Exercises and Further Results
2. The Carleson–Jacobs Theorem
To complete the proof of Corollary 1.8, notice thatG = f F/B0 is an H1 function on the annulusr <|z|<1 ifr >|zj|. Using (1.10) and using the lemma locally, we see that G is analytic across T, and thatG is in fact a rational function. Since B0G is analytic acrossT, and since f/f∞ is an inner function, Theorem II.6.3 shows thatf is analytic acrossT. Consequently f/f∞ is a Blaschke product of finite degree, andFis a rational function.
NowB0hasnzeros inDandFhas a zero atz=0. By (1.10) and the argument principle, it follows thatf has at mostn−1 zeros.
In Corollary 1.8 the extremal functionf(z)=cB(z) can have fewer thann− 1 zeros. For example, supposegis the function of minimum norm interpolating
g(zj)=wj, 1≤ j ≤n−1,
and takewn =g(zn). Then f =ghas at mostn−2 zeros. Similarly, the ex- tremal function in Corollary 1.9 can have fewer thannzeros. However, if the in- terpolation (1.9) has two distinct solutions f1, f2withf1∞≤1,f2∞≤1, then the extremal function f0in Corollary 1.9 is a Blaschke product of degree n. For the proof, notice that (1.9) then has a solution f withf∞ <1, by the uniqueness asserted in Corollary 1.8. Then by Rouch´e’s theorem, f0andf0− f have the same number of zeros in|z|<1. Since f0(zj)= f(zj),1≤ j ≤n, it follows that f0has at leastnzeros in|z|<1.
Corollary 2.4 Chapter I, contains more information than we have obtained here, but the duality methods of this section apply to a wider range of problems.
f(θ)∈C such thatωf(δ)≤ω(δ)but such that the best approximation to f in H∞is not continuous.
We prove Theorem 2.1 first. In doing so we can assume f −g∞ =dist(f,H∞)=1. By Theorem 1.7, there isF∈ H01,F1 =1, such that
(f −g)F= |F| a.e (2.1)
All information used in the proof will follow from (2.1) and the continuity of f. We need two lemmas.
Lemma 2.3. Let G=u+iv∈ H1and let I be an arc on T such that almost everywhere on I
u>0, |v| ≤αu,
whereα >0. Let J be a relatively compact subarc of I and let V be the domain {r eiθ :r0 <r <1,0∈ J}. Then G∈Hp(V)if
arctanα < π/2p.
The statementG ∈ Hp(V) means that|G|phas a harmonic majorant inV.
Proof. Recall from Corollary 2.5 in Chapter III that an H1 function whose boundary values lie in the sector S= {x >0,|y|< αx} is in Hp if arctan α < π/2p. EnlargingJ, we may assumeGhas a finite radial limit at the end- points ofJ. ThenM =sup{|G(z)|:z∈∂V,|z|<1}is finite, and
g= M(1+1/α)+G
has, at almost every point of∂V, boundary value in the coneS(see Figure IV.1).
Figure IV.1. S= {|y|< αx,x>0}. On∂V,ghas values in the union of the disc and the shaded cone.
Becauseg∈ H1(V), the Poisson integral formula (applied after conformally mappingVonto the unit disc) shows thatg(V)⊂S. By Corollary III.2.5 and the conformal invariance of the definition of Hp(V), we see thatg∈Hp(V) and hence thatG ∈ Hp(V).
Lemma 2.4. Let f(θ)∈C, let g(θ)∈H∞and let F(θ)∈H01 be functions such that(2.1)holds. Then
(a) F ∈Hp for all p<∞, and (b) ifτ ∈[0,2π]and if
fτ = f − f(τ), gτ =g− f(τ), then there isδ >0and r0 >0such that|gr(z)| ≥ 12 on
Wτ = {r eiθ :|θ−τ|< δ/2,r0 <r <1}.
Proof. To prove (a) let p<∞ and let ε >0 satisfy arctan(ε/(1−ε))<
π/2p. Chooseδso that|f(θ)− f(τ)|< εwhen|θ−τ|< δ. Then on Iτ = {θ :|θ−τ|< δ}, we have by (2.1)
−gτF=(f −g)F− fτF= |F| − fτF.
Consequently, Re(−gτF)>0 and |Im gτF|< ε/(1−ε) Re(−gτF) almost everywhere onIτ. By Lemma 2.3 we havegτF∈ Hp(Wτ). If we replaceWτ
by the intersection of two discs whose boundaries cross inside Iτ, a simple conformal mapping can be used to show
|θ−τ|<δ/2|gτF|pdθ <∞.
Since|gτ| ≥ f −g∞−ω(δ)≥1−εonIτ, this means
|θ−τ|<δ/2|F|pdθ <
∞, so thatF∈Lp. HenceF∈ Hp by Corollary 11.4.3.
