Abstract
By applying Pitt’s inequality we prove a weighted \(L^p\) version of Gallagher’s inequality for trigonometric series. Furthermore, we consider a family of weights generated by a smoothing process, via convolution operation, whose first steps are the indicator function of a compact interval and the so-called Cesàro weight supported in the same interval. The eventual aim is the comparison of such weights in view of possible refinements of our inequality for \(p=2\).
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
The trigonometric series considered here are of the type
where \(e(x):= e^{2\pi ix}\) and \(a(\nu )\in {{\mathbb {C}}}\) for a strictly increasing sequence \(\nu \in {{\mathbb {R}}}\). For any given \(T\in (0,+\infty )\) and \(p\in [1,+\infty )\), let us denote the \(L^p\)-norm of \(f:{{\mathbb {R}}}\rightarrow {{\mathbb {C}}}\) by \(\Vert f\Vert _{p}:=\big (\int _{{{\mathbb {R}}}}|f(t)|^p\,dt\big )^{1/p}\) and set \(\Vert f\Vert _{p,T}:=\big (\int _{-T}^T|f(t)|^p\,dt\big )^{1/p}\). The space of the complex-valued functions defined on \({{\mathbb {R}}}\) with a finite \(L^p\)-norm is denoted by \(L^p({{\mathbb {R}}},{{\mathbb {C}}})\). By \(f\in L^1_{loc}({{\mathbb {R}}},B)\), with \(B\subseteq {{\mathbb {C}}}\), we mean that \(f:{{\mathbb {R}}}\rightarrow B\) is summable on each compact subset of \({{\mathbb {R}}}\). The Fourier transform of an integrable \(f:{{\mathbb {R}}}\rightarrow {{\mathbb {C}}}\) is denoted by
The convolution of \(f,g\in L^1({{\mathbb {R}}},{{\mathbb {C}}})\) is
For any weight \(w:{{\mathbb {R}}} \rightarrow {{\mathbb {C}}}\) and any \(\delta >0\), we set \(w_\delta (x)=w(x)\) or 0 according as \(|x|\le \delta\) or not. Further, we denote the discrete convolution of \(a(\nu )\) and \(w_\delta (x)\) as
The function that is identically 1 we denote by \(\mathbf{1}\), so that \(\mathbf{1}_{\delta }\) is the characteristic function of the interval \((-\delta ,\delta )\). Recall that \(\widehat{\mathbf{1}_{\delta }}(y)=2\delta \hbox {sinc}(2\delta y)\), where
Definition 1.1
Let \(1<p\le q<\infty\) be fixed and \(p'=p/(p-1)\). If \(U,V\in L^1_{loc}({{\mathbb {R}}},{{\mathbb {R}}}_+)\) are even, U is non-decreasing on \((0,+\infty )\), V is non-increasing on \((0,+\infty )\) and
then we say that (U, V) is a (p, q)-pair of Pitt weights.
In Sect. 2 the following result is proved.
Theorem 1.1
Let \(\delta ,T\in (0,+\infty )\) and \(1<p\le q<\infty\) be fixed. If \(w\in L^1_{loc}({{\mathbb {R}}},{{\mathbb {C}}})\) is even, (U, V) is a (p, q)-pair of Pitt weights and \(S(t)=\sum _{\nu } a(\nu )e(\nu t)\) is absolutely convergent, then there exists \(C=C(p,q,U.V)>0\) such that
More generally, the same inequality holds if \(w_\delta\) is replaced by \(w\in L^1({{\mathbb {R}}},{{\mathbb {C}}})\).
Remark 1.1
Unless otherwise stated, hereafter the constant C is not necessarily the same at each occurrence. Further, let us assume henceforth that \(\displaystyle {\min _{|t|\le T}|\widehat{w_\delta }(t)|}\not =0\), for otherwise (1.1) is trivial.
