Abstract
We develop a special multilinear complex interpolation theorem that allows us to prove an optimal version of the bilinear Hörmander multiplier theorem concerning symbols that lie in the Sobolev space \(L^r_s({\mathbb {R}}^{2n})\), \(2\le r<\infty \), \(rs>2n\), uniformly over all annuli. More precisely, given such a symbol with smoothness index s, we find the largest open set of indices \((1/p_1,1/p_2 )\) for which we have boundedness for the associated bilinear multiplier operator from \(L^{p_1}({\mathbb {R}}^{ n})\times L^{p_2} ({\mathbb {R}}^{ n})\) to \( L^p({\mathbb {R}}^{ n})\) when \(1/p=1/p_1+1/p_2\), \(1<p_1,p_2<\infty \).
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Multipliers are linear operators of the form
where f is a Schwartz function on \({\mathbb {R}}^n\) and \(\widehat{f}(\xi ) = \int _{{\mathbb {R}}^n} f(x) e^{-2\pi i x\cdot \xi }dx\) is its Fourier transform.
Let \(\Psi \) be a Schwartz function whose Fourier transform is supported in the annulus of the form \(\{\xi : 1/2< |\xi |< 2\}\) which satisfies \(\sum _{j\in {\mathbb {Z}}} \widehat{\Psi }(2^{-j}\xi )=1\) for all \(\xi \ne 0\). We denote by \(\Delta \) the Laplacian and by \((I-\Delta )^{s/2} \) the operator given on the Fourier transform by multiplication by \((1+4\pi ^2 |\xi |^2)^{s/2}\); also for \(s>0\), and we denote by \(L^r_s\) the Sobolev space of all functions h on \({\mathbb {R}}^n\) with norm \( \Vert h\Vert _{L^r_s}:=\Vert (I-\Delta )^{s/2} h \Vert _{L^r} <\infty . \) Extending an earlier result of Mikhlin [15], the optimal version of the Hörmander multiplier theorem says that if
and
then \(T_\sigma \) is bounded from \(L^p({\mathbb {R}}^n)\) to itself for \(1<p<\infty \). Hörmander’s [13] original version of this theorem stated boundedness in the entire interval \(1<p<\infty \) provided \(s> n/2\). A restriction on the indices first appeared in Calderón and Torchinsky [1], while condition (2) appeared in [5]; this condition is sharp as examples are given in [5] indicating that the theorem fails in general when \(\big | \frac{1}{p} -\frac{1}{2} \big | > \frac{s}{n}\). Moreover, recently Slavíková [19] provided an example showing that boundedness may also fail even on the critical line \(\big | \frac{1}{p} -\frac{1}{2} \big | = \frac{s}{n}\).
In this paper we provide bilinear analogues of these results. The study of the Hörmander multiplier theorem in the multilinear setting was initiated by Tomita [21] and was further studied by Fujita, Grafakos, Miyachi, Nguyen, Si, Tomita (see [2, 7, 8, 11, 17, 18]) among others. For a given function \(\sigma \) on \({\mathbb {R}}^{2n}\) we define a bilinear operator
originally defined on pairs of \({\mathcal {C}}_0^\infty \) functions \(f_1,f_2\) on \({\mathbb {R}}^n\). We fix a Schwartz function \(\Psi \) on \({\mathbb {R}}^{2n}\) whose Fourier transform is supported in the annulus \(1/2\le |( \xi _1,\xi _2) |\le 2\) and satisfies
The following theorem is the main result of this paper:
Theorem 1.1
Let \(2\le r<\infty \), \(s>\frac{2n}{r}\), \(1<p_1,p_2\le \infty \) and let \(1/p=1/p_1+1/p_2>0\).
