Introduction
Let D = { z ∈ C : ∣ z ∣ < 1 } \mathbb{D}=\{z\in\mathbb{C}:|z|<1\} D = { z ∈ C : ∣ z ∣ < 1 } , let T = ∂ D \mathbb{T}=\partial\mathbb{D} T = ∂ D , and let d A ( z ) = d x d y / π dA(z)=dx\,dy/\pi d A ( z ) = d x d y / π be normalized area measure. Write Hol ( D ) \operatorname{Hol}(\mathbb{D}) Hol ( D ) for the analytic functions on D \mathbb{D} D . For a ∈ D a\in\mathbb{D} a ∈ D , set
φ a ( z ) = a − z 1 − a ‾ z , 1 − ∣ φ a ( z ) ∣ 2 = ( 1 − ∣ a ∣ 2 ) ( 1 − ∣ z ∣ 2 ) ∣ 1 − a ‾ z ∣ 2 . \varphi_a(z)=\frac{a-z}{1-\overline{a}z},\qquad1-|\varphi_a(z)|^2=\frac{(1-|a|^2)(1-|z|^2)}{|1-\overline{a}z|^2}. φ a ( z ) = 1 − a z a − z , 1 − ∣ φ a ( z ) ∣ 2 = ∣1 − a z ∣ 2 ( 1 − ∣ a ∣ 2 ) ( 1 − ∣ z ∣ 2 ) .
For p > 0 p>0 p > 0 , the space Q p Q_p Q p consists of all f ∈ Hol ( D ) f\in\operatorname{Hol}(\mathbb{D}) f ∈ Hol ( D ) such that
∥ f ∥ Q p = ∣ f ( 0 ) ∣ + ∥ f ∥ Q p , ∗ < ∞ , \|f\|_{Q_p}=|f(0)|+\|f\|_{Q_p,*}<\infty, ∥ f ∥ Q p = ∣ f ( 0 ) ∣ + ∥ f ∥ Q p , ∗ < ∞ ,
∥ f ∥ Q p , ∗ 2 = sup a ∈ D ∫ D ∣ f ′ ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p d A ( z ) . \|f\|_{Q_p,*}^2=\sup_{a\in\mathbb{D}}\int_{\mathbb{D}}|f'(z)|^2(1-|\varphi_a(z)|^2)^p\,dA(z). ∥ f ∥ Q p , ∗ 2 = a ∈ D sup ∫ D ∣ f ′ ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p d A ( z ) .
We use this norm throughout. It is equivalent to the usual norm defined by the Green function log ( 1 / ∣ φ a ( z ) ∣ ) \log(1/|\varphi_a(z)|) log ( 1/∣ φ a ( z ) ∣ ) ; see [19 Section 2.2] . The space Q p Q_p Q p is a Banach space and its seminorm is Möbius invariant: for every automorphism ψ \psi ψ of D \mathbb{D} D ,
∥ f ∘ ψ ∥ Q p , ∗ = ∥ f ∥ Q p , ∗ . \|f\circ\psi\|_{Q_p,*}=\|f\|_{Q_p,*}. ∥ f ∘ ψ ∥ Q p , ∗ = ∥ f ∥ Q p , ∗ .
The scale contains two classical spaces:
Q 1 = BMOA , Q p = B ( p > 1 ) , Q_1=\operatorname{BMOA},\qquad Q_p=\mathcal{B}\quad(p>1), Q 1 = BMOA , Q p = B ( p > 1 ) ,
with equivalent norms, where BMOA is the analytic subspace of boundary BMO and B \mathcal{B} B is the Bloch space with norm
∥ f ∥ B = ∣ f ( 0 ) ∣ + sup z ∈ D ( 1 − ∣ z ∣ 2 ) ∣ f ′ ( z ) ∣ . \|f\|_{\mathcal{B}}=|f(0)|+\sup_{z\in\mathbb{D}}(1-|z|^2)|f'(z)|. ∥ f ∥ B = ∣ f ( 0 ) ∣ + z ∈ D sup ( 1 − ∣ z ∣ 2 ) ∣ f ′ ( z ) ∣.
Moreover Q r ⊆ Q p Q_r\subseteq Q_p Q r ⊆ Q p when 0 < r ≤ p 0<r\le p 0 < r ≤ p . We write Q ˚ p = { f ∈ Q p : f ( 0 ) = 0 } \mathring{Q}_p=\{f\in Q_p:f(0)=0\} Q ˚ p = { f ∈ Q p : f ( 0 ) = 0 } .
Let u ∈ Hol ( D ) u\in\operatorname{Hol}(\mathbb{D}) u ∈ Hol ( D ) and let φ : D → D \varphi:\mathbb{D}\to\mathbb{D} φ : D → D be holomorphic. The weighted composition operator induced by u u u and φ \varphi φ is
W u , φ f ( z ) = u ( z ) f ( φ ( z ) ) . W_{u,\varphi}f(z)=u(z)f(\varphi(z)). W u , φ f ( z ) = u ( z ) f ( φ ( z )) .
2020 Mathematics Subject Classification. 30H25, 47B33, 42B35.
Key words and phrases. Q p Q_p Q p space, weighted composition operator, dyadic capacity, compactness. For u = 1 u=1 u = 1 it is the composition operator C φ C_{\varphi} C φ , and for φ = id \varphi=\operatorname{id} φ = id the multiplication operator M u M_u M u . We seek necessary and sufficient conditions on u u u and φ \varphi φ for
W u , φ : Q p 1 ⟶ Q p 2 W_{u,\varphi}: Q_{p_1}\longrightarrow Q_{p_2} W u , φ : Q p 1 ⟶ Q p 2
to be bounded or compact. The parameters p 1 , p 2 p_1,p_2 p 1 , p 2 describe the weights in the derivative norms. Throughout, p 1 > 0 p_1>0 p 1 > 0 and 0 < p 2 < 1 0<p_2<1 0 < p 2 < 1 , and the results are stated separately for 0 < p 1 < 1 0<p_1<1 0 < p 1 < 1 , p 1 = 1 p_1=1 p 1 = 1 , and p 1 > 1 p_1>1 p 1 > 1 .
Background and the main question
Weighted composition operators are often studied by means of measure embeddings. If ν \nu ν is a positive measure and q > 0 q>0 q > 0 , define
μ ( B ) = ∫ φ − 1 ( B ) ∣ u ( z ) ∣ q d ν ( z ) ( B ⊆ D Borel ) . \mu(B)=\int_{\varphi^{-1}(B)}|u(z)|^q\,d\nu(z)\qquad(B\subseteq\mathbb{D}\ \text{Borel}). μ ( B ) = ∫ φ − 1 ( B ) ∣ u ( z ) ∣ q d ν ( z ) ( B ⊆ D Borel ) .
Then
∫ D ∣ u ( z ) f ( φ ( z ) ) ∣ q d ν ( z ) = ∫ D ∣ f ( w ) ∣ q d μ ( w ) . \int_{\mathbb{D}}|u(z)f(\varphi(z))|^q\,d\nu(z)=\int_{\mathbb{D}}|f(w)|^q\,d\mu(w). ∫ D ∣ u ( z ) f ( φ ( z )) ∣ q d ν ( z ) = ∫ D ∣ f ( w ) ∣ q d μ ( w ) .
This identity relates the operator to a Carleson measure problem. It is used in the Hardy-space results of Contreras and Hernández-Díaz [5 , 6 ] and the Bergman-space results of Čučković and Zhao [7 ] . Other descriptions of the induced measure include Berezin transforms for large Fock spaces [1 ] and sparse domination forms for weighted Bergman spaces [9 ] .
For spaces defined by derivatives, the product rule gives two terms:
( W u , φ f ) ′ = U f + V f , U f = u ′ ( f ∘ φ ) , V f = u φ ′ ( f ′ ∘ φ ) . (1) (W_{u,\varphi}f)'=Uf+Vf,\qquad Uf=u'(f\circ\varphi),\qquad Vf=u\varphi'(f'\circ\varphi). \tag*{(1)} ( W u , φ f ) ′ = U f + V f , U f = u ′ ( f ∘ φ ) , V f = u φ ′ ( f ′ ∘ φ ) . (1)
Pointwise estimates lead to explicit conditions on Bloch and Bloch-type spaces [17 , 16 ] . Integral derivative norms, however, require both terms to be controlled within the same integral; the same difficulty appears for the Dirichlet space [3 ] . Cheng, Kumar and Zhou [4 ] study weighted composition operators on Dirichlet-type spaces and related Q p Q_p Q p spaces under additional assumptions on the weight. Boundedness, essential norms and compactness on BMOA are studied in [13 , 14 , 15 ] .
The identity Q 1 = B M O A Q_1=\mathrm{BMOA} Q 1 = BMOA relates these questions to the Q p Q_p Q p scale. Here the operator takes values in Q p 2 Q_{p_2} Q p 2 with 0 < p 2 < 1 0<p_2<1 0 < p 2 < 1 . Since Q p 2 ⊆ B M O A Q_{p_2}\subseteq\mathrm{BMOA} Q p 2 ⊆ BMOA , boundedness into BMOA alone does not give the required estimate, and we keep the Q p 2 Q_{p_2} Q p 2 norm in both terms of (1) .
For p > 0 p>0 p > 0 , let
N p = { h ∈ Hol ( D ) : ∥ h ∥ N p 2 = sup a ∈ D ∫ D ∣ h ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p d A ( z ) < ∞ } . \mathcal{N}_p=\left\{h\in\operatorname{Hol}(\mathbb{D}):\|h\|_{\mathcal{N}_p}^2=\sup_{a\in\mathbb{D}}\int_{\mathbb{D}}|h(z)|^2(1-|\varphi_a(z)|^2)^p\,dA(z)<\infty\right\}. N p = { h ∈ Hol ( D ) : ∥ h ∥ N p 2 = a ∈ D sup ∫ D ∣ h ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p d A ( z ) < ∞ } .
These spaces were studied on the disc by Palmberg [18 ] and on the ball by Hu and Khoi [8 ] ; see also [12 ] . With our norms,
∥ f ′ ∥ N p = ∥ f ∥ Q p , ∗ , ∥ ∫ 0 z h ( ζ ) d ζ ∥ Q p = ∥ h ∥ N p . \|f'\|_{\mathcal{N}_p}=\|f\|_{Q_p,*},\qquad\left\|\int_0^z h(\zeta)\,d\zeta\right\|_{Q_p}=\|h\|_{\mathcal{N}_p}. ∥ f ′ ∥ N p = ∥ f ∥ Q p , ∗ , ∫ 0 z h ( ζ ) d ζ Q p = ∥ h ∥ N p .
We therefore study boundedness and compactness of the two maps in (1) from Q p 1 Q_{p_1} Q p 1 to N p 2 \mathcal{N}_{p_2} N p 2 . When u = 1 u=1 u = 1 , only the derivative term remains. Hu and Zhou [11 ] characterize composition operators on Q p Q_p Q p by a dyadic derivative embedding, and Hu, Luo, Xiao and Zhou [10 ] develop the corresponding theory for Q K Q_K Q K spaces. The embedding is treated by different capacities in [2 ] . We use these two embedding theorems as tools; their definitions and the estimates needed here are given in § 2 .
The main question is how to pass from an estimate for U f + V f Uf+Vf U f + V f to separate estimates for its two terms. The triangle inequality gives a bound for the sum from separate bounds, but does not give the converse, because
∣ U f + V f ∣ 2 = ∣ U f ∣ 2 + ∣ V f ∣ 2 + 2 Re ( U f V f ‾ ) . |Uf+Vf|^2=|Uf|^2+|Vf|^2+2\operatorname{Re}(Uf\overline{Vf}). ∣ U f + V f ∣ 2 = ∣ U f ∣ 2 + ∣ V f ∣ 2 + 2 Re ( U f V f ) .
The separation theorem proves
W u , φ : Q p 1 → Q p 2 bounded ⟺ U , V : Q p 1 → N p 2 bounded , W u , φ : Q p 1 → Q p 2 compact ⟺ U , V : Q p 1 → N p 2 compact . \begin{aligned}
W_{u,\varphi}:Q_{p_1}&\to Q_{p_2}\text{ bounded}\quad\Longleftrightarrow\quad U,V:Q_{p_1}\to\mathcal{N}_{p_2}\text{ bounded},\\
W_{u,\varphi}:Q_{p_1}&\to Q_{p_2}\text{ compact}\quad\Longleftrightarrow\quad U,V:Q_{p_1}\to\mathcal{N}_{p_2}\text{ compact}.
\end{aligned} W u , φ : Q p 1 W u , φ : Q p 1 → Q p 2 bounded ⟺ U , V : Q p 1 → N p 2 bounded , → Q p 2 compact ⟺ U , V : Q p 1 → N p 2 compact .
Two analytic functions on each set of a fixed partition give a uniform estimate for their values and derivatives. Averaging their finite sums over independent signs gives the estimate for V V V ; then U f = ( W u , φ f ) ′ − V f Uf=(W_{u,\varphi}f)'-Vf U f = ( W u , φ f ) ′ − V f . The corresponding estimates near the boundary give compactness. No further assumption on φ \varphi φ is needed.
Main results
We first specify the measures and quantities in the statements. For a ∈ D a\in\mathbb{D} a ∈ D , 0 < t < 1 0<t<1 0 < t < 1 , and a Borel set B ⊆ D B\subseteq\mathbb{D} B ⊆ D , put
μ a , t 0 ( B ) = ∫ { ∣ z ∣ < t : φ ( z ) ∈ B } ∣ u ′ ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) . \mu^0_{a,t}(B)=\int_{\{ |z|<t:\,\varphi(z)\in B\}}|u'(z)|^2(1-|\varphi_a(z)|^2)^{p_2}\,dA(z). μ a , t 0 ( B ) = ∫ { ∣ z ∣ < t : φ ( z ) ∈ B } ∣ u ′ ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) .
μ a 1 ( B ) = ∫ φ − 1 ( B ) ∣ u ( z ) φ ′ ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) . \mu^1_a(B)=\int_{\varphi^{-1}(B)}|u(z)\varphi'(z)|^2(1-|\varphi_a(z)|^2)^{p_2}\,dA(z). μ a 1 ( B ) = ∫ φ − 1 ( B ) ∣ u ( z ) φ ′ ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) .
The restriction ∣ z ∣ < t |z|<t ∣ z ∣ < t makes μ a , t 0 \mu^0_{a,t} μ a , t 0 finite; we take the supremum over 0 < t < 1 0<t<1 0 < t < 1 below.
Section 2 defines the value capacities C p , R ( b ) ( ν ; E ) C_{p,\mathcal{R}}^{(b)}(\nu;E) C p , R ( b ) ( ν ; E ) and D p , R ( b ) ( ν ; E ) D_{p,\mathcal{R}}^{(b)}(\nu;E) D p , R ( b ) ( ν ; E ) , the derivative gauge ∥ h ∥ B p 1 2 ( D ) \|h\|_{B^2_{p_1}(\mathcal{D})} ∥ h ∥ B p 1 2 ( D ) , and h ν p ( I ) = ν ( Q I u p ) / ∣ I ∣ p + 2 h^p_\nu(I)=\nu(Q_I^{\mathrm{up}})/|I|^{p+2} h ν p ( I ) = ν ( Q I up ) /∣ I ∣ p + 2 . Here D \mathcal{D} D is a fixed dyadic grid on T \mathbb{T} T , R \mathcal{R} R is the polar partition used for reconstruction, and b = b ( p 1 ) b=b(p_1) b = b ( p 1 ) is fixed as in (5) . For Borel E ⊆ D E\subseteq\mathbb{D} E ⊆ D , set
C 0 p 1 , p 2 ( E ) = sup a ∈ D , 0 < t < 1 C p 1 , R ( b ) ( μ a , t 0 ; E ) . (2) C_0^{p_1,p_2}(E)=\sup_{\substack{a\in\mathbb{D},\,0<t<1}}C_{p_1,\mathcal{R}}^{(b)}(\mu^0_{a,t};E). \tag*{(2)} C 0 p 1 , p 2 ( E ) = a ∈ D , 0 < t < 1 sup C p 1 , R ( b ) ( μ a , t 0 ; E ) . (2)
C 1 p 1 , p 2 ( E ) = sup a ∈ D ∥ h 1 E μ a , 2 1 p 1 ∥ B p 1 2 ( D ) . C_1^{p_1,p_2}(E)=\sup_{a\in\mathbb{D}}\left\|h^{p_1}_{\mathbf{1}_E\mu^1_{a,2}}\right\|_{B^2_{p_1}(\mathcal{D})}. C 1 p 1 , p 2 ( E ) = a ∈ D sup h 1 E μ a , 2 1 p 1 B p 1 2 ( D ) .
The subscripts 0 and 1 correspond to function values and derivatives. Both quantities are defined by the symbols and the explicit coefficient constraints in Section § 2 .
Write E ρ = { w ∈ D : ∣ w ∣ > ρ } E_\rho=\{w\in\mathbb{D}:|w|>\rho\} E ρ = { w ∈ D : ∣ w ∣ > ρ } , and let d m dm d m be normalized arc measure on T \mathbb{T} T . For p 1 = 1 p_1=1 p 1 = 1 , put
M p 2 ( E ) = sup a ∈ D ∫ T sup I ∈ D ξ ∈ I μ a 1 ( E ∩ Q I u p ) ∣ I ∣ 3 d m ( ξ ) . M_{p_2}(E)=\sup_{a\in\mathbb{D}}\int_{\mathbb{T}}\sup_{\substack{I\in\mathcal{D}\\ \xi\in I}}\frac{\mu^1_a(E\cap Q_I^{\mathrm{up}})}{|I|^3}\,dm(\xi). M p 2 ( E ) = a ∈ D sup ∫ T I ∈ D ξ ∈ I sup ∣ I ∣ 3 μ a 1 ( E ∩ Q I up ) d m ( ξ ) .
Here Q I u p Q_I^{\mathrm{up}} Q I up are the upper half tents defined in (3) . For p 1 > 1 p_1>1 p 1 > 1 , put
J p 2 ( E ) = sup a ∈ D ∫ φ − 1 ( E ) ∣ u ( z ) φ ′ ( z ) ∣ 2 ( 1 − ∣ φ ( z ) ∣ 2 ) 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) . J_{p_2}(E)=\sup_{a\in\mathbb{D}}\int_{\varphi^{-1}(E)}\frac{|u(z)\varphi'(z)|^2}{(1-|\varphi(z)|^2)^2}(1-|\varphi_a(z)|^2)^{p_2}\,dA(z). J p 2 ( E ) = a ∈ D sup ∫ φ − 1 ( E ) ( 1 − ∣ φ ( z ) ∣ 2 ) 2 ∣ u ( z ) φ ′ ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) .
All these quantities are allowed to be infinite.
Theorem 1.1. Let 0 < p 1 < 1 0<p_1<1 0 < p 1 < 1 and 0 < p 2 < 1 0<p_2<1 0 < p 2 < 1 . Then W u , φ : Q p 1 → Q p 2 W_{u,\varphi}:Q_{p_1}\to Q_{p_2} W u , φ : Q p 1 → Q p 2 is bounded if and only if
u ∈ Q p 2 , C 0 p 1 , p 2 ( D ) + C 1 p 1 , p 2 ( D ) < ∞ u\in Q_{p_2},\qquad C_0^{p_1,p_2}(\mathbb{D})+C_1^{p_1,p_2}(\mathbb{D})<\infty u ∈ Q p 2 , C 0 p 1 , p 2 ( D ) + C 1 p 1 , p 2 ( D ) < ∞ .
Under boundedness, it is compact if and only if
lim ρ ↑ 1 [ C 0 p 1 , p 2 ( E ρ ) + C 1 p 1 , p 2 ( E ρ ) ] = 0. \lim_{\rho\uparrow1}\left[C_0^{p_1,p_2}(E_\rho)+C_1^{p_1,p_2}(E_\rho)\right]=0. ρ ↑ 1 lim [ C 0 p 1 , p 2 ( E ρ ) + C 1 p 1 , p 2 ( E ρ ) ] = 0.
Theorem 1.2. Let p 1 = 1 p_1=1 p 1 = 1 and 0 < p 2 < 1 0<p_2<1 0 < p 2 < 1 . Then W u , φ : B M O A → Q p 2 W_{u,\varphi}:\mathrm{BMOA}\to Q_{p_2} W u , φ : BMOA → Q p 2 is bounded if and only if
u ∈ Q p 2 , C 0 1 , p 2 ( D ) + M p 2 ( D ) < ∞ . u\in Q_{p_2},\qquad C_0^{1,p_2}(\mathbb{D})+M_{p_2}(\mathbb{D})<\infty. u ∈ Q p 2 , C 0 1 , p 2 ( D ) + M p 2 ( D ) < ∞. Under boundedness, it is compact if and only if
lim ρ ↑ 1 [ C 0 1 , p 2 ( E ρ ) + M p 2 ( E ρ ) ] = 0. \lim_{\rho\uparrow1}\left[C_0^{1,p_2}(E_\rho)+M_{p_2}(E_\rho)\right]=0. ρ ↑ 1 lim [ C 0 1 , p 2 ( E ρ ) + M p 2 ( E ρ ) ] = 0. In fact, for every Borel set E ⊆ D E\subseteq\mathbb{D} E ⊆ D ,
C 1 1 , p 2 ( E ) = M p 2 ( E ) . C_1^{1,p_2}(E)=M_{p_2}(E). C 1 1 , p 2 ( E ) = M p 2 ( E ) .
Theorem 1.3. Let p 1 > 1 p_1>1 p 1 > 1 and 0 < p 2 < 1 0<p_2<1 0 < p 2 < 1 . Then W u , φ : Q p 1 → Q p 2 W_{u,\varphi}:Q_{p_1}\to Q_{p_2} W u , φ : Q p 1 → Q p 2 , equivalently W u , φ : B → Q p 2 W_{u,\varphi}:\mathcal{B}\to Q_{p_2} W u , φ : B → Q p 2 , is bounded if and only if
u ∈ Q p 2 , C 0 2 , p 2 ( D ) + J p 2 ( D ) < ∞ . u\in Q_{p_2},\qquad C_0^{2,p_2}(\mathbb{D})+J_{p_2}(\mathbb{D})<\infty. u ∈ Q p 2 , C 0 2 , p 2 ( D ) + J p 2 ( D ) < ∞. Under boundedness, it is compact if and only if
lim ρ ↑ 1 [ C 0 2 , p 2 ( E ρ ) + J p 2 ( E ρ ) ] = 0. \lim_{\rho\uparrow1}\left[C_0^{2,p_2}(E_\rho)+J_{p_2}(E_\rho)\right]=0. ρ ↑ 1 lim [ C 0 2 , p 2 ( E ρ ) + J p 2 ( E ρ ) ] = 0.
The value capacity is defined with p = 2 p=2 p = 2 , since Q p 1 = Q 2 = B Q_{p_1}=Q_2=\mathcal{B} Q p 1 = Q 2 = B with equivalent norms for p 1 > 1 p_1>1 p 1 > 1 .
The condition u ∈ Q p 2 u\in Q_{p_2} u ∈ Q p 2 follows from W u , φ 1 = u W_{u,\varphi}1=u W u , φ 1 = u . The suprema defining the capacities are taken before the limit ρ ↑ 1 \rho\uparrow1 ρ ↑ 1 . The value capacity C C C may be replaced by its equivalent form D D D in all three statements. For u = 1 u=1 u = 1 , the value term is zero, and the criteria reduce to the derivative embedding condition for composition operators. The proofs also give two-sided norm estimates for W u , φ W_{u,\varphi} W u , φ ; see (31) .