In order to prove Theorem 1.1, we apply a weighted Fourier transform norm inequality, called Pitt’s inequality, which is an old result in Fourier analysis ([15]). Such an inequality is quoted in Sect. 2, before the proof of Theorem 1.1, by referring to the version essentially given by Heinig et al. ([8]), [10, 14] in 1983–1984 (see also [2, 6] for advanced and more recent versions). It is well-known that Pitt’s inequality yields the Hardy-Littlewood inequality, the Hausdorff-Young inequality and the Plancherel identity. In the latter case, i.e., \(p=q=2\), \(U=V=\mathbf{1}\), \(C=1\), the inequality (1.1) becomes
In the present paper we continue our study of (1.2), that was started in [5]. We point out that most of the next considerations can be easily adapted to the general case (1.1). Let us recall that a first consequence of (1.2) is the following majorant properties established in [5]. Let \(\delta , T\in (0,+\infty )\) and let \(w\in L^1_{loc}({{\mathbb {R}}},{{\mathbb {C}}})\) be even such that
If \(S(t)=\sum _{\nu } a(\nu )e(\nu t)\) and \(B(y)=\sum _\nu b(\nu )e(\nu y)\) are absolutely convergent, and the first sum is a majorant for the second one, i.e., \(|a(\nu )|\le b(\nu )\) for all \(\nu\), then
Now, it is worthwhile to recall that by a straightforward application of the Hardy-Littlewood majorant principle (see [13, Ch.7] or [11]) one gets
The combination of such an inequality with (1.2) for B(y) gives a weaker inequality than (1.3) because of the factor 3 in the right-hand side. Moreover, we recall that such a factor is best possible in the Hardy-Littlewood majorant principle (see [12]).
Example 1.1
A well-known instance of (1.2) is given by taking the weight \(u_{\delta }\) associated to the unit step function
It is easily seen that \(\widehat{u_{\delta }}(y)=e(-\delta y/2)\, \delta \, \hbox {sinc}(\delta y)\). Therefore, for \(0<\delta T<1\), from (1.2) we find the Gallagher inequality [7]
We refer the reader to [9] for a first introduction to the applications of such an inequality within the analytic number theory.
Example 1.2
Another remarkable instance of (1.2) is the special case of the Cesàro weight
whose Fourier transform is \(\widehat{C_{\delta }}(y)=\delta \hbox { sinc}^2(\delta y)\). In the literature, this is known as Fejer kernel [4]. For \(\delta T\in (0,1)\), the inequality (1.2) gives
In [5] we have applied the latter inequality to the Dirichlet polynomials and to the distribution of certain arithmetical functions in short intervals.
In the present paper we focus on some further aspects of (1.2), mostly concerning the optimal choice of the weight w. To this aim and for the convenience of the reader, we need to quote some definitions and properties already introduced in [5]. First, for every \(w\in L^1_{loc}({{\mathbb {R}}},{{\mathbb {C}}})\) let us define the normalized self-convolution of \(w_\delta\), with \(\delta >0\), as
For example, the Cesàro weight (1.4) is the normalized self-convolution of the restriction to \([-\delta /2,\delta /2]\) of \(\mathbf{1}\):
It is well-known that the iteration of the self-convolution gives rise to a process of smoothing (see [4]). Moreover, the support of \(\widetilde{w_\delta }\) is doubled with respect to the support of \(w_\delta\) in the sense that it is a subset of \([-2\delta ,2\delta ]\). Because of the normalizing factor \((2\delta )^{-1}\), that takes into account the length of the integration interval, the magnitude of \(w_\delta\) is not altered much by the normalized self-convolution. More precisely, if one has \(w_{\delta }\asymp 1\), i.e., \(1\ll w_{\delta }\ll 1\), in an interval of length \(\gg \delta\), then there exists an interval of length \(\gg \delta\) (not necessarily the same) where \(\widetilde{w_\delta }\asymp 1\). Recall that \(A\ll _{\theta } B\) stands for \(|A|\le cB\), where \(c>0\) is an unspecified constant depending on \(\theta\).
From another well-known property of the convolution it follows that the Fourier transform of \(\widetilde{w_\delta }\) is \(\widehat{\widetilde{w_\delta }}(y)= (2\delta )^{-1}\widehat{w_{\delta }}(y)^2\). In particular, from \(\widehat{\mathbf{1}_{\delta }}(y)=2\delta \hbox {sinc}(2\delta y)\) we rediscover that \(\widehat{C_{\delta }}(y)=\widehat{\widetilde{\mathbf{1}_{\delta /2}}}(y)= \delta ^{-1}\widehat{\mathbf{1}_{\delta /2}}(y)^2= \delta \hbox {sinc}^2(\delta y)\).
