\) for all \(n\in {\mathbb {N}}\). Furthermore,

Lemma 1

Each straight line \(\varepsilon ^{j/4}\cdot \{\zeta = t\ ,\ t\in R\}\), \(j=0,1,\dots ,4n-1\) is the line of symmetry of \(K_{{\mathcal {F}}^{(n)}}\) and \(L_{{\mathcal {F}}^{(n)}}\) for every positive odd integer \(n\).

Proof

Let \(n\) be a positive odd integer and let \(D\) be one of the two sets: \(K_{{\mathcal {F}}^{(n)}}\) or \(L_{{\mathcal {F}}^{(n)}}\).

Since \(D\) is symmetric with respect to the real axis and the positive real half-axis contains one side of the sector \(\Lambda _0\), each rotation of the real axis about a multiple of \(2\pi /n\) is the line of symmetry of \(D\). Because of the equality

$$\begin{aligned} \{\zeta = t\ ,\ t\in R\}\cdot \varepsilon ^{1/2} = \{\zeta = t\ ,\ t\in R\}\cdot \varepsilon ^{(n+1)/2}\ , \end{aligned}$$

our claim is true for all even \(j\), \(j=0,1,\dots ,4n-1\).

Let \(n=4k+1\), \(k\ge 1\). The bisector of \(\Lambda _{(n-1)/4}\) divides this sector into two subsectors: \(\{w: \arg w \in [\pi /2-\pi /2n, \pi /2+\pi /2n]\}\) and \(\{w: \arg w \in [\pi /2+\pi /2n, \pi /2+3\pi /2n]\}\). Hence the imaginary axis is the bisector of the former. For this reason, each rotation of the imaginary axis about a multiple of \(2\pi /n\) is the line of symmetry of \(D\). Moreover,

$$\begin{aligned} \{\zeta = it\ ,\ t\in R\}\cdot \varepsilon ^{1/2} = \{\zeta = it\ ,\ t\in R\}\cdot \varepsilon ^{(n+1)/2}. \end{aligned}$$

Hence our claim is valid also for all odd \(j\), \(j=0,1,\dots ,4n-1\).

If \(n=4k+3\), \(k\ge 0\), then the bisector of \(\Lambda _{(n-3)/4}\) divides this sector into two subsectors: \(\{w: \arg w \in [\pi /2-3\pi /2n, \pi /2-\pi /2n]\}\) and \(\{w: \arg w \in [\pi /2-\pi /2n, \pi /2+\pi /2n]\}\). The imaginary axis is the bisector of the latter. Similar argument to the one for \(n=4k+1\) completes the proof for this choice of \(n\). \(\square \)

Theorem 3

The sets \(K_{{\mathcal {F}}^{(n)}}\) and \(L_{{\mathcal {F}}^{(n)}}\), for positive odd integers \(n\), are \(2n\)-fold symmetric.

Proof

Let \(D\) be one of the two sets: \(K_{{\mathcal {F}}^{(n)}}\) or \(L_{{\mathcal {F}}^{(n)}}\). Let \(w_0=|w_0|e^{i\varphi _0}\) be an arbitrary point belonging to the boundary of \(D\) such that \(\arg \varphi _0\in [0,\pi /2n]\).

It is sufficient to apply Lemma 1. Firstly, the symmetric point to \(w_0\) with respect to the straight line \(\varepsilon ^{1/4}\cdot \{\zeta = t\ ,\ t\in R\}\) is \(w_1=|w_0|e^{i(\pi /n-\varphi _0)}\). Secondly, the symmetric point to \(w_0\) with respect to the real axis is \(w_2=|w_0|e^{-i\varphi _0}\). Both points \(w_1\), \(w_2\) also belong to the boundary of \(D\). Consequently,

$$\begin{aligned} w_1=\varepsilon ^{1/2}\cdot w_2\ , \end{aligned}$$

which results in

$$\begin{aligned} D=\varepsilon ^{1/2}\cdot D. \end{aligned}$$

\(\square \)

Remark 1

On the basis of this lemma, we can observe that in order to find sets \(K\) and \(L\), it is enough to determine their boundaries only in the sector of the measure \(\pi /2n\).

2 Extremal Polygons and Functions

Let \(n\) be a fixed positive odd integer, \(n\ge 5\), and let \(K\) denote the Koebe set for \({\mathcal {F}}^{(n)}\).

We denote by \({\mathcal {W}}\), the family of \(n\)-fold symmetric polygons \(W\) convex in the direction of the real axis and which have \(2n\) sides and interior angles \(\pi +\pi /n\) and \(\pi -3\pi /n\) alternately.

Suppose that \(w_*\in \partial K \cap \Lambda ^*\), where \(\partial K\) stands for the boundary of \(K\). According to Theorem 1, \(-\overline{w_*}\in \partial K\). Consider the straight horizontal line containing a segment \(I=\{-\lambda \overline{w_*}+(1-\lambda )w_*:\lambda \in (0,1)\}\). There are two possibilities: the intersection of this line with \(K\) is either \(I\) or the empty set. Assume now that \(I\subset K\). We shall see that the second case holds only if \({{\mathrm{Re}}}w_*=0\).

Since \(K\) is \(n\)-fold symmetric, all points \(w_*\cdot \varepsilon ^j\), \(j=0,1,\ldots , n-1\) belong to \(\partial K\). On the one hand, \(w_*\in \Lambda ^*\) means that \(w_*\) has the greatest imaginary part among points \(w_*\cdot \varepsilon ^j\). On the other hand, \(w_*\in \partial K\) means that there exists \(f_*\in {\mathcal {F}}^{(n)}\) such that \(w_*\in \partial f_*(\Delta )\).

From the convexity of \(f_*\) in the direction of the real axis, at least one of the two horizontal rays emanating from \(w_*\) is disjoint from \(f_*(\Delta )\). Since \(\{w_*-t: t\ge 0\}\cap K=I\), it is a ray \(l=\{w_*+t: t\ge 0\}\) is disjoint from \(f_*(\Delta )\). Taking into account the \(n\)-fold symmetry of \(f_*\), all rays \(l\cdot \varepsilon ^j\), \(j=0,1,\ldots , n-1\) are disjoint from \(f_*(\Delta )\).

Observe that the point \(w_*\cdot \varepsilon ^{(n+1)/2}\) has the lowest imaginary part among points \(w_*\cdot \varepsilon ^j\), \(j=0,1,\ldots , n-1\). Only if \(\arg w_*=\pi /2\), this point is one of two points with the same imaginary part. The convexity in the direction of the real axis of \(f_*\) implies that one of two horizontal rays emanating from \(w_*\cdot \varepsilon ^{(n+1)/2}\) is also disjoint from \(f_*(\Delta )\). If this ray is of the form \(k_1=\{w_*\cdot \varepsilon ^{(n+1)/2}+t: t\ge 0\}\), then \((k_1\cdot \varepsilon ^j) \cap f_*(\Delta ) =\emptyset \) and, conseqeuntly, \(f_*(\Delta )\) is included in a polygon of the family \({\mathcal {W}}\). Indeed, the rays \(l\) and \(k_1\cdot \varepsilon ^{(n-1)/2}\) form a sector with the vertex in \(w_*\) and the opening angle \(\pi +\pi /n\). From the \(n\)-fold symmetry of \(f_*\), we obtain the polygon mentioned above. The conjugate angle to this opening angle is the vertex angle of the polygon at \(w_*\). It is easy to check that the angle of the polygon between \(l\cdot \varepsilon \) and \(k_1\cdot \varepsilon ^{(n+1)/2}\) has the measure \(\pi -3\pi /n\).