-
(a)
Let \(n/2<s\le n\). Suppose that
$$\begin{aligned} \frac{1}{p_1}<\frac{s}{n},\, \frac{1}{p_2}<\frac{s}{n},\, 1-\frac{s}{n}<\frac{1}{p}<\frac{s}{n}+\frac{1}{2} . \end{aligned}$$(3)Then for all \({\mathcal {C}}_0^\infty ({\mathbb {R}}^n)\) functions \(f_1,f_2\) we have
$$\begin{aligned} \Vert T_{\sigma }(f_1,f_2)\Vert _{L^p({\mathbb {R}}^n)}\le C\sup _{j\in {\mathbb {Z}}} \Vert \sigma (2^j\cdot ){\widehat{\Psi }}\Vert _{L^r_s({\mathbb {R}}^{2n})}\Vert f_1\Vert _{L^{p_1}({\mathbb {R}}^n)}\Vert f_2\Vert _{L^{p_2}({\mathbb {R}}^n)}. \end{aligned}$$(4)Moreover, if (4) holds for all \(f_1,f_2 \in \mathcal C_0^\infty \) and all \(\sigma \) satisfying (1), then we must necessarily have
$$\begin{aligned} \frac{1}{p_1}\le \frac{s}{n},\, \frac{1}{p_2}\le \frac{s}{n},\, 1-\frac{s}{n}\le \frac{1}{p}\le \frac{s}{n}+\frac{1}{2} . \end{aligned}$$(5) -
(b)
Let \(n <s\le 3n/2\) and satisfy
$$\begin{aligned} \frac{1}{p } <\frac{s}{n}+\frac{1}{2}\, . \end{aligned}$$(6)Then (4) holds. Moreover, if (4) holds for all \(f_1,f_2 \in {\mathcal {C}}_0^\infty \) and all \(\sigma \) satisfying (1), then we must necessarily have
$$\begin{aligned} \frac{1}{p } \le \frac{s}{n}+\frac{1}{2}\, . \end{aligned}$$(7) -
(c)
If \(s>\frac{3n}{2}\) then (4) holds for all \(1<p_1,p_2<\infty \) and \(\frac{1}{2}< p<\infty \).
This theorem uses two main tools: First, the optimal n / 2-derivative result in the local \(L^2\)-case contained in [6] and a special type of multilinear interpolation suitable for the purposes of this problem (see Theorem 3.1 below). Figure 1 (Sect. 4), plotted on a slanted \((1/p_1,1/p_2)\) plane, shows the regions of boundedness for \(T_\sigma \) in the two cases \(n/2<s\le n\) and \(n <s\le 3n/2\). Note also that in the former case, the condition \(1-\frac{s}{n}<\frac{1}{p}\) is only needed when \(p>2\).
Finally, we mention that the necessity of conditions (3), (5), and (7) in Theorem 1.1 are consequences of Theorems 2 and 3 in [6]; these say that if boundedness holds, then we must necessarily have
Also, if \(T_{\sigma }\) maps \(L^{p_1}\times L^{p_2}\) to \( L^p\) and \(p>2\), then duality implies that \(T_{\sigma }\) maps \(L^{p'}\times L^{p_2}\) to \(L^{p_1'}\). Now \(p'\) plays the role of \(p_1\) and so constraint \(\frac{1}{p_1} \le \frac{s}{n}\) becomes \(1-\frac{s}{n}\le \frac{1}{p}\). This proves (5). So the main contribution of this work is the sufficiency of the conditions in (3) and (6).
2 Preliminary material for interpolation
In this section we briefly discuss three lemmas needed in our interpolation.
Lemma 2.1
Let \(0< p_0<p<p_1 \le \infty \) be related as in \(1/p=(1-\theta )/p_0+\theta /p_1\) for some \(\theta \in (0,1)\). Given \(f\in {{\mathcal {C}}}_0^\infty ({\mathbb {R}}^n)\) and \(\varepsilon >0,\) there exist smooth functions \(h_j^\varepsilon \), \(j=1,\dots , N_\varepsilon \), supported in cubes with pairwise disjoint interiors, and nonzero complex constants \(c_j^\varepsilon \) such that the functions
satisfy
and
where \(\varepsilon '\) depends on \(\varepsilon ,p_0,p_1,p, \Vert f\Vert _{L^p}\) and tends to zero as \(\varepsilon \rightarrow 0\).
Proof
Given \(f\in {{\mathcal {C}}}_0^\infty ({\mathbb {R}}^n)\) and \( \varepsilon >0\), by uniform continuity there are \(N_\varepsilon \) cubes \(Q_j^\varepsilon \) (with disjoint interiors) and nonzero complex constants \(c_j^\varepsilon \) such that
and
Find smooth functions \(g_j^\varepsilon \) satisfying \(0\le g_j^\varepsilon \le \chi _{Q_j^\varepsilon }\) such that
where the last estimate is required only when \(p_1<\infty \). We set \(h_j^\varepsilon = e^{i\phi _j^\varepsilon } g_j^\varepsilon \), where \(\phi _j^\varepsilon \) is the argument of the complex number \(c_j^\varepsilon \). Then \(h_j^\varepsilon \) is that function claimed in (8). Observe that
satisfies (9) when \(p_1<\infty \); in the case \(p_1=\infty \) we have
Now we have
having made use of (10).
