betaFeedback

WorksMathematics

The Koebe radius for univalent cubics

Abstract

Let p(z)=z+a2z2+a3z3p(z)=z+a_2z^2+a_3z^3 be a polynomial univalent in the unit disk ΔΔ. We determine the largest disk centered at the origin that is contained in the image p(Δ)p(Δ) of every such polynomial. Its squared radius is the unique real root of 27q3−54q2+63q−8=0. 27q^3-54q^2+63q-8=0. We also find the extremal polynomial p∗p_*.

Let S\mathcal{S} denote the class of functions that are analytic and univalent (one-to-one) in the unit disk Δ\Delta of the complex plane, normalized by the conditions f(0)=0f(0)=0, f′(0)=1f'(0)=1. Thus

f(z)=z+a2z2+a3z3+⋯(1)f(z)=z+a_2z^2+a_3z^3+\cdots \tag*{(1)}

In his 1907 paper on the uniformization of analytic curves, Koebe proved that there is a number r>0r>0, independent of ff, such that

Δr={z∈C:∣z∣<r}\Delta_r=\{z\in\mathbb{C}:|z|<r\}

is contained in f(Δ)f(\Delta) for every f∈Sf\in\mathcal{S} [9]. This estimate was needed to prevent normalized univalent maps from degenerating in a limiting construction for the general uniformization problem [9]. Koebe conjectured that the exact value is r=1/4r=1/4 and that an extremal function is

k(z)=z(1−z)2=z+2z2+3z3+⋯(2)k(z)=\frac{z}{(1-z)^2}=z+2z^2+3z^3+\cdots \tag*{(2)}

which maps Δ\Delta conformally onto the slit plane C∖(−∞,−1/4]\mathbb{C}\setminus(-\infty,-1/4].

Bieberbach proved this result in 1916 by deriving it from the sharp coefficient estimate ∣a2∣≤2|a_2|\leq2 [2] (see also [8 Chapter 2]). In the same paper, Bieberbach formulated his celebrated conjecture ∣am∣≤m|a_m|\leq m, m=2,3,…m=2,3,\ldots, for every function in S\mathcal{S}, with the Koebe function as an extremal function.

This conjecture shaped much of the subsequent development of univalent function theory in the twentieth century and was finally proved by Louis de Branges. His proof was published in 1985 [5].

To the best of our knowledge, univalent polynomials were first studied systematically by Alexander in his fascinating 1915 paper [1]. He related univalence to the locations of the zeros of a polynomial and its derivative and established several sufficient conditions for univalence.

Much of the subsequent work on normalized univalent polynomials concentrated on the coefficient problem, in the spirit of the Bieberbach conjecture. In particular, in 1969 Suffridge [11] constructed a family of polynomials from SS whose coefficients tend to those of the Koebe function (2).

The general problem of describing the coefficient regions of univalent polynomials appears to be extremely difficult and seems unlikely to admit a tractable description. For polynomials of low degree, however, such a description is possible. In particular, Brannan in 1967 [3], and independently Cowling and Royster in 1968 [4], described all complex pairs (a2,a3)(a_2,a_3) for which

p(z)=z+a2z2+a3z3p(z)=z+a_2z^2+a_3z^3

is univalent; see also the later account of Rahman and Waniurski [10 Introduction].

A normalized rotation of pp is the polynomial pθ(z)=e−iθp(eiθz)p_\theta(z)=e^{-i\theta}p(e^{i\theta}z), where θ\theta is real. We have pθ′(0)=1p_\theta'(0)=1 and pθ(Δ)=e−iθp(Δ)p_\theta(\Delta)=e^{-i\theta}p(\Delta), so the largest origin-centered disk in the image has the same radius. The cubic coefficient becomes e2iθa3e^{2i\theta}a_3. Choosing θ\theta so that this coefficient equals b=∣a3∣b=|a_3|, we may write

p(z)=z+az2+bz3,a=u+iv,(3)p(z)=z+az^2+bz^3,\qquad a=u+iv, \tag*{(3)}

where uu and vv are real. Then Brannan’s criterion [3] says that pp is univalent in Δ\Delta if and only if

0≤b≤13,(4)0\leq b\leq\frac{1}{3}, \tag*{(4)}

and, for every s∈[0,4]s\in[0,4] with s=4s=4 omitted when b=1/3b=1/3,

su2(1+b(s−1))2+sv2(1−b(s−1))2≤1.(5)\frac{su^2}{(1+b(s-1))^2}+\frac{sv^2}{(1-b(s-1))^2}\leq1. \tag*{(5)}
The coefficient body of normalized univalent cubic polynomials

Figure 1. The coefficient body of normalized univalent cubic polynomials.

