Exercises and Further Results
5. Parametrization of K
maps this annulus into the domain bounded by the slit [−4,0] and an ellipse, and the circle |w| =1 is mapped onto the slit [−4,0]. Also ϕ(G(0))>0.
Therefore
g= ϕ◦G ϕ◦G(0)
hasg(0)=1 andgis real and negative almost everywhere onT\E.
Consequently
J(eiθ)= F(eiθ)Re(u0(eiθ))=1
2λG(e¯ iθ)u0(eiθ)}{2 Reu0(eiθ) (5.3)
= 12λG(e¯ iθ)(1+u20(eiθ)), andJ ∈H1. Almost everywhere we have
|F±J| = |F|(1±Reu0), so that (5.2) gives
F+J1 = F−J1 = F1 =1.
SinceFis not outer,u0is not constant, and J =0. Then F = F+J
2 + F−J 2 andFis not an extreme point. Furthermore,
J±F= 12λG(1¯ ±2u0+u20)= 12λG¯ (1±u0)2 is an outer function, because 1±u0 is outer, and hence (5.1) holds.
Now assume F is an outer function, F1=1. If F=t F1+(1− t)F2,F1,F2 ∈ball(H1),0<t <1, then
1= F1 =t F¯
|F| F1 dθ
2π +(1−t) F¯
|F| F2 dθ 2π
≤tF11+(1−t)F21≤1,
and equality must hold throughout these inequalities. HenceF11 = F21 = 1, and
F¯
|F|Fjdθ
2π = Fj1, j =1,2.
Since|F|>0 almost everywhere, that means Fj =kjF, wherekj >0 and tk1+(1−t)k2=1. But then by the subharmonicity of log|Fj|,
|Fj(0)| ≤ |F(0)|exp
logkjdθ 2π
,
becauseFis outer. Since|F(0)| ≤t|F1(0)| +(1−t)|F2(0)|, Jensen’s inequal- ity now yields
1≤t exp
logk1dθ 2π
+(1−t) exp
logk2dθ 2π
≤t
k1dθ
2π +(1−t)
k2dθ 2π =1.
Because the exponential is strictly convex we conclude thatkjis constant and kj = Fj1/F1 =1. HenceF1 =F2 =FandFis an extreme point.
It is unfortunately the case that anH1 functionFof unit norm is not de- termined by its argument (which is defined modulo 2π almost everywhere, because|F|>0 almost everywhere). For example, ifFis not outer, then the two outer functions in (5.1) have the same argument asF(and as each other).
That follows from the construction ofF1 andF2or from the observation that by (5.1),
F¯
|F|Fjdθ
2π =1= Fj1, j =1,2,
which means ( ¯F/|F|)Fj = |Fj|almost everywhere. (See Example 1.4 above for another counterexample.) Whenh∈L∞and|h| =1 almost everywhere, we define
Sh = {F∈ H1:F1=1,F/|F| =ha.e}.
Geometrically,Sh is the intersection of ball(H1) and the hyperplane
F:
h F¯ dθ 2π =1
, (5.4)
and so Sh is a convex set. Of course, sometimesSh is empty, but we are interested in the caseSh =∅. WhenSh contains exactly one functionF, the hyperplane (5.4) touches ball(H1) only at F, which means Fis an exposed point of ball(H1). There is no good characterization of the exposed points of ball(H1), or equivalently of those F ∈ball(H1) such thatSF/|F| = {F}.
However, ifShcontains two functions, thenShis very large.
Theorem 5.2. Let h∈L∞,|h| =1almost everywhere, and assumeShcon- tains at least two distinct functions. Let z0∈ D. Then{F(z0) :F∈Sh}con- tains a disc centered at the origin.
Proof. When|z0| ≤1,|z1| ≤1, (z−z1)(1−¯z1z)
(z−z0)(1−¯z0z) = (z−z1)z(¯z−z¯1) (z−z0)z(¯z−z¯0)
is real and nonnegative onT. If there exists F∈Sh having a zero of orderk atz0, then
Fz1(z)=
(z−z1)(1−¯z1z) (z−z0)(1−¯z0z)
k
F(z) is inSh and the valuesFz1(z0) fill a disc about 0.
