New Proof of Hörmander multiplier Theorem on compact manifolds without boundary Xiangjin Xu Department of Mathematics Johns Hopkins University Baltimore, MD, 21218, USA xxu@math.jhu.edu Abstract On compact manifolds (M, g) without boundary of dimension n ≥ 2, the gradient estimates for unit band spectral projection operators χλ is proved for any second order elliptic differential operators L by maximum principle. A new proof of Hörmander Multiplier Theorem on the eigenfunction expansion of the operator L is given in this setting by using the gradient estimates and the Calderón-Zygmund argument. Keywords: gradient estimate, spectral projection operator, Hörmander Multiplier Theorem, Calderón-Zygmund decomposition Mathematics Subject Classification 2000: 58J40, 35P20, 35J25, 1 1 Introduction Let (M, g) be a compact boundaryless manifolds (M, g) of dimension n ≥ 2, and suppose that L is a second order elliptic differential operator which is positive and self-adjoint with respect to the C ∞ density dx associated to the Reimannian metric g. The purpose of this paper is to give a simple proof of sharp gradient estimates for the eigenfunctions of L and then to use these estimates to give a new proof of Hörmander multiplier theorem for the eigenfunction expansions on the compact boundaryless manifold (M, g). Thus, we shall consider the eigenvalue problem (L + λ2 )u(x) = 0, x ∈ M, (1) Recall that the spectrum of L is discrete and tends to infinity. Let 0 ≤ λ21 ≤ λ22 ≤ λ23 ≤ · · · denote the eigenvalues, so that {λj } is the spectrum of the first order √ operator P = −L. Let {ej (x)} ⊂ L2 (M ) be an associated real orthogonormal basis, define the unit band spectral projection operators, χλ f = X ej (f ), where ej (f )(x) = ej (x) λj ∈[λ,λ+1) Z M f (y)ej (y)dy. In [5] and [6], Sogge proved the following L∞ estimates on χλ , ||χλ f ||∞ ≤ Cλ(n−1)/2 ||f ||2 , λ ≥ 1, for a uniform constant C, which is equivalent to X ej (x)2 ≤ Cλn−1 , ∀x ∈ M, λj ∈[λ,λ+1) Here based on the above L∞ estimates on χλ f , by maximum principle, we show the following gradient estimates for χλ f 2 Theorem 1.1 Fix a compact boundaryless Riemannian manifold (M, g), there is a uniform constant C so that ||∇χλ f ||∞ ≤ Cλ(n+1)/2 ||f ||2 , λ ≥ 1, which is equivalent to X |∇ej (x)|2 ≤ Cλn+1 , ∀x ∈ M. λj ∈[λ,λ+1) We shall show the gradient estimates following the ideas of the interior gradient estimates for Poisson’s equation in [1], where for Poisson’s equation ∆u = f , there are gradient estimates for the interior point x0 as |∇u(x0 )| ≤ C sup |u| + Cd sup |f |, d ∂B B by using maximum principle in a cube B centered at x0 with length d. Our other main result is Hörmander Multiplier Theorem for the eigenfunction expansions on compact manifolds without boundary. Given a bounded function m(λ) ∈ L∞ (R), one can define the multiplier operator, m(P ), by m(P )f = ∞ X m(λj )ej (f ) (2) j=1 such an operator is always bounded on L2 (M ). However, if one considers on any other space Lp (M ), it is known that some smoothness assumptions on the function m(λ) are needed to ensure the boundedness of m(P ) : Lp (M ) → Lp (M ). When m(λ) is C ∞ and, moreover, in the symbol class S 0 , i.e., |( d α ) m(λ)| ≤ Cα (1 + |λ|)−α , dλ 3 α = 0, 1, 2, · · · (3) It has been known for some time that (3) holds for all 1 < p < ∞ on compact manifolds (see [10]). One assumes the following regularity assumption: Suppose that m ∈ L∞ (R), let L2s (R) denote the usual Sobolev space and fix β ∈ C0∞ ((1/2, 2)) satisfying P∞ −∞ β(2j t) = 1, t > 0, for s > n/2, there is sup λ−1+s ||β(·/λ)m(·)||2L2s = sup ||β(·)m(λ·)||2L2s < ∞. λ>0 (4) λ>0 One can see this is the sharp assumption to ensure the boundedness of m(P ) on Lp (M ) for all 1 < p < ∞, since for a special class of multiplier operators, Riesz means Sλδ f = P λj ≤λ (1 − λ2j /λ2 )δ ej (f ), such an assumption is sharp (see in [6]). Many authors have studied this problem under different setting. Hörmander [2] first proved the boundedness of m(P ) for Rn under the assumption (4), using the Calderón-Zygmund decomposition and the estimates on the integral kernel of the multiplier operator. Stein and Weiss [9] studied the boundedness of m(P ) for multiple Fourier series, which can be regarded as the case on the flat torus Tn . By studying the paramatrix of the wave kernel of m(P ), Seeger and Sogge [4] and Sogge [6] proved the following Hömander multiplier Theorem for the eigenfunction expansions on compact manifolds without boundary: Theorem 1.2 Let m ∈ L∞ (R) satisfy (4), then there are constants Cp such that ||m(P )f ||Lp (M ) ≤ Cp ||f ||Lp (M ) , 1 < p < ∞. (5) Here we shall give a new proof for above theorem by using the L∞ estimates and the gradient estimates on χλ f , without using the paramatrix of the wave kernel of m(P ) as done in [4] and [6]. Since the complex conjugate of m satisfies the same hypotheses (4), we need only to prove Theorem 1.2 for exponents 1 < p ≤ 2. This 4 will allow us to exploit orthogonality, and since m(P ) is bounded on L2 (M ), also reduce Theorem 1.2 to show that m(P ) is weak-type (1, 1) by the Marcinkiewicz Interpolation Theorem. The weak-type (1,1) estimates of m(P ) will involve a splitting of m(P ) into two pieces m(P ) = m̃(P ) + r(P ). For the remainder r(P ), one can obtain the strong (1,1) estimates by the L∞ estimates on χλ f as done in [6]. For the main term m̃(P ), in [4] and [6], the authors used the paramatrix of the wave kernel of m̃(P ) to get the required estimates on the integral kernel of m̃(P ), and applying these estimates, proved the weak-type (1,1) estimates on m̃(P ). As people know, the the paramatrix construction of the wave equation does not work well for general compact manifolds with boundary unless one assume the boundary is geodesically concave. Here we use another approach to prove the weak-type (1,1) estimates on m̃(P ), which works for any compact manifolds with boundary (One may see it from [11] and [12] for the proof of Hörmander Multiplier Theorem of Dirichlet Laplacian and Neumann Laplacian on compact manifolds with boundary). We make a second decomposition {Kλ,l (x, y)}∞ l=−∞ for each term Kλ (x, y), which comes from the dyadic decomposition of the integral kernel K(x, y) of m̃(P ) K(x, y) = ∞ X K2k (x, y) + K0 (x, y), (∗) k=1 such that Kλ (x, y) = m̃(P )f (x) = P∞ ∞ X ∞ X l=−∞ Kλ,l (x, y) and T2k ,l (P )f (x), where Tλ,l (P )f (x) = k=0 l=−∞ Z M Kλ,l (x, y)f (y)dy. We shall give the