PROCEEDINGS OF THE AMERICAN MATHEMATICAL SOCIETY Volume 124, Number 7, July 1996 HERMITE MULTIPLIERS AND PSEUDO-MULTIPLIERS JAY EPPERSON Communicated by J. Marshall Ash) Abstract. We prove a multiplier theorem for the Hermite-Triebel-Lizorkin spaces introduced by Epperson in [Studia Math. 114 1995), 87 103]. This extends Thangavelu s theorem [Revist. Mat. Ibero 3 1987), 1 24; Math. Notes, vol. 42, 1993] on Hermite multipliers for L p spaces. We also prove an L p boundedness result for a class of Hermite pseudo-multipliers. 1. Introduction and main results We begin with a review of some of the notation and results from [1]. Consult [6] for background information on Hermite expansions. Let h k x) denotethe k th L 2 R)-normalized Hermite function, k N 0 = {0, 1, 2,...}. Recall that the collection {h k } is a complete orthonormal basis for L2,andthath k x)isan eigenfunction of the Hermite operator H = d2 dx + x 2 with corresponding eigenvalue 2k +1. Ifm:R + Cis a bounded function, then we let mh) denotethe 2 bounded linear operator on L 2 defined by mh)h k = m2k +1)h k. Now suppose ϕ : R C is C and satisfies i) supp ϕ [ 1 2, 2], ii) ϕx) c>0ifx [ 3 4,7 4 ]. For each µ N 0 define the operator Q µ = ϕ2 µ H). Let L 2 f denote the space of finite linear combinations of Hermite functions. For g L 2 f define the Hermite- Triebel-Lizorkin norm g H αq = 2 µα Q p µ g ) q ) 1/q Lp R). See [7, 8] for a detailed description of the Triebel-Lizorkin spaces which occur in Fourier analysis. The parameters α, q, p are assumed to satisfy α R, 1<p<, and 1 <q, with the usual interpretation if q =. ThespaceHp αq is defined to be the completion of L 2 f with respect to the Hp αq norm. One of the main results in [1] is that the space Hp αq is essentially independent of the particular choice of ϕ chosen to satisfy conditions i), ii). To be precise, suppose ϕ 1), ϕ 2) are two different C functions satisfying i), ii), and let Hp αq 1), Hαq p 2) denote the corresponding spaces. Then Theorem 1.1 of [1] states that Hp αq 1) and Hp αq 2) are identical as sets and have equivalent norms. Theorem 1.2 of [1] Received by the editors January 3, 1995. 1991 Mathematics Subject Classification. Primary 42C10. 2061 c 1996 American Mathematical Society
2062 JAY EPPERSON states that the spaces Hp 02 and L p are isomorphic and have equivalent norms, as is expected. A function m : R + C will be called an Hp αq Hermite multiplier if the operator mh) :L 2 f L2 f has a bounded linear extension to Hαq p. Theorem 1. Let α R, 1 <p,q<. Suppose m : R + C is bounded and satisfies m κ) cκ 1. Then m is an Hp αq Hermite multiplier. Note that this is directly analogous to Mihlin s theorem [2] for Fourier multipliers. Thangavelu [5, 6] first proved this theorem for L p spaces the α =0,q= 2 case) using special g-functions based on the Hermite semigroup. Section 2 of this paper contains a natural, alternative approach to the proof of Theorem 1. Of course the derivative condition on m in Theorem 1 can be replaced by a difference condition. Let m2k +1):=m2k +1)+1) m2k + 1). In the proof we only need m to satisfy m2k +1) c1 + k) 1 for k N 0, which is certainly implied by the condition given on m. Next we consider pseudo-multipliers. Let a : R R + C be bounded, and for g L 2 f define 1) Agx) = ax, 2k +1) g, h k h k x). Theorem 2. Suppose ax, κ) is measurable in the x variable for each fixed κ, and satisfies κax, γ κ) c1 + κ) γ for 0 γ 5. If the operator A is bounded on L 2,thenAalso extends to a bounded operator on L p for 1 <p<2. Using a method from [1] we establish uniform weak-l 1,L 1 ) bounds on certain truncated versions of A, from which Theorem 2 follows by Marcinkiewicz interpolation. See Section 3. 2. Multipliers We begin by describing the main steps toward proving Theorem 1. As in [1], let ψ : R C satisfy the same conditions i), ii) as ϕ, and the condition ϕ2 µ x)ψ2 µ x) = 1 for all x 1. Let ρx) =ϕ2x)ψ2x) +ϕx)ψx)+ϕ2 1 x)ψ2 1 x). For µ N 0 define T µ = ρ2 µ H). Note that Q µ = T µ Q µ. Now let m be as in Theorem 1, and for each µ N 0 let W µ = mh)t µ. Let L 2 l 2 ) f denote the subspace of L 2 l 2 ) consisting of sequences {g µ } such that only finitely many g µ are nonvanishing. Finally, define W : L 2 l 2 ) f L 2 l 2 ) f by W {g µ } )={W µg µ }. It is easy to see that for g L2 f, mh)g H αq = W {2 µα Q p µ g}) L p l q ). Therefore, to prove Theorem 1 it suffices to show that W is bounded on L 2 l 2 ) f L p l q )inthel p l q ) norm topology. Lemma 2.1. Let 1 <q<. Then W has a bounded linear extension to L q l q ). Lemma 2.2. Let 1 <q<. Then W is weak-l 1 l q ),L 1 l q )) bounded.