For each bb satisfying (4), let CbC_b be the region of admissible coefficients a=u+iva=u+iv. It is the intersection of the ellipses in (5) for 0<s≤40<s\leq4, with s=4s=4 omitted when b=1/3b=1/3. At s=0s=0 the inequality holds identically. A direct calculation shows that, for 0≤b≤1/50\leq b\leq1/5, this intersection is itself an ellipse; when b=0b=0, it is a circle. For 1/5<b<1/31/5<b<1/3, no single ellipse determines the intersection: its boundary consists of an arc of the ellipse corresponding to s=4s=4 and the envelope of a nontrivial subfamily of (5), as in Figure 2. When b=1/3b=1/3, letting s↑4s\uparrow4 forces v=0v=0; the remaining inequalities give the segment ∣u∣≤22/3|u|\leq2\sqrt{2}/3.

One half of the section $C_b$ of the coefficient body at $b=b_*\approx 0.2896$

Figure 2. One half of the section CbC_b of the coefficient body at b=b∗≈0.2896b=b_*\approx0.2896.

Brannan’s criterion immediately gives

∣a2∣≤223,∣a3∣≤13,|a_2| \le\frac{2\sqrt{2}}{3}, \qquad|a_3| \le\frac{1}{3},

and the cubic Suffridge polynomial

pS(z)=z+223z2+13z3p_S(z)=z+\frac{2\sqrt{2}}{3}z^2+\frac{1}{3}z^3

maximizes both coefficient moduli [3, 4, 11]. Thus the coefficient problem for cubic polynomials was completely solved.

To the best of our knowledge, Dimitrov first posed a polynomial analogue of Koebe’s one-quarter theorem in 2002 [6]. Namely, for each nn, determine the largest disk centered at the origin that is contained in p(Δ)p(\Delta) for every polynomial p∈Sp\in\mathcal{S} of degree nn. He suggested that the Suffridge polynomials might be extremal. We call the extremizers in Dimitrov’s problem Koebe polynomials.

For a Suffridge polynomial pSp_S, let rSr_S denote the radius of the largest origin-centered disk contained in pS(Δ)p_S(\Delta). For the cubic Suffridge polynomial above, a direct calculation gives

rS=233.r_S=\frac{2}{3\sqrt{3}}.
The image $p_S(\Delta)$ of the cubic Suffridge polynomial and a magnification of the boxed region

Figure 3. The image pS(Δ)p_S(\Delta) of the cubic Suffridge polynomial and a magnification of the boxed region.

In 2019, Dmitrishin, Dyakonov and Stokolos proved [7 Theorem 5.1] that the Koebe radius for cubic polynomials from S\mathcal{S} with real coefficients is

rD=3−52,r_D = \frac{3-\sqrt{5}}{2},

and that a Koebe polynomial in this subclass is

pD(z)=z+25z2+12(1−15)z3.p_D(z) = z + \frac{2}{\sqrt{5}}z^2 + \frac{1}{2}\left(1-\frac{1}{\sqrt{5}}\right)z^3.
The image $p_D(\Delta)$ of the real Koebe polynomial and a magnification of the boxed region

Figure 4. The image pD(Δ)p_D(\Delta) of the real Koebe polynomial and a magnification of the boxed region.

Since rD<rSr_D < r_S, this result shows that the cubic Suffridge polynomial is not a Koebe polynomial even within the subclass of polynomials with real coefficients. The Koebe problem for the full class of univalent cubic polynomials with complex coefficients remained open.

In this paper we determine the Koebe radius r∗r_* for cubic polynomials from S\mathcal{S} and construct a cubic polynomial p∗p_* attaining it.

The image $p_*(\Delta)$ and a magnification of the boxed region

Figure 5. The image p∗(Δ)p_*(\Delta) and a magnification of the boxed region.

The exact value of r∗r_* is given in Theorem 1. In particular, rS>rD>r∗r_S > r_D > r_*, where r∗r_* is the Koebe radius for univalent cubic polynomials. One critical point of p∗p_* lies on ∂Δ\partial\Delta, while the other lies outside Δ\Delta. The image of the boundary critical point is a cusp of p∗(∂Δ)p_*(\partial\Delta) and lies on ∂Δr∗\partial\Delta_{r_*}.

Before stating and proving the theorem, we describe the main ideas that lead to the solution. For each univalent cubic pp, the largest disk centered at the origin and contained in p(Δ)p(\Delta) has radius

r(p)=min⁡∣z∣=1∣p(z)∣.r(p) = \min_{|z|=1} |p(z)|.

For pp in the form (3), we have p(z)/z2=a+z‾+bzp(z)/z^2 = a + \overline{z} + bz on the unit circle. The curve traced by z‾+bz\overline{z} + bz is centrally symmetric, so

r(p)=min⁡∣z∣=1∣a−(z‾+bz)∣.(6)r(p) = \min_{|z|=1} |a - (\overline{z} + bz)|. \tag*{(6)}

Since z‾+bz\overline{z} + bz, as zz ranges over the unit circle, describes the boundary of the ellipse

Eb={ξ+iη∈C:ξ2(1+b)2+η2(1−b)2≤1},(7)E_b = \left\{ \xi+ i\eta\in\mathbb{C} : \frac{\xi^2}{(1+b)^2} + \frac{\eta^2}{(1-b)^2} \le1 \right\}, \tag*{(7)}

we obtain

r(p)=dist⁡(a,∂Eb).(8)r(p) = \operatorname{dist}(a,\partial E_b). \tag*{(8)}

For fixed bb, we therefore seek the point of CbC_b nearest to ∂Eb\partial E_b, and then minimize the resulting distance over bb.