Given \(a,c>0\) and \(\varepsilon >0\) set \(\varepsilon '=\varepsilon '(\varepsilon ,a,c)= (\varepsilon ^a+c^a)^{1/a}-c\). Then \( (\varepsilon ^a+c^a)^{1/a} \le \varepsilon '+c\) and \(\varepsilon '\rightarrow 0 \) as \(\varepsilon \rightarrow 0\). Then for a suitable \(\varepsilon '\) that only depends on \(\varepsilon , p,p_0,p_1, \Vert f\Vert _{L^p}\), the preceding estimate gives: \( \Vert f^{it,\varepsilon } \Vert _{L^{p_0}}^{p_0} \le \Vert f \Vert _{L^p}^p +\varepsilon '\) and analogously \( \Vert f^{1+it,\varepsilon } \Vert _{L^{p_1}} \le \big ( \Vert f \Vert _{L^p}^p +\varepsilon ' \big )^{1/p_1}\) when \(p_1<\infty \); notice that if \(p_1=\infty \) then \(\Vert f^{1+it,\varepsilon } \Vert _{L^{\infty }}\le 1\) and the right hand side of the inequality is equal to 1, thus the inequality is still valid. \(\square \)
Lemma 2.2
Given a domain \(\Omega \) on the complex plane and \((M,\mu ) \) a measure space, let \(V : \Omega \times M\rightarrow {\mathbb {C}}\) be a function such that \(V(\cdot ,x)\) is analytic on \(\Omega \) for almost every \(x\in M\). If the function
is integrable over M for each \(z\in \Omega \), then the mapping \(z\longmapsto V(z,\cdot )\) is an analytic function from \(\Omega \) to the Banach space \(L^1(M,d\mu )\).
Proof
Fix \(z\in \Omega \) and denote \(r_z = \frac{1}{2}\mathrm {dist}(z,\partial \Omega ).\) It is enough to show that
The assumption yields that for some set \(M_0\) with \(\mu (M{\setminus } M_0)=0\), we have
for all \(x\in M_0\). Thus for each \(x\in M_0\) and \(h\in {\mathbb {C}}\) with \( |h|<r_z\) we can write
Since \(V^*(z,\cdot )\) is integrable on \(M_0\), the Lebesgue dominated convergence theorem yields
This yields (12) and completes the proof, as the last integral is over the entire space M. \(\square \)
Lemma 2.3
Given \(0<a<b<\infty \), \(\Omega =\{z\in {\mathbb {C}}\ :\ a<\mathfrak {R}(z)<b\}\), and a measure space \((M,\mu )\) of finite measure, let \(H: \Omega \times {\mathbb {R}}^d\times M\rightarrow {\mathbb {C}}\) be a measurable function so that \(H(\cdot ,\xi ,x)\) be analytic on \(\Omega \) and continuous on \({\overline{\Omega }}\) for each \((\xi ,x)\in {\mathbb {R}}^d\times M.\) Suppose that
for all \((\xi ,x)\in {\mathbb {R}}^d\times M\). If \(\varphi \) be a bounded measurable function on \({\mathbb {R}}^d\), then the mapping \(z\longmapsto V(z,\cdot )\), defined by
is an analytic function from \(\Omega \) to the Banach space \(L^1(M,d\mu )\) and is continuous on \({\overline{\Omega }}\).
Proof
Let \(K=\{\xi \in {\mathbb {R}}^d:\,\, \varphi (\xi )\ne 0\}\). By assumption, for each \(x\in M\) we have
As for each \(z\in \Omega \) we have
and H satisfies assumption (13), the associated function \(V^*(z,\cdot )\) defined in (11) is bounded and thus integrable over M. Therefore, using Lemma 2.2 we deduce that \(z\longmapsto V(z,\cdot )\) is analytic from \(\Omega \) to \(L^1(M,d\mu )\).
