Abstract
We estimate the \(L^{p}\) norms of the discrepancy between the volume and the number of integer points in \(r\Omega -x\), a dilated by a factor r and translated by a vector x of a convex body \(\Omega \) in \({\mathbb {R}}^{d}\) with smooth boundary with strictly positive curvature,
where \(\mu \) is a Borel measure compactly supported on the positive real axis and \(R\rightarrow +\infty \).
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
The discrepancy between the volume and the number of integer points in \(r\Omega -x\), a dilated by a factor r and translated by a vector x of bounded domain \(\Omega \) in \({\mathbb {R}}^{d}\), is
Here \(\chi _{r\Omega -x}(y)\) denotes the characteristic function of \(r\Omega -x\) and \(\left| \Omega \right| \) the measure of \(\Omega \). A classical problem is to estimate the size of \({\mathcal {D}}\left( \Omega ,r,x\right) \), as \(r\rightarrow +\infty \). For a survey see e.g. [10, 12] or [19].
By a classical result of D. G. Kendall, the \(L^{2}\) norm with respect to the translation variable x of the discrepancy \({\mathcal {D}}\left( \Omega , r,x\right) \) of an oval \(\Omega \) is of the order of \(r^{\left( d-1\right) /2}\). See [11] and what follows. For this reason we shall call \(r^{-\left( d-1\right) /2}{\mathcal {D}}\left( \Omega ,r,x\right) \) the normalized discrepancy. Our main result below is an estimate of the fractal dimension of the set of values of the dilation variable r where this normalized discrepancy may be large.
Throughout the paper, we shall assume that \(\Omega \) is a convex body in \({\mathbb {R}}^d\) with smooth boundary with strictly positive Gaussian curvature such that the origin belongs to the interior of \(\Omega \). We will also assume that \(\mu \) is a positive Borel measure with compact support contained in \(\{0\le r <+\infty \}\) and with Fourier transform \(|\widehat{\mu }(\xi )|\le C(1+|\xi |)^{-\beta }\) for some \(\beta \ge 0\). We recall that the Fourier transform of \(\mu \) is defined by
We also recall that the Fourier dimension of a compact set \(K\subset {\mathbb {R}}\) is the supremum of all \(\delta \in [0,1]\) for which there exists a probability measure \(\nu \) supported in K with a positive constant C such that \(\left| \widehat{\nu }\left( \xi \right) \right| \le C\left| \xi \right| ^{-\delta /2}\). It is known that the Fourier dimension is smaller than or equal to the Hausdorff dimension, and indeed they may differ. See [4, Section 4.4] and [14, Section 12.17].
Also, for any \(p\ge 1\) and for any \(R\ge 2\) we define
where the translated measure \(d\mu \left( r-R\right) \) is defined by
Theorem 1.1
Let \(d=2\).
If \(0\le \beta <2/5\) then there exists a constant C such that for every \(R\ge 2\),
If \(\beta =2/5\) then there exists a constant C such that for every \(R\ge 2\),
If \(2/5<\beta <1/2\) then there exists a constant C such that for every \(R\ge 2\),
If \(\beta =1/2\) then there exists a constant C such that for every \(R\ge 2\),
If \(\beta >1/2\) then there exists a constant C such that for every \(R\ge 2\),
Theorem 1.2
Let \(d\ge 3\).
If \(0\le \beta <1\) then there exists a constant C such that for every \(R\ge 2\),
If \(\beta =1\) then there exists a constant C such that for every \(R\ge 2\),
If \(\beta >1\) then there exists a constant C such that for every \(R\ge 2\),
The case \(d=2\) can be improved in the range \(\beta >2/5\) when \(\Omega \) is an ellipse E. More precisely we have the following result.
Theorem 1.3
Let E be an ellipse in the plane.
If \(0\le \beta <1\) then there exists a constant C such that for every \(R\ge 2\),
If \(\beta =1\) then there exists a constant C such that for every \(R\ge 2\),
If \(\beta >1\) then there exists a constant C such that for every \(R\ge 2\),
When the measure \(\mu \) is the Dirac delta \(\delta _0\) centered at 0, then
In this case, \(|\widehat{\delta }_0(\xi )|=1\) so that \(\beta =0\), and the above Theorems 1.1 and 1.2 can be restated as
Corollary 1.4
This recovers recent results of M. Huxley [9] for the case \(d=2\) and L. Brandolini, L. Colzani, G. Gigante, G. Travaglini [2] for the general dimension d.
