Abstract
Let \(L\) be a homogeneous sublaplacian on a \(2\)-step stratified Lie group \(G\) of topological dimension \(d\) and homogeneous dimension \(Q\). By a theorem due to Christ and to Mauceri and Meda, an operator of the form \(F(L)\) is bounded on \(L^p\) for \(1 < p < \infty \) if \(F\) satisfies a scale-invariant smoothness condition of order \(s > Q/2\). Under suitable assumptions on \(G\) and \(L\), here we show that a smoothness condition of order \(s > d/2\) is sufficient. This extends to a larger class of \(2\)-step groups the results for the Heisenberg and related groups by Müller and Stein and by Hebisch and for the free group \(N_{3,2}\) by Müller and the author.
Similar content being viewed by others
1 Introduction
Let \(L\) be a homogeneous sublaplacian on a stratified Lie group \(G\) of homogeneous dimension \(Q\). Since \(L\) is a positive self-adjoint operator on \(L^2(G)\), a functional calculus for \(L\) is defined via the spectral theorem and, for all Borel functions \(F : {\mathbb {R}}\rightarrow {\mathbb {C}}\), the operator \(F(L)\) is bounded on \(L^2(G)\) whenever the “spectral multiplier” \(F\) is bounded. As for the \(L^p\)-boundedness for \(p \ne 2\) of \(F(L)\), a sufficient condition in terms of smoothness properties of the multiplier \(F\) is given by a theorem of Mihlin–Hörmander type due to Christ [4] and Mauceri and Meda [20]: the operator \(F(L)\) is of weak type \((1,1)\) and bounded on \(L^p(G)\) for all \(p \in ]1,\infty [\) whenever
for some \(s > Q/2\), where \(W_2^s({\mathbb {R}})\) is the \(L^2\) Sobolev space of fractional order \(s\), and \(\eta \in C^\infty _c(]0,\infty [)\) is a nontrivial auxiliary function.
A natural question that arises is whether the smoothness condition \(s > Q/2\) is sharp. This is clearly true when \(G\) is abelian, so \(Q\) coincides with the topological dimension \(d\) of \(G\), and \(L\) is essentially the Laplace operator on \({\mathbb {R}}^d\). Take, however, the smallest nonabelian example of a stratified group, that is, the Heisenberg group \({\mathrm {H}}_1\), which is defined by endowing \({\mathbb {R}}\times {\mathbb {R}}\times {\mathbb {R}}\) with the group law
and with the automorphic dilations
\({\mathrm {H}}_1\) is a \(2\)-step stratified group, and the homogeneous dimension of \({\mathrm {H}}_1\) is \(4\). Nevertheless, a result by Müller and Stein [23] and Hebisch [12] shows that, for a homogeneous sublaplacian on \({\mathrm {H}}_1\), the smoothness condition on the multiplier can be pushed down to \(s > d/2\), where \(d = 3\) is the topological dimension of \({\mathrm {H}}_1\) (in [23], it is also proved that the condition \(s > d/2\) is sharp). Such an improvement of the Christ–Mauceri–Meda theorem holds not only for \({\mathrm {H}}_1\), but for the larger class of Métivier groups (and for direct products of Métivier and abelian groups), and also for differential operators other than sublaplacians (see, e.g., [13, 17]); moreover, as shown subsequently by Cowling and Sikora [5] (see also [6]), the sharp result on \({\mathrm {H}}_1\) can be obtained by transplantation from an analogous result for a distinguished sublaplacian on the (nonstratified) group \(\mathrm {SU}_2\) (which in turn improves, in the case of \(\mathrm {SU}_2\), an extension of the Christ–Mauceri–Meda theorem to spaces of homogeneous type [1, 7, 11]). However, it is still an open question whether, for a general stratified Lie group (or even for a general \(2\)-step stratified group), the homogeneous dimension in the smoothness condition can be replaced by the topological dimension.
The aim of this paper is to extend the class of the \(2\)-step stratified groups and sublaplacians for which the smoothness condition in the multiplier theorem can be pushed down to half the topological dimension.
Take for instance the Heisenberg–Reiter group \({\mathrm {H}}_{{d_1},{d_2}}\) (cf. [27]), defined by endowing \({\mathbb {R}}^{{d_2}\times {d_1}} \times {\mathbb {R}}^{d_1}\times {\mathbb {R}}^{d_2}\) with the group law (1) and the automorphic dilations (2); here, however, \({\mathbb {R}}^{{d_2}\times {d_1}}\) is the set of the real \({d_2}\times {d_1}\) matrices, and the products \(xy',x'y\) in (1) are interpreted in the sense of matrix multiplication. \({\mathrm {H}}_{{d_1},{d_2}}\) is a \(2\)-step stratified group of homogeneous dimension \(Q = {d_1}{d_2}+ {d_1}+ 2{d_2}\) and topological dimension \(d = {d_1}{d_2}+ {d_1}+ {d_2}\). Despite the formal similarity with \({\mathrm {H}}_1\), the group \({\mathrm {H}}_{{d_1},{d_2}}\) does not fall into the class of Métivier groups, unless \({d_2}= 1\) (in fact, \({\mathrm {H}}_{{d_1},1}\) is the \((2{d_1}+1)\)-dimensional Heisenberg group \({\mathrm {H}}_{d_1}\)). Nevertheless, the technique presented here allows one to handle the case \({d_2}> 1\) too.
Namely, let \(X_{1,1},\dots ,X_{{d_2},{d_1}},Y_1,\dots ,Y_{d_1},U_1,\dots ,U_{d_2}\) be the left-invariant vector fields on \({\mathrm {H}}_{{d_1},{d_2}}\) extending the standard basis of \({\mathbb {R}}^{{d_2}\times {d_1}} \times {\mathbb {R}}^{d_1}\times {\mathbb {R}}^{d_2}\) at the identity, and define the homogeneous sublaplacian \(L\) by
Then, a particular instance of our main result reads as follows.
Theorem 1
Suppose that a function \(F : {\mathbb {R}}\rightarrow {\mathbb {C}}\) satisfies
for some \(s > d/2\). Then, the operator \(F(L)\) is of weak type \((1,1)\) and bounded on \(L^p({\mathrm {H}}_{{d_1},{d_2}})\) for all \(p \in ]1,\infty [\).
To the best of our knowledge, this result is new, at least in the case \({d_2}> {d_1}\). In fact, in the case \({d_2}\le {d_1}\), the extension described in [17] of the technique of [12, 13] would give the same result. However, the technique presented here is different, and yields the result irrespective of the parameters \({d_1},{d_2}\).
The left quotient of \({\mathrm {H}}_{{d_1},{d_2}}\) by the subgroup \({\mathbb {R}}^{{d_2}\times {d_1}} \times \{0\} \times \{0\}\) gives a homogeneous space diffeomorphic to \({\mathbb {R}}^{d_1}\times {\mathbb {R}}^{d_2}\), and the sublaplacian \(L\) corresponds in the quotient to a Grushin operator. In recent joint works with Sikora [18] and Müller [14], we proved for these Grushin operators on \({\mathbb {R}}^{d_1}\times {\mathbb {R}}^{d_2}\) a sharp spectral multiplier theorem of Mihlin–Hörmander type, where the smoothness requirement is again half the topological dimension of the ambient space.
The proofs in [14, 18] rely heavily on properties of the eigenfunction expansions for the Hermite operators. Since a homogeneous sublaplacian on a \(2\)-step stratified group reduces to a Hermite operator in almost all irreducible unitary representations of the group, it is conceivable that an adaptation of the methods of [14, 18] may give an improvement to the multiplier theorem for \(2\)-step stratified groups, even outside of the Métivier setting. A first result in this direction is shown in [19], where the free \(2\)-step nilpotent Lie group \(N_{3,2}\) on three generators is considered, and properties of Laguerre polynomials are exploited (somehow in the spirit of [21, 23, 24]). The argument presented here refines and extends the one in [19].
Theorem 1 above is just a particular case of the result presented here, and we refer the reader to the next section for a precise statement. We remark that the analog of Theorem 1 holds on \({\mathrm {H}}_{{d_1},{d_2}}\) when the sublaplacian \(L\) has the more general form
where \(X_{0,j} = Y_j\) and \((a^j_{k,k'})_{k,k'=0,\dots ,{d_2}}\) is a positive-definite symmetric matrix for all \(j\in \{1,\dots ,{d_1}\}\). Other groups can be considered too, e.g., the complexification of a Heisenberg–Reiter group, or the quotient of the direct product of \({\mathrm {H}}_{1,3}\) and \(N_{3,2}\) given by identifying the respective centers.
2 The general setting
Let \(G\) be a connected, simply connected nilpotent Lie group of step \(2\). Recall that, via exponential coordinates, \(G\) may be identified with its Lie algebra \(\mathfrak {g}\), that is, the tangent space of \(G\) at the identity. In turn, \(\mathfrak {g}\) may be identified with the Lie algebra of left-invariant vector fields on \(G\). We refer to [9] for the basic definitions and further details.