HERMITE MULTIPLIERS AND PSEUDO-MULTIPLIERS 2063 By Marcinkiewicz interpolation, these two lemmas suffice to show that W is bounded on L 2 l 2 ) f L p l q )inthel p l q ) norm topology, for 1 <p q<. Thecase1<q p< follows from the facts that: 1) L p l q ) is the dual of L p l q ), and 2) Lemmas 2.1 and 2.2 continue to hold if m and ρ are replaced by their complex conjugates. The proofs of Lemmas 2.1 and 2.2 depend on integral estimates for the kernels of the W µ operators. First we need Lemma 2.3. There exist constants c 1,c 2 >0 independent of L 1 such that L h 2 kx) 1 L 1/2 e c2l 1 x 2. Proof. We recall the argument used to prove Lemma 3.2.1 in [6]. If 0 <r<1, then by Mehler s formula L h 2 kx) r L Substituting r = e 1/L we get r k h 2 kx) =π 1/2 r L 1 r 2 ) 1/2 e 1 r 1+r x2. L h 2 k x) 1L 1/2 e c2l 1 x 2. Lemma 2.4. There exist constants 0 <c 1,c 2 < such that for every t>0, µ N 0,andy R, W µ x, y) dx 1 2 µ/2 t) 1/2 e c22 µ y 2) 2. x y t Proof. Inequality 2) follows from 3) x y) 2 W µ x, y) 2 dx 1 2 µ/2 e c22 µ y 2 by an application of Schwarz s inequality. To prove 3) we use a simple case of Thangavelu s Lemma 3.2.3 [6], p. 72): 4) x y)w µ x, y) = 1 2 B A) W µx, y). Here A = x + x, B = y + y, and W µ x, y) := m2k +1)ρ2 µ 2k + 1)))h k x)h k y). Identity 4) is easily derived from the recursion relation together with the fact that 2xh k x) =2k+2) 1/2 h k+1 x)+2k) 1/2 h k 1 x), d dx + x)h kx) =2k+2) 1/2 h k+1 x).
2064 JAY EPPERSON Substituting 4) in 3), we get 5) x y) 2 W µ x, y) 2 dx B W µ x, y) 2 dx + c A W µ x, y) 2 dx m2k +1)ρ2 µ 2k +1))+m2k +3) ρ2 µ 2k +1)) 2 2k +2)h 2 k+1y)+h 2 ky)) 1 + k) 1 ρ2 µ 2k +1)) + m2k +3)2 µ ρ 2 µ ξk)) ) 2 2k +2)h 2 k+1 y)+h2 k y)), where each ξk) is between 2k +1 and 2k+3. Since ρ is compactly supported away from the origin, there exist integers 0 <N 1 <N 2 independent of µ N 0 such that the terms in 5) vanish unless 2 µ N 1 k 2 µ N 2. Thus 5) is bounded by 2 µ N 2+1 c 2 2µ 2 µ h 2 ky) 1 2 µ/2 e c22 µ y 2, k=2 µ N 1 by an application of Lemma 2.3. Lemma 2.5. There exists a constant c< such that for every t>0,µ N 0, and y, z R with y z t, 6) x z 2t W µ x, y) W µ x, z)+y z)d 2 W µ x, z)) dx 2 µ/2 t) 3/2. Proof. Let J denote the interval with endpoints y, z. We can rewrite the left side of 6) as y u)d2w 2 µ x, u)du dx x z 2t J y u J D2 2 W µx, u) dxdu x z 2t t 2 sup D2W 2 µ x, u) dx. u J x z 2t We estimate the last integral using D2W 2 µ x, u) =D2 u 2 2 )W µ x, u)+u 2 W µ x, u) =I+II. Note that I = 2 µ m2k +1)ρ2 µ 2k + 1))2 µ 2k +1))h k x)h k y). Hence, by the method of proving Lemma 2.4 we get t 2 sup u J I dx t 2 sup u J I dx t 2 2 µ 2 µ/2 t) 1/2. x z 2t x u t
HERMITE MULTIPLIERS AND PSEUDO-MULTIPLIERS 2065 Again using Lemma 2.4, we have t 2 sup II dx 2 µ/2 t) 2 sup 2 µ u 2 2 µ/2 t) 1/2 e c22 µ u 2 2 µ/2 t) 3/2. u J u J x z 2t Proof of Lemma 2.1. It suffices to show that the operators W µ, µ N 0, are uniformly bounded on L q. Thisistrivialforq= 2, so it suffices by interpolation and duality) to show that the W µ operators are uniformly weak-l 1,L 1 ) bounded. To do this we must show that there exists a constant c< independent of f L 1, λ>0, and µ N 0 such that {x : W µ fx) >λ} λ f L 1. This will be a routine application of Lemmas 2.4 and 2.5. So fix f L 1 and λ>0, and apply the Calderón-Zygmund lemma to get a collection of disjoint dyadic open intervals { } such that a) fx) λfor a.e. x R \ j, b) j 1 λ f L 1, c) λ 1 fx) dx 2λ for all j. Let z j denote the centerpoint of,andforx let gx) = 1 fy)dy + 12x z j) 3 fy)y z j ) dy. Also, if x,letbx)=fx) gx). For x/ j,letgx)=fx)andbx)=0. Thus f = g + b everywhere. Note that if x,then gx) 8λ. Also, for a.e. x/ j, gx) λ. By the standard argument g 2 L λ f 2 L 1. So, by Chebyshev s inequality {x : W µ gx) >λ/2} 4 λ 2 W µg 2 L 2 λ f L 1. Next we have to prove the correct sort of estimate for {x : W µ bx) >λ/2}. Define Ij =z j,z j + ). Since j I j 2 λ f L1, it suffices to estimate {x R \ j I j : W µ bx) >λ/2}. For each j let b j = b χ Ij. Then b = j b j a.e., b j x)x z j ) dx =0,and bj x)dx = 0. By Chebyshev s inequality {x R \ Ij : W µbx) >λ/2} 2 7) W µ b j x) dx. λ j j R\Ij For each j define the kernel { L j Wµ x, y) if 2 µx, y) = µ/2 1, W µ x, y) W µ x, z j )+y z j )D 2 W µ x, z j )) if 2 µ/2 < 1. Because of the vanishing moment conditions imposed on b j,wehave W µ b j x) dx = L j µ x, y)b jy)dy dx R\Ij R\Ij b j y) L j µ x, y) dxdy.
2066 JAY EPPERSON Now according to Lemmas 2.4 and 2.5 we see that 7) is bounded by c min{2 µ/2 ) 1/2, 2 µ/2 ) 3/2 } b j y) dy λ λ f L 1. j Proof of Lemma 2.2. We need to show that there exists a constant c< independent of {f µ } L 1 l q ), λ>0 such that {x : W µ f µ x) q ) 1/q >λ} λ {f µ} L 1 l q ). So fix {f µ } L 1 l q )andλ>0, let hx) = f µx) q ) 1/q, and apply the Calderón-Zygmund lemma to get a collection of disjoint open intervals { } such that a) hx) λfor a.e. x R \ j, b) j 1 λ h L 1 c) λ 1 hx) dx 2λ for all j. Again let z j denote the centerpoint of,andforx let g µ x) = 1 f µ y)dy + 12x z j) 3 f µ y)y z j )dy. For x let b µ x) =f µ x) g µ x), and for x/ j let g µ x) =f µ x), b µ x) =0. If x we have by Minkowski s inequality g µ x) q ) 1/q 1 f µ y)dy q ) 1/q + 12x z j) I j I j 3 f µ y)y z j )dy q ) 1/q 1 f µ y) q ) 1/q dy + 3 f µ y) q ) 1/q dy I j 8λ. It follows that {g µ } q L q l q ) λq 1 {f µ } L 1 l q ), and therefore by Chebyshev s inequality and Lemma 2.1, {x : W µ g µ x) q ) 1/q >λ/2} 2q λ q {W µg µ } q L q l q ) λ {f µ} L1 l q ). Next we have to estimate {x : W µb µ x) q ) 1/q >λ/2}. As in the proof of Lemma 2.1 it suffices to handle {x R \ Ij : W µ b µ x) q ) 1/q >λ/2} j 8) 2 W µ b µ,j x) q ) 1/q dx. λ j
HERMITE MULTIPLIERS AND PSEUDO-MULTIPLIERS 2067 Here of course b µ,j = b µ χ Ij. Let L j µ be as in the proof of Lemma 2.1. Then by Minkowski s inequality and lemmas Lemmas 2.4 and 2.5 W µ b µ,j x) q ) 1/q dx = L j µ x, y)b µ,jy)dy q ) 1/q dx L j µx, y)b µ,j y) q ) 1/q dydx b µ,j y) q ) 1/q R\Ij L j µx, y) dx) dy b µ,j y) q ) 1/q min{2 µ/2 ) 1/2, 2 µ/2 ) 3/2 }) dy b µ,j y) q ) 1/q dy f µ y) q ) 1/q dy with c independent of. Substituting this in 8) finishes the proof. 