Notes on Complex Analysis
Export PDFGitHub

§8.5 The Great Picard, Bloch, Landau, and Schottky Theorems

8.5 The Great Picard, Bloch, Landau, and Schottky Theorems

Recall the Casorati–Weierstrass Theorem, one of the earliest results on the value distribution near essential singularities:

Theorem 4.2.4 (Casorati–Sokhotski–Weierstrass).

Let 𝑎∈ℂ and 𝑈⊆ℂ be an open region. Suppose 𝑓:𝑈∖{𝑎}→ℂ is holomorphic with an essential singularity at 𝑎. Then the set of values that 𝑓 attains on any open punctured neighborhood of 𝑎 is dense. In other words, ∀𝜀,𝛿>0, ∀𝑤∈ℂ, ∃𝑧∈𝐷∗(𝑎,𝛿) such that |𝑓(𝑧)−𝑤|<𝜀.

We will now prove a more advanced characterization of this distribution by methods of differential geometry.

Theorem 8.5.1 (Picard’s Great Theorem).

Suppose 𝑓 is holomorphic on a punctured neighborhood 𝐷∗(𝑧0,𝛿) of 𝑧0. If 𝑧0 is an essential singularity of 𝑓, then 𝑓(𝐷∗(𝑧0,𝛿)) omits at most one value of ℂ.
Proof.

Without loss of generality, assume 𝑧0=0 and that 𝑓 omits the values 0 and 1 (otherwise, consider 𝑧↦1𝛽−𝛼(𝑓(𝑧+𝑧0)−𝛼), where 𝛼 and 𝛽 are the omitted values). Define the family

ℱ︀={𝑧↦𝑓(𝑧𝑛):𝑛∈ℕ}

of holomorphic functions on 𝐷∗(0,𝛿). Since 𝑓 omits 0 and 1, each element of ℱ︀ does as well. By the Fundamental Normality Test (Theorem 8.4.7Theorem 8.4.7Theorem 8.4.7), ℱ︀ is spherically normal. Thus, there exists a subsequence {𝑓𝑛𝑘}𝑘∈ℕ⊆ℱ︀ that converges locally uniformly on 𝐷∗(0,𝛿) in the spherical metric. By Proposition 8.4.5Proposition 8.4.5Proposition 8.4.5, this subsequence converges locally uniformly either to a holomorphic function on 𝐷∗(0,𝛿) or to ∞ thereon.

  1. Suppose {𝑓𝑛𝑘}𝑘∈ℕ converges locally uniformly to a holomorphic function on 𝐷∗(0,𝛿). Then {𝑓𝑛𝑘}𝑘∈ℕ is uniformly bounded on 𝜕𝐷(0,𝛿2). Hence, there exists 𝑀>0 such that

    |𝑓(𝑧𝑛𝑘)|=|𝑓𝑛𝑘(𝑧)|<𝑀∀𝑧∈𝜕𝐷(0,𝛿2),𝑘∈ℕ.

    In other words, 𝑓 is bounded by 𝑀 on every circle 𝜕𝐷(0,𝛿2𝑛𝑘) for 𝑘∈ℕ. By the Maximum Modulus Principle (Theorem 3.4.2Theorem 3.4.2Theorem 3.4.2), 𝑓 is then bounded by 𝑀 on each annulus 𝐷(0,𝛿2𝑛𝑘)‾∖𝐷(0,𝛿2𝑛𝑘+1) for 𝑘∈ℕ. As

    ⋃𝑘∈ℕ𝐷(0,𝛿2𝑛𝑘)‾∖𝐷(0,𝛿2𝑛𝑘+1)=𝐷(0,𝛿2𝑛1)‾∖{0},
    it follows that 𝑓 is bounded on 𝐷∗(0,𝛿2). By Riemann’s Removable Singularity Theorem (Theorem 3.2.9Theorem 3.2.9Theorem 3.2.9), 𝑓 therefore extends holomorphically to 0.
  2. Suppose {𝑓𝑛𝑘}𝑘∈ℕ converges locally uniformly to ∞ on 𝐷∗(0,𝛿). Then, for every 𝜀>0, there exists 𝑁∈ℕ such that, for all 𝑘>𝑁,

    |1𝑓(𝑧𝑛𝑘)|=|1𝑓𝑛𝑘(𝑧)|<𝜀∀𝑧∈𝜕𝐷(0,𝛿2).