The paper is organized as follows. § 2 collects the capacity definitions and the auxiliary estimates. § 3 proves the separation theorem and the compactness criteria, and then deduces Theorems 1.1–1.3.
We write A ≲ B A\lesssim B A ≲ B if A ≤ C B A\le CB A ≤ C B and A ≍ B A\asymp B A ≍ B if both A ≲ B A\lesssim B A ≲ B and B ≲ A B\lesssim A B ≲ A . Unless otherwise stated, b b b and M M M are chosen as in (5) with p = p 1 p=p_1 p = p 1 .
Preliminaries and lemmas
This section has two purposes: to fix the capacity notation used in the main results and to collect the estimates needed for the separation argument and the compactness criteria. The two terms in (1) lead to the embedding problems
∫ D ∣ f ∣ 2 d ν and ∫ D ∣ f ′ ∣ 2 d ν . \int_{\mathbb{D}} |f|^2\,d\nu\qquad\text{and}\qquad\int_{\mathbb{D}} |f'|^2\,d\nu. ∫ D ∣ f ∣ 2 d ν and ∫ D ∣ f ′ ∣ 2 d ν .
The first uses reconstruction of f − f ( 0 ) f-f(0) f − f ( 0 ) from f ′ f' f ′ , the second the distribution of ν \nu ν on dyadic tents. Throughout this section p > 0 p>0 p > 0 is fixed, and all function-space norms in the lemmas use this same p p p .
The two embedding conditions
Identify T \mathbb{T} T with R / Z \mathbb{R}/\mathbb{Z} R / Z . Let
D = { I n , k : n ≥ 0 , 0 ≤ k < 2 n } , I n , k = { e 2 π i t : k 2 − n ≤ t < ( k + 1 ) 2 − n } . \mathcal{D}=\{I_{n,k}:n\geq0,\ 0\leq k<2^n\},\qquad I_{n,k}=\{e^{2\pi it}:k2^{-n}\leq t<(k+1)2^{-n}\}. D = { I n , k : n ≥ 0 , 0 ≤ k < 2 n } , I n , k = { e 2 π i t : k 2 − n ≤ t < ( k + 1 ) 2 − n } .
Arc length is normalized so that ∣ T ∣ = 1 |\mathbb{T}|=1 ∣ T ∣ = 1 . For any arc I ⊆ T I\subseteq\mathbb{T} I ⊆ T , define
Q I = { r e i θ : e i θ ∈ I , 1 − ∣ I ∣ ≤ r < 1 } , Q I u p = { r e i θ : e i θ ∈ I , 1 − ∣ I ∣ ≤ r < 1 − ∣ I ∣ / 2 } . (3) Q_I=\{re^{i\theta}:e^{i\theta}\in I,\ 1-|I|\leq r<1\},\qquad Q_I^{\mathrm{up}}=\{re^{i\theta}:e^{i\theta}\in I,\ 1-|I|\leq r<1-|I|/2\}. \tag*{(3)} Q I = { r e i θ : e i θ ∈ I , 1 − ∣ I ∣ ≤ r < 1 } , Q I up = { r e i θ : e i θ ∈ I , 1 − ∣ I ∣ ≤ r < 1 − ∣ I ∣/2 } . (3)
The sets Q I u p Q_I^{\mathrm{up}} Q I up , I ∈ D I\in\mathcal{D} I ∈ D , partition D \mathbb{D} D ; see [11 Section 1] .
Given a measurable function F F F , set
∥ F ∥ T p 2 = sup I ⊆ T ∫ Q I ∣ F ( w ) ∣ 2 ( 1 − ∣ w ∣ ∣ I ∣ ) p d A ( w ) . \|F\|_{\mathcal{T}_p}^2=\sup_{I\subseteq\mathbb{T}}\int_{Q_I}|F(w)|^2\left(\frac{1-|w|}{|I|}\right)^p dA(w). ∥ F ∥ T p 2 = I ⊆ T sup ∫ Q I ∣ F ( w ) ∣ 2 ( ∣ I ∣ 1 − ∣ w ∣ ) p d A ( w ) .
The supremum is over all arcs, including T \mathbb{T} T . Restricting it to D \mathcal{D} D gives ∥ F ∥ T p , D \|F\|_{\mathcal{T}_{p,\mathcal{D}}} ∥ F ∥ T p , D . The standard tent-space characterization, recorded in [2 (9)–(11)] , gives
∥ F ∥ T p , D ≍ ∥ F ∥ T p , ∥ f ~ ∥ T p ≍ ∥ f ∥ Q p , ∗ . (4) \|F\|_{\mathcal{T}_{p,\mathcal{D}}}\asymp\|F\|_{\mathcal{T}_p},\qquad\|\widetilde f\|_{\mathcal{T}_p}\asymp\|f\|_{\mathcal{Q}_{p,*}}. \tag*{(4)} ∥ F ∥ T p , D ≍ ∥ F ∥ T p , ∥ f ∥ T p ≍ ∥ f ∥ Q p , ∗ . (4)
Choose integers b b b and M M M once for this p p p , with
b ≥ max { b p , ⌈ p ⌉ + 3 } , M = b + 1. (5) b\geq\max\{b_p,\lceil p\rceil+3\},\qquad M=b+1. \tag*{(5)} b ≥ max { b p , ⌈ p ⌉ + 3 } , M = b + 1. (5)
Choose b p b_p b p so that the projection estimates hold for every b ≥ b p b\geq b_p b ≥ b p ; see [2 Section 2.3] and [10 Lemmas 2.2 and 2.3] . Define
d A b ( w ) = ( b + 1 ) ( 1 − ∣ w ∣ 2 ) b d A ( w ) , k b ( ζ , w ) = ( 1 − ζ ‾ w ) − b − 2 . dA_b(w)=(b+1)(1-|w|^2)^b dA(w),\qquad k_b(\zeta,w)=(1-\overline{\zeta}w)^{-b-2}. d A b ( w ) = ( b + 1 ) ( 1 − ∣ w ∣ 2 ) b d A ( w ) , k b ( ζ , w ) = ( 1 − ζ w ) − b − 2 .
The kernel used to reconstruct function values is
L b ( ζ , w ) = ∫ 0 ζ k b ( ξ , w ) d ξ = { ( 1 − ζ w ‾ ) − b − 1 − 1 ( b + 1 ) w ‾ , w ≠ 0 , ζ , w = 0. L_b(\zeta,w)=\int_0^\zeta k_b(\xi,w)\,d\xi=
\begin{cases}
\dfrac{(1-\zeta\overline{w})^{-b-1}-1}{(b+1)\overline{w}}, & w\ne0,\\
\zeta, & w=0.
\end{cases} L b ( ζ , w ) = ∫ 0 ζ k b ( ξ , w ) d ξ = ⎩ ⎨ ⎧ ( b + 1 ) w ( 1 − ζ w ) − b − 1 − 1 , ζ , w = 0 , w = 0.
Now set
P b , 0 F ( ζ ) = ∫ D L b ( ζ , w ) F ( w ) d A b ( w ) . P_{b,0}F(\zeta)=\int_{\mathbb{D}}L_b(\zeta,w)F(w)\,dA_b(w). P b , 0 F ( ζ ) = ∫ D L b ( ζ , w ) F ( w ) d A b ( w ) .
By [2 Lemma 2.12 and Proposition 2.13] ,
∥ P b , 0 F ∥ Q p ≤ C p , b ∥ F ∥ T p , P b , 0 ( f ′ ~ ) = f − f ( 0 ) . (6) \|P_{b,0}F\|_{\mathcal{Q}_p}\leq C_{p,b}\|F\|_{\mathcal{T}_p},\qquad P_{b,0}(\widetilde{f'})=f-f(0). \tag*{(6)} ∥ P b , 0 F ∥ Q p ≤ C p , b ∥ F ∥ T p , P b , 0 ( f ′ ) = f − f ( 0 ) . (6)
For the value capacity, use the polar partitions from [2 Section 2.2] :
R N = { R N , j , k : 0 ≤ j , k < 2 N } , N ≥ 0 , \mathcal{R}_N=\{R_{N,j,k}:0\leq j,k<2^N\},\qquad N\geq0, R N = { R N , j , k : 0 ≤ j , k < 2 N } , N ≥ 0 ,
where
R N , j , k = { r e i θ : j 2 − N ≤ r 2 < ( j + 1 ) 2 − N , 2 π k 2 − N ≤ θ < 2 π ( k + 1 ) 2 − N } . R_{N,j,k}=\{re^{i\theta}:j2^{-N}\leq r^2<(j+1)2^{-N},\ 2\pi k2^{-N}\leq\theta<2\pi(k+1)2^{-N}\}. R N , j , k = { r e i θ : j 2 − N ≤ r 2 < ( j + 1 ) 2 − N , 2 π k 2 − N ≤ θ < 2 π ( k + 1 ) 2 − N } .
The origin is assigned to R N , 0 , 0 R_{N,0,0} R N , 0 , 0 , and the sets are ordered by their indices. Write R = { R N } N ≥ 0 \mathcal{R}=\{\mathcal{R}_N\}_{N\geq0} R = { R N } N ≥ 0 . For a coefficient vector c = ( c R ) R ∈ R N c=(c_R)_{R\in\mathcal{R}_N} c = ( c R ) R ∈ R N , define
∥ c ∥ X p , N 2 = ∥ ∑ R ∈ R N c R 1 R ∥ T p 2 = sup I ⊆ T c ∗ A I , N ( p ) c , \|c\|_{X_{p,N}}^2=\left\|\sum_{R\in\mathcal{R}_N}c_R\mathbf{1}_R\right\|_{\mathcal{T}_p}^2=\sup_{I\subseteq\mathbb{T}}c^*A_{I,N}^{(p)}c, ∥ c ∥ X p , N 2 = R ∈ R N ∑ c R 1 R T p 2 = I ⊆ T sup c ∗ A I , N ( p ) c ,
with diagonal matrix
A I , N ( p ) = diag ( ∫ R ∩ Q I ( 1 − ∣ w ∣ ∣ I ∣ ) p d A ( w ) : R ∈ R N ) . A_{I,N}^{(p)}=\operatorname{diag}\left(\int_{R\cap Q_I}\left(\frac{1-|w|}{|I|}\right)^p dA(w):R\in\mathcal{R}_N\right). A I , N ( p ) = diag ( ∫ R ∩ Q I ( ∣ I ∣ 1 − ∣ w ∣ ) p d A ( w ) : R ∈ R N ) .
Accordingly, c c c is admissible exactly when ∥ c ∥ X p , N ≤ 1 \lVert c\rVert_{X_{p,N}}\le1 ∥ c ∥ X p , N ≤ 1 .
Let ν \nu ν be a finite positive Borel measure and E ⊆ D E\subseteq\mathbb{D} E ⊆ D a Borel set. Define the holomorphic functions
Ψ R , E ( b ) ( ζ ) = ∫ R ∩ E L b ( ζ , w ) d A b ( w ) . \Psi_{R,E}^{(b)}(\zeta)=\int_{R\cap E}L_b(\zeta,w)\,dA_b(w). Ψ R , E ( b ) ( ζ ) = ∫ R ∩ E L b ( ζ , w ) d A b ( w ) .
and the associated Gram matrix
Γ ν , N ( b ) ( E ) ( R , S ) = ∫ D Ψ R , E ( b ) ( ζ ) ‾ Ψ S , E ( b ) ( ζ ) d ν ( ζ ) . \Gamma_{\nu,N}^{(b)}(E)(R,S)=\int_{\mathbb{D}}\overline{\Psi_{R,E}^{(b)}(\zeta)}\Psi_{S,E}^{(b)}(\zeta)\,d\nu(\zeta). Γ ν , N ( b ) ( E ) ( R , S ) = ∫ D Ψ R , E ( b ) ( ζ ) Ψ S , E ( b ) ( ζ ) d ν ( ζ ) .
In particular,
c ∗ Γ ν , N ( b ) ( E ) c = ∫ D ∣ ∑ R c R Ψ R , E ( b ) ∣ 2 d ν . c^*\Gamma_{\nu,N}^{(b)}(E)c=\int_{\mathbb{D}}\left|\sum_R c_R\Psi_{R,E}^{(b)}\right|^2\,d\nu. c ∗ Γ ν , N ( b ) ( E ) c = ∫ D R ∑ c R Ψ R , E ( b ) 2 d ν .
We employ the following localized versions of the capacities from [2 Definitions 3.1 and 3.3] :
C p , R ( b ) ( ν ; E ) = sup N ≥ 0 sup ∥ c ∥ X p , N ≤ 1 c ∗ Γ ν , N ( b ) ( E ) c . C_{p,\mathcal{R}}^{(b)}(\nu;E)=\sup_{N\ge0}\sup_{\lVert c\rVert_{X_{p,N}}\le1}c^*\Gamma_{\nu,N}^{(b)}(E)c. C p , R ( b ) ( ν ; E ) = N ≥ 0 sup ∥ c ∥ X p , N ≤ 1 sup c ∗ Γ ν , N ( b ) ( E ) c .
and
D p , R ( b ) ( ν ; E ) = sup N ≥ 0 inf { ∑ j = 1 m λ j : Γ ν , N ( b ) ( E ) ⪯ ∑ j = 1 m λ j A I j , N ( p ) } . D_{p,\mathcal{R}}^{(b)}(\nu;E)=\sup_{N\ge0}\inf\left\{\sum_{j=1}^{m}\lambda_j:\Gamma_{\nu,N}^{(b)}(E)\preceq\sum_{j=1}^{m}\lambda_j A_{I_j,N}^{(p)}\right\}. D p , R ( b ) ( ν ; E ) = N ≥ 0 sup inf { j = 1 ∑ m λ j : Γ ν , N ( b ) ( E ) ⪯ j = 1 ∑ m λ j A I j , N ( p ) } .
Here m ∈ N m\in\mathbb{N} m ∈ N , λ j ≥ 0 \lambda_j\ge0 λ j ≥ 0 , and each I j ⊆ T I_j\subseteq\mathbb{T} I j ⊆ T is an arc. The ordering X ⪯ Y X\preceq Y X ⪯ Y means v ∗ ( Y − X ) v ≥ 0 v^*(Y-X)v\ge0 v ∗ ( Y − X ) v ≥ 0 for every complex vector v v v . The infimum over an empty collection is taken to be + ∞ +\infty + ∞ . For E = D E=\mathbb{D} E = D , these are the cited capacities, and we omit E E E . The restriction to a Borel set E E E is used for compactness.
The value embedding theorem, assembled from [2 Theorems 3.2, 3.5 and 3.6] , asserts
sup f ( 0 ) = 0 ∥ f ∥ Q p , ∗ ≤ 1 ∫ D ∣ f ∣ 2 d ν ≍ C p , R ( b ) ( ν ) ≍ D p , R ( b ) ( ν ) . \sup_{\substack{f(0)=0\\\lVert f\rVert_{Q_{p,*}}\le1}}\int_{\mathbb{D}}|f|^2\,d\nu\asymp C_{p,\mathcal{R}}^{(b)}(\nu)\asymp D_{p,\mathcal{R}}^{(b)}(\nu). f ( 0 ) = 0 ∥ f ∥ Q p , ∗ ≤ 1 sup ∫ D ∣ f ∣ 2 d ν ≍ C p , R ( b ) ( ν ) ≍ D p , R ( b ) ( ν ) .
The form adapted for compactness is
sup ∥ F ∥ T p ≤ 1 ∫ D ∣ P b , 0 ( 1 E F ) ∣ 2 d ν ≍ C p , R ( b ) ( ν ; E ) ≍ D p , R ( b ) ( ν ; E ) , (7) \sup_{\lVert F\rVert_{T_p}\le1}\int_{\mathbb{D}}|P_{b,0}(1_EF)|^2\,d\nu\asymp C_{p,\mathcal{R}}^{(b)}(\nu;E)\asymp D_{p,\mathcal{R}}^{(b)}(\nu;E), \tag*{(7)} ∥ F ∥ T p ≤ 1 sup ∫ D ∣ P b , 0 ( 1 E F ) ∣ 2 d ν ≍ C p , R ( b ) ( ν ; E ) ≍ D p , R ( b ) ( ν ; E ) , (7)
with comparison constants independent of E E E and ν \nu ν . For the first comparison, use the conditional expectations
E N ( b ) F = ∑ R ∈ R N ∫ R F d A b A b ( R ) 1 R , A b ( R ) = ∫ R d A b . E_N^{(b)}F=\sum_{R\in\mathcal{R}_N}\frac{\int_R F\,dA_b}{A_b(R)}1_R,\qquad A_b(R)=\int_RdA_b. E N ( b ) F = R ∈ R N ∑ A b ( R ) ∫ R F d A b 1 R , A b ( R ) = ∫ R d A b .
By [2 Lemma 2.8 and the proof of Theorem 3.2] ,
∥ E N ( b ) F ∥ T p ≤ C p , b ∥ F ∥ T p , ∥ 1 E ( E N ( b ) F − F ) ∥ L 1 ( d A b ) ⟶ 0. \lVert E_N^{(b)}F\rVert_{T_p}\le C_{p,b}\lVert F\rVert_{T_p},\qquad\lVert1_E(E_N^{(b)}F-F)\rVert_{L^1(dA_b)}\longrightarrow0. ∥ E N ( b ) F ∥ T p ≤ C p , b ∥ F ∥ T p , ∥ 1 E ( E N ( b ) F − F ) ∥ L 1 ( d A b ) ⟶ 0.
For fixed ζ \zeta ζ , the kernel L b ( ζ , ⋅ ) L_b(\zeta,\cdot) L b ( ζ , ⋅ ) is bounded. Fatou’s lemma therefore gives
∫ D ∣ P b , 0 ( 1 E F ) ∣ 2 d ν ≤ C p C p , R ( b ) ( ν ; E ) ∥ F ∥ T p 2 . \int_{\mathbb{D}}|P_{b,0}(1_EF)|^2\,d\nu\le C_pC_{p,\mathcal{R}}^{(b)}(\nu;E)\lVert F\rVert_{T_p}^2. ∫ D ∣ P b , 0 ( 1 E F ) ∣ 2 d ν ≤ C p C p , R ( b ) ( ν ; E ) ∥ F ∥ T p 2 .
The reverse inequality follows by taking F = ∑ R c R 1 R F=\sum_Rc_R1_R F = ∑ R c R 1 R . The second equivalence is a consequence of the proof of [2 Theorem 3.5] : its matrix argument applies to every positive semidefinite matrix with the same diagonal constraints. Apply it to Γ ν , N ( b ) ( E ) \Gamma_{\nu,N}^{(b)}(E) Γ ν , N ( b ) ( E ) at each level N N N , and then take the supremum over N N N . The constants do not depend on E E E .
For the derivative embedding, we use the capacity gauge from [10 (1.2)] . Given h : D → [ 0 , ∞ ] h:\mathcal{D}\to[0,\infty] h : D → [ 0 , ∞ ] , define
∥ h ∥ B p 2 ( D ) = sup d ∑ I ∈ D d I h ( I ) , \lVert h\rVert_{B_p^2(\mathcal{D})}=\sup_{d}\sum_{I\in\mathcal{D}}d_Ih(I), ∥ h ∥ B p 2 ( D ) = d sup I ∈ D ∑ d I h ( I ) ,
where the admissible weights satisfy
d I ≥ 0 , ∑ I ⊂ J d I ≤ ∣ J ∣ p ( J ∈ D ) . (8) d_I\ge0,\qquad\sum_{I\subset J}d_I\le|J|^p\quad(J\in\mathcal{D}). \tag*{(8)} d I ≥ 0 , I ⊂ J ∑ d I ≤ ∣ J ∣ p ( J ∈ D ) . (8)
All sums are over nonnegative terms, with the convention 0 ⋅ ∞ = 0 0\cdot\infty=0 0 ⋅ ∞ = 0 . No square root is taken in this definition. For a positive Borel measure ν \nu ν , set
h ν , 2 p ( I ) = ν ( Q I u p ) ∣ I ∣ p + 2 . (9) h_{\nu,2}^p(I)=\frac{\nu(Q_I^{\mathrm{up}})}{|I|^{p+2}}. \tag*{(9)} h ν , 2 p ( I ) = ∣ I ∣ p + 2 ν ( Q I up ) . (9)
The derivative embedding theorem gives
sup ∥ f ∥ Q p , ∗ ≤ 1 ∫ D ∣ f ′ ∣ 2 d ν ≍ ∥ h ν , 2 p ∥ B p 2 ( D ) . (10) \sup_{\lVert f\rVert_{Q_p,*}\le1}\int_{\mathbb{D}}|f'|^2\,d\nu\asymp\lVert h_{\nu,2}^p\rVert_{B_p^2(\mathcal{D})}. \tag*{(10)} ∥ f ∥ Q p , ∗ ≤ 1 sup ∫ D ∣ f ′ ∣ 2 d ν ≍ ∥ h ν , 2 p ∥ B p 2 ( D ) . (10)
For absolutely continuous measures this is [10 Theorem 1.1] , with exponent two. The proof after Lemma 2.3 gives the same result for positive Borel measures. The constants are unchanged under restriction of ν \nu ν .
The two restrictions have different meanings:
C p , R ( b ) ( ν ; E ) : F ↦ P b , 0 ( 1 E F ) , h 1 E ν , 2 p : ν ↦ 1 E ν . C_{p,R}^{(b)}(\nu;E):\ F\mapsto P_{b,0}(\mathbf{1}_E F),\qquad h_{1_E\nu,2}^p:\ \nu\mapsto\mathbf{1}_E\nu. C p , R ( b ) ( ν ; E ) : F ↦ P b , 0 ( 1 E F ) , h 1 E ν , 2 p : ν ↦ 1 E ν .
Local estimates and finite sums
We introduce the positive integral operator
B b + F ( ζ ) = ∫ D ∣ F ( w ) ∣ ∣ 1 − ζ ‾ w ∣ b + 2 d A b ( w ) . B_b^+F(\zeta)=\int_{\mathbb{D}}\frac{|F(w)|}{|1-\overline{\zeta}w|^{b+2}}\,dA_b(w). B b + F ( ζ ) = ∫ D ∣1 − ζ w ∣ b + 2 ∣ F ( w ) ∣ d A b ( w ) .
For b b b as in (5), the absolute-kernel estimate in [10 Lemma 2.2 and the proof of Lemma 2.3] gives
∥ B b + F ∥ T p ≤ C p , b ∥ F ∥ T p . (11) \lVert B_b^+F\rVert_{T_p}\le C_{p,b}\lVert F\rVert_{T_p}. \tag*{(11)} ∥ B b + F ∥ T p ≤ C p , b ∥ F ∥ T p . (11)
Since b > p b>p b > p and A b ( D ) = 1 A_b(\mathbb{D})=1 A b ( D ) = 1 , we further obtain
∥ F ∥ L 1 ( d A b ) ≤ ∥ F ∥ L 2 ( d A b ) ≤ C p , b ∥ F ∥ T p , (12) \lVert F\rVert_{L^1(dA_b)}\le\lVert F\rVert_{L^2(dA_b)}\le C_{p,b}\lVert F\rVert_{T_p}, \tag*{(12)} ∥ F ∥ L 1 ( d A b ) ≤ ∥ F ∥ L 2 ( d A b ) ≤ C p , b ∥ F ∥ T p , (12)
along with the standard growth estimates recorded in [19 , 20 ]
( 1 − ∣ z ∣ 2 ) ∣ f ′ ( z ) ∣ ≤ C p ∥ f ∥ Q p , ∗ , ∣ f ( z ) ∣ ≤ C p log e 1 − ∣ z ∣ 2 ∥ f ∥ Q p . (13) (1-|z|^2)|f'(z)|\le C_p\lVert f\rVert_{Q_{p,*}},\qquad|f(z)|\le C_p\log\frac{e}{1-|z|^2}\lVert f\rVert_{Q_p}. \tag*{(13)} ( 1 − ∣ z ∣ 2 ) ∣ f ′ ( z ) ∣ ≤ C p ∥ f ∥ Q p , ∗ , ∣ f ( z ) ∣ ≤ C p log 1 − ∣ z ∣ 2 e ∥ f ∥ Q p . (13)
The norm of the point-evaluation functional satisfies
∥ ev z ∥ Q p ∗ ≍ p log e 1 − ∣ z ∣ 2 . \lVert\operatorname{ev}_z\rVert_{Q_p^*}\asymp_p\log\frac{e}{1-|z|^2}. ∥ ev z ∥ Q p ∗ ≍ p log 1 − ∣ z ∣ 2 e .
