Introduction
Motivation and background
Let S ⊂ R n + 1 S \subset\mathbb{R}^{n+1} S ⊂ R n + 1 be a smooth hypersurface and let d σ ( y ˉ ) d\sigma(\bar{y}) d σ ( y ˉ ) be a smooth measure supported in a compact subset S 0 S_0 S 0 of S S S . For a Schwartz function f f f we consider the family of averaging operators
A t f ( x ˉ ) = ∫ S f ( x ˉ − t y ˉ ) d σ ( y ˉ ) , t > 0 , A_t f(\bar{x})=\int_S f(\bar{x}-t\bar{y})\,d\sigma(\bar{y}), \qquad t>0, A t f ( x ˉ ) = ∫ S f ( x ˉ − t y ˉ ) d σ ( y ˉ ) , t > 0 ,
where x ˉ = ( x , x n + 1 ) ∈ R n + 1 \bar{x}=(x,x_{n+1})\in\mathbb{R}^{n+1} x ˉ = ( x , x n + 1 ) ∈ R n + 1 . The study of such operators has a long history in harmonic analysis, beginning with the seminal works of Stein [27 ] (n > 1 n>1 n > 1 ), and Bourgain [2 ] (n = 1 n=1 n = 1 ) on spherical maximal functions. A fundamental question, raised in [28 ] , is to determine the range of exponents p p p for which the maximal operator M f = sup t > 0 ∣ A t f ∣ Mf=\sup_{t>0}|A_t f| M f = sup t > 0 ∣ A t f ∣ is bounded on L p L^p L p .
For general hypersurfaces the curvature plays a decisive role. Greenleaf [8 ] proved that if at least k k k principal curvatures are non-zero everywhere, then the maximal operator is bounded for p > ( k + 1 ) / k p>(k+1)/k p > ( k + 1 ) / k . The case k = 1 k=1 k = 1 was later treated by Sogge [29 ] . For convex hypersurfaces of finite type, Iosevich and Sawyer [13 , 14 ] characterized L p L^p L p -boundedness in terms of the distance to tangent planes, and refined results relating the boundedness index to the oscillation index of the surface were obtained by Ikromov, Müller and others [10 , 5 , 6 , 11 ] .
In recent years the attention has shifted to stronger variational norm estimates, which capture oscillation along the parameter and directly imply pointwise convergence without requiring a dense class. For a family { a t ( x ˉ ) } t > 0 \{a_t(\bar{x})\}_{t>0} { a t ( x ˉ ) } t > 0 and r ∈ [ 1 , ∞ ) r \in[1,\infty) r ∈ [ 1 , ∞ ) the r r r -variation seminorm is defined by
V r ( a t : t > 0 ) ( x ˉ ) = sup N ∈ N sup t 1 < ⋯ < t N ( ∑ i = 1 N − 1 ∣ a t i + 1 ( x ˉ ) − a t i ( x ˉ ) ∣ r ) 1 / r . V_r(a_t:t>0)(\bar{x})=\sup_{N\in\mathbb{N}}\sup_{t_1<\cdots<t_N}\left(\sum_{i=1}^{N-1}\lvert a_{t_{i+1}}(\bar{x})-a_{t_i}(\bar{x})\rvert^r\right)^{1/r}. V r ( a t : t > 0 ) ( x ˉ ) = N ∈ N sup t 1 < ⋯ < t N sup ( i = 1 ∑ N − 1 ∣ a t i + 1 ( x ˉ ) − a t i ( x ˉ ) ∣ r ) 1/ r .
with the obvious modification for r = ∞ r=\infty r = ∞ . The concept of r r r -variation was first used in harmonic analysis by Bourgain [3 ] in the context of ergodic averages; its roots lie in the probability and martingale theory works of Lépingle [15 ] and Pisier-Xu [26 ] . Subsequent work by Jones, Seeger and Wright [16 ] established sharp variational bounds for spherical means, showing that V r ( A t : t > 0 ) V_r(A_t:t>0) V r ( A t : t > 0 ) is bounded on L p L^p L p for ( n + 1 ) / n < p < r ( n + 1 ) (n+1)/n<p<r(n+1) ( n + 1 ) / n < p < r ( n + 1 ) , with the upper bound p < r ( n + 1 ) p<r(n+1) p < r ( n + 1 ) being necessary. Endpoint estimates were later obtained by Beltran, Oberlin, Roncal and Seeger [1 ] . In a recent paper, Nie [25 ] studied variational operators associated with planar curves of finite type, obtaining sharp L p L^p L p bounds for isotropic and non-isotropic dilations; that work extended earlier results of Iosevich [12 ] and Li [17 ] to the variational setting. The present article continues this line of research by considering averages over higher-dimensional hypersurfaces, allowing the curvature to vanish in a controlled way.
In addition to the aforementioned results on hypersurfaces, sharp variation-norm estimates for oscillatory integrals related to Carleson’s theorem were obtained by Guo, Roos, and Yung [9 ] , and variational inequalities for generalized spherical means were established by Li, Yang, and Zhang [18 ] . Dimension-free variational inequalities for averaging operators have been studied by Bourgain, Mirek, Stein and collaborators [4 ] , while discrete analogues have been systematically developed by Mirek, Stein, Trojan and Zorin-Kranich [19 , 20 , 21 , 22 ] . These developments highlight the growing interest in variational estimates for a wide class of averaging operators and their applications to pointwise convergence problems.
Main results
Let Φ ∈ C ∞ ( R n ) \Phi\in C^\infty(\mathbb{R}^n) Φ ∈ C ∞ ( R n ) satisfy Φ ( 0 ) = 0 \Phi(0)=0 Φ ( 0 ) = 0 and ∇ Φ ( 0 ) = 0 \nabla\Phi(0)=0 ∇Φ ( 0 ) = 0 . For a fixed D ∈ R + D\in\mathbb{R}^+ D ∈ R + define the averaging operator
A t f ( x , x n + 1 ) = ∫ R n f ( x − t y , x n + 1 − t ( D + Φ ( y ) ) ) ρ ( y ) d y . A_t f(x,x_{n+1})=\int_{\mathbb{R}^n}f\bigl(x-t y,x_{n+1}-t(D+\Phi(y))\bigr)\rho(y)\,dy. A t f ( x , x n + 1 ) = ∫ R n f ( x − t y , x n + 1 − t ( D + Φ ( y )) ) ρ ( y ) d y .
where ρ \rho ρ is a smooth cutoff supported near the origin. Our first result concerns hypersurfaces with a prescribed number of non-zero principal curvatures.
Theorem 1.1 Suppose the hypersurface S = { ( x , x n + 1 ) : x n + 1 = Φ ( x ) + D } S=\{(x,x_{n+1}):x_{n+1}=\Phi(x)+D\} S = {( x , x n + 1 ) : x n + 1 = Φ ( x ) + D } has exactly k k k non-zero principal curvatures at the point ( 0 , D ) (0,D) ( 0 , D ) (with 1 ≤ k ≤ n 1\leq k\leq n 1 ≤ k ≤ n ). Then for every r > 2 r>2 r > 2 the r r r -variation operator satisfies
∥ V r ( A t f : t > 0 ) ∥ L p ( R n + 1 ) ≤ C ( p , r , D ) ∥ f ∥ L p ( R n + 1 ) (1) \lVert V_r(A_t f:t>0)\rVert_{L^p(\mathbb{R}^{n+1})}\leq C(p,r,D)\lVert f\rVert_{L^p(\mathbb{R}^{n+1})} \tag*{(1)} ∥ V r ( A t f : t > 0 ) ∥ L p ( R n + 1 ) ≤ C ( p , r , D ) ∥ f ∥ L p ( R n + 1 ) (1) for all
( k + 1 ) / k < p < r ( k + 1 ) . (k+1)/k<p<r(k+1). ( k + 1 ) / k < p < r ( k + 1 ) .
where the operator norm satisfies the estimate
C ( p , r , D ) ≤ C ε ( 1 + D 1 / ( r ∧ p ) + ε ) (2) C(p,r,D) \le C_{\varepsilon}\left(1+D^{1/(r\wedge p)+\varepsilon}\right) \tag*{(2)} C ( p , r , D ) ≤ C ε ( 1 + D 1/ ( r ∧ p ) + ε ) (2)
and the growth order 1 / ( r ∧ p ) 1/(r\wedge p) 1/ ( r ∧ p ) in D D D is optimal up to an arbitrarily small ε \varepsilon ε . Conversely, when k = n k=n k = n , if there exists a constant C C C such that
∥ V r ( A t f : t > 0 ) ∥ L p ( R n + 1 ) ≤ C ∥ f ∥ L p ( R n + 1 ) , \left\|V_r(A_t f:t>0)\right\|_{L^p(\mathbb{R}^{n+1})} \le C\|f\|_{L^p(\mathbb{R}^{n+1})}, ∥ V r ( A t f : t > 0 ) ∥ L p ( R n + 1 ) ≤ C ∥ f ∥ L p ( R n + 1 ) ,
then p ≤ r ( n + 1 ) p\le r(n+1) p ≤ r ( n + 1 ) .
When k = n k=n k = n (i.e., non-vanishing Gaussian curvature) this recovers and extends the classical spherical variational estimates of Jones-Seeger-Wright [16 ] . Moreover, the case k < n k<n k < n was not previously known for general hypersurfaces. Theorem 1.1 thus provides a simultaneous generalization of Greenleaf’s maximal inequality [8 ] and the variational results for spheres.
Our second result deals with convex hypersurfaces of finite type, a class intensively studied by Iosevich and Sawyer [13 ] . In the proof of this result, the sharp estimate for the dependence on the translation parameter D D D obtained in Theorem 1.1 will play a crucial role.
Theorem 1.2. Let Φ \Phi Φ be a smooth convex function and let S = { ( x , x n + 1 ) : x n + 1 = Φ ( x ) + D } S=\{(x,x_{n+1}):x_{n+1}=\Phi(x)+D\} S = {( x , x n + 1 ) : x n + 1 = Φ ( x ) + D } be its graph. Assume S S S has no tangent line of infinite order at the point ( 0 , D ) (0,D) ( 0 , D ) . Denote by p 0 p_0 p 0 the contact index of S S S at ( 0 , D ) (0,D) ( 0 , D ) , defined by
p 0 = inf p { p : ∫ Φ ( y ) − 1 / p ρ ( y ) d y < ∞ } (3) p_0=\inf_{p}\left\{p:\int\Phi(y)^{-1/p}\rho(y)\,dy<\infty\right\} \tag*{(3)} p 0 = p inf { p : ∫ Φ ( y ) − 1/ p ρ ( y ) d y < ∞ } (3) and suppose p 0 > 2 p_0>2 p 0 > 2 . Then for every r > p 0 r>p_0 r > p 0 the estimate
∥ V r ( A t f : t > 0 ) ∥ L p ( R n + 1 ) ≤ C ∥ f ∥ L p ( R n + 1 ) \left\|V_r(A_t f:t>0)\right\|_{L^p(\mathbb{R}^{n+1})}\le C\|f\|_{L^p(\mathbb{R}^{n+1})} ∥ V r ( A t f : t > 0 ) ∥ L p ( R n + 1 ) ≤ C ∥ f ∥ L p ( R n + 1 ) holds for p 0 < p < 2 r p_0<p<2r p 0 < p < 2 r .
Strategies and novelties
The proof of Theorem 1.1 follows a well-established strategy for variational estimates which used in [16 , 25 , 9 , 18 ] , which involves a long-short decomposition of the variation operator. The novelty of our approach lies in a meticulous and sharp analysis of the dependence on the translation parameter D D D .
Long-Short Decomposition: We first decompose the r r r -variation into two parts: the “long variation”, which captures changes across dyadic scales, and the “short variation”, which captures changes within each dyadic scale. The long variation is handled by straightforward decay estimates for the Fourier transform of the surface measure, leading to a simple bound.
Reduction to L t 2 L^2_t L t 2 Estimates: The core of the proof is the analysis of the short variation. We employ a powerful pointwise estimate that bounds the short variation at a dyadic scale by the product of an L t 2 L^2_t L t 2 norm of the family and an L t 2 L^2_t L t 2 norm of its time derivative. This cleverly reduces the problem to proving L t 2 L^2_t L t 2 estimates for the averaging operators A k , t f A_{k,t}f A k , t f and their derivatives.
Vector-Valued Fourier Multiplier and Local Smoothing: To obtain these L t 2 L^2_t L t 2 estimates, we combine several harmonic analysis tools. Plancherel’s theorem translates the problem into frequency space. A detailed stationary phase analysis provides precise asymptotic for the Fourier multiplier m m m associated with the hypersurface. This, in turn, allows us to apply a vector-valued multiplier theorem and, crucially, local smoothing estimates for Fourier integral operators. These estimates are essential for handling the curvature and achieving the full range of exponents p p p .
Key Novelty - Sharp Dependence on D D D : A critical innovation in our argument is the careful tracking of the translation parameter D D D through every step—from the stationary phase expansion, through the multiplier estimates, to the final L p L^p L p bounds. This level of detail is not present in previous works. The resulting estimate C ( p , r , D ) ≲ 1 + D 1 r ∧ p + ϵ C(p,r,D) \lesssim1+D^{\frac{1}{r\wedge p}+\epsilon} C ( p , r , D ) ≲ 1 + D r ∧ p 1 + ϵ is not only optimal in its growth order, as we prove via counterexamples, but also becomes an indispensable tool for extending our results to more complex surfaces. This sharp D D D -dependence is what allows us to apply a scaling argument to convex finite-type hypersurfaces in Theorem 1.2 , where the translation parameter is effectively scaled up to 2 k D 2^kD 2 k D .
Organisation of the paper
The paper is structured as follows.
Section 2: Preliminaries. We introduce the key notational conventions used throughout the paper. We then define the long and short variational decompositions and recall a fundamental estimate for bounding the short variation in terms of L t 2 L^2_t L t 2 norms.
Section 3: The case k = n k=n k = n . This section is dedicated to the core proof of Theorem 1.1 for hypersurfaces with non-vanishing Gaussian curvature. It is divided into several subsections that handle the long variation, the short variation, and the necessary L 2 L^2 L 2 , L 1 L^1 L 1 , and L p L^p L p estimates for the decomposed operators. It concludes with the proof of the necessary condition p ≤ r ( n + 1 ) p \le r(n+1) p ≤ r ( n + 1 ) .
Section 4: The case k < n k<n k < n . Here, we extend the result of § 5 to hypersurfaces with only k k k non-zero principal curvatures. The main tool is a rotation argument (Lemma 4.1 ), which averages the operator over rotations in the “flat” directions, reducing the problem to the previously established full-curvature case.
Section 5: Convex finite-type surfaces. We prove Theorem 1.2 for convex hypersurfaces of finite type. This is achieved by using a non-isotropic dyadic decomposition and a powerful scaling argument. The scaling property, combined with the sharp D D D -dependence from Theorem 1.1 , yields the desired variational bounds.
Section 6: Appendix. The appendix contains the proof of the auxiliary estimates for the Poisson-like operator S t S_t S t , which were used in § 5 to simplify the main operator.
Preliminaries
This section collects the essential definitions, notations, and preliminary lemmas that will be used throughout the paper.
Notation
For any two non-negative quantities x , y x,y x , y we write x ≲ y x\lesssim y x ≲ y (or x ≳ y x\gtrsim y x ≳ y ) to denote that there exists an absolute constant C > 0 C>0 C > 0 such that x ≤ C y x\le Cy x ≤ C y (or x ≥ C y x\ge Cy x ≥ C y ). If the constant depends on a parameter, e.g., ϵ \epsilon ϵ , we write x ≲ ϵ y x\lesssim_\epsilon y x ≲ ϵ y .
For a sequence { a k } k ∈ Z \{a_k\}_{k\in\mathbb{Z}} { a k } k ∈ Z , we denote its ℓ 2 \ell^2 ℓ 2 -norm by ∥ a k ∥ ℓ k r : = ( ∑ k ∈ Z ∣ a k ∣ r ) 1 / r \|a_k\|_{\ell_k^r}:=\left(\sum_{k\in\mathbb{Z}}|a_k|^r\right)^{1/r} ∥ a k ∥ ℓ k r := ( ∑ k ∈ Z ∣ a k ∣ r ) 1/ r . For a family of functions indexed by a continuous parameter t > 0 t>0 t > 0 , we use the notation ∥ a t ∥ L t r : = ( ∫ 0 ∞ ∣ a t ∣ r d t t ) 1 / r \|a_t\|_{L_t^r}:=\left(\int_0^\infty|a_t|^r\frac{dt}{t}\right)^{1/r} ∥ a t ∥ L t r := ( ∫ 0 ∞ ∣ a t ∣ r t d t ) 1/ r .
We fix a smooth cutoff function ψ : R → [ 0 , 1 ] \psi:\mathbb{R}\to[0,1] ψ : R → [ 0 , 1 ] supported in [ 1 / 2 , 2 ] [1/2,2] [ 1/2 , 2 ] such that ∑ j ∈ Z ψ j ( t ) = 1 \sum_{j\in\mathbb{Z}}\psi_j(t)=1 ∑ j ∈ Z ψ j ( t ) = 1 for all t > 0 t>0 t > 0 . We set ψ j ( t ) = ψ ( 2 − j t ) \psi_j(t)=\psi\left(2^{-j}t\right) ψ j ( t ) = ψ ( 2 − j t ) . This partition of unity allows us to define dyadic frequency projections on R n + 1 \mathbb{R}^{n+1} R n + 1 . For each j ∈ Z j\in\mathbb{Z} j ∈ Z , let P j P_j P j be the Fourier multiplier operator defined by
P j f ^ ( ξ , ξ n + 1 ) = ψ j ( ξ , ξ n + 1 ) f ^ ( ξ , ξ n + 1 ) . \widehat{P_jf}(\xi,\xi_{n+1})=\psi_j(\xi,\xi_{n+1})\widehat{f}(\xi,\xi_{n+1}). P j f ( ξ , ξ n + 1 ) = ψ j ( ξ , ξ n + 1 ) f ( ξ , ξ n + 1 ) .
The Long-Short decomposition
A key tool in proving variational estimates is the decomposition of the variation into “long" and “short" parts. For a family of operators { A t } t > 0 \{A_t\}_{t>0} { A t } t > 0 , define the short variation on the dyadic interval [ 2 j , 2 j + 1 ] [2^j,2^{j+1}] [ 2 j , 2 j + 1 ] as
V r j ( A t f ( x ‾ ) ) : = V r ( A t f ( x ‾ ) : t ∈ [ 2 j , 2 j + 1 ] ) . V_r^j(A_tf(\overline{x})):=V_r\left(A_tf(\overline{x}):t\in[2^j,2^{j+1}]\right). V r j ( A t f ( x )) := V r ( A t f ( x ) : t ∈ [ 2 j , 2 j + 1 ] ) .
The total short variation is then given by the l r l^r l r -sum of these dyadic contributions:
S r ( A t f ( x ‾ ) : t > 0 ) : = ( ∑ j ∈ Z ∣ V r j ( A t f ( x ‾ ) ) ∣ r ) 1 / r . S_r(A_tf(\overline{x}):t>0):=\left(\sum_{j\in\mathbb{Z}}\left|V_r^j(A_tf(\overline{x}))\right|^r\right)^{1/r}. S r ( A t f ( x ) : t > 0 ) := j ∈ Z ∑ V r j ( A t f ( x )) r 1/ r .
The following lemma justifies this decomposition.
Lemma 2.1 For any r ∈ [ 1 , ∞ ) r\in[1,\infty) r ∈ [ 1 , ∞ ) , there exists a constant C r > 0 C_r>0 C r > 0 such that
V r ( A t f ( x ‾ ) : t > 0 ) ≤ C r S r ( A t f ( x ‾ ) : t > 0 ) + C r V r ( A 2 j f ( x ‾ ) : j ∈ Z ) . (4) V_r(A_tf(\overline{x}):t>0)\le C_rS_r(A_tf(\overline{x}):t>0)+C_rV_r(A_{2^j}f(\overline{x}):j\in\mathbb{Z}). \tag*{(4)} V r ( A t f ( x ) : t > 0 ) ≤ C r S r ( A t f ( x ) : t > 0 ) + C r V r ( A 2 j f ( x ) : j ∈ Z ) . (4) where V r ( A 2 j f ( x ‾ ) : j ∈ Z ) V_r(A_{2^j}f(\overline{x}):j\in\mathbb{Z}) V r ( A 2 j f ( x ) : j ∈ Z ) is the r r r -variation of the sequence of operators sampled at the dyadic points t = 2 j t=2^j t = 2 j .
The first term on the right-hand side captures the oscillation of A t f A_tf A t f within each dyadic scale, while the second term captures the longer-range variation between scales. This decomposition reduces the problem to estimating these two separate components.
A pointwise estimate for the short variation
A crucial estimate for controlling the short variation is provided by the following proposition. It bounds the short variation at a given scale by the L t r L_t^r L t r norms of the function and its time derivative.
Proposition 2.2 For any r > 1 r>1 r > 1 , let r ′ r' r ′ be its conjugate exponent (i.e., 1 r + 1 r ′ = 1 \frac{1}{r}+\frac{1}{r'}=1 r 1 + r ′ 1 = 1 ). Then there exists a constant C r > 0 C_r>0 C r > 0 such that for any differentiable family { a t } t > 0 \{a_t\}_{t>0} { a t } t > 0 ,
S r ( a t : t > 0 ) ≤ C r ∥ a t ∥ L t r 1 / r ∥ t ⋅ a t ′ ∥ L t r ′ 1 / r ′ . (5) S_r(a_t:t>0)\le C_r\lVert a_t\rVert_{L_t^r}^{1/r}\lVert t\cdot a_t'\rVert_{L_t^{r'}}^{1/r'}. \tag*{(5)} S r ( a t : t > 0 ) ≤ C r ∥ a t ∥ L t r 1/ r ∥ t ⋅ a t ′ ∥ L t r ′ 1/ r ′ . (5)
A more convenient form for our purposes, which follows directly from Hölder’s inequality, is the following pointwise estimate for the dyadic short variation:
S 2 ( A t f ( x ‾ ) : t > 0 ) ≲ ( ∫ 0 ∞ ∣ A t f ( x ‾ ) ∣ 2 d t t ) 1 / 4 ( ∫ 0 ∞ ∣ t d d t A t f ( x ‾ ) ∣ 2 d t t ) 1 / 4 . S_2(A_tf(\overline{x}):t>0)\lesssim\left(\int_0^\infty|A_tf(\overline{x})|^2\frac{dt}{t}\right)^{1/4}\left(\int_0^\infty\left|t\frac{d}{dt}A_tf(\overline{x})\right|^2\frac{dt}{t}\right)^{1/4}. S 2 ( A t f ( x ) : t > 0 ) ≲ ( ∫ 0 ∞ ∣ A t f ( x ) ∣ 2 t d t ) 1/4 ( ∫ 0 ∞ t d t d A t f ( x ) 2 t d t ) 1/4 .
This inequality is fundamental as it transforms the problem of estimating a complicated variational norm into the more tractable problem of estimating L t 2 L_t^2 L t 2 norms of the operator and its derivative. The majority of the technical work in Section 3 will be dedicated to establishing L p L^p L p bounds for these L 2 L^2 L 2 norms.
The proof of Theorem 1.1: the case k = n k=n k = n
In this section we consider the case k = n k=n k = n in Theorem 1.1 . Since the norm ∥ ⋅ ∥ L p → L p \|\,\cdot\,\|_{L^p\to L^p} ∥ ⋅ ∥ L p → L p of operator V r ( A t : t > 0 ) V_r(A_t:t>0) V r ( A t : t > 0 ) remains unchanged under orthogonal transformation, we can assume
Φ ( x ) = ∑ i = 1 n λ i x i 2 + R ( x ) \Phi(x)=\sum_{i=1}^{n}\lambda_i x_i^2+R(x) Φ ( x ) = i = 1 ∑ n λ i x i 2 + R ( x )
where λ i ≠ 0 \lambda_i\ne0 λ i = 0 and D α R ( 0 ) = 0 D^\alpha R(0)=0 D α R ( 0 ) = 0 for ∣ α ∣ ≤ 2 |\alpha|\le2 ∣ α ∣ ≤ 2 . Without loss of generality, assume ∫ R n ρ ( x ) d x = 1 \int_{\mathbb{R}^n}\rho(x)dx=1 ∫ R n ρ ( x ) d x = 1 . Thus, if we let μ ( ξ , ξ n + 1 ) = ∫ R n e − i ( x ⋅ ξ + ξ n + 1 ( D + Φ ( x ) ) ) ρ ( x ) d x \mu(\xi,\xi_{n+1})=\int_{\mathbb{R}^n}e^{-i(x\cdot\xi+\xi_{n+1}(D+\Phi(x)))}\rho(x)dx μ ( ξ , ξ n + 1 ) = ∫ R n e − i ( x ⋅ ξ + ξ n + 1 ( D + Φ ( x ))) ρ ( x ) d x , then we obtain μ ( 0 , 0 ) = 1 \mu(0,0)=1 μ ( 0 , 0 ) = 1 .
In this section we shall fix r ∈ ( 2 , ∞ ) r\in(2,\infty) r ∈ ( 2 , ∞ ) . Denote
F ( S t f ) ( ξ , ξ n + 1 ) = e − t ∣ ( ξ , ξ n + 1 ) ∣ e − i t D ξ n + 1 f ^ ( ξ , ξ n + 1 ) \mathcal{F}(S_tf)(\xi,\xi_{n+1})=e^{-t|(\xi,\xi_{n+1})|}e^{-itD_{\xi_{n+1}}}\widehat{f}(\xi,\xi_{n+1}) F ( S t f ) ( ξ , ξ n + 1 ) = e − t ∣ ( ξ , ξ n + 1 ) ∣ e − i t D ξ n + 1 f ( ξ , ξ n + 1 )
and
F ( A t f ) ( ξ , ξ n + 1 ) = m ( t ξ , t ξ n + 1 ) f ^ ( ξ , ξ n + 1 ) \mathcal{F}(A_tf)(\xi,\xi_{n+1})=m(t\xi,t\xi_{n+1})\widehat{f}(\xi,\xi_{n+1}) F ( A t f ) ( ξ , ξ n + 1 ) = m ( t ξ , t ξ n + 1 ) f ( ξ , ξ n + 1 )
where m ( ξ , ξ n + 1 ) = μ ( ξ , ξ n + 1 ) − e − ∣ ( ξ , ξ n + 1 ) ∣ e − i D ξ n + 1 m(\xi,\xi_{n+1})=\mu(\xi,\xi_{n+1})-e^{-|(\xi,\xi_{n+1})|}e^{-iD_{\xi_{n+1}}} m ( ξ , ξ n + 1 ) = μ ( ξ , ξ n + 1 ) − e − ∣ ( ξ , ξ n + 1 ) ∣ e − i D ξ n + 1 . For every f ∈ L p ( R n + 1 ) f\in L^p(\mathbb{R}^{n+1}) f ∈ L p ( R n + 1 ) , we obtain
∥ V r ( A t f : t > 0 ) ∥ L p ≤ C r ∥ V r ( S t f : t > 0 ) ∥ L p + C r ∥ V r ( A t f : t > 0 ) ∥ L p . (6) \|V_r(A_tf:t>0)\|_{L^p}\le C_r\|V_r(S_tf:t>0)\|_{L^p}+C_r\|V_r(A_tf:t>0)\|_{L^p}. \tag*{(6)} ∥ V r ( A t f : t > 0 ) ∥ L p ≤ C r ∥ V r ( S t f : t > 0 ) ∥ L p + C r ∥ V r ( A t f : t > 0 ) ∥ L p . (6)
In Appendix, we shall prove the estimate
∥ V r ( S t f : t > 0 ) ∥ L p ≤ C ( p , r , D ) ∥ f ∥ L p \|V_r(S_tf:t>0)\|_{L^p}\le C(p,r,D)\|f\|_{L^p} ∥ V r ( S t f : t > 0 ) ∥ L p ≤ C ( p , r , D ) ∥ f ∥ L p
for p > 1 p>1 p > 1 , where C ( p , r , D ) C(p,r,D) C ( p , r , D ) satisfies (1.2). Therefore, it remains to obtain L p L^p L p bounds for the second term in (3.6). For this purpose, we use (2.4). Indeed, we shall show that,
∥ V r ( A 2 j f : j ∈ Z ) ∥ L p ≤ C ∥ f ∥ L p (7) \|V_r(A_{2^j}f:j\in\mathbb{Z})\|_{L^p}\le C\|f\|_{L^p} \tag*{(7)} ∥ V r ( A 2 j f : j ∈ Z ) ∥ L p ≤ C ∥ f ∥ L p (7)
for p > 1 p>1 p > 1 and
∥ S r ( A t f : t > 0 ) ∥ L p ≤ C ( p , r , D ) ∥ f ∥ L p , (8) \|S_r(A_tf:t>0)\|_{L^p}\le C(p,r,D)\|f\|_{L^p}, \tag*{(8)} ∥ S r ( A t f : t > 0 ) ∥ L p ≤ C ( p , r , D ) ∥ f ∥ L p , (8)
for ( n + 1 ) / n < p < r ( n + 1 ) (n+1)/n<p<r(n+1) ( n + 1 ) / n < p < r ( n + 1 ) .
Estimates for long variations: proof of (3.7)
Recall that V r ( A 2 j f : j ∈ Z ) ≤ ∥ A 2 j f ∥ ℓ r ≤ ∥ A 2 j f ∥ ℓ 2 V_r(A_{2^j}f:j\in\mathbb{Z})\le\|A_{2^j}f\|_{\ell^r}\le\|A_{2^j}f\|_{\ell^2} V r ( A 2 j f : j ∈ Z ) ≤ ∥ A 2 j f ∥ ℓ r ≤ ∥ A 2 j f ∥ ℓ 2 . To prove (3.7) we shall show the following bounds
∥ A 2 j f ∥ L p ( ℓ j 2 ) ≤ C ∥ f ∥ L p , (9) \|A_{2^j}f\|_{L^p(\ell_j^2)}\le C\|f\|_{L^p}, \tag*{(9)} ∥ A 2 j f ∥ L p ( ℓ j 2 ) ≤ C ∥ f ∥ L p , (9)
for p > 1 p>1 p > 1 . In order to justify (3.9), we shall use the Littlewood-Paley operator P j P_j P j . Indeed, observe that
∥ A 2 j f ∥ L p ( ℓ j 2 ) = ∥ ∑ k ∈ Z A 2 j P − ( j + k ) f ∥ L p ( ℓ j 2 ) ≤ ∑ k ∈ Z ∥ A 2 j P − ( j + k ) f ∥ L p ( ℓ j 2 ) ≤ C ∑ k ∈ Z 2 − δ p ∣ k ∣ ∥ f ∥ L p ≤ C ∥ f ∥ L p . (10) \begin{aligned}
\|A_{2^j}f\|_{L^p(\ell_j^2)}
&=\left\|\sum_{k\in\mathbb{Z}}A_{2^j}P_{-(j+k)}f\right\|_{L^p(\ell_j^2)}\\
&\le\sum_{k\in\mathbb{Z}}\|A_{2^j}P_{-(j+k)}f\|_{L^p(\ell_j^2)}\\
&\le C\sum_{k\in\mathbb{Z}}2^{-\delta_p|k|}\|f\|_{L^p}
\le C\|f\|_{L^p}. \tag*{(10)}
\end{aligned} ∥ A 2 j f ∥ L p ( ℓ j 2 ) = k ∈ Z ∑ A 2 j P − ( j + k ) f L p ( ℓ j 2 ) ≤ k ∈ Z ∑ ∥ A 2 j P − ( j + k ) f ∥ L p ( ℓ j 2 ) ≤ C k ∈ Z ∑ 2 − δ p ∣ k ∣ ∥ f ∥ L p ≤ C ∥ f ∥ L p . (10)
The proof of the last but one inequality in (3.10) follows from
∥ A 2 j P − ( j + k ) f ∥ L 2 ( ℓ j 2 ) ≤ C 2 − δ ∣ k ∣ ∥ f ∥ L 2 , (11) \|A_{2^j}P_{-(j+k)}f\|_{L^2(\ell_j^2)}\le C2^{-\delta|k|}\|f\|_{L^2}, \tag*{(11)} ∥ A 2 j P − ( j + k ) f ∥ L 2 ( ℓ j 2 ) ≤ C 2 − δ ∣ k ∣ ∥ f ∥ L 2 , (11)
and
∥ A 2 j P − ( j + k ) f ∥ L p ( l j 2 ) ≤ C ∥ f ∥ L p . (12) \left\|\mathcal{A}_{2j}P_{-(j+k)}f\right\|_{L^p(l_j^2)}\leq C\|f\|_{L^p}. \tag*{(12)} A 2 j P − ( j + k ) f L p ( l j 2 ) ≤ C ∥ f ∥ L p . (12)
To prove (11) , we firstly show that
∣ m ( ξ , ξ n + 1 ) ∣ ≤ C ∣ ( ξ , ξ n + 1 ) ∣ , ∣ m ( ξ , ξ n + 1 ) ∣ ≤ C ( 1 + ∣ ( ξ , ξ n + 1 ) ∣ ) − n 2 . (13) |m(\xi,\xi_{n+1})|\leq C|(\xi,\xi_{n+1})|,\qquad|m(\xi,\xi_{n+1})|\leq C\left(1+|(\xi,\xi_{n+1})|\right)^{-\frac{n}{2}}. \tag*{(13)} ∣ m ( ξ , ξ n + 1 ) ∣ ≤ C ∣ ( ξ , ξ n + 1 ) ∣ , ∣ m ( ξ , ξ n + 1 ) ∣ ≤ C ( 1 + ∣ ( ξ , ξ n + 1 ) ∣ ) − 2 n . (13)
The first estimate in (13) follows from
∣ μ ( ξ , ξ n + 1 ) − e − i D ξ n + 1 ∣ = ∣ ∫ ( e − i ( x ⋅ ξ + ξ n + 1 Φ ( x ) ) − 1 ) e − i D ξ n + 1 ρ ( x ) d x ∣ ≤ C ∣ ( ξ , ξ n + 1 ) ∣ \left|\mu(\xi,\xi_{n+1})-e^{-iD\xi_{n+1}}\right|
=\left|\int\left(e^{-i(x\cdot\xi+\xi_{n+1}\Phi(x))}-1\right)e^{-iD\xi_{n+1}}\rho(x)\,dx\right|
\leq C|(\xi,\xi_{n+1})| μ ( ξ , ξ n + 1 ) − e − i D ξ n + 1 = ∫ ( e − i ( x ⋅ ξ + ξ n + 1 Φ ( x )) − 1 ) e − i D ξ n + 1 ρ ( x ) d x ≤ C ∣ ( ξ , ξ n + 1 ) ∣
and
∣ e − ∣ ( ξ , ξ n + 1 ) ∣ e − i D ξ n + 1 − e − i D ξ n + 1 ∣ ≤ C ∣ ( ξ , ξ n + 1 ) ∣ . \left|e^{-|(\xi,\xi_{n+1})|}e^{-iD\xi_{n+1}}-e^{-iD\xi_{n+1}}\right|\leq C|(\xi,\xi_{n+1})|. e − ∣ ( ξ , ξ n + 1 ) ∣ e − i D ξ n + 1 − e − i D ξ n + 1 ≤ C ∣ ( ξ , ξ n + 1 ) ∣.