The normalized self-convolution generates recursively the family of weights:
with the base steps \(C^{\langle 0\rangle }_{\delta }(x):= \mathbf{1}_{\delta }(x)\) and \(C^{\langle 1\rangle }_{\delta }(x):= C_{\delta }(x)\).
In [5] we proved the following inductive formula for the Fourier transform of such weights:
Consequently, \(\widehat{C^{\langle j\rangle }_\delta }(y) \asymp _j \delta\) at least for \(0\le |y|\le 2^{j-2}\delta ^{-1}\).
Such a process of continuous smoothing through the self-convolution of a weight \(w_\delta\) has a discrete counterpart given by the autocorrelation of \(w_\delta\) (since no confusion can arise in the following, we will use the simpler term correlation):
For example, since it turns out that
the Cesàro weight (1.4) is the normalized correlation of the unit step weight \(u_\delta\). Note that the Cesàro weight (1.4) is generated by both types of smoothing from the function \(\mathbf{1}\). Moreover, through an iteration of the normalized correlation one might parallel the self-convolution process to generate the whole family of weights \(C^{\langle j\rangle }_\delta\), \(j\ge 1\).
An important aspect is that if the coefficients of a trigonometric series are correlations of w, then such a series is non-negative. More precisely,
A particularly well-known case is the Fejér kernel (compare the \(\widehat{C_\delta }\) formula)
Such a positivity property is the complete analogous of the aforementioned fact that the Fourier transform of a self-convolution is a square.
Now, let us introduce our comparison argument for the weight w in view of possible refinements of (1.2). First, recall that
by assumption in order to avoid triviality. Thus, we define
and write
In order to take advantage of such an inequality one would require \({{\mathcal {E}}}_{w}\) either with a small measure or free of peaks of S(y), if it does not support a sufficiently small \(L^2\)-norm. For example, let us examine the possible scenario for the weights (1.5). To this aim, recall that
The aforementioned triviality for (1.2) does not occur if we assume \(\delta T\in (0,1/2)\), for otherwise it would be \(\delta t_0=1/2\) for some \(t_0\in (0,T]\) and consequently
On the other hand, it is plain that if \(\delta T\in (0,1/2)\), then one has
for some \(T_0\ge T\). It is easily seen that analogous considerations hold for the whole family \(\{C^{\langle j\rangle }_{\delta }\}_{j\ge 0}\). However, in several applications of (1.2), like in the case of Dirichlet polynomials (see [5]), the contribution from the tail \(|y|>T\) might amount to a remainder term. On the other hand, note that the inequality
trivially holds for any choice of \(w_\delta\) such that \(m_{\delta ,T}\not =0\). Even for this reason, it is worthwhile to compare weights in view of possible refinements of the first term on the right-hand side of (1.6). To this end, we give the following definition.
Definition 1.2
Let us suppose that \(w_\delta , v_\delta\) are such that \(\widehat{w_\delta }(t)\not =0\), \(\widehat{v_\delta }(t)\not =0\), \(\forall t\in [-T,T]\). We say that \(v_\delta\) is T-better than \(w_\delta\) if
If so, we write \(v_\delta \prec _{_T} w_\delta\).
If \(v_\delta \prec _{_T} w_\delta\), it is plain that \(v_\delta\) yields a refinement of the upper bound of \(\Vert S\Vert ^2_{2,T}\) with respect to \(w_\delta\), at least in \([-T,T]\). In §3 we prove the next result on the family of weights (1.5) and discuss the comparison involving also some weights as \(C_{\delta }*\mathbf{1}_\delta\) and the Lanczos weight.