Thus finding the sharp Koebe radius and the extremal polynomial reduces to minimizing the distance between the region CbC_b of admissible second coefficients and the boundary of the ellipse EbE_b determined by the third coefficient bb.

For 0≤b≤1/40 \le b \le1/4, the ellipse KbK_b corresponding to s=3s = 3 in Brannan’s family contains CbC_b. Comparing KbK_b with ∂Eb\partial E_b gives the uniform estimate

r(p)≥748.(9)r(p) \ge\sqrt{\frac{7}{48}}. \tag*{(9)}

This lower bound already exceeds the Koebe radius attained in the second range.

For 1/4≤b<1/31/4 \le b < 1/3, we obtain the exact distance. Both estimates apply at b=1/4b = 1/4. Two ellipses in Brannan’s family, corresponding to s=s0(b)s = s_0(b) and s=4s = 4, meet at a corner aba_b of CbC_b. A positive linear combination of their defining quadratic forms gives an ellipse LbL_b that contains CbC_b and passes through aba_b. The shortest segment from LbL_b to ∂Eb\partial E_b can be taken to start at aba_b, so

dist⁡(Lb,∂Eb)=dist⁡(Cb,∂Eb)=dist⁡(ab,∂Eb).\operatorname{dist}(L_b,\partial E_b)=\operatorname{dist}(C_b,\partial E_b)=\operatorname{dist}(a_b,\partial E_b).

The case b=1/3b = 1/3 follows by a limit. Minimizing the distance over bb gives the extremal parameter b∗b_* and the radius r∗r_*.

Theorem 1. Let

p(z)=z+a2z2+a3z3,a2,a3∈C,p(z)=z+a_2z^2+a_3z^3,\qquad a_2,a_3\in\mathbb{C},

be univalent in Δ\Delta. Then

Δr∗⊂p(Δ),\Delta_{r_*}\subset p(\Delta),

where r∗2r_*^2 is the unique real root of

27q3−54q2+63q−8=0.27q^3-54q^2+63q-8=0.

Namely,

r∗=2+63−93−63+933.(10)r_*=\sqrt{\frac{2+\sqrt[3]{6\sqrt{3}-9}-\sqrt[3]{6\sqrt{3}+9}}{3}}. \tag*{(10)}

The radius is sharp: there exists an extremizer p∗p_* such that Δρ⊄p∗(Δ)\Delta_\rho\not\subset p_*(\Delta) for every ρ>r∗\rho>r_*. Up to normalized rotations and complex conjugation, this extremizer is unique and is given by

p∗(z)=z+(u∗+iv∗)z2+b∗z3,p_*(z)=z+(u_*+iv_*)z^2+b_*z^3,

where

b∗=r∗4−r∗23+19,b_*=\sqrt{r_*^4-\frac{r_*^2}{3}+\frac{1}{9}},
u∗=(1+3b∗)(1+6b∗−3b∗2)3/212b∗(1+3b∗2),u_*=\frac{(1+3b_*)(1+6b_*-3b_*^2)^{3/2}}{12\sqrt{b_*}(1+3b_*^2)},
v∗=(1−3b∗)(3b∗2+6b∗−1)3/212b∗(1+3b∗2).v_*=\frac{(1-3b_*)(3b_*^2+6b_*-1)^{3/2}}{12\sqrt{b_*}(1+3b_*^2)}.

Proof. By (8), it suffices to estimate the distance from CbC_b to ∂Eb\partial E_b for each bb. First suppose that 0≤b≤1/40 \le b \le1/4. The inequality at s=3s = 3 gives Cb⊆KbC_b \subseteq K_b, where

Kb={u+iv∈C:3u2(1+2b)2+3v2(1−2b)2≤1},K_b=\left\{u+iv\in\mathbb{C}:\frac{3u^2}{(1+2b)^2}+\frac{3v^2}{(1-2b)^2}\le1\right\},

as illustrated in Figure 6.

The coefficient region $C_b$ and its containing ellipse $K_b$ for $b = 10/41$, together with a magnification of the boxed region

Figure 6. The coefficient region CbC_b and its containing ellipse KbK_b for b=10/41b = 10/41, together with a magnification of the boxed region.

We now have to estimate the distance from KbK_b to ∂Eb\partial E_b. The ellipses ∂Kb\partial K_b and ∂Eb\partial E_b are concentric and have the same axes, but they are not confocal when b>0b > 0. Consequently, a shortest segment joining them need not lie on the uu-axis. In fact, it lies on this axis precisely when 0≤b≤(3−1)/40 \le b \le(\sqrt{3}-1)/4 (see Figure 7).