If \(\mu \) is the uniformly distributed measure in the interval \(\left\{ 0<r<1\right\} \), then
The Fourier transform of the uniformly distributed measure in \(\left\{ 0<r<1\right\} \) has decay \(\beta =1\),
On the other hand, if \(\psi \left( r\right) \) is a non negative smooth function with integral one and support in \(0\le r\le 1\), one can consider a smoothed average
This smoothed average is equivalent to the uniform average over \(\left\{ R<r<R+1\right\} \), but the decay of the Fourier transform \(\widehat{\psi } \) is faster than any power \(\beta \). Hence for the uniformly distributed measure in \(\left\{ 0<r<1\right\} \) Theorems 1.1 and 1.2 apply with the indices corresponding to \(\beta >1\), and one obtains the following
Corollary 1.5
As mentioned before, when \(d=2\) and \(\Omega =E\) (an ellipse) this result can be improved. By Theorem 1.3,
Corollary 1.6
Observe that the range of indices for which the \(L^{p}\) norm remains uniformly bounded with this choice of \(\mu \) is larger than the range of indices in [2, 9] (those that we obtain when \(\mu =\delta _0\)).
As an intermediate case between the two preceeding examples, one can consider a measure \(d\mu \left( r\right) =r^{-\alpha }\chi _{\left\{ 0<r<1\right\} }\left( r\right) dr\), with \(0<\alpha <1\). In this case \(\left| \widehat{\mu }\left( \xi \right) \right| \le C\left( 1+\left| \xi \right| \right) ^{\alpha -1}\), that is \(\beta =1-\alpha \). As a more sophisticated intermediate example, recall that R. Salem has proved by means of a random construction that there exist compact sets with Fourier dimension equal to the Hausdorff dimension, for every dimension between 0 and 1. See [15] and also [14, Section 12.17]. The above theorems assert that the discrepancy cannot be too large in mean on translations of these sets. On the other hand, there are cases when the Fourier dimension is strictly less than the Hausdorff dimension. An example is the triadic Cantor set, with Fourier dimension 0 and Hausdorff dimension \(\log (2)/\log (3)\). In this particular case, and in similar cases with \(\beta =0\), the above theorems are unable to exploit the average over a set of positive Hausdorff dimension, and they do not improve the results already contained in [2].
The techniques used to prove the above theorems also apply to the estimates of mixed \(L^{p}(L^2)\) norms of the discrepancy:
This has been done by the authors in [3]. Here it suffices to remark that the set of p’s that give bounded mixed \(L^p(L^2)\) norms is larger than the set of p’s that give bounded pure \(L^p\) norms. See also [5].
We would like to thank Luca Brandolini and Giancarlo Travaglini for several discussions on this subject during the early stages of the preparation of this paper.
2 Preliminary Lemmas
The proofs of the theorems will be splitted into a number of lemmas, some of them well known. The starting point is the observation of D. G. Kendall that the discrepancy \({\mathcal {D}}\left( \Omega ,r,x\right) \) is a periodic function of the translation, and it has a Fourier expansion with coefficients that are a sampling of the Fourier transform of \(\Omega \),
Lemma 2.1
The number of integer points in \(r\Omega -x\), a translated by a vector \(x\in {\mathbb {R}}^{d}\) and dilated by a factor \(r>0\) of a domain \(\Omega \) in the d dimensional Euclidean space, is a periodic function of the translation with Fourier expansion
In particular,
Proof
This is a particular case of the Poisson summation formula. \(\square \)
Remark 2.2
We emphasize that the Fourier expansion of the discrepancy converges at least in \(L^{2}\left( {\mathbb {T}}^{d}\right) \), but we are not claiming that it converges pointwise. Indeed, the discrepancy is discontinuous, hence the associated Fourier expansion does not converge absolutely or uniformly. To overcome this problem, one can introduce a mollified discrepancy. If the domain \(\Omega \) is convex and contains the origin, then there exists \(\varepsilon >0\) such that if \(\varphi \left( x\right) \) is a non negative smooth radial function with support in \(\left\{ \left| x\right| \le \varepsilon \right\} \) and with integral 1, and if \(0<\delta \le 1\) and \(r\ge 1\), then
and therefore
One has \(\left| \left( r+\delta \right) ^{d}-r^{d}\right| \le Cr^{d-1}\delta \), and one can work with the mollified discrepancy defined by
Observe that since \(\left| \widehat{\varphi }\left( \zeta \right) \right| \le C\left( 1+\left| \zeta \right| \right) ^{-\lambda }\) for every \(\lambda >0\), the mollified Fourier expansion has no problems of convergence.
Let us recall that the support function of a convex set \(\Omega \subset {\mathbb {R}}^d\) is defined by \(g(x)=\sup _{y\in \Omega }\left\{ x\cdot y\right\} \). When \(\Omega \) is strictly convex with smooth boundary, the point y that realizes the supremum in this definition is the point of \(\partial \Omega \) where the outer normal is parallel to x (see Lemma 3.1 (2) and Fig. 1). For example, if \(\Omega \) is the unit ball centered at the origin, then \(g(x)=|x|\).
