We establish on Rn×Rm a Fefferman–Stein type distribution function inequality in which the nontangential maximal function of the product Poisson extension is controlled by the Lusin area integral function: ∣{N>λ}∣≤C(∣{A>λ}∣+λ−2∫{A≤λ}A2dx). The central idea is to establish a local L2 estimate for the radial maximal function. This yields a new proof of the corresponding one-parameter inequality, avoiding Green's formula and the treatment of sawtooth regions, and extends to the multiparameter setting.
∣{N>λ}∣≤C(∣{A>λ}∣+λ−2∫{A≤λ}A2dx).
The central idea is to establish a local L2 estimate for the radial maximal function. This yields a new proof of the corresponding one-parameter inequality, avoiding Green’s formula and the treatment of sawtooth regions, and extends to the multiparameter setting.
Introduction and main results
Maximal functions and square functions are basic tools in the study of Hardy space theory. Comparing these functions is a fundamental problem in this theory, particularly in connection with the following two Fefferman–Stein type distribution function inequalities for the area integral function A and the nontangential maximal function N:
∣{A>λ}∣≤C∣{N>λ}∣+λ2C∫{N≤λ}N2dx.(1)
∣{N>λ}∣≤C∣{A>λ}∣+λ2C∫{A≤λ}A2dx.(2)
Both inequalities have been studied systematically in the one-parameter setting. In the multiparameter setting, however, (1.1) has been extensively investigated, whereas progress on (1.2) has remained elusive. In this paper we study the latter inequality on product spaces and establish (1.2).
In the one-parameter theory, distribution function comparisons in both directions between the area integral function and the nontangential maximal function have long been classical. Burkholder and Gundy [3] studied distribution function inequalities for the area integral using probabilistic and martingale methods, connecting localized distribution control with global integral estimates. Fefferman and Stein [8] established the corresponding real-variable framework in their theory of Hp spaces of several variables. By using Green’s formula and a subtle treatment of sawtooth regions, they obtained distribution estimates in both directions, namely the one-parameter versions of (1.1) and (1.2). Subsequent developments included weighted comparisons, geometric extensions to Lipschitz domains, and sharper ratio estimates. The weighted theory of Gundy and Wheeden [11], Dahlberg’s results on Lipschitz domains [7], and the ratio estimates of Fefferman, Gundy, Silverstein, and Stein [9] illustrate the role of distribution comparisons in weighted theory, complex boundary geometry, and stronger integral control, respectively. Murai and Uchiyama [15] studied good-λ estimates, while the work of Bañuelos and Moore [1, 2] further addressed sharp comparisons and the density of the area integral.
The difficulty in the multiparameter setting arises because the two groups of spatial variables have independent time variables, while the area integral function involves mixed derivatives in the two groups of space-time variables. The Hp theory on the bidisc developed by Malliavin and Malliavin [13] and by Gundy and Stein [10] was pioneering. In [10], Gundy and Stein first established (1.1) and a probabilistic version of (1.2) on the bidisc. Subsequently, Merryfield [14], Sato [16], and Chen [4] (also see Chen [6] for its proof strategy in the one-parameter case) independently obtained distribution estimates of type (1.1), controlling the area integral function by the nontangential maximal function on various product domains;. Chen and Wang [5] extended (1.1) to products of complete Riemannian manifolds with nonnegative Ricci curvature, while Li [12] extended it to product Shilov boundaries. For the reverse distribution inequality (1.2), however, there has been no further breakthrough.
Let d1=n, d2=m, where n,m≥1, and set X=Rd1×Rd2. For i=1,2, let Δi denote the Laplacian in the ith group of variables, and define
Vectors and tensors between the two groups of variables are equipped with their natural Hilbert norms (which are simply Euclidean norms). For α=(α1,α2) and β=(β1,β2)∈(0,∞)2, define
For a nonnegative function F, write dF(λ)=∣{x:F(x)>λ}∣, and set
Θλ(s)=min{s2,λ2},Φλ(F)=∫XΘλ(F(x))dx.(7)
We have the following result.
Theorem 1.1.For every f∈Cc∞(Rn×Rm) and λ>0, we have
Φλ(Nαf)≤CΦλ(Aβf),(8)
where C depends only on the dimensions and on α,β, and Φλ is defined by (1.7). The corresponding one-parameter result holds for Hilbert-valued functions, with a constant independent of the target Hilbert space.