definitions of Kλ,l (x, y) in the proof of Theorem 1.2. And for each operator Tλ,l (P ), we shall prove the L1 → L2 estimates by a rescaling argument of the proof for the remainder r(P ) using the L∞ estimates and gradient estimates on 5 χλ f which we obtained in Theorem 1.1. With the support properties and the finite propagation speed properties, one has the relation (10) between λ and l, which is one key observation when we apply the Calderón-Zygmund decomposition to show the weak-type (1,1) estimates on m̃(P ). In what follows we shall use the convention that C will denote a constant that is not necessarily the same at each occurrence. Acknowledgement: The author would like to thank his advisor, Professor C.D. Sogge, brings the problems to him and a number helpful conversations on these problems and his research. The results in this paper come from part of the author’s Ph.D. thesis [12] in Johns Hopkins University. 2 Gradient estimates In this section, we shall prove Theorem 1.1. by using maximum principle. Proof of Theorem 1.1. Now we fix x0 ∈ M . We shall use the maximum principle in the cube centered at x0 with length d = (λ + 1)−1 to prove the same gradient estimates for χλ f as for Poisson’s equation. Define the geodesic coordinates x = (x1 , · · · , xn ) centered at point x0 as following, fixed an orthogonormal basis {vi }ni=1 ⊂ Tx0 M , identity x = (x1 , · · · , xn ) ∈ Rn with Pn the point exp( i=1 xi vi ) ∈ M . In this coordinate, the elliptic operator L can be written as L= n X i,j=1 aij (x) n X ∂2 ∂ + bi (x) + c(x), ∂xi ∂xj i=1 ∂xi where aij (x), bi (x), c(x) ∈ C ∞ (M ), and c(x) ≤ 0 which comes from L is an elliptic 6 operator. Now define the cube Q = {x = (x1 , · · · , xn ) ∈ Rn | |xi | < d, i = 1, · · · , n} ⊂ M. Denote u(x; f ) = χλ f (x), we have u ∈ C 2 (Q), and λ2j ej (f ) := h(x; f ). X Lu(x; f ) = − λj ∈[λ,λ+1) From the L∞ estimate in [5], and Cauchy-Schwarz inequality, we have |h(x; f )|2 = ( Z (λ2j ej (x))( X λj ∈[λ,λ+1) ≤ λ4j e2j (x) X ej (y)f (y)dy))2 X ≤ (λ + 1) ( X Z ( λj ∈[λ,λ+1) λj ∈[λ,λ+1) 4 M M ej (y)f (y)dy)2 e2j (x))||χλ f ||2L2 (M ) λj ∈[λ,λ+1) ≤ C(λ + 1) n+3 ||χλ f ||2L2 (M ) Here we estimate |Dn u(0; f )| = | ∂x∂n u(0; f )| only, and the same estimate holds for |Di u(0; f )| with i = 1, · · · , n − 1 also. Now in the half-cube Q0 = {x = (x1 , · · · , xn ) ∈ Rn | |xi | < d, i = 1, · · · , n − 1, 0 < xn < d.} ⊂ M, Consider the function 1 ϕ(x0 , xn ; f ) = [u(x0 , xn ; f ) − u(x0 , −xn ; f )], 2 where we write x = (x0 , xn ) = (x1 , · · · , xn−1 , xn ). One sees that ϕ(x0 , 0; f ) = 0, sup∂Q0 |ϕ| ≤ sup∂Q |u| := A, and |Lϕ| ≤ supQ |h| := N in Q0 . Now consider the function ψ(x0 , xn ) = A 02 [|x | + αxn (nd − (n − 1)xn )] + βN xn (d − xn ) ≥ 0 d2 7 defined on the half-cube Q0 , where α ≥ 1 and β ≥ 1 will be determined below. Obviously ψ(x0 , xn ) ≥ 0 on xn = 0 and ψ(x0 , xn ) ≥ A in the remaining portion of ∂Q0 . Lψ(x) = n X A [2tr(a (x)) − (2nα − 2α + 1)a (x) + 2 bi (x)xi + bn (x)(nαd ij nn d2 i=1 −(2nα − 2α + 1)xn )] + N β[−2ann (x) + bn (x)(d − 2xn )] + c(x)ψ(x). Since in M , tr(aij (x)) and bi (x) are bounded uniformly, c(x) ≤ 0 and ann (x) is positive, then for a large α, we can make 2tr(aij (x)) − (2nα − 2α + 1)ann (x) ≤ −1, Fix