Using Lebesgue’s dominated convergence theorem and the fist part of assumption (13) we easily deduce that \(V(z,\cdot )\) is continuous up to the boundary of \(\Omega \). \(\square \)
Lemma 2.4
[3] Let F be analytic on the open strip \(S=\left\{ z\in {\mathbb {C}}\ :\ 0<\mathfrak {R}(z)<1\right\} \) and continuous on its closure. Assume that for all \(0\le \tau \le 1\) there exist functions \(A_\tau \) on the real line such that
and suppose that there exist constants \(A>0\) and \(0<a<\pi \) such that for all \(t\in {\mathbb {R}}\) we have
Then for \(0<\theta <1 \) we have
In calculations it is crucial to note that
3 Multilinear interpolation
In this section we prove the main tool needed to derive Theorem 1.1 by interpolation. We denote by \(\vec \xi =(\xi _1,\dots , \xi _m) \) elements of \({\mathbb {R}}^{mn}\), where \(\xi _j\in {\mathbb {R}}^{n}\). We fix a Schwartz function \(\Psi \) on \({\mathbb {R}}^{mn}\) whose Fourier transform is supported in the annulus \(1/2\le |\vec \xi \, |\le 2\) and satisfies
Theorem 3.1
Let \(0<p_1^0,\dots , p_m^0 \le \infty \), \(0<p_1^1,\dots , p_m^1 \le \infty \), \(0<q_0,q_1\le \infty \), \(0\le s_0,s_1<\infty \), \(1<r_0,r_1<\infty \), \(0<\theta <1\), and let
for \(l=1,\dots , m\). Assume \(r_0s_0>mn\), and \(r_1s_1>mn\) and that for all \(f_l\in {\mathcal {C}}_0^\infty ({\mathbb {R}}^n)\), \(l=1,\dots , m\), we have
for \(k=0,1\) where \(K_0,K_1\) are positive constants. Then the intermediate estimate holds:
for all \(f_l\in {\mathcal {C}}_0^\infty ({\mathbb {R}}^n)\), where \(C_*\) depends on all the indices, on \(\theta \), and on the dimension.
Consequently, if \(p_l<\infty \) for all \(l\in \{1,\dots , m\}\), then \(T_\sigma \) admits a bounded extension from \(L^{p_1}\times \cdots \times L^{p_m} \) to \(L^q\) that satisfies (14).
Proof
Fix a smooth function \({\widehat{\Phi }}\) on \({\mathbb {R}}^{mn}\) such that \({{\,\mathrm{\mathrm {supp}}\,}}(\Phi )\subset \big \{\frac{1}{4}\le |{\vec \xi \, }|\le 4\big \}\) and \(\widehat{\Phi }\equiv 1\) on the support of the function \({\widehat{\Psi }}.\) Denote \( \varphi _j = (I-\Delta )^{\frac{s}{2}}[\sigma (2^j\cdot ){\widehat{\Psi }}] \) and define
This sum has only finitely many terms and we now estimate its \(L^\infty \) norm. \(\square \)
Fix \(\vec \xi \in {\mathbb {R}}^{mn}\). Then there is a \(j_0\) such that \(|\vec \xi \,| \approx 2^{j_0}\) and there are only two terms in the sum in (15). For these terms we estimate the \(L^\infty \) norm of \((I-\Delta )^{-\frac{s_0(1-z)+s_1 z}{2}} \big [ |\varphi _j|^{r(\frac{1-z}{r_0}+\frac{z}{r_1})}e^{i \text {Arg } (\varphi _j)} \big ]\). For \(z=\tau +it\) with \(0\le \tau \le 1\), let \(s_\tau = (1-\tau )s_0+\tau s_1\) and \(1/r_\tau = (1-\tau )/r_0+\tau /r_1\). By the Sobolev embedding theorem we have
It follows from this that
Let \(T_{\sigma _z}\) be the family of operators associated to the multipliers \(\sigma _z.\) Let \(\varepsilon \) be given.
Suppose first that \( \min ( p^0_l , p^1_l)<\infty \) for all \(l\in \{1,\dots , m\}\). This forces \(p_l<\infty \) for all l.
Case I:\(\varvec{\min }({{\varvec{q}}}_{\mathbf{0}},{{\varvec{q}}}_{\mathbf{1}})>\mathbf{1}\) This assumption implies that \(q>1\), hence \(q',q_0',q_1'<\infty \). Fix \(f_l, g\in {\mathcal {C}}_0^\infty ({\mathbb {R}}^n)\). For given \(\varepsilon >0\), for every \(l\in \{1,\dots , m\}\), by Lemma 2.1 there exist functions \(f_l^{z,\varepsilon } \) and \( {g}^{z,\varepsilon }\) of the form (8) such that
when \(\max ( p_l^0,p_l^1)<\infty \), while one of the first two inequalities is replaced by \(\Vert f_l^{\theta ,\varepsilon } \Vert _{L^\infty } \le \Vert f_l \Vert _{L^{p_l^k} }+\varepsilon = \Vert f_l \Vert _{L^{\infty } }+\varepsilon \) when \(p_l^k=\max ( p_l^0,p_l^1)=\infty \), and that
Define
Notice that
is equal to a finite sum (over \(k_1,\dots , k_m,l\)) of terms of the form
which we call \( H(z,\vec \xi \, )\), where \(\zeta _{k_1,\dots , k_m,l}\) are Schwartz functions. Thus \( H(z,\vec \xi \, )\) is an analytic function in z. Moreover \( H(z,\vec \xi \, ) \) can be thought of as a function of three variables \( H(z,\vec \xi , x_0) \), being constant in the variable \(x_0\), where \(\{x_0\}\) is a measure space of one element equipped with counting measure. With this interpretation, it is not hard to verify that \(H(z,\vec \xi , x_0) \) satisfies (13).