Theorem 1.1 implies the reverse distribution estimate (2) and the corresponding one-sided strong and weak Lp estimates for 0<p<2. The central idea of its proof is to establish the following local L2 estimate for the radial maximal function, thereby reducing the study of the nontangential maximal function to that of the radial maximal function.
Let Mi be the centered Hardy–Littlewood maximal operator on the ith spatial factor Rdi, and let M1M2 denote their composition in the indicated order.
Theorem 1.2. There exists δ∈(0,1), depending only on n,m,β, such that for every measurable E⊂X with ∣Ec∣<∞, the set
GE={x:M1M21Ec(x)≤δ}
satisfies
∣GEc∣≤C∣Ec∣,∫GE(N+f)2dx≤C∫E(Aβf)2dx.(9)
This holds for every f∈L2(X). The corresponding one-parameter Hilbert-valued result also holds.
One-parameter preliminaries
This section presents the one-parameter results needed below.
Lemma 2.1. Suppose c1,…,ck≥0. For all s1,…,sk≥0, we have
Θλ(j=1∑kcjsj)≤kj=1∑kmax{1,cj2}Θλ(sj).(10)
If nonnegative functions Fj→F in L2(X), then Φλ(Fj)→Φλ(F) for every λ>0.
Proof. By the Cauchy–Schwarz inequality, (∑jcjsj)2≤k∑j(cjsj)2. Considering separately the case where some cjsj≥λ and the case where all cjsj<λ, we obtain
min{kj∑(cjsj)2,λ2}≤kj∑min{(cjsj)2,λ2}.
The first assertion follows from Θλ(cs)≤max{1,c2}Θλ(s). Also,
∣Θλ(a)−Θλ(b)∣≤∣a−b∣(a+b),a,b≥0.(11)
Integrating and applying Cauchy–Schwarz inequality gives the final assertion.
For every dimension d≥1, fix a radial function φd∈Cc∞(B(0,1/4)) with φd≥0, ∫φd=1, and φd>0 on B(0,1/8). Set
ψd(x)=φd(x)−2−dφd(x/2).(12)
Then ∫ψd=0 and suppψd⊂B(0,1/2). Choose also a nonnegative χd∈Cc∞(B(0,1)) with ∫χd=1. Write ψd,t(x)=t−dψd(x/t) and χd,t(x)=t−dχd(x/t).
Let H be a Hilbert space (it may be a Euclidean space, whose dimension depends on the gradient and differs between the one- and two-parameter settings, or an L2 space. For simplicity, we use the same notation throughout). Define
where ψ(x)=ψ(−x). This proves (2.6). We next prove (2.7). We can write
EsH(x)=∫0∞(es,t∗Ht)(x)dt,(18)
where es,t=1{t>s}ψt−s−χs∗ψt. Choose a sufficiently large constant C0 and put ar(x)=r−d1{∣x∣<C0r}(x). Taylor’s formula and the compact supports of ψ,χ yield
This value is independent of θ. We have the following reconstruction formula.
Lemma 2.3. Let q∈L2(Rd;H) and v(x,t)=Ptq(x), where Pt is the Poisson semigroup on Rd. For every s>0, the following identity holds in L2:
v(x,s)=−cd−1Ts(∂tv(⋅,t))(x).(20)
Proof. For ξ=0, the Fourier multiplier on the right-hand side of (20) is
−cd−1∫0∞ψd(rξ)(−∣ξ∣)e−(s+r)∣ξ∣dr=e−s∣ξ∣.
Here we use the substitution r∣ξ∣↦r and the definition of cd. Since ∂tv∈Ed(H), Lemma 2.2 justifies (20). □
Lemma 2.4. Let F be a measurable function on R+d+1 with values in a Hilbert space H, and, for c>0, define
ScF(x)=(∫Γc(x)∥F(y,t)∥H2vd(ct)dtdydt)1/2.