The second estimate in (13) follows from the property of e − ∣ ( ξ , ξ n + 1 ) ∣ e^{-|(\xi,\xi_{n+1})|} e − ∣ ( ξ , ξ n + 1 ) ∣ and
∣ μ ( ξ , ξ n + 1 ) ∣ ≤ C ( 1 + ∣ ( ξ , ξ n + 1 ) ∣ ) − n 2 , |\mu(\xi,\xi_{n+1})|\leq C\left(1+|(\xi,\xi_{n+1})|\right)^{-\frac{n}{2}}, ∣ μ ( ξ , ξ n + 1 ) ∣ ≤ C ( 1 + ∣ ( ξ , ξ n + 1 ) ∣ ) − 2 n ,
which can be obtained by the method of stationary phase. Therefore, using Plancherel’s theorem we have
∥ A 2 j P − ( j + k ) f ∥ L 2 ( l j 2 ) = ∥ m ( 2 j ⋅ ) ψ ( 2 j + k ⋅ ) f ^ ( ⋅ ) ∥ L 2 ( l j 2 ) ≤ sup ( ⋅ ) ∈ R n + 1 ∥ m ( 2 j ⋅ ) ψ ( 2 j + k ⋅ ) ∥ l j 2 ∥ f ∥ L 2 . (14) \left\|\mathcal{A}_{2j}P_{-(j+k)}f\right\|_{L^2(l_j^2)}
=\left\|m(2^j\cdot)\psi(2^{j+k}\cdot)\widehat{f}(\cdot)\right\|_{L^2(l_j^2)}
\leq\sup_{(\cdot)\in\mathbb{R}^{n+1}}\left\|m(2^j\cdot)\psi(2^{j+k}\cdot)\right\|_{l_j^2}\|f\|_{L^2}. \tag*{(14)} A 2 j P − ( j + k ) f L 2 ( l j 2 ) = m ( 2 j ⋅ ) ψ ( 2 j + k ⋅ ) f ( ⋅ ) L 2 ( l j 2 ) ≤ ( ⋅ ) ∈ R n + 1 sup m ( 2 j ⋅ ) ψ ( 2 j + k ⋅ ) l j 2 ∥ f ∥ L 2 . (14)
Fix ( ⋅ ) ∈ R n + 1 (\cdot)\in\mathbb{R}^{n+1} ( ⋅ ) ∈ R n + 1 and k ∈ Z k\in\mathbb{Z} k ∈ Z , (13) implies
∥ m ( 2 j ⋅ ) ψ ( 2 j + k ⋅ ) ∥ l j 2 ≤ C min { 2 − k , ( 1 + 2 − k ) − n / 2 } ≤ C min { 2 − k , ( 1 + 2 − k ) − 1 / 2 } ≤ C 2 − 1 2 ∣ k ∣ . (15) \left\|m(2^j\cdot)\psi(2^{j+k}\cdot)\right\|_{l_j^2}
\leq C\min\left\{2^{-k},(1+2^{-k})^{-n/2}\right\}
\leq C\min\left\{2^{-k},(1+2^{-k})^{-1/2}\right\}\leq C2^{-\frac{1}{2}|k|}. \tag*{(15)} m ( 2 j ⋅ ) ψ ( 2 j + k ⋅ ) l j 2 ≤ C min { 2 − k , ( 1 + 2 − k ) − n /2 } ≤ C min { 2 − k , ( 1 + 2 − k ) − 1/2 } ≤ C 2 − 2 1 ∣ k ∣ . (15)
(14) and (15) yield (11) .
To prove (12) , it is enough to show
∥ A 2 j g j ∥ L p ( l j 2 ) ≤ C ∥ g j ∥ L p ( l j 2 ) . (16) \left\|\mathcal{A}_{2j}g_j\right\|_{L^p(l_j^2)}\leq C\|g_j\|_{L^p(l_j^2)}. \tag*{(16)} ∥ A 2 j g j ∥ L p ( l j 2 ) ≤ C ∥ g j ∥ L p ( l j 2 ) . (16)
Thus in view of (16) and the Littlewood-Paley theorem, we conclude
∥ A 2 j P − ( j + k ) f ∥ L p ( l j 2 ) ≤ C ∥ P − ( j + k ) f ∥ L p ( l j 2 ) ≤ C ∥ f ∥ L p \left\|\mathcal{A}_{2j}P_{-(j+k)}f\right\|_{L^p(l_j^2)}
\leq C\left\|P_{-(j+k)}f\right\|_{L^p(l_j^2)}
\leq C\|f\|_{L^p} A 2 j P − ( j + k ) f L p ( l j 2 ) ≤ C P − ( j + k ) f L p ( l j 2 ) ≤ C ∥ f ∥ L p
which proves (12) .
Now, we estimate (16) . This method comes from [7 ] . By duality, it suffices to consider the case p ≥ 2 p\geq2 p ≥ 2 . Choose q = ( p / 2 ) ′ q=(p/2)' q = ( p /2 ) ′ and let H \mathcal{H} H denote the tempered distribution whose Fourier transform is m m m and H ( k ) ( ⋅ ) = 2 − k ( n + 1 ) H ( 2 − k ⋅ ) \mathcal{H}^{(k)}(\cdot)=2^{-k(n+1)}\mathcal{H}(2^{-k}\cdot) H ( k ) ( ⋅ ) = 2 − k ( n + 1 ) H ( 2 − k ⋅ ) . Then there exists g ∈ L q g\in L^q g ∈ L q with unit norm such that
∥ A 2 k g k ∥ L p ( l k 2 ) 2 = ∫ ∑ k ∈ Z ∣ A 2 k g k ∣ 2 g = ∫ ∑ k ∈ Z ∣ H ( k ) ∗ g k ∣ 2 g ≤ ∑ k ∈ Z ∫ ∣ H ( k ) ∣ ∗ ∣ g k ∣ 2 g ≤ ∫ ∑ k ∈ Z ∣ g k ∣ 2 ∣ H ( k ) ∣ ∗ g ≤ ∥ g k ∥ L p ( l k 2 ) 2 ∥ sup k ∈ Z ∣ H ( k ) ∗ g ∣ ∥ L q . (17) \left\|\mathcal{A}_{2k}g_k\right\|_{L^p(l_k^2)}^2
=\int\sum_{k\in\mathbb{Z}}|\mathcal{A}_{2k}g_k|^2g
=\int\sum_{k\in\mathbb{Z}}|\mathcal{H}^{(k)}*g_k|^2g
\leq\sum_{k\in\mathbb{Z}}\int|\mathcal{H}^{(k)}|*|g_k|^2g
\leq\int\sum_{k\in\mathbb{Z}}|g_k|^2|\mathcal{H}^{(k)}|*g
\leq\|g_k\|_{L^p(l_k^2)}^2\left\|\sup_{k\in\mathbb{Z}}|\mathcal{H}^{(k)}*g|\right\|_{L^q}. \tag*{(17)} ∥ A 2 k g k ∥ L p ( l k 2 ) 2 = ∫ k ∈ Z ∑ ∣ A 2 k g k ∣ 2 g = ∫ k ∈ Z ∑ ∣ H ( k ) ∗ g k ∣ 2 g ≤ k ∈ Z ∑ ∫ ∣ H ( k ) ∣ ∗ ∣ g k ∣ 2 g ≤ ∫ k ∈ Z ∑ ∣ g k ∣ 2 ∣ H ( k ) ∣ ∗ g ≤ ∥ g k ∥ L p ( l k 2 ) 2 k ∈ Z sup ∣ H ( k ) ∗ g ∣ L q . (17)
Recall that (3.11) implies, for q = 2 q=2 q = 2 ,
∥ sup k H ( k ) ∗ g ∥ L q ≤ ∥ A 2 k g ∥ L q ( ℓ 2 ) ≤ C ∥ g ∥ L q . (18) \left\|\sup_k \mathcal{H}^{(k)} * g\right\|_{L^q} \le\left\|A_{2k}g\right\|_{L^q(\ell^2)} \le C\|g\|_{L^q}. \tag*{(18)} k sup H ( k ) ∗ g L q ≤ ∥ A 2 k g ∥ L q ( ℓ 2 ) ≤ C ∥ g ∥ L q . (18)
(3.17) and (3.18) yield (3.16) holds for 4 / 3 < p < 4 4/3<p<4 4/3 < p < 4 . This result implies (3.18) holds for 4 / 3 < p < 4 4/3<p<4 4/3 < p < 4 in turn. Thus we can obtain a series of intervals
( 4 3 , 4 ) ⊂ ( 8 7 , 8 ) ⊂ ⋯ ⊂ ( 2 j 2 j − 1 , 2 j ) ⊂ ⋯ \left(\frac{4}{3},4\right)\subset\left(\frac{8}{7},8\right)\subset\cdots\subset\left(\frac{2j}{2j-1},2j\right)\subset\cdots ( 3 4 , 4 ) ⊂ ( 7 8 , 8 ) ⊂ ⋯ ⊂ ( 2 j − 1 2 j , 2 j ) ⊂ ⋯
which ensure (3.16) holds for p ∈ ( 2 j 2 j − 1 , 2 j ) p\in\left(\frac{2j}{2j-1},2j\right) p ∈ ( 2 j − 1 2 j , 2 j ) . It is easy to show lim j → ∞ 2 j = ∞ \lim_{j\to\infty}2j=\infty lim j → ∞ 2 j = ∞ . Thus (3.16) holds for p > 1 p>1 p > 1 . The proof of (3.7) is completed.
Estimates for short variations: proof of (3.8)
This subsection is devoted to proving (3.8). We firstly present an elementary decomposition of A t f A_t f A t f . Using the partition of unity ∑ k ∈ Z ψ ( 2 − k ⋅ ) = 1 \sum_{k\in\mathbb{Z}}\psi(2^{-k}\cdot)=1 ∑ k ∈ Z ψ ( 2 − k ⋅ ) = 1 , we get
A t f ( x ‾ ) : = ∑ k ∈ Z A k , t f ( x ‾ ) (19) A_t f(\overline{x}):=\sum_{k\in\mathbb{Z}} A_{k,t}f(\overline{x}) \tag*{(19)} A t f ( x ) := k ∈ Z ∑ A k , t f ( x ) (19)
where
F ( A k , t f ) ( ξ , ξ n + 1 ) = m ( t ξ , t ξ n + 1 ) ψ ( 2 − k t ξ , 2 − k t ξ n + 1 ) f ^ ( ξ , ξ n + 1 ) . \mathcal{F}(A_{k,t}f)(\xi,\xi_{n+1})=m(t\xi,t\xi_{n+1})\psi(2^{-k}t\xi,2^{-k}t\xi_{n+1})\widehat{f}(\xi,\xi_{n+1}). F ( A k , t f ) ( ξ , ξ n + 1 ) = m ( t ξ , t ξ n + 1 ) ψ ( 2 − k t ξ , 2 − k t ξ n + 1 ) f ( ξ , ξ n + 1 ) .
Using interpolation, it suffice to prove Lemmas 3.1, 3.2, 3.3 and 3.4 below.
Lemma 3.1 For k ∈ Z k\in\mathbb{Z} k ∈ Z , we have
∥ V 2 j ( A k f ) ∥ L 2 ( ℓ j 2 ) ≤ C ( 1 + D 1 / 2 ) min { 2 − k n − 1 2 , 2 k } ∥ f ∥ L 2 . (20) \left\|V_2^j(A_kf)\right\|_{L^2(\ell_j^2)}\le C(1+D^{1/2})\min\left\{2^{-k\frac{n-1}{2}},2^k\right\}\|f\|_{L^2}. \tag*{(20)} V 2 j ( A k f ) L 2 ( ℓ j 2 ) ≤ C ( 1 + D 1/2 ) min { 2 − k 2 n − 1 , 2 k } ∥ f ∥ L 2 . (20)
Lemma 3.2 For k ∈ Z k\in\mathbb{Z} k ∈ Z and an arbitrary positive number ϵ \epsilon ϵ , there exists a constant C ϵ C_\epsilon C ϵ such that
∥ V 2 j ( A k f ) ∥ L 1 ( ℓ j 2 ) ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k ϵ , 2 − k ϵ } ∥ f ∥ H 1 . (21) \left\|V_2^j(A_kf)\right\|_{L^1(\ell_j^2)}\le C_\epsilon(1+D^{1+\epsilon})\max\left\{2^{k\epsilon},2^{-k\epsilon}\right\}\|f\|_{H^1}. \tag*{(21)} V 2 j ( A k f ) L 1 ( ℓ j 2 ) ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k ϵ , 2 − k ϵ } ∥ f ∥ H 1 . (21)
Lemma 3.3 Suppose that n ≥ 2 n\ge2 n ≥ 2 and p > 2 + 4 n p>2+\frac{4}{n} p > 2 + n 4 , or n = 1 n=1 n = 1 , p ≥ 4 p\ge4 p ≥ 4 . Then for k ∈ Z k\in\mathbb{Z} k ∈ Z , we obtain
∥ V 2 j ( A k f ) ∥ L p ( ℓ j 2 ) ≤ C ϵ ( 1 + D 1 / 2 ) min { 2 − k ( n + 1 p − 1 2 − ϵ ) , 2 k ϵ } ∥ f ∥ L p . (22) \left\|V_2^j(A_kf)\right\|_{L^p(\ell_j^2)}\le C_\epsilon(1+D^{1/2})\min\left\{2^{-k\left(\frac{n+1}{p}-\frac{1}{2}-\epsilon\right)},2^{k\epsilon}\right\}\|f\|_{L^p}. \tag*{(22)} V 2 j ( A k f ) L p ( ℓ j 2 ) ≤ C ϵ ( 1 + D 1/2 ) min { 2 − k ( p n + 1 − 2 1 − ϵ ) , 2 k ϵ } ∥ f ∥ L p . (22)
Lemma 3.4 For n = 1 n=1 n = 1 and k ∈ Z k\in\mathbb{Z} k ∈ Z , we have
∥ V 4 j ( A k f ) ∥ L 4 ( ℓ j 4 ) ≤ C ϵ ( 1 + D 1 / 4 ) min { 2 − k ( 1 8 − ϵ ) , 1 } ∥ f ∥ L 4 . (23) \left\|V_4^j(A_kf)\right\|_{L^4(\ell_j^4)}\le C_\epsilon(1+D^{1/4})\min\left\{2^{-k\left(\frac{1}{8}-\epsilon\right)},1\right\}\|f\|_{L^4}. \tag*{(23)} V 4 j ( A k f ) L 4 ( ℓ j 4 ) ≤ C ϵ ( 1 + D 1/4 ) min { 2 − k ( 8 1 − ϵ ) , 1 } ∥ f ∥ L 4 . (23)
Now we can prove (3.8), assuming the validity of lemmas above.
Proof of (3.8): We first consider the case n ≥ 2 n\ge2 n ≥ 2 . Interpolating (3.20) and (3.21), we obtain that there exists a positive constant δ p \delta_p δ p such that
∥ V 2 j ( A k f ) ∥ L p ( ℓ j 2 ) ≤ C ϵ ( 1 + D 1 / p + ϵ ) 2 − δ p ∣ k ∣ ∥ f ∥ L p , (24) \left\|V_2^j(A_kf)\right\|_{L^p(\ell_j^2)}\le C_\epsilon(1+D^{1/p+\epsilon})2^{-\delta_p|k|}\|f\|_{L^p}, \tag*{(24)} V 2 j ( A k f ) L p ( ℓ j 2 ) ≤ C ϵ ( 1 + D 1/ p + ϵ ) 2 − δ p ∣ k ∣ ∥ f ∥ L p , (24)
for ( n + 1 ) / n < p ≤ 2 (n+1)/n < p \le2 ( n + 1 ) / n < p ≤ 2 . Interpolating (20) and (22) , we get there exists a constant δ p \delta_p δ p such that
∥ V 2 j ( A k f ) ∥ L p ( l j 2 ) ≤ C ( 1 + D 1 / 2 ) 2 − δ p ∣ k ∣ ∥ f ∥ L p , (25) \left\|V_2^j(\mathcal{A}_k f)\right\|_{L^p(l_j^2)} \le C(1+D^{1/2})2^{-\delta_p|k|}\|f\|_{L^p}, \tag*{(25)} V 2 j ( A k f ) L p ( l j 2 ) ≤ C ( 1 + D 1/2 ) 2 − δ p ∣ k ∣ ∥ f ∥ L p , (25)
for 2 < p < 2 ( n + 1 ) 2 < p < 2(n+1) 2 < p < 2 ( n + 1 ) . Thus (24) and (25) yield, for ( n + 1 ) / n < p < 2 ( n + 1 ) (n+1)/n < p < 2(n+1) ( n + 1 ) / n < p < 2 ( n + 1 ) ,
∥ V 2 j ( A f ) ∥ L p ( l j 2 ) ≤ C ( p , 2 , D ) ∥ f ∥ L p . (26) \left\|V_2^j(\mathcal{A}f)\right\|_{L^p(l_j^2)} \le C(p,2,D)\|f\|_{L^p}. \tag*{(26)} V 2 j ( A f ) L p ( l j 2 ) ≤ C ( p , 2 , D ) ∥ f ∥ L p . (26)
(8) can be deduced since ∥ ⋅ ∥ l p 1 ≤ ∥ ⋅ ∥ l p 2 \|\cdot\|_{l^{p_1}} \le\|\cdot\|_{l^{p_2}} ∥ ⋅ ∥ l p 1 ≤ ∥ ⋅ ∥ l p 2 for p 2 < p 1 p_2 < p_1 p 2 < p 1 and
∥ V ∞ j ( A f ) ∥ L ∞ ( l j ∞ ) ≤ C ∥ f ∥ L ∞ . (27) \left\|V_\infty^j(\mathcal{A}f)\right\|_{L^\infty(l_j^\infty)} \le C\|f\|_{L^\infty}. \tag*{(27)} V ∞ j ( A f ) L ∞ ( l j ∞ ) ≤ C ∥ f ∥ L ∞ . (27)
Now we consider the case n = 1 n=1 n = 1 . By interpolation, (20) and (22) imply
∥ V 2 j ( A k f ) ∥ L p ( l j 2 ) ≤ C ϵ ( 1 + D 1 / 2 ) 2 ϵ k ∥ f ∥ L p (28) \left\|V_2^j(\mathcal{A}_k f)\right\|_{L^p(l_j^2)} \le C_\epsilon(1+D^{1/2})2^{\epsilon k}\|f\|_{L^p} \tag*{(28)} V 2 j ( A k f ) L p ( l j 2 ) ≤ C ϵ ( 1 + D 1/2 ) 2 ϵ k ∥ f ∥ L p (28)
for 2 < p ≤ 4 2 < p \le4 2 < p ≤ 4 . Using interpolation again, (23) and (28) yield
∥ V r j ( A k f ) ∥ L p ( l j 2 ) ≤ C ϵ ( 1 + D 1 / r ) 2 − δ r , p ∣ k ∣ ∥ f ∥ L p (29) \left\|V_r^j(\mathcal{A}_k f)\right\|_{L^p(l_j^2)} \le C_\epsilon(1+D^{1/r})2^{-\delta_{r,p}|k|}\|f\|_{L^p} \tag*{(29)} V r j ( A k f ) L p ( l j 2 ) ≤ C ϵ ( 1 + D 1/ r ) 2 − δ r , p ∣ k ∣ ∥ f ∥ L p (29)
for 2 < r ≤ 4 2 < r \le4 2 < r ≤ 4 and r < p ≤ 4 r < p \le4 r < p ≤ 4 . Thus (29) and (27) yield
∥ V r j ( A f ) ∥ L p ( l j r ) ≤ C ( 1 + D 1 / r ) ∥ f ∥ L p , (30) \left\|V_r^j(\mathcal{A}f)\right\|_{L^p(l_j^r)} \le C(1+D^{1/r})\|f\|_{L^p}, \tag*{(30)} V r j ( A f ) L p ( l j r ) ≤ C ( 1 + D 1/ r ) ∥ f ∥ L p , (30)
for 2 < p < 2 r 2 < p < 2r 2 < p < 2 r . Then (8) follows from (30) and the proof is complete.
Proof of Lemma 3.1
Using (5) and Hölder inequality, we get
∥ V 2 j ( A k f ) ∥ L 2 ( l j 2 ) ≤ C ∥ A k , t f ∥ L 2 ( L t 2 ) 1 / 2 ∥ t d d t A k , t f ∥ L 2 ( L t 2 ) 1 / 2 . (31) \left\|V_2^j(\mathcal{A}_k f)\right\|_{L^2(l_j^2)} \le C\left\|\mathcal{A}_{k,t}f\right\|_{L^2(L_t^2)}^{1/2}\left\|t\frac{d}{dt}\mathcal{A}_{k,t}f\right\|_{L^2(L_t^2)}^{1/2}. \tag*{(31)} V 2 j ( A k f ) L 2 ( l j 2 ) ≤ C ∥ A k , t f ∥ L 2 ( L t 2 ) 1/2 t d t d A k , t f L 2 ( L t 2 ) 1/2 . (31)
Thus, we obtain from (31) via Plancherel’s theorem that
∥ V 2 j ( A k f ) ∥ L 2 ( l j 2 ) ≤ C ∥ m k ( t ⋅ ) f ^ ( ⋅ ) ∥ L 2 ( L t 2 ) 1 / 2 ∥ ( t ⋅ ) ∇ m k ( t ⋅ ) f ^ ( ⋅ ) ∥ L 2 ( L t 2 ) 1 / 2 ≤ C sup ∣ ⋅ ∣ = 1 ∥ m k ( t ⋅ ) ∥ L t 2 1 / 2 sup ∣ ⋅ ∣ = 1 ∥ ( t ⋅ ) ∇ m k ( t ⋅ ) ∥ L t 2 1 / 2 ∥ f ∥ L 2 . (32) \begin{aligned}
\left\|V_2^j(\mathcal{A}_k f)\right\|_{L^2(l_j^2)}
&\le C\left\|m_k(t\cdot)\widehat{f}(\cdot)\right\|_{L^2(L_t^2)}^{1/2}
\left\|(t\cdot)\nabla m_k(t\cdot)\widehat{f}(\cdot)\right\|_{L^2(L_t^2)}^{1/2}\\
&\le C\sup_{|\cdot|=1}\left\|m_k(t\cdot)\right\|_{L_t^2}^{1/2}
\sup_{|\cdot|=1}\left\|(t\cdot)\nabla m_k(t\cdot)\right\|_{L_t^2}^{1/2}\|f\|_{L^2}. \tag*{(32)}
\end{aligned} V 2 j ( A k f ) L 2 ( l j 2 ) ≤ C m k ( t ⋅ ) f ( ⋅ ) L 2 ( L t 2 ) 1/2 ( t ⋅ ) ∇ m k ( t ⋅ ) f ( ⋅ ) L 2 ( L t 2 ) 1/2 ≤ C ∣ ⋅ ∣ = 1 sup ∥ m k ( t ⋅ ) ∥ L t 2 1/2 ∣ ⋅ ∣ = 1 sup ∥ ( t ⋅ ) ∇ m k ( t ⋅ ) ∥ L t 2 1/2 ∥ f ∥ L 2 . (32)
where m k ( ⋅ ) = m ( ⋅ ) ψ ( 2 − k ⋅ ) m_k(\cdot)=m(\cdot)\psi(2^{-k}\cdot) m k ( ⋅ ) = m ( ⋅ ) ψ ( 2 − k ⋅ ) . Thus it is important to estimate ∥ m k ∥ L ∞ \|m_k\|_{L^\infty} ∥ m k ∥ L ∞ and ∥ ( ⋅ ) ∇ m k ( ⋅ ) ∥ L ∞ \|(\cdot)\nabla m_k(\cdot)\|_{L^\infty} ∥ ( ⋅ ) ∇ m k ( ⋅ ) ∥ L ∞ , since both m k m_k m k and ( ⋅ ) ∇ m k ( ⋅ ) (\cdot)\nabla m_k(\cdot) ( ⋅ ) ∇ m k ( ⋅ ) are supported where ∣ ⋅ ∣ ∼ 2 k |\cdot|\sim2^k ∣ ⋅ ∣ ∼ 2 k .
By (13) , we can deduce that
∥ m k ∥ L ∞ ≤ C min { 2 − n k / 2 , 2 k } . (33) \|m_k\|_{L^\infty}\le C\min\{2^{-nk/2},2^k\}. \tag*{(33)} ∥ m k ∥ L ∞ ≤ C min { 2 − nk /2 , 2 k } . (33)
Rewrite m ( ξ , ξ n + 1 ) = e − i D ξ n + 1 ( ∫ e − i ( x ⋅ ξ + ξ n + 1 Φ ( x ) ) ρ ( x ) d x − e − ∣ ( ξ , ξ n + 1 ) ∣ ) m(\xi,\xi_{n+1})=e^{-iD\xi_{n+1}}\left(\int e^{-i(x\cdot\xi+\xi_{n+1}\Phi(x))}\rho(x)\,dx-e^{-|(\xi,\xi_{n+1})|}\right) m ( ξ , ξ n + 1 ) = e − i D ξ n + 1 ( ∫ e − i ( x ⋅ ξ + ξ n + 1 Φ ( x )) ρ ( x ) d x − e − ∣ ( ξ , ξ n + 1 ) ∣ ) . Using the method of stationary phase for ∣ ( ξ , ξ n + 1 ) ∣ |(\xi,\xi_{n+1})| ∣ ( ξ , ξ n + 1 ) ∣ near ∞ \infty ∞ and computing directly for ∣ ( ξ , ξ n + 1 ) ∣ |(\xi,\xi_{n+1})| ∣ ( ξ , ξ n + 1 ) ∣ near zero, we obtain
∣ ( ξ , ξ n + 1 ) ⋅ ∇ m ( ξ , ξ n + 1 ) ∣ ≤ C ( 1 + D ) min { ∣ ( ξ , ξ n + 1 ) ∣ , ∣ ( ξ , ξ n + 1 ) ∣ − n 2 + 1 } . (34) |(\xi,\xi_{n+1})\cdot\nabla m(\xi,\xi_{n+1})|\le C(1+D)\min\left\{|(\xi,\xi_{n+1})|,|(\xi,\xi_{n+1})|^{-\frac n2+1}\right\}. \tag*{(34)} ∣ ( ξ , ξ n + 1 ) ⋅ ∇ m ( ξ , ξ n + 1 ) ∣ ≤ C ( 1 + D ) min { ∣ ( ξ , ξ n + 1 ) ∣ , ∣ ( ξ , ξ n + 1 ) ∣ − 2 n + 1 } . (34)
It follows from (33) and (34) that
∥ m k ( t ⋅ ) ∥ L t 2 ≤ C min { 2 − n k / 2 , 2 k } \|m_k(t\cdot)\|_{L_t^2}\le C\min\{2^{-nk/2},2^k\} ∥ m k ( t ⋅ ) ∥ L t 2 ≤ C min { 2 − nk /2 , 2 k }
and
∥ ( t ⋅ ) ( ∇ m k ) ( t ⋅ ) ∥ L t 2 ≤ C ( 1 + D ) min { 2 − n k / 2 + k , 2 k } . (35) \|(t\cdot)(\nabla m_k)(t\cdot)\|_{L_t^2}\le C(1+D)\min\{2^{-nk/2+k},2^k\}. \tag*{(35)} ∥ ( t ⋅ ) ( ∇ m k ) ( t ⋅ ) ∥ L t 2 ≤ C ( 1 + D ) min { 2 − nk /2 + k , 2 k } . (35)
(32) and (35) yield (20) . This completes the proof.
Proof of Lemma 3.2
Denote K ^ k ( ξ , ξ n + 1 ) = μ ( ξ , ξ n + 1 ) ψ ( 2 − k ξ , 2 − k ξ n + 1 ) \widehat{K}_{k}(\xi,\xi_{n+1})=\mu(\xi,\xi_{n+1})\psi(2^{-k}\xi,2^{-k}\xi_{n+1}) K k ( ξ , ξ n + 1 ) = μ ( ξ , ξ n + 1 ) ψ ( 2 − k ξ , 2 − k ξ n + 1 ) and K k , t ( x ˉ ) = t − ( n + 1 ) K k ( x ˉ t ) K_{k,t}(\bar{x})=t^{-(n+1)}K_k\left(\frac{\bar{x}}{t}\right) K k , t ( x ˉ ) = t − ( n + 1 ) K k ( t x ˉ ) . Moreover, let G ~ k ( ξ , ξ n + 1 ) = e − ∣ ξ ∣ − i D ξ n + 1 ψ ( 2 − k ξ , 2 − k ξ n + 1 ) \widetilde{G}_k(\xi,\xi_{n+1})=e^{-|\xi|-iD\xi_{n+1}}\psi(2^{-k}\xi,2^{-k}\xi_{n+1}) G k ( ξ , ξ n + 1 ) = e − ∣ ξ ∣ − i D ξ n + 1 ψ ( 2 − k ξ , 2 − k ξ n + 1 ) and H k = K k − G k H_k=K_k-G_k H k = K k − G k . To achieve (21) , we shall use vector-valued singular integral theory for the convolution operators { A k , t f = H k , t ∗ f } t > 0 \{A_{k,t}f=H_{k,t}*f\}_{t>0} { A k , t f = H k , t ∗ f } t > 0 . Here we take the vector valued kernel H k = { H k , t } t > 0 H_k=\{H_{k,t}\}_{t>0} H k = { H k , t } t > 0 and denote the norm by S 2 ( H k , t : t > 0 ) S_2(H_{k,t}:t>0) S 2 ( H k , t : t > 0 ) . Recalling that (20) gives
∥ S 2 ( H k , t ∗ f : t > 0 ) ∥ L 2 ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k + ϵ , 2 − k ϵ } ∥ f ∥ L 2 . \|S_2(H_{k,t}*f:t>0)\|_{L^2}\leq C_\epsilon(1+D^{1+\epsilon})\max\left\{2^{k+\epsilon},2^{-k\epsilon}\right\}\|f\|_{L^2}. ∥ S 2 ( H k , t ∗ f : t > 0 ) ∥ L 2 ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k + ϵ , 2 − k ϵ } ∥ f ∥ L 2 .
By Calderón-Zygmund theory, it remains to prove
I H ( y ˉ ) : = ∫ ∣ x ˉ ∣ > 2 ∣ y ˉ ∣ S 2 ( H k , t ( x ˉ − y ˉ ) − H k , t ( x ˉ ) : t > 0 ) d x ˉ ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k + ϵ , 2 − k ϵ } . (36) I_H(\bar{y}):=\int_{|\bar{x}|>2|\bar{y}|}S_2(H_{k,t}(\bar{x}-\bar{y})-H_{k,t}(\bar{x}):t>0)\,d\bar{x}\leq C_\epsilon(1+D^{1+\epsilon})\max\left\{2^{k+\epsilon},2^{-k\epsilon}\right\}. \tag*{(36)} I H ( y ˉ ) := ∫ ∣ x ˉ ∣ > 2∣ y ˉ ∣ S 2 ( H k , t ( x ˉ − y ˉ ) − H k , t ( x ˉ ) : t > 0 ) d x ˉ ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k + ϵ , 2 − k ϵ } . (36)
Without loss of generality, we only prove
I ( y ˉ ) : = ∫ ∣ x ˉ ∣ > 2 ∣ y ˉ ∣ S 2 ( K k , t ( x ˉ − y ˉ ) − K k , t ( x ˉ ) : t > 0 ) d x ˉ ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k + ϵ , 2 − k ϵ } . (37) I(\bar{y}):=\int_{|\bar{x}|>2|\bar{y}|}S_2(K_{k,t}(\bar{x}-\bar{y})-K_{k,t}(\bar{x}):t>0)\,d\bar{x}\leq C_\epsilon(1+D^{1+\epsilon})\max\left\{2^{k+\epsilon},2^{-k\epsilon}\right\}. \tag*{(37)} I ( y ˉ ) := ∫ ∣ x ˉ ∣ > 2∣ y ˉ ∣ S 2 ( K k , t ( x ˉ − y ˉ ) − K k , t ( x ˉ ) : t > 0 ) d x ˉ ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k + ϵ , 2 − k ϵ } . (37)
Using S 2 ( a t : t > 0 ) ≤ C ∥ t a t ′ ∥ L t 1 S_2(a_t:t>0)\leq C\|ta'_t\|_{L^1_t} S 2 ( a t : t > 0 ) ≤ C ∥ t a t ′ ∥ L t 1 , it can be deduced that
I ( y ˉ ) ≤ C ∫ ∣ x ˉ ∣ > 2 ∣ y ˉ ∣ ∫ 0 ∞ t ∣ d d t ( K k , t ( x ˉ − y ˉ ) − K k , t ( x ˉ ) ) ∣ d t t d x ˉ ≤ C ( n + 1 ) ∫ 0 ∞ ∫ ∣ x ˉ ∣ > 2 ∣ y ˉ ∣ ∣ K k , t ( x ˉ − y ˉ ) − K k , t ( x ˉ ) ∣ d x ˉ d t t + ∫ 0 ∞ ∫ ∣ x ˉ ∣ > 2 ∣ y ˉ ∣ ∣ K ~ k , t ( x ˉ − y ˉ ) − K ~ k , t ( x ˉ ) ∣ d x ˉ d t t \begin{aligned}
I(\bar{y})&\leq C\int_{|\bar{x}|>2|\bar{y}|}\int_0^\infty t\left|\frac{d}{dt}\left(K_{k,t}(\bar{x}-\bar{y})-K_{k,t}(\bar{x})\right)\right|\frac{dt}{t}\,d\bar{x}\\
&\leq C(n+1)\int_0^\infty\int_{|\bar{x}|>2|\bar{y}|}|K_{k,t}(\bar{x}-\bar{y})-K_{k,t}(\bar{x})|\,d\bar{x}\frac{dt}{t}\\
&\quad+\int_0^\infty\int_{|\bar{x}|>2|\bar{y}|}|\widetilde{K}_{k,t}(\bar{x}-\bar{y})-\widetilde{K}_{k,t}(\bar{x})|\,d\bar{x}\frac{dt}{t}
\end{aligned} I ( y ˉ ) ≤ C ∫ ∣ x ˉ ∣ > 2∣ y ˉ ∣ ∫ 0 ∞ t d t d ( K k , t ( x ˉ − y ˉ ) − K k , t ( x ˉ ) ) t d t d x ˉ ≤ C ( n + 1 ) ∫ 0 ∞ ∫ ∣ x ˉ ∣ > 2∣ y ˉ ∣ ∣ K k , t ( x ˉ − y ˉ ) − K k , t ( x ˉ ) ∣ d x ˉ t d t + ∫ 0 ∞ ∫ ∣ x ˉ ∣ > 2∣ y ˉ ∣ ∣ K k , t ( x ˉ − y ˉ ) − K k , t ( x ˉ ) ∣ d x ˉ t d t
where
K ~ k , t ( x ˉ ) = x ˉ ⋅ ∇ x ˉ K k , t ( x ˉ ) . \widetilde{K}_{k,t}(\bar{x})=\bar{x}\cdot\nabla_{\bar{x}}K_{k,t}(\bar{x}). K k , t ( x ˉ ) = x ˉ ⋅ ∇ x ˉ K k , t ( x ˉ ) .