For the lower bound, use the logarithm g ( w ) = log ( 1 / ( 1 − w ) ) g(w)=\log(1/(1-w)) g ( w ) = log ( 1/ ( 1 − w )) , normalized so that g ( 0 ) = 0 g(0)=0 g ( 0 ) = 0 . Given an arc I I I with length h = ∣ I ∣ h=|I| h = ∣ I ∣ ,
∫ Q I ( 1 − ∣ w ∣ ) p ∣ 1 − w ∣ 2 d A ( w ) ≤ C ∫ 1 − h 1 ( 1 − r ) p ( ∫ − π π d θ ∣ 1 − r e i θ ∣ 2 ) r d r ≤ C ∫ 1 − h 1 ( 1 − r ) p − 1 d r = C p h p . \int_{Q_I}\frac{(1-|w|)^p}{|1-w|^2}\,dA(w)\le C\int_{1-h}^{1}(1-r)^p\left(\int_{-\pi}^{\pi}\frac{d\theta}{|1-re^{i\theta}|^2}\right)r\,dr\le C\int_{1-h}^{1}(1-r)^{p-1}\,dr=\frac{C}{p}h^p. ∫ Q I ∣1 − w ∣ 2 ( 1 − ∣ w ∣ ) p d A ( w ) ≤ C ∫ 1 − h 1 ( 1 − r ) p ( ∫ − π π ∣1 − r e i θ ∣ 2 d θ ) r d r ≤ C ∫ 1 − h 1 ( 1 − r ) p − 1 d r = p C h p .
The angular integral equals 2 π / ( 1 − r 2 ) 2\pi/(1-r^2) 2 π / ( 1 − r 2 ) . Applying (4) , we obtain ∥ g ∥ Q p , ∗ ≤ C p \|g\|_{Q_p,*} \le C_p ∥ g ∥ Q p , ∗ ≤ C p . For any nonzero z ∈ D z \in\mathbb{D} z ∈ D , define
g z ( w ) = g ( z ‾ ∣ z ∣ φ z ( w ) ) − g ( ∣ z ∣ ) . g_z(w)=g\left(\frac{\overline{z}}{|z|}\varphi_z(w)\right)-g(|z|). g z ( w ) = g ( ∣ z ∣ z φ z ( w ) ) − g ( ∣ z ∣ ) .
Möbius invariance then implies
g z ( 0 ) = 0 , ∥ g z ∥ Q p ≤ C p , ∣ g z ( z ) ∣ = log 1 1 − ∣ z ∣ . g_z(0)=0,\qquad\|g_z\|_{Q_p}\le C_p,\qquad|g_z(z)|=\log\frac{1}{1-|z|}. g z ( 0 ) = 0 , ∥ g z ∥ Q p ≤ C p , ∣ g z ( z ) ∣ = log 1 − ∣ z ∣ 1 .
Combining this family with the constant function 1 1 1 , we obtain
∥ ev z ∥ Q p ∗ ≥ max { 1 , C p − 1 log 1 1 − ∣ z ∣ } ≳ p log e 1 − ∣ z ∣ 2 . \|\operatorname{ev}_z\|_{Q_p^*}\ge\max\left\{1,C_p^{-1}\log\frac{1}{1-|z|}\right\}\gtrsim_p \log\frac{e}{1-|z|^2}. ∥ ev z ∥ Q p ∗ ≥ max { 1 , C p − 1 log 1 − ∣ z ∣ 1 } ≳ p log 1 − ∣ z ∣ 2 e .
The case z = 0 z=0 z = 0 is immediate, and (13) gives the upper bound.
For each integer j ≥ 0 j\ge0 j ≥ 0 and 0 ≤ l < 2 j + 3 0\le l<2^{j+3} 0 ≤ l < 2 j + 3 , define
V j , l = { r e i θ : 1 − 2 − j ≤ r < 1 − 2 − j − 1 , 2 π l 2 j + 3 ≤ θ < 2 π ( l + 1 ) 2 j + 3 } , V_{j,l}=\left\{re^{i\theta}:1-2^{-j}\le r<1-2^{-j-1},\ \frac{2\pi l}{2^{j+3}}\le\theta<\frac{2\pi(l+1)}{2^{j+3}}\right\}, V j , l = { r e i θ : 1 − 2 − j ≤ r < 1 − 2 − j − 1 , 2 j + 3 2 π l ≤ θ < 2 j + 3 2 π ( l + 1 ) } ,
a j , l = ( 1 − 3 4 ⋅ 2 − j ) exp ( 2 π i ( l + 1 / 2 ) 2 j + 3 ) . a_{j,l}=\left(1-\frac{3}{4}\cdot2^{-j}\right)\exp\left(\frac{2\pi i(l+1/2)}{2^{j+3}}\right). a j , l = ( 1 − 4 3 ⋅ 2 − j ) exp ( 2 j + 3 2 π i ( l + 1/2 ) ) .
Assign the origin to V 0 , 0 V_{0,0} V 0 , 0 and use half-open boundaries, so that the sets form a Borel partition of D \mathbb{D} D . Write λ = ( j , l ) \lambda=(j,l) λ = ( j , l ) , V λ = V j , l V_\lambda=V_{j,l} V λ = V j , l , and a λ = a j , l a_\lambda=a_{j,l} a λ = a j , l .
The following bounds will be used for this partition.
Lemma 2.1. Let λ = ( j , l ) \lambda=(j,l) λ = ( j , l ) . For every w ∈ V j , l w\in V_{j,l} w ∈ V j , l ,
15 16 2 − j ≤ 1 − ∣ a j , l ∣ 2 ≤ 3 2 2 − j , 2 − 2 j 32 ≤ A ( V j , l ) ≤ 2 − 2 j 8 , (14) \frac{15}{16}2^{-j}\le1-|a_{j,l}|^2\le\frac{3}{2}2^{-j},\qquad\frac{2^{-2j}}{32}\le A(V_{j,l})\le\frac{2^{-2j}}{8}, \tag*{(14)} 16 15 2 − j ≤ 1 − ∣ a j , l ∣ 2 ≤ 2 3 2 − j , 32 2 − 2 j ≤ A ( V j , l ) ≤ 8 2 − 2 j , (14) 1 2 ( 1 − ∣ a λ ∣ 2 ) ≤ ∣ 1 − a λ ‾ w ∣ ≤ 4 ( 1 − ∣ a λ ∣ 2 ) , ∣ a λ ∣ ≥ 1 4 . \frac{1}{2}(1-|a_\lambda|^2)\le|1-\overline{a_\lambda}w|\le4(1-|a_\lambda|^2),\qquad|a_\lambda|\ge\frac{1}{4}. 2 1 ( 1 − ∣ a λ ∣ 2 ) ≤ ∣1 − a λ w ∣ ≤ 4 ( 1 − ∣ a λ ∣ 2 ) , ∣ a λ ∣ ≥ 4 1 . For arbitrary ζ ∈ D \zeta\in\mathbb{D} ζ ∈ D ,
1 3 ∣ 1 − a λ ‾ ζ ∣ ≤ ∣ 1 − w ‾ ζ ∣ ≤ 2 ∣ 1 − a λ ‾ ζ ∣ . \frac{1}{3}|1-\overline{a_\lambda}\zeta|\le|1-\overline{w}\zeta|\le2|1-\overline{a_\lambda}\zeta|. 3 1 ∣1 − a λ ζ ∣ ≤ ∣1 − w ζ ∣ ≤ 2∣1 − a λ ζ ∣. If we set
V λ ∗ = { z : dist ( z , V λ ) < ( 1 − ∣ a λ ∣ 2 ) / 16 } , V_\lambda^*=\{z:\operatorname{dist}(z,V_\lambda)<(1-|a_\lambda|^2)/16\}, V λ ∗ = { z : dist ( z , V λ ) < ( 1 − ∣ a λ ∣ 2 ) /16 } , then V λ ∗ ⊂ D V_\lambda^*\subset\mathbb{D} V λ ∗ ⊂ D , the overlap multiplicity of the enlarged sets is bounded by an absolute constant, and 1 − ∣ z ∣ ≍ 1 − ∣ a λ ∣ 2 1-|z|\asymp1-|a_\lambda|^2 1 − ∣ z ∣ ≍ 1 − ∣ a λ ∣ 2 uniformly on V λ ∗ V_\lambda^* V λ ∗ .
Proof. Take w = r e i θ ∈ V j , l w=re^{i\theta}\in V_{j,l} w = r e i θ ∈ V j , l , with θ \theta θ lying in the corresponding angular sector. Direct calculation gives
1 − ∣ a j , l ∣ 2 = 3 4 2 − j ( 2 − 3 4 2 − j ) , 1-|a_{j,l}|^2=\frac{3}{4}2^{-j}\left(2-\frac{3}{4}2^{-j}\right), 1 − ∣ a j , l ∣ 2 = 4 3 2 − j ( 2 − 4 3 2 − j ) ,
A ( V j , l ) = 2 − j − 3 [ ( 1 − 2 − j − 1 ) 2 − ( 1 − 2 − j ) 2 ] = 2 − 2 j 8 ( 1 − 3 4 2 − j ) , A(V_{j,l})=2^{-j-3}\left[(1-2^{-j-1})^2-(1-2^{-j})^2\right]=\frac{2^{-2j}}{8}\left(1-\frac{3}{4}2^{-j}\right), A ( V j , l ) = 2 − j − 3 [ ( 1 − 2 − j − 1 ) 2 − ( 1 − 2 − j ) 2 ] = 8 2 − 2 j ( 1 − 4 3 2 − j ) ,
∣ w − a j , l ∣ ≤ ∣ ∣ w ∣ − ∣ a j , l ∣ ∣ + ∣ θ − arg a j , l ∣ ≤ ( 1 4 + π 8 ) 2 − j . |w-a_{j,l}|\le||w|-|a_{j,l}||+|\theta-\arg a_{j,l}|\le\left(\frac{1}{4}+\frac{\pi}{8}\right)2^{-j}. ∣ w − a j , l ∣ ≤ ∣∣ w ∣ − ∣ a j , l ∣∣ + ∣ θ − arg a j , l ∣ ≤ ( 4 1 + 8 π ) 2 − j .
Angular differences are measured inside the sector defining V j , l V_{j,l} V j , l . The estimates (14) follow immediately, and
∣ 1 − a ‾ j , l w ∣ ≥ 1 − ∣ a j , l ∣ = 3 4 2 − j ≥ 1 2 ( 1 − ∣ a j , l ∣ 2 ) , |1-\overline{a}_{j,l}w| \ge1-|a_{j,l}|=\frac{3}{4}2^{-j}\ge\frac{1}{2}(1-|a_{j,l}|^2), ∣1 − a j , l w ∣ ≥ 1 − ∣ a j , l ∣ = 4 3 2 − j ≥ 2 1 ( 1 − ∣ a j , l ∣ 2 ) ,
∣ 1 − a ‾ j , l w ∣ ≤ 1 − ∣ a j , l ∣ 2 + ∣ a j , l − w ∣ < 3 ⋅ 2 − j < 4 ( 1 − ∣ a j , l ∣ 2 ) . |1-\overline{a}_{j,l}w|\le1-|a_{j,l}|^2+|a_{j,l}-w|<3\cdot2^{-j}<4(1-|a_{j,l}|^2). ∣1 − a j , l w ∣ ≤ 1 − ∣ a j , l ∣ 2 + ∣ a j , l − w ∣ < 3 ⋅ 2 − j < 4 ( 1 − ∣ a j , l ∣ 2 ) .
Moreover, ∣ a j , l ∣ ≥ 1 / 4 |a_{j,l}|\ge1/4 ∣ a j , l ∣ ≥ 1/4 . For any ζ ∈ D \zeta\in\mathbb{D} ζ ∈ D ,
∣ 1 − w ‾ ζ ∣ ≥ 2 − j − 1 , ∣ 1 − a ‾ j , l ζ ∣ ≥ 3 4 2 − j , |1-\overline{w}\zeta|\ge2^{-j-1},\qquad|1-\overline{a}_{j,l}\zeta|\ge\frac{3}{4}2^{-j}, ∣1 − w ζ ∣ ≥ 2 − j − 1 , ∣1 − a j , l ζ ∣ ≥ 4 3 2 − j ,
∣ ∣ 1 − w ‾ ζ ∣ − ∣ 1 − a ‾ j , l ζ ∣ ∣ ≤ ( 1 4 + π 8 ) 2 − j . \bigl||1-\overline{w}\zeta|-|1-\overline{a}_{j,l}\zeta|\bigr|\le\left(\frac{1}{4}+\frac{\pi}{8}\right)2^{-j}. ∣1 − w ζ ∣ − ∣1 − a j , l ζ ∣ ≤ ( 4 1 + 8 π ) 2 − j .
We then obtain
∣ 1 − w ‾ ζ ∣ ≤ [ 1 + 4 3 ( 1 4 + π 8 ) ] ∣ 1 − a ‾ j , l ζ ∣ < 2 ∣ 1 − a ‾ j , l ζ ∣ , |1-\overline{w}\zeta|\le\left[1+\frac{4}{3}\left(\frac{1}{4}+\frac{\pi}{8}\right)\right]|1-\overline{a}_{j,l}\zeta|<2|1-\overline{a}_{j,l}\zeta|, ∣1 − w ζ ∣ ≤ [ 1 + 3 4 ( 4 1 + 8 π ) ] ∣1 − a j , l ζ ∣ < 2∣1 − a j , l ζ ∣ ,
∣ 1 − a ‾ j , l ζ ∣ ≤ [ 1 + 2 ( 1 4 + π 8 ) ] ∣ 1 − w ‾ ζ ∣ < 3 ∣ 1 − w ‾ ζ ∣ . |1-\overline{a}_{j,l}\zeta|\le\left[1+2\left(\frac{1}{4}+\frac{\pi}{8}\right)\right]|1-\overline{w}\zeta|<3|1-\overline{w}\zeta|. ∣1 − a j , l ζ ∣ ≤ [ 1 + 2 ( 4 1 + 8 π ) ] ∣1 − w ζ ∣ < 3∣1 − w ζ ∣.
For z ∈ V j , l ∗ z\in V^*_{j,l} z ∈ V j , l ∗ , the dilation radius is bounded above by 3 ⋅ 2 − j / 32 3\cdot2^{-j}/32 3 ⋅ 2 − j /32 , so
13 32 2 − j < 1 − ∣ z ∣ < 35 32 2 − j . \frac{13}{32}2^{-j}<1-|z|<\frac{35}{32}2^{-j}. 32 13 2 − j < 1 − ∣ z ∣ < 32 35 2 − j .
For a fixed point z z z , this restricts j j j to a finite set of values. At each admissible scale, the angular extension has length at most C 2 − j C2^{-j} C 2 − j , while the original sectors have angular width 2 π 2 − j − 3 2\pi2^{-j-3} 2 π 2 − j − 3 . Consequently
# { j : ∃ l , z ∈ V j , l ∗ } ≤ C , sup j # { l : z ∈ V j , l ∗ } ≤ C , ∑ j , l 1 V j , l ∗ ( z ) ≤ C . \#\{j:\exists l,\ z\in V^*_{j,l}\}\le C,\qquad\sup_j\#\{l:z\in V^*_{j,l}\}\le C,\qquad\sum_{j,l}\mathbf{1}_{V^*_{j,l}}(z)\le C. # { j : ∃ l , z ∈ V j , l ∗ } ≤ C , j sup # { l : z ∈ V j , l ∗ } ≤ C , j , l ∑ 1 V j , l ∗ ( z ) ≤ C .
The level j = 0 j=0 j = 0 has only eight sectors. This proves all assertions. □
Given a coefficient sequence c = ( c λ ) c=(c_\lambda) c = ( c λ ) , define
F c ( w ) = ∑ λ c λ ( 1 − ∣ a λ ∣ 2 ) 1 V λ ( w ) , ∥ c ∥ t p = ∥ F c ∥ T p . (15) F_c(w)=\sum_\lambda\frac{c_\lambda}{(1-|a_\lambda|^2)}\mathbf{1}_{V_\lambda}(w),\qquad\|c\|_{t_p}=\|F_c\|_{T_p}. \tag*{(15)} F c ( w ) = λ ∑ ( 1 − ∣ a λ ∣ 2 ) c λ 1 V λ ( w ) , ∥ c ∥ t p = ∥ F c ∥ T p . (15)
This norm depends solely on the moduli ∣ c λ ∣ |c_\lambda| ∣ c λ ∣ and does not increase upon removing coordinates from the sequence. We also have
sup λ ∣ c λ ∣ ≤ C p ∥ c ∥ t p . (16) \sup_\lambda|c_\lambda|\le C_p\|c\|_{t_p}. \tag*{(16)} λ sup ∣ c λ ∣ ≤ C p ∥ c ∥ t p . (16)
To justify (16) , take an arc containing the angular projection of V λ V_\lambda V λ , with length min { 8 ( 1 − ∣ a λ ∣ 2 ) , 1 } \min\{8(1-|a_\lambda|^2),1\} min { 8 ( 1 − ∣ a λ ∣ 2 ) , 1 } . Its tent contains V λ V_\lambda V λ , and on this set
( 1 − ∣ w ∣ ∣ I ∣ ) p ≥ 24 − p . \left(\frac{1-|w|}{|I|}\right)^p\ge24^{-p}. ( ∣ I ∣ 1 − ∣ w ∣ ) p ≥ 2 4 − p .
An application of Lemma 2.1 gives
∥ c ∥ t p 2 ≥ ∣ c λ ∣ 2 A ( V λ ) ( 1 − ∣ a λ ∣ 2 ) 2 24 − p ≥ c p ∣ c λ ∣ 2 . \|c\|_{t_p}^2\ge\frac{|c_\lambda|^2A(V_\lambda)}{(1-|a_\lambda|^2)^2}24^{-p}\ge c_p|c_\lambda|^2. ∥ c ∥ t p 2 ≥ ( 1 − ∣ a λ ∣ 2 ) 2 ∣ c λ ∣ 2 A ( V λ ) 2 4 − p ≥ c p ∣ c λ ∣ 2 .
For later reference, pairwise disjointness of the sets gives the exact representation
∥ c ∥ t p 2 = sup I ∑ λ ∣ c λ ∣ 2 ( 1 − ∣ a λ ∣ 2 ) 2 ∫ V λ ∩ Q I ( 1 − ∣ w ∣ ∣ I ∣ ) p d A ( w ) . \|c\|_{t_p}^2=\sup_I\sum_\lambda\frac{|c_\lambda|^2}{(1-|a_\lambda|^2)^2}\int_{V_\lambda\cap Q_I}\left(\frac{1-|w|}{|I|}\right)^p dA(w). ∥ c ∥ t p 2 = I sup λ ∑ ( 1 − ∣ a λ ∣ 2 ) 2 ∣ c λ ∣ 2 ∫ V λ ∩ Q I ( ∣ I ∣ 1 − ∣ w ∣ ) p d A ( w ) .
It follows that for any index set Λ \Lambda Λ and any choice of unimodular scalars θ λ \theta_\lambda θ λ ,
∥ ( c λ 1 Λ ( λ ) ) ∥ t p ≤ ∥ c ∥ t p , ∥ ( θ λ c λ ) ∥ t p = ∥ c ∥ t p . \left\|\left(c_\lambda\mathbf{1}_{\Lambda}(\lambda)\right)\right\|_{t_p} \le\|c\|_{t_p}, \qquad\|(\theta_\lambda c_\lambda)\|_{t_p}=\|c\|_{t_p}. ∥ ( c λ 1 Λ ( λ ) ) ∥ t p ≤ ∥ c ∥ t p , ∥ ( θ λ c λ ) ∥ t p = ∥ c ∥ t p .
Lemma 2.2. For f ∈ Q p f\in Q_p f ∈ Q p , set
c λ ( f ) = ( 1 − ∣ a λ ∣ 2 ) sup w ∈ V λ ∣ f ′ ( w ) ∣ . c_\lambda(f)=(1-|a_\lambda|^2)\sup_{w\in V_\lambda}|f'(w)|. c λ ( f ) = ( 1 − ∣ a λ ∣ 2 ) w ∈ V λ sup ∣ f ′ ( w ) ∣.
Then ∥ c ( f ) ∥ t p ≤ C p ∥ f ∥ Q p , ∗ \|c(f)\|_{t_p}\le C_p\|f\|_{Q_{p,*}} ∥ c ( f ) ∥ t p ≤ C p ∥ f ∥ Q p , ∗ .
Proof. For w ∈ V λ w\in V_\lambda w ∈ V λ , the disk D ( w , ( 1 − ∣ a λ ∣ 2 ) / 32 ) D(w,(1-|a_\lambda|^2)/32) D ( w , ( 1 − ∣ a λ ∣ 2 ) /32 ) is contained in V λ ∗ V_\lambda^* V λ ∗ . The submean property of subharmonic functions gives
sup V λ ∣ f ′ ∣ 2 ≤ C ( 1 − ∣ a λ ∣ 2 ) 2 ∫ V λ ∗ ∣ f ′ ( ζ ) ∣ 2 d A ( ζ ) . \sup_{V_\lambda}|f'|^2\le\frac{C}{(1-|a_\lambda|^2)^2}\int_{V_\lambda^*}|f'(\zeta)|^2\,dA(\zeta). V λ sup ∣ f ′ ∣ 2 ≤ ( 1 − ∣ a λ ∣ 2 ) 2 C ∫ V λ ∗ ∣ f ′ ( ζ ) ∣ 2 d A ( ζ ) .
Fix an arc I I I and write h = ∣ I ∣ h=|I| h = ∣ I ∣ . Let J J J denote the concentric arc with length min { 16 h , 1 } \min\{16h,1\} min { 16 h , 1 } , and take J = T J=\mathbb{T} J = T whenever 16 h ≥ 1 16h\ge1 16 h ≥ 1 . From the definitions of V λ V_\lambda V λ and V λ ∗ V_\lambda^* V λ ∗ ,
V λ ∩ Q I ≠ ∅ ⟹ 1 − ∣ a λ ∣ 2 ≤ 3 h , V λ ∗ ⊂ Q J , V_\lambda\cap Q_I\ne\varnothing\quad\Longrightarrow\quad1-|a_\lambda|^2\le3h,\qquad V_\lambda^*\subset Q_J, V λ ∩ Q I = ∅ ⟹ 1 − ∣ a λ ∣ 2 ≤ 3 h , V λ ∗ ⊂ Q J ,
( 1 − ∣ w ∣ h ) p ≤ C p ( 1 − ∣ ζ ∣ ∣ J ∣ ) p ( w ∈ V λ ∩ Q I , ζ ∈ V λ ∗ ) . \left(\frac{1-|w|}{h}\right)^p\le C_p\left(\frac{1-|\zeta|}{|J|}\right)^p\qquad(w\in V_\lambda\cap Q_I,\ \zeta\in V_\lambda^*). ( h 1 − ∣ w ∣ ) p ≤ C p ( ∣ J ∣ 1 − ∣ ζ ∣ ) p ( w ∈ V λ ∩ Q I , ζ ∈ V λ ∗ ) .