For fixed a,b>0, we have
Φλ(SaF)≤CΦλ(SbF).(21)
Proof. The assertion is immediate if the right-hand side is infinite. Otherwise, set E={SbF≤λ}, B=Ec, and G={M1B≤ε}, where M is the centered Hardy–Littlewood maximal operator. Its weak type (1,1) bound gives ∣Gc∣≤Cε−1∣B∣. For any (y,t) such that G∩B(y,at)=∅, choose x∈G∩B(y,at). Since B(y,bt)⊂B(x,(a+b)t),
1−∣B(y,bt)∣∣E∩B(y,bt)∣≤Ca,bM1B(x)≤Ca,bε.
Fix ε sufficiently small that ∣E∩B(y,bt)∣/∣B(y,bt)∣≥1/2. Thus
∣B(y,at)∣∣G∩B(y,at)∣≤1≤2∣B(y,bt)∣∣E∩B(y,bt)∣.
Tonelli’s theorem gives ∫G(SaF)2≤2∫E(SbF)2. Using Φλ(F)≤∫GF2+λ2∣Gc∣ and B={SbF>λ} proves (21). □
Lemma 2.5. Let v be a Hilbert-valued harmonic function on R+d+1. If y∈B(x,at) and η>a, then
∥v(y,t)−v(x,t)∥H≤CAηv(x),(22)
where Aηv is the one-parameter area integral function.
Proof. Choose c>0, depending only on a,η, sufficiently small that every ball of radius ct centered on the horizontal segment from (x,t) to (y,t) lies in Γη(x). By subharmonicity of ∥∇v∥H2 and the mean-value inequality on balls,
Integrating along the horizontal segment proves the assertion. □
We now give the one-parameter Hilbert-valued results required in the proofs of Theorems 1.2 and 1.1. For q∈L2(Rd;H), write v(x,t)=Ptq(x) and define
N+q(x)=t>0sup∥v(x,t)∥H,
Naq(x)=t>0,∣y−x∣<atsup∥v(y,t)∥H,
Abq(x)=(∫Γb(x)∥∇v(y,t)∥H2vd(bt)dtdydt)1/2.
Lemma 2.6. For fixed b>0, there exists δ>0 such that, for any measurable E⊂Rd with ∣Ec∣<∞, the corresponding set GE={x:M1Ec(x)≤δ} satisfies
∣GEc∣≤C∣Ec∣,∫GE(N+q)2dx≤C∫E(Abq)2dx,(23)
with constants independent of the Hilbert space H and b,q,E.
Proof. Put Wb(y,t)=B(y,bt) and ρE(y,t)=∣E∩Wb(y,t)∣/∣Wb(y,t)∣. Tonelli’s theorem yields
∫E(Abq)2dx=∫Rd∫0∞∥∇v(y,t)∥H2ρE(y,t)tdtdy.
Since ∥∂tv∥H≤∥∇v∥H, setting H=1{ρE≥1/2}∂tv gives ∥H∥Ed(H)2≤2∫E(Abq)2dx. For GE={M1Ec≤δ}, the weak type (1,1) bound of M gives ∣GEc∣≤Cδ−1∣Ec∣. If ∣y−x∣<t/2, then B(y,bt)⊂B(x,(b+1/2)t), so
1−ρE(y,t)≤CbM1Ec(x).
Taking δ≤(2Cb)−1 ensures ρE(y,t)≥1/2. The support of the reconstruction kernel ψt−s satisfies ∣y−x∣<(t−s)/2<t/2. Hence, for x∈GE, all derivative evaluation points in the reconstruction lie in {ρE≥1/2}. By the one-parameter reconstruction formula (Lemma 2.3) and Lemma 2.2,
The preceding proposition yields the one-parameter version of (2). Thus Proposition 2.7 gives a new proof of that inequality, avoiding Green’s formula and the treatment of sawtooth regions. We next extend this approach to the multiparameter setting, treating the two-parameter case as an example.
Proofs of the multiparameter theorems
Building on the one-parameter Hilbert-valued estimates in § 2, we establish a product operator estimate, a product reconstruction formula, and geometric lemmas for the product setting, and then prove the main theorems.
Two-parameter reconstruction and product geometric lemmas
Apply the preceding constructions in each spatial factor. Write ψi,r=ψdi,r and χi,r=χdi,r, with corresponding operators Si,Ti,s,Ei,s, positive kernels Ki,t, and averaging operators Vi,sh=χi,s∗ih. For a Hilbert space H, define
The L2 boundedness of M2 therefore gives ∥U10∥22≤C∥H∥En,m(H)2. The third term is treated symmetrically.