Let \(\mathfrak {g}\) be decomposed as \({\mathfrak {v}}\oplus {\mathfrak {z}}\), where \({\mathfrak {z}}\) is the center of \(\mathfrak {g}\), and let \(\langle \cdot , \cdot \rangle \) be an inner product on \({\mathfrak {v}}\). The sublaplacian \(L\) associated with the inner product is defined by \(L = -\sum \nolimits _j X_j^2\), where \(\{X_j\}_j\) is any orthonormal basis of \({\mathfrak {v}}\). Note that, vice versa, by the Poincaré–Birkhoff–Witt theorem, any second-order operator \(L\) of the form \(-\sum \nolimits _j X_j^2\) for some basis \(\{X_j\}_j\) of \(\mathfrak {g}\) modulo \({\mathfrak {z}}\) determines uniquely a linear complement \({\mathfrak {v}}= {{\mathrm{{\mathrm {span}}}}}\{X_j\}_j\) of \({\mathfrak {z}}\) and an inner product on \({\mathfrak {v}}\) such that \(\{X_j\}_j\) is orthonormal.
Let \({\mathfrak {z}}^*\) be the dual of \(\mathfrak {z}\) and, for all \(\eta \in {\mathfrak {z}}^*\), define \(J_\eta \) as the linear endomorphism of \({\mathfrak {v}}\) such that \(\eta ([z,z']) = \langle J_\eta z, z' \rangle \) for all \(z,z'\in {\mathfrak {v}}\). Clearly, \(J_\eta \) is skewadjoint with respect to the inner product; hence, \(J_\eta ^2\) is self-adjoint and negative semidefinite, with even rank, for all \(\eta \in {\mathfrak {z}}^*\). Set moreover \({\mathfrak {\dot{z}}}= {\mathfrak {z}}^* {\setminus } \{0\}\).
Assumption (A)
There exist integers \(r_1,\dots ,r_{d_1}> 0\) and an orthogonal decomposition \({\mathfrak {v}}= {\mathfrak {v}}_1 \oplus \dots \oplus {\mathfrak {v}}_{d_1}\) such that, if \(P_1,\dots ,P_{d_1}\) are the corresponding orthogonal projections, then \(J_\eta P_j= P_jJ_\eta \) and \(J^2_\eta P_j\) has rank \(2r_j\) and a unique nonzero eigenvalue for all \(\eta \in {\mathfrak {\dot{z}}}\) and all \(j\in \{1,\dots ,{d_1}\}\).
Note that from Assumption \(({\mathrm {A}})\) it follows that \(J_\eta \ne 0\) for all \(\eta \in {\mathfrak {\dot{z}}}\). Therefore \([{\mathfrak {v}},{\mathfrak {v}}] = {\mathfrak {z}}\), that is, the decomposition \(\mathfrak {g} = {\mathfrak {v}}\oplus {\mathfrak {z}}\) is a stratification of \(\mathfrak {g}\), and the sublaplacian \(L\) is hypoelliptic.
In fact, \(J_\eta \) has constant rank \(2(r_1 + \dots + r_k)\) for all \(\eta \in {\mathfrak {\dot{z}}}\). If \(J_\eta \) is invertible for all \(\eta \in {\mathfrak {\dot{z}}}\), then \(G\) is a Métivier group, and if in particular \(J_\eta ^2 = -|\eta |^2 {\mathrm {id}}_{{\mathfrak {v}}}\) for some inner product norm \(|\cdot |\) on \({\mathfrak {z}}^*\), then \(G\) is an H-type group. The main novelty of our Assumption \(({\mathrm {A}})\) is that it allows \(J_\eta \) to have a nonzero kernel when \(\eta \in {\mathfrak {\dot{z}}}\), although the dimension of the kernel must be constant.
The fact that \(J_\eta \) has constant rank for \(\eta \in {\mathfrak {\dot{z}}}\) depends only on the algebraic structure of \(G\). What depends on the inner product, that is, on the sublaplacian \(L\), are the values and multiplicities of the eigenvalues of the \(J_\eta \). The above Assumption \(({\mathrm {A}})\) asks for a sort of simultaneous diagonalizability of the \(J_\eta \).
Under our Assumption \(({\mathrm {A}})\) on the group \(G\) and the sublaplacian \(L\), we are able to prove the following multiplier theorem.
Theorem 2
Suppose that a function \(F : {\mathbb {R}}\rightarrow {\mathbb {C}}\) satisfies
for some \(s > (\dim G)/2\). Then, the operator \(F(L)\) is of weak type \((1,1)\) and bounded on \(L^p(G)\) for all \(p \in ]1,\infty [\).
The previously mentioned Heisenberg–Reiter groups \({\mathrm {H}}_{{d_1},{d_2}}\) satisfy Assumption \(({\mathrm {A}})\), where the inner product is determined by the sublaplacian (3), and the orthogonal decomposition of the first layer is given by the natural isomorphism \({\mathbb {R}}^{{d_2}\times {d_1}} \times {\mathbb {R}}^{{d_1}} \cong ({\mathbb {R}}^{{d_2}} \times {\mathbb {R}})^{d_1}\). Other examples are the free \(2\)-step nilpotent Lie group \(N_{3,2}\) on \(3\) generators, considered in [19], and its complexification \(N_{3,2}^{\mathbb {C}}\). Moreover, if \(G_1\) and \(G_2\) satisfy Assumption \(({\mathrm {A}})\), and their centers have the same dimension, then the quotient of \(G_1 \times G_2\) given by any linear identification of the centers satisfy Assumption \(({\mathrm {A}})\). Note that the direct product \(G_1 \times G_2\) itself does not satisfy Assumption \(({\mathrm {A}})\), but an adaptation of the argument presented here allows one to consider that case too. We postpone to the end of this paper a more detailed discussion of these remarks.
From now on, unless otherwise specified, we assume that \(G\) and \(L\) are a \(2\)-step stratified group and a homogeneous sublaplacian on \(G\) satisfying Assumption \(({\mathrm {A}})\). Since \(L\) is a left-invariant operator, so is any operator of the form \(F(L)\). Let \({{\mathrm{{\mathcal {K}}}}}_{F(L)}\) denote the convolution kernel of \(F(L)\). As shown, e.g., by [17, Theorem 4.6], the previous theorem is a consequence of the following estimate.
Proposition 3
For all \(s > (\dim G)/2\), there exists a weight \(w_s : G \rightarrow [1,\infty [\) such that \(w_s^{-1} \in L^2(G)\) and, for all compact sets \(K \subseteq {\mathbb {R}}\) and for all functions \(F : {\mathbb {R}}\rightarrow {\mathbb {C}}\) with \({{\mathrm{{\mathrm {supp}}}}}F \subseteq K\),
in particular,
The rest of the paper, except for the last section, is devoted to the proof of this estimate.
3 The joint functional calculus
Let \({d_2}= \dim {\mathfrak {z}}\), and let \(U_1,\dots ,U_{d_2}\) be any basis of the center \({\mathfrak {z}}\). Let moreover the “partial sublaplacian” \(L_j\) be defined as \(L_j= - \sum \nolimits _l X^2_{j,l}\), where \(\{X_{j,l}\}_l\) is any orthonormal basis of \({\mathfrak {v}}_j\), for all \(j\in \{1,\dots ,{d_1}\}\); in particular \(L = L_1 + \dots + L_{d_1}\). Then, the left-invariant differential operators
are essentially self-adjoint and commute strongly; hence, they admit a joint functional calculus (see, e.g., [16]). Therefore, if \({\mathbf {L}}\) and \({\mathbf {U}}\) denote the “vectors of operators” \((L_1,\dots ,L_{d_1})\) and \((-iU_1,\dots ,-iU_{d_2})\), and if we identify \({\mathfrak {z}}^*\) with \({\mathbb {R}}^{d_2}\) via the dual basis of \(U_1,\dots ,U_n\), then, for all bounded Borel functions \(H : {\mathbb {R}}^{d_1}\times {\mathfrak {z}}^* \rightarrow {\mathbb {C}}\), the operator \(H({\mathbf {L}},{\mathbf {U}})\) is defined and bounded on \(L^2(G)\). Moreover, \(H({\mathbf {L}},{\mathbf {U}})\) is left-invariant, and we can express its convolution kernel \({{\mathrm{{\mathcal {K}}}}}_{H({\mathbf {L}},{\mathbf {U}})}\) in terms of Laguerre functions.