Lemma 2.3 guarantees that F(z) is analytic on the strip \(0<\mathfrak {R}(z)<1\) and continuous up to the boundary. Furthermore, by Hölder’s inequality,
and noting that only the terms with \(j=k-1,k,k+1\) survive in the sum in (15) for \(\sigma _{it}(2^k\cdot ){\widehat{\Psi }}\), the Kato–Ponce inequality [10, 14] applied as \(\Vert (I-\Delta )^{s/2} (F\widehat{\Phi }) \Vert _{L^{r_0}} \le C \Vert (I-\Delta )^{s/2} (F ) \Vert _{L^{r_0}}\) yields
Thus, for some constant \(C=C(m,n,r_0,s_0,s_1)\) we have
Similarly, we can choose the constant \(C=C(m,n,r_1,s_0,s_1)\) above large enough so that
Note that F(z) is a combination of finite terms of the form
where
and \(h_{j_1}^{1,\varepsilon }\), \(g_j^\varepsilon \) are smooth functions with compact support. Thus for \(z=\tau +it\), \(t\in {\mathbb {R}}\) and \(0\le \tau \le 1\) it follows from (16) and from the definition of F(z) that
As \(A_\tau (t)\le \exp (A e^{a|t|}) \), the admissible growth hypothesis of Lemma 2.4 is satisfied. Applying Lemma 2.4 we obtain
But
and then we have
A telescoping identity yields
For every fixed l, applying the hypothesis that \(T_\sigma \) is bounded from \(L^{p^k_1}\times \cdots \times L^{p^k_m}\) to \(L^{q_k}\) for \(k=0,1\) we obtain
In view of the inequality \(\Vert h\Vert _{L^q}\le \Vert h\Vert _{L^{q_0}}^{1-\theta }\Vert h\Vert _{L^{q_1}}^{\theta }\) these estimates yield
As \(0<\theta <1\) and one of \(p_l^0 \) or \(p_l^1 \) is strictly less than infinity, the expression on the right above is bounded by a constant multiple of \(\varepsilon ^{\min (\theta , 1-\theta )}\) and hence it tends to zero as \(\varepsilon \rightarrow 0\) because of (9). This proves that (in fact for all \(0<q<\infty \))
where \(E_\varepsilon \rightarrow 0\) as \(\varepsilon \rightarrow 0\). Returning to (19) and using (18) and Hölder’s inequality we write
Recalling (17) and using that each \( \Vert f_l^{\theta ,\varepsilon } \Vert _{L^{p_l^0}}\) remains bounded as \(\varepsilon \rightarrow 0\) we obtain
by letting \(\varepsilon \rightarrow 0\). Taking the supremum over all functions \(g\in L^{q'}\) with \(\Vert g\Vert _{L^{q'}}=1 \) yields the sought estimate (14) in Case I.
Case II: \(\varvec{\min }({{\varvec{q}}}_{\mathbf{0}},{{\varvec{q}}}_{\mathbf{1}}) \le \mathbf{1}\)
Here we will make use of two following lemmas proved by Stein and Weiss [20].
Lemma 3.2
([20]) Let \(U:{\overline{S}}\longrightarrow {\mathbb {R}}\) be an upper semi-continuous function of admissible growth and subharmonic in the unit strip S. Then for \(z_0=x_0+iy_0\in S\) we have
where
Lemma 3.3
([20]) Let \(0<c\le 1\) and let \((M,\mu )\) be a measure space with finite measure. If a function \(V(z,\cdot )\) is analytic from the unit strip S to \(L^1(M,\mu )\), then \(\log \int _{M}\left| V(z,x)\right| ^cd\mu \) is subharmonic on S.