Three nested ellipses for $b=10/41$ and $\lambda=(1-2b^2)/2$

Figure 7. The three nested ellipses for b=10/41b=10/41 and λ=(1−2b2)/2\lambda=(1-2b^2)/2. The middle ellipse is Qλ=1Q_\lambda=1, while KbK_b is the level region Qλ(a)≤2/3Q_\lambda(a)\le2/3. The double arrow from a∈∂Kba\in\partial K_b has length δb=dist⁡(Kb,∂Eb)\delta_b=\operatorname{dist}(K_b,\partial E_b).

This distance can be studied in two ways. The first is to construct parallel supporting lines to the two ellipses and minimize the distance between them over the direction of their common normal. We take the second route and introduce an intermediate family of auxiliary ellipses Qλ(a)=cQ_\lambda(a)=c. This method gives the required uniform estimate directly and will also be useful in the second case, where 1/4≤b<1/31/4 \le b < 1/3.

Denote the semiaxes of EbE_b by

α=1+b,β=1−b.\alpha= 1+b,\qquad\beta= 1-b.

Then every point of ∂Eb\partial E_b can be written as

w=αx+iβy,x2+y2=1.w=\alpha x+i\beta y,\qquad x^2+y^2=1.

Fix a point a=u+iv∈Eba=u+iv\in E_b. We now use the completion-of-squares argument underlying the Lagrange multiplier equations for this constrained nearest-point problem. For every λ\lambda different from both α2\alpha^2 and β2\beta^2, and every w∈∂Ebw\in\partial E_b, the identity

∣w−a∣2=λ(1−Qλ(a))+(α2−λ)(x−αuα2−λ)2+(β2−λ)(y−βvβ2−λ)2(11)\lvert w-a\rvert^2=\lambda(1-Q_\lambda(a))+(\alpha^2-\lambda)\left(x-\frac{\alpha u}{\alpha^2-\lambda}\right)^2+(\beta^2-\lambda)\left(y-\frac{\beta v}{\beta^2-\lambda}\right)^2 \tag*{(11)}

holds, where

Qλ(a)=u2α2−λ+v2β2−λ.(12)Q_\lambda(a)=\frac{u^2}{\alpha^2-\lambda}+\frac{v^2}{\beta^2-\lambda}. \tag*{(12)}

The two squares in (11) vanish simultaneously precisely when

(α2−λ)x=αu,(β2−λ)y=βv.(13)(\alpha^2-\lambda)x=\alpha u,\qquad(\beta^2-\lambda)y=\beta v. \tag*{(13)}

In this case, λ\lambda is the corresponding Lagrange multiplier. If 0<λ<β20<\lambda<\beta^2, then both coefficients of the squares in (11) are positive. Hence

∣w−a∣2≥λ(1−Qλ(a)).(14)\lvert w-a\rvert^2\ge\lambda(1-Q_\lambda(a)). \tag*{(14)}

Equality holds precisely when (13) holds. Thus, for every a∈Eba \in E_b and every λ\lambda satisfying 0<λ<β20 < \lambda< \beta^2,

dist⁡(a,∂Eb)2≥λ(1−Qλ(a)).(15)\operatorname{dist}(a,\partial E_b)^2 \ge\lambda(1-Q_\lambda(a)). \tag*{(15)}

The identity (11), together with the equality condition (13), will be used not only to estimate this distance from below but also later to identify the critical point.

Our task is now to obtain a uniform lower bound for the distance in (15) as aa ranges over the ellipse KbK_b. For this purpose, we choose λ\lambda and cc so that

Kb={a∈C:Qλ(a)≤c}.K_b=\{a\in\mathbb{C}:Q_\lambda(a)\le c\}.

In other words, a level ellipse of QλQ_\lambda must coincide with KbK_b. Matching the semiaxes requires

c(α2−λ)=(1+2b)23,c(β2−λ)=(1−2b)23.c(\alpha^2-\lambda)=\frac{(1+2b)^2}{3},\qquad c(\beta^2-\lambda)=\frac{(1-2b)^2}{3}.

For b>0b>0, these equations give

c=23,λ=1−2b22,c=\frac{2}{3},\qquad\lambda=\frac{1-2b^2}{2},

and the same choice works when b=0b=0. Thus, for the chosen λ\lambda, every a∈Kba\in K_b satisfies Qλ(a)≤2/3Q_\lambda(a)\le2/3. Hence, by (15),

dist⁡(a,∂Eb)2≥λ(1−23)=1−2b26≥748,0≤b≤14.(16)\operatorname{dist}(a,\partial E_b)^2\ge\lambda\left(1-\frac{2}{3}\right)=\frac{1-2b^2}{6}\ge\frac{7}{48},\qquad0\le b\le\frac{1}{4}. \tag*{(16)}

Now let 1/4≤b<1/31/4\le b<1/3. We seek a sharp bound for this fixed bb. The point where the arc of the ellipse s=4s=4 meets the envelope of Brannan’s family is a natural candidate for the minimum; see Figure 8.