Namely, for all \(n,k \in {\mathbb {N}}\), let
be the \(n\)-th Laguerre polynomial of type \(k\), and define
Note that, by Assumption \(({\mathrm {A}})\), for all \(\eta \in {\mathfrak {\dot{z}}}\) and \(j\in \{1,\dots ,{d_1}\}\),
for some orthogonal projection \(P_j^\eta \) of rank \(2r_j\) and some \(b_j^\eta > 0\). Set moreover
Modulo reordering the \({\mathfrak {v}}_j\) in the decomposition of \({\mathfrak {v}}\), we may suppose that there exists \({\tilde{d_1}}\in \{0,\dots ,{d_1}\}\) such that \(\dim {\mathfrak {v}}_j> 2r_j\) if \(j\le {\tilde{d_1}}\), and \(\dim {\mathfrak {v}}_j= 2r_j\) if \(j > {\tilde{d_1}}\). In particular, \(\bar{P}^\eta _j= 0\) and \(P^\eta _j= P_j\) for all \(j> {\tilde{d_1}}\) and \(\eta \in {\mathfrak {\dot{z}}}\). We will also use the abbreviations \(r = (r_1,\dots ,r_{d_1}),\,{\mathbb {R}}^r = {\mathbb {R}}^{r_1} \times \dots \times {\mathbb {R}}^{r_{d_1}},\,{\mathbb {N}}^r = {\mathbb {N}}^{r_1} \times \dots \times {\mathbb {N}}^{r_{d_1}},\,|r| = r_1 + \dots + r_{d_1}\). Moreover \(\langle \cdot , \cdot \rangle \) will also denote the duality pairing \(\mathfrak {z}^* \times \mathfrak {z} \rightarrow {\mathbb {R}}\).
Proposition 4
Let \(H : {\mathbb {R}}^{{d_1}} \times {\mathfrak {z}}^* \rightarrow {\mathbb {C}}\) be in the Schwartz class, and set
for all \(n \in {\mathbb {N}}^{d_1},\,\mu \in {\mathbb {R}}^{{\tilde{d_1}}},\,\eta \in {\mathfrak {\dot{z}}}\). Then, for all \((z,u) \in G\),
Proof
For all \(\eta \in {\mathfrak {\dot{z}}}\) and \(j\in \{1,\dots ,{d_1}\}\), let \(E^\eta _{j,1},\bar{E}^\eta _{j,1},\dots ,E^\eta _{j,r_j}, \bar{E}^\eta _{j,r_j}\) be an orthonormal basis of the range of \(P^\eta _j\) such that
Hence, for all \(z \in {\mathfrak {v}},\,\eta \in {\mathfrak {\dot{z}}}\), and \(j\in \{1,\dots ,{d_1}\}\), we can write
for some uniquely determined \(z^\eta _{j,l},\bar{z}^\eta _{j,l} \in {\mathbb {R}}\); set then \(z^\eta _j= (z^\eta _{j,1},\dots ,z^\eta _{j,r_j}),\,\bar{z}^\eta _j= (\bar{z}^\eta _{j,1},\dots ,\bar{z}^\eta _{j,r_j})\), and moreover \(z^\eta = (z^\eta _1,\dots ,z^\eta _{d_1})\) and \(\bar{z}^\eta = (\bar{z}^\eta _1,\dots ,\bar{z}^\eta _{d_1})\).
For all \(\eta \in {\mathfrak {\dot{z}}}\) and all \(\rho \in \ker J_\eta \), an irreducible unitary representation \(\pi _{\eta ,\rho }\) of \(G\) on \(L^2({\mathbb {R}}^r)\) is defined by
for all \((z,u) \in G,\,v \in {\mathbb {R}}^r,\,\phi \in L^2({\mathbb {R}}^r)\), where \(\bar{P}^\eta = \bar{P}^\eta _1 + \dots + \bar{P}^\eta _{{\tilde{d_1}}}\) is the orthogonal projection onto \(\ker J_\eta \). Following, e.g., [2, §2], one can see that these representations are sufficient to write the Plancherel formula for the group Fourier transform of \(G\), and the corresponding Fourier inversion formula:
for all \(f : G \rightarrow {\mathbb {C}}\) in the Schwartz class and all \((z,u) \in G\), where \(\pi _{\eta ,\rho }(f) = \int _{G} f(g) \, \pi _{\eta ,\rho }(g^{-1}) \,dg\).
Fix \(\eta \in {\mathfrak {\dot{z}}}\) and \(\rho \in \ker J_\eta \). The operators (6) are represented in \(\pi _{\eta ,\rho }\) as
for all \(j\in \{1,\dots ,{d_1}\}\) and \(k\in \{1,\dots ,{d_2}\}\), where \(v_j\in {\mathbb {R}}^{r_j}\) denotes the \(j\)-th component of \(v \in {\mathbb {R}}^r\), and \(\varDelta _{v_j}\) denotes the corresponding partial Laplacian. Let \(h_\ell \) denote the \(\ell \)-th Hermite function, that is,
and, for all \(\omega \in {\mathbb {N}}^r\), define \(\tilde{h}_{\eta ,\omega } : {\mathbb {R}}^r \rightarrow {\mathbb {R}}\) by
for all \(j\in \{1,\dots ,{d_1}\}\), where \(\omega _{j,l}\) and \(v_{j,l}\) denote the \(l\)-th components of \(\omega _j\in {\mathbb {N}}^{r_j}\) and \(v_j\in {\mathbb {R}}^{r_j}\). Then, \(\{\tilde{h}_{\eta ,\omega }\}_{\omega \in {\mathbb {N}}^r}\) is a complete orthonormal system for \(L^2({\mathbb {R}}^r)\), made of joint eigenfunctions of the operators (10). In fact,
where \(|\omega _j| = \omega _{j,1} + \dots + \omega _{j,r_j}\); it should be observed that \(P_j\rho = 0\) if \(j> {\tilde{d_1}}\).
Since \(H : {\mathbb {R}}^{d_1}\times {\mathfrak {z}}^* \rightarrow {\mathbb {C}}\) is in the Schwartz class, \({{\mathrm{{\mathcal {K}}}}}_{H({\mathbf {L}},{\mathbf {U}})} : G \rightarrow {\mathbb {C}}\) is in the Schwartz class too (see [3, Theorem 5.2] or [15, §4.2]). Moreover,
by (11) and [22, Proposition 1.1]; hence, if \(\varphi _{\eta ,\rho ,\omega }(z,u) = \langle \pi _{\eta ,\rho }(z,u) \tilde{h}_{\eta ,\omega }, \tilde{h}_{\eta ,\omega } \rangle \) is the corresponding diagonal matrix coefficient of \(\pi _{\eta ,\rho }\), then
Therefore, (9) gives that
where
On the other hand,
The last integral is essentially the Fourier–Wigner transform of a pair of Hermite functions, whose bidimensional Fourier transform is a Fourier–Wigner transform too [10, formula (1.90)]. The parity properties of the Hermite functions then yield
Since the Fourier–Wigner transform of a pair of Hermite functions can be expressed in terms of Laguerre polynomials (see [10, Theorem 1.104] or [26, Theorem 1.3.4]), we obtain that
Consequently, for all \(n \in {\mathbb {N}}^{d_1}\),
[9, §10.12, formula (41)]. The conclusion then follows by plugging (13) into (12) and performing a change of variable by rotation in the inner integrals. \(\square \)
4 A weighted Plancherel estimate
Proposition 4 expresses the convolution kernel \({{\mathrm{{\mathcal {K}}}}}_{H({\mathbf {L}},{\mathbf {U}})}\) as the inverse Fourier transform of a function of the multiplier \(H\). Due to the properties of the Fourier transform, it is not unreasonable to think that multiplying the kernel by a polynomial weight might correspond to taking derivatives of the multiplier. As a matter of fact, the presence of the Laguerre expansion leads us to consider both “discrete” and “continuous” derivatives of the reparametrization \(m : {\mathbb {N}}^{d_1}\times {\mathbb {R}}^{{\tilde{d_1}}} \times {\mathfrak {\dot{z}}}\rightarrow {\mathbb {C}}\) of the multiplier \(H\) given by (7).
For convenience, set \({\mathcal {L}}_n^{(k)} = 0\) for all \(n < 0\). From the properties of Laguerre polynomials (see, e.g., [9, §10.12]), one can easily derive the following identities.
Lemma 5
For all \(k,n,m \in {\mathbb {N}}\) and \(t \in {\mathbb {R}}\),
Let \(e_1,\dots ,e_{d_1}\) denote the standard basis of \({\mathbb {R}}^{d_1}\). We introduce some operators on functions \(f : {\mathbb {N}}^{d_1}\times {\mathbb {R}}^{{\tilde{d_1}}} \times {\mathfrak {\dot{z}}}\rightarrow {\mathbb {C}}\):
for all \(j\in \{1,\dots ,{d_1}\},\,l \in \{1,\dots ,{\tilde{d_1}}\},\,k\in \{1,\dots ,{d_2}\}\).
For all \(h \in {\mathbb {N}}\) and all multiindices \(\alpha \in {\mathbb {N}}^h\), we denote by \(|\alpha |\) the length \(\alpha _1+\dots +\alpha _h\) of \(\alpha \). Inequalities between multiindices, such as \(\alpha \le \alpha '\), shall be interpreted componentwise. Set moreover \((\alpha )_+ = ((\alpha _1)_+,\dots ,(\alpha _h)_+)\), where \((\ell )_+ = \max \{\ell ,0\}\).