But there is another possibility, i.e., \(k_2\cap f_*(\Delta ) =\emptyset \), where \(k_2=\{w_*\cdot \varepsilon ^{(n+1)/2}-t: t\ge 0\}\).

If \(\arg w_*=\pi /2\), the set \({\mathbb {C}}\setminus \{k_2\cdot \varepsilon ^j, j=0,1,\ldots , n-1\}\) consists of two parts: an unbounded part and a bounded one which is a regular \(n\)-gon, see Fig. 1. A regular polygon is convex and it can be treated as the generalization of a set of the family \({\mathcal {W}}\). Every second side of this generalized polygon has the length 0.

Fig. 1
figure 1

Extremal \(n\)-gon for \(n=5\)

If \(\arg w_*\in [\pi /2-\pi /n,\pi /2)\), then the set \({\mathbb {C}}\setminus \{l\cdot \varepsilon ^j, k_2\cdot \varepsilon ^j, j=0,1,\ldots , n-1\}\) is not bounded and it is not convex in the direction of the real axis. Since \(w_*\cdot \varepsilon \in \partial f_*(\Delta )\), one of two horizontal rays emanating from this point is also disjoint from \(f_*(\Delta )\). If \(m_1=\{w_*\cdot \varepsilon +t: t\ge 0\}\) has no common points with \(f_*(\Delta )\), then \(w_*\notin \partial f_*(\Delta )\), because \({{\mathrm{Im}}}(w_*\varepsilon )<{{\mathrm{Im}}}w_*\) and \({{\mathrm{Re}}}(w_*\varepsilon )<{{\mathrm{Re}}}w_*\), a contradiction. For this reason, \(m_2\cap f_*(\Delta ) =\emptyset \), where \(m_2=\{w_*\cdot \varepsilon -t: t\ge 0\}\). Hence \((m_2\cdot \varepsilon ^j) \cap f_*(\Delta ) =\emptyset \).

In this way, we obtain \(3n\) rays emanating from \(n\) points: \(w_*\cdot \varepsilon ^j\), \(j=0,1,\ldots , n-1\). Let us take three rays starting from \(w_*\). These rays are: \(l=\{w_*+t: t\ge 0\}\), \(k_2\varepsilon ^{-(n+1)/2}=\{w_*-t\cdot \varepsilon ^{-(n+1)/2}: t\ge 0\}\) and \(m_2\varepsilon ^{-1}=\{w_*-t\cdot \varepsilon ^{-1}: t\ge 0\}\). The angles between them and the positive real half-axis are equal to: 0, \(-\pi /n\), \(\pi -2\pi /n\). It means that \(l\) lies in the sector with the vertex in \(w_*\) and with the sides \(k_2\varepsilon ^{-(n+1)/2}\) and \(m_2\varepsilon ^{-1}\). The opening angle of this sector is equal to \(\pi -3\pi /n\).

Consequently, \(f_*(\Delta )\) is included in a polygon generated by \(k_2\varepsilon ^{j}\) and \(m_2\varepsilon ^{j}\), \(j=0,1,\ldots , n-1\). This polygon belongs to the family \({\mathcal {W}}\). Two examples of members of \({\mathcal {W}}\) are shown in Fig. 2.

Fig. 2
figure 2

Extremal polygons for \(n=5\)

From now on, we assume that all members of \({\mathcal {W}}\) are open sets.

Let

$$\begin{aligned} f_\alpha (z)&= \int _0^z (1-\zeta ^n e^{-in\alpha })^{\frac{1}{n}} (1+\zeta ^n e^{-in\alpha /3})^{-\frac{3}{n}}d\zeta \quad ,\quad \alpha \in \left[ -\frac{3\pi }{n},\frac{3\pi }{n}\right] \ ,\end{aligned}$$
(4)
$$\begin{aligned} g_\alpha (z)&= \int _0^z (1-\zeta ^n e^{-in\alpha })^{\frac{1}{n}} (1-\zeta ^n e^{-i(n\alpha /3+5\pi /3)})^{-\frac{3}{n}}d\zeta \quad , \quad \alpha \in \left[ -\frac{2\pi }{n},\frac{4\pi }{n}\right] .\nonumber \\ \end{aligned}$$
(5)

We choose the principal branch of \(n\)-th root. Since the exponential function is periodic, in the above definitions we restrict the range of variability of \(\alpha \) to the intervals of length \(6\pi /n\). The choice of these intervals depends on the properties of \(f_\alpha \) and \(g_\alpha \). Some additional information will be given in Remark 2.

The definition of the family \({\mathcal {W}}\) may be extended for \(n=3\). In this case, the sets belonging to \({\mathcal {W}}\) may be treated as generalized polygons. The measure of the angles is equal to \(4\pi /3\) and 0 alternately. These sets have the shape of an unbounded three-pointed star, see Fig. 3. Moreover, for \(n=3\), the functions \(g_\alpha \) map \(\Delta \) onto these generalized polygons.

Fig. 3
figure 3

Extremal polygons for \(n=3\)

Lemma 2

All functions \(f_\alpha \), \(\alpha \in \left[ -\frac{3\pi }{n},\frac{3\pi }{n}\right] \), belong to \({\mathcal {F}}^{(n)}\) for \(n=4k+1\), \(k\ge 1\).

Lemma 3

All functions \(g_\alpha \), \(\alpha \in \left[ -\frac{2\pi }{n},\frac{4\pi }{n}\right] \), belong to \({\mathcal {F}}^{(n)}\) for \(n=4k+3\), \(k\ge 0\).

Proof of Lemma 2

At the beginning, we shall show that the functions \(f_\alpha \), \(\alpha \in \left[ -3\pi /n,3\pi /n\right] \) are univalent. Observe that

$$\begin{aligned} {f_\alpha ^{\prime }} (z)=p(z)\cdot \frac{h(z)}{z}\ , \end{aligned}$$
(6)

where

$$\begin{aligned} p(z)=\left( \frac{1-z^n e^{-ia}}{1+z^n e^{-ib}}\right) ^{1/n} \end{aligned}$$

and

$$\begin{aligned} h(z)=\frac{z}{(1+z^n e^{-ib})^{2/n}} \end{aligned}$$

with \(a=n\alpha \), \(b=n\alpha /3\).