The coefficient region $C_{b_*}$, the two members of Brannan’s family corresponding to $s=s_0$ and $s=4$, and the auxiliary ellipse $L_{b_*}$

Figure 8. The coefficient region Cb∗C_{b_*}, the two members of Brannan’s family corresponding to s=s0s=s_0 and s=4s=4, and the auxiliary ellipse Lb∗L_{b_*} constructed below. Their boundaries meet at the corner point aa.

Our next task is to find the two ellipses in Brannan’s family whose boundaries meet at the corner point aa. One ellipse is already known: it corresponds to s=4s=4. The second is unknown; denote its parameter by s0s_0. This second ellipse must not merely pass through aa, but must generate the envelope of the family at that point. Writing a=u+iva=u+iv, put

Φb(s;u,v)=su2(1+b(s−1))2+sv2(1−b(s−1))2.\Phi_b(s;u,v)=\frac{su^2}{(1+b(s-1))^2}+\frac{sv^2}{(1-b(s-1))^2}.

Thus Φb(s;u,v)=1\Phi_b(s;u,v)=1 is the boundary of the ellipse with parameter ss. We must therefore find uu, vv, s0s_0 satisfying the three conditions

Φb(4;u,v)=1,Φb(s0;u,v)=1,∂Φb∂s(s0;u,v)=0.\Phi_b(4;u,v)=1,\qquad\Phi_b(s_0;u,v)=1,\qquad\frac{\partial\Phi_b}{\partial s}(s_0;u,v)=0.

We first solve the two envelope equations

Φb(s;u,v)=1,∂Φb∂s(s;u,v)=0.\Phi_b(s;u,v)=1,\qquad\frac{\partial\Phi_b}{\partial s}(s;u,v)=0.

Since

∂∂ss(1±b(s−1))2=1∓b(1+s)(1±b(s−1))3,\frac{\partial}{\partial s}\frac{s}{(1\pm b(s-1))^2}=\frac{1\mp b(1+s)}{(1\pm b(s-1))^3},

solving for u2u^2 and v2v^2 gives

u2(s)=(1+b(1+s))(1+b(s−1))34bs2,u^2(s)=\frac{(1+b(1+s))(1+b(s-1))^3}{4bs^2},
v2(s)=(b(1+s)−1)(1−b(s−1))34bs2.v^2(s)=\frac{(b(1+s)-1)(1-b(s-1))^3}{4bs^2}.

We now impose the remaining condition Φb(4;u(s),v(s))=1\Phi_b(4;u(s),v(s))=1. Direct substitution gives

Φb(4;u(s),v(s))−1=−(s−4)2(6b2s−1−3b2)(2b2s−1+b2)s2(1−9b2)2.\Phi_b(4;u(s),v(s))-1=-\frac{(s-4)^2(6b^2s-1-3b^2)(2b^2s-1+b^2)}{s^2(1-9b^2)^2}.

The root s=4s=4 gives the ellipse already known, whereas

1−b22b2>4\frac{1-b^2}{2b^2}>4

when 1/4≤b<1/31/4 \le b < 1/3. Hence the only parameter of a distinct ellipse in Brannan’s family is

s0=1+3b26b2.(17)s_0=\frac{1+3b^2}{6b^2}. \tag*{(17)}

Here 2<s0≤19/6<42<s_0\le19/6<4. Setting s=s0s=s_0 in the envelope formulas gives the first-quadrant intersection of the two ellipses:

ub=(1+3b)(1+6b−3b2)3/212b(1+3b2),vb=(1−3b)(3b2+6b−1)3/212b(1+3b2),ab=ub+ivb,pb(z)=z+abz2+bz3.(18)\begin{aligned} u_b&=\frac{(1+3b)(1+6b-3b^2)^{3/2}}{12\sqrt{b}(1+3b^2)},\\ v_b&=\frac{(1-3b)(3b^2+6b-1)^{3/2}}{12\sqrt{b}(1+3b^2)},\\ a_b&=u_b+iv_b,\qquad p_b(z)=z+a_bz^2+bz^3. \tag*{(18)} \end{aligned}

Both radicals have positive arguments on the interval under consideration.

To check that aba_b satisfies every inequality in (5), put η=b2\eta=b^2, d=b(s−1)d=b(s-1) and

P(s)=(1−d2)2(1−Φb(s;ub,vb)).P(s)=(1-d^2)^2(1-\Phi_b(s;u_b,v_b)).

This is a quartic with leading coefficient η2\eta^2. By construction it vanishes at 44 and has a double zero at s0s_0. Since P(0)=(1−η)2P(0)=(1-\eta)^2,

P(s)=η2(s−4)(s−s0)2(s−s1),s1=(1−η)24η2s02=9(1−η)2(1+3η)2>4,P(s)=\eta^2(s-4)(s-s_0)^2(s-s_1),\qquad s_1=\frac{(1-\eta)^2}{4\eta^2s_0^2}=\frac{9(1-\eta)^2}{(1+3\eta)^2}>4,

where the last inequality follows from η<1/9\eta<1/9. Thus P(s)≥0P(s)\ge0 on [0,4][0,4]. Since ∣d∣<1|d|<1 there, ab∈Cba_b\in C_b and pbp_b is univalent.