Using the bound A ( V λ ) ≲ ( 1 − ∣ a λ ∣ 2 ) 2 A(V_\lambda)\lesssim(1-|a_\lambda|^2)^2 A ( V λ ) ≲ ( 1 − ∣ a λ ∣ 2 ) 2 , we compute
∫ Q I ∣ F c ( f ) ( w ) ∣ 2 ( 1 − ∣ w ∣ h ) p d A ( w ) = ∑ V λ ∩ Q I ≠ ∅ sup V λ ∣ f ′ ∣ 2 ∫ V λ ∩ Q I ( 1 − ∣ w ∣ h ) p d A ( w ) ≤ C p ∑ V λ ∩ Q I ≠ ∅ A ( V λ ∩ Q I ) ( 1 − ∣ a λ ∣ 2 ) 2 ∫ V λ ∗ ∣ f ′ ( ζ ) ∣ 2 ( 1 − ∣ ζ ∣ ∣ J ∣ ) p d A ( ζ ) ≤ C p ∫ Q J ∣ f ′ ( ζ ) ∣ 2 ( 1 − ∣ ζ ∣ ∣ J ∣ ) p ∑ λ 1 V λ ∗ ( ζ ) d A ( ζ ) ≤ C p ∥ f ′ ∥ T p 2 ≤ C p ∥ f ∥ Q p , ∗ 2 . \begin{aligned}
\int_{Q_I}|F_{c(f)}(w)|^2\left(\frac{1-|w|}{h}\right)^p\,dA(w)
&=\sum_{V_\lambda\cap Q_I\ne\varnothing}\sup_{V_\lambda}|f'|^2
\int_{V_\lambda\cap Q_I}\left(\frac{1-|w|}{h}\right)^p\,dA(w)\\
&\le C_p\sum_{V_\lambda\cap Q_I\ne\varnothing}
\frac{A(V_\lambda\cap Q_I)}{(1-|a_\lambda|^2)^2}
\int_{V_\lambda^*}|f'(\zeta)|^2
\left(\frac{1-|\zeta|}{|J|}\right)^p\,dA(\zeta)\\
&\le C_p\int_{Q_J}|f'(\zeta)|^2
\left(\frac{1-|\zeta|}{|J|}\right)^p
\sum_\lambda\mathbf{1}_{V_\lambda^*}(\zeta)\,dA(\zeta)\\
&\le C_p\|f'\|_{T_p}^2\le C_p\|f\|_{Q_{p,*}}^2.
\end{aligned} ∫ Q I ∣ F c ( f ) ( w ) ∣ 2 ( h 1 − ∣ w ∣ ) p d A ( w ) = V λ ∩ Q I = ∅ ∑ V λ sup ∣ f ′ ∣ 2 ∫ V λ ∩ Q I ( h 1 − ∣ w ∣ ) p d A ( w ) ≤ C p V λ ∩ Q I = ∅ ∑ ( 1 − ∣ a λ ∣ 2 ) 2 A ( V λ ∩ Q I ) ∫ V λ ∗ ∣ f ′ ( ζ ) ∣ 2 ( ∣ J ∣ 1 − ∣ ζ ∣ ) p d A ( ζ ) ≤ C p ∫ Q J ∣ f ′ ( ζ ) ∣ 2 ( ∣ J ∣ 1 − ∣ ζ ∣ ) p λ ∑ 1 V λ ∗ ( ζ ) d A ( ζ ) ≤ C p ∥ f ′ ∥ T p 2 ≤ C p ∥ f ∥ Q p , ∗ 2 .
Taking the supremum over I I I proves the estimate. □
Lemma 2.3. Let M M M be as in (5) . For q = M , M + 1 q=M,M+1 q = M , M + 1 , define
h λ , q ( ζ ) = ( 1 − ∣ a λ ∣ 2 1 − a λ ‾ ζ ) q , S q c = ∑ λ c λ h λ , q h_{\lambda,q}(\zeta)=\left(\frac{1-|a_\lambda|^2}{1-\overline{a_\lambda}\zeta}\right)^q,\qquad S_qc=\sum_\lambda c_\lambda h_{\lambda,q} h λ , q ( ζ ) = ( 1 − a λ ζ 1 − ∣ a λ ∣ 2 ) q , S q c = λ ∑ c λ h λ , q
on the space of finitely supported sequences c c c . Then
∥ S q c ∥ Q p ≤ C p , M ∥ c ∥ t p . (17) \|S_qc\|_{Q_p}\le C_{p,M}\|c\|_{t_p}. \tag*{(17)} ∥ S q c ∥ Q p ≤ C p , M ∥ c ∥ t p . (17)
If the support of c c c is confined to indices with j ≥ J j\ge J j ≥ J , then for every R < 1 R<1 R < 1 and integer k ≥ 0 k\ge0 k ≥ 0 ,
sup ∣ ζ ∣ ≤ R ∣ ( S q c ) ( k ) ( ζ ) ∣ ≤ C p , M , R , k 2 − J ( q − 1 ) ∥ c ∥ t p . (18) \sup_{|\zeta|\le R}|(S_qc)^{(k)}(\zeta)|\le C_{p,M,R,k}2^{-J(q-1)}\|c\|_{t_p}. \tag*{(18)} ∣ ζ ∣ ≤ R sup ∣ ( S q c ) ( k ) ( ζ ) ∣ ≤ C p , M , R , k 2 − J ( q − 1 ) ∥ c ∥ t p . (18)
Proof. For w ∈ V λ w \in V_\lambda w ∈ V λ , Lemma 2.1 guarantees
1 − ∣ w ∣ 2 ≍ 1 − ∣ a λ ∣ 2 , A ( V λ ) ≍ ( 1 − ∣ a λ ∣ 2 ) 2 . 1-|w|^2 \asymp1-|a_\lambda|^2,\qquad A(V_\lambda)\asymp(1-|a_\lambda|^2)^2. 1 − ∣ w ∣ 2 ≍ 1 − ∣ a λ ∣ 2 , A ( V λ ) ≍ ( 1 − ∣ a λ ∣ 2 ) 2 .
Accordingly, the positive integral over a single set satisfies
∫ V λ ∣ F c ( w ) ∣ ( 1 − ∣ w ∣ 2 ) q − 1 ∣ 1 − w ‾ ζ ∣ q + 1 d A ( w ) ≍ ∣ c λ ∣ ( 1 − ∣ a λ ∣ 2 ) q ∣ 1 − a λ ‾ ζ ∣ q + 1 . \int_{V_\lambda}\frac{|F_c(w)|(1-|w|^2)^{q-1}}{|1-\overline{w}\zeta|^{q+1}}\,dA(w)\asymp\frac{|c_\lambda|(1-|a_\lambda|^2)^q}{|1-\overline{a_\lambda}\zeta|^{q+1}}. ∫ V λ ∣1 − w ζ ∣ q + 1 ∣ F c ( w ) ∣ ( 1 − ∣ w ∣ 2 ) q − 1 d A ( w ) ≍ ∣1 − a λ ζ ∣ q + 1 ∣ c λ ∣ ( 1 − ∣ a λ ∣ 2 ) q .
Since
h λ , q ′ ( ζ ) = q a λ ‾ ( 1 − ∣ a λ ∣ 2 ) q ( 1 − a λ ‾ ζ ) q + 1 , h'_{\lambda,q}(\zeta)=\frac{q\overline{a_\lambda}(1-|a_\lambda|^2)^q}{(1-\overline{a_\lambda}\zeta)^{q+1}}, h λ , q ′ ( ζ ) = ( 1 − a λ ζ ) q + 1 q a λ ( 1 − ∣ a λ ∣ 2 ) q ,
summing these estimates gives
∣ ( S q c ) ′ ( ζ ) ∣ ≤ q ∑ λ ∣ c λ ∣ ( 1 − ∣ a λ ∣ 2 ) q ∣ 1 − a λ ‾ ζ ∣ q + 1 ≤ C q ∫ D ∣ F c ( w ) ∣ ( 1 − ∣ w ∣ 2 ) q − 1 ∣ 1 − w ‾ ζ ∣ q + 1 d A ( w ) ≤ C q B q − 1 + F c ( ζ ) . \begin{aligned}
|(S_qc)'(\zeta)|\leq q\sum_\lambda\frac{|c_\lambda|(1-|a_\lambda|^2)^q}{|1-\overline{a_\lambda}\zeta|^{q+1}} \\
&\leq C_q\int_{\mathbb{D}}\frac{|F_c(w)|(1-|w|^2)^{q-1}}{|1-\overline{w}\zeta|^{q+1}}\,dA(w)\leq C_qB^+_{q-1}F_c(\zeta).
\end{aligned} ∣ ( S q c ) ′ ( ζ ) ∣ ≤ q λ ∑ ∣1 − a λ ζ ∣ q + 1 ∣ c λ ∣ ( 1 − ∣ a λ ∣ 2 ) q ≤ C q ∫ D ∣1 − w ζ ∣ q + 1 ∣ F c ( w ) ∣ ( 1 − ∣ w ∣ 2 ) q − 1 d A ( w ) ≤ C q B q − 1 + F c ( ζ ) .
Applying (11) with exponent q − 1 q-1 q − 1 ,
∥ S q c ∥ Q p , ∗ ≲ ∥ ( S q c ) ′ ∥ T p ≲ ∥ B q − 1 + F c ∥ T p ≲ ∥ F c ∥ T p = ∥ c ∥ t p . \|S_qc\|_{Q_p,*}\lesssim\|(S_qc)'\|_{T_p}\lesssim\|B^+_{q-1}F_c\|_{T_p}\lesssim\|F_c\|_{T_p}=\|c\|_{t_p}. ∥ S q c ∥ Q p , ∗ ≲ ∥ ( S q c ) ′ ∥ T p ≲ ∥ B q − 1 + F c ∥ T p ≲ ∥ F c ∥ T p = ∥ c ∥ t p .
At the origin, (16) gives
∣ S q c ( 0 ) ∣ ≤ sup λ ∣ c λ ∣ ∑ j ≥ 0 ∑ l = 0 2 j + 3 − 1 ( 1 − ∣ a j , l ∣ 2 ) q ≤ C q sup λ ∣ c λ ∣ ∑ j ≥ 0 2 j + 3 2 − j q ≤ C p , q ∥ c ∥ t p . \begin{aligned}
|S_qc(0)|\leq\sup_\lambda|c_\lambda|\sum_{j\geq0}\sum_{l=0}^{2^{j+3}-1}(1-|a_{j,l}|^2)^q \\
&\leq C_q\sup_\lambda|c_\lambda|\sum_{j\geq0}2^{j+3}2^{-jq}\leq C_{p,q}\|c\|_{t_p}.
\end{aligned} ∣ S q c ( 0 ) ∣ ≤ λ sup ∣ c λ ∣ j ≥ 0 ∑ l = 0 ∑ 2 j + 3 − 1 ( 1 − ∣ a j , l ∣ 2 ) q ≤ C q λ sup ∣ c λ ∣ j ≥ 0 ∑ 2 j + 3 2 − j q ≤ C p , q ∥ c ∥ t p .
Because q > 1 q>1 q > 1 , the series converges absolutely. The same argument shows that for any unimodular θ λ \theta_\lambda θ λ ,
∥ ∑ λ θ λ c λ h λ , q ∥ Q p ≤ C p , M ∥ c ∥ t p . \left\|\sum_\lambda\theta_\lambda c_\lambda h_{\lambda,q}\right\|_{Q_p}\leq C_{p,M}\|c\|_{t_p}. λ ∑ θ λ c λ h λ , q Q p ≤ C p , M ∥ c ∥ t p .
The constant does not depend on the finite index set or on the choice of phases θ λ \theta_\lambda θ λ .
For ∣ ζ ∣ ≤ R |\zeta|\leq R ∣ ζ ∣ ≤ R and k ≥ 0 k\geq0 k ≥ 0 ,
∣ h λ , q ( k ) ( ζ ) ∣ ≤ C q , k ( 1 − R ) − q − k ( 1 − ∣ a λ ∣ 2 ) q . |h_{\lambda,q}^{(k)}(\zeta)|\leq C_{q,k}(1-R)^{-q-k}(1-|a_\lambda|^2)^q. ∣ h λ , q ( k ) ( ζ ) ∣ ≤ C q , k ( 1 − R ) − q − k ( 1 − ∣ a λ ∣ 2 ) q .
Hence, when supp c ⊂ { j ≥ J } \operatorname{supp}c\subset\{j\geq J\} supp c ⊂ { j ≥ J } ,
sup ∣ ζ ∣ ≤ R ∣ ( S q c ) ( k ) ( ζ ) ∣ ≤ C q , R , k sup λ ∣ c λ ∣ ∑ j ≥ J 2 j + 3 2 − j q ≤ C p , q , R , k ∥ c ∥ t p 2 − J ( q − 1 ) 1 − 2 1 − q . \begin{aligned}
\sup_{|\zeta|\leq R}|(S_qc)^{(k)}(\zeta)|\leq C_{q,R,k}\sup_\lambda|c_\lambda|\sum_{j\geq J}2^{j+3}2^{-jq} \\
&\leq C_{p,q,R,k}\|c\|_{t_p}\frac{2^{-J(q-1)}}{1-2^{1-q}}.
\end{aligned} ∣ ζ ∣ ≤ R sup ∣ ( S q c ) ( k ) ( ζ ) ∣ ≤ C q , R , k λ sup ∣ c λ ∣ j ≥ J ∑ 2 j + 3 2 − j q ≤ C p , q , R , k ∥ c ∥ t p 1 − 2 1 − q 2 − J ( q − 1 ) .
This proves (18) . □
The coefficient norm introduced in (15) gives another formula for the derivative gauge:
L p ( ν ) = sup c f i n i t e ∥ c ∥ t p ≤ 1 ∑ λ ν ( V λ ) ( 1 − ∣ a λ ∣ 2 ) 2 ∣ c λ ∣ 2 , (19) \mathcal{L}_p(\nu)=\sup_{\substack{c\ \mathrm{finite}\\ \|c\|_{t_p}\leq1}}\sum_\lambda\frac{\nu(V_\lambda)}{(1-|a_\lambda|^2)^2}|c_\lambda|^2, \tag*{(19)} L p ( ν ) = c finite ∥ c ∥ t p ≤ 1 sup λ ∑ ( 1 − ∣ a λ ∣ 2 ) 2 ν ( V λ ) ∣ c λ ∣ 2 , (19)
L p ( ν ) ≍ ∥ h ν , 2 p ∥ B p 2 ( D ) . (20) \mathcal{L}_p(\nu)\asymp\|h_{\nu,2}^p\|_{B_p^2(\mathbb{D})}. \tag*{(20)} L p ( ν ) ≍ ∥ h ν , 2 p ∥ B p 2 ( D ) . (20)
Each upper half Q I u p Q_I^{\mathrm{up}} Q I up contains exactly eight sets V λ V_\lambda V λ . Whenever V λ ⊂ Q I u p V_\lambda\subset Q_I^{\mathrm{up}} V λ ⊂ Q I up and I ⊆ J I \subseteq J I ⊆ J , we have 1 − ∣ a λ ∣ 2 ≍ ∣ I ∣ 1-|a_\lambda|^2 \asymp|I| 1 − ∣ a λ ∣ 2 ≍ ∣ I ∣ , 1 − ∣ w ∣ ≍ ∣ I ∣ 1-|w| \asymp|I| 1 − ∣ w ∣ ≍ ∣ I ∣ uniformly on V λ V_\lambda V λ , and A ( V λ ) ≍ ∣ I ∣ 2 A(V_\lambda) \asymp|I|^2 A ( V λ ) ≍ ∣ I ∣ 2 . It follows that
∫ V λ 1 ( 1 − ∣ a λ ∣ 2 ) 2 ( 1 − ∣ w ∣ ∣ J ∣ ) p d A ( w ) ≍ p ∣ I ∣ 2 ∣ I ∣ 2 ( ∣ I ∣ ∣ J ∣ ) p = ( ∣ I ∣ ∣ J ∣ ) p . \int_{V_\lambda} \frac{1}{(1-|a_\lambda|^2)^2}\left(\frac{1-|w|}{|J|}\right)^p dA(w) \asymp_p \frac{|I|^2}{|I|^2}\left(\frac{|I|}{|J|}\right)^p = \left(\frac{|I|}{|J|}\right)^p. ∫ V λ ( 1 − ∣ a λ ∣ 2 ) 2 1 ( ∣ J ∣ 1 − ∣ w ∣ ) p d A ( w ) ≍ p ∣ I ∣ 2 ∣ I ∣ 2 ( ∣ J ∣ ∣ I ∣ ) p = ( ∣ J ∣ ∣ I ∣ ) p .
For a dyadic interval J J J , its tent decomposes disjointly into the upper halves indexed by all I ⊆ J I\subseteq J I ⊆ J . Summing this estimate and invoking (4) gives
∥ c ∥ t p 2 2 ≳ p sup J ∈ D ∑ I ⊆ J ( ∣ I ∣ ∣ J ∣ ) p ∑ V λ ⊂ Q I u p ∣ c λ ∣ 2 . \|c\|_{t_p^2}^2 \gtrsim_p \sup_{J\in\mathcal{D}}\sum_{I\subseteq J}\left(\frac{|I|}{|J|}\right)^p\sum_{V_\lambda\subset Q_I^{\mathrm{up}}}|c_\lambda|^2. ∥ c ∥ t p 2 2 ≳ p J ∈ D sup I ⊆ J ∑ ( ∣ J ∣ ∣ I ∣ ) p V λ ⊂ Q I up ∑ ∣ c λ ∣ 2 .
Given a finite coefficient family, set d I = ∣ I ∣ p max V λ ⊂ Q I u p ∣ c λ ∣ 2 d_I=|I|^p\max_{V_\lambda\subset Q_I^{\mathrm{up}}}|c_\lambda|^2 d I = ∣ I ∣ p max V λ ⊂ Q I up ∣ c λ ∣ 2 . Then
sup J ∑ I ⊆ J ∣ J ∣ − p d I ≲ p ∥ c ∥ t p 2 2 , \sup_J\sum_{I\subseteq J}|J|^{-p}d_I\lesssim_p\|c\|_{t_p^2}^2, J sup I ⊆ J ∑ ∣ J ∣ − p d I ≲ p ∥ c ∥ t p 2 2 ,
∑ λ ν ( V λ ) ∣ c λ ∣ 2 ( 1 − ∣ a λ ∣ 2 ) 2 ≍ ∑ I d I ν ( Q I u p ) ∣ I ∣ p + 2 . \sum_\lambda\frac{\nu(V_\lambda)|c_\lambda|^2}{(1-|a_\lambda|^2)^2}\asymp\sum_I d_I\frac{\nu(Q_I^{\mathrm{up}})}{|I|^{p+2}}. λ ∑ ( 1 − ∣ a λ ∣ 2 ) 2 ν ( V λ ) ∣ c λ ∣ 2 ≍ I ∑ d I ∣ I ∣ p + 2 ν ( Q I up ) .
Conversely, for an admissible finite sequence d d d , assign c λ = ( d I / ∣ I ∣ p ) 1 / 2 c_\lambda=(d_I/|I|^p)^{1/2} c λ = ( d I /∣ I ∣ p ) 1/2 to each of the eight cells inside Q I u p Q_I^{\mathrm{up}} Q I up . One obtains
∥ c ∥ t p 2 2 ≲ p 8 sup J ∑ I ⊆ J ∣ J ∣ − p d I ≤ C p ′ , \|c\|_{t_p^2}^2\lesssim_p 8\sup_J\sum_{I\subseteq J}|J|^{-p}d_I\le C'_p, ∥ c ∥ t p 2 2 ≲ p 8 J sup I ⊆ J ∑ ∣ J ∣ − p d I ≤ C p ′ ,
∑ λ ν ( V λ ) ∣ c λ ∣ 2 ( 1 − ∣ a λ ∣ 2 ) 2 ≍ ∑ I d I ν ( Q I u p ) ∣ I ∣ p + 2 . \sum_\lambda\frac{\nu(V_\lambda)|c_\lambda|^2}{(1-|a_\lambda|^2)^2}\asymp\sum_I d_I\frac{\nu(Q_I^{\mathrm{up}})}{|I|^{p+2}}. λ ∑ ( 1 − ∣ a λ ∣ 2 ) 2 ν ( V λ ) ∣ c λ ∣ 2 ≍ I ∑ d I ∣ I ∣ p + 2 ν ( Q I up ) .
After a fixed rescaling of coefficients, these inequalities prove (20) . The constants are independent of ν \nu ν and of any measurable restriction of ν \nu ν .
Also, Lemmas 2.2 and 2.3 together imply (10) for positive Borel measures. Define
E p ( ν ) = sup ∥ f ∥ Q p , ∗ ≤ 1 ∫ D ∣ f ′ ∣ 2 d ν . E_p(\nu)=\sup_{\|f\|_{Q_{p,*}}\le1}\int_{\mathbb{D}}|f'|^2d\nu. E p ( ν ) = ∥ f ∥ Q p , ∗ ≤ 1 sup ∫ D ∣ f ′ ∣ 2 d ν .
A sampling argument combined with finite truncation and monotone convergence gives E p ( ν ) ≤ C p L p ( ν ) E_p(\nu)\le C_p\mathcal{L}_p(\nu) E p ( ν ) ≤ C p L p ( ν ) . For the reverse inequality, note that ∣ h λ , M ′ ( w ) ∣ ≥ c M / ( 1 − ∣ a λ ∣ 2 ) |h'_{\lambda,M}(w)|\ge c_M/(1-|a_\lambda|^2) ∣ h λ , M ′ ( w ) ∣ ≥ c M / ( 1 − ∣ a λ ∣ 2 ) on V λ V_\lambda V λ . Take a finitely supported sequence c c c and independent sign variables ε λ = ± 1 \varepsilon_\lambda=\pm1 ε λ = ± 1 . Then
∑ λ ∣ c λ ∣ 2 ν ( V λ ) ( 1 − ∣ a λ ∣ 2 ) 2 ≤ C M ∑ λ ∣ c λ ∣ 2 ∫ D ∣ h λ , M ′ ∣ 2 d ν = C M E ∫ D ∣ ( S M ( ε c ) ) ′ ∣ 2 d ν ≤ C p , M E p ( ν ) ∥ c ∥ t p 2 2 . \begin{aligned}
\sum_\lambda\frac{|c_\lambda|^2\nu(V_\lambda)}{(1-|a_\lambda|^2)^2}\le C_M\sum_\lambda|c_\lambda|^2\int_{\mathbb{D}}|h'_{\lambda,M}|^2d\nu\\
&=C_M\mathbb{E}\int_{\mathbb{D}}|(S_M(\varepsilon c))'|^2d\nu\\
&\le C_{p,M}E_p(\nu)\|c\|_{t_p^2}^2.
\end{aligned} λ ∑ ( 1 − ∣ a λ ∣ 2 ) 2 ∣ c λ ∣ 2 ν ( V λ ) ≤ C M λ ∑ ∣ c λ ∣ 2 ∫ D ∣ h λ , M ′ ∣ 2 d ν = C M E ∫ D ∣ ( S M ( ε c ) ) ′ ∣ 2 d ν ≤ C p , M E p ( ν ) ∥ c ∥ t p 2 2 .