Lemma 2.3
Assume that \(\Omega \) is a convex body in \({\mathbb {R}}^{d}\) with smooth boundary with strictly positive Gaussian curvature. Let \(g\left( x\right) =\sup _{y\in \Omega }\left\{ x\cdot y\right\} \) be the support function of \(\Omega \). Then, there exist functions \(\left\{ a_{j}\left( \xi \right) \right\} _{j=0}^{+\infty }\) and \(\left\{ b_{j}\left( \xi \right) \right\} _{j=0}^{+\infty }\) homogeneous of degree 0 and smooth in \({\mathbb {R}}^{d}\setminus \left\{ 0\right\} \) such that the Fourier transform of the characteristic function of \(\Omega \) for \(\left| \xi \right| \rightarrow +\infty \) has the asymptotic expansion
The functions \(a_{j}\left( \xi \right) \) and \(b_{j}\left( \xi \right) \) depend on a finite number of derivatives of a parame-trization of the boundary of \(\Omega \) at the points with outward unit normal \(\pm \xi /\left| \xi \right| \). In particular, \(a_{0}\left( \xi \right) \) and \(b_{0}\left( \xi \right) \) are, up to some absolute constants, equal to \(K\left( \pm \xi \right) ^{-1/2}\), with \(K\left( \pm \xi \right) \) the Gaussian curvature of \(\partial \Omega \) at the points with outward unit normal \(\pm \xi /\left| \xi \right| \).
Proof
This is a classical result. See e.g. [6,7,8, 17]. Here, as an explicit example, we just recall that the Fourier transform of a ball \(\left\{ x\in {\mathbb {R}}^{d}:\left| x\right| \le R\right\} \) can be expressed in terms of a Bessel function, and Bessel functions have simple asymptotic expansions in terms of trigonometric functions,
See [18]. More generally, also the Fourier transform of an ellipsoid, that is an affine image of a ball, can be expressed in terms of Bessel functions. \(\square \)
By the above lemma, the normalized discrepancy has a Fourier expansion of the form
By replacing \((d+1)/2\) with a complex variable z one obtains an analytic function of this complex variable with values in \(L^p\) spaces. This will allow us to estimate the norm of this discrepancy via complex interpolation.
Lemma 2.4
Assume that \(\Omega \) is a convex body in \({\mathbb {R}}^{d}\) with smooth boundary with strictly positive Gaussian curvature. Let z be a complex parameter, and for every \(h=0,1,2,\ldots \) and \(r\ge 1\), with the notation of the previous lemmas, let define the function \(\Phi \left( \delta , z,r,x\right) \) via the Fourier expansions
The convergence of the above series is absolute and uniform. With \(z=(d+1)/2\), define
Then, if \(h>\left( d-3\right) /2\) there exists C such that for every \(\delta >0\) and \(r\ge 1\),
Proof
This is a consequence of the previous lemmas. \(\square \)
Lemma 2.5
Let N be a positive integer, and \(\mu \) a positive measure with compact support in the positive real axis and with Fourier transform satisfying \(|\widehat{\mu }(\xi )|\le (1+|\xi |)^{-\beta }\) for some \(\beta \ge 0\). Then for every \(\lambda >0\) and for every \(z\in \mathbb C\) there exists \(C>0\) such that for every \(\delta >0\) and for every \(1<R<+\infty \),
The inner integrals are with respect to the surface measure on the \((N-1)d\) dimensional variety of N points with sum k. In other words, they are essentially Lebesgue integrals with respect to the first \(N-1\) variables, replacing the last variables \( m_N\) and \(n_N\) respectively with \(k-m_1-\cdots -m_{N-1}\) and \(k-n_1-\cdots -n_{N-1}\). The above final expression is a decreasing function of \({\mathrm {Re}}(z)\).