For the final term, apply (2.7) successively, first in the first variable and then in the second, moving the suprema over the time variables inside the nonnegative integrals. We obtain
The formula on the right uses only values of u at points satisfying ti>si and ∣yi−xi∣<(ti−si)/2<ti/2.
Proof. First apply Lemma 2.3 in the first spatial variable, treating the second spatial variable and its time variable as Hilbert-valued parameters. Apply the same lemma in the second spatial variable. Multiplying the constants from the two one-parameter reconstructions and canceling the two minus signs gives (3.3). Since ψi,ti−si is supported in B(0,(ti−si)/2), the reconstruction integral uses only values of u in the stated cones.
and β∈(0,∞)2. The identity remains valid if ∇1⊥∇2⊥u(y,t) is replaced by an arbitrary measurable function and Aβ by the square function Sβ in Lemma 3.5 below.
Proof. Substitute the definition of the area integral function and apply Tonelli’s theorem. The details are omitted.
Lemma 3.4. Suppose ai,βi>0. There exists δ>0, depending only on these numbers and the dimensions, such that for every E⊂X with ∣Ec∣<∞, the set GE={x:M1M21Ec(x)≤δ} satisfies
∣GEc∣≤C∣Ec∣,(30)
ρE(y,t)≥21if x∈GE,∣yi−xi∣<aiti(i=1,2).
Proof. Let B=Ec. By the L2 boundedness of M1M2,
∣{M1M21B>δ}∣≤δ−2∥M1M21B∥22≤Cδ−2∣B∣.
For the second estimate, put Bi∗=B(xi,(βi+ai)ti). Then B(yi,βiti)⊂Bi∗, so
Proof. The assertion is immediate if the right-hand side is infinite. Otherwise, set E={SbF≤λ}, B=Ec, and G={M1M21B≤ε}. Then ∣Gc∣≤Cε∣B∣. If G∩Wa(z) is nonempty, choose a point in this intersection as a base point. By Lemma 3.4, for a sufficiently small fixed ε we have ρEb(z)≥1/2, and hence
∣Wa(z)∣∣G∩Wa(z)∣≤2∣Wb(z)∣∣E∩Wb(z)∣.
The inequality also holds when its left-hand side is zero. Tonelli’s theorem gives ∫G(SaF)2≤2∫E(SbF)2. Splitting the integral defining Φλ into the parts over G and Gc proves (31).
Proof of Theorem 1.2
Proof of Theorem 1.2. By a limiting argument, we may assume that f∈Cc∞(Rn×Rm) and u=Pt1(1)Pt2(2)f. In Lemma 3.4, take a1=a2=1/2 and choose GE accordingly. Then
ρE(y,t)≥21whenever x∈GE,∣yi−xi∣<ti/2.
Set
ΩE={(y,t):ρE(y,t)≥1/2},H=1ΩE∂t1∂t2u.
Since ∂t1∂t2 is one component of ∇1⊥∇2⊥, Lemma 3.3 gives
∥H∥En,m2≤2∫E(Aβf)2dx.(32)
For x∈GE, Lemma 3.2 and the preceding geometric property show that every point used in the reconstruction belongs to ΩE. Thus ∂t1∂t2u can be replaced by H in the reconstruction integral. Proposition 3.1 gives
Thus qx2∈L2(Rn;H2) for almost every x2. The Hilbert-valued nontangential maximal function of its Poisson extension in the first variable is Q1(⋅,x2), and the corresponding area integral function is A(β1,η2)f(⋅,x2). Applying Proposition 2.7 for each x2, integrating, and then using Lemma 3.5, we obtain
Φλ(Q1)≤CΦλ(A(β1,η2)f)≤CΦλ(Aβf).
The estimate for Q2 is symmetric: fix x1 and use the one-parameter argument with values in
H1=L2(Rn×(0,∞),vn(η1t1)nt1dy1dt1;Cn+1).