Hence L p ( ν ) ≤ C p E p ( ν ) \mathcal{L}_p(\nu)\le C_pE_p(\nu) L p ( ν ) ≤ C p E p ( ν ) . Together with (20) , this proves the equivalence. The expectation involves only finite sums; truncation handles infinite integrals. The constants remain unchanged under measurable restrictions of ν \nu ν .
Lemma 2.4. For w ∈ V λ w\in V_\lambda w ∈ V λ and arbitrary x , y ∈ C x,y\in\mathbb{C} x , y ∈ C ,
∣ x ∣ 2 + ( 1 − ∣ a λ ∣ 2 ) − 2 ∣ y ∣ 2 ≤ C M ∑ q = M M + 1 ∣ x h λ , q ( w ) + y h λ , q ′ ( w ) ∣ 2 . (21) |x|^2+(1-|a_\lambda|^2)^{-2}|y|^2\le C_M\sum_{q=M}^{M+1}|xh_{\lambda,q}(w)+yh'_{\lambda,q}(w)|^2. \tag*{(21)} ∣ x ∣ 2 + ( 1 − ∣ a λ ∣ 2 ) − 2 ∣ y ∣ 2 ≤ C M q = M ∑ M + 1 ∣ x h λ , q ( w ) + y h λ , q ′ ( w ) ∣ 2 . (21)
Proof. For w ∈ V λ w \in V_\lambda w ∈ V λ , Lemma 2.1 gives
4 − q ≤ ∣ h λ , q ( w ) ∣ ≤ 2 q , q = M , M + 1 , 4^{-q} \le\lvert h_{\lambda,q}(w)\rvert\le2^q,\qquad q=M,M+1, 4 − q ≤ ∣ h λ , q ( w )∣ ≤ 2 q , q = M , M + 1 ,
and
1 16 ≤ ∣ ( 1 − ∣ a λ ∣ 2 ) a λ ‾ 1 − a λ ‾ w ∣ ≤ 2. \frac{1}{16} \le\left\lvert\frac{(1-\lvert a_\lambda\rvert^2)\overline{a_\lambda}}{1-\overline{a_\lambda}w}\right\rvert\le2. 16 1 ≤ 1 − a λ w ( 1 − ∣ a λ ∣ 2 ) a λ ≤ 2.
Compute the logarithmic derivative:
h λ , q ′ ( w ) h λ , q ( w ) = q a λ ‾ 1 − a λ ‾ w . \frac{h'_{\lambda,q}(w)}{h_{\lambda,q}(w)}=\frac{q\overline{a_\lambda}}{1-\overline{a_\lambda}w}. h λ , q ( w ) h λ , q ′ ( w ) = 1 − a λ w q a λ .
We then derive the identities
a λ ‾ λ y 1 − a λ ‾ w = x h λ , M + 1 ( w ) + y h λ , M + 1 ′ ( w ) h λ , M + 1 ( w ) − x h λ , M ( w ) + y h λ , M ′ ( w ) h λ , M ( w ) , \frac{\overline{a_\lambda}\lambda y}{1-\overline{a_\lambda}w}
=
\frac{xh_{\lambda,M+1}(w)+yh'_{\lambda,M+1}(w)}{h_{\lambda,M+1}(w)}
-
\frac{xh_{\lambda,M}(w)+yh'_{\lambda,M}(w)}{h_{\lambda,M}(w)}, 1 − a λ w a λ λ y = h λ , M + 1 ( w ) x h λ , M + 1 ( w ) + y h λ , M + 1 ′ ( w ) − h λ , M ( w ) x h λ , M ( w ) + y h λ , M ′ ( w ) ,
x = ( M + 1 ) x h λ , M ( w ) + y h λ , M ′ ( w ) h λ , M ( w ) − M x h λ , M + 1 ( w ) + y h λ , M + 1 ′ ( w ) h λ , M + 1 ( w ) . x=(M+1)\frac{xh_{\lambda,M}(w)+yh'_{\lambda,M}(w)}{h_{\lambda,M}(w)}
-
M\frac{xh_{\lambda,M+1}(w)+yh'_{\lambda,M+1}(w)}{h_{\lambda,M+1}(w)}. x = ( M + 1 ) h λ , M ( w ) x h λ , M ( w ) + y h λ , M ′ ( w ) − M h λ , M + 1 ( w ) x h λ , M + 1 ( w ) + y h λ , M + 1 ′ ( w ) .
Using the first identity and the lower bound 1 / 16 1/16 1/16 , we obtain
∣ y ∣ 1 − ∣ a λ ∣ 2 ≤ 16 ∑ q = M M + 1 4 q ∣ x h λ , q ( w ) + y h λ , q ′ ( w ) ∣ . \frac{\lvert y\rvert}{1-\lvert a_\lambda\rvert^2}
\le
16\sum_{q=M}^{M+1}4^q\lvert xh_{\lambda,q}(w)+yh'_{\lambda,q}(w)\rvert. 1 − ∣ a λ ∣ 2 ∣ y ∣ ≤ 16 q = M ∑ M + 1 4 q ∣ x h λ , q ( w ) + y h λ , q ′ ( w )∣ .
The second identity provides
∣ x ∣ ≤ ( M + 1 ) ∑ q = M M + 1 4 q ∣ x h λ , q ( w ) + y h λ , q ′ ( w ) ∣ . \lvert x\rvert\le(M+1)\sum_{q=M}^{M+1}4^q\lvert xh_{\lambda,q}(w)+yh'_{\lambda,q}(w)\rvert. ∣ x ∣ ≤ ( M + 1 ) q = M ∑ M + 1 4 q ∣ x h λ , q ( w ) + y h λ , q ′ ( w )∣ .
Square both estimates and apply the convexity bound ( s + t ) 2 ≤ 2 s 2 + 2 t 2 (s+t)^2\le2s^2+2t^2 ( s + t ) 2 ≤ 2 s 2 + 2 t 2 . Adding the resulting inequalities gives (21) . □
Compactness and reconstruction
Let
E ρ = { w ∈ D : ∣ w ∣ > ρ } , P ρ F = P b , 0 ( 1 E ρ F ) , 0 < ρ < 1. E_\rho=\{w\in\mathbb{D}:\lvert w\rvert>\rho\},\qquad
P_\rho F=P_{b,0}\left(1_{E_\rho}F\right),\qquad
0<\rho<1. E ρ = { w ∈ D : ∣ w ∣ > ρ } , P ρ F = P b , 0 ( 1 E ρ F ) , 0 < ρ < 1.
We introduce the positive quantity
η b ( ρ ) = ( ∫ ∣ w ∣ > ρ ( 1 − ∣ w ∣ 2 ) 2 b ( 1 − ∣ w ∣ ) p d A ( w ) ) 1 / 2 . \eta_b(\rho)=\left(\int_{\lvert w\rvert>\rho}
\frac{(1-\lvert w\rvert^2)^{2b}}{(1-\lvert w\rvert)^p}\,dA(w)\right)^{1/2}. η b ( ρ ) = ( ∫ ∣ w ∣ > ρ ( 1 − ∣ w ∣ ) p ( 1 − ∣ w ∣ 2 ) 2 b d A ( w ) ) 1/2 .
A direct integration gives the estimate
η b ( ρ ) ≤ C p , b ( 1 − ρ ) b + ( 1 − p ) / 2 ⟶ 0. \eta_b(\rho)\le C_{p,b}(1-\rho)^{b+(1-p)/2}\longrightarrow0. η b ( ρ ) ≤ C p , b ( 1 − ρ ) b + ( 1 − p ) /2 ⟶ 0.
The exponent appearing above is positive under the standing choice of b b b .
Lemma 2.5. There exists a constant such that, uniformly over all F F F satisfying ∥ F ∥ T p ≤ 1 \lVert F\rVert_{T_p}\le1 ∥ F ∥ T p ≤ 1 ,
∥ P ρ F ∥ Q p ≤ C p , b , \lVert P_\rho F\rVert_{Q_p}\le C_{p,b}, ∥ P ρ F ∥ Q p ≤ C p , b , and, for each fixed 0 < R < 1 0<R<1 0 < R < 1 ,
sup ∣ ζ ∣ ≤ R ( ∣ P ρ F ( ζ ) ∣ + ∣ ( P ρ F ) ′ ( ζ ) ∣ ) ≤ C p , b , R η b ( ρ ) . \sup_{\lvert\zeta\rvert\le R}\left(\lvert P_\rho F(\zeta)\rvert+\lvert(P_\rho F)'(\zeta)\rvert\right)\le C_{p,b,R}\eta_b(\rho). ∣ ζ ∣ ≤ R sup ( ∣ P ρ F ( ζ )∣ + ∣( P ρ F ) ′ ( ζ )∣ ) ≤ C p , b , R η b ( ρ ) .
Proof. From the contractive bound ∥ 1 E ρ F ∥ T p ≤ ∥ F ∥ T p \|1_{E_\rho}F\|_{T_p}\le\|F\|_{T_p} ∥ 1 E ρ F ∥ T p ≤ ∥ F ∥ T p together with identity (6) ,
∥ P ρ F ∥ Q p ≤ C p , b ∥ 1 E ρ F ∥ T p ≤ C p , b ∥ F ∥ T p , P ρ F ( 0 ) = 0. \|P_\rho F\|_{Q_p}\le C_{p,b}\|1_{E_\rho}F\|_{T_p}\le C_{p,b}\|F\|_{T_p},\qquad P_\rho F(0)=0. ∥ P ρ F ∥ Q p ≤ C p , b ∥ 1 E ρ F ∥ T p ≤ C p , b ∥ F ∥ T p , P ρ F ( 0 ) = 0.
For any R < 1 R<1 R < 1 , the kernel admits the uniform bound
sup ∣ ζ ∣ ≤ R w ∈ D ( ∣ L b ( ζ , w ) ∣ + ∣ k b ( ζ , w ) ∣ ) ≤ C b , R . \sup_{\substack{|\zeta|\le R\\w\in\mathbb D}}\left(|L_b(\zeta,w)|+|k_b(\zeta,w)|\right)\le C_{b,R}. ∣ ζ ∣ ≤ R w ∈ D sup ( ∣ L b ( ζ , w ) ∣ + ∣ k b ( ζ , w ) ∣ ) ≤ C b , R .
An application of the Cauchy–Schwarz inequality, combined with the definition of the root tent, gives
sup ∣ ζ ∣ ≤ R ( ∣ P ρ F ( ζ ) ∣ + ∣ ( P ρ F ) ′ ( ζ ) ∣ ) ≤ C b , R ∫ E ρ ∣ F ( w ) ∣ d A b ( w ) ≤ C b , R ( ∫ E ρ ∣ F ( w ) ∣ 2 ( 1 − ∣ w ∣ ) p d A ( w ) ) 1 / 2 ( ∫ E ρ ( 1 − ∣ w ∣ 2 ) 2 b ( 1 − ∣ w ∣ ) p d A ( w ) ) 1 / 2 ≤ C p , b , R ∥ F ∥ T p η b ( ρ ) . \begin{aligned}
\sup_{|\zeta|\le R}\left(|P_\rho F(\zeta)|+|(P_\rho F)'(\zeta)|\right)
\le C_{b,R}\int_{E_\rho}|F(w)|\,dA_b(w) \\
&\le C_{b,R}\left(\int_{E_\rho}|F(w)|^2(1-|w|)^p\,dA(w)\right)^{1/2}
\left(\int_{E_\rho}\frac{(1-|w|^2)^{2b}}{(1-|w|)^p}\,dA(w)\right)^{1/2} \\
&\le C_{p,b,R}\|F\|_{T_p}\eta_b(\rho).
\end{aligned} ∣ ζ ∣ ≤ R sup ( ∣ P ρ F ( ζ ) ∣ + ∣ ( P ρ F ) ′ ( ζ ) ∣ ) ≤ C b , R ∫ E ρ ∣ F ( w ) ∣ d A b ( w ) ≤ C b , R ( ∫ E ρ ∣ F ( w ) ∣ 2 ( 1 − ∣ w ∣ ) p d A ( w ) ) 1/2 ( ∫ E ρ ( 1 − ∣ w ∣ ) p ( 1 − ∣ w ∣ 2 ) 2 b d A ( w ) ) 1/2 ≤ C p , b , R ∥ F ∥ T p η b ( ρ ) .
We next examine the asymptotic behaviour of η b \eta_b η b . Changing variables to radial coordinates,
η b ( ρ ) 2 = 2 ∫ ρ 1 ( 1 − r ) 2 b − p ( 1 + r ) 2 b r d r ≤ 2 2 b + 1 2 b − p + 1 ( 1 − ρ ) 2 b − p + 1 ⟶ 0 , \begin{aligned}
\eta_b(\rho)^2=2\int_\rho^1(1-r)^{2b-p}(1+r)^{2b}r\,dr \\
&\le\frac{2^{2b+1}}{2b-p+1}(1-\rho)^{2b-p+1}\longrightarrow0,
\end{aligned} η b ( ρ ) 2 = 2 ∫ ρ 1 ( 1 − r ) 2 b − p ( 1 + r ) 2 b r d r ≤ 2 b − p + 1 2 2 b + 1 ( 1 − ρ ) 2 b − p + 1 ⟶ 0 ,
where convergence follows from the hypothesis 2 b − p + 1 > 0 2b-p+1>0 2 b − p + 1 > 0 .
We shall also need the part of reconstruction over ∣ w ∣ ≤ ρ |w|\le\rho ∣ w ∣ ≤ ρ .
Lemma 2.6. For 0 < ρ < 1 0<\rho<1 0 < ρ < 1 , the mapping
R ρ F = P b , 0 ( 1 { ∣ w ∣ ≤ ρ } F ) R_\rho F=P_{b,0}\left(1_{\{|w|\le\rho\}}F\right) R ρ F = P b , 0 ( 1 { ∣ w ∣ ≤ ρ } F ) is compact as an operator from T p T_p T p into Q p Q_p Q p . Consequently,
S ρ f = f ( 0 ) + R ρ ( f ′ ) S_\rho f=f(0)+R_\rho(f') S ρ f = f ( 0 ) + R ρ ( f ′ ) defines a compact operator on Q p Q_p Q p , whose operator norm remains bounded uniformly in ρ \rho ρ .
Proof. The reproducing kernel admits the series expansion
k b ( ζ , w ) = ∑ j = 0 ∞ a b , j ζ j + 1 w ‾ j , a b , j = ( b + j + 1 j ) ≤ C b ( j + 1 ) b + 1 . k_b(\zeta,w)=\sum_{j=0}^{\infty}a_{b,j}\zeta^{j+1}\overline{w}^{\,j},\qquad
a_{b,j}=\binom{b+j+1}{j}\le C_b(j+1)^{b+1}. k b ( ζ , w ) = j = 0 ∑ ∞ a b , j ζ j + 1 w j , a b , j = ( j b + j + 1 ) ≤ C b ( j + 1 ) b + 1 .
Introduce the finite-rank approximation
R ρ , N F ( ζ ) = ∑ j = 0 N a b , j j + 1 ζ j + 1 ∫ ∣ w ∣ ≤ ρ F ( w ) w ‾ j d A b ( w ) . R_{\rho,N}F(\zeta)=\sum_{j=0}^{N}\frac{a_{b,j}}{j+1}\zeta^{j+1}\int_{|w|\le\rho}F(w)\overline{w}^{\,j}\,dA_b(w). R ρ , N F ( ζ ) = j = 0 ∑ N j + 1 a b , j ζ j + 1 ∫ ∣ w ∣ ≤ ρ F ( w ) w j d A b ( w ) .
For ∥ F ∥ T p ≤ 1 \|F\|_{T_p}\le1 ∥ F ∥ T p ≤ 1 , inequality (12) gives
∣ a b , j ∫ ∣ w ∣ ≤ ρ F ( w ) w ‾ j d A b ( w ) ∣ ≤ C p , b ( j + 1 ) b + 1 ρ j . \left|a_{b,j}\int_{|w|\le\rho}F(w)\overline{w}^{\,j}\,dA_b(w)\right|
\le C_{p,b}(j+1)^{b+1}\rho^j. a b , j ∫ ∣ w ∣ ≤ ρ F ( w ) w j d A b ( w ) ≤ C p , b ( j + 1 ) b + 1 ρ j .
If g g g vanishes at the origin, then
∥ g ∥ Q p 2 = sup a ∫ D ∣ g ′ ∣ 2 ( 1 − ∣ φ a ∣ 2 ) p d A ≤ ∥ g ′ ∥ ∞ 2 . \lVert g\rVert_{Q_p}^{2}=\sup_{a}\int_{\mathbb{D}}|g'|^{2}(1-|\varphi_a|^{2})^{p}\,dA\leq\lVert g'\rVert_{\infty}^{2}. ∥ g ∥ Q p 2 = a sup ∫ D ∣ g ′ ∣ 2 ( 1 − ∣ φ a ∣ 2 ) p d A ≤ ∥ g ′ ∥ ∞ 2 .
which implies
∥ R ρ − R ρ , N ∥ T p → Q p ≤ sup ∥ F ∥ T p ≤ 1 ∥ ( R ρ F − R ρ , N F ) ′ ∥ ∞ ≤ C p , b ∑ j > N ( j + 1 ) b + 1 ρ j ⟶ 0. \lVert R_{\rho}-R_{\rho,N}\rVert_{T_p\to Q_p}\leq\sup_{\lVert F\rVert_{T_p}\leq1}\lVert(R_{\rho}F-R_{\rho,N}F)'\rVert_{\infty}
\leq C_{p,b}\sum_{j>N}(j+1)^{b+1}\rho^{j}\longrightarrow0. ∥ R ρ − R ρ , N ∥ T p → Q p ≤ ∥ F ∥ T p ≤ 1 sup ∥( R ρ F − R ρ , N F ) ′ ∥ ∞ ≤ C p , b j > N ∑ ( j + 1 ) b + 1 ρ j ⟶ 0.
Thus R ρ R_{\rho} R ρ is an operator-norm limit of finite-rank operators and is compact. The decomposition
S ρ f = f ( 0 ) + R ρ ( f ′ ) , ∥ S ρ f ∥ Q p ≤ ∣ f ( 0 ) ∣ + C p ∥ f ′ ∥ T p ≤ C p ∥ f ∥ Q p S_{\rho}f=f(0)+R_{\rho}(f'),\qquad\lVert S_{\rho}f\rVert_{Q_p}\leq|f(0)|+C_p\lVert f'\rVert_{T_p}\leq C_p\lVert f\rVert_{Q_p} S ρ f = f ( 0 ) + R ρ ( f ′ ) , ∥ S ρ f ∥ Q p ≤ ∣ f ( 0 ) ∣ + C p ∥ f ′ ∥ T p ≤ C p ∥ f ∥ Q p
proves compactness and uniform boundedness of S ρ S_{\rho} S ρ . □ \square □
The reconstruction formula gives the exact identity
f − S ρ f = P ρ ( f ′ ) . f-S_{\rho}f=P_{\rho}(f'). f − S ρ f = P ρ ( f ′ ) .
The endpoint maximal identity
When p = 1 p=1 p = 1 , the coefficient condition [8 ] reduces to ∑ I ⊆ J d I ≤ ∣ J ∣ \sum_{I\subseteq J}d_I\leq|J| ∑ I ⊆ J d I ≤ ∣ J ∣ . We give a direct proof of its maximal integral representation. Set
D N = { I ∈ D : ∣ I ∣ ≥ 2 − N } , N ≥ 0. D_N=\{I\in\mathcal{D}: |I|\geq2^{-N}\},\qquad N\geq0. D N = { I ∈ D : ∣ I ∣ ≥ 2 − N } , N ≥ 0.
Lemma 2.7. For every function h : D → [ 0 , ∞ ] h:\mathcal{D}\to[0,\infty] h : D → [ 0 , ∞ ] ,
sup d I ≥ 0 ∑ I ⊆ J d I ≤ ∣ J ∣ ( J ∈ D ) ∑ I ∈ D d I h ( I ) = ∫ T sup I ∈ D ξ ∈ I h ( I ) d m ( ξ ) . \sup_{\substack{d_I\geq0\\ \sum_{I\subseteq J}d_I\leq|J|\ (J\in\mathcal{D})}}\sum_{I\in\mathcal{D}}d_Ih(I)
=\int_{\mathbb{T}}\sup_{\substack{I\in\mathcal{D}\\ \xi\in I}}h(I)\,dm(\xi). d I ≥ 0 ∑ I ⊆ J d I ≤ ∣ J ∣ ( J ∈ D ) sup I ∈ D ∑ d I h ( I ) = ∫ T I ∈ D ξ ∈ I sup h ( I ) d m ( ξ ) . Both sides may be infinite.
Proof. Fix an integer N ≥ 0 N\geq0 N ≥ 0 and a real number R > 0 R>0 R > 0 , and define
h R ( I ) = min { h ( I ) , R } , M N , R ( ξ ) = max I ∈ D N ξ ∈ I h R ( I ) . h_R(I)=\min\{h(I),R\},\qquad M_{N,R}(\xi)=\max_{\substack{I\in D_N\\ \xi\in I}}h_R(I). h R ( I ) = min { h ( I ) , R } , M N , R ( ξ ) = I ∈ D N ξ ∈ I max h R ( I ) .
The maximum is attained because the family is finite and contains the root interval T \mathbb{T} T . Let
P N , R = sup { ∑ I ∈ D N d I h R ( I ) : d I ≥ 0 , ∑ I ∈ D N I ⊆ J d I ≤ ∣ J ∣ ( J ∈ D N ) } . P_{N,R}=\sup\left\{\sum_{I\in D_N}d_Ih_R(I):d_I\geq0,\ \sum_{\substack{I\in D_N\\ I\subseteq J}}d_I\leq|J|\ (J\in D_N)\right\}. P N , R = sup ⎩ ⎨ ⎧ I ∈ D N ∑ d I h R ( I ) : d I ≥ 0 , I ∈ D N I ⊆ J ∑ d I ≤ ∣ J ∣ ( J ∈ D N ) ⎭ ⎬ ⎫ .
We first verify
P N , R = ∫ T M N , R d m . (22) P_{N,R}=\int_{\mathbb{T}}M_{N,R}\,dm. \tag*{(22)} P N , R = ∫ T M N , R d m . (22)
Take any admissible coefficients ( d I ) I ∈ D N (d_I)_{I\in D_N} ( d I ) I ∈ D N . There exist pairwise disjoint measurable subsets E I ⊆ I E_I\subseteq I E I ⊆ I satisfying
m ( E I ) = d I ( I ∈ D N ) . (23) m(E_I)=d_I\qquad(I\in D_N). \tag*{(23)} m ( E I ) = d I ( I ∈ D N ) . (23)
These sets may be constructed inductively. Begin with arcs of generation N N N , selecting E I ⊆ I E_I \subseteq I E I ⊆ I with measure d I ≤ ∣ I ∣ d_I \leq|I| d I ≤ ∣ I ∣ . Suppose the sets for all arcs of generation strictly larger than n n n have already been chosen. For an arc J J J at generation n n n , define
F J = J ∖ ⋃ I ∈ D N I ⊊ J E I . F_J=J\setminus\bigcup_{\substack{I\in\mathcal{D}_N\\I\subsetneq J}}E_I. F J = J ∖ I ∈ D N I ⊊ J ⋃ E I .
The previously selected sets are pairwise disjoint, so
m ( F J ) = ∣ J ∣ − ∑ I ∈ D N I ⊊ J d I ≥ d J . m(F_J)=|J|-\sum_{\substack{I\in\mathcal{D}_N\\I\subsetneq J}}d_I\geq d_J. m ( F J ) = ∣ J ∣ − I ∈ D N I ⊊ J ∑ d I ≥ d J .