We now continue the proof of the second case. We fix functions \(f_l\) as in the previous case. Choose an integer \(\rho >1\) such that \(\rho \ge \rho \min (q_0,q_1)>q.\) Take an arbitrary positive simple function g with \(\left\| g\right\| _{L^{\rho '}}=1.\) Assume that \(g = \sum _{k=1}^N c_k\chi _{E_k},\) where \(c_k>0\) and \(E_k\) are pairwise disjoint measurable sets of finite measure and compact support. For \(z\in {\mathbb {C}},\) set
Now consider
Let \(V(z,x) =T_{\sigma _z}(f_1^{z,\varepsilon },\ldots ,f_m^{z,\varepsilon })(x).\) Then V(z, x) can be represented as a finite sum of terms of the form
where \(h_\kappa ^\varepsilon \) are the smooth functions with compact support in (8) and P is a polynomial. Setting
we note that \(H(z,\vec \xi ,x)\) is analytic in z, smooth in \(\xi \) and bounded in x, as long as x remains in a compact set. Moreover H satisfies (13). Applying Lemma 2.3 we obtain that for all \((\vec \xi ,x)\) the mapping \(H(\cdot , \vec \xi ,x) \) is analytic from S to \(L^1(E_k,dx)\) Then Lemma 3.3 applies and yields that \(\log G\) is subharmonic on S. Using Hölder’s inequality with indices \(\frac{\rho q_0}{q} \) and \(\big ( \frac{\rho q_0}{q} \big )'\) and the fact that the \(L^{\rho '}\)-norm of g is equal to 1, we have
Similarly, we can estimate
Applying Lemma 3.2 to \(U=\log G\) (with \(y_0=0\) and \(x_0=\theta \)) and using that for \(0<\theta <1\) we have
(see [3, Page 48]) we obtain
Notice that as
inequality (21) implies that
Finally, we use
and we note that for the second term we use (22), while the first term tends to zero, in view of (20). Letting \(\varepsilon \rightarrow 0\), we deduce (14).
We now turn to the case where \(\min (p^0_l,p^1_l)=\infty \) for some (but not all) l in \( \{1,\dots , m\}\). Then we must have \(p_l=\infty \) for these l, and for these l we set \(f_l^{z,\varepsilon }=f\), while for the remaining l the functions \(f_l^{z,\varepsilon }\) are defined as before; we notice that the preceding argument works with only minor modifications.
Finally we consider the case where \(p^0_l=p^1_l=\infty \) for all \(1\le l\le m\). Here we also take \(f_l^{z,\varepsilon }=f_l\) for all l in \( \{1,\dots , m\}\). Now (19) becomes
Hence, in Case I, when \(\min (q_0,q_1)>1\), we have
Passing the limit as \(\varepsilon \rightarrow 0\) to obtain
The result in Case II, which is when \(\min (q_0,q_1)\le 1\), can be obtained from that in Case I by choosing \(\rho >1\) such that \( \rho \min (q_0,q_1)>q \) and by arguing as before, replacing each term \( \big (\left\| f_{l}\right\| _{L^{p_ l}}^{p_l}+\varepsilon '\big )^{\frac{1}{p_l}}\) by \(\Vert f_l\Vert _{L^\infty }\). This concludes the proof of the theorem in all cases. \(\square \)
Note that the proof of Theorem 3.1 is much simpler in the case \(r_0=r_1=2\), and this was proved earlier in [8, Theorem 6.1, Step 1]; see also [9, Theorem 2.3]. In this case, the domains can be arbitrary Hardy spaces. We state the theorem in this case (without providing a proof):
Theorem 3.4
([8]) Let \(p^0_l,p^1_l,p_l,q_0,q_1,q, s_0,s_1,s\) and \(\theta \in (0,1)\) be as in Theorem 3.1 for \(l=1,\ldots ,m\). Assume that \(s_0,s_1>\frac{mn}{2}\), \(p^0_l,p^1_l <\infty \) for all l, and that
for \(k=0,1\) where \(K_0,K_1\) are positive constants. Then we have the intermediate estimate:
for all Schwartz functions \(f_l\) with vanishing moments of all orders, where \(C_*\) depends on all the indices, \(\theta \), and the dimension.
4 The proof of the main result via interpolation
We now turn to the proof of Theorem 1.1.
Proof
(a) Assume \(n/2<s\le n\) and let
We will prove that
for every \((\frac{1}{p_1},\frac{1}{p_2})\in \Gamma _1\), which is a convex set with vertices D, K, L, G, H and N (see Fig. 1a below). By multilinear real interpolation [4, Corollary 7.2.4], we only need to verify the boundedness of \(T_\sigma \) at points in \(\Gamma _1\) near its vertices D, K, L, G, H, N which do not lie in \( \Gamma _1\).