To prove (b) first takeεso small that onIτ
−gτF=exp(u+iv), wherev∞< π/4. then
h=exp(−ivχIτ +vχIτ)
is inH2by Corollary III.2.6. TheH1functiongτFhis real onIτ. By Lemma 1.10 and Theorem II.6.3, the inner factor ofgτFhis analytic acrossIt. Conse- quently, the inner factor ofgτ is analytic across Iτ. The representation for- mula for outer functions then shows |gτ(z)| ≥ 34 in a region {r0(τ)<r <
1,|θ−τ|< δ/2}. Sincegσ(z)=gτ(z)−(f(σ)− f(τ)), we have|gσ(z)|> 12 on the same region if|σ −τ|< δ. This means we may chooser0(τ) indepen- dent ofτ, and thus (b) is proved.
Proof of Theorem 2.1. By Lemma 2.4(b), gτ(z) has a single-valued loga- rithm onWτ, and the logarithm is defined by
loggτ(z)= 1 2π
|θ−τ|<δlog|gτ(θ)|eiθ +z
eiθ −z dθ+Rτ(z), (2.2)
where Rτ(z) is the same integral over|θ−τ| ≥δ plus the logarithm of the inner factor of gτ(z). Since|gτ(z)| ≥ 12 onWτ, the inner factor is bounded below on Wτ and it is analytic across {eiθ :|θ−τ|< δ/2}. Hence there is r1>0 such thatRτ(z) is bounded and analytic onτ = {|z−eiτ|<r1}. The radiusr1and the bound sup{|Rτ(z)|:z ∈τ}are independent ofτ.
Because|fτ−gτ| =1, we have
1− |gτ|2 = |fτ|2−2 Refτg¯τ ≤ Aω(|θ−τ|), so that
|log|gτ(θ) ≤cω(|θ−τ|). Let(τ) be the truncated cone
(τ)=
z: |z−eiτ|
1− |z| ≤2,r2 <|z|<1
,
wherer2 >0 is such that(τ)⊂τ. Forz ∈τ, eiθ+z
eiθ−z
≤ C
|θ−τ|, so that (2.2) yields
|loggτ(z)−Rτ(z)| ≤C δ
0
ω(t)
t dt, z∈(τ).
By Schwarz’s lemma and the uniform bound on Rτ(z),|Rτ(z)−Rτ(w)| ≤ c|z−w|,z, w∈(τ). Therefore
|gτ(z)−gτ(w)| ≤C|z−w| +η(δ), z, w∈τ, whereη(δ)→0(δ→0).
Now letτ andσ be so close together that there is a pointz∈(τ)∩(σ).
Taking nontangential limits, we then obtain (when 1− |z|is small)
|g(eiτ)−g(eiσ)| ≤ |g(eiτ)−g(z)| + |g(z)−g(eiσ)|
≤ |gτ(eiτ)−gτ(z)| + |gσ(z)−gσ(eiσ)|
≤2η(δ)+ |eiτ−z| + |eiσ −z|
≤2η(δ)+C|σ −τ|.
Fixingδ >0 with 2η(δ)< ε, we conclude that
σlim→τ|g(eiσ)−g(eiτ)|< ε, which means thatgis continuous.
Now letω(t) satisfy the hypotheses of Theorem 2.2.
Lemma 2.5. Letδ >0. Let f(eit)=
ω(t), 0≤t ≤δ, 0, −δ≤t <0.
Extend f(eit) to be smooth onδ <|t|< π and continuous on[−π, π]with f(−π)= f(π). Let g∈H∞be the best approximation of f. If g is continuous, then
g(1)= ±if −g∞.
Proof. Since f is real, f ∈/ H∞ and f −g∞>0. We may suppose f −g∞ =1, so that|f −g| =1 onT. We must prove Reg(1)=0. Since
|f −g| =1,log|g(t)| =0 on−δ <t <0. Suppose Re g(1)>0. Then for 0<t < δ,
log|g| = −log g− f
g
≤ −log
1− f Reg
|g|2
≤ −Cω(t).
Similarly, if Reg(1)<0, then on 0<t < δ,log|g| ≥Cω(t). Moreover,ghas a continuous logarithm on{|z−1|< δ,|z| ≤1}, again because|f −g| =1.
However, if Reg(1)=0, then by Lemma III.1.2,
r→1lim|argg(r)| = c+lim
ε→0
1 π
|t|>ε
log|g(t)| t dt
≤c+Clim
ε→0
1 π
|t|>ε
ω(t)
t dt = +∞.
This contradiction showsg(1)= ±i.
To prove Theorem 2.2 letδn =2−n, and let ωn(t)=ω
δn
2 −
δn+δn−1
2 −t
on [δn, δn−1], ωn(t)=0 off [δn, δn−1]. Let
f(eit)=∞
k=1
1
2ω4k+1(t)+i ∞ k=1
1
2ω4k+3(t).
Thenf2is real, so that f ∈/ H∞. It is not hard to verify thatωf(δ)≤ω(δ). If gis the best approximation off, then Lemma 2.5 says that
g(eiδ4k+1)= ±if −g∞, g(eiδ4k+3)= ±f −g∞. Thusg(eit) is not continuous att =0.