Since arc length measure is non-atomic, we can choose a measurable subset E J ⊆ F J E_J\subseteq F_J E J ⊆ F J with measure exactly d J d_J d J . Arcs of a fixed generation are disjoint, and sets chosen outside J J J do not intersect J J J . This completes the induction and gives (23) .
For ξ ∈ E I ⊆ I \xi\in E_I\subseteq I ξ ∈ E I ⊆ I , we have h R ( I ) ≤ M N , R ( ξ ) h_R(I)\leq M_{N,R}(\xi) h R ( I ) ≤ M N , R ( ξ ) . It follows that
∑ I ∈ D N d I h R ( I ) = ∑ I ∈ D N ∫ E I h R ( I ) d m ≤ ∑ I ∈ D N ∫ E I M N , R d m ≤ ∫ T M N , R d m . \begin{aligned}
\sum_{I\in\mathcal{D}_N}d_Ih_R(I)=\sum_{I\in\mathcal{D}_N}\int_{E_I}h_R(I)\,\mathrm{d}m \\
&\leq\sum_{I\in\mathcal{D}_N}\int_{E_I}M_{N,R}\,\mathrm{d}m\leq\int_{\mathbb{T}}M_{N,R}\,\mathrm{d}m.
\end{aligned} I ∈ D N ∑ d I h R ( I ) = I ∈ D N ∑ ∫ E I h R ( I ) d m ≤ I ∈ D N ∑ ∫ E I M N , R d m ≤ ∫ T M N , R d m .
Taking the supremum gives the upper bound in (22) .
For the reverse inequality, enumerate D N = { I 1 , … , I L } \mathcal{D}_N=\{I_1,\ldots,I_L\} D N = { I 1 , … , I L } . Put
A j = { ξ ∈ I j : h R ( I j ) = M N , R ( ξ ) } , E I j = A j ∖ ⋃ i < j A i . A_j=\{\xi\in I_j:h_R(I_j)=M_{N,R}(\xi)\},\qquad E_{I_j}=A_j\setminus\bigcup_{i<j}A_i. A j = { ξ ∈ I j : h R ( I j ) = M N , R ( ξ )} , E I j = A j ∖ i < j ⋃ A i .
These measurable sets form a partition of T \mathbb{T} T , because at each point at least one containing arc achieves the maximal value. Moreover,
E I j ⊆ I j , M N , R = h R ( I j ) on E I j . E_{I_j}\subseteq I_j,\qquad M_{N,R}=h_R(I_j)\text{ on }E_{I_j}. E I j ⊆ I j , M N , R = h R ( I j ) on E I j .
Set d I = m ( E I ) d_I=m(E_I) d I = m ( E I ) . For each J ∈ D N J\in\mathcal{D}_N J ∈ D N ,
∑ I ∈ D N I ⊆ J d I = m ( ⋃ I ∈ D N I ⊆ J E I ) ≤ ∣ J ∣ . \sum_{\substack{I\in\mathcal{D}_N\\I\subseteq J}}d_I=m\left(\bigcup_{\substack{I\in\mathcal{D}_N\\I\subseteq J}}E_I\right)\leq|J|. I ∈ D N I ⊆ J ∑ d I = m I ∈ D N I ⊆ J ⋃ E I ≤ ∣ J ∣.
Hence ( d I ) (d_I) ( d I ) is admissible, and
∑ I ∈ D N d I h R ( I ) = ∑ I ∈ D N ∫ E I M N , R d m = ∫ T M N , R d m . \sum_{I\in\mathcal{D}_N}d_Ih_R(I)=\sum_{I\in\mathcal{D}_N}\int_{E_I}M_{N,R}\,\mathrm{d}m=\int_{\mathbb{T}}M_{N,R}\,\mathrm{d}m. I ∈ D N ∑ d I h R ( I ) = I ∈ D N ∑ ∫ E I M N , R d m = ∫ T M N , R d m .
This proves equality (22) , and shows that the supremum is attained for the truncated finite problem.
Let A \mathcal{A} A denote the collection of admissible sequences appearing in (22). The restriction of any element of A \mathcal{A} A to D N \mathcal{D}_N D N satisfies the constraints defining P N , R P_{N,R} P N , R . Conversely, any vector admissible for P N , R P_{N,R} P N , R extends to an element of A \mathcal{A} A by assigning d I = 0 d_I=0 d I = 0 for indices outside D N \mathcal{D}_N D N . The constraints over J ∈ D N J\in\mathcal{D}_N J ∈ D N remain unchanged; if J J J has generation exceeding N N N , no arc from D N \mathcal{D}_N D N is contained in J J J , so the associated sum vanishes. Therefore
P N , R = sup d ∈ A ∑ I ∈ D N d I h R ( I ) . P_{N,R}=\sup_{d\in\mathcal{A}}\sum_{I\in\mathcal{D}_N}d_Ih_R(I). P N , R = d ∈ A sup I ∈ D N ∑ d I h R ( I ) .
All summands are nonnegative. Taking R R R to be a positive integer and interchanging suprema,
sup d ∈ A ∑ I ∈ D d I h ( I ) = sup d ∈ A sup N , R ∑ I ∈ D N d I h R ( I ) = sup N , R P N , R = sup N , R ∫ T M N , R d m . \sup_{d\in A}\sum_{I\in\mathcal{D}}d_Ih(I)=\sup_{d\in A}\sup_{N,R}\sum_{I\in\mathcal{D}_N}d_Ih_R(I)
=\sup_{N,R}P_{N,R}=\sup_{N,R}\int_{\mathbb{T}}M_{N,R}\,dm. d ∈ A sup I ∈ D ∑ d I h ( I ) = d ∈ A sup N , R sup I ∈ D N ∑ d I h R ( I ) = N , R sup P N , R = N , R sup ∫ T M N , R d m .
The sequence M n , n M_{n,n} M n , n increases pointwise to sup I ∋ ξ h ( I ) \sup_{I\ni\xi}h(I) sup I ∋ ξ h ( I ) , and for any choice of N , R N,R N , R there exists an integer n ≥ max { N , R } n\ge\max\{N,R\} n ≥ max { N , R } such that M N , R ≤ M n , n M_{N,R}\le M_{n,n} M N , R ≤ M n , n . The monotone convergence theorem gives
sup N , R ∫ T M N , R d m = lim n → ∞ ∫ T M n , n d m = ∫ T sup I ∋ ξ h ( I ) d m , \sup_{N,R}\int_{\mathbb{T}}M_{N,R}\,dm=\lim_{n\to\infty}\int_{\mathbb{T}}M_{n,n}\,dm=\int_{\mathbb{T}}\sup_{I\ni\xi}h(I)\,dm, N , R sup ∫ T M N , R d m = n → ∞ lim ∫ T M n , n d m = ∫ T I ∋ ξ sup h ( I ) d m ,
which proves (22) without additional finiteness assumptions. □
We apply Lemma 2.7 with h ( I ) = ν ( Q I u p ) / ∣ I ∣ 3 h(I)=\nu(Q_I^{\mathrm{up}})/|I|^3 h ( I ) = ν ( Q I up ) /∣ I ∣ 3 . Using (8) and (9) ,
∥ h ν , 2 1 ∥ B 1 2 ( D ) = ∫ T sup I ∈ D ξ ∈ I ν ( Q I u p ) ∣ I ∣ 3 d m ( ξ ) . (24) \lVert h^1_{\nu,2}\rVert_{B^2_1(\mathbb{D})}
=\int_{\mathbb{T}}\sup_{\substack{I\in\mathcal{D}\\ \xi\in I}}\frac{\nu(Q_I^{\mathrm{up}})}{|I|^3}\,dm(\xi). \tag*{(24)} ∥ h ν , 2 1 ∥ B 1 2 ( D ) = ∫ T I ∈ D ξ ∈ I sup ∣ I ∣ 3 ν ( Q I up ) d m ( ξ ) . (24)
Substitute ν = 1 E μ a 1 \nu=\mathbf{1}_E\mu^1_a ν = 1 E μ a 1 and then take the supremum over a ∈ D a\in\mathbb{D} a ∈ D :
C 1 1 , p 2 ( E ) = sup a ∈ D ∥ h 1 E μ a 1 , 2 1 ∥ B 1 2 ( D ) = sup a ∈ D ∫ T sup I ∈ D ξ ∈ I μ a 1 ( E ∩ Q I u p ) ∣ I ∣ 3 d m ( ξ ) = M p 2 ( E ) . C_1^{1,p_2}(E)=\sup_{a\in\mathbb{D}}\lVert h^1_{\mathbf{1}_E\mu^1_a,2}\rVert_{B^2_1(\mathbb{D})}
=\sup_{a\in\mathbb{D}}\int_{\mathbb{T}}\sup_{\substack{I\in\mathcal{D}\\ \xi\in I}}\frac{\mu^1_a(E\cap Q_I^{\mathrm{up}})}{|I|^3}\,dm(\xi)=\mathcal{M}_{p_2}(E). C 1 1 , p 2 ( E ) = a ∈ D sup ∥ h 1 E μ a 1 , 2 1 ∥ B 1 2 ( D ) = a ∈ D sup ∫ T I ∈ D ξ ∈ I sup ∣ I ∣ 3 μ a 1 ( E ∩ Q I up ) d m ( ξ ) = M p 2 ( E ) .
This proves the identity in Theorem 1.2, valid both for E = D E=\mathbb{D} E = D and for E = E ρ E=E_\rho E = E ρ .
Proofs of the main results
We first prove the separation theorem, then translate boundedness and compactness of U \mathcal{U} U and V \mathcal{V} V into capacity conditions, and finally deduce Theorems 1.1–1.3.
Separation of the two terms
Throughout this section p 1 > 0 p_1>0 p 1 > 0 and 0 < p 2 < 1 0<p_2<1 0 < p 2 < 1 are fixed. We apply the lemmas of Section 2 with p = p 1 p=p_1 p = p 1 and choose b b b and M M M accordingly. The following theorem separates the two terms.
Theorem 3.1. Let p 1 > 0 p_1>0 p 1 > 0 and 0 < p 2 < 1 0<p_2<1 0 < p 2 < 1 , let A , B A,B A , B be holomorphic functions on D \mathbb{D} D , and let φ : D → D \varphi:\mathbb{D}\to\mathbb{D} φ : D → D be holomorphic. Define
T f = A ( f ∘ φ ) + B ( f ′ ∘ φ ) , T 0 f = A ( f ∘ φ ) , T 1 f = B ( f ′ ∘ φ ) . Tf=A(f\circ\varphi)+B(f'\circ\varphi),\qquad T_0f=A(f\circ\varphi),\qquad T_1f=B(f'\circ\varphi). T f = A ( f ∘ φ ) + B ( f ′ ∘ φ ) , T 0 f = A ( f ∘ φ ) , T 1 f = B ( f ′ ∘ φ ) . Then T : Q p 1 → N p 2 T:Q_{p_1}\to\mathcal{N}_{p_2} T : Q p 1 → N p 2 is bounded if and only if both T 0 T_0 T 0 and T 1 T_1 T 1 are bounded. In this case
∥ T 0 ∥ + ∥ T 1 ∥ ≤ C p 1 ∥ T ∥ . \lVert T_0\rVert+\lVert T_1\rVert\le C_{p_1}\lVert T\rVert. ∥ T 0 ∥ + ∥ T 1 ∥ ≤ C p 1 ∥ T ∥ . Moreover, T T T is compact if and only if both T 0 T_0 T 0 and T 1 T_1 T 1 are compact. Consequently W u , φ : Q p 1 → Q p 2 W_{u,\varphi}:Q_{p_1}\to Q_{p_2} W u , φ : Q p 1 → Q p 2 is bounded (respectively compact) if and only if the two maps in (1) are bounded (respectively compact) as operators into N p 2 \mathcal{N}_{p_2} N p 2 .
We use the following compactness test for the operators in this theorem and for W u , φ W_{u,\varphi} W u , φ [19 , 20 ] :
sup n ∥ f n ∥ Q p 1 < ∞ , f n → 0 uniformly on compact subsets of D ⟹ ∥ S f n ∥ → 0. (25) \sup_{n}\|f_n\|_{Q_{p_1}}<\infty,\quad f_n\to0\ \text{uniformly on compact subsets of }\mathbb{D}\quad\Longrightarrow\quad\|Sf_n\|\to0. \tag*{(25)} n sup ∥ f n ∥ Q p 1 < ∞ , f n → 0 uniformly on compact subsets of D ⟹ ∥ S f n ∥ → 0. (25)
For bounded S S S , (25) is equivalent to compactness. Indeed, norm convergence in Q p 2 Q_{p_2} Q p 2 or N p 2 \mathcal{N}_{p_2} N p 2 implies locally uniform convergence, and the operators considered here preserve locally uniform convergence. Conversely, by Montel’s theorem and Fatou’s lemma, every bounded sequence in Q p 1 Q_{p_1} Q p 1 has a subsequence converging locally uniformly to some f ∈ Q p 1 f\in Q_{p_1} f ∈ Q p 1 , and (25) applies to the differences f n − f f_n-f f n − f .
Proof. Assume first that T : Q p 1 → N p 2 T:Q_{p_1}\to\mathcal{N}_{p_2} T : Q p 1 → N p 2 is bounded. For each a ∈ D a\in\mathbb{D} a ∈ D , introduce the positive Borel measure
ν a ( E ) = ∫ φ − 1 ( E ) ∣ B ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) . \nu_a(E)=\int_{\varphi^{-1}(E)}|B(z)|^2(1-|\varphi_a(z)|^2)^{p_2}\,dA(z). ν a ( E ) = ∫ φ − 1 ( E ) ∣ B ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) .
Fix a finite index set Λ \Lambda Λ and coefficients c = ( c λ ) λ ∈ Λ \mathbf{c}=(c_\lambda)_{\lambda\in\Lambda} c = ( c λ ) λ ∈ Λ . Give { − 1 , 1 } Λ \{-1,1\}^{\Lambda} { − 1 , 1 } Λ the uniform probability measure, and write ε λ \varepsilon_\lambda ε λ for its coordinate random signs. Then
E X = 2 − ∣ Λ ∣ ∑ ε ∈ { − 1 , 1 } Λ X ( ε ) , E ( ε λ ε η ) = { 1 , λ = η , 0 , λ ≠ η . \mathbb{E}X=2^{-|\Lambda|}\sum_{\varepsilon\in\{-1,1\}^{\Lambda}}X(\varepsilon),\qquad
\mathbb{E}(\varepsilon_\lambda\varepsilon_\eta)=
\begin{cases}
1,&\lambda=\eta,\\
0,&\lambda\ne\eta.
\end{cases} E X = 2 − ∣Λ∣ ε ∈ { − 1 , 1 } Λ ∑ X ( ε ) , E ( ε λ ε η ) = { 1 , 0 , λ = η , λ = η .
For q = M , M + 1 q=M,M+1 q = M , M + 1 , set f q , ε = S q ( ε c ) f_{q,\varepsilon}=S_q(\varepsilon\mathbf{c}) f q , ε = S q ( ε c ) . Sign changes do not alter the norm of the coefficient sequence, and Lemma 2.3 gives
sup q , ε ∥ f q , ε ∥ Q p 1 ≤ C p 1 ∥ c ∥ t p 1 . \sup_{q,\varepsilon}\|f_{q,\varepsilon}\|_{Q_{p_1}}\le C_{p_1}\|\mathbf{c}\|_{t_{p_1}}. q , ε sup ∥ f q , ε ∥ Q p 1 ≤ C p 1 ∥ c ∥ t p 1 .
Expanding the squared modulus and taking the expectation gives
E ∣ T f q , ε ( z ) ∣ 2 = ∑ λ ∈ Λ ∣ c λ ∣ 2 ∣ A ( z ) h λ , q ( φ ( z ) ) + B ( z ) h λ , q ′ ( φ ( z ) ) ∣ 2 . \mathbb{E}|Tf_{q,\varepsilon}(z)|^2=\sum_{\lambda\in\Lambda}|c_\lambda|^2|A(z)h_{\lambda,q}(\varphi(z))+B(z)h'_{\lambda,q}(\varphi(z))|^2. E ∣ T f q , ε ( z ) ∣ 2 = λ ∈ Λ ∑ ∣ c λ ∣ 2 ∣ A ( z ) h λ , q ( φ ( z )) + B ( z ) h λ , q ′ ( φ ( z )) ∣ 2 .
For z ∈ φ − 1 ( V λ ) z\in\varphi^{-1}(V_\lambda) z ∈ φ − 1 ( V λ ) , apply Lemma 2.4 with w = φ ( z ) w=\varphi(z) w = φ ( z ) , x = A ( z ) x=A(z) x = A ( z ) , and y = B ( z ) y=B(z) y = B ( z ) . Multiply through by ∣ c λ ∣ 2 |c_\lambda|^2 ∣ c λ ∣ 2 and integrate to obtain
∑ λ ∈ Λ ∣ c λ ∣ 2 ∫ φ − 1 ( V λ ) ( ∣ A ( z ) ∣ 2 + ∣ B ( z ) ∣ 2 ( 1 − ∣ a λ ∣ 2 ) 2 ) ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) ≤ C M ∑ q = M M + 1 ∑ λ ∈ Λ ∣ c λ ∣ 2 ∫ D ∣ T h λ , q ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) = C M ∑ q = M M + 1 E ∫ D ∣ T f q , ε ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) ≤ C M ∥ T ∥ 2 ∑ q = M M + 1 E ∥ f q , ε ∥ Q p 1 2 ≤ C p 1 ∥ T ∥ 2 ∥ c ∥ t p 1 2 . (26) \begin{aligned}
&\sum_{\lambda\in\Lambda}|c_\lambda|^2\int_{\varphi^{-1}(V_\lambda)}
\left(|A(z)|^2+\frac{|B(z)|^2}{(1-|a_\lambda|^2)^2}\right)
(1-|\varphi_a(z)|^2)^{p_2}\,dA(z)\\
&\le C_M\sum_{q=M}^{M+1}\sum_{\lambda\in\Lambda}|c_\lambda|^2
\int_{\mathbb{D}}|Th_{\lambda,q}(z)|^2(1-|\varphi_a(z)|^2)^{p_2}\,dA(z)\\
&=C_M\sum_{q=M}^{M+1}\mathbb{E}\int_{\mathbb{D}}|Tf_{q,\varepsilon}(z)|^2(1-|\varphi_a(z)|^2)^{p_2}\,dA(z)\\
&\le C_M\|T\|^2\sum_{q=M}^{M+1}\mathbb{E}\|f_{q,\varepsilon}\|_{Q_{p_1}}^2
\le C_{p_1}\|T\|^2\|\mathbf{c}\|_{t_{p_1}}^2. \tag*{(26)}
\end{aligned} λ ∈ Λ ∑ ∣ c λ ∣ 2 ∫ φ − 1 ( V λ ) ( ∣ A ( z ) ∣ 2 + ( 1 − ∣ a λ ∣ 2 ) 2 ∣ B ( z ) ∣ 2 ) ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) ≤ C M q = M ∑ M + 1 λ ∈ Λ ∑ ∣ c λ ∣ 2 ∫ D ∣ T h λ , q ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) = C M q = M ∑ M + 1 E ∫ D ∣ T f q , ε ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) ≤ C M ∥ T ∥ 2 q = M ∑ M + 1 E ∥ f q , ε ∥ Q p 1 2 ≤ C p 1 ∥ T ∥ 2 ∥ c ∥ t p 1 2 . (26)
Since ∣ A ∣ 2 ≥ 0 |A|^2\ge0 ∣ A ∣ 2 ≥ 0 , we obtain
sup a ∈ D ∑ λ ∈ Λ ν a ( V λ ) ( 1 − ∣ a λ ∣ 2 ) 2 ∣ c λ ∣ 2 ≤ C p 1 ∥ T ∥ 2 ∥ c ∥ t p 1 2 . (27) \sup_{a\in\mathbb{D}}\sum_{\lambda\in\Lambda}\frac{\nu_a(V_\lambda)}{(1-|a_\lambda|^2)^2}|c_\lambda|^2
\le C_{p_1}\|T\|^2\|\mathbf{c}\|_{t_{p_1}}^2. \tag*{(27)} a ∈ D sup λ ∈ Λ ∑ ( 1 − ∣ a λ ∣ 2 ) 2 ν a ( V λ ) ∣ c λ ∣ 2 ≤ C p 1 ∥ T ∥ 2 ∥ c ∥ t p 1 2 . (27)
Now take arbitrary f ∈ Q p 1 f\in Q_{p_1} f ∈ Q p 1 and define
c λ ( f ) = ( 1 − ∣ a λ ∣ 2 ) sup w ∈ V λ ∣ f ′ ( w ) ∣ . c_\lambda(f)=(1-|a_\lambda|^2)\sup_{w\in V_\lambda}|f'(w)|. c λ ( f ) = ( 1 − ∣ a λ ∣ 2 ) w ∈ V λ sup ∣ f ′ ( w ) ∣.
By Lemma 2.2 and monotonicity of the coefficient norm,
∥ ( c j , l ( f ) 1 { j ≤ N } ) ∥ t p 1 ≤ ∥ c ( f ) ∥ t p 1 ≤ C p 1 ∥ f ∥ Q p 1 , ∗ . \left\|(c_{j,l}(f)1_{\{j\le N\}})\right\|_{t_{p_1}} \le\|c(f)\|_{t_{p_1}} \le C_{p_1}\|f\|_{Q_{p_1},*}. ( c j , l ( f ) 1 { j ≤ N } ) t p 1 ≤ ∥ c ( f ) ∥ t p 1 ≤ C p 1 ∥ f ∥ Q p 1 , ∗ .
Using (27) together with monotone convergence, for every a ∈ D a\in\mathbb{D} a ∈ D ,
∑ λ ν a ( V λ ) ∣ c λ ( f ) ∣ 2 ( 1 − ∣ a λ ∣ 2 ) 2 = lim N → ∞ ∑ j ≤ N , l ν a ( V j , l ) ∣ c j , l ( f ) ∣ 2 ( 1 − ∣ a j , l ∣ 2 ) 2 ≤ C p 1 ∥ T ∥ 2 ∥ f ∥ Q p 1 , ∗ 2 . \sum_{\lambda}\frac{\nu_a(V_\lambda)|c_\lambda(f)|^2}{(1-|a_\lambda|^2)^2}
=\lim_{N\to\infty}\sum_{j\le N,l}\frac{\nu_a(V_{j,l})|c_{j,l}(f)|^2}{(1-|a_{j,l}|^2)^2}
\le C_{p_1}\|T\|^2\|f\|_{Q_{p_1},*}^2. λ ∑ ( 1 − ∣ a λ ∣ 2 ) 2 ν a ( V λ ) ∣ c λ ( f ) ∣ 2 = N → ∞ lim j ≤ N , l ∑ ( 1 − ∣ a j , l ∣ 2 ) 2 ν a ( V j , l ) ∣ c j , l ( f ) ∣ 2 ≤ C p 1 ∥ T ∥ 2 ∥ f ∥ Q p 1 , ∗ 2 .