We now determine the distance from aba_b to ∂Eb\partial E_b. We choose 0<λ<β20<\lambda<\beta^2 and wb∈∂Ebw_b\in\partial E_b so that both squares in (11) vanish. This gives equality in (15) at aba_b. We then show that Qλ(a)≤Qλ(ab)Q_\lambda(a)\le Q_\lambda(a_b) for every a∈Cba\in C_b, so aba_b minimizes the distance over CbC_b.

The equality Φb(4;ub,vb)=1\Phi_b(4;u_b,v_b)=1 shows that

xb=2ub1+3b,yb=2vb1−3bx_b=\frac{2u_b}{1+3b},\qquad y_b=\frac{2v_b}{1-3b}

satisfy xb2+yb2=1x_b^2+y_b^2=1. Thus wb=αxb+iβybw_b=\alpha x_b+i\beta y_b lies on ∂Eb\partial E_b. We choose λ\lambda so that the Lagrange equations (13) hold for aba_b and wbw_b. The first equation gives

λ=α2−αubxb=1−b22,\lambda=\alpha^2-\frac{\alpha u_b}{x_b}=\frac{1-b^2}{2},

and the second gives the same value. Indeed,

α2−λ=(1+b)(1+3b)2,β2−λ=(1−b)(1−3b)2.\alpha^2-\lambda=\frac{(1+b)(1+3b)}{2},\qquad \beta^2-\lambda=\frac{(1-b)(1-3b)}{2}.

Both quantities are positive when 1/4≤b<1/31/4\le b<1/3. Hence the two squares in (11) vanish at ab,wba_b,w_b.

It remains to prove that QλQ_\lambda is largest on CbC_b at aba_b. For 0<s≤40 < s \le4, denote the left-hand side of Brannan’s inequality by

Bs(a)=A+(s)u2+A−(s)v2,A±(s)=s(1±b(s−1))2.\mathcal{B}_s(a)=A_+(s)u^2+A_-(s)v^2,\qquad A_\pm(s)=\frac{s}{\left(1\pm b(s-1)\right)^2}.

Every point of CbC_b satisfies Bs(a)≤1\mathcal{B}_s(a)\le1, while Bs0(ab)=B4(ab)=1\mathcal{B}_{s_0}(a_b)=\mathcal{B}_4(a_b)=1. We shall express QλQ_\lambda as a positive linear combination of Bs0\mathcal{B}_{s_0} and B4\mathcal{B}_4. Such a representation immediately gives the desired maximum at aba_b.

Write R(s)=A+(s)/A−(s)R(s)=A_+(s)/A_-(s). Both coefficients of each form Bs\mathcal{B}_s are positive, so the required combination exists precisely when

R(4)<k<R(s0),k=β2−λα2−λ=(1−b)(1−3b)(1+b)(1+3b).R(4)<k<R(s_0),\qquad k=\frac{\beta^2-\lambda}{\alpha^2-\lambda}=\frac{(1-b)(1-3b)}{(1+b)(1+3b)}.

Substituting s0s_0 from (17) gives

k−R(4)=4b(1−3b)(1+b)(1+3b)2.k-R(4)=\frac{4b(1-3b)}{(1+b)(1+3b)^2}.
R(s0)−k=16b(18b4+15b2−1)(1+b)(1+3b)(1+6b−3b2)2.R(s_0)-k=\frac{16b(18b^4+15b^2-1)}{(1+b)(1+3b)(1+6b-3b^2)^2}.

The first difference is positive for 0<b<1/30<b<1/3. The numerator of the second equals 1/321/32 at b=1/4b=1/4 and increases for b≥1/4b\ge1/4. Thus for 1/4≤b<1/31/4\le b<1/3 we can choose positive μb,νb\mu_b,\nu_b such that

Qλ=μbBs0+νbB4.Q_\lambda=\mu_b\mathcal{B}_{s_0}+\nu_b\mathcal{B}_4.

The lower threshold for both weights to be positive is b=(33−5)/12=0.249092…b=\sqrt{(\sqrt{33}-5)/12}=0.249092\ldots, just below 1/41/4. At this value R(s0)=kR(s_0)=k and the weight of B4\mathcal{B}_4 vanishes.