    By the same reasoning as in the previous case, |1𝑓|<𝜀 on

    ⋃𝑘>𝑁𝐷(0,𝛿2𝑛𝑘)‾∖𝐷(0,𝛿2𝑛𝑘+1)⊃𝐷∗(0,𝛿2𝑛𝑁+1).
    Thus, by the definition of the limit, lim𝑧→01𝑓(𝑧)=0, so 𝑓 has a pole at 0.

In either case, we have derived a meromorphic continuation of 𝑓 to 0, contradicting the assumption that 0 is an essential singularity of 𝑓. ⁠□

Corollary 8.5.1.1.

Suppose that 𝑓 is meromorphic on a punctured neighborhood 𝐷∗(𝑧0,𝛿) of 𝑧0. If 𝑓(𝐷∗(𝑧0,𝛿)) omits at least three different values of ℂ̂, then 𝑓 has a meromorphic continuation to 𝑧0.
Proof.A linear fractional transformation maps the omitted values to 0, 1, and ∞, mapping 𝑓 so that it exhibits holomorphy. Similar to Corollary 8.4.7.1Corollary 8.4.7.1Corollary 8.4.7.1, the preceding result is preserved under the inverse linear fractional transformation. ⁠□

Remark.

An accumulation point of poles is an essential singularity on the Riemann sphere.

Picard’s Great Theorem is also a generalization of Picard’s Little Theorem (Theorem 8.3.6Theorem 8.3.6Theorem 8.3.6):

Theorem 8.3.6 (Picard’s Little Theorem).

Let 𝑓:ℂ→𝑈 be entire such that ℂ∖𝑈 contains two or more points. Then 𝑓 is constant.
Proof.

Let 𝑔(𝑧)=𝑓(1𝑧) with an isolated singularity at 0 and a removable singularity at ∞. By Picard’s Great Theorem (Theorem 8.5.1Theorem 8.5.1Theorem 8.5.1), 𝑔(𝑧) has a meromorphic extension to 𝑧=0. If 𝑧=0 is removable, by virtue of Proposition 4.3.1Proposition 4.3.1Proposition 4.3.1 and Theorem 3.2.5Theorem 3.2.5Theorem 3.2.5, the constancy of 𝑔 and 𝑓 follows.

If instead 𝑧=0 is a pole of 𝑔, then 𝑧=∞ is a pole of 𝑓, and hence 𝑓 is a polynomial. Assume, for the sake of contradiction that 𝑓 is non-constant. Then ∀𝑤∈ℂ, the Fundamental Theorem of Algebra (Theorem 3.3.1Theorem 3.3.1Theorem 3.3.1) gives the existence of some 𝑧∈ℂ such that 𝑓(𝑧)=𝑤. Hence, 𝑓 attains every value 𝑤∈ℂ. This contradicts the statement and hence 𝑓 is constant. ⁠□

The efforts of many mathematicians resulted in several alternative proofs following that of Picard; the geometric realization of Ahlfors (Theorem 8.3.1Theorem 8.3.1Theorem 8.3.1) was followed by results discovered by R. M. Robinson. Other approaches from Nevanlinna theory appeared later in the 20th century.

Picard’s original proof, providing an advanced characterization of the value distribution at essential singularities, relied primarily on the properties of the elliptic modular function (as a “covering map”). From this, Picard deduced that the function would necessarily extend holomorphically across the singularity, contradicting its essential nature. Thus, his proof established that near an essential singularity, a holomorphic function attains every complex value, with at most one exception, infinitely often.

More importantly, we have shown the utility of even seemingly fundamental differential geometry, which can also be used in the proof of many other important results.

The methods of differential geometry can also be used to prove the statements of the following theorems (which can also be independently used to prove the Picard theorems), but is made meaningful with the notion of Riemann surfaces.

Theorem 8.5.2 (Bloch).