Proof
The function \(\Phi (\delta ,z,r,x)\) is a sum of terms of the form
The expression \(\exp \left( \mp 2\pi ig\left( \pm n\right) r\right) \) has to be intended either as \(\exp \left( -2\pi ig\left( n\right) r\right) \) or \(\exp \left( +2\pi ig\left( - n\right) r\right) \). Assume that the first exponential is \(\exp (-2\pi ig\left( n\right) r)\). It follows that for every positive integer N,
For a proof, just observe that since \(\widehat{\varphi }(\xi )\) has a fast decay at infinity, all series involved are absolutely convergent, and one can freely expand the Nth power and rearrange the terms. Then, by Parseval equality,
Expanding the square and integrating in the variable r, one obtains
The last integral is the Fourier transform of the measure \(r^{-2jN}d\mu (r-R)\) and it is easy to see that if \(R>1\), so that the singularity of \(r^{-2jN}\) is far from the support of \(d\mu (r-R)\), it satisfies the same estimates as the Fourier transform of \(d\mu (r)\). Indeed
where q(s) is the Fourier transform of a smooth extension outside the support of \(d\mu \) of the function \((r+R)^{-2jN}\). Hence,
The function \(\varphi (x)\) is smooth, so that \(|\widehat{\varphi }(\xi )|\le C\left( 1+|\xi |\right) ^{-\lambda }\) for every \(\lambda >0\). Hence for every j the above quantity is dominated up to a constant by
In this formula there is no cutoff in the variable k. In order to obtain a cutoff in k, observe that, if \(m_1+\cdots +m_N=k\), then
In particular, some of the cutoff functions \((1+\delta |m_1|)^{-\sigma }\cdots (1+\delta |m_N|)^{-\sigma }\) can be replaced with \((1+\delta |k|)^{-\sigma }\). Finally, in the above formulas one can replace the sums with integrals. Indeed, there exist positive constants A and B such that for every integer point \(m\ne 0\) and every \(x\in Q\), the cube centered at the origin with sides parallel to the axes and of length one,
This implies that the function \(|m+x|^{-{\mathrm {Re}}(z)}\) is slowly varying in the cube Q. Moreover, also the function \( \left( 1+| g(m+x)+\cdots |\right) ^{-\beta } \) is slowly varying. Hence, one can replace a sum over m with an integral over the union of cubes \(m+Q\),
Finally, with a change of variables one can transform the domain of integration \(\left\{ |x|>1/2\right\} \) into \(\left\{ |y|>1\right\} \), and the thesis follows immediately. Indeed, if \(|x|>1\) then \(|x|^{-{\mathrm {Re}}(z)}\) decreases as \({\mathrm {Re}}(z)\) increases. \(\square \)
3 The Case of a Generic Convex Set with Non Vanishing Curvature
This section contains the proof of Theorems 1.1 and 1.2.
Lemma 3.1
Let \(g\left( x\right) =\sup _{y\in \Omega }\left\{ x\cdot y\right\} \) be the support function of a convex set \(\Omega \) which contains the origin, and with a smooth boundary with strictly positive Gaussian curvature.
(1) The support function is convex, homogeneous of degree one, positive and smooth away from the origin, and it is equivalent to the Euclidean norm, that is there exist \(0<A<B\) such that for every x,
(2) For every \(x\in {\mathbb {R}}^d\setminus \{0\}\) there exist a unique point \(z(x)\in \partial \Omega \) such that \(x\cdot z(x)=g(x)\). In particular, z(x) is the unique point in \(\partial \Omega \) with outer normal x. Furthermore z(x) is homogeneous of degree 0 in the variable x.
(3) There exist two positive constants c and C such that for every \(\vartheta ,\omega \) with \(\left| \vartheta \right| =\left| \omega \right| =1\)
Proof
(1) is trivial and (2) follows from the smoothness and positive curvature of the boundary of \(\Omega \). The estimate in (3) can be written as
A compactness argument implies that there exist two radii \(c\le C\) such that at every point of \(\partial \Omega \), \(\Omega \) contains the ball with radius c tangent to \(\partial \Omega \) and is contained in the ball with radius C tangent to \(\partial \Omega \), and (3) follows. \(\square \)
Lemma 3.2
In the hypotheses of the above lemma, assume that \(\beta \ge 0\), \(\delta ,\gamma >0\) and \(Y\ge 1\). Set
Then, for every \(\vartheta \) with \(\left| \vartheta \right| =1\) and every \(k\ne 0\), if we call \(\Delta =g\left( \vartheta \right) -\nabla g\left( k\right) \cdot \vartheta \),
Remark 3.3
Here and in the rest of the paper we will use the Kronecker delta notation
Proof
The case \(\beta =0\) is trivial. When \(\beta >0\), then \(\left( 1+Y\Delta \right) ^{-\tau }\log ^{\sigma }\left( 2+Y\Delta \right) \) is the gain with respect to the trivial estimate
Fix \(\vartheta \) and k and define
The quadratic form \(\vartheta ^{t}\cdot \nabla ^{2}g\left( x\right) \cdot \vartheta \) is positive semidefinite, with a 0 eigenvalue in the direction given by \(\nabla g\left( x\right) \), and strictly positive eigenvalues in the other directions. See [16]. Let \(\lambda >0\) be the minimum of all the \(d-1\) positive eigenvalues of \(\vartheta ^{t}\cdot \nabla ^{2}g\left( x\right) \cdot \vartheta \) over the sphere \(\left\{ \left| x\right| =1\right\} \). Splitting \(\vartheta =\vartheta _{0}+\vartheta _{1}\) with \(\vartheta _{0}\) parallel to \(\nabla g\left( k-\rho \vartheta \right) \) and \(\vartheta _{1}\) orthogonal to \(\nabla g\left( k-\rho \vartheta \right) \), and since \(\nabla ^{2}g\left( x\right) \) is homogeneous of degree \(-1\), we therefore have
In particular, \(F^{\prime \prime }\left( \rho \right) \ge 0\), so that \(F^{\prime }\left( \rho \right) \) is increasing and if \(\rho \ge 0\),
If \(z\left( x\right) \) is the unique point in the boundary of \(\Omega \) such that \(x\cdot z\left( x\right) =g\left( x\right) \), as described in Lemma 3.1, then
The last equality follows from Euler’s formula, since \(z\left( x\right) \) is homogeneous of degree 0. Hence,
This follows by the definition of \(g\left( \vartheta \right) \) as the maximum of \(y\cdot \vartheta \) for \(y\in \Omega \).