We only need to estimate the last double integrals. In doing so, without loss of generality, we shall show
∫ 0 1 ∫ ∣ x ˉ ∣ > 2 ∣ y ˉ ∣ ∣ K ~ k , t ( x ˉ − y ˉ ) − K ~ k , t ( x ˉ ) ∣ d x ˉ d t t ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k + ϵ , 2 − k ϵ } (38) \int_0^1\int_{|\bar{x}|>2|\bar{y}|}|\widetilde{K}_{k,t}(\bar{x}-\bar{y})-\widetilde{K}_{k,t}(\bar{x})|\,d\bar{x}\frac{dt}{t}\leq C_\epsilon(1+D^{1+\epsilon})\max\left\{2^{k+\epsilon},2^{-k\epsilon}\right\} \tag*{(38)} ∫ 0 1 ∫ ∣ x ˉ ∣ > 2∣ y ˉ ∣ ∣ K k , t ( x ˉ − y ˉ ) − K k , t ( x ˉ ) ∣ d x ˉ t d t ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k + ϵ , 2 − k ϵ } (38)
and
∫ 1 ∞ ∫ ∣ x ˉ ∣ > 2 ∣ y ˉ ∣ ∣ K ~ k , t ( x ˉ − y ˉ ) − K ~ k , t ( x ˉ ) ∣ d x ˉ d t t ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k + ϵ , 2 − k ϵ } . (39) \int_1^\infty\int_{|\bar{x}|>2|\bar{y}|}|\widetilde{K}_{k,t}(\bar{x}-\bar{y})-\widetilde{K}_{k,t}(\bar{x})|\,d\bar{x}\frac{dt}{t}\leq C_\epsilon(1+D^{1+\epsilon})\max\left\{2^{k+\epsilon},2^{-k\epsilon}\right\}. \tag*{(39)} ∫ 1 ∞ ∫ ∣ x ˉ ∣ > 2∣ y ˉ ∣ ∣ K k , t ( x ˉ − y ˉ ) − K k , t ( x ˉ ) ∣ d x ˉ t d t ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k + ϵ , 2 − k ϵ } . (39)
To achieve (38) , it is no restriction to assume that ∣ y ˉ ∣ = 1 |\bar{y}|=1 ∣ y ˉ ∣ = 1 . Thus we obtain
∫ 0 1 ∫ ∣ x ˉ ∣ > 2 ∣ K ~ k , t ( x ˉ − y ˉ ) − K ~ k , t ( x ˉ ) ∣ d x ˉ d t t = ∫ 0 1 ∫ ∣ x ˉ ∣ > 2 / t ∣ K ~ k ( x ˉ − y ˉ t ) − K ~ k ( x ˉ ) ∣ d x ˉ d t t ≤ 2 ∫ 0 1 ∫ ∣ x ˉ ∣ > 1 / t ∣ K ~ k ( x ˉ ) ∣ d x ˉ d t t = 2 ∫ ∣ x ˉ ∣ > 1 ∣ K ~ k ( x ˉ ) ∣ log ∣ x ˉ ∣ d x ˉ . (40) \begin{aligned}
\int_0^1\int_{|\bar{x}|>2}|\widetilde{K}_{k,t}(\bar{x}-\bar{y})-\widetilde{K}_{k,t}(\bar{x})|\,d\bar{x}\frac{dt}{t}
&=\int_0^1\int_{|\bar{x}|>2/t}\left|\widetilde{K}_k\left(\bar{x}-\frac{\bar{y}}{t}\right)-\widetilde{K}_k(\bar{x})\right|\,d\bar{x}\frac{dt}{t}\\
&\leq2\int_0^1\int_{|\bar{x}|>1/t}|\widetilde{K}_k(\bar{x})|\,d\bar{x}\frac{dt}{t}
=2\int_{|\bar{x}|>1}|\widetilde{K}_k(\bar{x})|\log|\bar{x}|\,d\bar{x}. \tag*{(40)}
\end{aligned} ∫ 0 1 ∫ ∣ x ˉ ∣ > 2 ∣ K k , t ( x ˉ − y ˉ ) − K k , t ( x ˉ ) ∣ d x ˉ t d t = ∫ 0 1 ∫ ∣ x ˉ ∣ > 2/ t K k ( x ˉ − t y ˉ ) − K k ( x ˉ ) d x ˉ t d t ≤ 2 ∫ 0 1 ∫ ∣ x ˉ ∣ > 1/ t ∣ K k ( x ˉ ) ∣ d x ˉ t d t = 2 ∫ ∣ x ˉ ∣ > 1 ∣ K k ( x ˉ ) ∣ log ∣ x ˉ ∣ d x ˉ . (40)
To continue our proof, let us analyze the expression of K ~ k \widetilde{K}_k K k . Denote ϕ ( x ˉ ) = ψ ~ ( x ˉ ) \phi(\bar{x})=\widetilde{\psi}(\bar{x}) ϕ ( x ˉ ) = ψ ( x ˉ ) and ϕ i ( x ˉ ) = ∂ ∂ x i ϕ ( x ˉ ) \phi_i(\bar{x})=\frac{\partial}{\partial x_i}\phi(\bar{x}) ϕ i ( x ˉ ) = ∂ x i ∂ ϕ ( x ˉ ) . Then we have
K k ( x ˉ ) = μ ~ ∗ ϕ ( 2 k ) ( x ˉ ) = 2 k ( n + 1 ) ∫ ϕ ( 2 k ( x − y , x n + 1 − ( D + Φ ( y ) ) ) ) ρ ( y ) d y . (41) K_k(\bar{x})=\widetilde{\mu}*\phi_{(2^k)}(\bar{x})=2^{k(n+1)}\int\phi(2^k(x-y,x_{n+1}-(D+\Phi(y))))\rho(y)\,dy. \tag*{(41)} K k ( x ˉ ) = μ ∗ ϕ ( 2 k ) ( x ˉ ) = 2 k ( n + 1 ) ∫ ϕ ( 2 k ( x − y , x n + 1 − ( D + Φ ( y )))) ρ ( y ) d y . (41)
where ϕ ( 2 k ) ( x ˉ ) = ϕ ( 2 k x ˉ ) \phi_{(2^k)}(\bar{x})=\phi(2^k\bar{x}) ϕ ( 2 k ) ( x ˉ ) = ϕ ( 2 k x ˉ ) . It follows that
K ~ k ( x ˉ ) = ∑ i = 1 n + 1 2 k ( n + 2 ) x i ∫ ϕ i ( 2 k ( x − y , x n + 1 − ( D + Φ ( y ) ) ) ) ρ ( y ) d y . (42) \widetilde{K}_k(\bar{x})=\sum_{i=1}^{n+1}2^{k(n+2)}x_i\int\phi_i(2^k(x-y,x_{n+1}-(D+\Phi(y))))\rho(y)\,dy. \tag*{(42)} K k ( x ˉ ) = i = 1 ∑ n + 1 2 k ( n + 2 ) x i ∫ ϕ i ( 2 k ( x − y , x n + 1 − ( D + Φ ( y )))) ρ ( y ) d y . (42)
Using (40) and (42) , we have
∫ ∣ x ˉ ∣ > 1 ∣ K ~ k ( x ˉ ) ∣ log ∣ x ˉ ∣ d x ˉ = ∬ ∣ x ˉ ∣ > 1 ∣ ∑ i = 1 n + 1 2 k ( n + 2 ) x i ϕ i ( 2 k ( x − y , x n + 1 − ( D + Φ ( y ) ) ) ) ∣ log ∣ x ˉ ∣ d x ˉ ρ ( y ) d y = ∬ ∣ x ˉ ∣ > 2 k ∣ ∑ i = 1 n + 1 x i ϕ i ( x − 2 k y , x n + 1 − 2 k ( D + Φ ( y ) ) ) ∣ log ∣ 2 − k x ˉ ∣ d x ˉ ρ ( y ) d y . \begin{aligned}
\int_{|\bar{x}|>1}|\widetilde{K}_k(\bar{x})|\log|\bar{x}|\,d\bar{x}
&=\iint_{|\bar{x}|>1}\left|\sum_{i=1}^{n+1}2^{k(n+2)}x_i\phi_i(2^k(x-y,x_{n+1}-(D+\Phi(y))))\right|\log|\bar{x}|\,d\bar{x}\rho(y)\,dy\\
&=\iint_{|\bar{x}|>2^k}\left|\sum_{i=1}^{n+1}x_i\phi_i(x-2^ky,x_{n+1}-2^k(D+\Phi(y)))\right|\log|2^{-k}\bar{x}|\,d\bar{x}\rho(y)\,dy.
\end{aligned} ∫ ∣ x ˉ ∣ > 1 ∣ K k ( x ˉ ) ∣ log ∣ x ˉ ∣ d x ˉ = ∬ ∣ x ˉ ∣ > 1 i = 1 ∑ n + 1 2 k ( n + 2 ) x i ϕ i ( 2 k ( x − y , x n + 1 − ( D + Φ ( y )))) log ∣ x ˉ ∣ d x ˉ ρ ( y ) d y = ∬ ∣ x ˉ ∣ > 2 k i = 1 ∑ n + 1 x i ϕ i ( x − 2 k y , x n + 1 − 2 k ( D + Φ ( y ))) log ∣ 2 − k x ˉ ∣ d x ˉ ρ ( y ) d y .
By inequality log ( ∣ x ∣ ) ≤ C ϵ ∣ x ∣ ϵ \log(|x|)\le C_\epsilon|x|^\epsilon log ( ∣ x ∣ ) ≤ C ϵ ∣ x ∣ ϵ for ∣ x ∣ ≥ 1 |x|\ge1 ∣ x ∣ ≥ 1 , it can be deduced that
∫ ∣ x ˉ ∣ > 1 ∣ K ~ k ( x ˉ ) ∣ log ∣ x ˉ ∣ d x ˉ ≤ C ϵ ∬ ∑ i = 1 n + 1 ∣ x i ( 2 − k ∣ x ˉ ∣ ) ϵ ϕ i ( x − 2 k y , x n + 1 − 2 k ( D + Φ ( y ) ) ) ∣ d x ˉ ρ ( y ) d y ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k , 2 − k ϵ } . \begin{aligned}
\int_{|\bar{x}|>1}|\widetilde{K}_k(\bar{x})|\log|\bar{x}|\,d\bar{x}
&\le C_\epsilon\iint\sum_{i=1}^{n+1}\left|x_i(2^{-k}|\bar{x}|)^\epsilon\phi_i(x-2^ky,x_{n+1}-2^k(D+\Phi(y)))\right|\,d\bar{x}\rho(y)\,dy\\
&\le C_\epsilon(1+D^{1+\epsilon})\max\{2^k,2^{-k\epsilon}\}.
\end{aligned} ∫ ∣ x ˉ ∣ > 1 ∣ K k ( x ˉ ) ∣ log ∣ x ˉ ∣ d x ˉ ≤ C ϵ ∬ i = 1 ∑ n + 1 x i ( 2 − k ∣ x ˉ ∣ ) ϵ ϕ i ( x − 2 k y , x n + 1 − 2 k ( D + Φ ( y ))) d x ˉ ρ ( y ) d y ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k , 2 − k ϵ } .
Now, we prove (39) . We first rewrite ,
∫ 1 ∞ ∫ ∣ x ˉ ∣ > 2 ∣ K ~ k , t ( x ˉ − y ˉ ) − K ~ k , t ( x ˉ ) ∣ d x ˉ d t t = ∫ 1 ∞ ∫ ∣ x ˉ ∣ > 2 / t ∣ K ~ k ( x ˉ − y ˉ t ) − K ~ k ( x ˉ ) ∣ d x ˉ d t t ≤ ∫ 1 ∞ ω 1 ( K ~ k , 1 / t ) d t t = ∫ 0 1 ω 1 ( K ~ k , t ) d t t , \begin{aligned}
\int_1^\infty\int_{|\bar{x}|>2}\left|\widetilde{K}_{k,t}(\bar{x}-\bar{y})-\widetilde{K}_{k,t}(\bar{x})\right|\,d\bar{x}\frac{dt}{t}
&=\int_1^\infty\int_{|\bar{x}|>2/t}\left|\widetilde{K}_k\left(\bar{x}-\frac{\bar{y}}{t}\right)-\widetilde{K}_k(\bar{x})\right|\,d\bar{x}\frac{dt}{t}\\
&\le\int_1^\infty\omega_1(\widetilde{K}_k,1/t)\frac{dt}{t}
=\int_0^1\omega_1(\widetilde{K}_k,t)\frac{dt}{t},
\end{aligned} ∫ 1 ∞ ∫ ∣ x ˉ ∣ > 2 K k , t ( x ˉ − y ˉ ) − K k , t ( x ˉ ) d x ˉ t d t = ∫ 1 ∞ ∫ ∣ x ˉ ∣ > 2/ t K k ( x ˉ − t y ˉ ) − K k ( x ˉ ) d x ˉ t d t ≤ ∫ 1 ∞ ω 1 ( K k , 1/ t ) t d t = ∫ 0 1 ω 1 ( K k , t ) t d t ,
where
ω 1 ( K ~ k , t ) = sup ∣ y ˉ ∣ = 1 ∫ ∣ K ~ k ( x ˉ − t y ˉ ) − K ~ k ( x ˉ ) ∣ d x ˉ . \omega_1(\widetilde{K}_k,t)=\sup_{|\bar{y}|=1}\int\left|\widetilde{K}_k(\bar{x}-t\bar{y})-\widetilde{K}_k(\bar{x})\right|\,d\bar{x}. ω 1 ( K k , t ) = ∣ y ˉ ∣ = 1 sup ∫ K k ( x ˉ − t y ˉ ) − K k ( x ˉ ) d x ˉ .
Using (40) , it can be deduced that
ω 1 ( K ~ k , t ) = sup ∣ y ˉ ∣ = 1 ∣ ∫ ∣ ∑ i = 1 n + 1 2 k ( n + 2 ) ( x i − t y i ) ϕ i ( 2 k ( x − t y − z , x n + 1 − t y n + 1 − ( D + Φ ( z ) ) ) ) − ∑ i = 1 n + 1 2 k ( n + 2 ) x i ϕ i ( 2 k ( x − z , x n + 1 − ( D + Φ ( z ) ) ) ) ρ ( z ) d z ∣ d x ˉ ≤ ∫ sup ∣ y ˉ ∣ = 1 ∣ ∫ ∣ ∑ i = 1 n + 1 2 k ( n + 2 ) ( x i − t y i ) ϕ i ( 2 k ( x − t y − z , x n + 1 − t y n + 1 − ( D + Φ ( z ) ) ) ) − ∑ i = 1 n + 1 2 k ( n + 2 ) x i ϕ i ( 2 k ( x − z , x n + 1 − ( D + Φ ( z ) ) ) ) d x ˉ ∣ ρ ( z ) d z ≤ ∫ sup ∣ y ˉ ∣ = 1 ∣ ∫ ∣ ∑ i = 1 n + 1 2 k ( n + 2 ) ( x i − t y i ) ϕ i ( 2 k ( x ˉ − t y ˉ ) ) − ∑ i = 1 n + 1 2 k ( n + 2 ) x i ϕ i ( 2 k x ˉ ) ∣ d x ˉ ∣ ρ ( z ) d z + ∫ sup ∣ y ˉ ∣ = 1 ∣ ∑ i = 1 n 2 k ( n + 2 ) z i ϕ i ( 2 k ( x ˉ − t y ˉ ) ) + 2 k ( n + 2 ) ( D + Φ ( z ) ) ϕ n + 1 ( 2 k ( x ˉ − t y ˉ ) ) − ∑ i = 1 n 2 k ( n + 2 ) z i ϕ i ( 2 k x ˉ ) − 2 k ( n + 2 ) ( D + Φ ( z ) ) ϕ n + 1 ( 2 k x ˉ ) ∣ d x ˉ ρ ( z ) d z ≤ C ( 1 + D ) max { 2 k t , min { 2 2 k t , 2 k } } ≤ C ( 1 + D ) max { 1 , 2 k } min { 1 , 2 k t } . \begin{aligned}
\omega_1(\widetilde{K}_k,t)=\sup_{|\bar y|=1}|\int|\sum_{i=1}^{n+1}2^{k(n+2)}(x_i-ty_i)\phi_i(2^k(x-ty-z,x_{n+1}-ty_{n+1}-(D+\Phi(z))))
-\sum_{i=1}^{n+1}2^{k(n+2)}x_i\phi_i(2^k(x-z,x_{n+1}-(D+\Phi(z))))\rho(z)\,dz|\,d\bar x \\
&\leq\int\sup_{|\bar y|=1}|\int|\sum_{i=1}^{n+1}2^{k(n+2)}(x_i-ty_i)\phi_i(2^k(x-ty-z,x_{n+1}-ty_{n+1}-(D+\Phi(z))))
-\sum_{i=1}^{n+1}2^{k(n+2)}x_i\phi_i(2^k(x-z,x_{n+1}-(D+\Phi(z))))\,d\bar x|\rho(z)\,dz \\
&\leq\int\sup_{|\bar y|=1}\left|\int\left|\sum_{i=1}^{n+1}2^{k(n+2)}(x_i-ty_i)\phi_i\left(2^k(\bar x-t\bar y)\right)-\sum_{i=1}^{n+1}2^{k(n+2)}x_i\phi_i(2^k\bar x)\right|\,d\bar x\right|\rho(z)\,dz \\
&+\int\sup_{|\bar y|=1}\left|\sum_{i=1}^{n}2^{k(n+2)}z_i\phi_i\left(2^k(\bar x-t\bar y)\right)+2^{k(n+2)}(D+\Phi(z))\phi_{n+1}\left(2^k(\bar x-t\bar y)\right)
-\sum_{i=1}^{n}2^{k(n+2)}z_i\phi_i(2^k\bar x)-2^{k(n+2)}(D+\Phi(z))\phi_{n+1}(2^k\bar x)\right|\,d\bar x\,\rho(z)\,dz \\
&\leq C(1+D)\max\{2^kt,\min\{2^{2k}t,2^k\}\}\leq C(1+D)\max\{1,2^k\}\min\{1,2^kt\}.
\end{aligned} ω 1 ( K k , t ) = ∣ y ˉ ∣ = 1 sup ∣ ∫ ∣ i = 1 ∑ n + 1 2 k ( n + 2 ) ( x i − t y i ) ϕ i ( 2 k ( x − t y − z , x n + 1 − t y n + 1 − ( D + Φ ( z )))) − i = 1 ∑ n + 1 2 k ( n + 2 ) x i ϕ i ( 2 k ( x − z , x n + 1 − ( D + Φ ( z )))) ρ ( z ) d z ∣ d x ˉ ≤ ∫ ∣ y ˉ ∣ = 1 sup ∣ ∫ ∣ i = 1 ∑ n + 1 2 k ( n + 2 ) ( x i − t y i ) ϕ i ( 2 k ( x − t y − z , x n + 1 − t y n + 1 − ( D + Φ ( z )))) − i = 1 ∑ n + 1 2 k ( n + 2 ) x i ϕ i ( 2 k ( x − z , x n + 1 − ( D + Φ ( z )))) d x ˉ ∣ ρ ( z ) d z ≤ ∫ ∣ y ˉ ∣ = 1 sup ∫ i = 1 ∑ n + 1 2 k ( n + 2 ) ( x i − t y i ) ϕ i ( 2 k ( x ˉ − t y ˉ ) ) − i = 1 ∑ n + 1 2 k ( n + 2 ) x i ϕ i ( 2 k x ˉ ) d x ˉ ρ ( z ) d z + ∫ ∣ y ˉ ∣ = 1 sup i = 1 ∑ n 2 k ( n + 2 ) z i ϕ i ( 2 k ( x ˉ − t y ˉ ) ) + 2 k ( n + 2 ) ( D + Φ ( z )) ϕ n + 1 ( 2 k ( x ˉ − t y ˉ ) ) − i = 1 ∑ n 2 k ( n + 2 ) z i ϕ i ( 2 k x ˉ ) − 2 k ( n + 2 ) ( D + Φ ( z )) ϕ n + 1 ( 2 k x ˉ ) d x ˉ ρ ( z ) d z ≤ C ( 1 + D ) max { 2 k t , min { 2 2 k t , 2 k }} ≤ C ( 1 + D ) max { 1 , 2 k } min { 1 , 2 k t } .
Thus for k ≥ 0 k\geq0 k ≥ 0 ,
∫ 0 1 ω 1 ( K ~ k , t ) d t t ≤ C ( 1 + D ) max { 1 , 2 k } ( ∫ 0 2 − k 2 k d t + ∫ 2 − k 1 d t t ) ≤ C ( 1 + D ) 2 k ( 1 + k ) , \int_0^1\omega_1(\widetilde{K}_k,t)\frac{dt}{t}\leq C(1+D)\max\{1,2^k\}\left(\int_0^{2^{-k}}2^k\,dt+\int_{2^{-k}}^1\frac{dt}{t}\right)\leq C(1+D)2^k(1+k), ∫ 0 1 ω 1 ( K k , t ) t d t ≤ C ( 1 + D ) max { 1 , 2 k } ( ∫ 0 2 − k 2 k d t + ∫ 2 − k 1 t d t ) ≤ C ( 1 + D ) 2 k ( 1 + k ) ,
and for k < 0 k<0 k < 0
∫ 0 1 ω 1 ( K ~ k , t ) d t t ≤ C ( 1 + D ) max { 1 , 2 k } ∫ 0 1 2 k d t ≤ C ( 1 + D ) 2 k . \int_0^1\omega_1(\widetilde{K}_k,t)\frac{dt}{t}\leq C(1+D)\max\{1,2^k\}\int_0^1 2^k\,dt\leq C(1+D)2^k. ∫ 0 1 ω 1 ( K k , t ) t d t ≤ C ( 1 + D ) max { 1 , 2 k } ∫ 0 1 2 k d t ≤ C ( 1 + D ) 2 k .
The proof is complete.
The proof of Lemma 3.3
We first state a vector-valued variant of a theorem about multipliers with localized bounds, which was introduced by Seeger. Readers can find the proof of this theorem in Jones-Seeger-Wright[16 ] and Guo[9 ] .
Proposition 3.5 Let I ⊂ R I\subset\mathbb{R} I ⊂ R be a compact interval. Let m ~ u ( ξ ˉ ) \widetilde m_u(\bar\xi) m u ( ξ ˉ ) , u ∈ I u\in I u ∈ I be a family of Fourier multipliers on R n + 1 \mathbb{R}^{n+1} R n + 1 , each of which is compactly supported on { ξ ˉ : 1 2 ≤ ∣ ξ ˉ ∣ ≤ 2 } \{\bar\xi:\frac{1}{2}\leq|\bar\xi|\leq2\} { ξ ˉ : 2 1 ≤ ∣ ξ ˉ ∣ ≤ 2 } and satisfies
sup u ∈ I ∣ ∂ ξ τ m ~ u ( ξ ˉ ) ∣ ≤ B for each 0 ≤ ∣ τ ∣ ≤ n + 2 \sup_{u\in I}\left|\partial_\xi^\tau\widetilde m_u(\bar\xi)\right|\leq B\quad\text{for each }0\leq|\tau|\leq n+2 u ∈ I sup ∂ ξ τ m u ( ξ ˉ ) ≤ B for each 0 ≤ ∣ τ ∣ ≤ n + 2 for some constant B B B . For u ∈ I u\in I u ∈ I and j ∈ Z j\in\mathbb{Z} j ∈ Z , denote by T u , j T_{u,j} T u , j the multiplier operator with multiplier m ~ u ( 2 − j ξ ˉ ) \widetilde m_u(2^{-j}\bar\xi) m u ( 2 − j ξ ˉ ) . Fix some p ∈ [ 2 , ∞ ) p\in[2,\infty) p ∈ [ 2 , ∞ ) . Assume that there exists some constant A A A such that
sup j ∈ Z ∥ ∥ T u , j f ∥ L 2 ( I ) ∥ L s ( R n + 1 ) ≤ A ∥ f ∥ L s ( R n + 1 ) \sup_{j\in\mathbb{Z}}\left\|\left\|T_{u,j}f\right\|_{L^2(I)}\right\|_{L^s(\mathbb{R}^{n+1})}\leq A\|f\|_{L^s(\mathbb{R}^{n+1})} j ∈ Z sup ∥ T u , j f ∥ L 2 ( I ) L s ( R n + 1 ) ≤ A ∥ f ∥ L s ( R n + 1 )
for both s = p s=p s = p and s = 2 s=2 s = 2 . Then
∥ ∥ ∥ T u , j f ∥ L 2 ( I ) ∥ ℓ 2 ( Z ) ∥ L p ( R n + 1 ) ≲ A ∣ log ( 2 + B A ) ∣ 1 2 − 1 p ∥ f ∥ L p ( R n + 1 ) . \|\|\|T_{u,j}f\|_{L^{2}(I)}\|_{\ell^{2}(\mathbb{Z})}\|_{L^{p}(\mathbb{R}^{n+1})}\lesssim A\left|\log\left(2+\frac{B}{A}\right)\right|^{\frac{1}{2}-\frac{1}{p}}\|f\|_{L^{p}(\mathbb{R}^{n+1})}. ∥∥∥ T u , j f ∥ L 2 ( I ) ∥ ℓ 2 ( Z ) ∥ L p ( R n + 1 ) ≲ A log ( 2 + A B ) 2 1 − p 1 ∥ f ∥ L p ( R n + 1 ) .
Firstly, we consider the case k ≤ 0 k\leq0 k ≤ 0 . Using inequality (2.5), we obtain
∥ V 2 j ( A k f ) ∥ ℓ j 2 ≤ C ∥ A k , t f ∥ L t 2 1 / 2 ∥ t d d t ( A k , t f ) ∥ L t 2 1 / 2 ≤ C ( 1 + D 1 / 2 ) ∥ A k , t f ∥ L t 2 + ( 1 + D 1 / 2 ) − 1 ∥ t d d t ( A k , t f ) ∥ L t 2 . \begin{aligned}
\|V_{2}^{j}(\mathcal{A}_{k}f)\|_{\ell_{j}^{2}}\leq C\|\mathcal{A}_{k,t}f\|_{L_{t}^{2}}^{1/2}\left\|t\frac{d}{dt}(\mathcal{A}_{k,t}f)\right\|_{L_{t}^{2}}^{1/2} \\
&\leq C(1+D^{1/2})\|\mathcal{A}_{k,t}f\|_{L_{t}^{2}}+(1+D^{1/2})^{-1}\left\|t\frac{d}{dt}(\mathcal{A}_{k,t}f)\right\|_{L_{t}^{2}}.
\end{aligned} ∥ V 2 j ( A k f ) ∥ ℓ j 2 ≤ C ∥ A k , t f ∥ L t 2 1/2 t d t d ( A k , t f ) L t 2 1/2 ≤ C ( 1 + D 1/2 ) ∥ A k , t f ∥ L t 2 + ( 1 + D 1/2 ) − 1 t d t d ( A k , t f ) L t 2 .
Recall that the kernel function of A k , t \mathcal{A}_{k,t} A k , t is H k , t = K k , t − G k , t H_{k,t}=K_{k,t}-G_{k,t} H k , t = K k , t − G k , t . As in the proof of Lemma 3.2, we only prove
∥ K k , t ∗ f ∥ L p ( L t 2 ) ≤ C 2 k ϵ ∥ f ∥ L p (43) \|K_{k,t}*f\|_{L^{p}(L_{t}^{2})}\leq C2^{k\epsilon}\|f\|_{L^{p}} \tag*{(43)} ∥ K k , t ∗ f ∥ L p ( L t 2 ) ≤ C 2 k ϵ ∥ f ∥ L p (43)
and
∥ t d d t K k , t ∗ f ∥ L p ( L t 2 ) ≤ C ( 1 + D ) 2 k ϵ ∥ f ∥ L p (44) \left\|t\frac{d}{dt}K_{k,t}*f\right\|_{L^{p}(L_{t}^{2})}\leq C(1+D)2^{k\epsilon}\|f\|_{L^{p}} \tag*{(44)} t d t d K k , t ∗ f L p ( L t 2 ) ≤ C ( 1 + D ) 2 k ϵ ∥ f ∥ L p (44)
for p > 2 p>2 p > 2 .
Using Minkowski’s inequality, it can be deduced from (3.41) that ∥ K k ∥ L 1 ≤ 1 \|K_{k}\|_{L^{1}}\leq1 ∥ K k ∥ L 1 ≤ 1 . Thus, by Hölder inequality on t t t , we obtain
( ∫ R n + 1 ( ∫ 1 2 ∣ A k , t f ( x ‾ ) ∣ 2 d t ) p / 2 d x ‾ ) 1 / p ≤ ( ∫ R n + 1 ∫ 1 2 ∣ A k , t f ( x ‾ ) ∣ p d t d x ‾ ) 1 / p = ( ∫ 1 2 ∫ R n + 1 ∣ A k , t f ( x ‾ ) ∣ p d x ‾ d t ) 1 / p = ( ∫ 1 2 ∫ R n + 1 ∣ H k , t ∗ f ( x ‾ − t D ) ∣ p d x ‾ d t ) 1 / p ≤ C ∥ f ∥ L p . \begin{aligned}
\left(\int_{\mathbb{R}^{n+1}}\left(\int_{1}^{2}|\mathcal{A}_{k,t}f(\overline{x})|^{2}dt\right)^{p/2}d\overline{x}\right)^{1/p}\leq\left(\int_{\mathbb{R}^{n+1}}\int_{1}^{2}|\mathcal{A}_{k,t}f(\overline{x})|^{p}dtd\overline{x}\right)^{1/p} \\
&=\left(\int_{1}^{2}\int_{\mathbb{R}^{n+1}}|\mathcal{A}_{k,t}f(\overline{x})|^{p}d\overline{x}dt\right)^{1/p} \\
&=\left(\int_{1}^{2}\int_{\mathbb{R}^{n+1}}|H_{k,t}*f(\overline{x}-tD)|^{p}d\overline{x}dt\right)^{1/p} \\
&\leq C\|f\|_{L^{p}}.
\end{aligned} ( ∫ R n + 1 ( ∫ 1 2 ∣ A k , t f ( x ) ∣ 2 d t ) p /2 d x ) 1/ p ≤ ( ∫ R n + 1 ∫ 1 2 ∣ A k , t f ( x ) ∣ p d t d x ) 1/ p = ( ∫ 1 2 ∫ R n + 1 ∣ A k , t f ( x ) ∣ p d x d t ) 1/ p = ( ∫ 1 2 ∫ R n + 1 ∣ H k , t ∗ f ( x − t D ) ∣ p d x d t ) 1/ p ≤ C ∥ f ∥ L p .
Then, we get (3.43) if we using Proposition 3.5 with B = O ( 2 k ( n + 2 ) ) B=\mathcal{O}(2^{k(n+2)}) B = O ( 2 k ( n + 2 ) ) and A A A being the constant above.
Using Minkowski’s inequality again, it can be deduced from (3.42) that ∥ K ~ k ∥ L 1 ≤ ( 1 + D ) \|\widetilde{K}_{k}\|_{L^{1}}\leq(1+D) ∥ K k ∥ L 1 ≤ ( 1 + D ) . Note that t d d t K k , t = − ( n + 1 ) K k , t − K ~ k , t t\frac{d}{dt}K_{k,t}=-(n+1)K_{k,t}-\widetilde{K}_{k,t} t d t d K k , t = − ( n + 1 ) K k , t − K k , t . These results imply (3.44).
It remains for us to consider the case k ≥ 1 k\geq1 k ≥ 1 . Using inequality (2.5), we obtain
∥ V 2 j ( A k f ) ∥ ℓ j 2 ≤ C ∥ A k , t f ∥ L t 2 1 / 2 ∥ t d d t ( A k , t f ) ∥ L t 2 1 / 2 ≤ C ( 1 + D 1 / 2 ) 2 k 2 ∥ A k , t f ∥ L t 2 + ( 1 + D 1 / 2 ) − 1 2 − k 2 ∥ t d d t ( A k , t f ) ∥ L t 2 . \begin{aligned}
\|V_{2}^{j}(\mathcal{A}_{k}f)\|_{\ell_{j}^{2}}\leq C\|\mathcal{A}_{k,t}f\|_{L_{t}^{2}}^{1/2}\left\|t\frac{d}{dt}(\mathcal{A}_{k,t}f)\right\|_{L_{t}^{2}}^{1/2} \\
&\leq C(1+D^{1/2})2^{\frac{k}{2}}\|\mathcal{A}_{k,t}f\|_{L_{t}^{2}}+(1+D^{1/2})^{-1}2^{-\frac{k}{2}}\left\|t\frac{d}{dt}(\mathcal{A}_{k,t}f)\right\|_{L_{t}^{2}}.
\end{aligned} ∥ V 2 j ( A k f ) ∥ ℓ j 2 ≤ C ∥ A k , t f ∥ L t 2 1/2 t d t d ( A k , t f ) L t 2 1/2 ≤ C ( 1 + D 1/2 ) 2 2 k ∥ A k , t f ∥ L t 2 + ( 1 + D 1/2 ) − 1 2 − 2 k t d t d ( A k , t f ) L t 2 .