Let 𝑓:𝔻→ℂ be holomorphic such that |𝑓′(0)|=1. Then there is a region 𝑆⊆𝔻 on which 𝑓 is univalent such that 𝑓(𝑆) contains a disk with a radius of at least 34 (known as a schlicht disk).

Remark.

Bloch’s constant 𝐵 is defined as the supremum of the radii of such disks that can be contained in 𝑓(𝔻) for any holomorphic function 𝑓:𝔻→ℂ satisfying 𝑓′(0)=1.

The precise value of 𝐵 remains unknown to this day. In 1937, H. Grunsky and L. Ahlfors established the bound

𝐵≤Γ(13)Γ(1112)Γ(14)3−12,

where Γ denotes the Gamma function (as in (12.1.1)(12.1.1)(12.1.1)). Later the lower bound of 34 was given, then to be refined to 𝐵≥34+10−1213 by M. Bonk, which was further improved to 𝐵≥34+15000 in 1996 by H. Chen and P. M. Gauthier.

Grunsky and Ahlfors actually conjectured that the upper bound in their inequality is exact – that is, 𝐵=Γ(13)Γ(1112)Γ(14)3−12.

Theorem 8.5.3 (Landau).

The image of any holomorphic function 𝑓 in 𝔻 satisfying 𝑓(0)=0 and 𝑓′(0)=1 contains a disk with radius of at least 12.

Remark.

Similarly, the estimate 12 is not optimal. It was established that the corresponding Landau’s constant lies between 12 and Γ(13)Γ(56)Γ(16).

Without Riemann surfaces, the proof of the two aforesaid results are rather difficult, as a distinction must be established for a point 𝑤∈𝑓(𝔻) which two values 𝑧1,𝑧2 map to. Otherwise, when we describe a schlicht disk at a point in the image, we may be talking about different “sheets” or “branches,” although each fixed sheet may describe perfectly well-defined analytic functions, although they describe different “copies.” More details may be found in [2].

Hence, for simplicity, we entertain a much simpler case without algebraic branch points.

Theorem 8.5.4.

Let 𝑓:𝔻→ℂ be univalent such that |𝑓′(0)|=1. Then 𝑓(𝔻) contains a disk with a radius of at least 34.
Proof.

For 𝑤∈𝑓(𝔻), let 𝜙(𝑤) denote the radius of the largest schlicht disk in 𝑓(𝔻) centered at 𝑤 (it is mapped to univalently by 𝑓 on some subdomain). Trivially, 𝜙 is 𝐶0 and vanishes toward the boundary of 𝑓(𝔻).

Define the metric

𝜌(𝑤)=𝐴𝜙(𝑤)(𝐴2−𝜙(𝑤)),d𝑠=𝜌(𝑤)|d𝑤|

for 𝑤∈𝑓(𝔻) and 𝜌(𝑤)≠0, where 𝐴2>sup𝑤𝜙(𝑤) is a constant. We may assume that sup𝑤𝜙(𝑤) is finite, since otherwise the theorem is already proved for 𝑓.

For every point 𝑤0∈𝑓(𝔻), the bounding circle corresponding to 𝜙(𝑤0) passes through a (at least one) boundary point, denoted by 𝑏=𝑏𝑤0. Let 𝜙𝑤0(𝑤)=|𝑤−𝑏| and let

𝜌𝑤0(𝑤)=𝐴𝜙𝑤0(𝑤)(𝐴2−𝜙𝑤0(𝑤)),𝑤∈𝐷(𝑤0,𝜙(𝑤0)).

By the definition of 𝜙, we have 𝜙𝑤0≥𝜙 everywhere in this neighborhood. Since 𝜌𝑤0=(𝜙𝑤0↦𝜙𝑤0)∗(𝜑↦2𝐴𝐴2−𝜑2) is the pullback of the hyperbolic metric in (8.3.4)(8.3.4)(8.3.4), the metric 𝜌𝑤0(𝑤) has the constant negative curvature of −1.