A function \(\varPsi : {\mathfrak {\dot{z}}}\!\times \! {\mathfrak {v}}\!\rightarrow \! {\mathbb {C}}\) will be called multihomogeneous if there exist \(h_0,h_1,\dots ,h_{d_1}\!\in \! {\mathbb {R}}\) such that
for all \(\eta \in {\mathfrak {\dot{z}}},\,\xi \in {\mathfrak {v}},\,\lambda _0,\lambda _1,\dots ,\lambda _{d_1}\in ]0,\infty [\); the homogeneity degrees \(h_0,h_1,\dots ,h_{d_1}\) of \(\varPsi \) will also be denoted as \(\deg _{\mathfrak {z}}\varPsi , \deg _{{\mathfrak {v}}_1} \varPsi , \dots , \deg _{{\mathfrak {v}}_{d_1}} \varPsi \). Note that, if \(\varPsi \) is multihomogeneous and continuous, then \(\deg _{{\mathfrak {v}}_j} \varPsi \ge 0\) for all \(j\in \{1,\dots ,{d_1}\}\).
Proposition 6
Let \(H : {\mathbb {R}}^{d_1}\times {\mathfrak {z}}^* \rightarrow {\mathbb {C}}\) be smooth and compactly supported in \({\mathbb {R}}^{d_1}\times {\mathfrak {\dot{z}}}\), and let \(m(n,\mu ,\eta )\) be defined by (7). For all \(\alpha \in {\mathbb {N}}^{d_2}\),
for almost all \((z,u) \in G\), where \(I_\alpha \) is a finite set and, for all \(\iota \in I_\alpha \),
-
\(\gamma ^\iota \in {\mathbb {N}}^{d_2},\,\theta ^\iota \in {\mathbb {N}}^{{\tilde{d_1}}},\,\beta ^\iota \in {\mathbb {N}}^{d_1},\,\gamma ^\iota \le \alpha \),
-
\(\varPsi _\iota = \varPsi _{\iota ,0} \varPsi _{\iota ,1} \dots \varPsi _{\iota ,{d_1}}\), where \(\varPsi _{\iota ,j} : {\mathfrak {\dot{z}}}\times {\mathfrak {v}}\rightarrow {\mathbb {C}}\) is smooth and multihomogeneous for all \(j\in \{0,\dots ,{d_1}\}\),
-
\(\deg _{\mathfrak {z}}\varPsi _\iota = |\gamma ^\iota | - |\alpha | - |\beta ^\iota |\) and \(\deg _{{\mathfrak {v}}_j} \varPsi _\iota = 2\beta ^\iota _j+ 2\theta ^\iota _j\) for all \(j\in \{1,\dots ,{d_1}\}\),
-
for all \(j\in \{1,\dots ,{d_1}\},\,\varPsi _{\iota ,j}(\eta ,\xi )\) is a product of factors of the form \(|P_j^\eta \xi |^2\) or \(\partial _{\eta _k} |P^\eta _j\xi |^2\) for \(k\in \{1,\dots ,{d_2}\}\),
-
\(|\gamma ^\iota | + |\theta ^\iota | + |\beta ^\iota | + \sum \nolimits _{j=1}^{{d_1}} (\beta ^\iota _j- (\deg _{{\mathfrak {v}}_j} \varPsi _{\iota ,j}) /2)_+ \le |\alpha |\).
Proof
By Proposition 4 and the properties of the Fourier transform, we are reduced to proving that, for all \(\alpha \in {\mathbb {N}}^{{d_2}},\,\eta \in {\mathfrak {\dot{z}}},\,\xi \in {\mathfrak {v}}\),
where \(I_\alpha ,\,\gamma ^\iota ,\,\theta ^\iota ,\,\beta ^\iota ,\,\varPsi _\iota \) are as in the above statement.
This is easily proved by induction on \(|\alpha |\). For \(|\alpha | = 0\), it is trivially verified. For the inductive step, one applies Leibniz’ rule and exploits the following observations:
-
when a derivative \(\partial _{\eta _k}\) hits a Laguerre function, by the identity (15) and summation by parts, the type of the Laguerre function is increased by \(1\), as well as the corresponding component of \(\beta ^\iota \);
-
for all \(j\in \{1,\dots ,{d_1}\},\,b^\eta _j= \sqrt{{{\mathrm{{\mathrm {tr}}}}}(-J_\eta ^2 P_j)/(2r_j)}\) is a smooth function of \(\eta \in {\mathfrak {\dot{z}}}\), homogeneous of degree \(1\);
-
for all \(j\in \{1,\dots ,{d_1}\},\,P^\eta _j= -J_\eta ^2 P_j/(b^\eta _j)^2\) is a smooth function of \(\eta \in {\mathfrak {\dot{z}}}\), homogeneous of degree \(0\), and in fact it is constant if \(j> {\tilde{d_1}}\);
-
for all \(j\in \{1,\dots ,{\tilde{d_1}}\},\,|P^\eta _j\xi |^2 = \langle P^\eta _jP_j\xi , P_j\xi \rangle \) is a smooth bihomogeneous function of \((\eta ,P_j\xi ) \in {\mathfrak {\dot{z}}}\times {\mathfrak {v}}_j\) of bidegree \((0,2)\), and moreover
$$\begin{aligned}&|\bar{P}^\eta _j\xi |^2 = |P_j\xi |^2 - |P^\eta _j\xi |^2, \qquad \partial _{\eta _k} |\bar{P}^\eta _j\xi |^2 = - \partial _{\eta _k} |P^\eta _j\xi |^2, \\&\partial _{\eta _k} \left( |P^\eta _j\xi |^2/b^\eta _j\right) = |P^\eta _j\xi |^2 \partial _{\eta _k} \left( 1/b^\eta _j\right) + \left( \partial _{\eta _k} |P^\eta _j\xi |^2\right) /b^\eta _j\end{aligned}$$for all \(k\in \{1,\dots ,{d_2}\}\).
The conclusion follows. \(\square \)
Note that, for all \(j\!\in \{1,\dots ,{d_1}\},\,\mu \!\in {\mathbb {R}}^{{\tilde{d_1}}},\,\eta \in {\mathfrak {\dot{z}}}\), the quantities \(\tau _jf(\cdot ,\mu ,\eta ),\,\delta _jf(\cdot ,\mu ,\eta )\) depend only on \(f(\cdot ,\mu ,\eta )\); in other words, \(\tau _j\) and \(\delta _j\) can be considered as operators on functions \({\mathbb {N}}^{d_1}\rightarrow {\mathbb {C}}\).
The following lemma exploits the orthogonality properties (16) of the Laguerre functions, together with (14), and shows that a mismatch between the type of the Laguerre function and the exponent of the weight attached to the measure may be turned in some cases into discrete differentiation.
Lemma 7
For all \(h,k \in {\mathbb {N}}^{d_1}\) and all compactly supported \(f : {\mathbb {N}}^{d_1}\rightarrow {\mathbb {C}}\),
Proof
Via an inductive argument, we may reduce to the case \({d_1}= 1\).
Note that, if \(f\) is compactly supported, then \(\tau ^l f\) is null for all sufficiently large \(l \in {\mathbb {N}}\). Hence, the operator \(1+\tau \), when restricted to the set of compactly supported functions, is invertible, with inverse given by
Then by (14), we deduce that, for all \(k \in {\mathbb {N}}\),
and consequently, for all \(h,k \in {\mathbb {N}}\),
Thus, the orthogonality properties (16) of the Laguerre functions give us that
where \(\langle n \rangle = 1+n\).
In the case \(h \ge k,\,(1+\tau )^{h-k}\) is given by the finite sum
and the conclusion follows immediately by the triangular inequality.
In the case \(h < k\), instead, since \(\delta = \tau - 1\), from the identity \(1-\tau ^2 = (1-\tau )(1+\tau )\), we deduce that
hence
by the Cauchy–Schwarz inequality, and we are done. \(\square \)
Let \(|\cdot |\) denote any Euclidean norm on \(\mathfrak {z}^*\). The previous lemma, together with Plancherel’s formula for the Fourier transform, yields the following \(L^2\)-estimate.
Proposition 8
Under the hypotheses of Proposition 6, for all \(\alpha \in {\mathbb {N}}^{d_2}\),
where \(\tilde{I}_\alpha \) is a finite set and, for all \(\iota \in \tilde{I}_\alpha \),
-
\(\gamma ^\iota \in {\mathbb {N}}^{d_2},\,\theta ^\iota \in {\mathbb {N}}^{{\tilde{d_1}}},\,a^\iota , \beta ^\iota \in {\mathbb {N}}^{d_1}\),
-
\(\gamma ^\iota \le \alpha ,\,|\gamma ^\iota | + |\theta ^\iota | + |\beta ^\iota | \le |\alpha |\),
-
\(\sigma _\iota \) is a regular Borel measure on \([0,\infty [^{{\tilde{d_1}}}\).