Theorem 1.2
Let \(\delta , T\in (0,+\infty )\). For every \(j\ge 0\), if \(\delta T<2^{j-1}\), then \(C^{\langle j+1\rangle }_\delta \prec _{_T} C^{\langle j\rangle }_\delta\). Moreover, the inequality
still holds for all \(\displaystyle { |y|\in \bigcup _{k\in {{\mathbb {N}}}} \big [2^{j-1}(2k+1)/\delta +\varepsilon '_k,2^{j-1}(2k+3)/\delta -\varepsilon ''_k\big ]}\), where
are such that \(\varepsilon '_k, \varepsilon ''_k\sim \delta T^2/(2^jk)-T/(2k)\) as \(\delta T\rightarrow 2^{j-1}\), \(\forall j,k\in {{\mathbb {N}}}\).
Remark 1.2
For any fixed \(j,k\in {{\mathbb {N}}}\), the interval \(\big [2^{j-1}(2k+1)\delta ^{-1}+\varepsilon '_k,2^{j-1}(2k+3)\delta ^{-1}-\varepsilon ''_k\big ]\) tends to cover the whole interval \([(2k+1)T,(2k+3)T]\) as \(\delta T\rightarrow 2^{j-1}\). Somehow this means that (1.8) holds almost everywhere in \({{\mathbb {R}}}\setminus [-T,T]\) as long as \(\delta T\rightarrow 2^{j-1}\) and it gives a further chance to get even a refinement of the second term on the left-hand side of (1.6) by replacing \(C^{\langle j\rangle }_\delta\) with \(C^{\langle j+1\rangle }_\delta\). Furthermore, such a refinement might rely also on the value of the series \(S((2k+1)T)\), \(\forall k\in {{\mathbb {Z}}}\).
Remark 1.3
An effective use of (1.2) with \(C^{\langle j\rangle }_{\delta }\) requires finding explicit expressions of such weights. For example, the normalized self-convolution of \(C_{\delta /2}\) is the so-called Jackson-de La Vallé Poussin weight \(C^{\langle 2\rangle }_{\delta }\), that is the following cubic spline [4, Problem 5.1.2 (v)]:
Note that the support of \(C^{\langle 2\rangle }_{\delta }\) is \([-\delta ,\delta ]\), as expected.
2 Proof of Theorem 1.1
Here we recall Pitt’s inequality:
Let \(1<p\le q<\infty\) be fixed and (U, V) be a (p, q)-pair of Pitt weights. There exists \(C=C(p,q,U.V)>0\) such that
for all f such that \(U^{1/p}f\in L^p({{\mathbb {R}}},{{\mathbb {C}}})\).
Proof
First, note that from the hypotheses on S and w it follows that
Therefore, by applying Lebesgue’s dominated convergence theorem [16, Th.1.38] it turns out that \(a\star w_\delta \in L^1({{\mathbb {R}}},{{\mathbb {C}}})\) and
Then, observe that we can assume \(U^{1/p} (a\star w_\delta )\in L^p({{\mathbb {R}}},{{\mathbb {C}}})\) for \(1<p<\infty\), for otherwise the inequality
would be trivial. Hence, (2.2) follows by applying Pitt’s inequality (2.1), being plain that (recall that w, V are even functions)
By taking any fixed \(T\in (0,+\infty )\), clearly (2.2) implies (1.1).
Now, let us prove that
As before, we can clearly assume that \(a\star w\in L^p({{\mathbb {R}}},{{\mathbb {C}}})\) and apply (2.2) with \(\delta =n\in {{\mathbb {N}}}\) to write
Since \(w_n(t)e(-ty)\) converges to \(w(t) e(-ty)\), with \(|w_n(t)e(-ty)|\le |w(t)|\) and \(w\in L^1({{\mathbb {R}}},{{\mathbb {C}}})\), the dominated convergence theorem yields
On the other hand, the same theorem implies that
Hence, (2.3) follows from (2.4) after passage to the limit as \(n\rightarrow \infty\). Again, by taking any fixed \(T\in (0,+\infty )\), it is plain that (1.1) holds if \(w_\delta\) is replaced by \(w\in L^1({{\mathbb {R}}},{{\mathbb {C}}})\). \(\square\)
Remark 2.1
It is well konwn that, by taking \(V(x)=|x|^{-\alpha q}\), \(U(x)=|x|^{\beta p}\) with
Pitt’s inequality (2.1) yields three classical inequalities in Fourier analysis:
-
the Hardy-Littlewood inequality if \(p=q\ge 2\), \(\alpha =0\) or \(1<p=q\le 2\), \(\beta =0\);
-
the Hausdorff-Young inequality if \(q=1-1/p\ge 2\), \(\alpha =\beta =0\);
-
the Plancherel identity if \(p=q=2\), \(\alpha =\beta =0\).