A Möbius function \(p_1(z)=(1-z e^{-ia})/(1+z e^{-ib})\ ,a,b\in {\mathbb {R}}\) satisfies the condition \({{\mathrm{Re}}}e^{i\beta } p_1(z)>0\) with some \(\beta \in {\mathbb {R}}\); hence, for all \(n\in {\mathbb {N}}\), the inequality \({{\mathrm{Re}}}e^{i\beta } p(z)={{\mathrm{Re}}}e^{i\beta } p_1(z^n)^{1/n}>0\) holds with the same \(\beta \). A function \(h_1(z)=z/(1+z e^{-ib})^2\), \(b\in {\mathbb {R}}\) is starlike, so is \(h(z)=\root n \of {h_1(z^n)}\). Combining these two facts with (6), we conclude that

$$\begin{aligned} {{\mathrm{Re}}}e^{i\beta } p(z)={{\mathrm{Re}}}e^{i\beta } \frac{{zf_\alpha ^{\prime }}(z)}{h(z)}>0\ , \end{aligned}$$

which means that \(f_\alpha \) is close-to-convex and consequently univalent.

Next, we claim that each polygon \(f_\alpha (\Delta )\) is a set which is convex in the direction of the real axis. It is sufficient to discuss the argument of the tangent line to \(\partial f_\alpha (\Delta )\). Observe that

$$\begin{aligned} \arg \left( \frac{\partial }{\partial \varphi }\left( f_\alpha (e^{i\varphi })\right) \right)&= \arg \left( {f_\alpha ^{\prime }}(e^{i\varphi })ie^{i\varphi }\right) \nonumber \\ \!&= \!\frac{1}{n}\arg \left( 1+e^{i(\pi +n\varphi -n\alpha )}\right) -\frac{3}{n}\arg \left( 1+e^{i(n\varphi -n\alpha /3)}\right) +\frac{\pi }{2}+\varphi .\nonumber \\ \end{aligned}$$
(7)

Let \(\varphi \in [\alpha /3-\pi /n,\alpha )\). Then, \(\pi +n\varphi -n\alpha \) as well as \(n\varphi -n\alpha /3\) are in \([-\pi ,\pi )\) and from (7) we get

$$\begin{aligned} \arg \left( \frac{\partial }{\partial \varphi }\left( f_\alpha (e^{i\varphi })\right) \right) \!&= \! \frac{1}{2n}\left( \pi +n\varphi -n\alpha \right) \!-\!\frac{3}{2n}\left( n\varphi -n\frac{\alpha }{3}\right) +\frac{\pi }{2}+\varphi = \frac{\pi }{2}+\frac{\pi }{2n}.\nonumber \\ \end{aligned}$$
(8)

The above means that the tangent for \(\varphi \) in \((\alpha /3-\pi /n,\alpha )\) has the constant argument \(\pi /2+\pi /2n\). Since \(f_\alpha (\Delta )\) is a polygon with angles measuring \(\pi +\pi /n\) and \(\pi -3\pi /n\) alternately, the argument of the tangent line takes values \(\pi /2+\pi /2n+2j\pi /n\) and \(\pi /2-\pi /2n+2j\pi /n\), \(j=1,2,\ldots ,n\) alternately. What is more, putting \(j=k\) in \(\pi /2+\pi /2n+2j\pi /n\), we obtain the argument equal to \(\pi \) and putting \(j=3k+1\) in \(\pi /2-\pi /2n+2j\pi /n\), we obtain the argument equal to \(2\pi \). Hence two of the sides of \(f_\alpha (\Delta )\) are horizontal; consequently \(f_\alpha \in {\mathcal {F}}^{(n)}\). \(\square \)

The proof of Lemma 3 is similar.

Remark 2

A polygon \(f_\alpha (\Delta )\) has vertices in points \(f_\alpha (e^{i\alpha })\cdot \varepsilon ^j\) and \(f_\alpha (e^{i(\alpha /3+\pi /n)})\cdot \varepsilon ^j\), \(j=0,1,\ldots ,n-1\). These vertices correspond to angles measuring \(\pi +\pi /n\) and \(\pi -3\pi /n\), respectively. It is worth pointing out some particular cases of polygons belonging to \({\mathcal {W}}\). For \(\alpha =-3\pi /2n\) and \(\alpha =3\pi /2n\), they become regular \(n\)-gons and for \(\alpha =-3\pi /n\), \(\alpha =0\), and \(\alpha =3\pi /n\) these sets are \(n\)-pointed stars symmetric with respect to the real axes. The functions \(f_{-3\pi /n}\), \(f_0\), and \(f_{3\pi /n}\) have real coefficients. In all other cases, coefficients are nonreal. These particular functions are as follows:

$$\begin{aligned}&f_{-3\pi /2n}(z) = \int _0^z (1+i\zeta ^n)^{-\frac{2}{n}} d\zeta \quad ,\quad f_{3\pi /2n}(z) = \int _0^z (1-i\zeta ^n)^{-\frac{2}{n}} d\zeta \ ,\\&f_0(z) = \int _0^z (1-\zeta ^n)^{\frac{1}{n}}(1+\zeta ^n)^{-\frac{3}{n}} d\zeta ,\\&f_{-3\pi /n}(z) = f_{3\pi /n}(z) = \int _0^z (1+\zeta ^n)^{\frac{1}{n}}(1-\zeta ^n)^{-\frac{3}{n}} d\zeta . \end{aligned}$$

Likewise, for \(\alpha =-\pi /2n\) and \(\alpha =5\pi /2n\) the sets \(g_\alpha (\Delta )\) are regular \(n\)-gons, and for \(\alpha =-2\pi /n\), \(\alpha =\pi /n\) and \(\alpha =4\pi /n\) these sets are \(n\)-pointed stars. These functions \(g_\alpha \) which map \(\Delta \) on \(n\)-pointed stars have real coefficients.

Let \(n=4k+1, k\ge 1\) be fixed. Let us denote by \(W_\alpha \), a set \(f_\alpha (\Delta )\) for a fixed \(\alpha \in \left[ -3\pi /n,3\pi /n\right] \).

Observe that the following equalities hold:

$$\begin{aligned} f_{-\alpha }^{\prime }(z) = \overline{f_{\alpha }^{\prime }(\overline{z})}\quad \text {for}\; \alpha \in \left[ -\frac{3\pi }{n},\frac{3\pi }{n}\right] \end{aligned}$$

and

$$\begin{aligned} {f_{\frac{3\pi }{2n}-\gamma }^{\prime }}(-z) = \overline{{f_{\frac{3\pi }{2n}+\gamma }^{\prime }}(\overline{z})}\quad \text {for}\; \gamma \in \left[ 0,\frac{3\pi }{2n}\right] \ , \end{aligned}$$

which means that

$$\begin{aligned} f_{-\alpha }(z) = \overline{f_{\alpha }(\overline{z})}\quad \text {for}\quad \alpha \in \left[ -\frac{3\pi }{n},\frac{3\pi }{n}\right] \end{aligned}$$

and

$$\begin{aligned} -f_{\frac{3\pi }{2n}-\gamma }(-z) = \overline{f_{\frac{3\pi }{2n}+\gamma }(\overline{z})}\quad \text {for}\quad \gamma \in \left[ 0,\frac{3\pi }{2n}\right] . \end{aligned}$$

Consequently,

$$\begin{aligned} W_{-\alpha }=\overline{W_{\alpha }}\quad \text {and}\quad -W_{\frac{3\pi }{2n}-\gamma }=\overline{W_{\frac{3\pi }{2n}+\gamma }}\ . \end{aligned}$$

For this reason, a polygon \(W\) as well as \(\overline{W}\) and \(-\overline{W}\) belong to the family \({\mathcal {W}}\). From the geometric construction of polygons in \({\mathcal {W}}\), it follows that the ratio of lengths of any two adjacent sides of a polygon varies from 0 to infinity as \(\alpha \) is changing in \([-3\pi /n,3\pi /n]\); in each case a polygon is convex in the direction of the real axis. Multiplying sets \(W_{\alpha }\), \(\alpha \in \left[ -3\pi /n,3\pi /n\right] \) by \(\lambda >0\) we obtain all members of the set \({\mathcal {W}}\).