Proof
Note that, for all \(j\in \{1,\dots ,{d_1}\}\),
consequently, if \(\varPsi _\iota ,\varPsi _{\iota ,j},\gamma ^\iota ,\theta ^\iota ,\beta ^\iota \) are as in the statement of Proposition 6, then
for all \(j\in \{1,\dots ,{d_1}\}\), hence
and moreover, for all \(h \in {\mathbb {N}}^{d_1}\), if \(h_j \ge (\deg _{{\mathfrak {v}}_j} \varPsi _{\iota ,j})/2\) for all \(j \in \{1,\dots ,{d_1}\}\), then
By Proposition 6, Plancherel’s formula and the triangular inequality, we then obtain that the left-hand side of (17) is majorized by a finite sum of terms of the form
where \(\gamma \in {\mathbb {N}}^{d_2},\,\theta ,k \in {\mathbb {N}}^{{\tilde{d_1}}},\,\beta ,h \in {\mathbb {N}}^{d_1}\) and \(|\gamma |+|\theta |+|\beta + (\beta -h)_+| \le |\alpha |\). Simple changes of variables (rotation, polar coordinates and rescaling) allow one to rewrite (18) as a constant times
By exploiting the fact that the \(b^\eta _j\) are smooth functions of \(\eta \in {\mathfrak {\dot{z}}}\), homogeneous of degree \(1\) (see the proof of Proposition 6), and applying Lemma 7 to the inner integral, the last quantity is majorized by
and since the exponents \(k_j+(\dim {\mathfrak {v}}_j-2r_j)/2\) are strictly positive, while
and \(|\gamma |+|\theta |+|\beta + (\beta -h)_+| \le |\alpha |\), the conclusion follows by suitably renaming the multiindices. \(\square \)
5 From discrete to continuous
Via the fundamental theorem of integral calculus, finite differences can be estimated by continuous derivatives. The next lemma is a multivariate analog of [19, Lemma 6], and we omit the proof (see also [18, Lemma 7]).
Lemma 9
Let \(f : {\mathbb {N}}^{d_1}\rightarrow {\mathbb {C}}\) have a smooth extension \(\tilde{f} : [0,\infty [^{d_1}\rightarrow {\mathbb {C}}\), and let \(\beta \in {\mathbb {N}}^{d_1}\). Then,
for all \(n \in {\mathbb {N}}\), where \(J_\beta = \prod _{j=1}^{d_1}[0,\beta _j]\), and \(\nu _\beta \) is a Borel probability measure on \(J_\beta \). In particular,
for all \(n \in {\mathbb {N}}^{d_1}\).
We give now a simplified version of the right-hand side of (17), in the case we restrict to the functional calculus of \(L\) alone. In order to avoid issues of divergent series, it is, however, convenient at first to truncate the multiplier along the spectrum of \({\mathbf {U}}\).
Lemma 10
Let \(\chi \in C^\infty _c({\mathbb {R}})\) be supported in \([1/2,2 ],\,K \subseteq {\mathbb {R}}\) be compact and \(M \in ]0, \infty [\). If \(F : {\mathbb {R}}\rightarrow {\mathbb {C}}\) is smooth and supported in \(K\), and \(F_M : {\mathbb {R}}\times {\mathfrak {z}}^* \rightarrow {\mathbb {C}}\) is given by
then, for all \(r \in [0, \infty [\),
Proof
We may restrict to the case \(r \in {\mathbb {N}}\), the other cases being recovered a posteriori by interpolation. Hence, we need to prove that
for all \(\alpha \in {\mathbb {N}}^{{d_1}}\). On the other hand, if \(m\) is defined by
where \(\langle \ell \rangle _{j} = 2\ell + r_j\) and \(|\mu |_\varSigma = \sum \nolimits _{j=1}^{\tilde{d_1}}\mu _j\), then the left-hand side of (19) is majorized by the right-hand side of (17), and we are reduced to proving that
for all \(\iota \in \tilde{I}_\alpha \), where \(\tilde{I}_\alpha ,\,\gamma ^\iota ,\,\theta ^\iota ,\,\beta ^\iota ,\,a^\iota ,\,\sigma _\iota \) are as in Proposition 8.
Note that the right-hand side of (20) makes sense for all \(n \in {\mathbb {R}}^{d_1}\) and defines a smooth extension of \(m\), which we still denote by \(m\) by a slight abuse of notation. Hence, by Lemma 9,
where \(J_\iota = \prod _{j=1}^{d_1}[0,\beta ^\iota _j]\) and \(\nu _\iota \) is a suitable probability measure on \(J_\iota \). Moreover, the measure \(\sigma _\iota \) in (21) is finite on compacta, and the right-hand side of (22) vanishes when \(|\mu |_\varSigma > \max K\), because \({{\mathrm{{\mathrm {supp}}}}}F \subseteq K\). Consequently, (21) will be proved if we show that
for all \(s \in J_\iota \) and \(\mu \in [0,\max K ]^{{\tilde{d_1}}}\), uniformly in \(s\) and \(\mu \).
As observed in the proof of Proposition 6, the \(b_j^\eta \) are positive, smooth functions of \(\eta \in {\mathfrak {\dot{z}}}\), homogeneous of degree \(1\); therefore, for all \(n \in {\mathbb {N}}^{d_1},\,j\in \{1,\dots ,{d_1}\},\,\eta \in {\mathfrak {\dot{z}}}, s \in [0,\infty [^{d_1},\,\mu \in [0,\infty [^{{\tilde{d_1}}}\),
and the last quantity is bounded by the constant \(\max K\) whenever \((n+s,\mu ,\eta ) \in {{\mathrm{{\mathrm {supp}}}}}m\), because \({{\mathrm{{\mathrm {supp}}}}}F \subseteq K\). Hence, the factors \(|\eta | (1+ n_j)\) in the left-hand side of (23) can be discarded, that is, we are reduced to proving (23) in the case \(a^\iota = 0\).
From (20), it follows immediately that
and then it is easily proved inductively that
where \(\varPsi _{\iota ,\upsilon ,q} : {\mathfrak {\dot{z}}}\rightarrow {\mathbb {R}}\) is smooth and homogeneous of degree \(|\beta ^\iota | + |\upsilon | + q - |\gamma ^\iota |\). By exploiting again (24) and the fact that \({{\mathrm{{\mathrm {supp}}}}}F \subseteq K\), we can majorize the factors \(\langle n_j \rangle _j\) in the right-hand side by \(|\eta |^{-1} \sim M^{-1}\) and obtain that
where \(\tilde{\chi }\) is the characteristic function of \([1/2,2 ]\). Hence, the left-hand side of (23), when \(a^\iota = 0\), is majorized by
Let \(S\) denote the unit sphere in \({\mathfrak {z}}^*\). By passing to polar coordinates and exploiting the homogeneity of the \(b_j^\eta \), the integral in the above formula is majorized by
where \(\langle n \rangle _{\omega ,s} = \sum \nolimits _{j=1}^{d_1}b_j^\omega \langle n_j+s_j\rangle _j\sim 1+|n|\) uniformly in \(\omega \in S\) and \(s \in J_\iota \). Since \(\tilde{\chi }(\rho /(M \langle n \rangle _{\omega ,s}))\) vanishes unless \(\langle n \rangle _{\omega ,s} \sim \rho /M\), the sum in the right-hand side of (25) has at most \(C_\iota (\rho /M)^{d_1}\) nonvanishing summands, and the integral on \(S\) is majorized by \(C_\iota (\rho /M)^{d_1}\). In conclusion, the left-hand side of (23) is majorized by
because \(d_1 \ge 1,\,{{\mathrm{{\mathrm {supp}}}}}F \subseteq K\) and \(|\beta ^\iota | + |\theta ^\iota | + |\gamma ^\iota | \le |\alpha |\), and we are done. \(\square \)
Proposition 11
Let \(F : {\mathbb {R}}\rightarrow {\mathbb {C}}\) be smooth and such that \({{\mathrm{{\mathrm {supp}}}}}F \subseteq K\) for some compact set \(K \subseteq {\mathbb {R}}\). For all \(r \in [0, {d_2}/2 [\),
Proof
Take \(\chi \in C^\infty _c(]0,\infty [)\) such that \({{\mathrm{{\mathrm {supp}}}}}\chi \subseteq [1/2,2 ]\) and \(\sum \nolimits _{k \in {\mathbb {Z}}} \chi (2^{-k} t) = 1\) for all \(t \in ]0,\infty [\). If \(F_M\) is defined for all \(M \in ]0,\infty [\) as in Lemma 10, then \({{\mathrm{{\mathcal {K}}}}}_{F_M(L,{\mathbf {U}})}\) is given by the right-hand side of (8), where \(m\) is defined by (20), and moreover,
for all \(\eta \in {\mathfrak {\dot{z}}},\,\mu \in [0,\infty [^{\tilde{d_1}}\) and \(n \in {\mathbb {N}}^{d_1}\), therefore \(F_M(L,{\mathbf {U}}) = 0\) whenever \(M > 2C\max K\). Hence, if \(k_K \in {\mathbb {Z}}\) is sufficiently large so that \(2^{k_K} > 2 C \max K\), then
(with convergence in the strong sense). Consequently, an estimate for \({{\mathrm{{\mathcal {K}}}}}_{F(L)}\) can be obtained, via Minkowski’s inequality, by summing the corresponding estimates for \({{\mathrm{{\mathcal {K}}}}}_{F_{2^{k}}}(L,{\mathbf {U}})\) given by Lemma 10. If \(r < d/2\), then the series \(\sum \nolimits _{k \le k_K} (2^{k})^{{d_2}/2-r}\) converges, thus
The conclusion follows by combining the last inequality with the corresponding one for \(r = 0\). \(\square \)
Let \(|\cdot |_\delta \) be a \(\delta _t\)-homogeneous norm on \(G\); take, e.g., \(|(z,u)|_\delta = |z| + |u|^{1/2}\). Interpolation then allows us to improve the standard weighted estimate for a homogeneous sublaplacian on a stratified group.