This gives Φλ(Q2)≤CΦλ(Aβf). □
Proof of Theorem 1.1. We may assume that f∈Cc∞(Rn×Rm) and u is its two-parameter Poisson extension. Consider the decomposition
For the first difference, fix z1, regard u(z1,z2) as a harmonic function in the second group of variables, and apply Lemma 2.5 with a=α2 and η=η2. Its right-hand side is controlled by Q1(x). For the second difference, fix z2=(x2,t2) and apply the same lemma in the first group of variables with a=α1 and η=η1; its right-hand side is controlled by Q2(x). Taking suprema over z1,z2 gives
Proof. Since λ2∣{Nαf>λ}∣≤Φλ(Nαf)≤CΦλ(Aβf), splitting the right-hand side over {Aβf≤λ} and {Aβf>λ} gives (37). The strong type assertion follows from Tonelli’s theorem and
∫0∞λp−3min{s2,λ2}dλ=p(2−p)2sp,0<p<2.
For the weak type assertion, writing K=∥F∥Lp,∞, we have
Remark 3.9. For any fixed finite number of factors, one can formally expand the tensor product of the operators ViSi+Ei. Each term must be treated separately using the corresponding positive integrable kernel estimates and iterated maximal operator estimates. Thus estimate (2) extends to the multiparameter setting.
Acknowledgments. The author was supported by the National Natural Science Foundation of China (No. 12571109) and the National Key Research and Development Program of the Ministry of Science and Technology of China (No. 2022YFA1005700).
References
[1]R. Bañuelos and C. N. Moore, Sharp estimates for the nontangential maximal function and the Lusin area function in Lipschitz domains, Trans. Amer. Math. Soc. 312 (1989), no. 2, 641–662.
[2]R. Bañuelos and C. N. Moore, Distribution function inequalities for the density of the area integral, Ann. Inst. Fourier 41 (1991), no. 1, 137–171.DOI
[3]D. L. Burkholder and R. F. Gundy, Distribution function inequalities for the area integral, Studia Math. 44 (1972), no. 6, 527–544.DOI
[4]J. Chen, H^p spaces related to multiparameters and their characterizations (in Chinese), M.Sc. thesis, Hangzhou University, 1984.
[5]J. Chen and S. Wang, Comparison between maximal function and square function on positively curved product manifolds, in Chinese Mathematics into the 21st Century (Wu Wen-tsun and Cheng Min-de, eds.), Peking University Press, 1991.
[6]J. Chen, A simple proof of an inequality dominating the area integral by the nontangential maximal function, Pacific J. Math. 190 (1999), no. 1, 41–44.DOI
[7]B. E. J. Dahlberg, Weighted norm inequalities for the Lusin area integral and the nontangential maximal functions for functions harmonic in a Lipschitz domain, Studia Math. 67 (1980), no. 3, 297–314.
[8]C. Fefferman and E. M. Stein, H^p spaces of several variables, Acta Math. 129 (1972), 137–193.
[9]R. Fefferman, R. F. Gundy, M. L. Silverstein and E. M. Stein, Inequalities for ratios of functionals of harmonic functions, Proc. Natl. Acad. Sci. USA 79 (1982), no. 24, 7958–7960.pubmed.ncbi.nlm.nih.gov/16593265
[10]R. F. Gundy and E. M. Stein, Hp theory for the poly-disc, Proc. Natl. Acad. Sci. USA 76 (1979), no. 3, 1026–1029.DOI
[11]R. F. Gundy and R. L. Wheeden, Weighted integral inequalities for the nontangential maximal function, Lusin area integral, and Walsh–Paley series, Studia Math. 49 (1974), no. 2, 107–124.DOI
[12]J. Li, On Fefferman–Stein type inequality for Shilov boundaries and applications, Ann. Sc. Norm. Super. Pisa Cl. Sci. 27 (2026), no. 3, 1329–1354.
[13]M.-P. Malliavin and P. Malliavin, Intégrales de Lusin-Calderón pour les fonctions biharmoniques, Bull. Sci. Math. (2) 101 (1977), 357–384.
[14]K. G. Merryfield, On the area integral, Carleson measures and Hp in the polydisc, Indiana Univ. Math. J. 34 (1985), no. 3, 663–685.
[15]T. Murai and A. Uchiyama, Good λ inequalities for the area integral and the nontangential maximal function, Studia Math. 83 (1986), no. 3, 251–262.DOI
[16]S. Sato, Lusin functions and nontangential maximal functions in the Hp theory on the product of upper half-spaces, Tohoku Math. J. (2) 37 (1985), no. 1, 1–13.DOI