We have proved one part of the following lemma (the second one can be proved analogously)

Lemma 4

  1. 1.

    \({\mathcal {W}}=\left\{ \lambda \cdot f_\alpha (\Delta ), \lambda >0, \alpha \in \left[ -3\pi /n,3\pi /n\right] \right\} \quad \text {for}\quad n=4k+1, k\ge 1\) ,

  2. 2.

    \({\mathcal {W}}=\left\{ \lambda \cdot g_\alpha (\Delta ), \lambda >0, \alpha \in \left[ -2\pi /n,4\pi /n\right] \right\} \quad \text {for}\quad n=4k+3, k\ge 0\).

3 Koebe Sets for \({\mathcal {F}}^{(n)}\)

Let us define

$$\begin{aligned} F(\alpha )\equiv f_\alpha (e^{i\alpha }),\quad \alpha \in \left[ -\frac{3\pi }{n},\frac{3\pi }{n}\right] \end{aligned}$$

and

$$\begin{aligned} G(\alpha )\equiv g_\alpha (e^{i\alpha }),\quad \alpha \in \left[ -\frac{2\pi }{n},\frac{4\pi }{n}\right] \ . \end{aligned}$$

From (4) and (5)

$$\begin{aligned} F(\alpha )=e^{i\alpha }\int _0^1 (1-t^n)^{\frac{1}{n}} (1+t^n e^{2in\alpha /3})^{-\frac{3}{n}}dt \end{aligned}$$

and

$$\begin{aligned} G(\alpha )=e^{i\alpha }\int _0^1 (1-t^n)^{\frac{1}{n}} (1-t^n e^{i(2n\alpha /3-5\pi /3)})^{-\frac{3}{n}}dt. \end{aligned}$$

It can be easily checked that

$$\begin{aligned} \arg F(\alpha )&= \alpha \quad \text {for}\; \alpha \in \left\{ -\frac{3\pi }{n},-\frac{3\pi }{2n},0,\frac{3\pi }{2n},\frac{3\pi }{n}\right\} \ , \end{aligned}$$
(9)
$$\begin{aligned} \arg G(\alpha )&= \alpha \quad \text {for}\; \alpha \in \left\{ -\frac{2\pi }{n},-\frac{\pi }{2n},\frac{\pi }{n},\frac{5\pi }{2n},\frac{4\pi }{n}\right\} \ . \end{aligned}$$
(10)

Theorem 4

The Koebe set \(K_{{\mathcal {F}}^{(n)}}\), for a fixed \(n=4k+1\), \(k\in {\mathbb {N}}\), is a bounded and \(2n\)-fold symmetric domain such that

$$\begin{aligned} \partial K_{{\mathcal {F}}^{(n)}}\cap \left\{ w:\arg w\in \left[ -\frac{\pi }{2n},\frac{\pi }{2n}\right] \right\} = F\left( \left[ -\alpha _F,\alpha _F\right] \right) \ , \end{aligned}$$
(11)

where \(\alpha _F\) is the only solution of the equation

$$\begin{aligned} \arg F\left( \alpha \right) =\frac{\pi }{2n} \end{aligned}$$
(12)

in \(\left[ 0,3\pi /2n\right] \).

Proof

Let \(K\) denote the Koebe set for \({\mathcal {F}}^{(n)}\).

Let us consider a polygon \(V_{\alpha }=f_\alpha (\Delta )\) belonging to \({\mathcal {W}}\), such that one of its vertices, let say \(v_*\), lies in \(\Lambda ^*\) (its argument is in \([\pi /2-\pi /n,\pi /2]\)) and the interior angle at \(v_*\) has the measure \(\pi (1+1/n)\). Suppose additionaly that \(w_*\) is a point of the boundary of \(K\), such that \(\arg w_*= \arg v_*\) and \(|w_*| < |v_*|\). We denote the quotient \(w_*/v_*= |w_*|/|v_*|\) by \(\lambda \). Hence \(\lambda <1\).

Since \(w_*\in \partial K\), there exists \(f_*\in {\mathcal {F}}^{(n)}\) such that

$$\begin{aligned} f_*(\Delta )\subset \lambda V_\alpha \subsetneq V_\alpha =f_\alpha (\Delta ). \end{aligned}$$
(13)

Therefore, \(f_*\prec f_\alpha \) and \(1={f_*^{\prime }}(0)\le {f_\alpha ^{\prime }}(0)=1\). Consequently \(f_*= f_\alpha \), which contradicts (13). It means that \(v_*=w_*\), or in other words, \(w_*\) coincides with some vertex of \(f_\alpha (\Delta )\). Hence \(w_*\) is equal to \(f_\alpha (e^{i\alpha })\) rotated about a multiple of \(2\pi /n\), namely about \(2\pi /n\cdot (n-1)/4\). However, it is true only for those \(\alpha \), for which \(w_*=F(\alpha )\cdot \varepsilon ^{(n-1)/4}\) is in \(\Lambda ^*\).

Observe that for \(\alpha \in \left[ -3\pi /n,3\pi /n\right] \), we have

$$\begin{aligned} F\left( -\alpha \right) = \overline{F\left( \alpha \right) }\ , \end{aligned}$$

that is,

$$\begin{aligned} \arg F\left( -\alpha \right) = -\arg F\left( \alpha \right) . \end{aligned}$$

From this and (9), (12) it follows that

$$\begin{aligned} \left\{ F(\alpha )\cdot \varepsilon ^{(n-1)/4}: \alpha \in \left[ -\alpha _F,\alpha _F\right] \right\} \end{aligned}$$

is the boundary of the Koebe set for \({\mathcal {F}}^{(n)}\) in \(\Lambda ^*\). Combining this with Theorem 3, the equality (11) follows.

Finally, we claim that \(\alpha _F\) is the only solution of (12) in \(\left[ 0,3\pi /2n\right] \). On the contrary, assume that there exist two different numbers \(\alpha _1, \alpha _2\in [0,3\pi /2n]\) such that

$$\begin{aligned} \arg F(\alpha _1) = \arg F(\alpha _2)\ , \end{aligned}$$

or equivalently,

$$\begin{aligned} \arg f_{\alpha _1}(e^{i\alpha _1}) = \arg f_{\alpha _2}(e^{i\alpha _2}). \end{aligned}$$

The sets \(W_{\alpha _1}=f_{\alpha _1}(\Delta )\), \(W_{\alpha _2}=f_{\alpha _2}(\Delta )\) are polygons of the family \({\mathcal {W}}\). This and the definition of \({\mathcal {W}}\) (or Lemma 4) result in

$$\begin{aligned} W_{\alpha _1}\subset W_{\alpha _2}\quad \text {or}\quad W_{\alpha _2}\subset W_{\alpha _1}. \end{aligned}$$

The normalization of \(f_{\alpha _1}\) and \(f_{\alpha _2}\) leads to \(W_{\alpha _1}= W_{\alpha _2}\). Hence \(\alpha _1=\alpha _2\), a contradiction. This means that (12) has only one solution in the set \([0,3\pi /2n]\). \(\square \)

The above proof gives more. Namely, \(F\) is starlike for \(\alpha \in [-3\pi /2n,3\pi /2n]\). Moreover,

$$\begin{aligned} \arg F\left( \alpha +\frac{3\pi }{n}\right) = \arg F(\alpha )+\frac{3\pi }{n}. \end{aligned}$$

It implies that \(F\) is starlike for \(\alpha \in [-\pi ,\pi ]\).