Proposition 12
Let \(F : {\mathbb {R}}\rightarrow {\mathbb {C}}\) be smooth and such that \({{\mathrm{{\mathrm {supp}}}}}F \subseteq K\) for some compact set \(K \subseteq {\mathbb {R}}\). For all \(r \in [0, {d_2}/2 [,\,\alpha \ge 0\) and \(\beta > \alpha + r\),
Proof
Note that \(1+|u| \le C (1+|(z,u)|_\delta )^2\). Hence, in the case \(\alpha \ge 0,\,\beta > \alpha + 2r\), the inequality (26) follows by the mentioned standard estimate (see [21, Lemma 1.2] or [17, Theorem 2.7]). On the other hand, if \(\alpha = 0\) and \(\beta \ge r\), then (26) is given by Proposition 11. The full range of \(\alpha \) and \(\beta \) is then obtained by interpolation. \(\square \)
We can finally prove the crucial estimate.
Proof of Proposition 3
Take \(r \in ](\dim G)/2 + {d_2}/2-s,{d_2}/2 [\). Then,
hence we can find \(\alpha _1 > (\dim {\mathfrak {v}})/2\) and \(\alpha _2 > {d_2}-2r\) such that \(s - r > \alpha _1 + \alpha _2\). Set \(w_s(z,u) = (1+|(z,u)|_\delta )^\alpha \, (1+|u|)^r\). The \(L^2\)-estimate (4) then follows from Proposition 12. On the other hand, for all \((z,u) \in G\),
and the right-hand side is integrable over \(G \cong \mathfrak {v} \times \mathfrak {z}\) since \(2\alpha _1 > \dim {\mathfrak {v}}\) and \(\alpha _2 + 2r > {d_2}= \dim {\mathfrak {z}}\). Therefore, \(w_s^{-1} \in L^2(G)\), and the \(L^1\)-estimate (5) follows from (4) and Hölder’s inequality. \(\square \)
6 Remarks on the validity of the assumption and direct products
In this section, we do no longer suppose that \(G\) and \(L\) are a \(2\)-step stratified Lie group and a sublaplacian satisfying Assumption \(({\mathrm {A}})\).
As observed in Sect. 2, a necessary condition for the validity of Assumption \(({\mathrm {A}})\) is that the skewadjoint endomorphism \(J_\eta \) of the first layer \(\mathfrak {v}\) has constant rank for \(\eta \) ranging in \({\mathfrak {\dot{z}}}= {\mathfrak {z}}^* {\setminus } \{0\}\). Here, we show that this condition is also sufficient when the rank is minimal.
Proposition 13
Let \(G\) be a \(2\)-step nilpotent Lie group, with Lie algebra \(\mathfrak {g} = \mathfrak {v} \oplus \mathfrak {z}\), and let \(\langle \cdot ,\cdot \rangle \) be an inner product on \(\mathfrak {v}\). Suppose that the skewadjoint endomorphism \(J_\eta \) of \(\mathfrak {v}\) has rank \(2\) for all \(\eta \in {\mathfrak {\dot{z}}}\). Then, \(G\) satisfies Assumption \(({\mathrm {A}})\) with the sublaplacian \(L\) associated to the given inner product, and also with any other sublaplacian associated to an inner product on a complement of \(\mathfrak {z}\).
Let moreover \(G_{\mathbb {C}}\) be the complexification of \(G\), considered as a real \(2\)-step group, with Lie algebra \(\mathfrak {g}_{\mathbb {C}}= \mathfrak {v}_{\mathbb {C}}\oplus \mathfrak {z}_{\mathbb {C}}\), and let \(\mathfrak {v}_{\mathbb {C}}\) be endowed with the real inner product induced by the inner product on \(\mathfrak {v}\). Then, \(G_{\mathbb {C}}\), with the sublaplacian associated to the given inner product, satisfies Assumption \(({\mathrm {A}})\).
Proof
From the normal form for skewadjoint endomorphisms, it follows immediately that, if \(J_\eta \) has rank \(2\), then \(J_\eta ^2\) has exactly one nonzero eigenvalue, and Assumption \(({\mathrm {A}})\) is trivially verified. Moreover, if \(\mathfrak {v}\) is identified with \(\mathfrak {g}/\mathfrak {z}\), then \(\ker J_\eta \) corresponds to the subspace
of \(\mathfrak {g}/\mathfrak {z}\); hence, the rank condition on \(J_\eta \) can be rephrased by saying that \(N_\eta \) has codimension \(2\) for all \(\eta \in {\mathfrak {\dot{z}}}\), and this condition does not depend on the sublaplacian \(L\) chosen on \(G\).
Let \(R(J_\eta )\) denote the range of \(J_\eta \). We show now that, for all \(\eta ,\eta ' \in {\mathfrak {\dot{z}}}\), the intersection \(R(J_\eta ) \cap R(J_{\eta '})\) is nontrivial. If it were trivial, since \(J_{\eta +\eta '} = J_\eta + J_\eta '\), we would have \(\ker J_{\eta +\eta '} = \ker J_\eta \cap \ker J_{\eta '}\), hence
thus \(J_{\eta +\eta '}\) would have rank \(4\), contradiction.
Consider now the complexification \(\mathfrak {g}_{\mathbb {C}}= \mathfrak {g} \oplus i \mathfrak {g}\). Via the linear identifications \(\mathfrak {g}_{\mathbb {C}}= \mathfrak {g} \times \mathfrak {g},\,\mathfrak {z}^*_{\mathbb {C}}= \mathfrak {z}^* \times \mathfrak {z}^*,\,\mathfrak {v}_{\mathbb {C}}= \mathfrak {v} \times \mathfrak {v}\), the skewsymmetric endomorphism \(\tilde{J}_\eta \) of the first layer \(\mathfrak {v}_{\mathbb {C}}\) corresponding to the element \(\eta =(\eta _R,\eta _I) \in \mathfrak {z}_{\mathbb {C}}^*\) is given by
Take now \(\eta = (\eta _R,\eta _I) \in {\mathfrak {\dot{z}}}_{\mathbb {C}}\); we want to show that \(\tilde{J}_\eta ^2\) has rank \(4\) and a unique nonzero eigenvalue. We distinguish several cases.
If \(\eta _I = 0\), then \(\tilde{J}_\eta = J_{\eta _R} \times (-J_{\eta _R})\), hence \(\tilde{J}_\eta ^2 = J_{\eta _R}^2 \times J_{\eta _R}^2\) satisfies the condition. The same argument gives the conclusion in the case \(\eta _R = 0\).
If both \(\eta _R,\eta _I \in {\mathfrak {\dot{z}}}\), then \(R(J_{\eta _R}) \cap R(J_{\eta _I}) \ne 0\), hence \(\dim (R(J_{\eta _R}) \cap R(J_{\eta _I}))\) is either \(2\) or \(1\). In the first case, \(R(J_{\eta _R}) = R(J_{\eta _I})\), so \(J_{\eta _R}\) and \(J_{\eta _I}\) commute and (27) implies that
since \(J_{\eta _R}^2\) and \(J_{\eta _I}^2\) are negative multiples of the same orthogonal projection, the conclusion follows.
Suppose now that \(R(J_{\eta _R}) \cap R(J_{\eta _I}) = {\mathbb {R}}x\) for some unit vector \(x \in \mathfrak {v}\), and set \(y_R = J_{\eta _R}x,\,y_I = J_{\eta _I}x,\,b_R = |y_R|,\,b_I = |y_I|\); in particular, \(J_{\eta _R}^2 x = -b_R^2 x\) and \(J_{\eta _I}^2 x = -b_I^2 x\). Since \(J_{\eta _R}\) and \(J_{\eta _I}\) are skewadjoint and of rank \(2\), necessarily \(J_{\eta _R} x, J_{\eta _I}x \in x^\perp \) and \(J_{\eta _R}(x^\perp ) = J_{\eta _I}(x^\perp ) = {\mathbb {R}}x\), therefore \(J_{\eta _R} J_{\eta _I} x\) and \(J_{\eta _I} J_{\eta _R} x\) are both multiples of \(x\); on the other hand,
hence \(J_{\eta _R} J_{\eta _I} x = J_{\eta _I} J_{\eta _R} x\). This identity, together with (27), allows us easily to show that
Note that \(b_R^2 + b_I^2\) is the squared norm of both \((y_R,y_I)\) and \((y_I,-y_R)\). Hence, we would be done if we knew that \(R(\tilde{J}_\mu )\) coincides with the linear span \(W\) of \((x,0),\,(0,x),\,(y_R,y_I),\,(y_I,-y_R)\).
