Titchmarsh theorems for Fourier transforms of Hölder–Lipschitz functions on compact homogeneous manifolds

In this paper we extend classical Titchmarsh theorems on the Fourier transform of Hölder–Lipschitz functions to the setting of compact homogeneous manifolds. As an application, we derive a Fourier multiplier theorem for L2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L^2$$\end{document}-Hölder–Lipschitz spaces on compact Lie groups. We also derive conditions and a characterisation for Dini–Lipschitz classes on compact homogeneous manifolds in terms of the behaviour of their Fourier coefficients.


Introduction
The studies of the convergence and of the rate of decay of Fourier coefficients are among the most classical problems in Fourier analysis. Starting from the Riemann-Lebesgue theorem relating the integrability of a function on the torus T 1 and the convergence of its Fourier coefficients, through the Hausdorff-Young inequality relating the integrability of a function and of its Fourier transform. Furthermore one can relate the smoothness of a function on the torus and the rate of decay of its Fourier coefficients: indeed if f ∈ C k (T 1 ) for some k ≥ 1, one has that is, | j| k | f ( j)| → 0 as j → ∞. An extension of such statement below the index k = 1 can be obtained in the setting of Hölder regularity. If f satisfies the Hölder condition of order 0 < α ≤ 1 then This time we only get big-O (Landau's notation). For the case α > 1 2 this estimate can be improved as a consequence of the classical Bernstein theorem (see Bernstein [3] or Zygmund [44, vol. I, p. 240]): if f satisfies the Hölder condition of order α > 1 2 then its Fourier series converges absolutely. It was also established by Bernstein that the index 1 2 is sharp. An introduction to the classical topics discussed above can be found in [34].
The Hausdorff-Young inequality on a locally compact abelian group G 1 states that f ∈ L q if f ∈ L p (G 1 ) for 1 ≤ p ≤ 2 and 1 p + 1 q = 1. In [37], Edward Charles Titchmarsh studied the problem of how much this fact can be strengthened if f additionally satisfies a Lipschitz condition in the case of G 1 = R. In fact, Titchmarsh showed that the information on f can be significantly improved. In [39,41,43], M. Younis extended the Titchmarsh theorems to the case of the circle and to the very specific case of compact groups, namely, the 0-dimensional groups. Other studies of Lipschitz conditions have been done in [19] in term of the modulus of continuity. However, our aim in this paper is to derive suitable characterisations in terms of the Fourier coefficients of the functions.
The aim of this paper is to extend the classical Titchmarsh theorems to the setting of general compact homogeneous manifolds. In particular, this includes higher dimensional tori, compact Lie groups, real, complex or quaternionic spheres, projective spaces, and many others. As an application of such an extension, we derive a Fourier multiplier theorem for L 2 -Lipschitz spaces.
The analysis of various spaces and functional inequalities on compact Lie groups and compact homogeneous manifolds, despite its own interest and long history, has also a range of applications linking the local properties of functions with the global spectral properties of the group and to its representation theory, see e.g. Pesenson [21] for Bernstein-Nikolskii inequalities and links to the approximation theory, [22] for links to frames construction and Geller and Pesenson [17] for Besov spaces, Kushpel [18] for cubature formulae, [5] for Gevrey spaces with further links to weakly hyper-bolic partial differential equations involving Laplacian and sub-Laplacian, [20] for Nikolskii inequality and Besov spaces, as well as further references in these papers. The convergence of the Fourier series of functions on compact Lie groups has a long history, see e.g., to mention only very few, Taylor [36], Sugiura [35], and some review in Faraut's book [12]. We can also refer to [2] for conditions for Lipschitz functions on compact totally disconnected groups, [23] on symmetric spaces of rank 1, [25] for Fourier series on Vilenkin groups (compact, metrisable, 0-dimensional, abelian groups), and results for Lipschitz multipliers in [4,24,26,27] on compact 0dimensional, locally compact abelian, and abelian 0-dimensional groups, respectively.