Hence
∥ T 1 f ∥ N p 2 2 = sup a ∫ D ∣ f ′ ( w ) ∣ 2 d ν a ( w ) ≤ sup a ∑ λ ν a ( V λ ) sup V λ ∣ f ′ ∣ 2 = sup a ∑ λ ν a ( V λ ) ( 1 − ∣ a λ ∣ 2 ) 2 ∣ c λ ( f ) ∣ 2 ≤ C p 1 ∥ T ∥ 2 ∥ f ∥ Q p 1 , ∗ 2 . \begin{aligned}
\|T_1f\|_{N_{p_2}}^2=\sup_a\int_{\mathbb{D}}|f'(w)|^2\,d\nu_a(w) \\
&\le\sup_a\sum_\lambda\nu_a(V_\lambda)\sup_{V_\lambda}|f'|^2 \\
&=\sup_a\sum_\lambda\frac{\nu_a(V_\lambda)}{(1-|a_\lambda|^2)^2}|c_\lambda(f)|^2
\le C_{p_1}\|T\|^2\|f\|_{Q_{p_1},*}^2.
\end{aligned} ∥ T 1 f ∥ N p 2 2 = a sup ∫ D ∣ f ′ ( w ) ∣ 2 d ν a ( w ) ≤ a sup λ ∑ ν a ( V λ ) V λ sup ∣ f ′ ∣ 2 = a sup λ ∑ ( 1 − ∣ a λ ∣ 2 ) 2 ν a ( V λ ) ∣ c λ ( f ) ∣ 2 ≤ C p 1 ∥ T ∥ 2 ∥ f ∥ Q p 1 , ∗ 2 .
It follows that
∥ T 1 ∥ ≤ C p 1 ∥ T ∥ , ∥ T 0 ∥ = ∥ T − T 1 ∥ ≤ ( 1 + C p 1 ) ∥ T ∥ . \|T_1\|\le C_{p_1}\|T\|,\qquad\|T_0\|=\|T-T_1\|\le(1+C_{p_1})\|T\|. ∥ T 1 ∥ ≤ C p 1 ∥ T ∥ , ∥ T 0 ∥ = ∥ T − T 1 ∥ ≤ ( 1 + C p 1 ) ∥ T ∥.
The converse direction follows directly from the decomposition T = T 0 + T 1 T=T_0+T_1 T = T 0 + T 1 .
Suppose now that T T T is compact. Then both T 0 T_0 T 0 and T 1 T_1 T 1 are bounded. Set
ε J = sup { ∥ T S q c ∥ N p 2 : q ∈ { M , M + 1 } , c has finite support , ∥ c ∥ t p 1 ≤ 1 , supp c ⊆ { j ≥ J } } . \varepsilon_J=\sup\left\{\|TS_qc\|_{N_{p_2}}:q\in\{M,M+1\},\ c\text{ has finite support},\ \|c\|_{t_{p_1}}\le1,\ \operatorname{supp}c\subseteq\{j\ge J\}\right\}. ε J = sup { ∥ T S q c ∥ N p 2 : q ∈ { M , M + 1 } , c has finite support , ∥ c ∥ t p 1 ≤ 1 , supp c ⊆ { j ≥ J } } .
Take any sequence J n → ∞ J_n\to\infty J n → ∞ , q n ∈ { M , M + 1 } q_n\in\{M,M+1\} q n ∈ { M , M + 1 } , and finitely supported coefficients c ( n ) c^{(n)} c ( n ) satisfying ∥ c ( n ) ∥ t p 1 ≤ 1 \|c^{(n)}\|_{t_{p_1}}\le1 ∥ c ( n ) ∥ t p 1 ≤ 1 and supp c ( n ) ⊆ { j ≥ J n } \operatorname{supp}c^{(n)}\subseteq\{j\ge J_n\} supp c ( n ) ⊆ { j ≥ J n } . Lemma 2.3 gives
sup n ∥ S q n c ( n ) ∥ Q p 1 ≤ C p 1 , sup ∣ w ∣ ≤ R ∣ S q n c ( n ) ( w ) ∣ ≤ C p 1 , M , R 2 − J n ( M − 1 ) ⟶ 0 ( R < 1 ) . \sup_n\|S_{q_n}c^{(n)}\|_{Q_{p_1}}\le C_{p_1},\qquad
\sup_{|w|\le R}|S_{q_n}c^{(n)}(w)|\le C_{p_1,M,R}2^{-J_n(M-1)}\longrightarrow0\quad(R<1). n sup ∥ S q n c ( n ) ∥ Q p 1 ≤ C p 1 , ∣ w ∣ ≤ R sup ∣ S q n c ( n ) ( w ) ∣ ≤ C p 1 , M , R 2 − J n ( M − 1 ) ⟶ 0 ( R < 1 ) .
Compactness combined with criterion (25) gives
∥ T S q n c ( n ) ∥ N p 2 ⟶ 0 , hence ε J ⟶ 0. \|TS_{q_n}c^{(n)}\|_{N_{p_2}}\longrightarrow0,\qquad\text{hence}\qquad\varepsilon_J\longrightarrow0. ∥ T S q n c ( n ) ∥ N p 2 ⟶ 0 , hence ε J ⟶ 0.
For finitely supported c c c supported in j ≥ J j\ge J j ≥ J , homogeneity gives
∥ T S q ( ε c ) ∥ N p 2 ≤ ε J ∥ c ∥ t p 1 , q = M , M + 1. \|TS_q(\varepsilon c)\|_{N_{p_2}}\le\varepsilon_J\|c\|_{t_{p_1}},\qquad q=M,M+1. ∥ T S q ( ε c ) ∥ N p 2 ≤ ε J ∥ c ∥ t p 1 , q = M , M + 1.
Repeating the estimate (26) with this bound gives
sup a ∑ j ≥ J , l ν a ( V j , l ) ( 1 − ∣ a j , l ∣ 2 ) 2 ∣ c j , l ∣ 2 ≤ C p 1 ε J 2 ∥ c ∥ t p 1 2 . (28) \sup_a\sum_{j\ge J,l}\frac{\nu_a(V_{j,l})}{(1-|a_{j,l}|^2)^2}|c_{j,l}|^2
\le C_{p_1}\varepsilon_J^2\|c\|_{t_{p_1}}^2. \tag*{(28)} a sup j ≥ J , l ∑ ( 1 − ∣ a j , l ∣ 2 ) 2 ν a ( V j , l ) ∣ c j , l ∣ 2 ≤ C p 1 ε J 2 ∥ c ∥ t p 1 2 . (28)
Applying the bounded operator T 1 T_1 T 1 to the identity map id ( w ) = w \operatorname{id}(w)=w id ( w ) = w gives
sup a ν a ( D ) = ∥ T 1 ( id ) ∥ N p 2 2 < ∞ . \sup_a\nu_a(\mathbb{D})=\|T_1(\operatorname{id})\|_{N_{p_2}}^2<\infty. a sup ν a ( D ) = ∥ T 1 ( id ) ∥ N p 2 2 < ∞.
Let ( f n ) (f_n) ( f n ) be a bounded sequence in Q p 1 Q_{p_1} Q p 1 converging uniformly to zero on compact subsets. Write r J = 1 − 2 − J r_J=1-2^{-J} r J = 1 − 2 − J . The sets indexed by j ≥ J j\ge J j ≥ J cover the region { ∣ w ∣ ≥ r J } \{|w|\ge r_J\} { ∣ w ∣ ≥ r J } . By Lemma 2.2, inequality (28) , and monotone convergence,
∥ T 1 f n ∥ N p 2 2 ≤ sup ∣ w ∣ ≤ r J ∣ f n ′ ( w ) ∣ 2 sup a ν a ( D ) + C p 1 ε J 2 ∥ f n ∥ Q p 1 , ∗ 2 . \|T_1f_n\|_{N_{p_2}}^2\le\sup_{|w|\le r_J}|f_n'(w)|^2\sup_a\nu_a(\mathbb{D})+C_{p_1}\varepsilon_J^2\|f_n\|_{Q_{p_1},*}^2. ∥ T 1 f n ∥ N p 2 2 ≤ ∣ w ∣ ≤ r J sup ∣ f n ′ ( w ) ∣ 2 a sup ν a ( D ) + C p 1 ε J 2 ∥ f n ∥ Q p 1 , ∗ 2 .
For fixed J J J , Cauchy’s integral formula guarantees sup ∣ w ∣ ≤ r J ∣ f n ′ ( w ) ∣ → 0 \sup_{|w|\le r_J}|f_n'(w)|\to0 sup ∣ w ∣ ≤ r J ∣ f n ′ ( w ) ∣ → 0 . Therefore
0 ≤ lim sup n → ∞ ∥ T 1 f n ∥ N p 2 2 ≤ C p 1 ε J 2 sup n ∥ f n ∥ Q p 1 , ∗ 2 → J → ∞ 0. 0\le\limsup_{n\to\infty}\|T_1f_n\|_{\mathcal{N}_{p_2}}^2\le C_{p_1}\varepsilon_J^2\sup_n\|f_n\|_{\mathcal{Q}_{p_1,*}}^2\xrightarrow{J\to\infty}0. 0 ≤ n → ∞ lim sup ∥ T 1 f n ∥ N p 2 2 ≤ C p 1 ε J 2 n sup ∥ f n ∥ Q p 1 , ∗ 2 J → ∞ 0.
By (25) , T 1 T_1 T 1 and T 0 = T − T 1 T_0=T-T_1 T 0 = T − T 1 are compact. Conversely, the sum of two compact operators is compact.
Finally set A = u ′ A=u' A = u ′ and B = u φ ′ B=u\varphi' B = u φ ′ . Differentiation carries a bounded or compact W u , φ W_{u,\varphi} W u , φ to an operator T T T with the same property. Conversely,
W u , φ f ( z ) = u ( 0 ) f ( φ ( 0 ) ) + ∫ 0 z U f ( ξ ) d ξ + ∫ 0 z V f ( ξ ) d ξ . W_{u,\varphi}f(z)=u(0)f(\varphi(0))+\int_0^z Uf(\xi)\,d\xi+\int_0^z Vf(\xi)\,d\xi. W u , φ f ( z ) = u ( 0 ) f ( φ ( 0 )) + ∫ 0 z U f ( ξ ) d ξ + ∫ 0 z V f ( ξ ) d ξ .
Integration defines a bounded mapping from N p 2 \mathcal{N}_{p_2} N p 2 into Q ˚ p 2 \mathring{\mathcal{Q}}_{p_2} Q ˚ p 2 , while the initial term is a rank-one bounded operator. This completes the proof of the final claim. □ \square □
Compactness of the two embeddings
By Theorem 3.1,
W u , φ compact ⟺ U , V : Q p 1 → N p 2 compact . W_{u,\varphi}\text{ compact}\quad\Longleftrightarrow\quad U,V:\mathcal{Q}_{p_1}\to\mathcal{N}_{p_2}\text{ compact}. W u , φ compact ⟺ U , V : Q p 1 → N p 2 compact .
We now express these conditions through the two capacities.
Substitute p = p 1 p=p_1 p = p 1 and ν = μ a 1 \nu=\mu_a^1 ν = μ a 1 into (10) and take the supremum over all a a a . This gives
∥ V : Q p 1 → N p 2 ∥ 2 ≍ C p 1 , p 2 1 ( D ) . \|V:\mathcal{Q}_{p_1}\to\mathcal{N}_{p_2}\|^2\asymp C_{p_1,p_2}^1(\mathbb{D}). ∥ V : Q p 1 → N p 2 ∥ 2 ≍ C p 1 , p 2 1 ( D ) .
Proposition 3.2. Assume that V : Q p 1 → N p 2 V:\mathcal{Q}_{p_1}\to\mathcal{N}_{p_2} V : Q p 1 → N p 2 is bounded. Then
V is compact ⟺ lim r ↑ 1 sup a ∈ D ∥ h 1 { ∣ w ∣ > r } p 1 μ a 1 ∥ B p 1 2 ( D ) = 0. (29) V\text{ is compact}\quad\Longleftrightarrow\quad\lim_{r\uparrow1}\sup_{a\in\mathbb{D}}\|h_{1_{\{|w|>r\}}}^{p_1}\mu_a^1\|_{B_{p_1}^2(\mathbb{D})}=0. \tag*{(29)} V is compact ⟺ r ↑ 1 lim a ∈ D sup ∥ h 1 { ∣ w ∣ > r } p 1 μ a 1 ∥ B p 1 2 ( D ) = 0. (29)
Proof. Suppose V V V is compact. Define
ε J = sup { ∥ V S M c ∥ N p 2 : c finite , ∥ c ∥ ℓ p 1 ≤ 1 , supp c ⊂ { j ≥ J } } . \varepsilon_J=\sup\{\|VS_Mc\|_{\mathcal{N}_{p_2}}:c\text{ finite},\ \|c\|_{\ell_{p_1}}\le1,\ \operatorname{supp}c\subset\{j\ge J\}\}. ε J = sup { ∥ V S M c ∥ N p 2 : c finite , ∥ c ∥ ℓ p 1 ≤ 1 , supp c ⊂ { j ≥ J }} .
Combining (17) , (18) , and (25) , we obtain ε J → 0 \varepsilon_J\to0 ε J → 0 as J → ∞ J\to\infty J → ∞ . Consequently, for every finitely supported coefficient sequence c c c supported in j ≥ J j\ge J j ≥ J ,
∥ V S M c ∥ N p 2 ≤ ε J ∥ c ∥ ℓ p 1 . \|VS_Mc\|_{\mathcal{N}_{p_2}}\le\varepsilon_J\|c\|_{\ell_{p_1}}. ∥ V S M c ∥ N p 2 ≤ ε J ∥ c ∥ ℓ p 1 .
For coefficients supported on indices j ≥ J j\ge J j ≥ J , averaging over random sign choices together with the pointwise lower bound ∣ h λ , M ′ ∣ ≥ c M / ( 1 − ∣ a λ ∣ 2 ) |h'_{\lambda,M}|\ge c_M/(1-|a_\lambda|^2) ∣ h λ , M ′ ∣ ≥ c M / ( 1 − ∣ a λ ∣ 2 ) on V λ V_\lambda V λ gives
∑ j ≥ J , l μ a 1 ( V j , l ) ( 1 − ∣ a j , l ∣ 2 ) 2 ∣ c j , l ∣ 2 ≤ C M E ∫ D ∣ u ( z ) φ ′ ( z ) ( S M ( ε c ) ) ′ ( φ ( z ) ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) ≤ C M ε J 2 ∥ c ∥ ℓ p 1 2 . \begin{aligned}
\sum_{j\ge J,l}\frac{\mu_a^1(V_{j,l})}{(1-|a_{j,l}|^2)^2}|c_{j,l}|^2
\le C_M\mathbb{E}\int_{\mathbb{D}}|u(z)\varphi'(z)(S_M(\varepsilon c))'(\varphi(z))|^2(1-|\varphi_a(z)|^2)^{p_2}\,dA(z) \\
&\le C_M\varepsilon_J^2\|c\|_{\ell_{p_1}}^2.
\end{aligned} j ≥ J , l ∑ ( 1 − ∣ a j , l ∣ 2 ) 2 μ a 1 ( V j , l ) ∣ c j , l ∣ 2 ≤ C M E ∫ D ∣ u ( z ) φ ′ ( z ) ( S M ( ε c ) ) ′ ( φ ( z )) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) ≤ C M ε J 2 ∥ c ∥ ℓ p 1 2 .
The union of these sets is { ∣ w ∣ ≥ r J } \{|w|\ge r_J\} { ∣ w ∣ ≥ r J } , where r J = 1 − 2 − J r_J=1-2^{-J} r J = 1 − 2 − J . The definition (19) and (20) therefore yield
sup a ∥ h 1 { ∣ w ∣ ≥ r J } p 1 μ a 1 ∥ B p 1 2 ( D ) ≤ C p 1 ε J 2 ⟶ 0. \sup_a\|h_{1_{\{|w|\ge r_J\}}}^{p_1}\mu_a^1\|_{B_{p_1}^2(\mathbb{D})}\le C_{p_1}\varepsilon_J^2\longrightarrow0. a sup ∥ h 1 { ∣ w ∣ ≥ r J } p 1 μ a 1 ∥ B p 1 2 ( D ) ≤ C p 1 ε J 2 ⟶ 0.
Monotonicity of the expression in r r r extends this limit to any sequence r ↑ 1 r\uparrow1 r ↑ 1 .
Conversely, assume the limit in (29) holds. Testing the bounded operator V V V against the function f ( w ) = w f(w)=w f ( w ) = w gives
M 1 : = sup a μ a 1 ( D ) = ∥ u φ ′ ∥ N p 2 2 < ∞ . M_1:=\sup_a\mu_a^1(\mathbb{D})=\|u\varphi'\|_{\mathcal{N}_{p_2}}^2<\infty. M 1 := a sup μ a 1 ( D ) = ∥ u φ ′ ∥ N p 2 2 < ∞.
Let { f n } \{f_n\} { f n } be a bounded sequence in Q p 1 Q_{p_1} Q p 1 that converges to zero uniformly on every compact subset of D \mathbb{D} D . For fixed r < 1 r<1 r < 1 , the comparison (10) gives
∥ V f n ∥ N p 2 2 = sup a ∫ D ∣ f n ′ ( w ) ∣ 2 d μ a 1 ( w ) ≤ M 1 sup ∣ w ∣ ≤ r ∣ f n ′ ( w ) ∣ 2 + C p 1 ∥ f n ∥ Q p 1 , ∗ 2 sup a ∥ h 1 p 1 1 { ∣ w ∣ > r } μ a , 2 1 ∥ B p 1 2 ( D ) . \|\mathcal{V}f_n\|_{N_{p_2}}^2=\sup_a\int_{\mathbb{D}}|f_n'(w)|^2\,d\mu_a^1(w)
\le M_1\sup_{|w|\le r}|f_n'(w)|^2+C_{p_1}\|f_n\|_{Q_{p_1,*}}^2\sup_a\|h_1^{p_1}\mathbf{1}_{\{|w|>r\}}\mu_{a,2}^1\|_{B_{p_1}^2(\mathbb{D})}. ∥ V f n ∥ N p 2 2 = a sup ∫ D ∣ f n ′ ( w ) ∣ 2 d μ a 1 ( w ) ≤ M 1 ∣ w ∣ ≤ r sup ∣ f n ′ ( w ) ∣ 2 + C p 1 ∥ f n ∥ Q p 1 , ∗ 2 a sup ∥ h 1 p 1 1 { ∣ w ∣ > r } μ a , 2 1 ∥ B p 1 2 ( D ) .
Writing C = sup n ∥ f n ∥ Q p 1 , ∗ < ∞ C=\sup_n\|f_n\|_{Q_{p_1,*}}<\infty C = sup n ∥ f n ∥ Q p 1 , ∗ < ∞ , we obtain
lim sup n → ∞ ∥ V f n ∥ N p 2 2 ≤ C p 1 C sup a ∥ h 1 p 1 1 { ∣ w ∣ > r } μ a , 2 1 ∥ B p 1 2 ( D ) ⟶ 0. \limsup_{n\to\infty}\|\mathcal{V}f_n\|_{N_{p_2}}^2\le C_{p_1}C\sup_a\|h_1^{p_1}\mathbf{1}_{\{|w|>r\}}\mu_{a,2}^1\|_{B_{p_1}^2(\mathbb{D})}\longrightarrow0. n → ∞ lim sup ∥ V f n ∥ N p 2 2 ≤ C p 1 C a sup ∥ h 1 p 1 1 { ∣ w ∣ > r } μ a , 2 1 ∥ B p 1 2 ( D ) ⟶ 0.
Apply the test (25) . □
For any Borel subset E ⊆ D E\subseteq\mathbb{D} E ⊆ D , define
L E F ( z ) = u ′ ( z ) ∫ E L b ( φ ( z ) , w ) F ( w ) d A b ( w ) . \mathcal{L}_E F(z)=u'(z)\int_E L_b(\varphi(z),w)F(w)\,dA_b(w). L E F ( z ) = u ′ ( z ) ∫ E L b ( φ ( z ) , w ) F ( w ) d A b ( w ) .
We substitute p = p 1 p=p_1 p = p 1 and ν = μ a , t 0 \nu=\mu_{a,t}^0 ν = μ a , t 0 into (7) and then take the supremum over the parameters a a a and t t t . This gives
∥ L E : T p 1 → N p 2 ∥ 2 ≍ C 0 p 1 , p 2 ( E ) . \|\mathcal{L}_E:T_{p_1}\to\mathcal{N}_{p_2}\|^2\asymp C_0^{p_1,p_2}(E). ∥ L E : T p 1 → N p 2 ∥ 2 ≍ C 0 p 1 , p 2 ( E ) .
The implicit constants are independent of E E E , a a a , and t t t . In the special case E = D E=\mathbb{D} E = D , the reconstruction formula together with the value embedding theorem further gives
∥ U : Q ˙ p 1 → N p 2 ∥ 2 ≍ C 0 p 1 , p 2 ( D ) . \|\mathcal{U}:\dot{Q}_{p_1}\to\mathcal{N}_{p_2}\|^2\asymp C_0^{p_1,p_2}(\mathbb{D}). ∥ U : Q ˙ p 1 → N p 2 ∥ 2 ≍ C 0 p 1 , p 2 ( D ) .
Proposition 3.3. Suppose U : Q p 1 → N p 2 \mathcal{U}:Q_{p_1}\to\mathcal{N}_{p_2} U : Q p 1 → N p 2 is bounded. Then
U is compact ⟺ C 0 p 1 , p 2 ( E ρ ) ⟶ 0 ( ρ ↑ 1 ) . \mathcal{U}\text{ is compact}\quad\Longleftrightarrow\quad C_0^{p_1,p_2}(E_\rho)\longrightarrow0\quad(\rho\uparrow1). U is compact ⟺ C 0 p 1 , p 2 ( E ρ ) ⟶ 0 ( ρ ↑ 1 ) . Proof. First assume that U \mathcal{U} U is compact. Lemma 2.5 supplies the estimates
sup ρ < 1 sup ∥ F ∥ T p 1 ≤ 1 ∥ P ρ F ∥ Q p 1 < ∞ , \sup_{\rho<1}\sup_{\|F\|_{T_{p_1}}\le1}\|P_\rho F\|_{Q_{p_1}}<\infty, ρ < 1 sup ∥ F ∥ T p 1 ≤ 1 sup ∥ P ρ F ∥ Q p 1 < ∞ , sup ∥ F ∥ T p 1 ≤ 1 sup ∣ ζ ∣ ≤ R ∣ P ρ F ( ζ ) ∣ ≤ C p 1 , b , R η b ( ρ ) ⟶ 0 ( R < 1 ) . \sup_{\|F\|_{T_{p_1}}\le1}\sup_{|\zeta|\le R}|P_\rho F(\zeta)|\le C_{p_1,b,R}\eta_b(\rho)\longrightarrow0\quad(R<1). ∥ F ∥ T p 1 ≤ 1 sup ∣ ζ ∣ ≤ R sup ∣ P ρ F ( ζ ) ∣ ≤ C p 1 , b , R η b ( ρ ) ⟶ 0 ( R < 1 ) . If sup ∥ F ∥ T p 1 ≤ 1 ∥ U P ρ F ∥ N p 2 \sup_{\|F\|_{T_{p_1}}\le1}\|\mathcal{U}P_\rho F\|_{N_{p_2}} sup ∥ F ∥ T p 1 ≤ 1 ∥ U P ρ F ∥ N p 2 failed to vanish as ρ ↑ 1 \rho\uparrow1 ρ ↑ 1 , we could select
ρ n ↑ 1 , ∥ F n ∥ T p 1 ≤ 1 , ∥ U P ρ n F n ∥ N p 2 ≥ ε > 0 \rho_n\uparrow1,\qquad\|F_n\|_{T_{p_1}}\le1,\qquad\|\mathcal{U}P_{\rho_n}F_n\|_{N_{p_2}}\ge\varepsilon>0 ρ n ↑ 1 , ∥ F n ∥ T p 1 ≤ 1 , ∥ U P ρ n F n ∥ N p 2 ≥ ε > 0 for some fixed ε > 0 \varepsilon>0 ε > 0 . However, the preceding bounds combined with (25) force ∥ U P ρ n F n ∥ N p 2 → 0 \|\mathcal{U}P_{\rho_n}F_n\|_{N_{p_2}}\to0 ∥ U P ρ n F n ∥ N p 2 → 0 , which gives a contradiction. It follows that
C 0 p 1 , p 2 ( E ρ ) ≍ ∥ L E ρ ∥ T p 1 → N p 2 2 = ∥ U P ρ ∥ T p 1 → N p 2 2 ⟶ 0. C_0^{p_1,p_2}(E_\rho)\asymp\|\mathcal{L}_{E_\rho}\|_{T_{p_1}\to N_{p_2}}^2=\|\mathcal{U}P_\rho\|_{T_{p_1}\to N_{p_2}}^2\longrightarrow0. C 0 p 1 , p 2 ( E ρ ) ≍ ∥ L E ρ ∥ T p 1 → N p 2 2 = ∥ U P ρ ∥ T p 1 → N p 2 2 ⟶ 0. For the converse direction, Lemma 2.6 guarantees that S ρ S_\rho S ρ is compact. Moreover,
( U − U S ρ ) f = U P ρ ( f ′ ) = L E ρ ( f ′ ) , (\mathcal{U}-\mathcal{U}S_\rho)f=\mathcal{U}P_\rho(f')=\mathcal{L}_{E_\rho}(f'), ( U − U S ρ ) f = U P ρ ( f ′ ) = L E ρ ( f ′ ) , ∥ U − U S ρ ∥ Q p 1 → N p 2 ≤ C p 1 ∥ L E ρ ∥ T p 1 → N p 2 2 ≤ C p 1 C 0 p 1 , p 2 ( E ρ ) ⟶ 0. \|\mathcal{U}-\mathcal{U}S_\rho\|_{Q_{p_1}\to N_{p_2}}\le C_{p_1}\|\mathcal{L}_{E_\rho}\|_{T_{p_1}\to N_{p_2}}^2\le C_{p_1}C_0^{p_1,p_2}(E_\rho)\longrightarrow0. ∥ U − U S ρ ∥ Q p 1 → N p 2 ≤ C p 1 ∥ L E ρ ∥ T p 1 → N p 2 2 ≤ C p 1 C 0 p 1 , p 2 ( E ρ ) ⟶ 0. Thus U S ρ → U \mathcal{U}S_\rho\to\mathcal{U} U S ρ → U in operator norm. Since each U S ρ \mathcal{U}S_\rho U S ρ is compact, U \mathcal{U} U is compact. □
The capacity criterion and the main theorems
We first combine the two embedding conditions. The endpoint identity and the comparison below then give the three main theorems.