In fact, we just need to show that \(R(\tilde{J}_\eta )\) is contained in \(W\), or equivalently, that \(W^\perp \) is contained in \(\ker \tilde{J}_\eta \). On the other hand, if \(v = (v_R,v_I) \in W^\perp \), then \(v_R,v_I \in x^\perp \) and moreover
hence \(J_{\eta _R} v_R, J_{\eta _R} v_I, J_{\eta _I} v_R, J_{\eta _I} v_I \in {\mathbb {R}}x\), and
therefore \(J_{\eta _R} v_R = -J_{\eta _I} v_I\) and \(J_{\eta _I} v_R = J_{\eta _R} v_I\), from which it follows immediately that \(\tilde{J}_\eta (v_R,v_I) = 0\). \(\square \)
The next proposition shows how groups and sublaplacians satisfying Assumption \(({\mathrm {A}})\) may be “glued together”, so to give a higher-dimensional group and a sublaplacian that satisfy Assumption \(({\mathrm {A}})\) too.
Proposition 14
Suppose that, for \(j=1,2\), the sublaplacian \(L_j\) on the \(2\)-step stratified Lie group \(G_j\) satisfies Assumption \(({\mathrm {A}})\). Suppose further that the centers of \(G_1\) and \(G_2\) have the same dimension. Let \(G\) be the quotient of \(G_1 \times G_2\) given by any linear identification of the respective centers, and let \(L = L_1^\sharp + L_2^\sharp \), where \(L_j^\sharp \) is the pushforward of \(L_j\) to \(G\). Then, the sublaplacian \(L\) on the group \(G\) satisfies Assumption \(({\mathrm {A}})\).
Proof
Let \(\mathfrak {g}_j\) be the Lie algebra of \(G_j\), and let \({\mathfrak {v}}_j\) and \(\langle \cdot , \cdot \rangle _j\) be the linear complement of the center \({\mathfrak {z}}_j\) and the inner product on \({\mathfrak {v}}_j\) determined by the sublaplacian \(L_j\); denote moreover by \(J_{j,\eta }\) the skewadjoint endomorphism of \({\mathfrak {v}}_j\) determined by \(\eta \in {\mathfrak {z}}_j^*\).
The linear identification of the centers of \(G_1\) and \(G_2\) corresponds to a linear isomorphism \(\phi : \mathfrak {z}_1 \rightarrow \mathfrak {z}_2\), and the Lie algebra \(\mathfrak {g}\) of the quotient \(G\) can be identified with \(\mathfrak {v}_1 \times \mathfrak {v}_2 \times \mathfrak {z}_2\), with Lie bracket
Then, the sublaplacian \(L\) on \(G\) corresponds to the inner product \(\langle \cdot , \cdot \rangle \) on \(\mathfrak {v}_1 \times \mathfrak {v}_2\) defined by
In particular, if \(\phi ^* : \mathfrak {z}_2^* \rightarrow \mathfrak {z}_1^*\) denotes the adjoint map of \(\phi : \mathfrak {z}_1 \rightarrow \mathfrak {z}_2\), then it is easily checked that the skewadjoint endomorphism of the first layer \(\mathfrak {v}_1 \times \mathfrak {v}_2\) of \(\mathfrak {g}\) corresponding to an element \(\eta \) of the dual \(\mathfrak {z}_2^*\) of the center of \(\mathfrak {g}\) is given by \(J_\eta = J_{1,\phi ^* \eta } \times J_{2,\eta }\). Hence, the orthogonal decomposition of \(\mathfrak {v}_1 \times \mathfrak {v}_2\) giving the “simultaneous diagonalization” of the \(J_\eta \) for all \(\eta \in {\mathfrak {\dot{z}}}_2\) (in the sense of Sect. 2) is simply obtained by juxtaposing the corresponding orthogonal decompositions of \(\mathfrak {v}_1\) and \(\mathfrak {v}_2\). \(\square \)
Note that the direct product \(G_1 \times G_2\) itself need not satisfy Assumption \(({\mathrm {A}})\), even if the factors \(G_1\) and \(G_2\) do. However, a functional-analytic argument, as in [24, §4], can be used to deal with that case.
The key step in our proof of Theorem 2 is the weighted \(L^2\)-estimate (4) of Proposition 3. Let us now turn the conclusion of Proposition 3 into an assumption on a homogeneous sublaplacian \(L\) on a stratified group \(G\).
Assumption \((\hbox {B}_{t})\). For all \(s > t\), there exist a weight \(w_s : G \rightarrow [1,\infty [\) such that \(w_s^{-1} \in L^2(G)\) and, for all compact sets \(K \subseteq {\mathbb {R}}\) and all Borel functions \(F : {\mathbb {R}}\rightarrow {\mathbb {C}}\) with \({{\mathrm{{\mathrm {supp}}}}}F \subseteq K\),
Our Proposition 3 can then be rephrased by saying that Assumption \(({\mathrm {A}})\) implies Assumption \(({\mathrm {B}}_{t})\) for \(t = (\dim G)/2\). Note, on the other hand, that Assumption \(({\mathrm {B}}_{t})\) makes sense for homogeneous sublaplacians on stratified groups \(G\) of step other than \(2\). In fact, every homogeneous sublaplacian on a stratified group of homogeneous dimension \(Q\) satisfies Assumption \(({\mathrm {B}}_{t})\) for \(t = Q/2\), by [21, Lemma 1.2] (suitably extended so to admit multipliers that do not vanish in a neighborhood of the origin of \({\mathbb {R}}\); see, e.g., [24, Lemma 3.1] for the \(1\)-dimensional case, and [17, Theorem 2.7] for the higher-dimensional case).
Differently from Assumption \(({\mathrm {A}})\), the new Assumption \(({\mathrm {B}}_{t})\) “behaves well” under direct products.
Proposition 15
For \(j=1,\dots ,n\), let \(L_j\) be a homogeneous sublaplacian on a stratified Lie group \(G_j\) satisfying Assumption \(({\mathrm {B}}_{t_j})\) for some \(t_j > 0\). Let \(G = G_1 \times \dots \times G_n\) and \(L = L_1^\sharp + \dots + L_n^\sharp \), where \(L_j^\sharp \) is the pushforward to \(G\) of the operator \(L_j\). Then, the sublaplacian \(L\) on \(G\) satisfies Assumption \(({\mathrm {B}}_{t})\), where \(t = t_1+\dots +t_n\).
Proof
Take \(s > t\). Then, we can choose \(s_1,\dots ,s_n\) such that \(s_1 > t_1,\dots ,s_n> t_n\) and \(s = s_1+\dots +s_n\). Let then \(w_{j,s_j} : G_j \rightarrow [1,\infty [\) be the weight corresponding to \(s_j\) given by Assumption \(({\mathrm {B}}_{t_j})\) on \(G_j\) and \(L_j\), for \(j=1,\dots ,n\). In particular, \(w_{j,s_j}^{-1} \in L^2(G_j)\) and, for all \(\phi \in C^\infty _c({\mathbb {R}})\), the map \(F \mapsto {{\mathrm{{\mathcal {K}}}}}_{(\phi F)(L_j)}\) is a bounded linear map of Hilbert spaces \(W_2^{s_j}({\mathbb {R}}) \rightarrow L^2(G_j,w_{j,s_j}^2(x_j) \,dx_j)\), where \(dx_j\) denotes the Haar measure on \(G_j\).
The operators \(L_1^\sharp ,\dots ,L_n^\sharp \) are essentially self-adjoint and commute strongly, that is, they admit a joint spectral resolution and a joint functional calculus on \(L^2(G)\), and moreover, for all bounded Borel functions \(F_1,\dots ,F_n : {\mathbb {R}}\rightarrow {\mathbb {C}}\),
[16, Corollary 5.5]. Hence, for all \(\phi _1,\dots ,\phi _n \in C^\infty _c({\mathbb {R}})\), if \(\phi = \phi _1 \otimes \dots \otimes \phi _n\), then the map \(H \mapsto {{\mathrm{{\mathcal {K}}}}}_{(\phi H)(L_1^\sharp ,\dots ,L_n^\sharp )}\) is the tensor product of the maps \(F_j \mapsto {{\mathrm{{\mathcal {K}}}}}_{(\phi _j F_j)(L_j)}\). Since these maps are bounded \(W_2^{s_j}({\mathbb {R}}) \rightarrow L^2(G_j,w_{j,s_j}^2(x_j) \,dx_j)\), the map \(H \mapsto {{\mathrm{{\mathcal {K}}}}}_{(\phi H)(L_1^\sharp ,\dots ,L_n^\sharp )}\) is bounded \(S_2^{(s_1,\dots ,s_n)}W({\mathbb {R}}^n) \rightarrow L^2(G,w_s^2(x) \,dx)\), where \(S^{(s_1,\dots ,s_n)}_2 W({\mathbb {R}}^n) = W_2^{s_1}({\mathbb {R}}) \otimes \dots \otimes W_2^{s_n}({\mathbb {R}})\) is the \(L^2\) Sobolev space with dominating mixed smoothness [25] of order \((s_1,\dots ,s_n)\), and \(w_s = w_{1,s_1} \otimes \dots \otimes w_{n,s_n}\) is the product weight on \(G\). In particular, for all compact sets \(K \subseteq {\mathbb {R}}\), if we choose the cutoffs \(\phi _j \in C^\infty _c({\mathbb {R}})\) so that \(\phi _j|_K = 1\), then we deduce that, for all \(H : {\mathbb {R}}^n \rightarrow {\mathbb {C}}\) with \({{\mathrm{{\mathrm {supp}}}}}H \subseteq K^n\),
(cf. [17, Proposition 5.2]). Since
where \(\hat{f}\) denotes the Euclidean Fourier transform of \(f\), we see immediately that the estimate
holds true whenever \(K \subseteq {\mathbb {R}}\) is compact and \(H : {\mathbb {R}}^n \rightarrow {\mathbb {C}}\) is supported in \(K^n\).