Furthermore, it is not difficult to see that there do not exist two different points in the set \(\partial K_{{\mathcal {F}}^{(n)}}\cap \Lambda ^*\) with the same imaginary part. For contrary suppose that it is not the case, i.e., there exist \(v_1\) and \(v_2\) such that \(v_1\ne v_2\), \(v_1, v_2\in \partial K_{{\mathcal {F}}^{(n)}}\cap \Lambda ^*\) and \({{\mathrm{Im}}}v_1={{\mathrm{Im}}}v_2\).

With use of an argument similar to those in the proof of previous theorem, we can see that there exist two polygons \(V_1, V_2\in {\mathcal {W}}\) with vertices \(v_1\), \(v_2\), respectively. The angles at these vertices have the same measure. Hence the sides of these polygons are pairwise parallel and if \({{\mathrm{Re}}}v_1<{{\mathrm{Re}}}v_2\), then \(V_1\subset V_2\). This means that there exist \(f_1, f_2 \in {\mathcal {F}}^{(n)}\) that \(f_1(\Delta )=V_1\), \(f_2(\Delta )=V_2\), and \(f_1\prec f_2\). But the normalization of \(f_1\) and \(f_2\) is the same, hence \(f_1=f_2\); a contradiction.

Theorem 5

The Koebe set \(K_{{\mathcal {F}}^{(n)}}\), for a fixed \(n=4k+3\), \(k\ge 0\) is a bounded and \(2n\)-fold symmetric domain such that

$$\begin{aligned} \partial K_{{\mathcal {F}}^{(n)}}\cap \left\{ w:\arg w\in \left[ \frac{\pi }{2n},\frac{3\pi }{2n}\right] \right\} = G\left( \left[ \alpha _G,\frac{2\pi }{n}-\alpha _G\right] \right) \ , \end{aligned}$$
(14)

where \(\alpha _G\) is the only solution of the equation

$$\begin{aligned} \arg G\left( \alpha \right) =\frac{\pi }{2n} \end{aligned}$$
(15)

in \(\left[ -\pi /2n,\pi /n\right] \).

Proof

A consideration similar to the above shows that \(\alpha _G\) is the only solution of the Eq. (15) in \(\left[ -\pi /2n,\pi /n\right] \).

Suppose that \(w_*\in \partial K\cap \Lambda ^*\). The analogous argument to this in the proof of Theorem 4 yields that \(K\) is contained in some polygon \(W\) of the family \({\mathcal {W}}\).

Let \(g_*\) be a function from \({\mathcal {F}}^{(n)}\) for which \(w_*\in \partial g_*(\Delta )\). We have \(g_*(\Delta )\subset W=g_\alpha (\Delta )\) for some \(\alpha \in [-\pi /2n,5\pi /n]\). For this reason \(g_*\prec g_\alpha \), but taking into account the normalization of both functions, we obtain \(g_*= g_\alpha \). Hence \(w_*=g_\alpha (e^{i\alpha })\cdot \varepsilon ^{(n-3)/4}\), but only if \(w_*\in \Lambda ^*\).

For \(\alpha \in [-\pi /2n,\pi /n]\),

$$\begin{aligned} G\left( \frac{2\pi }{n}-\alpha \right) = \varepsilon \overline{G\left( \alpha \right) }\ , \end{aligned}$$

and so

$$\begin{aligned} \arg G\left( \frac{2\pi }{n}-\alpha \right) = \frac{2\pi }{n}-\arg G\left( \alpha \right) . \end{aligned}$$

From this, (10) and (15), we conclude that

$$\begin{aligned} \left\{ G(\alpha )\cdot \varepsilon ^{(n-3)/4}: \alpha \in \left[ \alpha _G,\frac{2\pi }{n}-\alpha _G\right] \right\} \end{aligned}$$

is the boundary of the Koebe set for \({\mathcal {F}}^{(n)}\) in \(\Lambda ^*\). Theorem 3 concludes the proof of our theorem. \(\square \)

Now, we can derive the Koebe constant for \({\mathcal {F}}^{(n)}\).

Theorem 6

For a fixed positive odd integer \(n\), \(n\ge 3\) and for every function \(f\in {\mathcal {F}}^{(n)}\) the disk \(\Delta _{r_n}\), where \(r_n=B(1/n,1/2n+1/2)/n\root n \of {4}\), is included in \(f(\Delta )\). The number \(r_n\) cannot be increased.

The symbol \(B\) stands for the Beta and \(\Delta _r\), \(r>0\) means \(\Delta _r =\{\zeta \in {\mathbb {C}}\,:\,\left| \zeta \right| <r\}\).

Proof

According to Theorems 4 and 5, the Koebe constant is equal to

$$\begin{aligned} \min \left\{ |F(\alpha )|: \alpha \in \left[ -\alpha _F,\alpha _F\right] \right\} \quad \text {for} \quad n=4k+1\ , \end{aligned}$$

or

$$\begin{aligned} \min \left\{ |G(\alpha )|: \alpha \in \left[ \alpha _G,\frac{2\pi }{n}-\alpha _G\right] \right\} \quad \text {for} \quad n=4k+3. \end{aligned}$$

But

$$\begin{aligned} |F(\alpha )|^2\ge \left( \int _0^1 (1-t^n)^{\frac{1}{n}} {{\mathrm{Re}}}(1+t^n e^{2in\alpha /3})^{-\frac{3}{n}}dt\right) ^2 \end{aligned}$$

and the integrand in this expression is nonnegative; thus

$$\begin{aligned} |F(\alpha )|\ge \int _0^1 (1-t^n)^{\frac{1}{n}} {{\mathrm{Re}}}q_F(\alpha ,t) dt,\quad q_F(\alpha ,t)=(1+t^n e^{2in\alpha /3})^{-\frac{3}{n}}. \end{aligned}$$

Likewise,

$$\begin{aligned} |G(\alpha )|\ge \int _0^1 (1-t^n)^{\frac{1}{n}} {{\mathrm{Re}}}q_G(\alpha ,t) dt,\quad q_G(\alpha ,t)=(1-t^n e^{i(2n\alpha /3-5\pi /3)})^{-\frac{3}{n}}. \end{aligned}$$

It is easy to check that for \(n\ge 3\) the functions \(p(z)=(1\pm t^n z)^{-3/n}\) are convex in \(\Delta \) and they have real coefficients. This means that

$$\begin{aligned} {{\mathrm{Re}}}(1\pm t^n z)^{-\frac{3}{n}}\ge (1+t^n)^{-\frac{3}{n}}. \end{aligned}$$
(16)