Without loss of generality, using Proposition 3.5 again, it remains for us to prove
∥ A k , t f ∥ L p ( L t 2 ( [ 1 , 2 ] ) ) ≤ C ε 2 − k n + 1 p + k ε ∥ f ∥ L p (45) \|\mathcal{A}_{k,t}f\|_{L^{p}(L_{t}^{2}([1,2]))}\leq C\varepsilon2^{-k\frac{n+1}{p}+k\varepsilon}\|f\|_{L^{p}} \tag*{(45)} ∥ A k , t f ∥ L p ( L t 2 ([ 1 , 2 ])) ≤ C ε 2 − k p n + 1 + k ε ∥ f ∥ L p (45)
and
∥ d d t A k , t ∗ f ∥ L p ( L t 2 ( [ 1 , 2 ] ) ) ≤ C ϵ ( 1 + D ) 2 − k ( n + 1 p − 1 ) + k ϵ ∥ f ∥ L p (46) \left\|\frac{d}{dt}A_{k,t}*f\right\|_{L^p(L_t^2([1,2]))}\le C_{\epsilon}(1+D)2^{-k\left(\frac{n+1}{p}-1\right)+k\epsilon}\|f\|_{L^p} \tag*{(46)} d t d A k , t ∗ f L p ( L t 2 ([ 1 , 2 ])) ≤ C ϵ ( 1 + D ) 2 − k ( p n + 1 − 1 ) + k ϵ ∥ f ∥ L p (46)
for p > 2 + 4 n p>2+\frac{4}{n} p > 2 + n 4 . Now, let us focus our attention on the proof of (45) . Recall that the multipliers associated to operator A k , t A_{k,t} A k , t is
m k ( ξ ˉ ) = μ ( ξ ˉ ) ψ ( 2 − k ∣ ξ ˉ ∣ ) − e − ∣ ξ ˉ ∣ − i D ξ n + 1 ψ ( 2 − k ∣ ξ ˉ ∣ ) . m_k(\bar{\xi})=\mu(\bar{\xi})\psi(2^{-k}|\bar{\xi}|)-e^{-|\bar{\xi}|-iD\xi_{n+1}}\psi(2^{-k}|\bar{\xi}|). m k ( ξ ˉ ) = μ ( ξ ˉ ) ψ ( 2 − k ∣ ξ ˉ ∣ ) − e − ∣ ξ ˉ ∣ − i D ξ n + 1 ψ ( 2 − k ∣ ξ ˉ ∣ ) .
and the expression of μ \mu μ is
μ ( ξ ˉ ) = ∫ e − i ( x ⋅ ξ + ξ n + 1 ( D + Φ ( x ) ) ) ρ ( x ) d x = e − i D ξ n + 1 ∫ e − i ξ n + 1 ( − x ⋅ s + Φ ( x ) ) ρ ( x ) d x \begin{aligned}
\mu(\bar{\xi})=\int e^{-i(x\cdot\xi+\xi_{n+1}(D+\Phi(x)))}\rho(x)\,dx \\
&=e^{-iD\xi_{n+1}}\int e^{-i\xi_{n+1}(-x\cdot s+\Phi(x))}\rho(x)\,dx
\end{aligned} μ ( ξ ˉ ) = ∫ e − i ( x ⋅ ξ + ξ n + 1 ( D + Φ ( x ))) ρ ( x ) d x = e − i D ξ n + 1 ∫ e − i ξ n + 1 ( − x ⋅ s + Φ ( x )) ρ ( x ) d x
where s = ( − ξ 1 ξ n + 1 , … , − ξ n ξ n + 1 ) s=\left(-\frac{\xi_1}{\xi_{n+1}},\ldots,-\frac{\xi_n}{\xi_{n+1}}\right) s = ( − ξ n + 1 ξ 1 , … , − ξ n + 1 ξ n ) for ξ n + 1 ≠ 0 \xi_{n+1}\ne0 ξ n + 1 = 0 . Put
Φ ∗ ( x , s ) = − x ⋅ s + Φ ( x ) . \Phi^*(x,s)=-x\cdot s+\Phi(x). Φ ∗ ( x , s ) = − x ⋅ s + Φ ( x ) .
Then we have
∂ x i Φ ∗ ( x , s ) = − s i + ∂ x i Φ ( x ) \partial_{x_i}\Phi^*(x,s)=-s_i+\partial_{x_i}\Phi(x) ∂ x i Φ ∗ ( x , s ) = − s i + ∂ x i Φ ( x )
and
∂ x i x j 2 Φ ∗ ( x , s ) = ∂ x i x j 2 Φ ( x ) . \partial^2_{x_ix_j}\Phi^*(x,s)=\partial^2_{x_ix_j}\Phi(x). ∂ x i x j 2 Φ ∗ ( x , s ) = ∂ x i x j 2 Φ ( x ) .
Recall that Φ ( x ) = ∑ i = 1 n λ i x i 2 + R ( x ) \Phi(x)=\sum_{i=1}^n\lambda_i x_i^2+R(x) Φ ( x ) = ∑ i = 1 n λ i x i 2 + R ( x ) . Thus we can get
∂ x i x j 2 Φ ∗ ( 0 , s ) = { 2 λ i , i = j , 0 , i ≠ j . (47) \partial^2_{x_ix_j}\Phi^*(0,s)=
\begin{cases}
2\lambda_i, & i=j,\\
0, & i\ne j.
\end{cases} \tag*{(47)} ∂ x i x j 2 Φ ∗ ( 0 , s ) = { 2 λ i , 0 , i = j , i = j . (47)
Since ∣ x ∣ |x| ∣ x ∣ is assumed to be sufficiently small, then the implicit function theorem and (47) imply that there exists a smooth solution x ∗ = x ∗ ( s ) x^*=x^*(s) x ∗ = x ∗ ( s ) of the equation ∂ x i Φ ∗ ( x , s ) = 0 \partial_{x_i}\Phi^*(x,s)=0 ∂ x i Φ ∗ ( x , s ) = 0 for i = 1 , … , n i=1,\ldots,n i = 1 , … , n . From the above arguments and Taylor expansion of the smooth functions Φ ( x ) \Phi(x) Φ ( x ) , the function x ∗ ( s ) x^*(s) x ∗ ( s ) can be written as
x ∗ ( s ) = ( s 1 2 λ 1 , … , s n 2 λ n ) + R p h ( s ) (48) x^*(s)=\left(\frac{s_1}{2\lambda_1},\ldots,\frac{s_n}{2\lambda_n}\right)+R_{ph}(s) \tag*{(48)} x ∗ ( s ) = ( 2 λ 1 s 1 , … , 2 λ n s n ) + R p h ( s ) (48)
where R p h ( 0 ) = D s R p h ( s ) ∣ s = 0 = 0 R_{ph}(0)=D_sR_{ph}(s)|_{s=0}=0 R p h ( 0 ) = D s R p h ( s ) ∣ s = 0 = 0 .
Denote Φ ∗ ∗ ( s ) = − s ⋅ x ∗ ( s ) + Φ ( x ∗ ( s ) ) \Phi^{**}(s)=-s\cdot x^*(s)+\Phi(x^*(s)) Φ ∗∗ ( s ) = − s ⋅ x ∗ ( s ) + Φ ( x ∗ ( s )) . By applying the method of stationary phase, we obtain
μ ( ξ ˉ ) = e − i D ξ n + 1 ( e i ξ n + 1 Φ ∗ ∗ ( s ) A m ( ξ ˉ ) ( 1 + ∣ ξ ˉ ∣ ) n / 2 χ ( s ) + A r e ( ξ ˉ ) ) . (49) \mu(\bar{\xi})=e^{-iD\xi_{n+1}}\left(e^{i\xi_{n+1}\Phi^{**}(s)}\frac{A_m(\bar{\xi})}{(1+|\bar{\xi}|)^{n/2}}\chi(s)+A_{re}(\bar{\xi})\right). \tag*{(49)} μ ( ξ ˉ ) = e − i D ξ n + 1 ( e i ξ n + 1 Φ ∗∗ ( s ) ( 1 + ∣ ξ ˉ ∣ ) n /2 A m ( ξ ˉ ) χ ( s ) + A r e ( ξ ˉ ) ) . (49)
where χ ( s ) \chi(s) χ ( s ) is a smooth function supported in [ 0 , c ] [0,c] [ 0 , c ] , A m ( ξ ˉ ) A_m(\bar{\xi}) A m ( ξ ˉ ) is a symbol of order of zero. More precisely, A m ( ξ ˉ ) A_m(\bar{\xi}) A m ( ξ ˉ ) satisfies
∣ D ξ ˉ α ˉ A m ( ξ ˉ ) ∣ ≤ C ( 1 + ∣ ξ ˉ ∣ ) − ∣ α ˉ ∣ . \left|D_{\bar{\xi}}^{\bar{\alpha}}A_m(\bar{\xi})\right|\le C(1+|\bar{\xi}|)^{-|\bar{\alpha}|}. D ξ ˉ α ˉ A m ( ξ ˉ ) ≤ C ( 1 + ∣ ξ ˉ ∣ ) − ∣ α ˉ ∣ .
Moreover, the remainder term A r e ( ξ ˉ ) A_{re}(\bar{\xi}) A r e ( ξ ˉ ) satisfies
∣ D ξ ˉ α ˉ A r e ( ξ ˉ ) ∣ ≤ C N , α ˉ ( 1 + ∣ ξ ˉ ∣ ) − N . \left|D_{\bar{\xi}}^{\bar{\alpha}}A_{re}(\bar{\xi})\right|\le C_{N,\bar{\alpha}}(1+|\bar{\xi}|)^{-N}. D ξ ˉ α ˉ A r e ( ξ ˉ ) ≤ C N , α ˉ ( 1 + ∣ ξ ˉ ∣ ) − N .
At last, the multiplier corresponding to A k , t \mathcal{A}_{k,t} A k , t can be written by
m k ( ξ ˉ ) = e − i D ξ n + 1 ( e i ξ n + 1 Φ ∗ ∗ ( s ) A m ( ξ ˉ ) ( 1 + ∣ ξ ˉ ∣ ) n 2 χ ( s ) ) ψ ( 2 − k ξ ˉ ) + e − i D ξ n + 1 A r e ( ξ ˉ ) ψ ( 2 − k ξ ˉ ) , m_k(\bar{\xi})=e^{-iD\xi_{n+1}}\left(e^{i\xi_{n+1}\Phi^{**}(s)}\frac{A_m(\bar{\xi})}{(1+|\bar{\xi}|)^{\frac{n}{2}}}\chi(s)\right)\psi(2^{-k}\bar{\xi})+e^{-iD\xi_{n+1}}A_{re}(\bar{\xi})\psi(2^{-k}\bar{\xi}), m k ( ξ ˉ ) = e − i D ξ n + 1 ( e i ξ n + 1 Φ ∗∗ ( s ) ( 1 + ∣ ξ ˉ ∣ ) 2 n A m ( ξ ˉ ) χ ( s ) ) ψ ( 2 − k ξ ˉ ) + e − i D ξ n + 1 A r e ( ξ ˉ ) ψ ( 2 − k ξ ˉ ) ,
where
∣ D ξ ˉ α 2 ˉ A r e ( ξ ˉ ) ∣ ≤ C N , α ˉ ( 1 + ∣ ξ ˉ ∣ ) − N . \left|D_{\bar{\xi}}^{\frac{\alpha}{\bar{2}}}A_{re}(\bar{\xi})\right|\leq C_{N,\bar{\alpha}}(1+|\bar{\xi}|)^{-N}. D ξ ˉ 2 ˉ α A r e ( ξ ˉ ) ≤ C N , α ˉ ( 1 + ∣ ξ ˉ ∣ ) − N .
Let
M k , t f ( x ˉ ) = 1 ( 2 π ) n + 1 ∫ R n + 1 e i t ( x ˉ ⋅ ξ ˉ − D ξ n + 1 + ξ n + 1 Φ ∗ ∗ ( s ) ) A m ( ξ ˉ ) ( 1 + ∣ ξ ˉ ∣ ) n 2 χ ( s ) ψ ( 2 − k ξ ˉ ) f ^ ( ξ ˉ ) d ξ ˉ \mathcal{M}_{k,t}f(\bar{x})=\frac{1}{(2\pi)^{n+1}}\int_{\mathbb{R}^{n+1}}e^{it(\bar{x}\cdot\bar{\xi}-D\xi_{n+1}+\xi_{n+1}\Phi^{**}(s))}\frac{A_m(\bar{\xi})}{(1+|\bar{\xi}|)^{\frac{n}{2}}}\chi(s)\psi(2^{-k}\bar{\xi})\widehat{f}(\bar{\xi})d\bar{\xi} M k , t f ( x ˉ ) = ( 2 π ) n + 1 1 ∫ R n + 1 e i t ( x ˉ ⋅ ξ ˉ − D ξ n + 1 + ξ n + 1 Φ ∗∗ ( s )) ( 1 + ∣ ξ ˉ ∣ ) 2 n A m ( ξ ˉ ) χ ( s ) ψ ( 2 − k ξ ˉ ) f ( ξ ˉ ) d ξ ˉ
denote the main term and let
B k , t f ( x ˉ ) = 1 ( 2 π ) n + 1 ∫ R n + 1 e i t ( x ˉ ⋅ ξ ˉ − D ξ n + 1 ) A r e ( ξ ˉ ) ψ ( 2 − k ξ ˉ ) f ^ ( ξ ˉ ) d ξ ˉ \mathcal{B}_{k,t}f(\bar{x})=\frac{1}{(2\pi)^{n+1}}\int_{\mathbb{R}^{n+1}}e^{it(\bar{x}\cdot\bar{\xi}-D\xi_{n+1})}A_{re}(\bar{\xi})\psi(2^{-k}\bar{\xi})\widehat{f}(\bar{\xi})d\bar{\xi} B k , t f ( x ˉ ) = ( 2 π ) n + 1 1 ∫ R n + 1 e i t ( x ˉ ⋅ ξ ˉ − D ξ n + 1 ) A r e ( ξ ˉ ) ψ ( 2 − k ξ ˉ ) f ( ξ ˉ ) d ξ ˉ
denote the remainder term. This reduces the matter to proving the associated desired estimates for M k , t \mathcal{M}_{k,t} M k , t and B k , t \mathcal{B}_{k,t} B k , t .
The estimate of the remainder term. Denote the kernel function of B k , t \mathcal{B}_{k,t} B k , t by
K e k ( x ˉ , t ) = 1 ( 2 π ) n + 1 ∫ R n + 1 e i t ( x ˉ ⋅ ξ ˉ − D ξ n + 1 ) A r e ( ξ ˉ ) ψ ( 2 − k ξ ˉ ) d ξ ˉ . K_{ek}(\bar{x},t)=\frac{1}{(2\pi)^{n+1}}\int_{\mathbb{R}^{n+1}}e^{it(\bar{x}\cdot\bar{\xi}-D\xi_{n+1})}A_{re}(\bar{\xi})\psi(2^{-k}\bar{\xi})d\bar{\xi}. K e k ( x ˉ , t ) = ( 2 π ) n + 1 1 ∫ R n + 1 e i t ( x ˉ ⋅ ξ ˉ − D ξ n + 1 ) A r e ( ξ ˉ ) ψ ( 2 − k ξ ˉ ) d ξ ˉ .
Integrating by parts in ξ ˉ \bar{\xi} ξ ˉ , since t ≈ 1 t\approx1 t ≈ 1 , we obtain
∣ K e k ( x ˉ , t ) ∣ ≲ N 2 − k N 2 k ( n + 1 ) 1 + ∣ 2 k x ˉ ∣ N + ( 2 k ∣ x n + 1 − D ∣ ) N . |K_{ek}(\bar{x},t)|\lesssim_N 2^{-kN}\frac{2^{k(n+1)}}{1+|2^k\bar{x}|^N+(2^k|x_{n+1}-D|)^N}. ∣ K e k ( x ˉ , t ) ∣ ≲ N 2 − k N 1 + ∣ 2 k x ˉ ∣ N + ( 2 k ∣ x n + 1 − D ∣ ) N 2 k ( n + 1 ) .
Using Hölder’s inequality and Young’s inequality for convolution, it can be deduced that
∥ B k , t f ∥ L p ( L 2 ( [ 1 , 2 ] ) ) = ∥ K e k ( x ˉ , t ) ∗ x ˉ f ∥ L p ( L 2 ( [ 1 , 2 ] ) ) ≲ N 2 − k N ∥ f ∥ L p (50) \|\mathcal{B}_{k,t}f\|_{L^p(L^2([1,2]))}=\|K_{ek}(\bar{x},t)*_{\bar{x}}f\|_{L^p(L^2([1,2]))}\lesssim_N 2^{-kN}\|f\|_{L^p} \tag*{(50)} ∥ B k , t f ∥ L p ( L 2 ([ 1 , 2 ])) = ∥ K e k ( x ˉ , t ) ∗ x ˉ f ∥ L p ( L 2 ([ 1 , 2 ])) ≲ N 2 − k N ∥ f ∥ L p (50)
for p ≥ 2 p\geq2 p ≥ 2 . In order to prove (3.45), it remains for us to consider the estimate associated to the main term.
The estimate of the main term. Before we begin our proof, we introduce two inequalities about local smoothing estimates which can be found in the proof of Theorem 3.1 and Theorem 3.2 in [24 ] .
Lemma 3.6 Suppose a ( ξ ˉ , x ˉ ) a(\bar{\xi},\bar{x}) a ( ξ ˉ , x ˉ ) is a symbol of order zero. That is
∣ ( ∂ ∂ ξ ˉ ) α ( ∂ ∂ x ˉ ) β a ( ξ ˉ , x ˉ ) ∣ ≤ C α , β ( 1 + ∣ ξ ˉ ∣ ) − ∣ α ∣ . \left|\left(\frac{\partial}{\partial\bar{\xi}}\right)^\alpha\left(\frac{\partial}{\partial\bar{x}}\right)^\beta a(\bar{\xi},\bar{x})\right|\leq C_{\alpha,\beta}(1+|\bar{\xi}|)^{-|\alpha|}. ( ∂ ξ ˉ ∂ ) α ( ∂ x ˉ ∂ ) β a ( ξ ˉ , x ˉ ) ≤ C α , β ( 1 + ∣ ξ ˉ ∣ ) − ∣ α ∣ . Denote
Q λ f ( x ˉ , t ) = ψ ~ ( t ) ∫ R n + 1 e i t ( x ˉ ⋅ ξ ˉ − h ( ξ ˉ ) ) a λ ( ξ ˉ , x ˉ , λ ) f ^ ( ξ ˉ ) d ξ ˉ , (51) Q_\lambda f(\bar{x},t)=\widetilde{\psi}(t)\int_{\mathbb{R}^{n+1}}e^{it(\bar{x}\cdot\bar{\xi}-h(\bar{\xi}))}a_\lambda(\bar{\xi},\bar{x},\lambda)\widehat{f}(\bar{\xi})d\bar{\xi}, \tag*{(51)} Q λ f ( x ˉ , t ) = ψ ( t ) ∫ R n + 1 e i t ( x ˉ ⋅ ξ ˉ − h ( ξ ˉ )) a λ ( ξ ˉ , x ˉ , λ ) f ( ξ ˉ ) d ξ ˉ , (51) where a λ ( ξ ˉ , x ˉ ) = a ( ξ ˉ , x ˉ ) ψ ( λ − 1 ∣ ξ ˉ ∣ ) a_\lambda(\bar{\xi},\bar{x})=a(\bar{\xi},\bar{x})\psi(\lambda^{-1}|\bar{\xi}|) a λ ( ξ ˉ , x ˉ ) = a ( ξ ˉ , x ˉ ) ψ ( λ − 1 ∣ ξ ˉ ∣ ) , h ( ⋅ ) h(\cdot) h ( ⋅ ) is homogeneous of degree 1 and ψ ~ ( t ) \widetilde{\psi}(t) ψ ( t ) is a positive smooth function supported in [ 1 / 2 , 5 / 2 ] [1/2,5/2] [ 1/2 , 5/2 ] and ψ ~ ( t ) = 1 \widetilde{\psi}(t)=1 ψ ( t ) = 1 for t ∈ [ 1 , 2 ] t\in[1,2] t ∈ [ 1 , 2 ] . Then we have
∥ Q λ f ∥ L p ( R n + 1 , L 2 ( R ) ) ≤ λ ( n + 1 ) ( 1 2 − 1 p ) − 1 2 ∥ f ∥ L 2 ( R n + 1 ) , (52) \|Q_\lambda f\|_{L^p(\mathbb{R}^{n+1},L^2(\mathbb{R}))}\leq\lambda^{(n+1)\left(\frac{1}{2}-\frac{1}{p}\right)-\frac{1}{2}}\|f\|_{L^2(\mathbb{R}^{n+1})}, \tag*{(52)} ∥ Q λ f ∥ L p ( R n + 1 , L 2 ( R )) ≤ λ ( n + 1 ) ( 2 1 − p 1 ) − 2 1 ∥ f ∥ L 2 ( R n + 1 ) , (52)
and
∥ Q λ f ∥ L p ( R n + 2 ) ≤ λ ( n + 1 ) ( 1 2 − 1 p ) − 1 p ∥ f ∥ L 2 ( R n + 1 ) , (53) \lVert Q_\lambda f\rVert_{L^p(\mathbb{R}^{n+2})}\leq\lambda^{(n+1)\left(\frac{1}{2}-\frac{1}{p}\right)-\frac{1}{p}}\lVert f\rVert_{L^2(\mathbb{R}^{n+1})}, \tag*{(53)} ∥ Q λ f ∥ L p ( R n + 2 ) ≤ λ ( n + 1 ) ( 2 1 − p 1 ) − p 1 ∥ f ∥ L 2 ( R n + 1 ) , (53)
for p > 2 + 4 n p>2+\frac{4}{n} p > 2 + n 4 .
Denote
M k , t ∗ f ( x ˉ ) = 1 ( 2 π ) n + 1 ∫ R n + 1 e i t ( x ˉ ⋅ ξ ˉ + ξ n + 1 Φ ∗ ∗ ( s ) ) A m ( ξ ˉ ) ( 1 + ∣ ξ ˉ ∣ ) n 2 χ ( s ) ψ ( 2 − k ∣ ξ ˉ ∣ ) f ^ ( ξ ˉ ) d ξ ˉ . \mathcal{M}_{k,t}^{*}f(\bar{x})=\frac{1}{(2\pi)^{n+1}}\int_{\mathbb{R}^{n+1}}e^{it(\bar{x}\cdot\bar{\xi}+\xi_{n+1}\Phi^{**}(s))}\frac{A_m(\bar{\xi})}{(1+|\bar{\xi}|)^{\frac{n}{2}}}\chi(s)\psi(2^{-k}|\bar{\xi}|)\widehat{f}(\bar{\xi})d\bar{\xi}. M k , t ∗ f ( x ˉ ) = ( 2 π ) n + 1 1 ∫ R n + 1 e i t ( x ˉ ⋅ ξ ˉ + ξ n + 1 Φ ∗∗ ( s )) ( 1 + ∣ ξ ˉ ∣ ) 2 n A m ( ξ ˉ ) χ ( s ) ψ ( 2 − k ∣ ξ ˉ ∣ ) f ( ξ ˉ ) d ξ ˉ .
In order to use Lemma 3.6, we let h ( ξ ˉ ) = − ξ n + 1 Φ ∗ ∗ ( s ) h(\bar{\xi})=-\xi_{n+1}\Phi^{**}(s) h ( ξ ˉ ) = − ξ n + 1 Φ ∗∗ ( s ) and
a ( ξ ˉ , x ˉ ) = 2 k ⋅ n 2 A m ( ξ ˉ ) ( 1 + ∣ ξ ˉ ∣ ) n 2 χ ( s ) ψ ( 2 − k ∣ ξ ˉ ∣ ) a(\bar{\xi},\bar{x})=2^{k\cdot\frac{n}{2}}\frac{A_m(\bar{\xi})}{(1+|\bar{\xi}|)^{\frac{n}{2}}}\chi(s)\psi(2^{-k}|\bar{\xi}|) a ( ξ ˉ , x ˉ ) = 2 k ⋅ 2 n ( 1 + ∣ ξ ˉ ∣ ) 2 n A m ( ξ ˉ ) χ ( s ) ψ ( 2 − k ∣ ξ ˉ ∣ )
since ∣ ξ ˉ ∣ ≈ 2 k |\bar{\xi}|\approx2^k ∣ ξ ˉ ∣ ≈ 2 k . It follows that,
∥ M k , t ∗ f ∥ L p ( R n + 1 , L 2 ( [ 1 , 2 ] ) ) ≤ C 2 − k ⋅ n + 1 p ∥ f ∥ L 2 ( R n + 1 ) , (54) \lVert\mathcal{M}_{k,t}^{*}f\rVert_{L^p(\mathbb{R}^{n+1},L^2([1,2]))}\leq C2^{-k\cdot\frac{n+1}{p}}\lVert f\rVert_{L^2(\mathbb{R}^{n+1})}, \tag*{(54)} ∥ M k , t ∗ f ∥ L p ( R n + 1 , L 2 ([ 1 , 2 ])) ≤ C 2 − k ⋅ p n + 1 ∥ f ∥ L 2 ( R n + 1 ) , (54)
and
∥ M k , t ∗ f ∥ L p ( R n + 1 , L p ( [ 1 , 2 ] ) ) ≤ C 2 k ( ( n + 1 ) ( 1 2 − 1 p ) − 1 p − n 2 ) ∥ f ∥ L 2 ( R n + 1 ) , (55) \lVert\mathcal{M}_{k,t}^{*}f\rVert_{L^p(\mathbb{R}^{n+1},L^p([1,2]))}\leq C2^{k\left((n+1)\left(\frac{1}{2}-\frac{1}{p}\right)-\frac{1}{p}-\frac{n}{2}\right)}\lVert f\rVert_{L^2(\mathbb{R}^{n+1})}, \tag*{(55)} ∥ M k , t ∗ f ∥ L p ( R n + 1 , L p ([ 1 , 2 ])) ≤ C 2 k ( ( n + 1 ) ( 2 1 − p 1 ) − p 1 − 2 n ) ∥ f ∥ L 2 ( R n + 1 ) , (55)
for p > 2 + 4 n p>2+\frac{4}{n} p > 2 + n 4 .
Recalling that h ( ⋅ ) h(\cdot) h ( ⋅ ) is homogeneous of degree 1, we have h ( ξ ˉ ) = ∑ i = 1 n + 1 ξ i ( ∂ i h ) ( ξ ˉ ∣ ξ ˉ ∣ ) h(\bar{\xi})=\sum_{i=1}^{n+1}\xi_i(\partial_i h)\left(\frac{\bar{\xi}}{|\bar{\xi}|}\right) h ( ξ ˉ ) = ∑ i = 1 n + 1 ξ i ( ∂ i h ) ( ∣ ξ ˉ ∣ ξ ˉ ) . Thus we can rewrite M k , t ∗ f \mathcal{M}_{k,t}^{*}f M k , t ∗ f as a convolution, that is,
M k , t ∗ f ( x ˉ ) = ( f ∗ x ˉ K m , k ∗ ) ( x ˉ , t ) , (56) \mathcal{M}_{k,t}^{*}f(\bar{x})=(f*_{\bar{x}}K_{m,k}^{*})(\bar{x},t), \tag*{(56)} M k , t ∗ f ( x ˉ ) = ( f ∗ x ˉ K m , k ∗ ) ( x ˉ , t ) , (56)
where the kernel function K m , k ∗ K_{m,k}^{*} K m , k ∗ is given by
K m , k ∗ ( x ˉ , t ) = 1 ( 2 π ) n + 1 ∫ R n + 1 e i t ( ∑ i = 1 n + 1 ξ i ( x i − ( ∂ i h ) ( ξ ˉ ∣ ξ ˉ ∣ ) ) ) A m ( ξ ˉ ) ( 1 + ∣ ξ ˉ ∣ ) n 2 χ ( s ) ψ ( 2 − k ∣ ξ ˉ ∣ ) d ξ ˉ . (57) K_{m,k}^{*}(\bar{x},t)=\frac{1}{(2\pi)^{n+1}}\int_{\mathbb{R}^{n+1}}e^{it\left(\sum_{i=1}^{n+1}\xi_i\left(x_i-(\partial_i h)\left(\frac{\bar{\xi}}{|\bar{\xi}|}\right)\right)\right)}\frac{A_m(\bar{\xi})}{(1+|\bar{\xi}|)^{\frac{n}{2}}}\chi(s)\psi(2^{-k}|\bar{\xi}|)d\bar{\xi}. \tag*{(57)} K m , k ∗ ( x ˉ , t ) = ( 2 π ) n + 1 1 ∫ R n + 1 e i t ( ∑ i = 1 n + 1 ξ i ( x i − ( ∂ i h ) ( ∣ ξ ˉ ∣ ξ ˉ ) ) ) ( 1 + ∣ ξ ˉ ∣ ) 2 n A m ( ξ ˉ ) χ ( s ) ψ ( 2 − k ∣ ξ ˉ ∣ ) d ξ ˉ . (57)
Let C = sup { ∣ ( ∂ i h ) ( θ ) ∣ : θ ˉ ∈ S n , i = 1 , … , n + 1 } C=\sup\{|(\partial_i h)(\theta)|:\bar{\theta}\in S^n,\ i=1,\ldots,n+1\} C = sup { ∣ ( ∂ i h ) ( θ ) ∣ : θ ˉ ∈ S n , i = 1 , … , n + 1 } . By integrating by parts in ξ ˉ \bar{\xi} ξ ˉ and t ≈ 1 t\approx1 t ≈ 1 , we can infer that for any N ∈ N N\in\mathbb{N} N ∈ N ,
∣ K m , k ∗ ( x ˉ , t ) ∣ ≲ N 2 − k N 2 k ( n + 1 ) 1 + ∣ 2 k x ˉ ∣ N (58) |K_{m,k}^{*}(\bar{x},t)|\lesssim_N 2^{-kN}\frac{2^{k(n+1)}}{1+|2^k\bar{x}|^N} \tag*{(58)} ∣ K m , k ∗ ( x ˉ , t ) ∣ ≲ N 2 − k N 1 + ∣ 2 k x ˉ ∣ N 2 k ( n + 1 ) (58)
whenever ∣ x ˉ ∣ ≥ 2 ( n + 1 ) C |\bar{x}|\geq2(n+1)C ∣ x ˉ ∣ ≥ 2 ( n + 1 ) C .
For z ˉ : = ( z 1 , ⋯ , z n + 1 ) \bar{z}:=(z_1,\cdots,z_{n+1}) z ˉ := ( z 1 , ⋯ , z n + 1 ) , denote by Q z ˉ Q_{\bar{z}} Q z ˉ the cube ∏ i = 1 n + 1 [ z i , z i + C ) \prod_{i=1}^{n+1}[z_i,z_i+C) ∏ i = 1 n + 1 [ z i , z i + C ) and denote by Q z ˉ ∗ Q_{\bar{z}}^{*} Q z ˉ ∗ the cube centered at z ˉ \bar{z} z ˉ with side-length 2 ( n + 1 ) C 2(n+1)C 2 ( n + 1 ) C . By using the notation f Q z ˉ ∗ : = f 1 Q z ˉ ∗ f_{Q_{\bar{z}}^{*}}:=f1_{Q_{\bar{z}}^{*}} f Q z ˉ ∗ := f 1 Q z ˉ ∗ , and recalling the expression (3.56), we can rewrite M k , t ∗ f ( x ˉ ) \mathcal{M}_{k,t}^{*}f(\bar{x}) M k , t ∗ f ( x ˉ ) as
∑ z ˉ ( f Q z ˉ ∗ ∗ x ˉ K m , k ∗ ) ( x ˉ , t ) = ∑ z ˉ ( f Q z ˉ ∗ ∗ x ˉ K m , k ∗ ) ( x ˉ , t ) 1 Q z ˉ ∗ ( x ˉ ) + ∑ z ˉ ( f Q z ˉ ∗ ∗ x ˉ K m , k ∗ ) ( x ˉ , t ) 1 ( Q z ˉ ∗ ) c ( x ˉ ) = : T 1 m k ( x ˉ , t ) + T 2 m k ( x ˉ , t ) . \sum_{\bar{z}}(f_{Q_{\bar{z}}^{*}}*_{\bar{x}}K_{m,k}^{*})(\bar{x},t)
=
\sum_{\bar{z}}(f_{Q_{\bar{z}}^{*}}*_{\bar{x}}K_{m,k}^{*})(\bar{x},t)1_{Q_{\bar{z}}^{*}}(\bar{x})
+
\sum_{\bar{z}}(f_{Q_{\bar{z}}^{*}}*_{\bar{x}}K_{m,k}^{*})(\bar{x},t)1_{(Q_{\bar{z}}^{*})^c}(\bar{x})
=:T_{1mk}(\bar{x},t)+T_{2mk}(\bar{x},t). z ˉ ∑ ( f Q z ˉ ∗ ∗ x ˉ K m , k ∗ ) ( x ˉ , t ) = z ˉ ∑ ( f Q z ˉ ∗ ∗ x ˉ K m , k ∗ ) ( x ˉ , t ) 1 Q z ˉ ∗ ( x ˉ ) + z ˉ ∑ ( f Q z ˉ ∗ ∗ x ˉ K m , k ∗ ) ( x ˉ , t ) 1 ( Q z ˉ ∗ ) c ( x ˉ ) =: T 1 mk ( x ˉ , t ) + T 2 mk ( x ˉ , t ) .
By (54) , we can get the estimate
∥ T 1 m k ∥ L p ( L t 2 ( [ 1 , 2 ] ) ) ≲ ( ∑ z ˉ ∥ M k , t ∗ f Q z ˉ ∥ L p ( L t 2 ( [ 1 , 2 ] ) ) p ) 1 / p ≲ 2 − k ⋅ n + 1 p ( ∑ z ˉ ∥ f Q z ˉ ∥ L p p ) 1 / p . (59) \begin{aligned}
\|\mathcal{T}_{1mk}\|_{L^{p}(L_{t}^{2}([1,2]))}
&\lesssim\left(\sum_{\bar z}\|\mathcal{M}_{k,t}^{*}f_{Q_{\bar z}}\|_{L^{p}(L_{t}^{2}([1,2]))}^{p}\right)^{1/p}\\
&\lesssim2^{-k\cdot\frac{n+1}{p}}\left(\sum_{\bar z}\|f_{Q_{\bar z}}\|_{L^{p}}^{p}\right)^{1/p}. \tag*{(59)}
\end{aligned} ∥ T 1 mk ∥ L p ( L t 2 ([ 1 , 2 ])) ≲ ( z ˉ ∑ ∥ M k , t ∗ f Q z ˉ ∥ L p ( L t 2 ([ 1 , 2 ])) p ) 1/ p ≲ 2 − k ⋅ p n + 1 ( z ˉ ∑ ∥ f Q z ˉ ∥ L p p ) 1/ p . (59)
Using Fubini’s theorem, we obtain ∥ f Q z ˉ ∥ L 2 ≲ ∥ f Q z ˉ ∥ L p \|f_{Q_{\bar z}}\|_{L^{2}}\lesssim\|f_{Q_{\bar z}}\|_{L^{p}} ∥ f Q z ˉ ∥ L 2 ≲ ∥ f Q z ˉ ∥ L p . It follows from (59) that
∥ T 1 m k ∥ L p ( L t 2 ( [ 1 , 2 ] ) ) ≲ 2 − k ⋅ n + 1 p ∥ f Q z ˉ ∥ L p . (60) \|\mathcal{T}_{1mk}\|_{L^{p}(L_{t}^{2}([1,2]))}\lesssim2^{-k\cdot\frac{n+1}{p}}\|f_{Q_{\bar z}}\|_{L^{p}}. \tag*{(60)} ∥ T 1 mk ∥ L p ( L t 2 ([ 1 , 2 ])) ≲ 2 − k ⋅ p n + 1 ∥ f Q z ˉ ∥ L p . (60)
On the other hand, using (58) , we immediately deduce the pointwise inequality
∣ T 2 m k ( x ‾ , t ) ∣ ≲ N 2 − k N ∣ f ∣ ∗ E k , N , t ( x ‾ ) , (61) |\mathcal{T}_{2mk}(\overline{x},t)|\lesssim_{N}2^{-kN}|f|*\mathcal{E}_{k,N,t}(\overline{x}), \tag*{(61)} ∣ T 2 mk ( x , t ) ∣ ≲ N 2 − k N ∣ f ∣ ∗ E k , N , t ( x ) , (61)
where
E k , N , t = 2 k ( n + 1 ) t n + 1 1 + ∣ 2 k t x ‾ ∣ N . \mathcal{E}_{k,N,t}=\frac{2^{k(n+1)}t^{n+1}}{1+|2^{kt}\overline{x}|^{N}}. E k , N , t = 1 + ∣ 2 k t x ∣ N 2 k ( n + 1 ) t n + 1 .