Set γb=Qλ(ab)\gamma_b=Q_\lambda(a_b). Since both Brannan inequalities are equalities at aba_b, we have γb=μb+νb\gamma_b=\mu_b+\nu_b. Hence

Qλ(a)≤γb,a∈Cb,Q_\lambda(a)\le\gamma_b,\qquad a\in C_b,

and equality forces both Brannan inequalities to be equalities. Thus the ellipse

Lb={a:Qλ(a)≤γb}L_b=\{a:Q_\lambda(a)\le\gamma_b\}

contains CbC_b and touches it at aba_b, as shown in Figure 8. Combining this inclusion with (15) gives

dist⁡(a,∂Eb)2≥λ(1−γb),a∈Cb.\operatorname{dist}(a,\partial E_b)^2\ge\lambda(1-\gamma_b),\qquad a\in C_b.

At aba_b, the two squares in (11) vanish at wbw_b, so equality holds. Hence pbp_b is extremal for this fixed value of bb.

It remains to compute the distance. Since xb2+yb2=1x_b^2+y_b^2=1,

γb=1+3b2(1+b)xb2+1−3b2(1−b)yb2,λ(1−γb)=(1+b)24−bxb2.\gamma_b=\frac{1+3b}{2(1+b)}x_b^2+\frac{1-3b}{2(1-b)}y_b^2,\qquad \lambda(1-\gamma_b)=\frac{(1+b)^2}{4}-bx_b^2.

Recalling that η=b2\eta=b^2 and substituting ubu_b from (18), we obtain the exact squared radius for this fixed bb:

r(pb)2=D(η)=27η3+108η2−9η+29(1+3η)2.(19)r(p_b)^2=D(\eta)=\frac{27\eta^3+108\eta^2-9\eta+2}{9(1+3\eta)^2}. \tag*{(19)}

At b=1/3b = 1/3, apply the bound to p(hz)/hp(hz)/h, whose third coefficient is h2/3h^2/3. Uniform convergence on the unit circle as h↑1h \uparrow1 gives r(p)2≥D(1/9)=4/27r(p)^2 \ge D(1/9) = 4/27, with equality for pSp_S.

It remains to minimize D(η)D(\eta) for 1/16≤η≤1/91/16 \le\eta\le1/9. Differentiation gives

D′(η)=27η3+27η2+81η−73(1+3η)3.D'(\eta) = \frac{27\eta^3 + 27\eta^2 + 81\eta- 7}{3(1+3\eta)^3}.

The polynomial in the numerator is strictly increasing for η≥0\eta\ge0 and has opposite signs at 1/161/16 and 1/91/9. Hence it has a unique zero η∗\eta_* in this interval, determined by

27η∗3+27η∗2+81η∗−7=0.(20)27\eta_*^3 + 27\eta_*^2 + 81\eta_* - 7 = 0. \tag*{(20)}

The sign of D′(η)D'(\eta) shows that DD decreases before η∗\eta_* and increases after it. Thus D(η∗)D(\eta_*) is the unique minimum.

Put q∗=D(η∗)q_* = D(\eta_*). Reducing this expression by (20) gives

q∗=1+15η∗15+9η∗,η∗=15q∗−115−9q∗.q_* = \frac{1+15\eta_*}{15+9\eta_*}, \qquad\eta_* = \frac{15q_* - 1}{15 - 9q_*}.

Substitution in (20), followed by clearing the denominator, gives

G(q∗)=0,G(q)=27q3−54q2+63q−8.G(q_*) = 0, \qquad G(q) = 27q^3 - 54q^2 + 63q - 8.

Moreover,

G′(q)=81(q−2/3)2+27>0,G'(q) = 81(q - 2/3)^2 + 27 > 0,

so q∗q_* is the unique real zero of GG. Consequently,

r∗=q∗=D(η∗)=0.378591…,b∗=η∗=0.2896….r_* = \sqrt{q_*} = \sqrt{D(\eta_*)} = 0.378591\ldots,\qquad b_* = \sqrt{\eta_*} = 0.2896\ldots.

Thus b∗>1/4b_* > 1/4. For b≤1/4b \le1/4, the preceding lower bound 7/48\sqrt{7/48} is strictly larger than r∗r_*, since

G(7/48)=5034096>0.G(7/48) = \frac{503}{4096} > 0.

The inequality is also strict at b=1/3b = 1/3, since 1/9≠η∗1/9 \ne\eta_*. Writing q=r∗2=q∗q = r_*^2 = q_*, the identity

(15−9q)(q2−q3+19)−(15q−1)=−13G(q)(15 - 9q)\left(q^2 - \frac{q}{3} + \frac{1}{9}\right) - (15q - 1) = -\frac{1}{3}G(q)

then gives

b∗=r∗4−r∗23+19.b_* = \sqrt{r_*^4 - \frac{r_*^2}{3} + \frac{1}{9}}.

Putting t=3r∗2−2t = 3r_*^2 - 2 reduces G(r∗2)=0G(r_*^2) = 0 to

t3+9t+18=0,t^3 + 9t + 18 = 0,

which yields (10).