Thus, \(F\left( \rho \right) \) is an increasing function. If \(Y\ge F\left( 0\right) \) define \(r=F^{-1}\left( Y\right) \), and if \(Y<F\left( 0\right) \) define \(r=0\). Observe that for \(r>0\) one has \(Y=F\left( r\right) \ge g\left( r\vartheta \right) \ge Cr\). Thus, in any case, \(0\le r\le CY\). Then
Hence,
Call \(\delta \Delta Y=T\) and \(\Delta r=S\). Then we need to show that
In any case we have
In particular, the thesis follows when \(T\le 4\).
If \(T\ge 4\) and \(0\le S\le 2\) then
If \(T\ge 4\) and \(2\le S\le T/2\) then
If \(T\ge 4\) and \(T/2\le S\le 2T\), then
If \(T\ge 4\) and \(S\ge 2T\), then
\(\square \)
Lemma 3.4
In the hypotheses of the previous lemma and with
one has
Proof
Fix k. By the previous lemma and with the notation \(\omega =k/\left| k\right| \) and \(\Delta \left( \vartheta ,\omega \right) =g\left( \vartheta \right) -\nabla g\left( \omega \right) \cdot \vartheta \), we have to estimate
By Lemma 3.1, since \(\Omega \) has everywhere positive curvature,
Therefore,
\(\square \)
Lemma 3.5
Let \(g\left( x\right) \) be the support function of \(\Omega \), and let \(d\ge 2\), \(d/2<\alpha <d\), \(\beta \ge 0\), \(\zeta =\zeta (d,\alpha ,\beta )=\min \left\{ 1,\beta ,d-\alpha ,\left( d-1\right) /2\right\} \) (Fig. 2). Finally, define \(\eta =\eta (d,\alpha ,\beta )\) as follows.
If \(d=2\) define
If \(d=3\) define
If \(d\ge 4\) define
Then there exists C such that for every \(k\in {\mathbb {R}}^{d}\setminus \left\{ 0\right\} \) and for every \(-\infty<Y<+\infty \),
Proof
Let us explain the numerology behind the lemma. If there is no cutoff \(\left( 1+\left| g\left( x\right) +g\left( k-x\right) -Y\right| \right) ^{-\beta }\), then the change of variables \(x=\left| k\right| z\) and \(k=\left| k\right| \omega \) gives
On the other hand, the cutoff \(\left( 1+\left| g\left( x\right) +g\left( k-x\right) -Y\right| \right) ^{-\beta }\) gives an extra decay. In particular, the integral with the cutoff \(\left( 1+\left| g\left( x\right) +g\left( k-x\right) -Y\right| \right) ^{-\beta }\) with \(\beta \) large is essentially over the \(d-1\) dimensional set \(\left\{ g\left( x\right) +g\left( k-x\right) =Y\right\} \), that is the cutoff reduces the space dimension by 1. Hence, at least when \(\beta \) is large, the integral with the cutoff can be seen as the convolution in \({\mathbb {R}}^{d-1}\) of two homogeneous functions of degree \(-\alpha \), and this suggests the decay \(\left| k\right| ^{d-1-2\alpha }\). Hence, if with \(\beta =0\) the decay is \(\left| k\right| ^{d-2\alpha }\), and if with \(\beta >1\) the decay is \(\left| k\right| ^{d-1-2\alpha }\), then, by interpolation, when \(0<\beta <1\) the decay is \(\left| k\right| ^{d-\beta -2\alpha }\). This is just a rough numerology, indeed also the parameter Y enters into play and the details of the proof are more delicate.