Then, taking the L p ( L t 2 ( [ 1 , 2 ] ) ) L^{p}(L_{t}^{2}([1,2])) L p ( L t 2 ([ 1 , 2 ])) norm on both sides of (61) , and using Hölder’s inequality and Young’s inequality for convolution, we can obtain
∥ T 2 m k ∥ L p ( L t 2 ( [ 1 , 2 ] ) ) ≲ N 2 − k N ∥ f ∥ L p . (62) \|\mathcal{T}_{2mk}\|_{L^{p}(L_{t}^{2}([1,2]))}\lesssim_{N}2^{-kN}\|f\|_{L^{p}}. \tag*{(62)} ∥ T 2 mk ∥ L p ( L t 2 ([ 1 , 2 ])) ≲ N 2 − k N ∥ f ∥ L p . (62)
It can be deduced from (60) and (62) that
∥ M k , t ∗ f ∥ L p ( R n + 1 , L 2 ( [ 1 , 2 ] ) ) ≤ C ϵ 2 − k ⋅ n + 1 p + k ϵ ∥ f ∥ L p ( R n + 1 ) . (63) \|\mathcal{M}_{k,t}^{*}f\|_{L^{p}(\mathbb{R}^{n+1},L^{2}([1,2]))}\leq C_{\epsilon}2^{-k\cdot\frac{n+1}{p}+k\epsilon}\|f\|_{L^{p}(\mathbb{R}^{n+1})}. \tag*{(63)} ∥ M k , t ∗ f ∥ L p ( R n + 1 , L 2 ([ 1 , 2 ])) ≤ C ϵ 2 − k ⋅ p n + 1 + k ϵ ∥ f ∥ L p ( R n + 1 ) . (63)
After going through the same steps to prove (63) , we obtain
∥ M k , t ∗ f ∥ L p ( R n + 1 , L p ( [ 1 , 2 ] ) ) ≤ C ϵ 2 k ( ( n + 1 ) ( 1 2 − 1 p ) − 1 p + n 2 + ϵ ) ∥ f ∥ L p ( R n + 1 ) . (64) \|\mathcal{M}_{k,t}^{*}f\|_{L^{p}(\mathbb{R}^{n+1},L^{p}([1,2]))}\leq C_{\epsilon}2^{k\left((n+1)\left(\frac{1}{2}-\frac{1}{p}\right)-\frac{1}{p}+\frac{n}{2}+\epsilon\right)}\|f\|_{L^{p}(\mathbb{R}^{n+1})}. \tag*{(64)} ∥ M k , t ∗ f ∥ L p ( R n + 1 , L p ([ 1 , 2 ])) ≤ C ϵ 2 k ( ( n + 1 ) ( 2 1 − p 1 ) − p 1 + 2 n + ϵ ) ∥ f ∥ L p ( R n + 1 ) . (64)
Note that M k , t f ( x ‾ ) = M k , t ∗ f ( x , x n + 1 − D ) \mathcal{M}_{k,t}f(\overline{x})=\mathcal{M}_{k,t}^{*}f(x,x_{n+1}-D) M k , t f ( x ) = M k , t ∗ f ( x , x n + 1 − D ) . Thus we can get from (63) ,
∥ M k , t f ∥ L p ( R n + 1 , L 2 ( [ 1 , 2 ] ) ) ≤ C ϵ , D 2 − k ⋅ n + 1 p + k ϵ ∥ f ∥ L p ( R n + 1 ) . (65) \|\mathcal{M}_{k,t}f\|_{L^{p}(\mathbb{R}^{n+1},L^{2}([1,2]))}\leq C_{\epsilon,D}2^{-k\cdot\frac{n+1}{p}+k\epsilon}\|f\|_{L^{p}(\mathbb{R}^{n+1})}. \tag*{(65)} ∥ M k , t f ∥ L p ( R n + 1 , L 2 ([ 1 , 2 ])) ≤ C ϵ , D 2 − k ⋅ p n + 1 + k ϵ ∥ f ∥ L p ( R n + 1 ) . (65)
On the other hand, using Hölder’s inequality, we have
∥ M k , t f ∥ L p ( R n + 1 , L 2 ( [ 1 , 2 ] ) ) ≤ ∥ M k , t f ∥ L p ( R n + 1 , L p ( [ 1 , 2 ] ) ) . \|\mathcal{M}_{k,t}f\|_{L^{p}(\mathbb{R}^{n+1},L^{2}([1,2]))}\leq\|\mathcal{M}_{k,t}f\|_{L^{p}(\mathbb{R}^{n+1},L^{p}([1,2]))}. ∥ M k , t f ∥ L p ( R n + 1 , L 2 ([ 1 , 2 ])) ≤ ∥ M k , t f ∥ L p ( R n + 1 , L p ([ 1 , 2 ])) .
Combining the inequality above with (64) , it can be deduced that
∥ M k , t f ∥ L p ( R n + 1 , L 2 ( [ 1 , 2 ] ) ) ≤ C ϵ 2 k ( ( n + 1 ) ( 1 2 − 1 p ) − 1 p + n 2 + ϵ ) ∥ f ∥ L p ( R n + 1 ) . (66) \|\mathcal{M}_{k,t}f\|_{L^{p}(\mathbb{R}^{n+1},L^{2}([1,2]))}\leq C_{\epsilon}2^{k\left((n+1)\left(\frac{1}{2}-\frac{1}{p}\right)-\frac{1}{p}+\frac{n}{2}+\epsilon\right)}\|f\|_{L^{p}(\mathbb{R}^{n+1})}. \tag*{(66)} ∥ M k , t f ∥ L p ( R n + 1 , L 2 ([ 1 , 2 ])) ≤ C ϵ 2 k ( ( n + 1 ) ( 2 1 − p 1 ) − p 1 + 2 n + ϵ ) ∥ f ∥ L p ( R n + 1 ) . (66)
It implies that the constant C D , ϵ C_{D,\epsilon} C D , ϵ is bounded by a constant C ϵ C_{\epsilon} C ϵ which is independent of D D D . Then we obtained (45) .
The proof of (3.46). Denote the multiplier associated to t d d t ( A k , t f ) \frac{t^d}{dt}(\mathcal{A}_{k,t}f) d t t d ( A k , t f ) by m ~ k ( t ξ ˉ ) \widetilde{m}_k(t\bar{\xi}) m k ( t ξ ˉ ) . Then
m ~ k ( ξ ˉ ) = ∑ i = 1 n + 1 ξ ˉ i ( ∂ ξ i m ) ( ξ ˉ ) ψ ( 2 − k ∣ ξ ˉ ∣ ) + 2 − k ∑ i = 1 n + 1 ξ ˉ i 2 ∣ ξ ˉ ∣ m ( ξ ˉ ) ( ∂ ξ i ψ ) ( 2 − k ∣ ξ ˉ ∣ ) . (67) \widetilde{m}_k(\bar{\xi})=\sum_{i=1}^{n+1}\bar{\xi}_i(\partial_{\xi_i}m)(\bar{\xi})\psi(2^{-k}|\bar{\xi}|)+2^{-k}\sum_{i=1}^{n+1}\frac{\bar{\xi}_i^2}{|\bar{\xi}|}m(\bar{\xi})(\partial_{\xi_i}\psi)(2^{-k}|\bar{\xi}|). \tag*{(67)} m k ( ξ ˉ ) = i = 1 ∑ n + 1 ξ ˉ i ( ∂ ξ i m ) ( ξ ˉ ) ψ ( 2 − k ∣ ξ ˉ ∣ ) + 2 − k i = 1 ∑ n + 1 ∣ ξ ˉ ∣ ξ ˉ i 2 m ( ξ ˉ ) ( ∂ ξ i ψ ) ( 2 − k ∣ ξ ˉ ∣ ) . (67)
Note that
( ∂ ξ i μ ) ( ξ ˉ ) = − i ∫ e − i ( x ⋅ ξ + ξ n + 1 ( D + Φ ( x ) ) ) x i ρ ( x ) d x , for i = 1 , … , n (68) (\partial_{\xi_i}\mu)(\bar{\xi})=-i\int e^{-i(x\cdot\xi+\xi_{n+1}(D+\Phi(x)))}x_i\rho(x)\,dx,\qquad\text{for }i=1,\ldots,n \tag*{(68)} ( ∂ ξ i μ ) ( ξ ˉ ) = − i ∫ e − i ( x ⋅ ξ + ξ n + 1 ( D + Φ ( x ))) x i ρ ( x ) d x , for i = 1 , … , n (68)
and
( ∂ ξ n + 1 μ ) ( ξ ˉ ) = − i ∫ e − i ( x ⋅ ξ + ξ n + 1 ( D + Φ ( x ) ) ) ( D + Φ ( x ) ) ρ ( x ) d x . (69) (\partial_{\xi_{n+1}}\mu)(\bar{\xi})=-i\int e^{-i(x\cdot\xi+\xi_{n+1}(D+\Phi(x)))}(D+\Phi(x))\rho(x)\,dx. \tag*{(69)} ( ∂ ξ n + 1 μ ) ( ξ ˉ ) = − i ∫ e − i ( x ⋅ ξ + ξ n + 1 ( D + Φ ( x ))) ( D + Φ ( x )) ρ ( x ) d x . (69)
Recall that m ( ξ ˉ ) = μ ( ξ ˉ ) − e − i D ξ n + 1 − ∣ ξ ˉ ∣ m(\bar{\xi})=\mu(\bar{\xi})-e^{-iD\xi_{n+1}-|\bar{\xi}|} m ( ξ ˉ ) = μ ( ξ ˉ ) − e − i D ξ n + 1 − ∣ ξ ˉ ∣ . By applying the method of stationary phase, it follows by (3.67), (3.68) and (3.69) that
m ~ k ( ξ ˉ ) = ( 1 + D ) 2 k e − i D ξ n + 1 ( e i ξ n + 1 Φ ∗ ∗ ( s ) A ~ m ( ξ ˉ ) ( 1 + ∣ ξ ˉ ∣ ) n / 2 χ ~ ( s ) ) ψ ( 2 − k ξ ˉ ) + e − i D ξ n + 1 A ~ r e ( ξ ˉ ) ψ ( 2 − k ξ ˉ ) , (70) \widetilde{m}_k(\bar{\xi})=(1+D)2^k e^{-iD\xi_{n+1}}\left(e^{i\xi_{n+1}\Phi^{**}(s)}\frac{\widetilde{A}_m(\bar{\xi})}{(1+|\bar{\xi}|)^{n/2}}\widetilde{\chi}(s)\right)\psi(2^{-k}\bar{\xi})+e^{-iD\xi_{n+1}}\widetilde{A}_{re}(\bar{\xi})\psi(2^{-k}\bar{\xi}), \tag*{(70)} m k ( ξ ˉ ) = ( 1 + D ) 2 k e − i D ξ n + 1 ( e i ξ n + 1 Φ ∗∗ ( s ) ( 1 + ∣ ξ ˉ ∣ ) n /2 A m ( ξ ˉ ) χ ( s ) ) ψ ( 2 − k ξ ˉ ) + e − i D ξ n + 1 A r e ( ξ ˉ ) ψ ( 2 − k ξ ˉ ) , (70)
where χ ~ ( s ) \widetilde{\chi}(s) χ ( s ) is a smooth function supported in [ 0 , c ~ ] [0,\widetilde{c}] [ 0 , c ] , A ~ m ( ξ ˉ ) \widetilde{A}_m(\bar{\xi}) A m ( ξ ˉ ) is a symbol of order of zero. Moreover, the remainder term A ~ r e ( ξ ˉ ) \widetilde{A}_{re}(\bar{\xi}) A r e ( ξ ˉ ) satisfies
∣ D ξ α A ~ r e ( ξ ˉ ) ∣ ≤ C N , α ( 1 + ∣ ξ ˉ ∣ ) − N . \left|D_{\xi}^{\alpha}\widetilde{A}_{re}(\bar{\xi})\right|\leq C_{N,\alpha}(1+|\bar{\xi}|)^{-N}. D ξ α A r e ( ξ ˉ ) ≤ C N , α ( 1 + ∣ ξ ˉ ∣ ) − N .
Using the same method as in proving (3.45), we can obtain (3.46). The proof is complete.
The proof of Lemma 3.4.
Denote δ ρ f ( x ˉ ) = f ( ρ x ˉ ) \delta_\rho f(\bar{x})=f(\rho\bar{x}) δ ρ f ( x ˉ ) = f ( ρ x ˉ ) . Since
A k , 2 j t f ( x ˉ ) = 1 ( 2 π ) 2 ∫ e i x ˉ ⋅ ξ ˉ m ( 2 j t ξ ˉ ) ψ ( 2 − k + j t ∣ ξ ˉ ∣ ) ψ ~ ( 2 − k + j ∣ ξ ˉ ∣ ) f ^ ( ξ ˉ ) d ξ ˉ = 1 ( 2 π ) 2 ∫ e i 2 − j x ˉ ⋅ ξ ˉ m ( t ξ ˉ ) ψ ( 2 − k t ∣ ξ ˉ ∣ ) F ( δ 2 j ( P ~ k − j f ) ) ( ξ ˉ ) d ξ ˉ = A k , t ( δ 2 j ( P ~ k − j f ) ) ( 2 − j ∣ x ˉ ∣ ) , \begin{aligned}
\mathcal{A}_{k,2^jt}f(\bar{x})=\frac{1}{(2\pi)^2}\int e^{i\bar{x}\cdot\bar{\xi}}m(2^jt\bar{\xi})\psi(2^{-k+j}t|\bar{\xi}|)\widetilde{\psi}(2^{-k+j}|\bar{\xi}|)\widehat{f}(\bar{\xi})\,d\bar{\xi} \\
&=\frac{1}{(2\pi)^2}\int e^{i2^{-j}\bar{x}\cdot\bar{\xi}}m(t\bar{\xi})\psi(2^{-k}t|\bar{\xi}|)\mathcal{F}\left(\delta_{2^j}(\widetilde{P}_{k-j}f)\right)(\bar{\xi})\,d\bar{\xi} \\
&=\mathcal{A}_{k,t}\left(\delta_{2^j}(\widetilde{P}_{k-j}f)\right)(2^{-j}|\bar{x}|),
\end{aligned} A k , 2 j t f ( x ˉ ) = ( 2 π ) 2 1 ∫ e i x ˉ ⋅ ξ ˉ m ( 2 j t ξ ˉ ) ψ ( 2 − k + j t ∣ ξ ˉ ∣ ) ψ ( 2 − k + j ∣ ξ ˉ ∣ ) f ( ξ ˉ ) d ξ ˉ = ( 2 π ) 2 1 ∫ e i 2 − j x ˉ ⋅ ξ ˉ m ( t ξ ˉ ) ψ ( 2 − k t ∣ ξ ˉ ∣ ) F ( δ 2 j ( P k − j f ) ) ( ξ ˉ ) d ξ ˉ = A k , t ( δ 2 j ( P k − j f ) ) ( 2 − j ∣ x ˉ ∣ ) ,
we have V 4 j ( A k f ) ( x ˉ ) = V 4 1 ( A k ( δ 2 j ( P ~ k − j f ) ) ) ( 2 − j x ˉ ) V_4^j(\mathcal{A}_kf)(\bar{x})=V_4^1\left(\mathcal{A}_k\left(\delta_{2^j}(\widetilde{P}_{k-j}f)\right)\right)(2^{-j}\bar{x}) V 4 j ( A k f ) ( x ˉ ) = V 4 1 ( A k ( δ 2 j ( P k − j f ) ) ) ( 2 − j x ˉ ) . Thus if we prove
∥ V 4 1 ( A k f ) ∥ L 4 ≤ C ϵ ( 1 + D 1 / 4 ) min { 2 − k ( 1 8 − ϵ ) , 1 } ∥ f ∥ L 4 , (71) \left\|V_4^1(\mathcal{A}_kf)\right\|_{L^4}\leq C_\epsilon(1+D^{1/4})\min\left\{2^{-k(\frac{1}{8}-\epsilon)},1\right\}\|f\|_{L^4}, \tag*{(71)} V 4 1 ( A k f ) L 4 ≤ C ϵ ( 1 + D 1/4 ) min { 2 − k ( 8 1 − ϵ ) , 1 } ∥ f ∥ L 4 , (71)
then we obtain
∥ V 4 j ( A k f ) ∥ L 4 ( ℓ j 4 ) = ∥ V 4 1 ( A k ( δ 2 j ( P ~ k − j f ) ) ) ( 2 − j ⋅ ) ∥ L 4 ( ℓ j 4 ) ≤ C ϵ ( 1 + D 1 / 4 ) min { 2 − k ( 1 8 − ϵ ) , 1 } ∥ P ~ k − j f ∥ L 4 ( ℓ j 4 ) ≤ C ( 1 + D 1 / 4 ) min { 2 − k ( 1 8 − ϵ ) , 1 } ∥ f ∥ L 4 . \begin{aligned}
\left\|V_4^j(\mathcal{A}_kf)\right\|_{L^4(\ell^4_j)}
=\left\|V_4^1\left(\mathcal{A}_k\left(\delta_{2^j}(\widetilde{P}_{k-j}f)\right)\right)(2^{-j}\cdot)\right\|_{L^4(\ell^4_j)} \\
&\leq C_\epsilon(1+D^{1/4})\min\left\{2^{-k(\frac{1}{8}-\epsilon)},1\right\}\left\|\widetilde{P}_{k-j}f\right\|_{L^4(\ell^4_j)} \\
&\leq C(1+D^{1/4})\min\left\{2^{-k(\frac{1}{8}-\epsilon)},1\right\}\|f\|_{L^4}.
\end{aligned} V 4 j ( A k f ) L 4 ( ℓ j 4 ) = V 4 1 ( A k ( δ 2 j ( P k − j f ) ) ) ( 2 − j ⋅ ) L 4 ( ℓ j 4 ) ≤ C ϵ ( 1 + D 1/4 ) min { 2 − k ( 8 1 − ϵ ) , 1 } P k − j f L 4 ( ℓ j 4 ) ≤ C ( 1 + D 1/4 ) min { 2 − k ( 8 1 − ϵ ) , 1 } ∥ f ∥ L 4 .
It remains for us to prove (71) .
Using inequality (5) , we obtain
V 4 j ( A k f ) ≤ C ∥ A k , t f ∥ L t 4 ( [ 1 , 2 ] ) 3 / 4 ∥ t d d t ( A k , t f ) ∥ L t 4 ( [ 1 , 2 ] ) 1 / 4 . V_4^j(A_kf) \le C\|A_{k,t}f\|_{L_t^4([1,2])}^{3/4}\left\|t\frac{d}{dt}(A_{k,t}f)\right\|_{L_t^4([1,2])}^{1/4}. V 4 j ( A k f ) ≤ C ∥ A k , t f ∥ L t 4 ([ 1 , 2 ]) 3/4 t d t d ( A k , t f ) L t 4 ([ 1 , 2 ]) 1/4 .
By Hölder’s inequality, we obtain
∥ V 4 j ( A k f ) ∥ L 4 ≤ C ∥ A k , t f ∥ L 4 ( L t 4 ( [ 1 , 2 ] ) ) 3 / 4 ∥ t d d t ( A k , t f ) ∥ L 4 ( L t 4 ( [ 1 , 2 ] ) ) 1 / 4 . \|V_4^j(A_kf)\|_{L^4} \le C\|A_{k,t}f\|_{L^4(L_t^4([1,2]))}^{3/4}\left\|t\frac{d}{dt}(A_{k,t}f)\right\|_{L^4(L_t^4([1,2]))}^{1/4}. ∥ V 4 j ( A k f ) ∥ L 4 ≤ C ∥ A k , t f ∥ L 4 ( L t 4 ([ 1 , 2 ])) 3/4 t d t d ( A k , t f ) L 4 ( L t 4 ([ 1 , 2 ])) 1/4 .
It is enough for us to prove
∥ A k , t f ∥ L 4 ( L t 4 ( [ 1 , 2 ] ) ) ≤ C ε min { 2 − 3 8 k + k ε , 1 } ∥ f ∥ L 4 (72) \|A_{k,t}f\|_{L^4(L_t^4([1,2]))} \le C_\varepsilon\min\left\{2^{-\frac{3}{8}k+k\varepsilon},1\right\}\|f\|_{L^4} \tag*{(72)} ∥ A k , t f ∥ L 4 ( L t 4 ([ 1 , 2 ])) ≤ C ε min { 2 − 8 3 k + k ε , 1 } ∥ f ∥ L 4 (72)
and
∥ t d d t A k , t ∗ f ∥ L 4 ( L t 4 ( [ 1 , 2 ] ) ) ≤ C ε ( 1 + D ) { 2 5 8 k + k ε , 1 } ∥ f ∥ L 4 . (73) \left\|t\frac{d}{dt}A_{k,t}*f\right\|_{L^4(L_t^4([1,2]))} \le C_\varepsilon(1+D)\left\{2^{\frac{5}{8}k+k\varepsilon},1\right\}\|f\|_{L^4}. \tag*{(73)} t d t d A k , t ∗ f L 4 ( L t 4 ([ 1 , 2 ])) ≤ C ε ( 1 + D ) { 2 8 5 k + k ε , 1 } ∥ f ∥ L 4 . (73)
Recall that the multipliers corresponding to A k , t A_{k,t} A k , t and t d d t A k , t t\frac{d}{dt}A_{k,t} t d t d A k , t can be written as
m k ( ξ ˉ ) = e − i D ξ n + 1 ( e i ξ n + 1 Φ ∗ ∗ ( s ) A m ( ξ ˉ ) ( 1 + ∣ ξ ˉ ∣ ) n / 2 χ ( s ) ) ψ ( 2 − k ξ ˉ ) + e − i D ξ n + 1 A r e ( ξ ˉ ) ψ ( 2 − k ξ ˉ ) , m_k(\bar{\xi})=e^{-iD\xi_{n+1}}\left(e^{i\xi_{n+1}\Phi^{**}(s)}\frac{A_m(\bar{\xi})}{(1+|\bar{\xi}|)^{n/2}}\chi(s)\right)\psi(2^{-k}\bar{\xi})+e^{-iD\xi_{n+1}}A_{re}(\bar{\xi})\psi(2^{-k}\bar{\xi}), m k ( ξ ˉ ) = e − i D ξ n + 1 ( e i ξ n + 1 Φ ∗∗ ( s ) ( 1 + ∣ ξ ˉ ∣ ) n /2 A m ( ξ ˉ ) χ ( s ) ) ψ ( 2 − k ξ ˉ ) + e − i D ξ n + 1 A r e ( ξ ˉ ) ψ ( 2 − k ξ ˉ ) ,
and
m ~ k ( ξ ˉ ) = ∑ i = 1 n + 1 ξ i ( ∂ ξ i m ) ( ξ ˉ ) ψ ( 2 − k ∣ ξ ˉ ∣ ) + 2 − k ∑ i = 1 n + 1 ξ i 2 ∣ ξ ˉ ∣ m ( ξ ˉ ) ( ∂ ξ i ψ ) ( 2 − k ∣ ξ ˉ ∣ ) . \widetilde{m}_k(\bar{\xi})=\sum_{i=1}^{n+1}\xi_i(\partial_{\xi_i}m)(\bar{\xi})\psi(2^{-k}|\bar{\xi}|)+2^{-k}\sum_{i=1}^{n+1}\frac{\xi_i^2}{|\bar{\xi}|}m(\bar{\xi})(\partial_{\xi_i}\psi)(2^{-k}|\bar{\xi}|). m k ( ξ ˉ ) = i = 1 ∑ n + 1 ξ i ( ∂ ξ i m ) ( ξ ˉ ) ψ ( 2 − k ∣ ξ ˉ ∣ ) + 2 − k i = 1 ∑ n + 1 ∣ ξ ˉ ∣ ξ i 2 m ( ξ ˉ ) ( ∂ ξ i ψ ) ( 2 − k ∣ ξ ˉ ∣ ) .
respectively. Using the same method as in the proof of Lemma 3.3, we can get (72) and (73) by the following lemma, introduced in the proof of Theorem 6.1 in [24 ] .
Lemma 3.7 Denote Q λ f ( x ˉ , t ) Q_\lambda f(\bar{x},t) Q λ f ( x ˉ , t ) as in Lemma 3.6. Then we have
∥ Q λ f ∥ L 4 ( R 3 ) ≤ C ε λ 1 8 + ε ∥ f ∥ L 4 ( R 2 ) . (74) \|Q_\lambda f\|_{L^4(\mathbb{R}^3)}\le C_\varepsilon\lambda^{\frac{1}{8}+\varepsilon}\|f\|_{L^4(\mathbb{R}^2)}. \tag*{(74)} ∥ Q λ f ∥ L 4 ( R 3 ) ≤ C ε λ 8 1 + ε ∥ f ∥ L 4 ( R 2 ) . (74)
The proof of the necessary condition
In this subsection, we establish the necessity of the condition p < r ( n + 1 ) p<r(n+1) p < r ( n + 1 ) in Theorem 1.1 for the case k = n k=n k = n . The argument follows a standard Knapp-type counterexample construction, adapted to the variational setting.
Preliminary work. Recall that the Fourier transform of the surface measure is given by
μ ( ξ , ξ n + 1 ) = e − i D ξ n + 1 ( e i ξ n + 1 Φ ∗ ∗ ( s ) A m ( ξ ˉ ) ( 1 + ∣ ξ ˉ ∣ ) n / 2 χ ( s ) + A r e ( ξ ˉ ) ) . \mu(\xi,\xi_{n+1})=e^{-iD\xi_{n+1}}\left(e^{i\xi_{n+1}\Phi^{**}(s)}\frac{A_m(\bar{\xi})}{(1+|\bar{\xi}|)^{n/2}}\chi(s)+A_{re}(\bar{\xi})\right). μ ( ξ , ξ n + 1 ) = e − i D ξ n + 1 ( e i ξ n + 1 Φ ∗∗ ( s ) ( 1 + ∣ ξ ˉ ∣ ) n /2 A m ( ξ ˉ ) χ ( s ) + A r e ( ξ ˉ ) ) .
We split the operator A t A_t A t into a main part and a remainder part,
A t f = M t f + E t f , A_tf=M_tf+E_tf, A t f = M t f + E t f ,
where
M a t f ( x ˉ ) = ∫ R n + 1 e − i t D ξ n + 1 e i t ξ n + 1 Φ ∗ ∗ ( s ) A m ( ξ ˉ ) ( 1 + ∣ ξ ˉ ∣ ) n 2 χ ( s ) f ^ ( ξ , ξ n + 1 ) e i ( x ⋅ ξ + x n + 1 ξ n + 1 ) d ξ ˉ . M_{a_t}f(\bar{x})=\int_{\mathbb{R}^{n+1}}e^{-itD\xi_{n+1}}e^{it\xi_{n+1}\Phi^{**}(s)}\frac{A_m(\bar{\xi})}{(1+|\bar{\xi}|)^{\frac n2}}\chi(s)\hat{f}(\xi,\xi_{n+1})e^{i(x\cdot\xi+x_{n+1}\xi_{n+1})}\,d\bar{\xi}. M a t f ( x ˉ ) = ∫ R n + 1 e − i t D ξ n + 1 e i t ξ n + 1 Φ ∗∗ ( s ) ( 1 + ∣ ξ ˉ ∣ ) 2 n A m ( ξ ˉ ) χ ( s ) f ^ ( ξ , ξ n + 1 ) e i ( x ⋅ ξ + x n + 1 ξ n + 1 ) d ξ ˉ .
and
E r t f ( x ˉ ) = ∫ R n + 1 e − i t D ξ n + 1 A r e ( ξ ˉ ) f ^ ( ξ , ξ n + 1 ) e i ( x ⋅ ξ + x n + 1 ξ n + 1 ) d ξ ˉ . E_{r_t}f(\bar{x})=\int_{\mathbb{R}^{n+1}}e^{-itD\xi_{n+1}}A_{re}(\bar{\xi})\hat{f}(\xi,\xi_{n+1})e^{i(x\cdot\xi+x_{n+1}\xi_{n+1})}\,d\bar{\xi}. E r t f ( x ˉ ) = ∫ R n + 1 e − i t D ξ n + 1 A r e ( ξ ˉ ) f ^ ( ξ , ξ n + 1 ) e i ( x ⋅ ξ + x n + 1 ξ n + 1 ) d ξ ˉ .
Consequently,
∥ V r ( A t f : t > 0 ) ∥ L p ≥ ∥ V r ( M a t f : t > 0 ) ∥ L p − ∥ V r ( E r t f : t > 0 ) ∥ L p . (75) \|V_r(A_tf:t>0)\|_{L^p}\geq\|V_r(M_{a_t}f:t>0)\|_{L^p}-\|V_r(E_{r_t}f:t>0)\|_{L^p}. \tag*{(75)} ∥ V r ( A t f : t > 0 ) ∥ L p ≥ ∥ V r ( M a t f : t > 0 ) ∥ L p − ∥ V r ( E r t f : t > 0 ) ∥ L p . (75)
Thus it suffices to obtain a lower bound for the main term and an upper bound for the remainder term, which will be negligible for large λ \lambda λ .
To proceed, we perform a change of variables that linearises the phase in t t t . Set
ξ i = − s i s n + 1 − D + Φ ∗ ∗ ( s ) ( i = 1 , … , n ) , ξ n + 1 = s n + 1 − D + Φ ∗ ∗ ( s ) . \xi_i=-\frac{s_i s_{n+1}}{-D+\Phi^{**}(s)}\quad(i=1,\ldots,n),\qquad\xi_{n+1}=\frac{s_{n+1}}{-D+\Phi^{**}(s)}. ξ i = − − D + Φ ∗∗ ( s ) s i s n + 1 ( i = 1 , … , n ) , ξ n + 1 = − D + Φ ∗∗ ( s ) s n + 1 .
Then the Jacobian of this transformation is ∣ s n + 1 ∣ n Ψ ∗ ( s ) |s_{n+1}|^n\Psi^*(s) ∣ s n + 1 ∣ n Ψ ∗ ( s ) for some smooth function Ψ ∗ \Psi^* Ψ ∗ arising from the change of variables. After this substitution, the main term and remainder term become
M a t f ( x ˉ ) = ∫ R n + 1 e i t s n + 1 A m ( − s n + 1 ⋅ s − D + Φ ∗ ∗ ( s ) , s n + 1 − D + Φ ∗ ∗ ( s ) ) ( 1 + ∣ ( − s n + 1 ⋅ s − D + Φ ∗ ∗ ( s ) , s n + 1 − D + Φ ∗ ∗ ( s ) ) ∣ ) n 2 χ ( s ) M_{a_t}f(\bar{x})=\int_{\mathbb{R}^{n+1}}e^{it s_{n+1}}\frac{A_m\left(-\frac{s_{n+1}\cdot s}{-D+\Phi^{**}(s)},\frac{s_{n+1}}{-D+\Phi^{**}(s)}\right)}{\left(1+\left|\left(-\frac{s_{n+1}\cdot s}{-D+\Phi^{**}(s)},\frac{s_{n+1}}{-D+\Phi^{**}(s)}\right)\right|\right)^{\frac n2}}\chi(s) M a t f ( x ˉ ) = ∫ R n + 1 e i t s n + 1 ( 1 + ( − − D + Φ ∗∗ ( s ) s n + 1 ⋅ s , − D + Φ ∗∗ ( s ) s n + 1 ) ) 2 n A m ( − − D + Φ ∗∗ ( s ) s n + 1 ⋅ s , − D + Φ ∗∗ ( s ) s n + 1 ) χ ( s )
× e i s n + 1 − s ⋅ x + x n + 1 − D + Φ ∗ ∗ ( s ) f ^ ( − s n + 1 ⋅ s − D + Φ ∗ ∗ ( s ) , s n + 1 − D + Φ ∗ ∗ ( s ) ) ∣ s n + 1 ∣ n Ψ ∗ ( s ) d s , {}\times e^{i s_{n+1}\frac{-s\cdot x+x_{n+1}}{-D+\Phi^{**}(s)}}\hat{f}\left(-\frac{s_{n+1}\cdot s}{-D+\Phi^{**}(s)},\frac{s_{n+1}}{-D+\Phi^{**}(s)}\right)|s_{n+1}|^n\Psi^*(s)\,ds, × e i s n + 1 − D + Φ ∗∗ ( s ) − s ⋅ x + x n + 1 f ^ ( − − D + Φ ∗∗ ( s ) s n + 1 ⋅ s , − D + Φ ∗∗ ( s ) s n + 1 ) ∣ s n + 1 ∣ n Ψ ∗ ( s ) d s ,
and similarly
E r t f ( x ˉ ) = ∫ R n + 1 e i t s n + 1 A r e ( − s n + 1 ⋅ s − D + Φ ∗ ∗ ( s ) , s n + 1 − D + Φ ∗ ∗ ( s ) ) E_{r_t}f(\bar{x})=\int_{\mathbb{R}^{n+1}}e^{it s_{n+1}}A_{re}\left(-\frac{s_{n+1}\cdot s}{-D+\Phi^{**}(s)},\frac{s_{n+1}}{-D+\Phi^{**}(s)}\right) E r t f ( x ˉ ) = ∫ R n + 1 e i t s n + 1 A r e ( − − D + Φ ∗∗ ( s ) s n + 1 ⋅ s , − D + Φ ∗∗ ( s ) s n + 1 )
× e i s n + 1 − s ⋅ x + x n + 1 − D + Φ ∗ ∗ ( s ) f ^ ( − s n + 1 ⋅ s − D + Φ ∗ ∗ ( s ) , s n + 1 − D + Φ ∗ ∗ ( s ) ) ∣ s n + 1 ∣ n Ψ ∗ ( s ) d s . {}\times e^{i s_{n+1}\frac{-s\cdot x+x_{n+1}}{-D+\Phi^{**}(s)}}\hat{f}\left(-\frac{s_{n+1}\cdot s}{-D+\Phi^{**}(s)},\frac{s_{n+1}}{-D+\Phi^{**}(s)}\right)|s_{n+1}|^n\Psi^*(s)\,ds. × e i s n + 1 − D + Φ ∗∗ ( s ) − s ⋅ x + x n + 1 f ^ ( − − D + Φ ∗∗ ( s ) s n + 1 ⋅ s , − D + Φ ∗∗ ( s ) s n + 1 ) ∣ s n + 1 ∣ n Ψ ∗ ( s ) d s .