It remains to show there isF ∈Sh with F(z0)=0. If the convex setSh contains two functions, then by Theorem 5.1 it contains a function F=uG
with nonconstant inner factoru(z). Let J(z) be the function defined in (5.3).
Then for 0<t <1,
F+t J =F(1+tReu0)
satisfiesF+t J1 =1, because of (5.2), andF+t J ∈Sh. By (5.3) we also have
F+t J = 12λt G¯
1+ 2u0
t +u20
. When 0<t <1, the equation
ζ2+ 2ζ
t +1=0
has a rootζ(t),|ζ(t)|<1, and these roots fill the segment (−1,0). IfF+t J has no zero inDfor eacht ∈[0,1), then the range ofu0 is disjoint from the segment (−1,0]. But then by Corollary 4.8, Chapter II,u0 is both an inner function and an outer function, so thatu0is constant. Thus there existstsuch thatF+t J has a zero at some pointz1∈ D. Then
(z−z0)(1−z¯0z)
(z−z1)(1−z¯1z)(F+t J) is a function inSh having a zero atz0.
Returning to the topic of the previous section, we fix a coseth0+H∞ of L∞/H∞and we assume
K = {h∈h0+H∞:h ≤1}
contains more than one function. By Theorem 4.3,Kthen contains a function, which we callh0, such that|h0| =1 almost everywhere. We will need to recall that after a change of coordinatesh0is an extremal function:
Re
h0dθ 2π =sup
h∈K Re
hdθ 2π. (5.5)
(See the proof of Theorem 4.3.) The Adamyan, Arov, and Krein parametriza- tion ofKwhich we have been seeking is (5.7) below.
Theorem 5.3. There exists a unique outer function F ∈H1,F1 =1, such that
h0 = F/|F|.
(5.6)
Defineχ ∈H∞by
1+χ(z) 1−χ(z) = 1
2π
eiθ+z
eiθ−z|F(eiθ)|dθ.
Then
K =
h0− F(1−χ)(1−w)
1−χw :w(z)∈H∞,w∞≤1
. (5.7)
As a consequence of the parametrization (5.7) we see that, forz∈D,
h Pzdθ
2π :h∈ K
is a nondegenerate closed disc. Another corollary of (5.7) is the description by Nevanlinna [1929] of all solutions f ∈H∞,f ≤1, of the interpolation problem
f(zj)=wj, j =1,2, . . . (see Section 6 below).
Condition (5.6) may seem to contradict Theorem 4.4, because whenh0+ H∞<1, we have claimed that the extremal functionh0has the two forms
h0 =F¯1/|F1|, h0 = F2/|F2|
withF1,F2 ∈H1,Fj1=1. However, ifF1∈ H1andg∈H∞, and if F¯1
|F1|−g
∞ =α <1,
then |arg(g F1)| ≤sin−1(1−α)< π/2, and Corollary III.2.5 shows that (g F1)−1∈ H1. Hence F1−1 =g(g F1)−1 ∈H1 and (5.6) holds with F2 = F1−1/F1−11.
The proof of Theorem 5.3 requires three lemmas; the first uses an idea from Koosis [1973].
Lemma 5.4. There exists an outer function F∈ H1,F1=1, such that h0 = F/|F|
almost everywhere.
Proof. We know |h0| =1 almost everywhere and we know there is g∈ H∞,g≡0, such that h0−g∞ ≤1. Then |1−h¯0g| ≤1 almost every- where. Letα=arg ¯h0g. Then|α| ≤π/2 and
|g| = |h¯0g| ≤2 cosα
as shown in Figure IV.3. Letϕ=eα−iα˜ . Thenϕ∈ Hp,p<1, by Theorem III.2.4. However, we also havegϕ∈H1, because
|ϕ(eiθ)g(eiθ)| ≤2|ϕ(eiθ) cosα(eiθ)| =2 Reϕ(eiθ), (5.8)
Figure IV.3. Why|g| ≤2 cosα.
and since Reϕ≥0, 1 2π
Reϕdθ ≤lim
r→1
Reϕ(r eiθ)dθ
2π ≤ Reϕ(0).
Henceϕg∈Hp∩L1= H1. ThenF0 =ϕg/ϕg1is in ball(H1) and ¯h0F0 ≥ 0, which means that
h0 =F0/|F0|.