Our goal is to construct 𝜌𝑤0 so that it is the function of a supporting metric for 𝜌 (satisfies the criteria for Theorem 8.3.3Theorem 8.3.3Theorem 8.3.3). For 𝜌𝑤0≤𝜌 to be satisfied, we consider

𝜙𝑤0(𝑤)(𝐴2−𝜙𝑤0(𝑤))≥𝜙(𝑤)(𝐴2−𝜙(𝑤)),𝜙𝑤0(𝑤)≥𝜙(𝑤).

In particular, we want

𝑡↦𝑡(𝐴2−𝑡)

to be increasing on [0,𝜙(𝑤0)+𝛿] for arbitrary 𝛿>0. The function itself can be calculated to be increasing for 𝑡≤𝐴23 by elementary methods (using derivative tests). Therefore, the conditions for a supporting metric are satisfied if 𝐴23≥sup𝑤∈𝑓(𝔻)𝜙(𝑤)+𝛿≥𝜙(𝑤0)+𝛿. Without loss of generality we let 𝛿→0+ and thus, under the condition that 𝐴23>sup𝑤𝜙(𝑤), Theorem 8.3.3Theorem 8.3.3Theorem 8.3.3 gives that

𝜌(𝑤)|d𝑤|≤2|d𝑧|1−|𝑧|2.

Let 𝑧=0, 𝑤=𝑓(0), so that by the theorem conditions, |d𝑤d𝑧|=1, and therefore

𝜌(𝑓(0))≤2⟹𝐴≤2𝜙(𝑓(0))[𝐴2−𝜙(𝑓(0))].

By the previous assumptions on 𝐴, the corresponding function on the right-hand side is increasing, and since 𝜙(𝑓(0))≤sup𝑤𝜙(𝑤), we have

𝐴2≤4sup𝑤𝜙(𝑤)(𝐴2−sup𝑤𝜙(𝑤))2.

As 𝐴23→sup𝑤𝜙(𝑤)+ (𝐴 was chosen arbitarily, so this is valid), it follows that

sup𝑤𝜙(𝑤)≥34.□

It is however notable that the proof follows similarly for general functions, but instead we consider functions 𝑓:𝔻→𝑊, where 𝑊 is a Riemann surface and the “singularities” are not only boundary points but also algebraic branch points (where 𝑓′=0).

Theorem 8.5.5 (Landau–Carathéodory).

Let 𝑓(𝑧)=∑𝑛=0∞𝑎𝑛𝑧𝑛 such that 𝑎1≠0 and 𝑓 is holomorphic on 𝐷(0,𝑟). If 𝑓 omits 0 and 1, then ∃𝑅 dependent only on 𝑎0 and 𝑎1 such that 𝑟≤𝑅.

Theorem 8.5.6 (Schottky).

Suppose that 𝑓:𝔻→ℂ is holomorphic and omits 0 and 1. Then

log|𝑓(𝑧)|<1+|𝑧|1−|𝑧|(7+log+|𝑓(0)|),
where log+(𝑥)=max{0,log𝑥} (common notation in Nevanlinna theory).
Proof.

Consider conformal map 𝜁1(𝑤):ℂ∖[0,1]→ℂ∖𝔻‾, which extends to 0, 1, and ∞ continuously such that 𝜁1(∞)=∞, 𝜁1(1)=1, 𝜁1(0)=−1. Explicitly, we have the relationship

𝜁1+1𝜁1=4𝑤−2

as an affine transformation of the inverse Joukowski transform (inverse of 𝑧↦𝑧+𝑧−1). The solution is given by

𝜁1=(2𝑤−1)+2𝑤1−1𝑤=(2𝑤−1)+2𝑤exp[12Log(1−1𝑤)],
where the branch cut of the square root is taken to be the negative real axis, which maps to [0,1] in terms of 𝑤 (and with the principal branch logarithm). Moreover, this explicit map maps [0,1] to 𝜕𝔻, since for 𝑤∈[0,1], the term involving the square root is purely imaginary thus |𝜁1|=4𝑤2+1−4𝑤−4𝑤2+4𝑤=1. Because as 𝑤→∞, 𝜁1→∞, 𝜁1 here is a valid conformal map. ⁠□