Now we choose a family of test functions f λ f_\lambda f λ concentrated in frequency near a point where the phase is stationary. Define f ^ λ \hat{f}_\lambda f ^ λ by
f ^ λ ( − s n + 1 ⋅ s − D + Φ ∗ ∗ ( s ) , s n + 1 − D + Φ ∗ ∗ ( s ) ) = χ 0 ( s ) χ 1 ( s n + 1 λ ) e − i s n + 1 2 2 λ . (76) \hat{f}_\lambda\left(-\frac{s_{n+1}\cdot s}{-D+\Phi^{**}(s)},\frac{s_{n+1}}{-D+\Phi^{**}(s)}\right)=\chi_0(s)\chi_1\left(\frac{s_{n+1}}{\lambda}\right)e^{-i\frac{s_{n+1}^2}{2\lambda}}. \tag*{(76)} f ^ λ ( − − D + Φ ∗∗ ( s ) s n + 1 ⋅ s , − D + Φ ∗∗ ( s ) s n + 1 ) = χ 0 ( s ) χ 1 ( λ s n + 1 ) e − i 2 λ s n + 1 2 . (76)
Here χ 0 \chi_0 χ 0 is a smooth cut-off supported in a small neighbourhood of s = 0 s=0 s = 0 , and χ 1 \chi_1 χ 1 is a smooth function supported in [ b / 2 , 2 b ] [b/2,2b] [ b /2 , 2 b ] for a small positive constant c c c . This choice makes f λ f_\lambda f λ essentially a wave packet localized in both space and frequency; the quadratic exponential factor will later create oscillations that produce a large variation.
We first dispose of the remainder term. Substituting the definition of f ^ λ \hat{f}_\lambda f ^ λ into E r t f λ E_{r_t}f_\lambda E r t f λ gives
E r t f λ ( x ˉ ) = λ n + 1 ∫ R n + 1 e i t λ s n + 1 A r e ( − λ s n + 1 ⋅ s − D + Φ ∗ ∗ ( s ) , λ s n + 1 − D + Φ ∗ ∗ ( s ) ) E_{r_t}f_\lambda(\bar{x})=\lambda^{n+1}\int_{\mathbb{R}^{n+1}}e^{it\lambda s_{n+1}}A_{re}\left(-\frac{\lambda s_{n+1}\cdot s}{-D+\Phi^{**}(s)},\frac{\lambda s_{n+1}}{-D+\Phi^{**}(s)}\right) E r t f λ ( x ˉ ) = λ n + 1 ∫ R n + 1 e i t λ s n + 1 A r e ( − − D + Φ ∗∗ ( s ) λ s n + 1 ⋅ s , − D + Φ ∗∗ ( s ) λ s n + 1 )
× e i λ s n + 1 − s ⋅ x + x n + 1 − D + Φ ∗ ∗ ( s ) χ 0 ( s ) χ 1 ( s n + 1 ) e − i λ s n + 1 2 2 ∣ s n + 1 ∣ n Ψ ∗ ( s ) d s . {}\times e^{i\lambda s_{n+1}\frac{-s\cdot x+x_{n+1}}{-D+\Phi^{**}(s)}}\chi_0(s)\chi_1(s_{n+1})e^{-i\frac{\lambda s_{n+1}^2}{2}}|s_{n+1}|^n\Psi^*(s)\,ds. × e iλ s n + 1 − D + Φ ∗∗ ( s ) − s ⋅ x + x n + 1 χ 0 ( s ) χ 1 ( s n + 1 ) e − i 2 λ s n + 1 2 ∣ s n + 1 ∣ n Ψ ∗ ( s ) d s .
Because A r e A_{re} A r e satisfies the rapid decay estimate
∣ D ξ α A r e ( ξ ˉ ) ∣ ≤ C N , α ( 1 + ∣ ξ ˉ ∣ ) − N , (77) \left|D_{\xi}^{\alpha}A_{re}(\bar{\xi})\right| \leq C_{N,\alpha}(1+|\bar{\xi}|)^{-N}, \tag*{(77)} D ξ α A r e ( ξ ˉ ) ≤ C N , α ( 1 + ∣ ξ ˉ ∣ ) − N , (77)
a routine integration by parts argument (using the factor e i t λ s n + 1 e^{it\lambda s_{n+1}} e i t λ s n + 1 ) shows that E r t f λ Er_t f_{\lambda} E r t f λ is negligible. More precisely, for any r > 2 r>2 r > 2 and p > 1 p>1 p > 1 ,
∥ V r ( E r t f λ : t > 0 ) ∥ L p ≲ N λ − N ∥ f λ ∥ L p . (78) \left\|V_r(Er_t f_{\lambda}:t>0)\right\|_{L^p}\lesssim_N \lambda^{-N}\left\|f_{\lambda}\right\|_{L^p}. \tag*{(78)} ∥ V r ( E r t f λ : t > 0 ) ∥ L p ≲ N λ − N ∥ f λ ∥ L p . (78)
Now we turn to the main term and estimate ∥ f λ ∥ L p \left\|f_{\lambda}\right\|_{L^p} ∥ f λ ∥ L p and the variational norm of M λ f λ M_{\lambda}f_{\lambda} M λ f λ .
Proof of (83) : upper bound for ∥ f λ ∥ L p \left\|f_{\lambda}\right\|_{L^p} ∥ f λ ∥ L p . By Fourier inversion,
f λ ( x ˉ ) = 1 ( 2 π ) n + 1 ∫ e i ξ ˉ ⋅ x ˉ f ^ λ ( ξ ˉ ) d ξ ˉ . f_{\lambda}(\bar{x})=\frac{1}{(2\pi)^{n+1}}\int e^{i\bar{\xi}\cdot\bar{x}}\widehat{f}_{\lambda}(\bar{\xi})\,d\bar{\xi}. f λ ( x ˉ ) = ( 2 π ) n + 1 1 ∫ e i ξ ˉ ⋅ x ˉ f λ ( ξ ˉ ) d ξ ˉ .
Changing variables as before,
f λ ( x ˉ ) = 1 ( 2 π ) n + 1 ∫ e i s n + 1 − s ⋅ x ˙ + x n + 1 − D + Φ ∗ ∗ ( s ) f ^ λ ( s n + 1 ⋅ s − D + Φ ∗ ∗ ( s ) , s n + 1 − D + Φ ∗ ∗ ( s ) ) ∣ s n + 1 ∣ n Ψ ∗ ( s ) d s ˉ . f_{\lambda}(\bar{x})=\frac{1}{(2\pi)^{n+1}}\int e^{is_{n+1}\frac{-s\cdot\dot{x}+x_{n+1}}{-D+\Phi^{**}(s)}}\widehat{f}_{\lambda}\left(\frac{s_{n+1}\cdot s}{-D+\Phi^{**}(s)},\frac{s_{n+1}}{-D+\Phi^{**}(s)}\right)|s_{n+1}|^n\Psi^*(s)\,d\bar{s}. f λ ( x ˉ ) = ( 2 π ) n + 1 1 ∫ e i s n + 1 − D + Φ ∗∗ ( s ) − s ⋅ x ˙ + x n + 1 f λ ( − D + Φ ∗∗ ( s ) s n + 1 ⋅ s , − D + Φ ∗∗ ( s ) s n + 1 ) ∣ s n + 1 ∣ n Ψ ∗ ( s ) d s ˉ .
Inserting the definition (76) yields
f λ ( x ˉ ) = λ n + 1 ( 2 π ) n + 1 ∫ e i λ ( s n + 1 ( − s ⋅ x ˙ + x n + 1 ) − D + Φ ∗ ∗ ( s ) − s n + 1 2 2 ) χ 0 ( s ) χ 1 ( s n + 1 ) ∣ s n + 1 ∣ n Ψ ∗ ( s ) d s ˉ = λ n + 1 ( 2 π ) n + 1 ∫ e i λ ( − s n + 1 2 2 ) J ( λ x n + 1 s n + 1 , x ˙ ) χ 1 ( s n + 1 ) ∣ s n + 1 ∣ n d s n + 1 , (79) \begin{aligned}
f_{\lambda}(\bar{x})&=\frac{\lambda^{n+1}}{(2\pi)^{n+1}}\int e^{i\lambda\left(\frac{s_{n+1}\left(-s\cdot\dot{x}+x_{n+1}\right)}{-D+\Phi^{**}(s)}-\frac{s_{n+1}^2}{2}\right)}\chi_0(s)\chi_1(s_{n+1})|s_{n+1}|^n\Psi^*(s)\,d\bar{s}\\
&=\frac{\lambda^{n+1}}{(2\pi)^{n+1}}\int e^{i\lambda\left(-\frac{s_{n+1}^2}{2}\right)}J(\lambda x_{n+1}s_{n+1},\dot{x})\chi_1(s_{n+1})|s_{n+1}|^n\,ds_{n+1}, \tag*{(79)}
\end{aligned} f λ ( x ˉ ) = ( 2 π ) n + 1 λ n + 1 ∫ e iλ ( − D + Φ ∗∗ ( s ) s n + 1 ( − s ⋅ x ˙ + x n + 1 ) − 2 s n + 1 2 ) χ 0 ( s ) χ 1 ( s n + 1 ) ∣ s n + 1 ∣ n Ψ ∗ ( s ) d s ˉ = ( 2 π ) n + 1 λ n + 1 ∫ e iλ ( − 2 s n + 1 2 ) J ( λ x n + 1 s n + 1 , x ˙ ) χ 1 ( s n + 1 ) ∣ s n + 1 ∣ n d s n + 1 , (79)
where x ˙ = ( x 1 x n + 1 , … , x n x n + 1 ) \dot{x}=\left(\frac{x_1}{x_{n+1}},\ldots,\frac{x_n}{x_{n+1}}\right) x ˙ = ( x n + 1 x 1 , … , x n + 1 x n ) and
J ( μ , x ˙ ) = ∫ e i μ − s ⋅ x ˙ + 1 − D + Φ ∗ ∗ ( s ) χ 0 ( s ) Ψ ∗ ( s ) d s . (80) J(\mu,\dot{x})=\int e^{i\mu\frac{-s\cdot\dot{x}+1}{-D+\Phi^{**}(s)}}\chi_0(s)\Psi^*(s)\,ds. \tag*{(80)} J ( μ , x ˙ ) = ∫ e i μ − D + Φ ∗∗ ( s ) − s ⋅ x ˙ + 1 χ 0 ( s ) Ψ ∗ ( s ) d s . (80)
The phase function Λ ( x ˙ , s ) = − s ⋅ x ˙ + 1 − D + Φ ∗ ∗ ( s ) \Lambda(\dot{x},s)=\frac{-s\cdot\dot{x}+1}{-D+\Phi^{**}(s)} Λ ( x ˙ , s ) = − D + Φ ∗∗ ( s ) − s ⋅ x ˙ + 1 satisfies det ∂ s i s j 2 Λ ( x ˙ , 0 ) ≠ 0 \det\partial_{s_i s_j}^2\Lambda(\dot{x},0)\ne0 det ∂ s i s j 2 Λ ( x ˙ , 0 ) = 0 ; hence by the implicit function theorem there exists a unique smooth function s ∗ = s ∗ ( x ˙ ) s^*=s^*(\dot{x}) s ∗ = s ∗ ( x ˙ ) near 0 0 0 such that ∇ s Λ ( x ˙ , s ∗ ) = 0 \nabla_s\Lambda(\dot{x},s^*)=0 ∇ s Λ ( x ˙ , s ∗ ) = 0 for x ˙ \dot{x} x ˙ small. Applying the method of stationary phase to the integral J ( μ , x ˙ ) J(\mu,\dot{x}) J ( μ , x ˙ ) gives
J ( μ , x ˙ ) = e i μ Λ ( x ˙ , s ∗ ) H ( μ , x ˙ ) , (81) J(\mu,\dot{x})=e^{i\mu\Lambda(\dot{x},s^*)}H(\mu,\dot{x}), \tag*{(81)} J ( μ , x ˙ ) = e i μ Λ ( x ˙ , s ∗ ) H ( μ , x ˙ ) , (81)
where H H H is a symbol of order − n 2 -\frac{n}{2} − 2 n in μ \mu μ ; more precisely,
∣ ( ∂ ∂ μ ) α ( ∂ ∂ x ˙ ) β H ( μ , x ˙ ) ∣ ≤ C α , β ( 1 + ∣ μ ∣ ) − n 2 − ∣ α ∣ . \left|\left(\frac{\partial}{\partial\mu}\right)^{\alpha}\left(\frac{\partial}{\partial\dot{x}}\right)^{\beta}H(\mu,\dot{x})\right|\leq C_{\alpha,\beta}(1+|\mu|)^{-\frac{n}{2}-|\alpha|}. ( ∂ μ ∂ ) α ( ∂ x ˙ ∂ ) β H ( μ , x ˙ ) ≤ C α , β ( 1 + ∣ μ ∣ ) − 2 n − ∣ α ∣ .
Substituting this back, we obtain
f λ ( x ˉ ) = λ n + 1 ( 2 π ) n + 1 ∫ e i λ ( − s n + 1 2 2 + s n + 1 x n + 1 Λ ( x ˙ , s ∗ ) ) H ( λ x n + 1 s n + 1 , x ˙ ) χ 1 ( s n + 1 ) ∣ s n + 1 ∣ n d s n + 1 . (82) f_{\lambda}(\bar{x})=\frac{\lambda^{n+1}}{(2\pi)^{n+1}}\int e^{i\lambda\left(-\frac{s_{n+1}^2}{2}+s_{n+1}x_{n+1}\Lambda(\dot{x},s^*)\right)}H(\lambda x_{n+1}s_{n+1},\dot{x})\chi_1(s_{n+1})|s_{n+1}|^n\,ds_{n+1}. \tag*{(82)} f λ ( x ˉ ) = ( 2 π ) n + 1 λ n + 1 ∫ e iλ ( − 2 s n + 1 2 + s n + 1 x n + 1 Λ ( x ˙ , s ∗ ) ) H ( λ x n + 1 s n + 1 , x ˙ ) χ 1 ( s n + 1 ) ∣ s n + 1 ∣ n d s n + 1 . (82)
Because s ∗ ( 0 ) = 0 s^*(0)=0 s ∗ ( 0 ) = 0 and χ 0 \chi_0 χ 0 is supported near 0 0 0 , there exists a constant c > 0 c>0 c > 0 such that f λ ( x ˉ ) f_{\lambda}(\bar{x}) f λ ( x ˉ ) decays rapidly when ∣ x i x n + 1 ∣ > c \left|\frac{x_i}{x_{n+1}}\right|>c x n + 1 x i > c or ∣ x n + 1 ∣ > c |x_{n+1}|>c ∣ x n + 1 ∣ > c . For ∣ x i x n + 1 ∣ ≲ 1 \left|\frac{x_i}{x_{n+1}}\right|\lesssim1 x n + 1 x i ≲ 1 and ∣ x n + 1 ∣ ≲ 1 |x_{n+1}|\lesssim1 ∣ x n + 1 ∣ ≲ 1 , a final application of the stationary phase theorem (in the s n + 1 s_{n+1} s n + 1 integral) shows that ∣ f λ ( x ˉ ) ∣ ≤ C λ n + 1 2 |f_{\lambda}(\bar{x})|\leq C\lambda^{\frac{n+1}{2}} ∣ f λ ( x ˉ ) ∣ ≤ C λ 2 n + 1 . Consequently,
∥ f λ ∥ L p ≲ λ n + 1 2 . (83) \left\|f_{\lambda}\right\|_{L^p}\lesssim\lambda^{\frac{n+1}{2}}. \tag*{(83)} ∥ f λ ∥ L p ≲ λ 2 n + 1 . (83)
Proof of (3.87): lower bound for ∥ V r ( M a t f λ ) ∥ L p \|V_r(M_{at}f_\lambda)\|_{L^p} ∥ V r ( M a t f λ ) ∥ L p . We now substitute f λ f_\lambda f λ into the expression for M a t f M_{at}f M a t f . After the same change of variables, we obtain
M a t f λ ( x ˉ ) = ∫ R n + 1 e i t s n + 1 e i s n + 1 − s ⋅ x + x n + 1 − D + Φ ∗ ∗ ( s ) A m ( − s n + 1 ⋅ s − D + Φ ∗ ∗ ( s ) , s n + 1 − D + Φ ∗ ∗ ( s ) ) ( 1 + ∣ ( − s n + 1 ⋅ s − D + Φ ∗ ∗ ( s ) , s n + 1 − D + Φ ∗ ∗ ( s ) ) ∣ 2 ) n 2 χ ( s ) M_{at}f_\lambda(\bar{x})=\int_{\mathbb{R}^{n+1}}e^{its_{n+1}}e^{i s_{n+1}\frac{-s\cdot x+x_{n+1}}{-D+\Phi^{**}(s)}}\frac{A_m\left(-\frac{s_{n+1}\cdot s}{-D+\Phi^{**}(s)},\frac{s_{n+1}}{-D+\Phi^{**}(s)}\right)}{\left(1+\left|\left(-\frac{s_{n+1}\cdot s}{-D+\Phi^{**}(s)},\frac{s_{n+1}}{-D+\Phi^{**}(s)}\right)\right|^2\right)^{\frac n2}}\chi(s) M a t f λ ( x ˉ ) = ∫ R n + 1 e i t s n + 1 e i s n + 1 − D + Φ ∗∗ ( s ) − s ⋅ x + x n + 1 ( 1 + ( − − D + Φ ∗∗ ( s ) s n + 1 ⋅ s , − D + Φ ∗∗ ( s ) s n + 1 ) 2 ) 2 n A m ( − − D + Φ ∗∗ ( s ) s n + 1 ⋅ s , − D + Φ ∗∗ ( s ) s n + 1 ) χ ( s )
× χ 0 ( s ) χ 1 ( s n + 1 λ ) e − i s n + 1 2 2 λ ∣ s n + 1 ∣ n Ψ ∗ ( s ) d s = λ n + 1 ∫ R e − i λ ( s n + 1 2 2 − t s n + 1 ) χ 1 ( s n + 1 ) ∣ s n + 1 ∣ n G ( λ x n + 1 , t , s n + 1 , x ˙ ) d s n + 1 , (84) \begin{aligned}
\qquad\times\chi_0(s)\chi_1\left(\frac{s_{n+1}}{\lambda}\right)e^{-i\frac{s_{n+1}^2}{2\lambda}}|s_{n+1}|^n\Psi^*(s)\,ds \\
&=\lambda^{n+1}\int_{\mathbb{R}}e^{-i\lambda\left(\frac{s_{n+1}^2}{2}-ts_{n+1}\right)}\chi_1(s_{n+1})|s_{n+1}|^nG(\lambda x_{n+1},t,s_{n+1},\dot{x})\,ds_{n+1}, \tag*{(84)}
\end{aligned} × χ 0 ( s ) χ 1 ( λ s n + 1 ) e − i 2 λ s n + 1 2 ∣ s n + 1 ∣ n Ψ ∗ ( s ) d s = λ n + 1 ∫ R e − iλ ( 2 s n + 1 2 − t s n + 1 ) χ 1 ( s n + 1 ) ∣ s n + 1 ∣ n G ( λ x n + 1 , t , s n + 1 , x ˙ ) d s n + 1 , (84)
where
G ( λ x n + 1 , t , s n + 1 , x ˙ ) = ∫ R n e i t λ x n + 1 s n + 1 e − s ⋅ x ˙ + 1 − D + Φ ∗ ∗ ( s ) A m ( − λ s n + 1 ⋅ s − D + Φ ∗ ∗ ( s ) , λ s n + 1 − D + Φ ∗ ∗ ( s ) ) ( 1 + ∣ ( − λ s n + 1 ⋅ s − D + Φ ∗ ∗ ( s ) , λ s n + 1 − D + Φ ∗ ∗ ( s ) ) ∣ 2 ) n 2 χ ( s ) χ 0 ( s ) Ψ ∗ ( s ) d s . (85) G(\lambda x_{n+1},t,s_{n+1},\dot{x})=\int_{\mathbb{R}^n}e^{it\lambda x_{n+1}s_{n+1}}e^{\frac{-s\cdot\dot{x}+1}{-D+\Phi^{**}(s)}}\frac{A_m\left(-\frac{\lambda s_{n+1}\cdot s}{-D+\Phi^{**}(s)},\frac{\lambda s_{n+1}}{-D+\Phi^{**}(s)}\right)}{\left(1+\left|\left(-\frac{\lambda s_{n+1}\cdot s}{-D+\Phi^{**}(s)},\frac{\lambda s_{n+1}}{-D+\Phi^{**}(s)}\right)\right|^2\right)^{\frac n2}}\chi(s)\chi_0(s)\Psi^*(s)\,ds. \tag*{(85)} G ( λ x n + 1 , t , s n + 1 , x ˙ ) = ∫ R n e i t λ x n + 1 s n + 1 e − D + Φ ∗∗ ( s ) − s ⋅ x ˙ + 1 ( 1 + ( − − D + Φ ∗∗ ( s ) λ s n + 1 ⋅ s , − D + Φ ∗∗ ( s ) λ s n + 1 ) 2 ) 2 n A m ( − − D + Φ ∗∗ ( s ) λ s n + 1 ⋅ s , − D + Φ ∗∗ ( s ) λ s n + 1 ) χ ( s ) χ 0 ( s ) Ψ ∗ ( s ) d s . (85)
We now restrict x x x to a region where x ˙ \dot{x} x ˙ stays away from zero and x n + 1 x_{n+1} x n + 1 is small: take ∣ x ˙ ∣ ≈ 1 |\dot{x}|\approx1 ∣ x ˙ ∣ ≈ 1 , t ≈ 1 t\approx1 t ≈ 1 and x n + 1 ≈ λ − 1 x_{n+1}\approx\lambda^{-1} x n + 1 ≈ λ − 1 . Under these conditions s n + 1 s_{n+1} s n + 1 is approximately 1 on the support of χ 1 \chi_1 χ 1 , and a direct estimation using the symbol properties of A m A_m A m gives
∣ ( ∂ ∂ λ ) α ( ∂ ∂ s n + 1 ) β G ( λ x n + 1 , t , s n + 1 , x ˙ ) ∣ ≤ C α , β λ − n 2 − α . (86) \left|\left(\frac{\partial}{\partial\lambda}\right)^\alpha\left(\frac{\partial}{\partial s_{n+1}}\right)^\beta G(\lambda x_{n+1},t,s_{n+1},\dot{x})\right|\leq C_{\alpha,\beta}\lambda^{-\frac n2-\alpha}. \tag*{(86)} ( ∂ λ ∂ ) α ( ∂ s n + 1 ∂ ) β G ( λ x n + 1 , t , s n + 1 , x ˙ ) ≤ C α , β λ − 2 n − α . (86)
Applying the stationary phase theorem to the outer s n + 1 s_{n+1} s n + 1 integral in (84) (the phase is s n + 1 2 2 − t s n + 1 \frac{s_{n+1}^2}{2}-ts_{n+1} 2 s n + 1 2 − t s n + 1 ) yields
M a t f λ ( x ˉ ) = c 0 λ n + 1 2 e i 2 λ t 2 χ 1 ( t ) ∣ t ∣ n G ( λ t , t , x n + 1 , x ˙ ) + O ( λ n 2 − 1 2 ) . M_{at}f_\lambda(\bar{x})=c_0\lambda^{n+\frac12}e^{\frac{i}{2}\lambda t^2}\chi_1(t)|t|^nG(\lambda t,t,x_{n+1},\dot{x})+O\left(\lambda^{\frac n2-\frac12}\right). M a t f λ ( x ˉ ) = c 0 λ n + 2 1 e 2 i λ t 2 χ 1 ( t ) ∣ t ∣ n G ( λ t , t , x n + 1 , x ˙ ) + O ( λ 2 n − 2 1 ) .
The error term is of lower order because of the rapid decay of the stationary phase expansion. Observe that for our chosen region, λ n 2 G ( λ x n + 1 , t , s n + 1 , x ˙ ) ≈ 1 \lambda^{\frac n2}G(\lambda x_{n+1},t,s_{n+1},\dot{x})\approx1 λ 2 n G ( λ x n + 1 , t , s n + 1 , x ˙ ) ≈ 1 , and ∣ x i ∣ ≈ λ − 1 |x_i|\approx\lambda^{-1} ∣ x i ∣ ≈ λ − 1 , ∣ x ˙ ∣ ≈ 1 |\dot{x}|\approx1 ∣ x ˙ ∣ ≈ 1 . Denote the main contribution by
I ( x ˉ , t , λ ) = λ n + 1 2 e i 2 λ t 2 χ 1 ( t ) ∣ t ∣ n G ( λ t , t , x n + 1 , x ˙ ) . I(\bar{x},t,\lambda)=\lambda^{n+\frac12}e^{\frac{i}{2}\lambda t^2}\chi_1(t)|t|^nG(\lambda t,t,x_{n+1},\dot{x}). I ( x ˉ , t , λ ) = λ n + 2 1 e 2 i λ t 2 χ 1 ( t ) ∣ t ∣ n G ( λ t , t , x n + 1 , x ˙ ) .
Now choose a sequence of times t k = 2 π k λ t_k=\sqrt{\frac{2\pi k}{\lambda}} t k = λ 2 π k for integers k k k with k ∼ b 2 λ k\sim b^2\lambda k ∼ b 2 λ . Because χ 1 ( t ) \chi_1(t) χ 1 ( t ) is supported away from 0, these t k t_k t k lie in the support of χ 1 \chi_1 χ 1 . Then
∣ I ( x ˉ , t k + 1 , λ ) − I ( x ˉ , t k , λ ) ∣ ≳ C 1 λ n + 1 2 − c 0 λ n − 1 2 , |I(\bar{x},t_{k+1},\lambda)-I(\bar{x},t_k,\lambda)|\gtrsim C_1\lambda^{\frac{n+1}{2}}-c_0\lambda^{\frac{n-1}{2}}, ∣ I ( x ˉ , t k + 1 , λ ) − I ( x ˉ , t k , λ ) ∣ ≳ C 1 λ 2 n + 1 − c 0 λ 2 n − 1 ,
since the phase difference is approximately π \pi π . Summing over the O ( λ ) O(\lambda) O ( λ ) values of k k k , we obtain
V r ( M a t f λ ) ( x ˉ ) ≳ C 1 λ n + 1 2 + 1 r − c 0 λ n − 1 2 + 1 r . V_r(M_{at}f_\lambda)(\bar{x})\gtrsim C_1\lambda^{\frac{n+1}{2}+\frac1r}-c_0\lambda^{\frac{n-1}{2}+\frac1r}. V r ( M a t f λ ) ( x ˉ ) ≳ C 1 λ 2 n + 1 + r 1 − c 0 λ 2 n − 1 + r 1 .
Integrating over the region of x x x (which has measure ≈ λ − ( n + 1 ) ) \approx\lambda^{-(n+1)}) ≈ λ − ( n + 1 ) ) gives
∥ V r ( M a t f λ ) ∥ L p ≳ λ n + 1 2 + 1 r − n + 1 p . (87) \|V_r(M_{at}f_\lambda)\|_{L^p}\gtrsim\lambda^{\frac{n+1}{2}+\frac1r-\frac{n+1}{p}}. \tag*{(87)} ∥ V r ( M a t f λ ) ∥ L p ≳ λ 2 n + 1 + r 1 − p n + 1 . (87)
Together with the estimate ∥ f λ ∥ L p ≲ λ n + 1 2 \|f_\lambda\|_{L^p} \lesssim\lambda^{\frac{n+1}{2}} ∥ f λ ∥ L p ≲ λ 2 n + 1 , the assumed boundedness ∥ V r ( A t f λ ) ∥ L p ≲ ∥ f λ ∥ L p \|V_r(A_t f_\lambda)\|_{L^p} \lesssim\|f_\lambda\|_{L^p} ∥ V r ( A t f λ ) ∥ L p ≲ ∥ f λ ∥ L p implies
λ n + 1 2 + 1 r − n + 1 p ≲ λ n + 1 2 . \lambda^{\frac{n+1}{2}+\frac{1}{r}-\frac{n+1}{p}} \lesssim\lambda^{\frac{n+1}{2}}. λ 2 n + 1 + r 1 − p n + 1 ≲ λ 2 n + 1 .
Letting λ → ∞ \lambda\to\infty λ → ∞ forces 1 r − n + 1 p ≤ 0 \frac{1}{r}-\frac{n+1}{p} \le0 r 1 − p n + 1 ≤ 0 , i.e., p ≤ r ( n + 1 ) p \le r(n+1) p ≤ r ( n + 1 ) . For k = n k=n k = n this is exactly p < r ( n + 1 ) p<r(n+1) p < r ( n + 1 ) (since p p p cannot equal the endpoint by a separate argument). This completes the proof of the necessary condition.
The proof of Theorem 1.1 for k < n k<n k < n
The proof of (1.1)
In this subsection we reduce the case of k k k non-zero principal curvatures to the already established full curvature case (k = n k=n k = n ) by a rotation argument. The key observation is that, after a suitable orthogonal transformation, the hypersurface can be locally represented as a graph in which the first k k k variables carry the non-degenerate quadratic part, while the remaining n − k n-k n − k variables only appear through a rotation that does not affect the L p L^p L p norms.
Let x ~ = ( x 1 , … , x k ) \tilde{x}=(x_1,\ldots,x_k) x ~ = ( x 1 , … , x k ) and x ~ ~ = ( x k + 2 , … , x n + 1 ) \tilde{\tilde{x}}=(x_{k+2},\ldots,x_{n+1}) x ~ ~ = ( x k + 2 , … , x n + 1 ) . We first state a decomposition lemma.
Lemma 4.1 The averaging operator A t f A_t f A t f can be represented as follows:
A t f ( x ˉ ) = ∫ R n − k R − θ ~ A ~ t θ ~ R θ ~ f ( x ˉ ) ψ ( θ ~ ~ ) d θ ~ ~ , (88) A_t f(\bar{x})=\int_{\mathbb{R}^{n-k}}R^{-\tilde{\theta}}\widetilde{A}_t^{\tilde{\theta}}R^{\tilde{\theta}}f(\bar{x})\psi(\tilde{\tilde{\theta}})\,d\tilde{\tilde{\theta}}, \tag*{(88)} A t f ( x ˉ ) = ∫ R n − k R − θ ~ A t θ ~ R θ ~ f ( x ˉ ) ψ ( θ ~ ~ ) d θ ~ ~ , (88) where ψ ( θ ~ ~ ) \psi(\tilde{\tilde{\theta}}) ψ ( θ ~ ~ ) is supported in a small neighbourhood of the origin, R θ ~ : = R θ k + 1 R θ k + 2 ⋯ R θ n R^{\tilde{\theta}}:=R^{\theta_{k+1}}R^{\theta_{k+2}}\cdots R^{\theta_n} R θ ~ := R θ k + 1 R θ k + 2 ⋯ R θ n , R − θ ~ : = R − θ n R − θ n − 1 ⋯ R − θ k + 1 R^{-\tilde{\theta}}:=R^{-\theta_n}R^{-\theta_{n-1}}\cdots R^{-\theta_{k+1}} R − θ ~ := R − θ n R − θ n − 1 ⋯ R − θ k + 1 , and for each fixed θ ~ ~ \tilde{\tilde{\theta}} θ ~ ~ ,
A ~ t θ ~ f ( x ~ , x k + 1 , x ~ ~ ) : = ∫ R k f ( x ~ − t y ~ , x k + 1 − t ( C ( θ ~ ~ ) + Φ ~ ( y ~ , θ ~ ~ ) ) , x ~ ~ ) ρ 1 ( y ~ , θ ~ ~ ) d y ~ . \widetilde{A}_t^{\tilde{\theta}}f(\tilde{x},x_{k+1},\tilde{\tilde{x}}):=\int_{\mathbb{R}^k}f\bigl(\tilde{x}-t\tilde{y},x_{k+1}-t(C(\tilde{\tilde{\theta}})+\widetilde{\Phi}(\tilde{y},\tilde{\tilde{\theta}})),\tilde{\tilde{x}}\bigr)\rho_1(\tilde{y},\tilde{\tilde{\theta}})\,d\tilde{y}. A t θ ~ f ( x ~ , x k + 1 , x ~ ~ ) := ∫ R k f ( x ~ − t y ~ , x k + 1 − t ( C ( θ ~ ~ ) + Φ ( y ~ , θ ~ ~ )) , x ~ ~ ) ρ 1 ( y ~ , θ ~ ~ ) d y ~ . Here C ( θ ~ ~ ) C(\tilde{\tilde{\theta}}) C ( θ ~ ~ ) is a smooth function with C ( 0 ) = D C(0)=D C ( 0 ) = D , Φ ~ ( y ~ , θ ~ ~ ) = Φ ( y ~ , y ~ ~ ( θ ~ ~ , y ~ ) ) \widetilde{\Phi}(\tilde{y},\tilde{\tilde{\theta}})=\Phi(\tilde{y},\tilde{\tilde{y}}(\tilde{\tilde{\theta}},\tilde{y})) Φ ( y ~ , θ ~ ~ ) = Φ ( y ~ , y ~ ~ ( θ ~ ~ , y ~ )) , and R θ j R^{\theta_j} R θ j (j = k + 1 , … , n j=k+1,\ldots,n j = k + 1 , … , n ) are rotation operators defined by
R θ j f ( x ˉ ) = f ( x 1 , … , x j , − x j sin θ j + x j + 1 cos θ j , − x j cos θ j − x j + 1 sin θ j , x j + 2 , … , x n + 1 ) . R^{\theta_j}f(\bar{x})=f(x_1,\ldots,x_j,-x_j\sin\theta_j+x_{j+1}\cos\theta_j,-x_j\cos\theta_j-x_{j+1}\sin\theta_j,x_{j+2},\ldots,x_{n+1}). R θ j f ( x ˉ ) = f ( x 1 , … , x j , − x j sin θ j + x j + 1 cos θ j , − x j cos θ j − x j + 1 sin θ j , x j + 2 , … , x n + 1 ) . The rotations R θ j R^{\theta_j} R θ j are isometries of R n + 1 \mathbb{R}^{n+1} R n + 1 ; consequently, the L p L^p L p norm of any operator constructed from them is invariant under these rotations. Moreover, since Φ \Phi Φ has exactly k k k non-zero principal curvatures at the origin, after a preliminary orthogonal transformation we may assume that the quadratic part of Φ \Phi Φ is diagonalised in the first k k k variables:
Φ ( x ) = ∑ i = 1 k λ i x i 2 + R ( x ) , λ i ≠ 0 , \Phi(x)=\sum_{i=1}^{k}\lambda_i x_i^2+R(x),\qquad\lambda_i\ne0, Φ ( x ) = i = 1 ∑ k λ i x i 2 + R ( x ) , λ i = 0 , and that D α R ( 0 ) = 0 D^\alpha R(0)=0 D α R ( 0 ) = 0 for ∣ α ∣ ≤ 2 |\alpha|\le2 ∣ α ∣ ≤ 2 . The remaining n − k n-k n − k variables appear only in higher order terms; their effect will be handled by the rotation averaging.