Proposition 3.4. Let p 1 > 0 p_1>0 p 1 > 0 and 0 < p 2 < 1 0<p_2<1 0 < p 2 < 1 . Suppose u ∈ Hol ( D ) u\in\operatorname{Hol}(\mathbb{D}) u ∈ Hol ( D ) and φ : D → D \varphi:\mathbb{D}\to\mathbb{D} φ : D → D is holomorphic. Then W u , φ : Q p 1 → Q p 2 W_{u,\varphi}:Q_{p_1}\to Q_{p_2} W u , φ : Q p 1 → Q p 2 is bounded if and only if
u ∈ Q p 2 , C 0 p 1 , p 2 ( D ) + C 1 p 1 , p 2 ( D ) < ∞ . (30) u\in Q_{p_2},\qquad C_0^{p_1,p_2}(\mathbb{D})+C_1^{p_1,p_2}(\mathbb{D})<\infty. \tag*{(30)} u ∈ Q p 2 , C 0 p 1 , p 2 ( D ) + C 1 p 1 , p 2 ( D ) < ∞. (30) In that case,
∥ W u , φ ∥ 2 ≍ ∣ u ( 0 ) ∣ 2 log 2 e 1 − ∣ φ ( 0 ) ∣ 2 + ∥ u ∥ Q p 2 , ∗ 2 + C 0 p 1 , p 2 ( D ) + C 1 p 1 , p 2 ( D ) . (31) \lVert W_{u,\varphi}\rVert^2\asymp|u(0)|^2\log^2\frac{e}{1-|\varphi(0)|^2}+\lVert u\rVert_{Q_{p_2},*}^2+C_0^{p_1,p_2}(\mathbb{D})+C_1^{p_1,p_2}(\mathbb{D}). \tag*{(31)} ∥ W u , φ ∥ 2 ≍ ∣ u ( 0 ) ∣ 2 log 2 1 − ∣ φ ( 0 ) ∣ 2 e + ∥ u ∥ Q p 2 , ∗ 2 + C 0 p 1 , p 2 ( D ) + C 1 p 1 , p 2 ( D ) . (31) A bounded W u , φ W_{u,\varphi} W u , φ is compact if and only if
lim ρ → 1 [ C 0 p 1 , p 2 ( E ρ ) + C 1 p 1 , p 2 ( E ρ ) ] = 0. \lim_{\rho\to1}\left[C_0^{p_1,p_2}(E_\rho)+C_1^{p_1,p_2}(E_\rho)\right]=0. ρ → 1 lim [ C 0 p 1 , p 2 ( E ρ ) + C 1 p 1 , p 2 ( E ρ ) ] = 0. All comparison constants depend only on p 1 , p 2 p_1,p_2 p 1 , p 2 and the fixed reconstruction parameter.
Proof. Write W = W u , φ W=W_{u,\varphi} W = W u , φ . If W W W is bounded, separation and the two embedding comparisons yield
u = W 1 ∈ Q p 2 , ∥ u ∥ Q p 2 , ∗ ≤ ∥ W ∥ , u=W1\in Q_{p_2},\qquad\lVert u\rVert_{Q_{p_2},*}\le\lVert W\rVert, u = W 1 ∈ Q p 2 , ∥ u ∥ Q p 2 , ∗ ≤ ∥ W ∥ ,
C 0 p 1 , p 2 ( D ) + C 1 p 1 , p 2 ( D ) ≲ ∥ U ∥ 2 + ∥ V ∥ 2 ≲ ∥ W ∥ 2 , C_0^{p_1,p_2}(\mathbb{D})+C_1^{p_1,p_2}(\mathbb{D})\lesssim\lVert U\rVert^2+\lVert V\rVert^2\lesssim\lVert W\rVert^2, C 0 p 1 , p 2 ( D ) + C 1 p 1 , p 2 ( D ) ≲ ∥ U ∥ 2 + ∥ V ∥ 2 ≲ ∥ W ∥ 2 ,
∣ u ( 0 ) ∣ log e 1 − ∣ φ ( 0 ) ∣ 2 ≍ ∣ u ( 0 ) ∣ ∥ ev φ ( 0 ) ∥ Q p 1 ∗ = ∥ ev 0 W ∥ ≤ ∥ W ∥ . |u(0)|\log\frac{e}{1-|\varphi(0)|^2}\asymp|u(0)|\lVert\operatorname{ev}_{\varphi(0)}\rVert_{Q_{p_1}^*}=\lVert\operatorname{ev}_0W\rVert\le\lVert W\rVert. ∣ u ( 0 ) ∣ log 1 − ∣ φ ( 0 ) ∣ 2 e ≍ ∣ u ( 0 ) ∣ ∥ ev φ ( 0 ) ∥ Q p 1 ∗ = ∥ ev 0 W ∥ ≤ ∥ W ∥ .
These are the lower estimates in (31) .
Conversely, suppose condition (30) holds. Using the decomposition f = f ( 0 ) + P b , 0 ( f ′ ) f=f(0)+P_{b,0}(f') f = f ( 0 ) + P b , 0 ( f ′ ) , we obtain
∥ U f ∥ N p 2 ≤ ∣ f ( 0 ) ∣ ∥ u ∥ Q p 2 , ∗ + C p 1 C 0 p 1 , p 2 ( D ) 1 / 2 ∥ f ∥ Q p 1 , ∗ , \lVert Uf\rVert_{N_{p_2}}\le|f(0)|\lVert u\rVert_{Q_{p_2},*}+C_{p_1}C_0^{p_1,p_2}(\mathbb{D})^{1/2}\lVert f\rVert_{Q_{p_1},*}, ∥ U f ∥ N p 2 ≤ ∣ f ( 0 ) ∣ ∥ u ∥ Q p 2 , ∗ + C p 1 C 0 p 1 , p 2 ( D ) 1/2 ∥ f ∥ Q p 1 , ∗ ,
∥ V f ∥ N p 2 ≤ C p 1 C 1 p 1 , p 2 ( D ) 1 / 2 ∥ f ∥ Q p 1 , ∗ , \lVert Vf\rVert_{N_{p_2}}\le C_{p_1}C_1^{p_1,p_2}(\mathbb{D})^{1/2}\lVert f\rVert_{Q_{p_1},*}, ∥ V f ∥ N p 2 ≤ C p 1 C 1 p 1 , p 2 ( D ) 1/2 ∥ f ∥ Q p 1 , ∗ ,
∣ ( W f ) ( 0 ) ∣ ≤ C p 1 ∣ u ( 0 ) ∣ log e 1 − ∣ φ ( 0 ) ∣ 2 ∥ f ∥ Q p 1 . |(Wf)(0)|\le C_{p_1}|u(0)|\log\frac{e}{1-|\varphi(0)|^2}\lVert f\rVert_{Q_{p_1}}. ∣ ( W f ) ( 0 ) ∣ ≤ C p 1 ∣ u ( 0 ) ∣ log 1 − ∣ φ ( 0 ) ∣ 2 e ∥ f ∥ Q p 1 .
Therefore
∥ W f ∥ Q p 2 ≤ ∣ ( W f ) ( 0 ) ∣ + ∥ U f ∥ N p 2 + ∥ V f ∥ N p 2 ≤ C p 1 , p 2 [ ∣ u ( 0 ) ∣ log e 1 − ∣ φ ( 0 ) ∣ 2 + ∥ u ∥ Q p 2 , ∗ + ∑ j = 0 1 C j p 1 , p 2 ( D ) 1 / 2 ] ∥ f ∥ Q p 1 . \begin{aligned}
\lVert Wf\rVert_{Q_{p_2}}\le|(Wf)(0)|+\lVert Uf\rVert_{N_{p_2}}+\lVert Vf\rVert_{N_{p_2}} \\
&\le C_{p_1,p_2}\left[|u(0)|\log\frac{e}{1-|\varphi(0)|^2}+\lVert u\rVert_{Q_{p_2},*}+\sum_{j=0}^{1}C_j^{p_1,p_2}(\mathbb{D})^{1/2}\right]\lVert f\rVert_{Q_{p_1}}.
\end{aligned} ∥ W f ∥ Q p 2 ≤ ∣ ( W f ) ( 0 ) ∣ + ∥ U f ∥ N p 2 + ∥ V f ∥ N p 2 ≤ C p 1 , p 2 [ ∣ u ( 0 ) ∣ log 1 − ∣ φ ( 0 ) ∣ 2 e + ∥ u ∥ Q p 2 , ∗ + j = 0 ∑ 1 C j p 1 , p 2 ( D ) 1/2 ] ∥ f ∥ Q p 1 .
Square this estimate and take the supremum over the unit ball. Assuming boundedness, Theorem 3.1 and Propositions 3.2–3.3 imply
W compact ⟺ U , V : Q p 1 → N p 2 are compact W\text{ compact}\Longleftrightarrow U,V:Q_{p_1}\to N_{p_2}\text{ are compact} W compact ⟺ U , V : Q p 1 → N p 2 are compact
⟺ C 0 p 1 , p 2 ( E ρ ) → 0 and C 1 p 1 , p 2 ( E ρ ) → 0 \Longleftrightarrow C_0^{p_1,p_2}(E_\rho)\to0\quad\text{and}\quad C_1^{p_1,p_2}(E_\rho)\to0 ⟺ C 0 p 1 , p 2 ( E ρ ) → 0 and C 1 p 1 , p 2 ( E ρ ) → 0
⟺ lim ρ → 1 [ C 0 p 1 , p 2 ( E ρ ) + C 1 p 1 , p 2 ( E ρ ) ] = 0. \Longleftrightarrow\lim_{\rho\to1}\left[C_0^{p_1,p_2}(E_\rho)+C_1^{p_1,p_2}(E_\rho)\right]=0. ⟺ ρ → 1 lim [ C 0 p 1 , p 2 ( E ρ ) + C 1 p 1 , p 2 ( E ρ ) ] = 0.
The last equivalence uses nonnegativity of both quantities.
Both limits are taken after the suprema in (2) .
Corollary 3.5. If p 1 > 1 p_1>1 p 1 > 1 , then for every Borel subset E ⊆ D E\subseteq\mathbb{D} E ⊆ D ,
C 1 p 1 , p 2 ( E ) ≍ sup a ∈ D ∫ φ − 1 ( E ) ∣ u ( z ) φ ′ ( z ) ∣ 2 ( 1 − ∣ φ ( z ) ∣ 2 ) 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) . C_1^{p_1,p_2}(E)\asymp\sup_{a\in\mathbb{D}}\int_{\varphi^{-1}(E)}\frac{|u(z)\varphi'(z)|^2}{(1-|\varphi(z)|^2)^2}(1-|\varphi_a(z)|^2)^{p_2}\,dA(z). C 1 p 1 , p 2 ( E ) ≍ a ∈ D sup ∫ φ − 1 ( E ) ( 1 − ∣ φ ( z ) ∣ 2 ) 2 ∣ u ( z ) φ ′ ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) . This applies both to E = D E=\mathbb{D} E = D and to E = E ρ E=E_\rho E = E ρ .
Proof. Let p 1 > 1 p_1>1 p 1 > 1 . Every admissible sequence satisfies d I ≤ ∣ I ∣ p 1 d_I\leq|I|^{p_1} d I ≤ ∣ I ∣ p 1 . For the reverse estimate, set d I = ( 1 − 2 1 − p 1 ) ∣ I ∣ p 1 d_I=(1-2^{1-p_1})|I|^{p_1} d I = ( 1 − 2 1 − p 1 ) ∣ I ∣ p 1 . For any interval J J J ,
∣ J ∣ − p 1 ∑ I ⊆ J d I = ( 1 − 2 1 − p 1 ) ∣ J ∣ − p 1 ∑ n = 0 ∞ ∑ I ⊆ J ∣ I ∣ = 2 − n ∣ J ∣ ∣ I ∣ p 1 = ( 1 − 2 1 − p 1 ) ∑ n = 0 ∞ 2 n 2 − n p 1 = 1. \begin{aligned}
|J|^{-p_1}\sum_{I\subseteq J}d_I=(1-2^{1-p_1})|J|^{-p_1}\sum_{n=0}^{\infty}\sum_{\substack{I\subseteq J\\|I|=2^{-n}|J|}}|I|^{p_1} \\
&=(1-2^{1-p_1})\sum_{n=0}^{\infty}2^n2^{-np_1}=1.
\end{aligned} ∣ J ∣ − p 1 I ⊆ J ∑ d I = ( 1 − 2 1 − p 1 ) ∣ J ∣ − p 1 n = 0 ∑ ∞ I ⊆ J ∣ I ∣ = 2 − n ∣ J ∣ ∑ ∣ I ∣ p 1 = ( 1 − 2 1 − p 1 ) n = 0 ∑ ∞ 2 n 2 − n p 1 = 1.
We therefore have, with values taken in [ 0 , ∞ ] [0,\infty] [ 0 , ∞ ] ,
( 1 − 2 1 − p 1 ) ∑ I ν ( Q I u p ) ∣ I ∣ 2 ≤ ∥ h ν , 2 ∥ B p 1 2 ( D ) 2 ≤ ∑ I ν ( Q I u p ) ∣ I ∣ 2 . (1-2^{1-p_1})\sum_I\frac{\nu(Q_I^{\mathrm{up}})}{|I|^2}\leq\|h_{\nu,2}\|_{B_{p_1}^2(\mathbb{D})}^2\leq\sum_I\frac{\nu(Q_I^{\mathrm{up}})}{|I|^2}. ( 1 − 2 1 − p 1 ) I ∑ ∣ I ∣ 2 ν ( Q I up ) ≤ ∥ h ν , 2 ∥ B p 1 2 ( D ) 2 ≤ I ∑ ∣ I ∣ 2 ν ( Q I up ) .
For any point w ∈ Q I u p w\in Q_I^{\mathrm{up}} w ∈ Q I up ,
∣ I ∣ 2 < 1 − ∣ w ∣ ≤ ∣ I ∣ , ∣ I ∣ 2 < 1 − ∣ w ∣ 2 ≤ 2 ∣ I ∣ . \frac{|I|}{2}<1-|w|\leq|I|,\qquad\frac{|I|}{2}<1-|w|^2\leq2|I|. 2 ∣ I ∣ < 1 − ∣ w ∣ ≤ ∣ I ∣ , 2 ∣ I ∣ < 1 − ∣ w ∣ 2 ≤ 2∣ I ∣.
The upper-half tiles Q I u p Q_I^{\mathrm{up}} Q I up form a partition of the unit disk. It follows that
1 4 ∫ D d ν ( w ) ( 1 − ∣ w ∣ 2 ) 2 ≤ ∑ I ν ( Q I u p ) ∣ I ∣ 2 ≤ 4 ∫ D d ν ( w ) ( 1 − ∣ w ∣ 2 ) 2 . \frac{1}{4}\int_{\mathbb{D}}\frac{d\nu(w)}{(1-|w|^2)^2}\leq\sum_I\frac{\nu(Q_I^{\mathrm{up}})}{|I|^2}\leq4\int_{\mathbb{D}}\frac{d\nu(w)}{(1-|w|^2)^2}. 4 1 ∫ D ( 1 − ∣ w ∣ 2 ) 2 d ν ( w ) ≤ I ∑ ∣ I ∣ 2 ν ( Q I up ) ≤ 4 ∫ D ( 1 − ∣ w ∣ 2 ) 2 d ν ( w ) .
Now apply this estimate to the measure ν = 1 E μ a 1 \nu=\mathbf{1}_E\mu_a^1 ν = 1 E μ a 1 and invoke the pullback identity:
∫ D d ( 1 E μ a 1 ) ( w ) ( 1 − ∣ w ∣ 2 ) 2 = ∫ φ − 1 ( E ) ∣ u ( z ) φ ′ ( z ) ∣ 2 ( 1 − ∣ φ ( z ) ∣ 2 ) 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) . \int_{\mathbb{D}}\frac{d(\mathbf{1}_E\mu_a^1)(w)}{(1-|w|^2)^2}=\int_{\varphi^{-1}(E)}\frac{|u(z)\varphi'(z)|^2}{(1-|\varphi(z)|^2)^2}(1-|\varphi_a(z)|^2)^{p_2}\,dA(z). ∫ D ( 1 − ∣ w ∣ 2 ) 2 d ( 1 E μ a 1 ) ( w ) = ∫ φ − 1 ( E ) ( 1 − ∣ φ ( z ) ∣ 2 ) 2 ∣ u ( z ) φ ′ ( z ) ∣ 2 ( 1 − ∣ φ a ( z ) ∣ 2 ) p 2 d A ( z ) .
Taking the supremum over all a ∈ D a\in\mathbb{D} a ∈ D , we obtain, uniformly with respect to E E E ,
1 − 2 1 − p 1 4 J p 2 ( E ) ≤ C 1 p 1 , p 2 ( E ) ≤ 4 J p 2 ( E ) . (32) \frac{1-2^{1-p_1}}{4}J_{p_2}(E)\leq C_1^{p_1,p_2}(E)\leq4J_{p_2}(E). \tag*{(32)} 4 1 − 2 1 − p 1 J p 2 ( E ) ≤ C 1 p 1 , p 2 ( E ) ≤ 4 J p 2 ( E ) . (32)
□ \square □
Proof of Theorems 1.1–1.3. When 0 < p 1 < 1 0<p_1<1 0 < p 1 < 1 , Proposition 3.4 directly gives Theorem 1.1. For p 1 = 1 p_1=1 p 1 = 1 , Lemma 2.7 together with (24) gives
C 1 1 , p 2 ( D ) = M p 2 ( D ) , C 1 1 , p 2 ( E ρ ) = M p 2 ( E ρ ) . C_1^{1,p_2}(\mathbb{D})=\mathcal{M}_{p_2}(\mathbb{D}),\qquad C_1^{1,p_2}(E_\rho)=\mathcal{M}_{p_2}(E_\rho). C 1 1 , p 2 ( D ) = M p 2 ( D ) , C 1 1 , p 2 ( E ρ ) = M p 2 ( E ρ ) .
Substitution into Proposition 3.4 proves Theorem 1.2, for both boundedness and compactness. If p 1 > 1 p_1>1 p 1 > 1 , the identification
Q p 1 = Q 2 = B , ∥ f ∥ Q p 1 ≍ p 1 ∥ f ∥ Q 2 , Q_{p_1}=Q_2=\mathcal{B},\qquad\|f\|_{Q_{p_1}}\asymp_{p_1}\|f\|_{Q_2}, Q p 1 = Q 2 = B , ∥ f ∥ Q p 1 ≍ p 1 ∥ f ∥ Q 2 ,
shows that the boundedness and compactness properties remain unchanged upon replacing Q p 1 Q_{p_1} Q p 1 by Q 2 Q_2 Q 2 . Applying Proposition 3.4 with p 1 = 2 p_1=2 p 1 = 2 and using (32) , we obtain
W is bounded ⟺ u ∈ Q p 2 , C 0 2 , p 2 ( D ) + J p 2 ( D ) < ∞ , W\text{ is bounded}\ \Longleftrightarrow\ u\in Q_{p_2},\quad C_0^{2,p_2}(\mathbb{D})+\mathcal{J}_{p_2}(\mathbb{D})<\infty, W is bounded ⟺ u ∈ Q p 2 , C 0 2 , p 2 ( D ) + J p 2 ( D ) < ∞ ,
W is compact ⟺ lim ρ ↑ 1 [ C 0 2 , p 2 ( E ρ ) + J p 2 ( E ρ ) ] = 0 provided W is bounded. W\text{ is compact}\ \Longleftrightarrow\ \lim_{\rho\uparrow1}\left[C_0^{2,p_2}(E_\rho)+\mathcal{J}_{p_2}(E_\rho)\right]=0\quad\text{provided }W\text{ is bounded.} W is compact ⟺ ρ ↑ 1 lim [ C 0 2 , p 2 ( E ρ ) + J p 2 ( E ρ ) ] = 0 provided W is bounded.
In this argument, the value capacity is defined with p = 2 p=2 p = 2 and with b b b chosen accordingly. □ \square □
Declarations
Funding. This work was supported by the National Natural Science Foundation of China (Grant No. 12371131).
Conflict of interest. The authors declare no competing interests.
Data availability. Data sharing is not applicable for this article as no datasets were generated or analyzed during the current study.
Use of Generative-AI tools declaration. During the preparation of this work, the authors used an artificial-intelligence tool for language editing, LaTeX formatting and local mathematical reasoning. All mathematical content was written, checked, and approved by the authors, who take full responsibility for the content of the paper.
References
Wujun Cao: Department of Mathematics, Shantou University, 515063, Shantou, Guangdong, P.R. China. Email address: 24wjcao@stu.edu.cn
Zhouyuan Jiang: Department of Mathematics, Shantou University, 515063, Shantou, Guangdong, P.R. China. Email address: 25zyjiang@stu.edu.cn
Songxiao Li: Department of Mathematics, Shantou University, 515063, Shantou, Guangdong, P.R. China. Email address: jyulsx@163.com