such a α, since d = (λ + 1)−1 , for large λ, we have 2 n X bi (x)xi + bn (x)(nαd − (2nα − 2α + 1)xn ) < 1. i=1 Then the first term is negative. For second term, let β large enough, we have β[−2ann (x) + bn (x)(d − 2xn )] < −1. For the third term, we have c(x) ≤ 0 and ψ(x) ≥ 0 in Q0 . Hence we have Lψ(x) ≤ −N in Q0 . Now we have L(ψ ± ϕ) ≤ 0 in Q0 and ψ ± ϕ ≥ 0 on ∂Q0 , from which it follows by the maximum principle that |ϕ(x0 , xn ; f )| ≤ |ψ(x0 , xn )| in Q0 . Letting x0 = 0 in the expressions for ψ and ϕ, then dividing by xn and letting xn tend to zero, we obtain |Dn u(0; f )| = lim | xn →0 αnA ϕ(0, xn ; f ) |≤ + βdN. xn d Note that d = (λ + 1)−1 , A ≤ C(λ + 1)(n−1)/2 , and N ≤ (λ + 1)(n+3)/2 , then we have the estimate |Dn u(0; f )| ≤ C(λ + 1)(n+1)/2 . 8 The same estimate holds for |Di u(0)|, i = 1, · · · , n − 1. Hence we have |∇u(0)| ≤ C(λ + 1)(n+1)/2 . Since the estimate is for any x0 ∈ M , Theorem 1.1 is proved. Q.E.D. For Riemannian manifolds without boundary, in [8], the authors proved that for (n−1)/2 generic metrics on any manifold one has the bounds ||ej ||L∞ (M ) = o(λj ) for L2 normalized eigenfunctions. For Laplace-Beltrami operator ∆g , there are eigenvalues {−λ2j }, where 0 ≤ λ20 ≤ λ21 ≤ · · · → ∞ are counted with multiplicity. Let {ej (x)} be an associated orthogonal basis of L2 normalized eigenfunctions. If λ2 is in the spectrum of −∆g , let Vλ = {u | ∆g u = −λ2 u} denote the corresponding eigenspace. We define the eigenfunction growth rate in term of L∞ (λ, g) = ||u||L∞ , sup u∈Vλ ;||u||L2 =1 and the gradient growth rate in term of L∞ (∇, λ, g) = sup ||∇u||L∞ . u∈Vλ ;||u||L2 =1 In [8], Sogge and Zelditch proved the following results (n−1)/2 L∞ (λ, g) = o(λj ) for a generic metric on any manifold without boundary. Here we have the following estimates on the gradient growth rate (n+1)/2 Theorem 2.1 L∞ (∇, λ, g) = o(λj ) for a generic metric on any manifold with- out boundary. And the bounds are uniform if there is a uniformly bound on the norm of tr(g ij (x)) for (M, g). 9 Proof. For a compact Riemannian manifold (M, g) without boundary, we can apply Theorem 1.1 to any point in M . From Theorem 1.4 in [8], we have (n−1)/2 L∞ (λ, g) = o(λj ) for a generic metric on any compact manifold without boundary. Fix any such a metric on the manifold M and a L2 normalized eigenfunction u(x), apply Theorem 1.1 to u(x) at each point x0 ∈ M , we have |∇u(x0 )| ≤ αnA + βdN. d where α and β are constants depending on the norm of tr(g ij (x)) at M only, which (n−1)/2 can been seen in the above proof of Theorem 1.1, and A = sup∂Q |u| = o(λj (n+3)/2 and N = supQ |λ2 u| = (λj ), ), where the cube Q = {x = (x1 , · · · , xn ) ∈ Rn | |xi | < d, i = 1, · · · , n} ⊂ M, √ we choose d = (λ + 1)−1 / n. Hence we have (n+1)/2 |∇u(x0 )| = o(λj ) holds for all L2 normalized eigenfunction u(x) ∈ Vλ , furthermore, the bounds are uniform when those metrics of (M, g) have a uniformly bound on the norm of tr(g ij (x)) from the proof. Hence we have our Theorem. 