We note that for convolution operators (on compact Lie groups) the rate of decay of Fourier coefficients of the kernel is also related to the Schatten-von Neumann properties of the operator, see e.g. [6,8].
We also note that anisotropic Hölder spaces (with respect to the group structure) have been also analysed on the Heisenberg group in [16] (see also [15]) with applications to partial differential equations, and some Fourier transform description was considered in [42]. Anisotropic Hölder spaces on more general Carnot groups were recently applied in [29] to questions in the potential theory.
Thus, the first Titchmarsh theorem that we deal with (Titchmarsh [37,Theorem 84]) in its version on the torus ([39, Theorem 2.6]) is recalled as Theorem A below. First, for 0 < α ≤ 1 and 1 < p ≤ ∞, we define the space Lip T 1 (α; p) by We will now briefly outline our main results in the setting of general compact homogeneous manifolds. If K is a closed subgroup of a compact Lie group G, the quotient space G/K can be canonically identified with an analytic manifold M. Henceforth M will denote a compact homogeneous manifold. In the case when K = {e} is the identity of the group G, we have M = G is a compact Lie group itself. To simplify the notation in the sequel we will identify representations of G with their equivalence classes.
We denote by G the unitary dual of G, and by G 0 the subset of G consisting of representations ξ of class I with respect to the subgroup K : this means that [ξ ] ∈ G 0 if ξ has at least one non-zero invariant vector with respect to K , i.e. that for some non-zero a in the representation space of ξ we have ξ(h)a = a for all h ∈ K . We will also denote by k ξ the dimension of the space of invariant vectors of the representation ξ with respect to K .
In particular, if K = {e} so that G/K = G is a compact Lie group, then k ξ = d ξ is the dimension of the representation ξ ∈ G. If the subgroup K is massive, that is if the dimension of the space of invariant vectors of the representation ξ is one for each class I representation, then we have k ξ = 1 for all ξ ∈ G 0 , and k ξ = 0 for all ξ ∈ G\ G 0 .
In order to extend the Definition (1.1) on the circle to a compact homogeneous manifold M = G/K we will employ the geodesic distance on G. For the distance from the 'unit' element e to h ∈ G we will write |h| = d(h, e). This convention is not restrictive since the analysis of our problems is local. Definition 1.1 Let 0 < α ≤ 1 and 1 ≤ p ≤ ∞. We define the space Lip G/K (α; p) by that is, by the condition that for 1 ≤ p < ∞, with a natural modification for p = ∞.
The space Lip G/K (α; p) endowed with the norm becomes a Banach space.
we say that f is (α; p)-Lipschitz. Since G acts on G/K from left, it is natural to consider left shifts f (h·) with respect to h in Definition 1.1. However, if K = {e} and our homogeneous space is the Lie group G itself, one can also consider right shifts by h, and to define the right-Lipschitz space by taking f (·h) instead of f (h·) in the Definition 1.1, namely, by the condition for 1 ≤ p < ∞, with a natural modification for p = ∞. Some remarks regarding the equivalence between these two notions in the case of groups are included at the end of Sect. 2 in Remark 2.3.