Applying (16) for both \(q_F\) and \(q_G\), we get

$$\begin{aligned} |F(\alpha )|\ge q_0\quad \text {and}\quad |G(\alpha )|\ge q_0\ , \end{aligned}$$

where

$$\begin{aligned} q_0 = \int _0^1 (1-t^n)^{\frac{1}{n}} (1+t^n)^{-\frac{3}{n}} dt. \end{aligned}$$
(17)

This results in

$$\begin{aligned} \min \left\{ |F(\alpha )|: \alpha \in \left[ -\alpha _F,\alpha _F\right] \right\} = \left| F\left( 0\right) \right| \end{aligned}$$

and

$$\begin{aligned} \min \left\{ |G(\alpha )|: \alpha \in \left[ \alpha _G,\frac{2\pi }{n}-\alpha _G\right] \right\} = \left| G\left( \frac{\pi }{n}\right) \right| . \end{aligned}$$

Moreover, substituting \(t^n=\tan ^2(x/2)\) in (17), we get

$$\begin{aligned} q_0=\frac{2}{n\root n \of {4}}\int _0^{\pi /2}(\sin x)^{\frac{2}{n}-1}(\cos x)^\frac{1}{n}\ dx =\frac{1}{n\root n \of {4}}B\left( \frac{1}{n},\frac{1}{2n}+\frac{1}{2}\right) \ . \end{aligned}$$

\(\square \)

Corollary 1

For a fixed positive odd integer \(n\), \(n\ge 3\), the Koebe constant for \({\mathcal {F}}^{(n)}\) is equal to \(r_n=B(1/n,1/2n+1/2)/n\root n \of {4}\).

The Koebe sets and the Koebe disks for \(n=3\) and \(n=5\) are shown in Fig. 4.

Fig. 4
figure 4

Koebe domain (solid line) and Koebe disk (dashed line) for \({\mathcal {F}}^{(3)}\) and \({\mathcal {F}}^{(5)}\)

Remark 3

The results established in Theorem 4 and in Corollary 1 are actually valid also for \(n=1\). They were obtained by Złotkiewicz and Reade in [6].

One can check that for \(n=1\), the function \(F\) takes the form

$$\begin{aligned} F(\alpha )=\,e^{i\alpha }\int _0^1 \frac{1-t}{(1+t e^{2i\alpha /3})^3}dt= \frac{e^{2i\alpha /3}}{4\cos (\alpha /3)}. \end{aligned}$$

The Eq. (12) gives \(\alpha _F=3\pi /4\); thus the boundary of the Koebe set in the upper half-plane can be written as follows

$$\begin{aligned} u=\frac{\cos (2\alpha /3)}{4\cos (\alpha /3)}\ ,\quad v=\frac{1}{2}\sin (\alpha /3)\ ,\quad \alpha \in \left[ -\frac{3\pi }{4},\frac{3\pi }{4}\right] . \end{aligned}$$

This fact, we can rewrite in a different way

$$\begin{aligned} K_{{\mathcal {F}}}=\left\{ w\in {\mathbb {C}}:8|w|\left( |w|+|{{\mathrm{Re}}}w|\right) <1\right\} . \end{aligned}$$

The extremal functions \(f_\alpha \) given by (4) are of the form

$$\begin{aligned} f_\alpha (z)=\frac{z+Bz^2}{(1+ze^{-i\alpha /3})^2}\ ,\ B=i\sin (\alpha /3)e^{-2i\alpha /3} \end{aligned}$$

and

$$\begin{aligned} g_\alpha (z)=-f_\alpha (-z)\ , \end{aligned}$$

where \(\alpha \in \left[ -3\pi /4,3\pi /4\right] \). The image set \(f_\alpha (\Delta )\) for a fixed \(\alpha \in (-3\pi /4,3\pi /4)\) coincides with the plane with a horizontal ray excluded. For \(\alpha =-3\pi /4,3\pi /4\), the sets \(f_\alpha (\Delta )\) are half-planes.

Moreover, \(r_1=B(1,1)/4=1/4\).

4 Covering Domains for \({\mathcal {F}}^{(n)}\)

Theorem 7

The covering set \(L_{{\mathcal {F}}^{(n)}}\) for odd \(n\ge 5\) is a bounded and \(2n\)-fold symmetric domain such that

$$\begin{aligned} \partial L_{{\mathcal {F}}^{(n)}}\cap \left\{ w:\arg w\in \left[ 0,\frac{\pi }{2n}\right] \right\} = H\left( \left[ 0,\frac{\pi }{2n}\right] \right) . \end{aligned}$$
(18)

In the proof of this theorem, we need the following lemma.

Lemma 5

Let \(n\ge 5\) be a fixed odd integer. If \(f\in {\mathcal {F}}^{(n)}\) and \(w\in f(\Delta )\cap \Lambda ^*\), then \(f(\Delta )\) contains a polygon \(W\in {\mathcal {W}}\) such that \(W\) has one of its vertices at \(w\) and the interior angle at \(w\) has the measure \(\pi -3\pi /n\).

Proof

Let \(f\in {\mathcal {F}}^{(n)}\) and \(w\in f(\Delta )\cap \Lambda ^*\), i.e., \(\arg w\in [\pi /2-\pi /n,\pi /2]\). Because of the \(n\)-fold symmetry of \(f\) every point \(w\cdot \varepsilon ^j\), \(j=0,1,\ldots , n-1\) belongs to \(f(\Delta )\).

It can be easily checked that

$$\begin{aligned} \max \left\{ {{\mathrm{Im}}}\left( w\cdot \varepsilon ^j\right) , j=0,1,\ldots , n-1\right\} = {{\mathrm{Im}}}\left( w\right) \end{aligned}$$

and

$$\begin{aligned} \min \left\{ {{\mathrm{Im}}}\left( w\cdot \varepsilon ^j\right) , j=0,1,\ldots , n-1\right\} = {{\mathrm{Im}}}\left( w\cdot \varepsilon ^{2k+1}\right) . \end{aligned}$$

Let \(w_1 = w\cdot \varepsilon \) and \(w_2 = w\cdot \varepsilon ^{2k}\). The point \(w_1\) has the second biggest imaginary part among points \(w,w\cdot \varepsilon ,\ldots , w\cdot \varepsilon ^{n-1}\). Likewise, \(w_2\) has the second lowest imaginary part among those points.

Let, moreover, \(l_1\) and \(l_2\) stand for two horizontal rays emanating from \(w_1\) and \(w_2\): \(l_1=\{w_1+t: t\ge 0\}\), \(l_2=\{w_2+t: t\ge 0\}\), respectively.

From the inequality \({{\mathrm{Im}}}w_2>{{\mathrm{Im}}}w\cdot \varepsilon ^{2k+1}\), we conclude that the point \(w\cdot \varepsilon ^{2k+1}\) lies on the opposite side of the straight line which contains \(l_2\) with respect to the origin. As a consequence, \(w_1\) lies on the other side of the straight line including \(l_2\cdot \varepsilon ^{-2k}\) with respect to the origin. Hence, two rays \(l_1\) and \(l_2\cdot \varepsilon ^{-2k}\) have a common point, let say \(w_0\).