3 Q.E.D. Hörmander Multiplier Theorem In this section, we shall see how the L∞ estimates and gradients estimates for χλ imply Hörmander multiplier Theorem. As discussed in Introduction, we reduce 10 Theorem 1.2 to show that m(P ) is weak-type (1, 1), i.e., µ{x : |m(P )f (x)| > α} ≤ α−1 ||f ||L1 , (6) where µ(E) denotes the dx measure of E ⊂ M . Since the all eigenvalues of L are non-negative, we may assume m(t) is an even function on R. Then we have m(P )f (x) = where P = √ 1Z 1 Z m̂(t)eitP f (x)dt = m̂(t) cos(tP )f (x)dt, 2π R π R+ −L, and the cosine transform u(t, x) = cos(tP )f (x) is the solution of the following Cauchy problem of the wave equation: ( ∂2 − L)u(t, x) = 0, ∂t2 u(0, x) = f (x), ut (0, x) = 0. We shall use the finite propagation speed of solutions of the wave equation in Part 2 of the proof to get the key observation (10). Proof of Theorem 1.2. The proof of the weak-type (1,1) estimate will involve a splitting of m(P ) into two pieces: a main piece which one need carefully study, plus a remainder which has strong (1,1) estimate by using the L∞ estimates for the unit spectral projection operators as done in [6]. Specifically, define ρ ∈ C0∞ (R) as ρ(t) = 1, f or |t| ≤ , 2 ρ(t) = 0, f or |t| ≥ . (7) where > 0 is a given small constant related to the manifold, which will be specified later. Write m(P ) = m̃(P ) + r(P ), where 1 Z itP e ρ(t)m̂(t)dt m̃(P ) = (m ∗ ρ̌)(P ) = 2π 1 Z itP e (1 − ρ(t))m̂(t)dt r(P ) = (m ∗ (1 − ρ)ˇ)(P ) = 2π 11 To estimate the main term and remainder, for λ = 2j , j = 1, 2, · · ·, define τ mλ (τ ) = β( )m(τ ). λ (8) Part 1: Strong (1, 1) estimate on the remainder ||r(P )f ||L1 ≤ C||f ||L1 . We first show ||r(P )f ||L2 ≤ C||f ||L1 . Here we follow the first part in proof of Theorem 5.3.1 in [6]. Define 1 Z itP rλ (P ) = (mλ ∗ (1 − ρ)ˇ)(P ) = e (1 − ρ(t))m̂λ (t)dt 2π Notice that r0 (P ) = r(P )− P j≥1 r2j (P ) is a bounded and rapidly decreasing function of P . Hence r0 (P ) is bounded from L1 to any Lp space. We need only to show ||rλ (P )f ||L2 ≤ Cλn/2−s ||f ||L1 , λ = 2j , j = 1, 2, · · · . Using the L∞ estimate on χk , see [5] or [6], we have ||rλ (P )f ||2L2 ≤ ∞ X ||rλ (P )χk f ||2L2 k=1 ≤C ∞ X sup |rλ (τ )|2 (1 + k)n−1 ||f ||2L1 k=1 τ ∈[k,k+1] Hence we need only to show ∞ X sup |rλ (τ )|2 (1 + k)n−1 ≤ Cλn−2s k=1 τ ∈[k,k+1] Notice since mλ (τ ) = 0, for τ ∈ / [λ/2, 2λ], we have m̃λ (τ ) = O((1 + |τ | + |λ|)−N ) rλ (τ ) = O((1 + |τ | + |λ|)−N ) 12 for any N when τ ∈ / [λ/4, 4λ]. Hence we need only to show 4λ X sup |rλ (τ )|2 (1 + k)n−1 ≤ Cλn−2s k=λ/4 τ ∈[k,k+1] that is 4λ X sup |rλ (τ )|2 ≤ Cλ1−2s k=λ/4 τ ∈[k,k+1] Using the fundamental theorem of calculus and the Cauchy-Schwartz inequality, we have 4λ X Z 2 sup |rλ (τ )| ≤ C( R k=λ/4 τ ∈[k,k+1] Z = C( R 2 |rλ (τ )| dτ + Z R |rλ0 (τ )|2 dτ ) |m̂λ (t)(1 − ρ(t))|2 dt + Recall that ρ(t) = 1, for |t| ≤ , 2 Z R |tm̂λ (t)(1 − ρ(t))|2 dt) by a change variables shows that this is dominated by λ −1−2s Z R |ts m̂λ (t/λ)|2 dt = λ−1−2s ||λβ(·)m(λ·)||2L2s = λ1−2s ||β(·)m(λ·)||2L2s ≤ Cλ1−2s Here the first equality comes from a change variables, the second equality comes from the definition of Sobolev norm of L2s (M ) and the third inequality comes from our assumption (4). Hence we have the estimate for the remainder ||r(P )f ||L2 ≤ C||f ||L1 . And since our manifold is compact, we have ||r(P )f ||L1 ≤ V ol(M )1/2 ||r(P )f ||L2 ≤ C||f ||L1 . 