On the other hand, for 1 ≤ p ≤ ∞ we can also define p-Lipschitz spaces on G via localisations. Given f ∈ L p (G/K ), we say that for all coordinate chart cut-off functions χ , where χ * denotes the pull-back by χ . We can observe that for p = ∞, with natural modifications of the definitions, we have In view of Lemma 2.4, it is enough to show this for compact Lie groups. The inclusion Lip G/K (α; ∞) ⊂ Lip G/K (α; ∞) is obvious due to the compactness of G/K . The converse also follows if we take h = exp(h j X j ), j = 1, . . . , dim G, for small real numbers h j , with exp : g → G the exponential mapping near e, and {X j } dim G j=1 a basis of g, e.g. orthonormalised with respect to the Killing form. Using that the pushforwards χ * by coordinate chart maps have non-vanishing Jacobians, compactness, and adapting the argument above in the other direction, the equality (1.5) can be extended to any p, namely, we have The main results of this work are Theorem A and Theorem B below, extending the statements of Theorem A and Theorem B, respectively, from the circle to compact homogeneous manifolds G/K . Theorem A . Let G/K be a compact homogeneous manifold of dimension n. Let 0 < α ≤ 1, 1 < p ≤ 2, and let q be such that 1 . Then Here for a sequence σ (ξ) is the usual Hilbert-Schmidt norm. This is a natural family of spaces on G 0 , extending the 2 -norm appearing in the Plancherel identity in (1.9), also satisfying the Hausdorff-Young inequalities on G/K and many other functional analytic properties. These spaces were analysed in [20,Section 2] in the setting of compact homogeneous manifolds, extending the corresponding definition on the compact Lie groups introduced earlier in [30, Section 10.3.3], in which case we would have G 0 = G and k ξ = d ξ : see Definition 2.2 for this case. The Fourier coefficients of functions on G/K , although matrices in C d ξ ×d ξ , have a lot of zeros, see (2.13) and (2.14). This explains the factor k ξ appearing in the denominator in (1.7). The spaces β ( G 0 ) are interpolation spaces, with embeddings into another known family of β -spaces defined using Schatten norms, we refer to [20, Section 2] for a description of functional analytic properties of these spaces and to [14,Section 2.1.4] for the embedding properties.
Clearly (see Remark 3.3) Theorem A implies Theorem A when G = T 1 and K = {0}, so that G/K = T 1 . Since both Theorem A and Theorem B absorb Titchmarsh theorems in the case of the torus they are also sharp.
Here ξ are the eigenvalues of the elliptic operator (I − L G/K ) 1/2 corresponding to the representations ξ ∈ G 0 , where L G/K is the Laplacian on G/K : namely, we have The choice of the range of the indices i, j happens in this way since these are the only representation coefficients interacting with the Fourier coefficients of functions on G/K , see (2.13) and (2.14). In the case of Lie groups the quantity ξ is also explained in detail in (2.4).
Theorem B implies Theorem B since the Plancherel formula on G/K takes the form so that the left hand side of (1.8) is the remainder of the Fourier series decomposition of f ∈ L 2 (G/K ), see also (2.6).
In view of Lemma 2.4, Theorem A and Theorem B follow from their respective versions on compact Lie groups formulated in Theorems 3.2 and 3.4, respectively.
As an application of the obtained characterisations, we will show a Fourier multiplier theorem for Hölder/Lipschitz spaces on compact Lie groups. Here, the Fourier multiplier A with symbol a(ξ ) ∈ C d ξ ×d ξ for each ξ acts by multiplication by a(ξ ) on the Fourier transform side, i.e. by The matrix a(ξ ) arises as a linear transformation a(ξ ) : the operator norm of this linear transformation, with C d ξ equipped with the usual Euclidean distance. If L G is the Laplacian on G (or the Casimir element of the universal enveloping algebra), we denote by ξ the eigenvalue of the elliptic pseudo-differential operator (I − L G ) 1/2 corresponding to ξ , i.e.
see (2.5) for more explanation.
Then the Fourier multiplier A by symbol a, is bounded for all α such that 0 < α < 1 − γ .
For example, the matrix symbol of the Bessel potential A similar result to Corollary 1.2 is true on homogeneous manifolds G/K , although in this case, in view of (2.13) and (2.14), the symbol a(ξ ) may be assumed to be zero outside the first k ξ × k ξ block, i.e. we would take a(ξ ) = 0 for ξ / ∈ G 0 , and a(ξ ) i j = 0 for all ξ ∈ G 0 and for i > k ξ or j > k ξ .
The multiplier theorem in Corollary 1.2 complements other known multiplier theorems on compact Lie groups, e.g. the Mihlin multiplier theorem [32,33], Hörmander multiplier theorem [13], or more general results for the boundedness of pseudodifferential operators in L p -spaces [7].