For every Y and k one has
Assume \(\left| k\right| +\left| Y\right| \ge 1.\) Since \(c\left| x\right| \le g\left( x\right) \le C\left| x\right| \), one has
Hence, if \(-\infty <Y\le \varepsilon \left| k\right| \) for a small enough \(\varepsilon >0\), one also has
In this case \(-\infty <Y\le \varepsilon \left| k\right| \),
Assume now \(\left| k\right| +\left| Y\right| \ge 1\) and \(Y\ge \varepsilon \left| k\right| \). Let us split the integral into the three sets \(\left\{ \left| x\right| +\left| k\right| \le \varepsilon Y\right\} \), \(\left\{ \varepsilon Y\le \left| x\right| +\left| k\right| \le \delta Y\right\} \) and \(\left\{ \delta Y\le \left| x\right| +\left| k\right| <+\infty \right\} \), with \(\varepsilon \) small and \(\delta \) large. In \(\left\{ \left| x\right| +\left| k\right| \le \varepsilon Y\right\} \) one has
Hence
In \(\left\{ \delta Y\le \left| x\right| +\left| k\right| <+\infty \right\} \) one has
Hence
It remains to estimate the integral over the spherical shell
Recall that \(Y\ge \varepsilon \left| k\right| \). Hence if \(\varepsilon \left| k\right| \le Y\le 4\left| k\right| /\varepsilon \), then
Similarly, if \(Y\ge 4\left| k\right| /\varepsilon \), then
In polar coordinates \(x=\rho \vartheta \) with \(\rho \ge 0\) and \(\left| \vartheta \right| =1\) the first integral takes the form
Similarly, the second integral takes the form
By Lemma 3.4, the first integral can be bounded by
where
Again by Lemma 3.4, the second integral can be bounded by
where
It is easy to show that when \(d\ge 3\) then \(\varsigma _{1}+\sigma _{1}\le \eta \), and that when \(d\ge 3\) and \(\alpha \le d-1\), then \(\varsigma _{2}+\sigma _{2}\le \eta \). This shows the lemma in the case \(d\ge 3\). The case \(d=2\) is more delicate. First observe that the integral
is a decreasing function of the variable \(\beta \). It follows that for \(1/2<\beta <1\) and \(\alpha =2-\beta \), one can take some \(\tilde{\beta }\in \left( 1/2,\beta \right) \) and obtain
This shows that for \(d=2\) and \(1/2<\beta <1\) and \(\alpha =2-\beta \) one can indeed take \(\eta =0.\) A similar argument shows that one can take \(\eta =0\) also when \(\beta =1\) and \(1<\alpha <3/2\). In the remaining cases, it is easy to show that \(\varsigma _{1}+\sigma _{1}\le \eta \), and that when \(\alpha \le 3/2\), then \(\varsigma _{2} +\sigma _{2}\le \eta \). \(\square \)
Lemma 3.6
Let \(z_2=d/2\). If \({\mathrm {Re}}\left( z\right) \ge z_2\), there exists \(C>0\) such that for every \(R\ge 1\) and \(0<\delta <1/2\),
Proof
By Plancherel formula applied to \(\Phi \left( \delta ,z,r,x\right) \) as a function of the variable x,
Observe that if \(R>1\) then the singularity of \(r^{-j}\) is outside the support of \(d\mu (r-R)\), and recall that \(\widehat{\varphi }(r)\) has fast decay at infinity. Hence for every \(\lambda >0\),
and the lemma follows. \(\square \)
The following lemma is an estimate of the \(L^p \) norms of the function \(\Phi \left( \delta ,z,r,x\right) \) when \(p=4\) and the space dimension \(d\ge 2\). In dimension \(d=2\) there is a second relevant exponent \(p=6\), and this will be considered later.
Lemma 3.7
Let \(z\in \mathbb C\). Define \(\nu =\min \{1,(d-1)/2\}\) and let \(\eta =\eta (d,{\mathrm {Re}}(z),\beta )\) be defined as in Lemma 3.5. Let \(z_4=\max \{(3d-\beta )/4,\,(3d-\nu )/4\}\) (Fig. 3). If \({\mathrm {Re}}(z) \ge z_4\), then there exists \(C>0\) such that for every \(R\ge 1\) and \(0<\delta <1/2\),
Proof
Call \(\alpha ={\mathrm {Re}}\left( z\right) \). By the above Lemma 2.5 with \(N=2\), it suffices to estimate
Notice that we have canceled all the cutoff functions in the variables m, \(k-m\), n, \(k-n\). The integral over the set \(\left\{ | k| \le 2\right\} \) is bounded by
Let us now consider the integral over the set \(\left\{ |k| \ge 2\right\} \),
By Lemma 3.5, the inner integral
is bounded by
Thus, the goal estimate becomes
The result now follows after the observation that \(\alpha >(3d-\zeta )/4\) if and only if \(\alpha >z_4\) and \(\alpha =(3d-\zeta )/4\) if and only if \(\alpha =z_4\). \(\square \)
In the following lemma the space dimension is \(d=2\).