13 Part 2: weak-type (1, 1) estimate on the main term µ{x : |m̃(P )f (x)| > α} ≤ α−1 ||f ||L1 . The weak-type (1, 1) estimate on m̃(P ) would follow from the integral operator 1 Z m̃(P )f (x) = m̂(t)ρ(t)eitP f (x)dt 2π R Z X 1 Z itλk = m̂(t)ρ(t) e eλk (x) eλk (y)f (y)dydt 2π R M k≥1 X 1 Z Z { m̂(t)ρ(t) eitλk eλk (x)eλk (y)dt}f (y)dy 2π M R k≥1 = with the kernel K(x, y) = Z m̂(t)ρ(t) R = X X eitλk eλk (x)eλk (y)dt k≥1 (m ∗ ρ̌)(λk )eλk (x)eλk (y) k≥1 is weak-type (1,1). Now define the dyadic decomposition Kλ (x, y) = Z R m̂λ (t)ρ(t) X eitλk eλk (x)eλk (y)dt k≥1 We have K(x, y) = ∞ X K2j (x, y) + K0 (x, y) j=1 where K0 is bounded and vanishes when dist(x, y) is larger than a fixed constant. In order to estimate Kλ (x, y), we make a second dyadic decomposition as follows Kλ,l (x, y) = Z R m̂λ (t)β(2−l λ|t|)ρ(t) X eitλk eλk (x)eλk (y)dt k≥1 We have Kλ (x, y) = ∞ X l=−∞ 14 Kλ,l (x, y) Define Tλ,l (P )f (x) = Z M Kλ,l (x, y)f (y)dy. From above two dyadic decompositions, we have m̃(P )f (x) = ∞ X ∞ X T2k ,l (P )f (x). (9) k=0 l=−∞ Note that, because of the support properties of ρ(t), Kλ,l (x, y) vanishes if l is larger than a fixed multiple of logλ. Now we exploit the fact that the finite propagation speed of the wave equation mentioned before implies that the kernels of the operators Tλ,l , Kλ,l must satisfy Kλ,l (x, y) = 0, if dist(x, y) ≥ C(2l λ−1 ), since cos(tP ) will have a kernel that vanishes on this set when t belongs to the support of the integral defining Kλ,l (x, y). Hence in each of the second sum of (9), there are uniform constants c, C > 0 such that cλdist(x, y) ≤ 2l ≤ Cλ (10) must be satisfied for each λ = 2k , we will use this key observation later. Now for Tλ,l (P )s, we have the following estimates: (a). ||Tλ,l (P )f ||L2 (M ) ≤ C(2l )−s λn/2 ||f ||L1 (M ) (b). ||Tλ,l (P )g||L2 (M ) ≤ C(2l )−s0 λn/2 [λ max dist(y, y0 )]||g||L1 (Ω) where Ω = support(g), y,y0 ∈Ω R Ω g(y)dy = 0 and n/2 < s0 < min{s, n/2 + 1}. We first show estimate (a). Notice that β(2−l λ|t|)ρ(t) = 0 when |t| ≤ 2l−1 λ−1 , we can use the same idea to prove estimate (a) as we prove the estimate on the 15 remainder r(P ) in Part 1, where 1 − ρ(t) = 0, for |t| ≤ 2 . Using orthogonality of χk for k ∈ N, and the L∞ estimates on χk in [5], we have ||Tλ,l (P )f ||2L2 ≤ ∞ X ||Tλ,l (P )χk f ||2L2 k=1 ∞ X ≤ C sup |Tλ,l (τ )|2 (1 + k)n−1 ||f ||2L1 k=1 τ ∈[k,k+1] Hence we need only to show ∞ X sup |Tλ,l (τ )|2 (1 + k)n−1 ≤ C(2l )−2s λn k=1 τ ∈[k,k+1] Notice since mλ (τ ) = 0, for τ ∈ / [λ/2, 2λ], we have Tλ,l (τ ) = O((1 + |τ | + |λ|)−N ) for any N when τ ∈ / [λ/4, 4λ]. Then we have X sup |Tλ,l (τ )|2 (1 + k)n−1 ≤ C τ ∈[k,k+1] k∈[λ/4,4λ] / X (1 + k + λ)−2N (1 + k)n−1 k∈[λ/4,4λ] / ≤ C Z x>1,x∈[λ/4,4λ] / xn−1 dx (x + λ)2N ≤ C(1 + λ)n−2N Since 2l ≤ Cλ from our observation (10) above, we need only to show 4λ X sup |Tλ,l (τ )|2 (1 + k)n−1 ≤ C(2l )−2s λn k=λ/4 τ ∈[k,k+1] that is 4λ X sup |Tλ,l (τ )|2 ≤ C(2l )−2s λ k=λ/4 τ ∈[k,k+1] 16 Using the same argument as in Part 1, Notice that β(2−l λ|t|)ρ(t) = 0 when |t| ≤ 2l−1 λ−1 , we have the estimate (a), ||Tλ,l (P )f ||L2 (M ) ≤ C(2l )−s λn/2 ||f ||L1 (M ) Next we prove the estimate (b). We will use the orthogonality of {ej }j∈N , Z M eλk (x)eλj (x)dx = δkj , and the gradient estimates on χk for all k ∈ N as in Theorem 1.1. Given