Accordingly, for \(V(x)=|x|^{-\alpha q}\) and \(U(x)=|x|^{\beta p}\) the inequality (2.2) turns into
where we have set \(\mathrm{v}(x)=|x|\). Thus, (1.1) specializes to the following instances paralleling the aforementioned three properties:
(in the latter \(C=p^{1/(2p)}/q^{1/(2q)}\) is the so-called Beckner’s constant);
Finally, we conclude this remark by recalling that the original version of Pitt’s inequality was tailored for the Fourier series and the power series [15]. In particular, by assuming that \(\nu \in {{\mathbb {Z}}}\) for our Fourier series S, Theorem 2 of [15] yields
where \(p,q,\alpha ,\beta\) are as in (2.5).
3 Proof of Theorem 1.2: comparison of weights
Proof
Let us start with \(j=0\) and show that \(C_\delta\) is T-better than \(\mathbf{1}_\delta\) by assuming that \(\delta T<1/2\), namely we prove that
holds for all \(y\in [-T,T]\). First, recall that \(\delta T<1/2\) yields
Since
we can assume that \(y\not =0\) and set \(x=y/T\), \(\theta =\delta T\). Thus, (3.1) becomes
It is easy to see that \(G_\theta (x)\) satisfies the following properties:
-
(i)
\(G_\theta (x)\) is even with respect to both x and \(\theta\)
-
(ii)
\(G_\theta (x)\) is strictly increasing with respect to \(x\in (0,1]\)
-
(iii)
\(\displaystyle {\lim _{x\rightarrow 0} G_\theta (x)=1/4}\) \(\forall \theta \in (0,1/2)\)
-
(iv)
\(G_\theta (1)\) is strictly increasing with respect to \(\theta \in (0,1/2)\)
-
(v)
\(\displaystyle {\lim _{\theta \rightarrow 0}G_\theta (1)=1/4}\), \(\displaystyle {\lim _{\theta \rightarrow 1/2}G_\theta (1)=+\infty }\)
-
(vi)
\(G_\theta (x)=0\Longleftrightarrow x=k/\theta \ \forall k\in {{\mathbb {Z}}}{\setminus }\{0\}\) (note that \(\forall k\in {{\mathbb {Z}}}\setminus \{0\}\) and \(\forall \theta \in (0,1/2)\) one has \(|k|/\theta > 2|k|\ge 2\))
-
(vii)
\(G_\theta (x)\rightarrow +\infty\) as \(x\rightarrow (2k+1)(2\theta )^{-1}\ \forall k\in {{\mathbb {Z}}}\).
Note that \(|2k+1|(2\theta )^{-1}> |2k+1|\ge 1\), \(\forall k\in {{\mathbb {Z}}}\) and \(\forall \theta \in (0,1/2)\).
From properties (i)-(v) it follows that \(G_\theta (x) \le G_\theta (1)\) for all \(x\in [-1,1]{\setminus }\{0\}\), that is to say that (3.1) holds for all \(|y|\le T\), i.e., \(C_\delta \prec _{_T} \mathbf{1}_\delta\).