Take now a compact set \(K \subseteq {\mathbb {R}}\) and choose a smooth cutoff \(\eta _K \in C^\infty _c({\mathbb {R}})\) such that \(\eta _K|_{[0,\max K ]} = 1\). Let \(F : {\mathbb {R}}\rightarrow {\mathbb {C}}\) be such that \({{\mathrm{{\mathrm {supp}}}}}F \subseteq K\), and define \(H : {\mathbb {R}}^n \rightarrow {\mathbb {C}}\) by
for all \((\lambda _1,\dots ,\lambda _n) \in {\mathbb {R}}^n\). Then, \({{\mathrm{{\mathrm {supp}}}}}H \subseteq ({{\mathrm{{\mathrm {supp}}}}}\eta _K)^n\), and
for all \((\lambda _1,\dots ,\lambda _n) \in [0,\infty [^n\). Since the operators \(L_1,\dots ,L_n\) are nonnegative, the joint spectrum of \(L_1^\sharp ,\dots ,L_n^\sharp \) is contained in \([0,\infty [^n\), hence
Consequently, by (29) and the smoothness of the map \((\lambda _1,\dots ,\lambda _n) \mapsto \lambda _1+\dots +\lambda _n\), we obtain that
Since clearly \(w_s^{-1} = w_{1,s_1}^{-1} \otimes \dots \otimes w_{n,s_n}^{-1} \in L^2(G)\), we are done. \(\square \)
The previous results, together with the known weighted estimates for abelian [24, Lemma 3.1] and Métivier [12, 13, 17] groups, then yield the following extension of Theorem 2.
Theorem 16
For \(j=1,\dots ,n\), suppose that \(L_j\) is a homogeneous sublaplacian on a stratified Lie group \(G_j\). Suppose further that, for each \(j \in \{1,\dots ,n\}\), at least one of the following conditions holds:
-
\(G_j\) and \(L_j\) satisfy Assumption \(({\mathrm {A}})\);
-
\(G_j\) is a Métivier group;
-
\(G_j\) is abelian.
Let \(G = G_1 \times \dots \times G_n\) and \(L = L_1^\sharp + \dots + L_n^\sharp \), as in Proposition 15. If \(F : {\mathbb {R}}\rightarrow {\mathbb {C}}\) satisfies
for some \(s > (\dim G)/2\), then \(F(L)\) is of weak type \((1,1)\) and bounded on \(L^p(G)\) for all \(p \in ]1,\infty [\).
References
Alexopoulos, G.: Spectral multipliers on Lie groups of polynomial growth. Proc. Am. Math. Soc. 120(3), 973–979 (1994)
Astengo, F., Cowling, M., Di Blasio, B., Sundari, M.: Hardy’s uncertainty principle on certain Lie groups. J. Lond. Math. Soc. (2) 62(2), 461–472 (2000)
Astengo, F., Di Blasio, B., Ricci, F.: Gelfand pairs on the Heisenberg group and Schwartz functions. J. Funct. Anal. 256(5), 1565–1587 (2009)
Christ, M.: \(L^{p}\) bounds for spectral multipliers on nilpotent groups. Trans. Am. Math. Soc. 328(1), 73–81 (1991)
Cowling, M., Sikora, A.: A spectral multiplier theorem for a sublaplacian on \({\rm SU(2)}\). Math. Z. 238(1), 1–36 (2001)
Cowling, M., Klima, O., Sikora, A.: Spectral multipliers for the Kohn sublaplacian on the sphere in \({\mathbb{C}}^{n}\). Trans. Am. Math. Soc. 363(2), 611–631 (2011)
Duong, X.T., Ouhabaz, E.M., Sikora, A.: Plancherel-type estimates and sharp spectral multipliers. J. Funct. Anal. 196(2), 443–485 (2002)
Erdélyi, A., Magnus, W., Oberhettinger, F., Tricomi, F.G.: Higher Transcendental Functions, vol. II. Robert E. Krieger Publishing Co., Inc., Melbourne (1981)
Folland, G.B., Stein, E.M.: Hardy Spaces on Homogeneous Groups, Mathematical Notes, vol. 28. Princeton University Press, Princeton (1982)
Folland, G.B.: Harmonic Analysis in Phase Space. Ann. Math. Stud. vol. 122. Princeton University Press, Princeton (1989)
Hebisch, W.: Functional calculus for slowly decaying kernels (1995). Preprint. Available on the web at http://www.math.uni.wroc.pl/~hebisch/
Hebisch, W.: Multiplier theorem on generalized Heisenberg groups. Colloq. Math. 65(2), 231–239 (1993)
Hebisch, W., Zienkiewicz, J.: Multiplier theorem on generalized Heisenberg groups. II. Colloq. Math. 69(1), 29–36 (1995)
Martini, A., Müller, D.: A sharp multiplier theorem for Grushin operators in arbitrary dimensions (2012). To appear in, Rev. Mat. Iberoam. arXiv:1210.3564
Martini, A.: Algebras of differential operators on Lie groups and spectral multipliers. Tesi di perfezionamento (PhD thesis), Scuola Normale Superiore, Pisa (2010). arXiv:1007.1119
Martini, A.: Spectral theory for commutative algebras of differential operators on Lie groups. J. Funct. Anal. 260(9), 2767–2814 (2011)
Martini, A.: Analysis of joint spectral multipliers on Lie groups of polynomial growth. Ann. Inst. Fourier (Grenoble) 62(4), 1215–1263 (2012)
Martini, A., Sikora, A.: Weighted Plancherel estimates and sharp spectral multipliers for the Grushin operators. Math. Res. Lett. 19(5), 1075–1088 (2012)
Martini, A., Müller, D.: \(L^{p}\) spectral multipliers on the free group \(N_{3,2}\). Studia Math. 217(1), 41–55 (2013)
Mauceri, G., Meda, S.: Vector-valued multipliers on stratified groups. Rev. Mat. Iberoam. 6(3–4), 141–154 (1990)
De Michele, L., Mauceri, G.: \(L^{p}\) multipliers on the Heisenberg group. Mich. Math. J. 26(3), 361–371 (1979)
Müller, D.: A restriction theorem for the Heisenberg group. Ann. Math. (2) 131(3), 567–587 (1990)
Müller, D., Stein, E.M.: On spectral multipliers for Heisenberg and related groups. J. Math. Pures Appl. (9) 73(4), 413–440 (1994)
Müller, D., Ricci, F., Stein, E.M.: Marcinkiewicz multipliers and multi-parameter structure on Heisenberg (-type) groups. II. Math. Z. 221(2), 267–291 (1996)
Schmeisser, H.J.: Recent developments in the theory of function spaces with dominating mixed smoothness. In: Nonlinear Analysis, Function Spaces and Applications. Proceedings of the Spring School held in Prague, May 30-June 6, 2006, vol. 8, pp. 145–204. Czech Academy of Sciences, Mathematical Institute, Praha (2007)
Thangavelu, S.: Lectures on Hermite and Laguerre Expansions, Mathematical Notes, vol. 42. Princeton University Press, Princeton (1993)
Torres Lopera, J.F.: The cohomology and geometry of Heisenberg–Reiter nilmanifolds. In: Differential geometry, Peñíscola 1985, Lecture Notes in Math., vol. 1209, pp. 292–301. Springer, Berlin (1986)
Acknowledgments
I gratefully acknowledge the financial support of the Alexander von Humboldt Foundation. I am deeply grateful to Detlef Müller for stimulating discussions and his continuous encouragement. I am also grateful for the opportunity of visiting the Mathematical Institute of the University of Wrocław in the first bimester of 2011; numerous discussions, particularly with Jacek Dziubański and Waldemar Hebisch, have greatly contributed to my understanding of problems related to spectral multipliers.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
About this article
Cite this article
Martini, A. Spectral multipliers on Heisenberg–Reiter and related groups. Annali di Matematica 194, 1135–1155 (2015). https://doi.org/10.1007/s10231-014-0414-6
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10231-014-0414-6
Keywords
- Nilpotent Lie groups
- Heisenberg–Reiter groups
- Spectral multipliers
- Sublaplacians
- Mihlin–Hörmander multipliers
- Singular integral operators