Lemma 3.8
Let \(d=2\) and let \(z_6=\max \{(10-\beta )/6,8/5\}\) (Fig. 3a). If \({\mathrm {Re}}(z) \ge z_6\), then there exists \(C>0\) such that for every \(R\ge 1\) and \(0<\delta <1/2\),
Proof
Call \(\alpha ={\mathrm {Re}}\left( z\right) \). By Lemma 2.5 with \(N=3\), it suffices to estimate
Split \({\mathbb {R}}^{2}\) as \(\left\{ |k| \le 2\right\} \cup \left\{ | k| \ge 2\right\} \). The integral over the disc \(\left\{ | k| \le 2\right\} \) is bounded by
since \(\alpha \ge z_6>4/3\). Consider now the case \(\left\{ | k| \ge 2\right\} \). Apply Lemma 3.5 to the integral with respect to \(n_2\) with k replaced with \(k-n_1\) and Y replaced with \(g(m_1)+g(m_2)+g(k-m_1-m_2)-g(n_1)\),
Here \(\eta =\eta (2,\alpha ,\beta )\) as defined in Lemma 3.5. Moreover,
Finally, the integral over \(\left\{ | k| \ge 2\right\} \) gives
The result now follows after the observation that \(\alpha >(10-\zeta )/6\) if and only if \(\alpha >z_6\) and \(\alpha =(10-\zeta )/6\) if and only if \(\alpha =z_6\). \(\square \)
Lemma 3.9
The notation is as in the previous lemmas.
-
(1)
Let \(d=3\). If \(\beta <1\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge 2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
$$\begin{aligned}&\left\{ \int _{{\mathbb {R}}} \int _{{\mathbb {T}}^{d}}| \Phi \left( \delta ,z,r,x\right) | ^{p}dxd\mu (r-R)\right\} ^{1/p}\\&\quad \le {\left\{ \begin{array}{ll} C &{} \quad \text {if }\, p<3+\beta /(2-\beta ),\\ C\log ^{1/p}\left( 1/\delta \right) &{} \quad \text {if }\, p=3+\beta /(2-\beta ). \end{array}\right. } \end{aligned}$$If \(\beta =1\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge 2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
$$\begin{aligned} \left\{ \int _{{\mathbb {R}}} \int _{{\mathbb {T}}^{d}}| \Phi \left( \delta ,z,r,x\right) | ^{p}dxd\mu (r-R)\right\} ^{1/p} \le {\left\{ \begin{array}{ll} C &{} \quad \text {if }\, p<4,\\ C\log ^{3/4}\left( 1/\delta \right) &{} \quad \text {if }\, p=4. \end{array}\right. } \end{aligned}$$If \(\beta >1\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge 2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
$$\begin{aligned} \left\{ \int _{{\mathbb {R}}} \int _{{\mathbb {T}}^{d}}| \Phi \left( \delta ,z,r,x\right) | ^{p}dxd\mu (r-R)\right\} ^{1/p} \le {\left\{ \begin{array}{ll} C &{} \quad \text {if }\, p<4,\\ C\log ^{1/2}\left( 1/\delta \right) &{} \quad \text {if }\, p=4. \end{array}\right. } \end{aligned}$$ -
(2)
Let \(d\ge 4\). If \(\beta <1\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge (d+1)/2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
$$\begin{aligned}&\left\{ \int _{{\mathbb {R}}} \int _{{\mathbb {T}}^{d}}| \Phi \left( \delta ,z,r,x\right) | ^{p}dxd\mu (r-R)\right\} ^{1/p}\\&\quad \le {\left\{ \begin{array}{ll} C &{} \quad \text {if }\, p<2(d-\beta )/(d-\beta -1),\\ C\log ^{1/p}\left( 1/\delta \right) &{} \text {if }p=2(d-\beta )/(d-\beta -1). \end{array}\right. } \end{aligned}$$If \(\beta =1\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge (d+1)/2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
$$\begin{aligned}&\left\{ \int _{{\mathbb {R}}} \int _{{\mathbb {T}}^{d}}| \Phi \left( \delta ,z,r,x\right) | ^{p}dxd\mu (r-R)\right\} ^{1/p}\\&\quad \le {\left\{ \begin{array}{ll} C &{} \quad \text {if }\, p<2(d-1)/(d-2),\\ C\log ^{1/2}\left( 1/\delta \right) &{} \quad \text {if }\, p=2(d-1)/(d-2). \end{array}\right. } \end{aligned}$$If \(\beta >1\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge (d+1)/2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
$$\begin{aligned}&\left\{ \int _{{\mathbb {R}}} \int _{{\mathbb {T}}^{d}}| \Phi \left( \delta ,z,r,x\right) | ^{p}dxd\mu (r-R)\right\} ^{1/p}\\&\quad \le {\left\{ \begin{array}{ll} C &{} \quad \text {if }\, p<2(d-1)/(d-2),\\ C\log ^{1/p}\left( 1/\delta \right) &{} \quad \text {if }\, p=2(d-1)/(d-2). \end{array}\right. } \end{aligned}$$
Proof