As showed in [4, 11], the Hörmander condition \(\sup _{j\in {\mathbb {Z}}}\Vert \sigma (2^j\cdot ){\widehat{\Psi }}\Vert _{L^r_s({\mathbb {R}}^{2n})}\) is invariant under duality. For \(1\le p<\infty \), by duality, if \(T_{\sigma }\) maps \(L^{p_1}\times L^{p_2}\rightarrow L^p\), then it also maps \(L^{p'}\times L^{p_2}\rightarrow L^{p_1'}\). Therefore, if \(T_{\sigma }\) is bounded near D, then \(T_{\sigma }\) is also bounded near N by duality. By symmetry, if \(T_\sigma \) is bounded near N, D and K then it is bounded near H, G and L as well. From these reductions, it remains to prove (24) at points in \(\Gamma _1\) near D and K.
With \(s_1>\frac{n}{2}\) and \(r_1s_1>2n\), we recall the following [6, Theorem 1]:
By duality it follows from (25) that when \(s_1>\frac{n}{2}\) and \(r_1s_1>2n\) we have
Theorem 1.1 in [17] (with \( s_1=s_2\) in [17] being \(\gamma \) below) implies that
for \(\gamma >\frac{n}{2}\), where \(1< q_1,q_2\le \infty \), \(\frac{1}{q}=\frac{1}{q_1}+\frac{1}{q_2}< \frac{2\gamma }{n}+ \frac{1}{2}\). Given \(s_2>n\), choose \(\gamma =\frac{s_2}{2}>\frac{n}{2}\) and observing the trivial estimate
we obtain
for all \(1< q_1,q_2\le \infty \), \(\frac{1}{q}=\frac{1}{q_1}+\frac{1}{q_2}< \frac{s_2}{n}+ \frac{1}{2}\).
We now use Theorem 3.1 to interpolate between (26) and (27) (for \(q_1=q\) near 1 and \(q_2=\infty \)). We obtain (24) at points \(D_1(\frac{1}{p_1},0)\) with \(\frac{1}{p_1}<\frac{s}{n}\) which are near the point \(D(\frac{s}{n},0)\). Similarly, interpolating between (25) and (27) (\(q_1\) near 1, \(q_2=2\)) yields (24) at points \(K_1(\frac{1}{p_1},\frac{1}{2})\) with \(\frac{1}{p_1}<\frac{s}{n}\) near \(K(\frac{s}{n},\frac{1}{2})\). This yields (24) on \(\Gamma _1\) and completes part (a).
(b) Assume \(n<s\le \frac{3n}{2}\). Since \(r\ge 2\), the Kato–Poince inequality [10] implies that
Combining estimates (28) and (27) yields (24) in the open pentagon OIRSJ union the open segments OI and OJ. This completes the second part of Theorem 1.1.
(c) In the last case when \(s>\frac{3n}{2}\), notice that condition (7) reduces to \(p> \frac{1}{2}\) and since
the case in part (b) applies and yields (24) for every point in the entire rhombus OITJ union the open segments OI and OJ. The proof of Theorem 1.1 is now complete. \(\square \)
5 An application
We consider the following multiplier on \({\mathbb {R}}^{2n}\): \(m_{a,b}(\xi _1,\xi _2) = \psi (\xi _1,\xi _2) |(\xi _1,\xi _2)|^{- b} e^{i|(\xi _1,\xi _2)|^a}\) where \(a > 0\), \(a \ne 1\), \(b > 0\), and \(\psi \) is a smooth function on \({\mathbb {R}}^{2n}\) which vanishes in a neighborhood of the origin and is equal to 1 in a neighborhood of infinity. One can verify that \(m_{a,b}\) satisfies (1) on \({\mathbb {R}}^{2n}\) with \(s = b/a\) and any \(r>2n/s\).
The range of p’s for which \(m_{a,b}\) is a bounded bilinear multiplier on \(L^p({\mathbb {R}}^{2n})\) can be completely described by the equation \( |\frac{1}{p}-\frac{1}{2}|\le \frac{b/a}{2n} \) (see Hirschman [12, comments after Theorem 3c], Wainger [22, Part II], and Miyachi [16, Theorem 3]); similar examples of multipliers of limited boundedness are contained in Miyachi and Tomita [17, Section 7].