In order to prove the second part of the theorem for \(j=0\), observe that properties (i), (vi) and (vii) imply that the above inequality for \(G_\theta\) is true for all
where \(0<\Delta '_{k}=\Delta '_{k}(\theta ), \Delta ''_{k}=\Delta ''_{k}(\theta )<(2\theta )^{-1}\) are such that the endpoints of the intervals \({{\mathcal {I}}}_{\theta ,k}\) are the solutions of the equation \(G_\theta (x)=G_\theta (1)\). In particular, this yields
with \(0<\theta \Delta '_{k}<1/2\). From the latter equation and from (v) we deduce that \(\Delta '_{k}\sim (\theta -1/2)k^{-1}\) as \(\theta =\delta T\rightarrow 1/2\), for all \(k\in {{\mathbb {N}}}\). An analogous property holds for \(\Delta ''_k\). Hence, we set \(\varepsilon '_k(0,\delta ,T)=T\Delta '_{k}, \varepsilon ''_k(0,\delta ,T)=T\Delta ''_{k}\) to get the theorem proved in the case \(j=0\). For \(j>0\) it sufficies to note that (see (1.5))
where we have set \(\theta =\delta T/2^j\), \(x=y/T\) and \(G_\theta (x)=(2\pi \theta x)^{-2}\tan ^2(\pi \theta x)\) as before. Since \(0<\delta T< 2^{j-1}\) if and only if \(0<\theta < 1/2\), the conclusion follows from what we have showed in the case \(j=0\). \(\square\)
Remark 3.1
Theorem 1.2 seems to indicate that one could generate more and more T-better weights by increasing the iteretad convolutions of \(\widehat{\mathbf{1}_\delta }\). However, the following examples suggest that one must exercise some caution about the choice of the weights to be compared.
Example 3.1
Let us compare \(C_{2\delta }\) and \(D_{2\delta }:= C_{\delta }*\mathbf{1}_\delta\). First, recall that
so that
Further, for \(\delta T<1/2\), one has
Since it turns out that
then we conclude that \(D_{2\delta }\prec _{_T}C_{2\delta }\) once \(\delta T<1/2\).
Note that both \(C_{2\delta }\) and \(D_{2\delta }\) have support in \([-2\delta ,2\delta ]\). On the other hand, by comparing \(C_{\delta }\) and \(D_{2\delta }\) we see that
and, for \(\delta T<1/2\), one has
that is to say, if \(\delta T<1/2\), then \(C_{\delta }\prec _{_T}D_{2\delta }\). Analogously, it is easy to see that \(C^{\langle j\rangle }_\delta \prec _{_T}C^{\langle j\rangle }_{2\delta }\) for all \(j\ge 0\) (and consequently \(C^{\langle j\rangle }_\delta \prec _{_T}C^{\langle j-1\rangle }_{2\delta }\) for all \(j\ge 1\), because of Theorem 1.2), as long as \(\delta T<2^{j-1}\).
Example 3.2
Given real numbers \(\delta \ge \gamma >0\), the Lanczos weight is defined as [4, Problem 5.1.2 (v)]
whose Fourier transform is
The diagram of \({{\mathcal {L}}}_{\delta ,\gamma }\) is an isosceles trapezium. Note that \(\mathbf{1}_\delta \ge {{\mathcal {L}}}_{\delta ,\gamma }\ge C_\delta ={{\mathcal {L}}}_{\delta ,\delta }\) for any \(\delta \ge \gamma >0\). Since \({{\mathcal {L}}}_{\delta ,\delta }=C_\delta\), then we can assume that \(\delta>\gamma >0\). Further, by taking \(\gamma T<\delta T<(2\delta -\gamma )T<1/2\) one has \(\widehat{{\mathbf{1}}_{\delta }}(y)\not =0\) and \(\widehat{{{\mathcal {L}}}_{\delta ,\gamma }}(y)\not =0\) for all \(y\in [-T,T]\). Therefore, under such assumptions we can compare \({{\mathcal {L}}}_{\delta ,\gamma }\) and \(\mathbf{1}_\delta\) by verifying the inequality
To this aim, let us write
where \(F(x):= (\pi x)\cot (\pi x)\). Since \(F(x)\le 1\) for \(|x|\le 1/2\), we get
Therefore, we can assume that \(y\not =0\) and write
Since it is readily seen that \(F(x)\ge F(x_0)\) for \(|x|\le x_0<1/2\), we conclude that \({{\mathcal {L}}}_{\delta ,\gamma }\prec _{_T}\mathbf{1}_\delta\) provided \(\gamma T<\delta T<(2\delta -\gamma )T<1/2\).