*Now apply Lemma 4.1. For each θ ~ ~ \tilde{\tilde{\theta}} θ ~ ~ the operator A ~ t θ ~ \widetilde{A}_t^{\tilde{\theta}} A t θ ~ is of the same form as the operator studied in Theorem 1.1 (the case k = n k=n k = n ), but with k k k variables instead of n n n and with translation parameter C ( θ ~ ~ ) ≈ D C(\tilde{\tilde{\theta}})\approx D C ( θ ~ ~ ) ≈ D (since θ ~ ~ \tilde{\tilde{\theta}} θ ~ ~ is small). Moreover, the phase function Φ ~ ( y ~ , θ ~ ~ ) \widetilde{\Phi}(\tilde{y},\tilde{\tilde{\theta}}) Φ ( y ~ , θ ~ ~ ) has a non-degenerate quadratic
part ∑ λ i y ~ i 2 \sum\lambda_i\widetilde{y}_i^2 ∑ λ i y i 2 and a remainder R ~ \widetilde{R} R that vanishes to second order at y ~ = 0 \widetilde{y}=0 y = 0 . Thus, by Theorem 1.1 (the case k = n k=n k = n ), for every r > 2 r>2 r > 2 and every p p p satisfying ( k + 1 ) / k < p < r ( k + 1 ) (k+1)/k<p<r(k+1) ( k + 1 ) / k < p < r ( k + 1 ) we have
∥ S r ( A ~ θ ~ f ) ∥ L p ( R n + 1 ) ≤ C ( p , r , D ) ∥ f ∥ L p ( R n + 1 ) , \|S_r(\widetilde{\mathcal{A}}^{\widetilde{\theta}}f)\|_{L^p(\mathbb{R}^{n+1})}\leq C(p,r,D)\|f\|_{L^p(\mathbb{R}^{n+1})}, ∥ S r ( A θ f ) ∥ L p ( R n + 1 ) ≤ C ( p , r , D ) ∥ f ∥ L p ( R n + 1 ) ,
where the constant C ( p , r , D ) C(p,r,D) C ( p , r , D ) enjoys the same dependence on D D D as in (1.2).
Using Minkowski’s inequality for integrals and the fact that the rotations R θ ~ R^{\widetilde{\theta}} R θ preserve L p L^p L p norms, we obtain
S r ( A f ) ( x ~ ) ≤ ∫ R n − k S r ( R − θ ~ A ~ θ ~ R θ ~ f ) ( x ~ ) ψ ( θ ~ ) d θ ~ . S_r(\mathcal{A}f)(\widetilde{x})\leq\int_{\mathbb{R}^{n-k}}S_r(R^{-\widetilde{\theta}}\widetilde{\mathcal{A}}^{\widetilde{\theta}}R^{\widetilde{\theta}}f)(\widetilde{x})\psi(\widetilde{\theta})\,d\widetilde{\theta}. S r ( A f ) ( x ) ≤ ∫ R n − k S r ( R − θ A θ R θ f ) ( x ) ψ ( θ ) d θ .
Integrating over x ~ \widetilde{x} x and applying the norm estimate for each fixed θ ~ \widetilde{\theta} θ (together with the uniform bound on ψ \psi ψ ) yields
∥ S r ( A f ) ∥ L p ≤ C ( p , r , D ) ∥ f ∥ L p . \|S_r(\mathcal{A}f)\|_{L^p}\leq C(p,r,D)\|f\|_{L^p}. ∥ S r ( A f ) ∥ L p ≤ C ( p , r , D ) ∥ f ∥ L p .
This completes the proof of (1.1).
The sharp exponent of D D D
In this subsection we demonstrate that the exponent of D D D appearing in (1.2) is optimal, up to the arbitrarily small ϵ \epsilon ϵ . More precisely, we show that if the estimate
∥ V r ( A t f : t > 0 ) ∥ L p ≤ C D s ∥ f ∥ L p \|V_r(A_t f:t>0)\|_{L^p}\leq C D^s\|f\|_{L^p} ∥ V r ( A t f : t > 0 ) ∥ L p ≤ C D s ∥ f ∥ L p
holds for some s s s independent of D D D , then necessarily s ≥ 1 / r s\geq1/r s ≥ 1/ r and s ≥ 1 / p s\geq1/p s ≥ 1/ p . Consequently s ≥ 1 / ( r ∧ p ) s\geq1/(r\wedge p) s ≥ 1/ ( r ∧ p ) , which matches the upper bound up to ϵ \epsilon ϵ .
We construct two families of test functions. The first one gives a lower bound on the variation in terms of D D D , leading to 1 / r ≤ s 1/r\leq s 1/ r ≤ s ; the second one exploits the support of the averaging operator to obtain 1 / p ≤ s 1/p\leq s 1/ p ≤ s .
First counterexample: 1 / r ≤ s 1/r\leq s 1/ r ≤ s . Choose a smooth cut-off function ψ \psi ψ supported in [ − b , b ] [-b,b] [ − b , b ] with the property that ∣ e i 2 η − 1 ∣ < δ |e^{i2\eta}-1|<\delta ∣ e i 2 η − 1∣ < δ for all ∣ η ∣ < b |\eta|<b ∣ η ∣ < b (with δ \delta δ a small number to be fixed later). Define a family { f N } \{f_N\} { f N } by
f N ^ ( ξ ) = N ψ ( N ( ξ n + 1 − 1 ) ) ∏ i = 1 n N ψ ( N ξ i ) . \widehat{f_N}(\xi)=N\psi\bigl(N(\xi_{n+1}-1)\bigr)\prod_{i=1}^{n}N\psi(N\xi_i). f N ( ξ ) = N ψ ( N ( ξ n + 1 − 1 ) ) i = 1 ∏ n N ψ ( N ξ i ) .
By the scaling properties of the Fourier transform, ∥ f N ∥ L p ≈ N n + 1 p \|f_N\|_{L^p}\approx N^{\frac{n+1}{p}} ∥ f N ∥ L p ≈ N p n + 1 . Recall that the multiplier of A t A_t A t is μ 0 ( t ξ ~ ) e − i t D ξ n + 1 \mu_0(t\widetilde{\xi})e^{-itD\xi_{n+1}} μ 0 ( t ξ ) e − i t D ξ n + 1 , where
μ 0 ( ξ ~ ) = ∫ R n + 1 e − i ( x ⋅ ξ ~ + Φ ( x ) ξ n + 1 ) ρ ( x ) d x . \mu_0(\widetilde{\xi})=\int_{\mathbb{R}^{n+1}}e^{-i(x\cdot\widetilde{\xi}+\Phi(x)\xi_{n+1})}\rho(x)\,dx. μ 0 ( ξ ) = ∫ R n + 1 e − i ( x ⋅ ξ + Φ ( x ) ξ n + 1 ) ρ ( x ) d x .
Hence
A t f N ( x ~ ) = 1 ( 2 π ) n + 1 ∫ e i x ~ ⋅ ξ ~ e − i t D ξ n + 1 μ 0 ( t ξ ~ ) N ψ ( N ( ξ n + 1 − 1 ) ) ∏ i = 1 n N ψ ( N ξ i ) d ξ ~ = e i x n + 1 e − i t D 1 ( 2 π ) n + 1 ∫ e i ( ∑ i = 1 n + 1 x i N ξ i + x n + 1 − t D ξ n + 1 N ) μ 0 ( t ξ ~ N , t ξ n + 1 N + 1 ) ∏ i = 1 n + 1 ψ ( ξ i ) d ξ ~ . \begin{aligned}
A_tf_N(\widetilde{x})=\frac{1}{(2\pi)^{n+1}}\int e^{i\widetilde{x}\cdot\widetilde{\xi}}e^{-itD\xi_{n+1}}\mu_0(t\widetilde{\xi})N\psi\bigl(N(\xi_{n+1}-1)\bigr)\prod_{i=1}^{n}N\psi(N\xi_i)\,d\widetilde{\xi} \\
&=e^{ix_{n+1}}e^{-itD}\frac{1}{(2\pi)^{n+1}}\int e^{i\left(\sum_{i=1}^{n+1}\frac{x_i}{N}\xi_i+x_{n+1}-\frac{tD\xi_{n+1}}{N}\right)}\mu_0\left(\frac{t\widetilde{\xi}}{N},\frac{t\xi_{n+1}}{N}+1\right)\prod_{i=1}^{n+1}\psi(\xi_i)\,d\widetilde{\xi}.
\end{aligned} A t f N ( x ) = ( 2 π ) n + 1 1 ∫ e i x ⋅ ξ e − i t D ξ n + 1 μ 0 ( t ξ ) N ψ ( N ( ξ n + 1 − 1 ) ) i = 1 ∏ n N ψ ( N ξ i ) d ξ = e i x n + 1 e − i t D ( 2 π ) n + 1 1 ∫ e i ( ∑ i = 1 n + 1 N x i ξ i + x n + 1 − N t D ξ n + 1 ) μ 0 ( N t ξ , N t ξ n + 1 + 1 ) i = 1 ∏ n + 1 ψ ( ξ i ) d ξ .
Now restrict x x x to the cube ∣ x i ∣ ≤ 2 N |x_i| \le2N ∣ x i ∣ ≤ 2 N (i = 1 , … , n + 1 i=1,\ldots,n+1 i = 1 , … , n + 1 ) and choose a sequence of times t k = k π / D t_k=k\pi/D t k = k π / D for k = 1 , … , [ D ] k=1,\ldots,[D] k = 1 , … , [ D ] (where [ D ] [D] [ D ] is the integer part of D D D ). Write the r r r -variation as a sum over consecutive times. A straightforward expansion gives
e − i x n + 1 ( 2 π ) n + 1 ( ∑ k ∣ A t k + 1 f N ( x ˉ ) − A t k f N ( x ˉ ) ∣ r ) 1 / r ≥ I 1 − I 2 − I 3 , e^{-ix_{n+1}}(2\pi)^{n+1}\left(\sum_k\left|A_{t_{k+1}}f_N(\bar{x})-A_{t_k}f_N(\bar{x})\right|^r\right)^{1/r}
\ge I_1-I_2-I_3, e − i x n + 1 ( 2 π ) n + 1 ( k ∑ ∣ A t k + 1 f N ( x ˉ ) − A t k f N ( x ˉ ) ∣ r ) 1/ r ≥ I 1 − I 2 − I 3 ,
where
I 1 : = ( ∑ k ∣ ( e − i t k + 1 D − e − i t k D ) ∫ R n + 1 μ 0 ( t k + 1 ξ N , t k + 1 ξ n + 1 N + 1 ) ∏ ψ ( ξ i ) d ξ ∣ r ) 1 / r , I_1:=\left(\sum_k\left|(e^{-it_{k+1}D}-e^{-it_kD})\int_{\mathbb{R}^{n+1}}\mu_0\left(\frac{t_{k+1}\xi}{N},\frac{t_{k+1}\xi_{n+1}}{N}+1\right)\prod\psi(\xi_i)\,d\xi\right|^r\right)^{1/r}, I 1 := ( k ∑ ( e − i t k + 1 D − e − i t k D ) ∫ R n + 1 μ 0 ( N t k + 1 ξ , N t k + 1 ξ n + 1 + 1 ) ∏ ψ ( ξ i ) d ξ r ) 1/ r ,
I 2 : = ( ∑ k ∣ ( e − i t k + 1 D − e − i t k D ) ∫ ( e i ( ⋯ ) − 1 ) μ 0 ( ⋯ ) ∏ ψ d ξ ∣ r ) 1 / r , I_2:=\left(\sum_k\left|(e^{-it_{k+1}D}-e^{-it_kD})\int(e^{i(\cdots)}-1)\mu_0(\cdots)\prod\psi\,d\xi\right|^r\right)^{1/r}, I 2 := ( k ∑ ( e − i t k + 1 D − e − i t k D ) ∫ ( e i ( ⋯ ) − 1 ) μ 0 ( ⋯ ) ∏ ψ d ξ r ) 1/ r ,
I 3 : = ( ∑ k ∣ ∫ ( e i ( ⋯ − t k + 1 D ξ n + 1 / D ) μ 0 ( t k + 1 ξ N , ⋅ ) − e i ( ⋯ − t k D ξ n + 1 / D ) μ 0 ( t k ξ N , ⋅ ) ) ∏ ψ d ξ ∣ r ) 1 / r . I_3:=\left(\sum_k\left|\int\left(e^{i(\cdots-t_{k+1}D\xi_{n+1}/D)}\mu_0\left(\frac{t_{k+1}\xi}{N},\cdot\right)-e^{i(\cdots-t_kD\xi_{n+1}/D)}\mu_0\left(\frac{t_k\xi}{N},\cdot\right)\right)\prod\psi\,d\xi\right|^r\right)^{1/r}. I 3 := ( k ∑ ∫ ( e i ( ⋯ − t k + 1 D ξ n + 1 / D ) μ 0 ( N t k + 1 ξ , ⋅ ) − e i ( ⋯ − t k D ξ n + 1 / D ) μ 0 ( N t k ξ , ⋅ ) ) ∏ ψ d ξ r ) 1/ r .
Because ∣ μ 0 ( 0 , 1 ) ∣ = 1 |\mu_0(0,1)|=1 ∣ μ 0 ( 0 , 1 ) ∣ = 1 , for sufficiently large N N N we have
∣ μ 0 ( t k ξ N , t k ξ n + 1 N + 1 ) ∣ > 1 / 2 \left|\mu_0\left(\frac{t_k\xi}{N},\frac{t_k\xi_{n+1}}{N}+1\right)\right|>1/2 μ 0 ( N t k ξ , N t k ξ n + 1 + 1 ) > 1/2
uniformly in k k k and ξ ∈ supp ψ \xi\in\operatorname{supp}\psi ξ ∈ supp ψ . Moreover, ∣ e − i t k + 1 D − e − i t k D ∣ = ∣ e − i π − 1 ∣ = 2 |e^{-it_{k+1}D}-e^{-it_kD}|=|e^{-i\pi}-1|=2 ∣ e − i t k + 1 D − e − i t k D ∣ = ∣ e − iπ − 1∣ = 2 , so I 1 ≳ D 1 / r I_1\gtrsim D^{1/r} I 1 ≳ D 1/ r .
For I 2 I_2 I 2 , note that when N N N is sufficiently large and b b b sufficiently small, the factor
e i ( ∑ x i ξ i / N − t k + 1 D ξ n + 1 / N ) e^{i(\sum x_i\xi_i/N-t_{k+1}D\xi_{n+1}/N)} e i ( ∑ x i ξ i / N − t k + 1 D ξ n + 1 / N )
is close to 1 1 1 because ∣ x i ∣ ≤ 2 N |x_i|\le2N ∣ x i ∣ ≤ 2 N , ∣ ξ i ∣ ≤ b |\xi_i|\le b ∣ ξ i ∣ ≤ b and t k + 1 ≤ D t_{k+1}\le D t k + 1 ≤ D ; indeed we can make the difference smaller than any prescribed ϵ \epsilon ϵ by taking N N N sufficiently large. Also μ 0 \mu_0 μ 0 is bounded, so I 2 ≲ ϵ D 1 / r I_2\lesssim\epsilon D^{1/r} I 2 ≲ ϵ D 1/ r . Choosing ϵ \epsilon ϵ small enough, this term is negligible compared to I 1 I_1 I 1 .
For I 3 I_3 I 3 , a direct estimate using the smoothness of μ 0 \mu_0 μ 0 and the fact that t k + 1 − t k = π / D t_{k+1}-t_k=\pi/D t k + 1 − t k = π / D shows that I 3 ≤ C N − 1 D 1 / r I_3\le CN^{-1}D^{1/r} I 3 ≤ C N − 1 D 1/ r . Hence, for large N N N ,
∥ V r ( A t f N ) ∥ L p ≳ D 1 / r N n + 1 p . \|V_r(A_tf_N)\|_{L^p}\gtrsim D^{1/r}N^{\frac{n+1}{p}}. ∥ V r ( A t f N ) ∥ L p ≳ D 1/ r N p n + 1 .
If the assumed estimate ∥ V r ( A t f ) ∥ L p ≤ D s ∥ f ∥ L p \|V_r(A_tf)\|_{L^p}\le D^s\|f\|_{L^p} ∥ V r ( A t f ) ∥ L p ≤ D s ∥ f ∥ L p holds, then
D 1 / r N n + 1 p ≲ D s N n + 1 p . D^{1/r}N^{\frac{n+1}{p}}\lesssim D^sN^{\frac{n+1}{p}}. D 1/ r N p n + 1 ≲ D s N p n + 1 .
Letting N → ∞ N\to\infty N → ∞ first and then D → ∞ D\to\infty D → ∞ yields 1 / r ≤ s 1/r\le s 1/ r ≤ s .
Second counterexample: 1 / p ≤ s 1/p\le s 1/ p ≤ s . Consider a fixed bump function f ~ ( x ˉ ) = ∏ i = 1 n + 1 χ [ − 1 , 1 ] ( x i ) \widetilde{f}(\bar{x})=\prod_{i=1}^{n+1}\chi_{[-1,1]}(x_i) f ( x ˉ ) = ∏ i = 1 n + 1 χ [ − 1 , 1 ] ( x i ) . Choose the cutoff ρ \rho ρ in the definition of A t A_t A t so that ρ ( y ) \rho(y) ρ ( y ) is supported in ∣ y ∣ ≤ 1 / 2 |y|\le1/2 ∣ y ∣ ≤ 1/2 and ∣ Φ ( y ) ∣ ≤ 1 |\Phi(y)|\le1 ∣Φ ( y ) ∣ ≤ 1 on that support. For x x x satisfying ∣ x i ∣ ≤ 1 / 2 |x_i|\le1/2 ∣ x i ∣ ≤ 1/2 (i = 1 , … , n i=1,\ldots,n i = 1 , … , n ) and 1 / 2 ≤ x n + 1 ≤ D / 2 1/2\le x_{n+1}\le D/2 1/2 ≤ x n + 1 ≤ D /2 , set t 0 = ( x n + 1 − 1 / 2 ) / D t_0=(x_{n+1}-1/2)/D t 0 = ( x n + 1 − 1/2 ) / D . Then t 0 ≤ 1 / 2 t_0\le1/2 t 0 ≤ 1/2 and
∣ x n + 1 − t 0 ( D + Φ ( y ) ) ∣ ≤ ∣ x n + 1 − t 0 D ∣ + t 0 ∣ Φ ( y ) ∣ ≤ 1 / 2 + 1 / 2 = 1 , |x_{n+1}-t_0(D+\Phi(y))|\le|x_{n+1}-t_0D|+t_0|\Phi(y)|\le1/2+1/2=1, ∣ x n + 1 − t 0 ( D + Φ ( y )) ∣ ≤ ∣ x n + 1 − t 0 D ∣ + t 0 ∣Φ ( y ) ∣ ≤ 1/2 + 1/2 = 1 ,
∣ x i − t 0 y i ∣ ≤ ∣ x i ∣ + t 0 ∣ y i ∣ ≤ 1 / 2 + 1 / 2 = 1. |x_i-t_0y_i|\le|x_i|+t_0|y_i|\le1/2+1/2=1. ∣ x i − t 0 y i ∣ ≤ ∣ x i ∣ + t 0 ∣ y i ∣ ≤ 1/2 + 1/2 = 1.
Hence A t 0 f ~ ( x ˉ ) = 1 A_{t_0}\widetilde{f}(\bar{x})=1 A t 0 f ( x ˉ ) = 1 . On the other hand, for the same x x x and for t = N t=N t = N large, we have
∣ x n + 1 − N ( D + Φ ( y ) ) ∣ ≥ N ( D − ∣ Φ ( y ) ∣ ) − ∣ x n + 1 ∣ ≥ N ( D − 1 ) − D > 1 |x_{n+1}-N(D+\Phi(y))| \ge N(D-|\Phi(y)|)-|x_{n+1}| \ge N(D-1)-D>1 ∣ x n + 1 − N ( D + Φ ( y )) ∣ ≥ N ( D − ∣Φ ( y ) ∣ ) − ∣ x n + 1 ∣ ≥ N ( D − 1 ) − D > 1
provided N > D / ( D − 1 ) N>D/(D-1) N > D / ( D − 1 ) . A precise calculation shows that for N N N sufficiently large, A N f ~ ( x ) = 0 A_N\widetilde{f}(x)=0 A N f ( x ) = 0 . Thus
V r ( A t f ~ ) ( x ) ≥ 1 for all such x . V_r(A_t\widetilde{f})(x)\ge1 \qquad\text{for all such }x. V r ( A t f ) ( x ) ≥ 1 for all such x .
The set of admissible x x x has measure ≳ D \gtrsim D ≳ D , therefore ∥ V r ( A t f ~ ) ∥ L p ≳ D 1 / p \|V_r(A_t\widetilde{f})\|_{L^p}\gtrsim D^{1/p} ∥ V r ( A t f ) ∥ L p ≳ D 1/ p . If the estimate ∥ V r ( A t f ) ∥ L p ≤ D s ∥ f ∥ L p \|V_r(A_tf)\|_{L^p}\le D^s\|f\|_{L^p} ∥ V r ( A t f ) ∥ L p ≤ D s ∥ f ∥ L p holds, then
D 1 / p ≲ D s ∥ f ~ ∥ L p ≲ D s , D^{1/p}\lesssim D^s\|\widetilde{f}\|_{L^p}\lesssim D^s, D 1/ p ≲ D s ∥ f ∥ L p ≲ D s ,
which forces 1 / p ≤ s 1/p\le s 1/ p ≤ s . Combining the two lower bounds we obtain s ≥ 1 / ( r ∧ p ) s\ge1/(r\wedge p) s ≥ 1/ ( r ∧ p ) . This shows that the dependence on D D D in (1.2) is optimal up to the arbitrarily small ϵ \epsilon ϵ .
Convex finite-type surface
Before we begin our proving, we first show some useful result which can be find in Iosevich-Sawyer [13 ]
Lemma 5.1 (Schulz) Let Φ \Phi Φ be a smooth convex function such that Φ ( 0 ) = 0 \Phi(0)=0 Φ ( 0 ) = 0 and ∇ Φ ( 0 ) = 0 \nabla\Phi(0)=0 ∇Φ ( 0 ) = 0 . Suppose that Φ \Phi Φ has no tangents of infinite order at the origin. Then, after perhaps applying a rotation, we can write
Φ ( x ) = Q ( x ) + R ( x ) , \Phi(x)=Q(x)+R(x), Φ ( x ) = Q ( x ) + R ( x ) , where Q ( x ) Q(x) Q ( x ) is a convex mixed homogeneous polynomial of degree ( a 1 , … , a n ) (a_1,\ldots,a_n) ( a 1 , … , a n ) , Q ( x ) ≠ 0 Q(x)\ne0 Q ( x ) = 0 for x ≠ 0 x\ne0 x = 0 , and R ( x ) R(x) R ( x ) is a remainder term in the sense that
lim s → 0 R ( s 1 / a 1 x 1 , … , s 1 / a n x n ) s = 0. \lim_{s\to0}\frac{R(s^{1/a_1}x_1,\ldots,s^{1/a_n}x_n)}{s}=0. s → 0 lim s R ( s 1/ a 1 x 1 , … , s 1/ a n x n ) = 0.
This result, due to Schulz, provides a decomposition of any smooth convex finite-type function into a mixed-homogeneous principal part Q Q Q and a remainder R R R that becomes negligible under the associated anisotropic dilations. The tuple ( a 1 , … , a n ) (a_1,\ldots,a_n) ( a 1 , … , a n ) is called the multi-type of Φ \Phi Φ at the origin.
Definition We say that a smooth function Q : R n → R Q:\mathbb{R}^n\to\mathbb{R} Q : R n → R is mixed homogeneous of degree ( a 1 , a 2 , … , a n ) (a_1,a_2,\ldots,a_n) ( a 1 , a 2 , … , a n ) , a j > 0 a_j>0 a j > 0 , if Q ( s 1 / a 1 x 1 , … , s 1 / a n x n ) = s Q ( x ) Q(s^{1/a_1}x_1,\ldots,s^{1/a_n}x_n)=sQ(x) Q ( s 1/ a 1 x 1 , … , s 1/ a n x n ) = s Q ( x ) , s > 0 s>0 s > 0 .
The exponents a j a_j a j describe how Q Q Q scales under different coordinate dilations; they encode the different rates of decay of Φ \Phi Φ in different directions.
Remark 5.1 For such a mixed homogeneous polynomial we have
1 a 1 + 1 a 2 + ⋯ + 1 a n = p 0 , \frac{1}{a_1}+\frac{1}{a_2}+\cdots+\frac{1}{a_n}=p_0, a 1 1 + a 2 1 + ⋯ + a n 1 = p 0 ,
where p 0 = inf { p : d ( x , H ) − 1 ∈ L l o c 1 / p ( S ) } p_0=\inf\{p:d(x,\mathcal{H})^{-1}\in L_{\mathrm{loc}}^{1/p}(S)\} p 0 = inf { p : d ( x , H ) − 1 ∈ L loc 1/ p ( S )} is the critical exponent for the associated maximal operator.
This equality follows from a straightforward computation in polar coordinates adapted to the mixed homogeneity. It links the multi-type to the integrability condition (1.3) that is known to be necessary and sufficient for the L p L^p L p boundedness of the maximal operator when p > 2 p > 2 p > 2 (see [13 ] ). Thus it remains for us to prove the following theorem.
Remark 5.2 For such a mixed homogeneous polynomial we have
1 a 1 + 1 a 2 + ⋯ + 1 a n = p 0 , \frac{1}{a_1}+\frac{1}{a_2}+\cdots+\frac{1}{a_n}=p_0, a 1 1 + a 2 1 + ⋯ + a n 1 = p 0 ,
where p 0 = inf { p : d ( x , H ) − 1 ∈ L l o c 1 / p ( S ) } p_0=\inf\{p:d(x,\mathcal{H})^{-1}\in L_{\mathrm{loc}}^{1/p}(S)\} p 0 = inf { p : d ( x , H ) − 1 ∈ L loc 1/ p ( S )} is the critical exponent for the associated maximal operator.
Thus it remains for us to prove the following theorem.
Theorem 5.2 Let Φ \Phi Φ be a smooth function with the following properties. Suppose that after perhaps applying a rotation, Φ ( x ) = Q ( x ) + R ( x ) \Phi(x)=Q(x)+R(x) Φ ( x ) = Q ( x ) + R ( x ) , where Q Q Q is mixed homogeneous of degree ( a 1 , … , a n ) (a_1,\ldots,a_n) ( a 1 , … , a n ) with ( 1 / a 1 ) + ⋯ + ( 1 / a n ) ≤ 1 (1/a_1)+\cdots+(1/a_n)\le1 ( 1/ a 1 ) + ⋯ + ( 1/ a n ) ≤ 1 , Q ( x ) ≠ 0 Q(x)\ne0 Q ( x ) = 0 for x ≠ 0 x\ne0 x = 0 , and
lim s → 0 R ( s 1 / a 1 x 1 , … , s 1 / a n x n ) s = 0. (89) \lim_{s\to0}\frac{R(s^{1/a_1}x_1,\ldots,s^{1/a_n}x_n)}{s}=0. \tag*{(89)} s → 0 lim s R ( s 1/ a 1 x 1 , … , s 1/ a n x n ) = 0. (89) Let S S S be the surface given as the graph of Φ ( x ) + D \Phi(x)+D Φ ( x ) + D , with a smooth cutoff ρ \rho ρ supported in a sufficiently small neighbourhood of the origin. Suppose that ( 1 / a 1 ) + ⋯ + ( 1 / a n ) ≤ 1 2 (1/a_1)+\cdots+(1/a_n)\le\frac{1}{2} ( 1/ a 1 ) + ⋯ + ( 1/ a n ) ≤ 2 1 . Then for any r > 1 ( 1 / a 1 ) + ⋯ + ( 1 / a n ) r>\frac{1}{(1/a_1)+\cdots+(1/a_n)} r > ( 1/ a 1 ) + ⋯ + ( 1/ a n ) 1 we have
∥ V r ( A f ) ∥ L p ( R n + 1 ) ≤ C ∥ f ∥ L p ( R n + 1 ) , 1 ( 1 / a 1 ) + ⋯ + ( 1 / a n ) < p < 2 r . \|V_r(Af)\|_{L^p(\mathbb{R}^{n+1})}\le C\|f\|_{L^p(\mathbb{R}^{n+1})},\qquad\frac{1}{(1/a_1)+\cdots+(1/a_n)}<p<2r. ∥ V r ( A f ) ∥ L p ( R n + 1 ) ≤ C ∥ f ∥ L p ( R n + 1 ) , ( 1/ a 1 ) + ⋯ + ( 1/ a n ) 1 < p < 2 r . Proof. We begin by decomposing the surface S S S into dyadic shells adapted to the non-isotropic dilation associated with the multi-type ( a 1 , … , a n ) (a_1,\ldots,a_n) ( a 1 , … , a n ) . Write
ρ ( x ) = ∑ k = 0 ∞ ρ k ( x ) = ∑ k = 0 ∞ ρ 0 ( 2 k / a 1 x 1 , … , 2 k / a n x n ) , \rho(x)=\sum_{k=0}^{\infty}\rho_k(x)=\sum_{k=0}^{\infty}\rho_0(2^{k/a_1}x_1,\ldots,2^{k/a_n}x_n), ρ ( x ) = k = 0 ∑ ∞ ρ k ( x ) = k = 0 ∑ ∞ ρ 0 ( 2 k / a 1 x 1 , … , 2 k / a n x n ) , where ρ k ( x ) = ρ 0 ( 2 k / a 1 x 1 , … , 2 k / a n x n ) \rho_k(x)=\rho_0(2^{k/a_1}x_1,\ldots,2^{k/a_n}x_n) ρ k ( x ) = ρ 0 ( 2 k / a 1 x 1 , … , 2 k / a n x n ) and ρ 0 \rho_0 ρ 0 is a smooth cut-off supported in the annulus { x : 1 ≤ ∣ x ∣ ≤ 2 } \{x:1\le|x|\le2\} { x : 1 ≤ ∣ x ∣ ≤ 2 } . For each k k k define the averaged operator
A k , t f ( x ˉ ) = ∫ R n f ( x − t y , x n + 1 − t ( D + Φ ( y ) ) ) ρ k ( y ) d y . A_{k,t}f(\bar{x})=\int_{\mathbb{R}^n}f(x-ty,x_{n+1}-t(D+\Phi(y)))\rho_k(y)\,dy. A k , t f ( x ˉ ) = ∫ R n f ( x − t y , x n + 1 − t ( D + Φ ( y ))) ρ k ( y ) d y . Let τ k \tau_k τ k be the isotropic dilation
τ k g ( x , x n + 1 ) = g ( 2 k / a 1 x 1 , … , 2 k / a n x n , 2 k x n + 1 ) . \tau_k g(x,x_{n+1})=g(2^{k/a_1}x_1,\ldots,2^{k/a_n}x_n,2^k x_{n+1}). τ k g ( x , x n + 1 ) = g ( 2 k / a 1 x 1 , … , 2 k / a n x n , 2 k x n + 1 ) . A direct computation yields
τ − k A k , t τ k f ( x ˉ ) = 2 − k ( 1 / a 1 + ⋯ + 1 / a n ) ∫ R n f ( x − t y , x n + 1 − t ( 2 k Φ ( 2 − k / a ∘ y ) + 2 k D ) ) ρ 0 ( y ) d y , \tau_{-k}A_{k,t}\tau_k f(\bar{x})=2^{-k(1/a_1+\cdots+1/a_n)}\int_{\mathbb{R}^n}f\left(x-ty,x_{n+1}-t\left(2^k\Phi(2^{-k/a}\circ y)+2^kD\right)\right)\rho_0(y)\,dy, τ − k A k , t τ k f ( x ˉ ) = 2 − k ( 1/ a 1 + ⋯ + 1/ a n ) ∫ R n f ( x − t y , x n + 1 − t ( 2 k Φ ( 2 − k / a ∘ y ) + 2 k D ) ) ρ 0 ( y ) d y ,
where 2 − k / a ∘ y = ( 2 − k / a 1 y 1 , … , 2 − k / a n y n ) 2^{-k/a}\circ y=(2^{-k/a_1}y_1,\ldots,2^{-k/a_n}y_n) 2 − k / a ∘ y = ( 2 − k / a 1 y 1 , … , 2 − k / a n y n ) . Using the decomposition Φ = Q + R \Phi=Q+R Φ = Q + R and the homogeneity of Q Q Q ,
2 k Φ ( 2 − k / a ∘ y ) = 2 k Q ( 2 − k / a ∘ y ) + 2 k R ( 2 − k / a ∘ y ) = Q ( y ) + 2 k R ( 2 − k / a ∘ y ) . 2^k\Phi(2^{-k/a}\circ y)=2^kQ(2^{-k/a}\circ y)+2^kR(2^{-k/a}\circ y)=Q(y)+2^kR(2^{-k/a}\circ y). 2 k Φ ( 2 − k / a ∘ y ) = 2 k Q ( 2 − k / a ∘ y ) + 2 k R ( 2 − k / a ∘ y ) = Q ( y ) + 2 k R ( 2 − k / a ∘ y ) .
By the remainder property (89) , we have lim k → ∞ 2 k R ( 2 − k / a ∘ y ) = 0 \lim_{k\to\infty}2^kR(2^{-k/a}\circ y)=0 lim k → ∞ 2 k R ( 2 − k / a ∘ y ) = 0 . Hence for sufficiently large k k k , the surface
{ ( y , y n + 1 ) : y n + 1 = Q ( y ) + 2 k R ( 2 − k / a ∘ y ) + 2 k D } \{(y,y_{n+1}):y_{n+1}=Q(y)+2^kR(2^{-k/a}\circ y)+2^kD\} {( y , y n + 1 ) : y n + 1 = Q ( y ) + 2 k R ( 2 − k / a ∘ y ) + 2 k D }
is a small perturbation of the graph { y n + 1 = Q ( y ) + 2 k D } \{y_{n+1}=Q(y)+2^kD\} { y n + 1 = Q ( y ) + 2 k D } and therefore possesses at least one non-vanishing principal curvature uniformly in k k k . Moreover, note that the translation parameter has become 2 k D 2^kD 2 k D .