In Sect. 4 we discuss analogues of the obtained results for the Dini-Lipschitz classes. The Dini-Lipschitz classes are a logarithmic extension on Lipschitz classes and these spaces have been analysed in different settings, e.g. in low dimensions [40], or using Helgason transforms on rank 1 symmetric spaces in [10,11]. Our analysis in this paper is based on a logarithmic extension of the Duren lemma allowing inclusion of the log-terms in the characterisations.

Preliminaries
In this section we recall some basic facts on the Fourier analysis and the notion of global matrix-symbols for pseudo-differential operators on compact Lie groups and more generally on compact homogeneous manifolds. We refer to [30,31] for a comprehensive account of such topics.
Since the proof of our main theorems will be reduced to the case of compact Lie groups as it will be explained at the end of this section, it will be convenient for our purposes to start with the setting of compact Lie groups. Given a compact Lie group G, we equip it with the normalised Haar measure μ ≡ dx on the Borel σ -algebra associated to the topology of the smooth manifold G. The Lie algebra of G will be denoted by g. We also denote by G the set of equivalence classes of continuous irreducible unitary representations of G and by Rep(G) the set of all such representations. Since G is compact, the set G is discrete. For [ξ ] ∈ G, by choosing a basis in the representation space of ξ , we can view ξ as a matrix-valued function ξ : G → C d ξ ×d ξ , where d ξ is the dimension of the representation space of ξ . By the Peter-Weyl theorem the collection is an orthonormal basis of L 2 (G). If f ∈ L 1 (G) we define its global Fourier transform at ξ by Thus, if ξ is a matrix representation, we have f (ξ ) ∈ C d ξ ×d ξ . The Fourier inversion formula is a consequence of the Peter-Weyl theorem: Given a sequence of matrices a(ξ ) ∈ C d ξ ×d ξ , we can define where the series can be interpreted in a distributional sense or absolutely depending on the growth of (the Hilbert-Schmidt norms of) a(ξ ). For a further discussion we refer the reader to [30].
For each [ξ ] ∈ G, the matrix elements of ξ are the eigenfunctions for the Laplacian L G (or the Casimir element of the universal enveloping algebra), with the same eigenvalue which we denote by −λ 2 [ξ ] , so that The weight for measuring the decay or growth of Fourier coefficients in this setting is the eigenvalues of the elliptic first-order pseudo-differential operator (I − L G ) 1 2 . The Parseval identity takes the form which gives the norm on 2 ( G).

For a linear continuous operator
where Aξ(x) ∈ C d ξ ×d ξ is understood as (Aξ(x)) i j = (Aξ i j )(x), i.e. by applying A to each component of the matrix ξ(x). Then one has ( [30,31]) the global quantization in the sense of distributions, and the sum is independent of the choice of a representation ξ from each equivalence class [ξ ] ∈ G. If A is a linear continuous operator from C ∞ (G) to C ∞ (G), the series (2.8) is absolutely convergent and can be interpreted in the pointwise sense. The symbol σ can be interpreted as a matrix-valued function on G × G. We refer to [30,31] for the consistent development of this quantization and the corresponding symbolic calculus. If the operator A is left-invariant then its symbol σ does not depend on x. We often simply call such operators invariant. In the sequel we will also need the Taylor expansion of functions on compact Lie groups. For f ∈ C ∞ (G), we have for some invariant differential operators D (α) of order |α|, for an admissible family of functions q α , with |x| denoting the geodesic distance from x to e, see [30, Section 10.6].
We note that Definition 1.1 in the case of compact Lie groups can be stated in the following form by taking the geodesic distance on G from the unit element e to h ∈ G: Definition 2.1 Let G be a compact Lie group. Let 0 < α ≤ 1 and 1 ≤ p ≤ ∞. We define the space Lip G (α; p) by For the analysis of the Fourier transform we will require the following function spaces on the unitary dual G. We refer the reader to [30, Section 10.3.3] for the basic properties of such spaces.

Definition 2.2
For 0 < p < ∞, we will write p ( G) for the space of all H = H (ξ ) ∈ C d ξ ×d ξ such that If 1 ≤ p < ∞ the quantity H p ( G) defines a norm and p ( G) endowed with it becomes a Banach space. If 0 < p < 1 we can associate a Fréchet metric and the associated space becomes a complete metric space.