function g ∈ L1 (M ) with Ω = support(g) and R M g(y)dy = 0, fix a point y0 ∈ Ω, we have ||Tλ,l (P )g||2L2 = = Z ZM | | Z ZΩ M Ω Kλ,l (x, y)g(y)dy|2 dx [Kλ,l (x, y) − Kλ,l (x, y0 )]g(y)dy|2 dx (here use the cancellation of g) = Z | Z X M = Ω k≥1 XZ k≥1 M | Tλ,l (λk )eλk (x)[eλk (y) − eλk (y0 )]g(y)dy|2 dx Z Ω {Tλ,l (λj )eλj (x)[eλj (y) − eλj (y0 )]}g(y)dy|2 dx X λj ∈[k,k+1) (here use the orthogonality) ≤ XZ y∈Ω k≥1 M = ||g||2L1 X max | XZ k≥1 M {Tλ,l (λj )eλj (x)[eλj (y) − eλj (y0 )]}|2 dx| Z λj ∈[k,k+1) X | Ω {Tλ,l (λj )eλj (x)[eλj (y1 ) − eλj (y0 )]}|2 dx λj ∈[k,k+1) (where the maximum achieves at y1 ) = ||g||2L1 = ||g||2L1 XZ k≥1 M XZ k≥1 M Tλ,l (λj )eλj (x)eλj (ȳ), y1 − y0 )|2 dx X |(∇y λj ∈[k,k+1) |{ X Tλ,l (λj )eλj (x)(∇eλj (ȳ), y1 − y0 )}|2 dx λj ∈[k,k+1) 17 |g(y)|dy|2 = ||g||2L1 X |Tλ,l (λj )(∇eλj (ȳ), y1 − y0 )|2 X k≥1 λj ∈[k,k+1) ≤ ||g||2L1 X k≥1 max |Tλ,l (τ )|2 { τ ∈[k,k+1) |∇eλj (ȳ)|2 dist(y1 , y0 )2 } X λj ∈[k,k+1) (here use the orthogonality) ≤ ||g||2L1 [ max dist(y, y0 )]2 y,y0 ∈Ω ≤ max |Tλ,l (τ )|2 { X k≥1 C||g||2L1 [ max dist(y, y0 )]2 y,y0 ∈Ω τ ∈[k,k+1) |∇eλj (ȳ)|2 } λj ∈[k,k+1) 2 max |Tλ,l (τ )| (1 + k)n+1 X k≥1 X τ ∈[k,k+1) Now using the same computation as to the estimate (a), for some constant s0 satisfying n/2 < s0 < min{s, n/2 + 1}, we have X k≥1 max |Tλ,l (τ )|2 (1 + k)n+1 ≤ C(2l )−2s0 λn+2 . τ ∈[k,k+1) Combine above two estimates, we proved the estimate (b), ||Tλ,l (P )g||L2 (M ) ≤ C(2l )−s0 λn/2 [λ max dist(y, y0 )]||g||L1 (Ω) y,y0 ∈Ω Now we use the estimates (a) and (b) to show Z m̃(P )f (x) = K(x, y)f (y)dy M is weak-type (1,1). We let f (x) = g(x) + P∞ k=1 bk (x) := g(x) + b(x) be the Calderón- Zygmund decomposition of f ∈ L1 (M ) at the level α using the same idea as Lemma 0.2.7 in [6]. Let Qk ⊃ supp(bk ) be the cube associated to bk on M, and we have ||g||L1 + ∞ X k=1 n ||bk ||L1 ≤ 3||f ||L1 |g(x)| ≤ 2 α almost everywhere, and for certain non-overlapping cubes Qk , bk (x) = 0 f or x ∈ / Qk ∞ X µ|Qk | ≤ α−1 ||f ||L1 . k=1 18 and Z M bk (x)dx = 0 Now we show the weak-type (1, 1) estimate for m̃(P ). Since {x : |m̃(P )f (x)| > α} ⊂ {x : |m̃(P )g(x)| > α/2} ∪ {x : |m̃(P )b(x)| > α/2} Notice Z |g|2 dx ≤ 2n α Z M |g|dx. M Hence we use the L2 boundedness of m̃(P ) and Tchebyshev’s inequality to get µ{x : |m̃(P )g(x)| > α/2} ≤ Cα−2 ||g||2L2 ≤ C 0 α−1 ||f ||L1 . Let Q∗k be the cube with the same center as Qk but twice the side-length. After possibly making a translation, we may assume that Qk = {x : max |xj | ≤ R}. Let O∗ = ∪Q∗k , we have µ|O∗ | ≤ 2n α−1 ||f ||L1 , and ∗ µ{x ∈ / O : |m̃(P )b(x)| > α/2} ≤ 2α −1 ≤ 2α−1 Z x∈O / ∗ ∞ XZ |m̃(P )b(x)|dx / ∗k k=1 x∈Q |m̃(P )bk (x)|dx Hence we need only to show Z x∈Q / ∗k |m̃(P )bk (x)|dx = Z x∈Q / ∗k ≤ C Z M | Z Qk K(x, y)bk (y)dy|dx |bk |dx. From the double dyadic decomposition (9), we show two estimates of Tλ,l (P )bk (x) on set {x ∈ M : x ∈ / O∗ }, l n/2−s (I) ||Tλ,l (P )bk ||L1 (x∈O ||bk ||L1 (Qk ) / ∗ ) ≤ C(2 ) l n/2−s0 [λ max dist(y, y0 )]||bk ||L1 (Qk ) (II) ||Tλ,l (P )bk ||L1 (x∈O / ∗ ) ≤ C(2 ) y,y0 ∈Qk 19 From our observation (10), as did in [7], it suffices to show that for all geodesic balls BRλ,l of radius Rλ,l = 2l λ−1 , one has the bounds (I)0 (II)0 l n/2−s ||Tλ,l (P )bk ||L1 ({x∈O ||bk ||L1 (Qk ) / ∗ }∩BRλ,l ) ≤ C(2 ) l n/2−s0 ||Tλ,l (P )bk ||L1 ({x∈O [λ max dist(y, y0 )]||bk ||L1 (Qk ) / ∗ }∩BRλ,l ) ≤ C(2 ) y,y0 ∈Qk To show (I)0 , using the estimate (a), and Hölder inequality, we get 1/2 ||Tλ,l (P )bk ||L1 ({x∈O ||Tλ,l (P )bk ||L2 / ∗ }∩BRλ,l ) ≤ V ol(BRλ,l ) ≤ C(2l λ−1 )n/2 (2l )−s λn/2 ||bk ||L1 = C(2l )n/2−s ||bk ||L1 To show (II)0 , using the cancellation property R Qk bk (y)dy = 0, the estimate (b), and Hölder inequality, we have 1/2 ||Tλ,l (P )bk ||L1 ({x∈O ||Tλ,l (P )bk ||L2 / ∗ }∩BRλ,l ) ≤ V ol(BRλ,l ) 2l ≤ C( )n/2 (2l )−s0 λn/2 [λ max dist(y, y0 )]||bk ||L1 (Qk ) y,y0 ∈Qk λ l n/2−s0 = C(2 ) [λ max dist(y, y0 )]||bk ||L1 (Qk ) y,y0 ∈Qk From our observation (10), and estimates (I), we have ∞ X X ||Tλ,l (P )bk ||L1 (x∈O / ∗) ≤ C (2l )n/2−s ||bk ||L1 (Qk ) 2l ≥cλdist(x,y) l=−∞ ≤ Cs (λdist(x, y))n/2−s ||bk ||L1 (Qk ) ≤ Cs (λR)n/2−s ||bk ||L1 (Qk ) , and from maxy,y0 ∈Qk dist(y, y0 ) ≤ CR, estimate (II), and n/2 < s0 < min{s, n/2 + 1}, we have ∞ X l=−∞ ||Tλ,l (P )bk ||L1 (x∈O / ∗) ≤ C (2l )n/2−s0 [λ max dist(y, y0 )]||bk ||L1 (Qk ) X 2l ≥cλdist(x,y) 20 y,y0 ∈Qk ≤ Cs0 (λdist(x, y))n/2−s0 [λ max dist(y, y0 )]||bk ||L1 (Qk ) y,y0 ∈Qk n/2+1−s0 ≤ Cs0 (λR) ||bk ||L1 (Qk ) . Therefore, we combine the above two estimate we conclude that Z x∈Q / ∗k |m̃(P )bk (x)|dx ≤ ∞ X ∞ X ||T2j ,l (P )bk ||L1 (x∈O / ∗) j=0 l=−∞ ≤ Cs X (λR)n/2−s ||bk ||L1 + Cs0 2j R>1 X (λR)n/2+1−s0 ||bk ||L1 2j R≤1 ≤ Cs ||bk ||L1 Hence we have the weak-type (1, 1) estimate on the main term µ{x : |m̃(P )f (x)| > α} ≤ α−1 ||f ||L1 . Combine Case 1 and Case 2, we have the weak-type estimate of m(P ) and we finish the proof of Theorem 1.2. Q.E.D. References [1] D. Gilbarg, N. S. Trudinger, Elliptic partial differential equations of second order, Springer-Verlag, 2001 [2] L. Hörmander, The spectral function of an elliptic operator, Acta Math. 88 (1968), 341-370. [3] L. Hörmander, The analysis of linear partial differential operators III, SpringerVerlag, 1985. [4] A. Seeger and C. D. Sogge, On the boundedness of functions of pseudodifferential operators on compact manifolds. Duke Math. J. 59 (1989), 709-736. 21 [5] C. D. Sogge, Concerning the Lp norm of spectral clusters for second-order elliptic operators on compact manifolds, J. Funct. Anal. 77 (1988), no. 1, 123134. [6] C. D. Sogge, Fourier integrals in classical analysis. Cambridge Tracts in Mathematics, 105. Cambridge University Press, Cambridge, 1993. [7] C. D. Sogge, Eigenfunction and Bochner Riesz estimates on manifolds with boundary. Mathematical Research Letter 9, 205-216 (2002). [8] C. D. Sogge, S. Zelditch, Riemannian manifolds with maximal eigenfunction growth. Duke Math. J. Vol. 114, No. 3, 387-437 (2002). [9] E. M. Stein and G. Weiss, Introduction to Fourier analysis on Euclidean spaces. Princeton Uniersity Press, Princeton, 1971. [10] M. Taylor, Pseudo-differential Operators. Princeton Univ. Press, Princeton N.J., 1981. [11] Xiangjin Xu, Gradient estimates for eigenfunctions of compact manifolds with boundary and the Hörmander Multiplier Theorem. (preprint) [12] Xiangjin Xu, Eigenfunction Estimates on Compact Manifolds with Boundary and Hörmander Multiplier Theorem. PhD Thesis, Johns Hopkins University, May, 2004. 22