Set m = 1 / a 1 + ⋯ + 1 / a n m=1/a_1+\cdots+1/a_n m = 1/ a 1 + ⋯ + 1/ a n . Applying Theorem 1.1 to the operator τ − k A k , t τ k \tau_{-k}A_{k,t}\tau_k τ − k A k , t τ k (which after the change of variables takes exactly the form studied there, with D D D replaced by 2 k D 2^kD 2 k D ) yields the estimate
∥ V r ( τ − k A k , t τ k f : t > 0 ) ∥ L p ( R n + 1 ) ≤ C 2 − k m 2 k r ∧ p ∥ f ∥ L p ( R n + 1 ) , 2 < p < 2 r . \left\|V_r(\tau_{-k}A_{k,t}\tau_kf:t>0)\right\|_{L^p(\mathbb{R}^{n+1})}\leq C2^{-km}2^{\frac{k}{r\wedge p}}\|f\|_{L^p(\mathbb{R}^{n+1})},\qquad2<p<2r. ∥ V r ( τ − k A k , t τ k f : t > 0 ) ∥ L p ( R n + 1 ) ≤ C 2 − k m 2 r ∧ p k ∥ f ∥ L p ( R n + 1 ) , 2 < p < 2 r .
The factor 2 k / ( r ∧ p ) 2^{k/(r\wedge p)} 2 k / ( r ∧ p ) reflects the dependence on the translation parameter 2 k D 2^kD 2 k D as in (1.4). Since τ k \tau_k τ k is an isometry on L p L^p L p up to a constant, the same bound holds for V r ( A k , t f : t > 0 ) V_r(A_{k,t}f:t>0) V r ( A k , t f : t > 0 ) after undoing the dilation. If r > 1 / m r>1/m r > 1/ m , then the geometric series converges, yielding
∥ V r ( A t f : t > 0 ) ∥ L p ≤ C ∥ f ∥ L p for all 1 m < p < 2 r . \left\|V_r(A_tf:t>0)\right\|_{L^p}\leq C\|f\|_{L^p}\qquad\text{for all }\frac{1}{m}<p<2r. ∥ V r ( A t f : t > 0 ) ∥ L p ≤ C ∥ f ∥ L p for all m 1 < p < 2 r .
The proof is complete. □ \square □
Appendix
In this section we prove (8) . Denote
F ( W t f ) ( ξ , ξ n + 1 ) = e − t ∣ ( ξ , ξ n + 1 ) ∣ f ^ ( ξ , ξ n + 1 ) \mathcal{F}(W_tf)(\xi,\xi_{n+1})=e^{-t|(\xi,\xi_{n+1})|}\widehat{f}(\xi,\xi_{n+1}) F ( W t f ) ( ξ , ξ n + 1 ) = e − t ∣ ( ξ , ξ n + 1 ) ∣ f ( ξ , ξ n + 1 )
and
F ( U t f ) ( ξ , ξ n + 1 ) = e − t ∣ ( ξ , ξ n + 1 ) ∣ ( e − i D t ξ n + 1 − 1 ) f ^ ( ξ , ξ n + 1 ) . \mathcal{F}(U_tf)(\xi,\xi_{n+1})=e^{-t|(\xi,\xi_{n+1})|}(e^{-iDt\xi_{n+1}}-1)\widehat{f}(\xi,\xi_{n+1}). F ( U t f ) ( ξ , ξ n + 1 ) = e − t ∣ ( ξ , ξ n + 1 ) ∣ ( e − i D t ξ n + 1 − 1 ) f ( ξ , ξ n + 1 ) .
For every f ∈ L p ( R n + 1 ) f\in L^p(\mathbb{R}^{n+1}) f ∈ L p ( R n + 1 ) , we obtain
∥ V r ( S t f : t > 0 ) ∥ L p ≤ C r ∥ S r ( W t f : t > 0 ) ∥ L p + C r ∥ V r ( U t f : t > 0 ) ∥ L p . (90) \left\|V_r(S_tf:t>0)\right\|_{L^p}\leq C_r\left\|S_r(W_tf:t>0)\right\|_{L^p}+C_r\left\|V_r(U_tf:t>0)\right\|_{L^p}. \tag*{(90)} ∥ V r ( S t f : t > 0 ) ∥ L p ≤ C r ∥ S r ( W t f : t > 0 ) ∥ L p + C r ∥ V r ( U t f : t > 0 ) ∥ L p . (90)
By Rota’s theorem (see Bourgain [4 ] ) and Lépingle’s inequality (see [15 , 26 ] ), the first term in (90) satisfies
∥ V r ( W t f : t > 0 ) ∥ L p ≤ C p ∥ f ∥ L p , p > 1. \left\|V_r(W_tf:t>0)\right\|_{L^p}\leq C_p\|f\|_{L^p},\qquad p>1. ∥ V r ( W t f : t > 0 ) ∥ L p ≤ C p ∥ f ∥ L p , p > 1.
Therefore, it remains to obtain L p L^p L p bounds for the second term in (90) . For this purpose, we will use the long-short decomposition (4) and establish
∥ V r ( U 2 j f : j ∈ Z ) ∥ L p ≤ C ϵ D ϵ ∥ f ∥ L p (91) \left\|V_r(U_{2^j}f:j\in\mathbb{Z})\right\|_{L^p}\leq C_\epsilon D^\epsilon\|f\|_{L^p} \tag*{(91)} ∥ V r ( U 2 j f : j ∈ Z ) ∥ L p ≤ C ϵ D ϵ ∥ f ∥ L p (91)
and
∥ S r ( U t f : t > 0 ) ∥ L p ≤ C ( p , r , D ) ∥ f ∥ L p , (92) \|S_r(U_t f:t>0)\|_{L^p} \le C(p,r,D)\|f\|_{L^p}, \tag*{(92)} ∥ S r ( U t f : t > 0 ) ∥ L p ≤ C ( p , r , D ) ∥ f ∥ L p , (92)
for p > 1 p>1 p > 1 .
Recall that V r ( U 2 j f : j ∈ Z ) ≤ ∥ U 2 j f ∥ ℓ j r ≤ ∥ U 2 j f ∥ ℓ j 2 V_r(U_{2^j}f:j\in\mathbb{Z})\le\|U_{2^j}f\|_{\ell^r_j}\le\|U_{2^j}f\|_{\ell^2_j} V r ( U 2 j f : j ∈ Z ) ≤ ∥ U 2 j f ∥ ℓ j r ≤ ∥ U 2 j f ∥ ℓ j 2 . To prove (91) it suffices to show
∥ U 2 j f ∥ L p ( ℓ j 2 ) ≤ C D ϵ ∥ f ∥ L p , p > 1. (93) \|U_{2^j}f\|_{L^p(\ell^2_j)}\le CD^\epsilon\|f\|_{L^p},\qquad p>1. \tag*{(93)} ∥ U 2 j f ∥ L p ( ℓ j 2 ) ≤ C D ϵ ∥ f ∥ L p , p > 1. (93)
We shall use the Littlewood-Paley operators P j P_j P j . Observe that
∥ U 2 j f ∥ L p ( ℓ j 2 ) = ∥ ( ∑ k ∈ Z U 2 j P k − j f ) ∥ L p ( ℓ j 2 ) ≤ C ϵ D ϵ ∑ k ∈ Z ∥ U 2 j P k − j f ∥ L p ( ℓ j 2 ) ≤ C ϵ D ϵ ∑ k ∈ Z 2 − δ p ∣ k ∣ ∥ f ∥ L p ≤ C D ϵ ∥ f ∥ L p . (94) \begin{aligned}
\|U_{2^j}f\|_{L^p(\ell^2_j)}
=\left\|\left(\sum_{k\in\mathbb{Z}}U_{2^j}P_{k-j}f\right)\right\|_{L^p(\ell^2_j)} \\
&\le C_\epsilon D^\epsilon\sum_{k\in\mathbb{Z}}\|U_{2^j}P_{k-j}f\|_{L^p(\ell^2_j)} \\
&\le C_\epsilon D^\epsilon\sum_{k\in\mathbb{Z}}2^{-\delta_p|k|}\|f\|_{L^p}
\le C D^\epsilon\|f\|_{L^p}. \tag*{(94)}
\end{aligned} ∥ U 2 j f ∥ L p ( ℓ j 2 ) = ( k ∈ Z ∑ U 2 j P k − j f ) L p ( ℓ j 2 ) ≤ C ϵ D ϵ k ∈ Z ∑ ∥ U 2 j P k − j f ∥ L p ( ℓ j 2 ) ≤ C ϵ D ϵ k ∈ Z ∑ 2 − δ p ∣ k ∣ ∥ f ∥ L p ≤ C D ϵ ∥ f ∥ L p . (94)
The penultimate inequality in (94) follows from the two estimates
∥ U 2 j P k − j f ∥ L 2 ( ℓ j 2 ) ≤ C ϵ 2 − ϵ ∣ k − k 0 ∣ D ϵ ∥ f ∥ L 2 , (95) \|U_{2^j}P_{k-j}f\|_{L^2(\ell^2_j)}
\le C_\epsilon2^{-\epsilon|k-k_0|}D^\epsilon\|f\|_{L^2}, \tag*{(95)} ∥ U 2 j P k − j f ∥ L 2 ( ℓ j 2 ) ≤ C ϵ 2 − ϵ ∣ k − k 0 ∣ D ϵ ∥ f ∥ L 2 , (95)
and
∥ U 2 j P k − j f ∥ L p ( ℓ j 2 ) ≤ C ∥ f ∥ L p . (96) \|U_{2^j}P_{k-j}f\|_{L^p(\ell^2_j)}
\le C\|f\|_{L^p}. \tag*{(96)} ∥ U 2 j P k − j f ∥ L p ( ℓ j 2 ) ≤ C ∥ f ∥ L p . (96)
To prove (95) , we first establish pointwise bounds for the multiplier u ( ξ , ξ n + 1 ) u(\xi,\xi_{n+1}) u ( ξ , ξ n + 1 ) :
∣ u ( ξ , ξ n + 1 ) ∣ ≤ C D ∣ ( ξ , ξ n + 1 ) ∣ , ∣ u ( ξ , ξ n + 1 ) ∣ ≤ 2 e − ∣ ( ξ , ξ n + 1 ) ∣ , (97) |u(\xi,\xi_{n+1})|\le CD|(\xi,\xi_{n+1})|,\qquad
|u(\xi,\xi_{n+1})|\le2e^{-|(\xi,\xi_{n+1})|}, \tag*{(97)} ∣ u ( ξ , ξ n + 1 ) ∣ ≤ C D ∣ ( ξ , ξ n + 1 ) ∣ , ∣ u ( ξ , ξ n + 1 ) ∣ ≤ 2 e − ∣ ( ξ , ξ n + 1 ) ∣ , (97)
where u ( ξ , ξ n + 1 ) = e − ∣ ( ξ , ξ n + 1 ) ∣ ( e − i D ξ n + 1 − 1 ) u(\xi,\xi_{n+1})=e^{-|(\xi,\xi_{n+1})|}(e^{-iD\xi_{n+1}}-1) u ( ξ , ξ n + 1 ) = e − ∣ ( ξ , ξ n + 1 ) ∣ ( e − i D ξ n + 1 − 1 ) . The first inequality follows from ∣ e − i D ξ n + 1 − 1 ∣ ≤ D ∣ ξ n + 1 ∣ |e^{-iD\xi_{n+1}}-1|\le D|\xi_{n+1}| ∣ e − i D ξ n + 1 − 1∣ ≤ D ∣ ξ n + 1 ∣ , the second is obvious. Using Plancherel’s theorem,
∥ U 2 j P k − j f ∥ L 2 ( ℓ j 2 ) = ∥ u ( 2 j ⋅ ) ψ ( 2 j − k ⋅ ) f ^ ( ⋅ ) ∥ L 2 ( ℓ j 2 ) ≤ sup η ∈ R n + 1 ∥ u ( 2 j η ) ψ ( 2 j − k η ) ∥ ℓ j 2 ∥ f ∥ L 2 . (98) \begin{aligned}
\|U_{2^j}P_{k-j}f\|_{L^2(\ell^2_j)}
=\|u(2^j\cdot)\psi(2^{j-k}\cdot)\widehat{f}(\cdot)\|_{L^2(\ell^2_j)} \\
&\le\sup_{\eta\in\mathbb{R}^{n+1}}\|u(2^j\eta)\psi(2^{j-k}\eta)\|_{\ell^2_j}\|f\|_{L^2}. \tag*{(98)}
\end{aligned} ∥ U 2 j P k − j f ∥ L 2 ( ℓ j 2 ) = ∥ u ( 2 j ⋅ ) ψ ( 2 j − k ⋅ ) f ( ⋅ ) ∥ L 2 ( ℓ j 2 ) ≤ η ∈ R n + 1 sup ∥ u ( 2 j η ) ψ ( 2 j − k η ) ∥ ℓ j 2 ∥ f ∥ L 2 . (98)
Fix η ∈ R n + 1 \eta\in\mathbb{R}^{n+1} η ∈ R n + 1 and k ∈ Z k\in\mathbb{Z} k ∈ Z . From (97) we obtain
∥ u ( 2 j η ) ψ ( 2 j − k η ) ∥ ℓ j 2 ≤ C min { 2 k D , e − 2 k } . (99) \|u(2^j\eta)\psi(2^{j-k}\eta)\|_{\ell^2_j}
\le C\min\{2^kD,e^{-2^k}\}. \tag*{(99)} ∥ u ( 2 j η ) ψ ( 2 j − k η ) ∥ ℓ j 2 ≤ C min { 2 k D , e − 2 k } . (99)
Let k 0 k_0 k 0 be such that 2 − k 0 ≈ D 2^{-k_0}\approx D 2 − k 0 ≈ D . Then 2 k D ≲ 2 k − k 0 2^kD\lesssim2^{k-k_0} 2 k D ≲ 2 k − k 0 for k ≤ k 0 k\le k_0 k ≤ k 0 , and using e − x ≤ C ϵ x − ϵ e^{-x}\le C_\epsilon x^{-\epsilon} e − x ≤ C ϵ x − ϵ for x > 0 x>0 x > 0 we have e − 2 k ≤ C ϵ 2 − ( k − k 0 ) ϵ D ϵ e^{-2^k}\le C_\epsilon2^{-(k-k_0)\epsilon}D^\epsilon e − 2 k ≤ C ϵ 2 − ( k − k 0 ) ϵ D ϵ . Consequently,
∥ u ( 2 j η ) ψ ( 2 j − k η ) ∥ ℓ j 2 ≤ C ϵ 2 − ϵ ∣ k − k 0 ∣ . (100) \|u(2^j\eta)\psi(2^{j-k}\eta)\|_{\ell^2_j}
\le C_\epsilon2^{-\epsilon|k-k_0|}. \tag*{(100)} ∥ u ( 2 j η ) ψ ( 2 j − k η ) ∥ ℓ j 2 ≤ C ϵ 2 − ϵ ∣ k − k 0 ∣ . (100)
Combining (98) and (100) yields (95) .
Estimate (96) can be proved by following exactly the same argument used for (12) in § 5.1 ; we omit the details.
To obtain the desired bounds for S r ( U t f ) S_r(U_tf) S r ( U t f ) , we shall use interpolation between the estimates collected in the following lemmas.
Lemma 6.1 For k ∈ Z k \in\mathbb{Z} k ∈ Z , we have
∥ V 2 j ( U k f ) ∥ L 2 ( ℓ j 2 ) ≤ C ( 1 + D 1 / 2 ) min { 2 k D e − 2 k , 2 k 2 e − 2 k } ∥ f ∥ L 2 . (101) \left\|V_2^j(U_kf)\right\|_{L^2(\ell_j^2)} \le C(1+D^{1/2})\min\left\{2^kDe^{-2^k},\,2^{\frac{k}{2}}e^{-2^k}\right\}\|f\|_{L^2}. \tag*{(101)} V 2 j ( U k f ) L 2 ( ℓ j 2 ) ≤ C ( 1 + D 1/2 ) min { 2 k D e − 2 k , 2 2 k e − 2 k } ∥ f ∥ L 2 . (101)
Lemma 6.2 For k ∈ Z k \in\mathbb{Z} k ∈ Z and arbitrary positive number ϵ \epsilon ϵ , there exists a constant C ϵ C_\epsilon C ϵ such that
∥ V 2 j ( U k f ) ∥ L 1 ( ℓ j 2 ) ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k + ϵ , 2 − k ϵ } ∥ f ∥ H 1 . (102) \left\|V_2^j(U_kf)\right\|_{L^1(\ell_j^2)} \le C_\epsilon(1+D^{1+\epsilon})\max\left\{2^{k+\epsilon},2^{-k\epsilon}\right\}\|f\|_{H^1}. \tag*{(102)} V 2 j ( U k f ) L 1 ( ℓ j 2 ) ≤ C ϵ ( 1 + D 1 + ϵ ) max { 2 k + ϵ , 2 − k ϵ } ∥ f ∥ H 1 . (102)
Lemma 6.3 For k ∈ Z k \in\mathbb{Z} k ∈ Z and arbitrary positive number ϵ \epsilon ϵ , there exists a constant C ϵ C_\epsilon C ϵ such that
∥ V 2 j ( U k f ) ∥ L p ( ℓ j 2 ) ≤ C ϵ ( 1 + D 1 / 2 ) 2 k ϵ ∥ f ∥ L p , (103) \left\|V_2^j(U_kf)\right\|_{L^p(\ell_j^2)} \le C_\epsilon(1+D^{1/2})2^{k\epsilon}\|f\|_{L^p}, \tag*{(103)} V 2 j ( U k f ) L p ( ℓ j 2 ) ≤ C ϵ ( 1 + D 1/2 ) 2 k ϵ ∥ f ∥ L p , (103) for p > 2 p>2 p > 2 .
Now we can prove (6.92), assuming the validity of the lemmas above.
Proof of (6.92): Denote k 0 k_0 k 0 such that 2 − k 0 ≈ D 2^{-k_0}\approx D 2 − k 0 ≈ D . Interpolating (6.101) and (6.102), we obtain
∥ V 2 j ( U k f ) ∥ L p ( ℓ j 2 ) ≤ { C ϵ ( 1 + D 1 / p + ϵ ) ( 2 k D e − 2 k ) θ ( 2 − k ϵ ) 1 − θ ∥ f ∥ L p , k ≤ k 0 , C ϵ ( 1 + D 1 / p + ϵ ) ( 2 k 2 e − 2 k ) θ ( 2 − k ϵ ) 1 − θ ∥ f ∥ L p , k 0 < k ≤ 0 , C ϵ ( 1 + D 1 / p + ϵ ) ( 2 k 2 e − 2 k ) θ ( 2 k + ϵ ) 1 − θ ∥ f ∥ L p , k > 0 , \left\|V_2^j(U_kf)\right\|_{L^p(\ell_j^2)}
\le
\begin{cases}
C_\epsilon(1+D^{1/p+\epsilon})(2^kDe^{-2^k})^\theta(2^{-k\epsilon})^{1-\theta}\|f\|_{L^p}, & k\le k_0,\\
C_\epsilon(1+D^{1/p+\epsilon})(2^{\frac{k}{2}}e^{-2^k})^\theta(2^{-k\epsilon})^{1-\theta}\|f\|_{L^p}, & k_0<k\le0,\\
C_\epsilon(1+D^{1/p+\epsilon})(2^{\frac{k}{2}}e^{-2^k})^\theta(2^{k+\epsilon})^{1-\theta}\|f\|_{L^p}, & k>0,
\end{cases} V 2 j ( U k f ) L p ( ℓ j 2 ) ≤ ⎩ ⎨ ⎧ C ϵ ( 1 + D 1/ p + ϵ ) ( 2 k D e − 2 k ) θ ( 2 − k ϵ ) 1 − θ ∥ f ∥ L p , C ϵ ( 1 + D 1/ p + ϵ ) ( 2 2 k e − 2 k ) θ ( 2 − k ϵ ) 1 − θ ∥ f ∥ L p , C ϵ ( 1 + D 1/ p + ϵ ) ( 2 2 k e − 2 k ) θ ( 2 k + ϵ ) 1 − θ ∥ f ∥ L p , k ≤ k 0 , k 0 < k ≤ 0 , k > 0 ,
for 1 < p ≤ 2 1<p\le2 1 < p ≤ 2 , where θ \theta θ satisfies 1 p = 1 − θ 1 + θ 2 \frac{1}{p}=\frac{1-\theta}{1}+\frac{\theta}{2} p 1 = 1 1 − θ + 2 θ . A straightforward computation shows that
∑ k ≤ k 0 ( 2 k D e − 2 k ) θ ( 2 − k ϵ ) 1 − θ ≲ D θ 2 k 0 ( θ − ϵ ( 1 − θ ) ) ≲ D − θ + 2 ϵ ( 1 − θ ) ≤ D ϵ , \sum_{k\le k_0}(2^kDe^{-2^k})^\theta(2^{-k\epsilon})^{1-\theta}\lesssim D^{\theta2^{k_0(\theta-\epsilon(1-\theta))}}\lesssim D^{-\theta+2\epsilon(1-\theta)}\le D^\epsilon, k ≤ k 0 ∑ ( 2 k D e − 2 k ) θ ( 2 − k ϵ ) 1 − θ ≲ D θ 2 k 0 ( θ − ϵ ( 1 − θ )) ≲ D − θ + 2 ϵ ( 1 − θ ) ≤ D ϵ ,
∑ k 0 < k ≤ 0 ( 2 k 2 e − 2 k ) θ ( 2 − k ϵ ) 1 − θ ≲ ∣ k 0 ∣ = log D log 2 ≲ ϵ D ϵ , \sum_{k_0<k\le0}(2^{\frac{k}{2}}e^{-2^k})^\theta(2^{-k\epsilon})^{1-\theta}\lesssim|k_0|=\frac{\log D}{\log2}\lesssim_\epsilon D^\epsilon, k 0 < k ≤ 0 ∑ ( 2 2 k e − 2 k ) θ ( 2 − k ϵ ) 1 − θ ≲ ∣ k 0 ∣ = log 2 log D ≲ ϵ D ϵ ,
and
∑ k > 0 ( 2 k 2 e − 2 k ) θ ( 2 k + ϵ ) 1 − θ ≲ 2 ϵ ( 1 − θ ) ≲ D ϵ . \sum_{k>0}(2^{\frac{k}{2}}e^{-2^k})^\theta(2^{k+\epsilon})^{1-\theta}\lesssim2^{\epsilon(1-\theta)}\lesssim D^\epsilon. k > 0 ∑ ( 2 2 k e − 2 k ) θ ( 2 k + ϵ ) 1 − θ ≲ 2 ϵ ( 1 − θ ) ≲ D ϵ .
Hence (6.92) holds for 1 < p ≤ 2 1<p\le2 1 < p ≤ 2 . Interpolating (6.101) and (6.103) yields the estimate for p > 2 p>2 p > 2 . This completes the proof of (6.92).
Using the same method as in the proof of Lemma 3.2, we obtain inequality (6.2). We now turn to the proofs of Lemma 6.1 and Lemma 6.3. Applying Proposition 2.2 and Hölder’s inequality,
∥ V 2 j ( U k f ) ∥ L 2 ( ℓ j 2 ) ≤ C ∥ U k , t f ∥ L 2 ( L t 2 ) 1 / 2 ∥ d d t U k , t f ∥ L 2 ( L t 2 ) 1 / 2 . (104) \left\|V_2^j(U_kf)\right\|_{L^2(\ell_j^2)}
\le C\left\|U_{k,t}f\right\|_{L^2(L_t^2)}^{1/2}
\left\|\frac{d}{dt}U_{k,t}f\right\|_{L^2(L_t^2)}^{1/2}. \tag*{(104)} V 2 j ( U k f ) L 2 ( ℓ j 2 ) ≤ C ∥ U k , t f ∥ L 2 ( L t 2 ) 1/2 d t d U k , t f L 2 ( L t 2 ) 1/2 . (104)
By Plancherel’s theorem,
∥ V 2 j ( U k f ) ∥ L 2 ( ℓ j 2 ) ≤ C ∥ u k ( t ⋅ ) f ^ ( ⋅ ) ∥ L 2 ( L t 2 ) 1 / 2 ∥ ( t ⋅ ) ∇ u k ( t ) f ^ ( ⋅ ) ∥ L 2 ( L t 2 ) 1 / 2 ≤ C sup ∣ η ∣ = 1 ∥ u k ( t η ) ∥ L t 2 1 / 2 sup ∣ η ∣ = 1 ∥ ( t η ) ∇ u k ( t η ) ∥ L t 2 1 / 2 ∥ f ∥ L 2 . (105) \begin{aligned}
\left\|V_2^j(U_kf)\right\|_{L^2(\ell_j^2)}
\le C\left\|u_k(t\cdot)\widehat{f}(\cdot)\right\|_{L^2(L_t^2)}^{1/2}
\left\|(t\cdot)\nabla u_k(t)\widehat{f}(\cdot)\right\|_{L^2(L_t^2)}^{1/2} \\
&\le C\sup_{|\eta|=1}\left\|u_k(t\eta)\right\|_{L_t^2}^{1/2}
\sup_{|\eta|=1}\left\|(t\eta)\nabla u_k(t\eta)\right\|_{L_t^2}^{1/2}\|f\|_{L^2}. \tag*{(105)}
\end{aligned} V 2 j ( U k f ) L 2 ( ℓ j 2 ) ≤ C u k ( t ⋅ ) f ( ⋅ ) L 2 ( L t 2 ) 1/2 ( t ⋅ ) ∇ u k ( t ) f ( ⋅ ) L 2 ( L t 2 ) 1/2 ≤ C ∣ η ∣ = 1 sup ∥ u k ( t η ) ∥ L t 2 1/2 ∣ η ∣ = 1 sup ∥ ( t η ) ∇ u k ( t η ) ∥ L t 2 1/2 ∥ f ∥ L 2 . (105)
where u k ( ⋅ ) = u ( ⋅ ) ψ ( 2 − k ⋅ ) u_k(\cdot)=u(\cdot)\psi(2^{-k}\cdot) u k ( ⋅ ) = u ( ⋅ ) ψ ( 2 − k ⋅ ) . Thus it suffices to estimate ∥ u k ∥ L ∞ \|u_k\|_{L^\infty} ∥ u k ∥ L ∞ and ∥ ( ⋅ ) ∇ u k ( ⋅ ) ∥ L ∞ \|(\cdot)\nabla u_k(\cdot)\|_{L^\infty} ∥ ( ⋅ ) ∇ u k ( ⋅ ) ∥ L ∞ , since both are supported where ∣ ⋅ ∣ ∼ 2 k |\cdot|\sim2^k ∣ ⋅ ∣ ∼ 2 k .
From (97) we obtain
∥ u k ∥ L ∞ ≤ C min { 2 k D , e − 2 k } . (106) \|u_k\|_{L^\infty}\le C\min\{2^kD,e^{-2^k}\}. \tag*{(106)} ∥ u k ∥ L ∞ ≤ C min { 2 k D , e − 2 k } . (106)
Rewrite u ( ξ , ξ n + 1 ) = ( e − i D ξ n + 1 − 1 ) e − ∣ ( ξ , ξ n + 1 ) ∣ u(\xi,\xi_{n+1})=(e^{-iD\xi_{n+1}}-1)e^{-|(\xi,\xi_{n+1})|} u ( ξ , ξ n + 1 ) = ( e − i D ξ n + 1 − 1 ) e − ∣ ( ξ , ξ n + 1 ) ∣ . Then
∣ ( ξ , ξ n + 1 ) ⋅ ∇ u ( ξ , ξ n + 1 ) ∣ ≤ C ( 1 + D ) ∣ ( ξ , ξ n + 1 ) ∣ e − ∣ ( ξ , ξ n + 1 ) ∣ . (107) |(\xi,\xi_{n+1})\cdot\nabla u(\xi,\xi_{n+1})|\le C(1+D)|(\xi,\xi_{n+1})|e^{-|(\xi,\xi_{n+1})|}. \tag*{(107)} ∣ ( ξ , ξ n + 1 ) ⋅ ∇ u ( ξ , ξ n + 1 ) ∣ ≤ C ( 1 + D ) ∣ ( ξ , ξ n + 1 ) ∣ e − ∣ ( ξ , ξ n + 1 ) ∣ . (107)
Combining (106) and (107) gives
∥ u k ( t ⋅ ) ∥ L t 2 ≤ C min { 2 k D , e − 2 k } , ∥ ( t ⋅ ) ( ∇ u k ) ( t ⋅ ) ∥ L t 2 ≤ C ( 1 + D ) 2 k e − 2 k . (108) \|u_k(t\cdot)\|_{L_t^2}\le C\min\{2^kD,e^{-2^k}\},\qquad\|(t\cdot)(\nabla u_k)(t\cdot)\|_{L_t^2}\le C(1+D)2^ke^{-2^k}. \tag*{(108)} ∥ u k ( t ⋅ ) ∥ L t 2 ≤ C min { 2 k D , e − 2 k } , ∥ ( t ⋅ ) ( ∇ u k ) ( t ⋅ ) ∥ L t 2 ≤ C ( 1 + D ) 2 k e − 2 k . (108)
Inserting these bounds into (105) yields (101) , completing the proof of Lemma 6.1.
In order to prove Lemma 6.3, we use inequality (2.5) to obtain
∥ V j 2 ( U k f ) ∥ L t 2 ≤ C ∥ U k , t f ∥ L t 2 1 / 2 ∥ t d d t U k , t f ∥ L t 2 1 / 2 ≤ C ( 1 + D 1 / 2 ) ∥ U k , t f ∥ L t 2 + ( 1 + D 1 / 2 ) − 1 ∥ t d d t U k , t f ∥ L t 2 . \begin{aligned}
\|V_j^2(U_kf)\|_{L_{t}^{2}} \le C\|U_{k,t}f\|_{L_t^2}^{1/2}\left\|t\frac{d}{dt}U_{k,t}f\right\|_{L_t^2}^{1/2} \\
&\le C(1+D^{1/2})\|U_{k,t}f\|_{L_t^2}+(1+D^{1/2})^{-1}\left\|t\frac{d}{dt}U_{k,t}f\right\|_{L_t^2}.
\end{aligned} ∥ V j 2 ( U k f ) ∥ L t 2 ≤ C ∥ U k , t f ∥ L t 2 1/2 t d t d U k , t f L t 2 1/2 ≤ C ( 1 + D 1/2 ) ∥ U k , t f ∥ L t 2 + ( 1 + D 1/2 ) − 1 t d t d U k , t f L t 2 .
Recall that the kernel of U k , t U_{k,t} U k , t is U k , t U_{k,t} U k , t . Set U ~ k , t ( x ˉ ) = x ˉ ⋅ ∇ U k , t ( x ˉ ) \widetilde U_{k,t}(\bar{x})=\bar{x}\cdot\nabla U_{k,t}(\bar{x}) U k , t ( x ˉ ) = x ˉ ⋅ ∇ U k , t ( x ˉ ) ; then t d d t U k , t = − ( n + 1 ) U k , t − U ~ k , t t\frac{d}{dt}U_{k,t}=-(n+1)U_{k,t}-\widetilde U_{k,t} t d t d U k , t = − ( n + 1 ) U k , t − U k , t . As in the proof of Lemma 3.2, it is enough to prove
∥ U k , t ∗ f ∥ L p ( L t 2 ) ≤ C 2 k ε ∥ f ∥ L p , (109) \|U_{k,t}*f\|_{L^p(L_t^2)}\le C2^{k\varepsilon}\|f\|_{L^p}, \tag*{(109)} ∥ U k , t ∗ f ∥ L p ( L t 2 ) ≤ C 2 k ε ∥ f ∥ L p , (109)
for p > 2 p>2 p > 2 .
By Minkowski’s inequality we have ∥ U k ∥ L 1 ≤ 1 \|U_k\|_{L^1}\le1 ∥ U k ∥ L 1 ≤ 1 . Thus, applying Hölder’s inequality in t t t ,
( ∫ R n + 1 ( ∫ 1 2 ∣ U k , t f ( x ˉ ) ∣ 2 d t ) p / 2 d x ˉ ) 1 / p ≤ ( ∫ R n + 1 ∫ 1 2 ∣ U k , t f ( x ˉ ) ∣ p d t d x ˉ ) 1 / p = ( ∫ 1 2 ∫ R n + 1 ∣ U k , t f ( x ˉ ) ∣ p d x ˉ d t ) 1 / p = ( ∫ 1 2 ∫ R n + 1 ∣ U k , t ∗ f ( x ˉ − t D ) ∣ p d x ˉ d t ) 1 / p ≤ C ∥ f ∥ L p . \begin{aligned}
\left(\int_{\mathbb{R}^{n+1}}\left(\int_1^2|U_{k,t}f(\bar{x})|^2dt\right)^{p/2}d\bar{x}\right)^{1/p}
\le\left(\int_{\mathbb{R}^{n+1}}\int_1^2|U_{k,t}f(\bar{x})|^pdt\,d\bar{x}\right)^{1/p} \\
&=\left(\int_1^2\int_{\mathbb{R}^{n+1}}|U_{k,t}f(\bar{x})|^p\,d\bar{x}\,dt\right)^{1/p} \\
&=\left(\int_1^2\int_{\mathbb{R}^{n+1}}|U_{k,t}*f(\bar{x}-tD)|^p\,d\bar{x}\,dt\right)^{1/p} \\
&\le C\|f\|_{L^p}.
\end{aligned} ( ∫ R n + 1 ( ∫ 1 2 ∣ U k , t f ( x ˉ ) ∣ 2 d t ) p /2 d x ˉ ) 1/ p ≤ ( ∫ R n + 1 ∫ 1 2 ∣ U k , t f ( x ˉ ) ∣ p d t d x ˉ ) 1/ p = ( ∫ 1 2 ∫ R n + 1 ∣ U k , t f ( x ˉ ) ∣ p d x ˉ d t ) 1/ p = ( ∫ 1 2 ∫ R n + 1 ∣ U k , t ∗ f ( x ˉ − t D ) ∣ p d x ˉ d t ) 1/ p ≤ C ∥ f ∥ L p .
Then (109) follows by applying Proposition 3.5 with B = O ( 2 k ( n + 2 ) ) B=O(2^{k(n+2)}) B = O ( 2 k ( n + 2 ) ) and A A A being the constant just obtained.
References