Finally, if r(p)=r∗r(p) = r_*, the strict bounds at b≤1/4b \le1/4 and b=1/3b = 1/3 and the unique minimum of DD force b=b∗b = b_*. At a nearest point of ∂Eb\partial E_b, both squares in (11) must vanish and Qλ(a)=γbQ_\lambda(a) = \gamma_b. Hence

Φb(s0;u,v)=Φb(4;u,v)=1.\Phi_b(s_0;u,v) = \Phi_b(4;u,v) = 1.

These two equations are linear in u2u^2, v2v^2. Their coefficient ratios R(s0)R(s_0) and R(4)R(4) are distinct, so their only solution is u2=ub2u^2=u_b^2, v2=vb2v^2=v_b^2. The normalized rotation p(z)↦−p(−z)p(z)\mapsto-p(-z) changes aa to −a-a; conjugation changes aa to a‾\overline{a}. Together they give all four choices of signs.

To locate the cusp of p∗(∂Δ)p_*(\partial\Delta), take b=b∗b=b_* and put ζ=−xb+iyb\zeta=-x_b+iy_b. Then ∣ζ∣=1|\zeta|=1, and the formulas for xb,ybx_b,y_b give 2ab+ζ‾+3bζ=02a_b+\overline{\zeta}+3b\zeta=0. It follows that p∗′(ζ)=0p_*'(\zeta)=0. The product of the two zeros of p∗′p_*' is 1/(3b)1/(3b), so the other critical point has modulus 1/(3b)>11/(3b)>1. Moreover, p∗(ζ)/ζ2=ab−wbp_*(\zeta)/\zeta^2=a_b-w_b, hence ∣p∗(ζ)∣=∣ab−wb∣=r∗|p_*(\zeta)|=|a_b-w_b|=r_*. Write γ(t)=p∗(ζeit)\gamma(t)=p_*(\zeta e^{it}) and h=3bζ2h=3b\zeta^2. Then γ′(0)=0\gamma'(0)=0, γ′′(0)=ζ(1−h)\gamma''(0)=\zeta(1-h), and γ′′′(0)=iζ(3−5h)\gamma'''(0)=i\zeta(3-5h). Since ℜh=3b(2xb2−1)\Re h=3b(2x_b^2-1), (18) gives

ℑ(γ′′(0)‾γ′′′(0))=3−8ℜh+5∣h∣2=(21b2−1)2(3b2+5)3(1+3b2)2>0.(21)\Im\left(\overline{\gamma''(0)}\gamma'''(0)\right)=3-8\Re h+5|h|^2=\frac{(21b^2-1)^2(3b^2+5)}{3(1+3b^2)^2}>0. \tag*{(21)}

Thus the quadratic and cubic terms in the Taylor expansion point in linearly independent directions, and γ\gamma has an ordinary cusp at p∗(ζ)p_*(\zeta).

Acknowledgement

The author is grateful to Dmitry Dmitrishin for drawing his attention to this problem. The author has also freely used results of mathematicians published in open sources, as well as various large language models.

Odesa Polytechnic National University, Odesa, Ukraine

Email address: [email protected]

References

  1. [1]J. W. Alexander, Functions which map the interior of the unit circle upon simple regions, Ann. of Math. (2) 17 (1915), no. 1, 12–22.DOI
  2. [2]L. Bieberbach, Über die Koeffizienten derjenigen Potenzreihen, welche eine schlichte Abbildung des Einheitskreises vermitteln, Sitzungsber. Preuss. Akad. Wiss. Berlin (1916), 940–955.
  3. [3]D. A. Brannan, Coefficient regions for univalent polynomials of small degree, Mathematika 14 (1967), 165–169.DOI
  4. [4]V. F. Cowling and W. C. Royster, Domains of variability for univalent polynomials, Proc. Amer. Math. Soc. 19 (1968), 767–772.DOI
  5. [5]L. de Branges, A proof of the Bieberbach conjecture, Acta Math. 154 (1985), 137–152.DOI
  6. [6]D. K. Dimitrov, Extremal positive trigonometric polynomials, in Approximation Theory: A Volume Dedicated to Blagovest Sendov, B. Bojanov, ed., DARBA, Sofia, 2002, 136–157.
  7. [7]D. Dmitrishin, K. Dyakonov and A. Stokolos, Univalent polynomials and Koebe’s one-quarter theorem, Anal. Math. Phys. 9 (2019), 991–1004.DOI
  8. [8]P. L. Duren, Univalent Functions, Grundlagen der mathematischen Wissenschaften 259, Springer-Verlag, New York, 1983.
  9. [9]P. Koebe, Über die Uniformisierung beliebiger analytischer Kurven, Nachr. Kgl. Ges. Wiss. Göttingen, Math.-Phys. Kl. (1907), 191–210.
  10. [10]Q. I. Rahman and J. Waniurski, Coefficient regions for univalent trinomials, Canad. J. Math. 32 (1980), 1–20.DOI
  11. [11]T. J. Suffridge, On univalent polynomials, J. London Math. Soc. 44 (1969), 496–504.DOI

Paper details

vlnc:W_5wzkd7d4
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

various large language models

View statement
Looking for related papers…
Contents