We shall show that \(w_0\) also belongs to \(f(\Delta )\). Suppose, contrary to our claim, that \(w_0\notin f(\Delta )\). The points \(w_0,w_1\) lie on the ray \(l_1\) and \(w_1\in f(\Delta )\). Therefore, taking into account the convexity in the direction of the real axis of \(f\), a ray \(m_1=\{w_0+t: t\ge 0\}\) is disjoint from \(f(\Delta )\).

Since \(w_0, w\) belong to \(l_2\cdot \varepsilon ^{-2k}\), the points \(w_0\cdot \varepsilon ^{2k}, w\cdot \varepsilon ^{2k}\) belong to \(l_2\). Moreover, \(w_0\cdot \varepsilon ^{2k}\notin f(\Delta )\) and \(w_2\in f(\Delta )\). Consequently, \(m_2=\{w_0\cdot \varepsilon ^{2k}+t: t\ge 0\}\) is disjoint from \(f(\Delta )\), and, generally, \(m_2\varepsilon ^j \cap f(\Delta )=\emptyset \), \(j=0,1,\ldots , n-1\).

We have proved that the rays \(m_1\) and \(m_2\varepsilon ^{-2k}\) with the common vertex \(w_0\) are disjoint from \(f(\Delta )\). It means that the reflex sector with the vertex in \(w_0\) and these two rays as the sides has no common points with \(f(\Delta )\). But \(w_1\) lies in this reflex sector; hence \(w_1\notin f(\Delta )\), a contradiction.

From the argument given above all points \(w\varepsilon ^j\), \(w_0\varepsilon ^j\), \(j=0,1,\ldots , n-1\) belong to \(f(\Delta )\). Applying \(n\)-fold symmetry and the convexity of \(f\) in the direction of the real axis, we can see that a polygon \(W\) with succeeding vertices at points \(w, w_0, w\varepsilon , w_0\varepsilon ,\ldots , w\varepsilon ^{n-1}, w_0\varepsilon ^{n-1}\) is contained in \(f(\Delta )\). It is easy to check that this polygon has the interior angles \(\pi -3\pi /n\) and \(\pi +\pi /n\) alternately. For this reason, \(W\) is in \({\mathcal {W}}\). \(\square \)

According to Lemmas 2 and 3, every function in \({\mathcal {F}}^{(n)}\) mapping \(\Delta \) onto a polygon of the family \({\mathcal {W}}\) has the form (4) and (5) with appropriately taken \(\alpha \). These functions may be written in the form

$$\begin{aligned} f_\beta (z)&= \int _0^z (1+\zeta ^n e^{-3in\beta })^{\frac{1}{n}} (1-\zeta ^n e^{-in\beta })^{-\frac{3}{n}}d\zeta ,\quad \beta \in \left[ 0,\frac{2\pi }{n}\right] \ ,\end{aligned}$$
(19)
$$\begin{aligned} g_\beta (z)&= \int _0^z (1+\zeta ^n e^{-3in\beta })^{\frac{1}{n}} (1-\zeta ^n e^{-in\beta })^{-\frac{3}{n}}d\zeta ,\quad \beta \in \left[ \frac{\pi }{n},\frac{3\pi }{n}\right] \ , \end{aligned}$$
(20)

equivalent to (4) and (5).

In fact, the functions defined by (4) and (19) are connected by the relation \(\beta =\alpha /3+\pi /n\) and the functions in (5) and (20) are connected by \(\beta =\alpha /3+5\pi /3n\).

Let us define

$$\begin{aligned} H(\beta )= f_\beta (e^{i\beta })\quad \text {for}\; \beta \in \left[ 0,\frac{2\pi }{n}\right] \ , \end{aligned}$$

and

$$\begin{aligned} H(\beta )= g_\beta (e^{i\beta })\quad \text {for}\; \beta \in \left[ \frac{\pi }{n},\frac{3\pi }{n}\right] . \end{aligned}$$

Hence

$$\begin{aligned} H(\beta )\equiv e^{i\beta }\int _0^1 (1+t^ne^{-2in\beta })^{\frac{1}{n}} (1-t^n)^{-\frac{3}{n}}dt\ ,\quad \beta \in {\mathbb {R}}. \end{aligned}$$

Observe that

$$\begin{aligned} \arg H(\beta ) = \beta \quad \text {for}\quad \beta =\frac{\pi }{2n}\cdot j\ ,\quad j=0,1,\ldots ,4n-1. \end{aligned}$$
(21)

Furthermore,

$$\begin{aligned} H\left( \beta +\frac{\pi }{n}\right) = e^{i\frac{\pi }{n}}H(\beta ). \end{aligned}$$
(22)

Now, we can prove Theorem 7.

Proof of Theorem 7

Let \(L\) denote the covering set for \({\mathcal {F}}^{(n)}\). We additionaly assume that \(n=4k+1, k\ge 1\). The proof for the case \(n=4k+3, k\ge 0\) is almost similar.

Let us consider a polygon \(W_{\beta }=f_\beta (\Delta )\) belonging to \({\mathcal {W}}\), such that one of its vertices, let say \(w^*\), lies in \(\Lambda ^*\) and the interior angle at \(w^*\) has the measure \(\pi (1-3/n)\). Suppose additionaly that \(v^*\) is a point of the boundary of \(L\) such that \(\arg v^*= \arg w^*\) and \(|v^*| > |w^*|\). We denote the quotient \(v^*/w^*= |v^*|/|w^*|\) by \(\mu \). Hence \(\mu >1\).

Since \(v^*\in \partial L\), there exists \(f^*\in {\mathcal {F}}^{(n)}\) such that \(v^*\) is a boundary point of \(f^*(\Delta )\). From Lemma 5

$$\begin{aligned} f^*(\Delta )\supset \mu W_\beta \supsetneq W_\beta =f_\beta (\Delta ). \end{aligned}$$
(23)

Therefore, \(f_\beta \prec f^*\) and \(1={f_\beta ^{\prime }}(0)\le {f^{{*}^{\prime }}}(0)=1\). Consequently \(f_\beta = f^*\), which contradicts (23). It means that \(w^*=v^*\), or in other words, \(v^*\) coincides with some vertex of \(f_\beta (\Delta )\). Hence \(v^*\) is equal to \(f_\beta (e^{i\beta })\) rotated about a multiple of \(2\pi /n\), namely about \(2\pi /n\cdot (n-1)/4\). It is enough to take such \(\beta \) that \(v^*=H(\beta )\cdot \varepsilon ^{(n-1)/4}\) is in \(\Lambda ^*\). From this, we conclude that \(\beta \in [0,\pi /2n]\). \(\square \)

Theorem 8

For a fixed odd integer \(n\ge 5\) and for every function \(f\in {\mathcal {F}}^{(n)}\), the set \(f(\Delta )\) is included in \(\Delta _{R_n}\), where \(R_n=B(1/n,1/2-3/2n)/n\root n \of {4}\). The number \(R_n\) cannot be decreased.