We record some asymptotic properties on a compact Lie group G of dimension n, that will be of use on several occasions: asymptotically as λ → ∞ we have These properties follow from the Weyl spectral asymptotic formula for the elliptic pseudo-differential operator I − L G , and we can refer to [1] for their proof. We also note the convergence criterion ξ ≤λ d 2 ξ ξ −s < ∞ if and only if s > n, (2.12) see [5] for the proof.
and since |h| |h −1 | we have obtained that f is a right-Lipschitz of type (α; p) function if and only if f · is left-Lipschitz of type (α; p).
Our statements will be stated for left-Lipschitz (α; p). A look at the arguments in the proofs of our results shows that all the statements also hold for right-Lipschitz (α; p) type. At the level of compact Lie groups both concepts agree when p = 2 as a consequence of Theorem 3.5: indeed, the criterion (3.7) is the same for both left-or right-Lipschitz functions.
We shall now explain how the proof of Theorems A and B can be reduced to the case of compact Lie groups. For a function f ∈ C ∞ (G/K ), its canonical lifting f is defined by f (x) = f (x K ) for all x ∈ G, so that f is constant on the right cosets.

Proof A simple calculation shows
with equality, if the Haar measure is normalised in such a way that the measure of K is equal to one.

The Fourier coefficients of canonical liftings satisfy
and f (ξ ) i j = 0 for all ξ ∈ G 0 , i > k ξ , (2.14) see e.g. Vilenkin [38]. We refer to [5] or [20] for more details on the Fourier analysis on compact homogeneous manifolds.

Titchmarsh theorems for Fourier transforms of Lipschitz functions
In this section we prove our results. We start with a lemma on Fourier transforms that we will apply in the proof of the extension of Theorem reftha. The notion of global symbol discussed in Sect. 2 will be useful in the proof of the first main result. Henceforth G will denote a compact Lie group of dimension n.
, and hence H ∈ β ( G) for all β ≥ β 0 . So we need to prove the case β < β 0 , which we now assume.
For β < β 0 we can write (2.12) and the Hölder inequality applied to the last sum for the indices β 0 β , β 0 β 0 −β we obtain For the convergence of the series in the last inequality we have used (2.12). To finish the proof we just note that the condition We now state our first main result which extends Theorem A to general compact Lie groups and consequently, in view of Lemma 2.4, to compact homogeneous manifolds.

Theorem 3.2 Let G be a compact Lie group of dimension n. Let
and let q be such that 1 Proof For f ∈ L p (G) we will denote by f h the function f h (x) = f (hx). We first note that Hence the Fourier transform of f h − f is given by where I d ξ is the identity matrix in C d ξ ×d ξ . By the Hausdorff-Young inequality we get We observe that We now apply Taylor expansions of type (2.9) to the terms For each entry in the matrices they are of the form where q α ∈ C ∞ (G) vanishes at e of order 1 and e is its isolated zero, i.e., there exist constants C 1 , C 2 > 0 such that In the matrix form we can write To estimate (3.2) we plug (3.3) into it. Then We observe that for each pair α, β the matrix D (α) ξ(e)D (β) ξ(e) * + D (β) ξ(e)D (α) ξ(e) * entering this sum, is self-adjoint for every ξ , thus it can be written in the form where α,β,ξ is diagonal and U is unitary. Hence and we can deduce that |α|=1,|β|=1 α,β,ξ is a diagonal positive second order elliptic symbol since D (α) 's can be chosen to give a basis of the Lie algebra of G, see [30, (10.25)]. The ellipticity is preserved by the action of an unitary matrix, due to the identities Therefore, grouping terms with respect to the pair (α, β) and (β, α), we have We have proved that for small enough |h|. By (3.1) we have and hence by (3.4) we obtain Therefore, we obtain (3.5) Now, for β ≤ q and N ∈ N we write so that the analysis for |h| → 0 will be translated to N → ∞. In order to apply the Hölder inequality and for the forthcoming analysis it is convenient to write (N ) in the form Then, by Hölder inequality, (3.5) and (2.10) with r = 0, we get (3.6)

Remark 3.3
In the case G = T 1 we observe that Theorem 3.2 for the first conclusion takes the form: for p pα−1 < β. For the second conclusion: Consequently Theorem 3.2 with the second conclusion is an extension of Titchmarsh theorem (cf. Titchmarsh [37,Theorem 84]) and Younis [39,Theorem 2.6]) to the setting of compact Lie groups. As it was already mentioned in the introduction the lower bound p α p+ p−1 in Theorem 3.2 is in general sharp as a counterexample can be constructed for p = 2 in the case G = T 1 .