Finally, by the remarks following Theorem 5.1 there is an outer functionF∈ ball(H1) such that (5.6) holds.
Lemma 5.5. Suppose F(z)is any H1 function, F≡0. Defineχ ∈H∞by 1+χ(z)
1−χ(z) = 1 2π
eiθ+z
eiθ−z|F(eiθ)|dθ.
(5.9)
Ifw(z)∈H∞,w∞≤1,then
g(z)= F(z)(1−χ(z))(1−w(z)) 1−χ(z)w(z) (5.10)
is in H∞and
F
|F| −g ∞≤1.
(5.11)
Proof. Note that
Re1+χ(z) 1−χ(z) =
Pz(θ)|F(θ)|dθ 2π >0. Let|w(z)|<1 and set
ϕ(z)= 1+χ(z)
1−χ(z)+ 1+w(z)
1−w(z) = 2(1−χ(z)w(z)) (1−χ(z))(1−w(z)). (5.12)
Thenϕis holomorphic and since
Re1+w(z) 1−w(z) ≥0, we have
Reϕ(z)≥
Pz(θ)|F(θ)|dθ
2π ≥ |F(z)| ≥0. (5.13)
A simple calculation from (5.10) and (5.12) gives g(z)=2F(z)/ϕ(z),
so thatg∈ H∞. Whenw≡1,g=0, and whenw≡1, ϕhas nontangential limits almost everywhere satisfying
|ϕ(eiθ)| ≥ |F(eiθ)|.
Notice that
F(eiθ)
|F(eiθ)|−g(eiθ) ≤1 if and only if
1
|F(eiθ)| − g(eiθ) F(eiθ)
≤ 1
|F(eiθ)|. (5.14)
The transformationζ →2/ζmaps the half plane Reζ ≥ |F(eiθ)|onto the disc F(eiθ)|−1−w| ≤ |F(eiθ)|−1. By (5.13), Re (2F/g)= Re ϕ≥ |F| almost everywhere, so that we have (5.14) and consequently (5.11).
Lemma 5.6. The function F∈H1,F1 =1, such that h0 = F/|F|
is unique.
Proof. Lemma 5.4 shows there exists at least one suchF. What we must show is thatSh0does not contain two functions. But ifSh0contains two functions, then by Theorem 5.2, there isF1 ∈Sh0 with ReF1(0)<0. SinceF11 =1, the functionχ associated withF1by (5.9) satisfiesχ(0)=0. Takingw≡ −1
in (5.10), we obtaing∈ H∞such that Reg(0)<0, and by (5.11),h0−g∈K. Then
Re
(h0−g)dθ 2π > Re
h0dθ
2π, contradicting (5.5).
Conclusion of the proof of Theorem 5.3. By Lemmas 5.4 and 5.6, there is a unique outer functionF ∈H1,F1 =1, such that (5.6) holds. By Lemma 5.5 every function of the form
h0− F(1−χ)(1−w)
1−χw ,
(5.15)
w∈H∞,w∞≤1, lies inK. Now letg∈ H∞,g≡0, be such thath0− g ≤1. We must showh0−ghas the form (5.15) forw∈ball(H∞). The idea is in the proof of Lemma 5.4. Settingα=arg ¯h0g, andϕ=eα−iα˜ , we have ϕg∈H1 and, by its uniqueness,
F = ϕg ϕg1. Now Reϕ(z)>0 and by (5.8)
Re 2ϕ(eiθ) ϕg1
≥ |ϕ(eiθ)g(eiθ)|
ϕg1
= |F(eiθ)|.
The positive harmonic function Re 2ϕ(z)/ϕg1 is the Poisson integral of a positive measure with absolutely continuous part exceeding|F(eiθ)|. Conse- quently
2ϕ(z) ϕg1
−1+χ(z) 1−χ(z) =k(z) has Rek(z)>0, so that
k(z)= 1+w(z) 1−w(z), w∈H∞,w∞≤1. A calculation then gives
g= ϕg1F
ϕ =2F
2ϕ
|ϕg|1
−1
= 2F
1+χ 1−χ
1+w 1−w
=F(1−χ)(1−w) 1−χw , and (5.15) holds.