As a consequence of Theorem 1.1 we obtain that the bilinear multiplier operator associated with \(m_{a,b}\) is bounded from \(L^{p_1}({\mathbb {R}}^n)\times L^{p_2}({\mathbb {R}}^n)\) to \(L^{p }(\mathbb R^n)\) in the following cases:
-
(i)
when \(n\ge b/a> n/2\) and
$$\begin{aligned} \frac{1}{p_1}<\frac{b}{an},\, \frac{1}{p_2}<\frac{b}{an},\, 1-\frac{b}{an}<\frac{1}{p}<\frac{b}{an}+\frac{1}{2}. \end{aligned}$$ -
(ii)
when \(3n/2\ge b/a> n \) and
$$\begin{aligned} \frac{1}{p } <\frac{b}{an}+\frac{1}{2}\, ; \end{aligned}$$ -
(iii)
when \(b/a >3n/2\) in the entire range of exponents \(1<p_1,p_2\le \infty \), \(\frac{1}{2}<p<\infty \).
The boundedness of this specific bilinear multiplier is unknown to us outside the above range of indices.
References
Calderón, A.P., Torchinsky, A.: Parabolic maximal functions associated with a distribution, II. Adv. Math. 24, 101–171 (1977)
Fujita, M., Tomita, N.: Weighted norm inequalities for multilinear Fourier multipliers. Trans. Am. Math. Soc. 364, 6335–6353 (2012)
Grafakos, L.: Classical Fourier Analysis, Graduate Texts in Mathematics, GTM 249, 3rd edn. Springer, New York (2014)
Grafakos, L.: Modern Fourier Analysis, Graduate Texts in Mathematics, GTM 250, 3rd edn. Springer, New York (2014)
Grafakos, L., He, D., Honzík, P., Nguyen, H.V.: The Hörmander multiplier theorem I: the linear case. Ill. J. Math. 61, 25–35 (2017)
Grafakos, L., He, D., Honzík, P.: The Hörmander multiplier theorem II: the local \(L^2\) case. Math. Zeit. 289, 875–887 (2018)
Grafakos, L., Miyachi, A., Nguyen, H.V., Tomita, N.: Multilinear Fourier multipliers with minimal Sobolev regularity, II. J. Math. Soc. Japan 69, 529–562 (2017)
Grafakos, L., Miyachi, A., Tomita, N.: On multilinear Fourier multipliers of limited smoothness. Can. J. Math. 65, 299–330 (2013)
Grafakos, L., Nguyen, H.V.: Multilinear Fourier multipliers with minimal Sobolev regularity, I. Colloq. Math. 144, 1–30 (2016)
Grafakos, L., Oh, S.: The Kato–Ponce inequality. Comm. PDE 39, 1128–1157 (2014)
Grafakos, L., Si, Z.: The Hörmander multiplier theorem for multilinear operators. J. Reine Angew. Math. 668, 133–147 (2012)
Hirschman Jr., I.I.: On multiplier transformations. Duke Math. J. 26, 221–242 (1959)
Hörmander, L.: Estimates for translation invariant operators in \(L^p\) spaces. Acta Math. 104, 93–139 (1960)
Kato, T., Ponce, G.: Commutator estimates and the Euler and Navier–Stokes equations. Comm. Pure Appl. Math. 41, 891–907 (1988)
Mikhlin, S.G.: On the multipliers of Fourier integrals. Dokl. Akad. Nauk SSSR (N.S.) 109, 701–703 (1956)
Miyachi, A.: On some Fourier multipliers for \(H^p({\mathbb{R}}^{n})\). J. Fac. Sci. Univ. Tokyo Sect. IA Math. 27, 157–179 (1980)
Miyachi, A., Tomita, N.: Minimal smoothness conditions for bilinear Fourier multipliers. Rev. Mat. Iberoam. 29, 495–530 (2013)
Miyachi, A., Tomita, N.: Boundedness criterion for bilinear Fourier multiplier operators. Tohoku Math. J. 66, 55–76 (2014)
Slavíková, L.: On the failure of the Hörmander multiplier theorem in a limiting case. Rev. Mat. Iberoamer (to appear)
Stein, E.M., Weiss, G.: On the interpolation of analytic families of operators acting on \(H^{p}\)-spaces. Tohoku Math. J. 9, 318–339 (1957)
Tomita, N.: A Hörmander type multiplier theorem for multilinear operators. J. Funct. Anal. 259, 2028–2044 (2010)
Wainger, S.: Special trigonometric series in k-dimensions. Mem. Am. Math. Soc. 59, 1–102 (1965)
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by G. Teschl.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The first author would like to thank the Simons Foundation.
Rights and permissions
About this article
Cite this article
Grafakos, L., Van Nguyen, H. The Hörmander multiplier theorem, III: the complete bilinear case via interpolation. Monatsh Math 190, 735–753 (2019). https://doi.org/10.1007/s00605-019-01300-x
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00605-019-01300-x