Proof

We have

$$\begin{aligned} \left| H(\beta ) \right| \le \int _0^1 \left| (1+t^ne^{-2in\beta })^{\frac{1}{n}} (1-t^n)^{-\frac{3}{n}} \right| dt \le \int _0^1 \frac{(1+t^n)^{\frac{1}{n}}}{(1-t^n)^{\frac{3}{n}}} dt = \left| H(0) \right| . \end{aligned}$$

It can be shown that \(H(0)=B(1/n,1/2-3/2n)/n\root n \of {4}\). \(\square \)

Corollary 2

For a fixed odd integer \(n\ge 5\), the covering constant for \({\mathcal {F}}^{(n)}\) is equal to \(R_n=B(1/n,1/2-3/2n)/n\root n \of {4}\).

The results presented above are valid for positive odd integers greater than or equal to 5. In the last part of this section, we turn to the case \(n=3\).

As it was said in Section 2 (see also Fig. 3) for \(n=3\) and \(\beta \in [\pi /3,\pi ]\setminus \{\pi /2,5\pi /6\}\), the functions given by (20) map \(\Delta \) onto the polygons with the interior angles \(4\pi /3\) and 0 alternately, and the vertices in points \(a\cdot \varepsilon ^j\), \(\infty \cdot a\cdot \varepsilon ^j\), \(j=0,1,2\) alternately, where \(a=g_\beta (e^{i\beta })=H(\beta )\). Both sides adjacent to every vertex in infinity are parallel. Hence \(g_\beta (\Delta )\) are star-shaped sets with three unbounded strips. The strips have the direction \(\pi /3\), \(\pi \), \(5\pi /3\) if \(\beta \in [\pi /3,\pi /2)\cup (5\pi /6,\pi ]\) and 0, \(2\pi /3\), \(4\pi /3\) if \(\beta \in (\pi /2,5\pi /6)\). The thickness of the strips is changing as \(\beta \) varies in \(\beta \in [\pi /3,\pi ]\setminus \{\pi /2,5\pi /6\}\), but when \(\beta \) tends to \(\pi /2\) or \(5\pi /6\), the thickness of the strips tends to 0.

For \(\beta =\pi /2\) and \(\beta =5\pi /6\), the functions

$$\begin{aligned} g_\frac{\pi }{2}(z) = \int _0^z \frac{1}{(1-i\zeta ^3)^{2/3}}d\zeta \end{aligned}$$

and

$$\begin{aligned} g_\frac{5\pi }{6}(z) = \int _0^z \frac{1}{(1+i\zeta ^3)^{2/3}}d\zeta \end{aligned}$$

map \(\Delta \) onto the equilateral triangles symmetric with respect to the imaginary axis. The first triangle has one of its vertices in the point \(ic\), the second one - in the point \(-ic\), where

$$\begin{aligned} c=\frac{B(\frac{1}{3},\frac{1}{6})}{3\root 3 \of {4}}=1.76\ldots . \end{aligned}$$

Let

$$\begin{aligned} \Omega _0=\left\{ w:{{\mathrm{Re}}}w\ge 0, |{{\mathrm{Im}}}w|< \frac{1}{2} c\right\} . \end{aligned}$$

Theorem 9

The covering domain \(L_{{\mathcal {F}}^{(3)}}\) is an unbounded and \(6\)-fold symmetric domain

$$\begin{aligned} L_{{\mathcal {F}}^{(3)}} = \bigcup _{j=0}^5 e^{j\frac{\pi }{3}i} \cdot \Omega _0. \end{aligned}$$

Proof

Let \(L\) denote the covering set for \({\mathcal {F}}^{(n)}\) and let \(L^*\) stand for \(\bigcup _{j=0}^5 e^{j\pi i/3} \cdot \Omega _0\).

At the beginning, we can see that \(L\) includes six-pointed star obtained as a union of \(g_{\pi /2}(\Delta )\) and \(g_{5\pi /6}(\Delta )\). We know that for \(\beta \in (\pi /2,5\pi /6)\) each set \(g_\beta (\Delta )\) contains a part of a horizontal strip between two rays emanating from \(a/\varepsilon \) and \(a\), where \(a=H(\beta )\). From (21), it follows that the arguments of these points vary continuously from \(-\pi /6\) to \(\pi /6\) for the point \(a/\varepsilon \) and from \(\pi /2\) to \(5\pi /6\) for the point \(a\). This and the symmetry of \(L\) with respect to the imaginary axis result in \(L^*\subset L\).

Now, we shall prove that \(L\subset L^*\). On the contrary, assume that \(w_0\notin L^*\) but \(w_0\in L\). It means that there exists a function \(f_0\in {\mathcal {F}}^{(3)}\) such that \(w_0\in f_0(\Delta )\). Without loss of generality, we can assume that \(\arg w_0\in (0,\pi /6)\) because of Lemma 1 and Remark 1.

From the 3-fold symmetry of \(f_0\), we know that \(w_0\varepsilon , w_0\varepsilon ^2\in f_0(\Delta )\). Moreover,

$$\begin{aligned} {{\mathrm{Im}}}w_0 = |w_0|\sin \varphi _0 < |w_0|\sin \left( \varphi _0+\frac{2\pi }{3}\right) = {{\mathrm{Im}}}(w_0\varepsilon )\ , \end{aligned}$$

because \(\varphi _0=\arg w_0\in (0,\pi /6)\).

Observe that the point \(w_1=\{w_0-t: t\ge 0\}\cap (\varepsilon \cdot \{w_0-t: t\ge 0\})\) also belongs to \(f_0(\Delta )\). If it were not the case, the points \(w_1\varepsilon , w_1\varepsilon ^2\) would not be in \(f_0(\Delta )\) either. But \(w_1, w_1\varepsilon ^2\in \{w_0-t: t\ge 0\}\). Combining \(w_1, w_1\varepsilon ^2\notin f_0(\Delta )\) with \(w_0\in f_0(\Delta )\) yields that the segment connecting \(w_1\) and \(w_1\varepsilon ^2\) has no common points with \(f_0(\Delta )\). From this and the 3-fold symmetry, all three segments connecting \(w_1, w_1\varepsilon , w_1\varepsilon ^2\) and, as a consequence, the equilateral triangle \(T\) with vertices in these points, would be disjoint with \(f_0(\Delta )\), a contradiction. This means that \(w_1, w_1\varepsilon , w_1\varepsilon ^2\in f_0(\Delta )\), which results in

$$\begin{aligned} T\subset f_0(\Delta ). \end{aligned}$$
(24)

But

$$\begin{aligned} g_{\frac{5\pi }{6}}(\Delta )\subset T\quad \text {and}\quad g_{\frac{5\pi }{6}}(\Delta )\ne T. \end{aligned}$$
(25)

From (24) and (25), \(g_{5\pi /6}\) is subordinated to \(f\), but \(g_{5\pi /6}\) and \(f\) have the same normalization, a contradiction. It means that if \(w_0\in L\), then \(w_0\in L^*\), which completes the proof. \(\square \)

Fig. 5
figure 5

Covering domains for \({\mathcal {F}}^{(3)}\) and \({\mathcal {F}}^{(5)}\)

The covering domains for \({\mathcal {F}}^{(3)}\) and \({\mathcal {F}}^{(5)}\) are shown in Fig. 5.