For the proof of the second Titchmarsh theorem we will be using the following lemma due to Duren (cf. We can now state our second main result which extends Theorem B to compact Lie groups and consequently, in view of Lemma 2.4, to compact homogeneous manifolds (Theorem B ).
Proof We first assume that f ∈ Lip G (α; 2). By (3.5) applied to the case p = q = 2 we obtain Keeping the notations from the proof of Theorem 3.2 for (N ) with β = 2 we write are the eigenvalues of −L G . Since W is discrete countable we can enumerate it as W = {λ k : k ∈ N}. Of course this enumeration also induces an enumeration of G and by the eigenvalue estimate λ k ≈ k 1 n (see e.g [1]) we obtain We now assume (3.7). By Plancherel theorem and with the notation for f h as left translation as in the proof of Theorem 3.2 we have Following the analysis of the proof of Theorem 3.2 we note that from the ellipticity argument used to obtain (3.4), we can now just use the fact that |α|=1,|β|=1 α,β,ξ is of the second order. Then we get Therefore the proof is now reduced to estimate But an analogous argument as in the above proof of the "only if" part and using the same notations yield with N = 1 |h| , which concludes the proof.
We now formulate an application of Theorem 3.5 to the regularity of Fourier multipliers on Hölder spaces mentioned in Corollary 1.2. For convenience of the reader we repeat its statement: Corollary 3.6 Let 0 ≤ γ < 1 and let a : G → d∈N C d×d be such that a(ξ ) ∈ C d ξ ×d ξ for each ξ and Let A be the Fourier multiplier with symbol a, i.e. A f (ξ ) = a(ξ ) f (ξ ) for all ξ ∈ G. Then is bounded, for all α such that 0 < α < 1 − γ . section we will consider a different condition, the so-called Dini-Lipschitz condition on L 2 and we will generalise the corresponding Titchmarsh theorems (cf. [37, Theorem 85]).
We shall first establish an extension of Duren's lemma (cf. [9, p. 101]), Lemma 3.4 in this paper, adapted to the Dini-Lipschitz condition.
where we have also used k −a − (k + 1) −a ≤ ck −a−1 in the last inequality. Letting M → ∞ the conclusion is reduced to estimate the series We observe that an integration by parts argument shows that We now distinguish two cases: where the number C N ,a,b,d is non-negative since d a−b ≤ 0 and t b−a−1 (log t) −d−1 > 0. If d > 0, we take the least positive integer such that d ≤ − 1. An inductive argument applying integration by parts to the integral on the right-hand side of (4.2) gives us Once more, the fact that t b−a−1 (log t) d− ≥ 0 and the choice of implies that which concludes the proof of the "only if part". Conversely, we assume the second assertion and write The sum N −1 k=1 k b−1 (log k) d can be estimated by integration in a similar way as we have obtained (4.2). The inductive argument explained above gives By using this formula, one can deduce that The proof is complete.
We now write N = 1/h. Then An analogous argument as in the proof of Theorem 3.5 to obtain (3.9) and using the same notations lead us to A look at the proof above shows that the converse also holds.
To end this work we now give a Dini-Lipschitz version of Theorem 3.2. Consequently, in both cases we have f ∈ γ ( G 0 ) for np α p + np − n < γ ≤ q.