Finally, we show that \(C_{\delta }\prec _{_T}{{\mathcal {L}}}_{\delta ,\gamma }\), under the same conditions on \(\gamma\), \(\delta\) and T. Indeed, we prove that the inequality
holds for all \(y\in [-T,T]\). To this aim, let us recall that, being concave in \((-\pi ,\pi )\), the function \(\hbox {sinc}\) is also logarithmically concave in the same interval (see [3, Ch.3]), i.e., for all \(\lambda \in [0,1]\) and all \(x_1,x_2\in (-1,1)\) one has
In particular, by taking \(\lambda =1/2\), \(x_1=\gamma T\) and \(x_2=(2\delta -\gamma )T\), with \(\gamma T<\delta T<(2\delta -\gamma )T<1/2\), this yields
Therefore, let assume that \(y\not =0\) and write
Thus, it is plain that we get the desired conclusion once we prove that, for any fixed real numbers a, b such that \(0<a<b<\pi /(2T)\), the function
is monotone increasing in (0, T]. Indeed, it turns out that
because \(0<a<b<\pi /(2T)\) and \(y\in (0,T]\) yield \(\cos (ay)-\cos (by)>0\) and \((ay)^{-1}\sin (ay)>(by)^{-1}\sin (by)\).
References
Beckner, W. 1975. Inequalities in Fourier analysis. Annals of Mathematics (2) 102 (1): 159–182.
Benedetto, J.J., and H.P. Heinig. 2003. Weighted Fourier inequalities: new proofs and generalizations. Journal of Fourier Analysis and Applications 9: 1–37.
Boyd, S.P., and L. Vandenberghe. 2004. Convex Optimization. Cambridge: Cambridge University Press.
Butzer, P.L., and R.J. Nessel. 1971. Fourier Analysis and Approximation, vol. I. Basel und Stuttgart: Birkhäuser Verlag.
Coppola, G., and M. Laporta. 2015. A generalization of Gallagher’s Lemma for exponential sums. Šiauliai Maths Seminar 10 (18): 1–19.
De Carli, L., D. Gorbachev, and S. Tikhonov. 2017. Pitt inequalities and restriction theorems for the Fourier transform. Revista Matematica Iberoamericana 33 (3): 789–808. https://doi.org/10.4171/RMI/955.
Gallagher, P.X. 1970. A large sieve density estimate near \(=1\). Inventiones Mathematicae 11: 329–339.
Heinig, H.P. 1984. Weighted norm inequalities for classes of operators. Indiana University Mathematics Journal 33: 573–582.
Huxley, M.N. 1972. The Distribution of Prime Numbers. Oxford: Oxford University Press.
Jurkat, W.B., and G. Sampson. 1984. On rearrangement and weight inequalities for the Fourier transform. Indiana University Mathematics Journal 33: 257–270.
Laporta, M. 2014. Some remarks on the Hardy–Littlewood majorant principle for exponential sums. Ricerche di Matematica 63 (1): 179–186. https://doi.org/10.1007/s11587-014-0196-6.
Logan, B.F. 1988. An interference problem for exponentials. Michigan Mathematical Journal 35: 369–393.
Montgomery, H.L. 1994. Ten Lectures on the Interface Between Analytic Number Theory and Harmonic Analysis. CBMS Regional Conf. Ser. in Math. 84, Providence, RI: American Mathematical Society.
Muckenhoupt, B. 1983. Weighted norm inequalities for the Fourier transform. Transactions of the American Mathematical Society 276: 729–742.
Pitt, H.R. 1937. Theorems on Fourier series and power series. Duke Mathematical Journal 3: 747–755.
Rudin, W. 1987. Real and Complex Analysis, 3rd edn. New York, NY: Mc Graw Hill International Editions.
Acknowledgements
The authors wish to thank Laura De Carli for helpful comments on an early version of the present paper. Further, they wish to emphasize the excellent and constructive attention received from the referee.
Funding
Open access funding provided by Università degli Studi di Napoli Federico II within the CRUI-CARE Agreement.
Author information
Authors and Affiliations
Contributions
All the authors contributed equally for the preparation of the present paper.
Corresponding author
Ethics declarations
Conflict of interest
The authors declare that they do not have any competing interests.
Ethical approval
This article does not contain any studies with human participants or animals performed by any of the authors.
Additional information
Communicated by S. Ponnusamy.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Coppola, G., Laporta, M. A weighted inequality for trigonometric series. J Anal 31, 1041–1056 (2023). https://doi.org/10.1007/s41478-022-00486-y
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s41478-022-00486-y