3. Pseudo-multipliers In this section we prove Theorem 2. Let P µ = π2 µ H), where πx) isac function supported on [ 1 2, 2] with the property that π2 µ x) = 1 for all x 1. Lemma 3.1. Let ax, κ) be as in the statement of Theorem 2, and suppose that the associated operator A is bounded on L 2. Then the operators N AP µ are weak-l 1,L 1 ) bounded, uniformly in N. Proof. We begin by recalling how to estimate the kernel K µ x, y) of the operator AP µ.leta µ x, κ) =ax, κ)π2 µ κ), and let â µ x, ξ) denote the Fourier transform of a µ in its second variable. As in the derivation of 7) in 3 of[1],wehavethe representation K µ x, y) =c â µ x, ξ/2)e iξ/2 1 e i2ξ ) 1/2 e i 2 x2 +y 2 )cotξ 2xy csc ξ) dξ, where c is some unimportant constant. Now the conditions on a imply that for every l N 0 there exists a constant c l independent of µ N 0 such that ξâµx, l ξ) c l 2 µ1+l) 1 + 2 µ ξ ) 5. It follows by inspection of the proof of Lemma 1.1 in [1] that there exists a constant c independent of µ N 0 such that K µ x, y) c2 µ/2 9) 1 + 2 µ/2 x + σy ) 4 σ=±1 and yk 2 µ x, y) c2 3µ/2 10) 1 + 2 µ/2 x + σy ). 2 σ=±1
2068 JAY EPPERSON The proof is finished with a simple modification of the proof of Lemma 2.1, which we very briefly indicate. Fix f L 1 and λ>0, and let { }, gx), bx), etc., be as in the proof of Lemma 2.1. From the L 2 -boundedness of A and the uniform L 2 -boundedness of the operators N P µ we get N {x : AP µ gx) >λ/2} λ f L 1. Now let L j µx, y) be defined as in the proof of Lemma 2.1, except with K µ x, y) in place of W µ x, y). Then from 9) and 10) we get N N AP µ b j x) dx b j y) L j µx, y) dx) dy R\Ij b j y) min{2 µ/2 ) 3, 2 µ/2 }) dy b j y) dy. Proof of Theorem 2. By Lemma 3.1 and the Marcinkiewicz interpolation theorem, each of the operators N AP µ has a bounded linear extension to L p,for1< p 2. Moreover, the operator norms N AP µ Lp Lp are bounded uniformly in N. Now let g L 2 f. Then Ag = N AP µg for a large enough choice of N. Hence Ag L p g L p. The proof is finished by recalling that L 2 f is dense in Lp see for example [3], Lemma 2). It would be interesting to find natural criteria for the L 2 -boundedness of a Hermite pseudo-multiplier, since the standard methods for obtaining L 2 -boundedness of an ordinary pseudo-differential operator as in [4]) do not seem to be applicable here. References 1. J. Epperson, Triebel-Lizorkin spaces for Hermite expansions, Studia Math. 114 1995), 87 103. 2. S.G. Mihlin, On the multipliers of Fourier integrals, Dokl. Acad. Nauk SSSR N.S. 109 1956), 701 703 Russian). MR 18:304a 3. B. Muckenhoupt, Mean convergence of Hermite and Laguerre series II, Trans. Amer. Math. Soc. 147 1970), 433-460. MR 41:711 4. E.M. Stein, Harmonic Analysis, Princeton University Press, 1993. MR 95c:42002 5. S. Thangavelu, Multipliers for Hermite expansions, Revist.Mat.Ibero.31987), 1 24. MR 90h:42043 6. S. Thangavelu, Lectures on Hermite and Laguerre Expansions, Mathematical Notes 42, Princeton University Press, 1993. MR 94i:42001 7. H. Triebel, Theory of Function Spaces, Birkhäuser Verlag, Basel Monographs in Mathematics, Vol. 78), 1983. MR 86j:46026 8. H. Triebel, Theory of Function Spaces II, Birkhäuser Verlag, Basel Monographs in Mathematics, Vol. 84), 1992. MR 93f:46029 Department of Mathematics and Statistics, University of New Mexico, Albuquerque, New Mexico 87131 E-mail address: jeppers@math.unm.edu