It is enough to prove the result for \(z=(d+1)/2\). The Lemma follows via complex interpolation. For the definition of the complex interpolation method and the complex interpolation of \(L^p \) spaces, see for example [1, Chapters 4 and 5]. Here we recall the relevant result: Let \({\mathbb {X}}\) be a measure space, \(1\le a<b\le +\infty \), \(-\infty<A<B<+\infty \), and let \(\Phi \left( z\right) \) be a function with values in \(L^{a}\left( {\mathbb {X}}\right) +L^{b}\left( {\mathbb {X}}\right) \), continuous and bounded on the closed strip \(\left\{ A\le {\mathrm {Re}}\left( z\right) \le B\right\} \) and analytic on the open strip \(\left\{ A<{\mathrm {Re}}\left( z\right) <B\right\} \). Assume that there exist constants H and K such that for every \(-\infty<t<+\infty \),
If \(1/p=\left( 1-\vartheta \right) /a+\vartheta /b\), with \(0<\vartheta <1\), then
In our case, the analytic function is \(\Phi \left( \delta ,z,r,x\right) \), the measure space is \({\mathbb {R}}\times {\mathbb {T}}^{d}\) with measure \(d\mu (r-R)dx\), \(a=2\), \(b=4\), \(A=z_2+\varepsilon \), \(B=z_4+\varepsilon \), with \(\varepsilon \ge 0\) (Fig. 3b). The norms H and K are given in Lemmas 3.6 and 3.7. Set
This gives
and
When \(\varepsilon >0\) and \(p<{2(d-\min \{\beta ,1\}) }/{(d-1-\min \{\beta ,1\})} \),
When \(\varepsilon =0\) and \(p={2(d-\min \{\beta ,1\}) }/{(d-1-\min \{\beta ,1\})} \),
where \(\eta =\eta (d,z_4,\beta )\) as in Lemma 3.5. \(\square \)
Lemma 3.10
The notation is as in the previous lemmas and let \(d=2\).
If \(0\le \beta <2/5\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge 3/2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
If \(\beta =2/5\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge 3/2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
If \(2/5<\beta <1/2\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge 3/2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
If \(\beta =1/2\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge 3/2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
If \(\beta >1/2\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge 3/2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
Proof
Again, it is enough to prove the result for \(z=3/2\). The case \(\beta =0\) is contained in Lemma 3.7. If \(\beta >0\) the proof follows by complex interpolation with \(a=4\), \(b=6\), \(A=z_4+\varepsilon \), \(B=z_6+\varepsilon \), with \(\varepsilon \ge 0\) (Fig. 3a). The norms H and K are given in Lemmas 3.7 and 3.8. Set
This gives
and
When \(\varepsilon >0\) and \(p<{24(z_6-z_4)}/{(6z_6-4z_4-3)} \),
When \(\varepsilon =0\) and \(p={24(z_6-z_4)}/{(6z_6-4z_4-3)} \),
where \(\eta =\eta (2,z_4,\beta )\) as in Lemma 3.5 and \(\omega \) is the exponent of the logarithm in Lemma 3.8, that is \(\omega =1\) if \(\beta \ne 2/5\) and \(\omega =2\) if \(\beta =2/5\). \(\square \)
Proof
(of the Theorems 1.1 and 1.2) By Remark 2.2, one has
If \(\delta =R^{-\left( d-1\right) /2}\) the last term is bounded. On the other hand, by Lemma 2.4
If \(h\ge \left( d-3\right) /2\) then the last term is bounded, while the first term can be written as
The theorem now follows from the two previous lemmas, with R replaced by \(R\pm \delta \).
\(\square \)
4 The Case of the Ellipse
Here we assume \(d=2\) and \(\Omega =E=\{x\in {\mathbb {R}}^2:|M^{-1}x|\le 1\}\), where M is a non singular \(2\times 2\) matrix. In this case the support function is \(g(x)=|M^Tx|\). By the change of variable \(M^Tx=y\) applied to all variables \(n_1,\ldots ,n_N,\,m_1,\ldots ,m_N\), in this case Lemma 2.5 can be restated as
Lemma 4.1
-
(1)
If \(\beta \ge 0\) and \(0<2-\alpha <\beta \), there exists C such that for every \(-\infty<X<+\infty \),
$$\begin{aligned} \int _{0}^{+\infty }\left( 1+| X-\tau | \right) ^{-\beta }\tau ^{1-\alpha }d\tau \le C(1+|X|)^{2-\alpha -\min \{\beta ,1\}}\log ^{\delta _{1,\beta }}(2+|X|). \end{aligned}$$ -
(2)
If \(\alpha <2\), then for every \(\beta \) there exists C such that for every \(-\infty<X<+\infty \),
$$\begin{aligned} \int _{0}^{1}\left( 1+| X-\tau | \right) ^{-\beta }\tau ^{1-\alpha }(-\log \left( \tau \right) ) d\tau \le C\left( 1+| X| \right) ^{-\beta }. \end{aligned}$$ -
(3)
If \(0\le \beta <1\) and \(2-\alpha \ge \beta \), then there exists C such that for every \(-\infty< X<+\infty \) and \(2\le T<+\infty \),
$$\begin{aligned} \int _{0}^{T}\left( 1+| X-\tau | \right) ^{-\beta }\tau ^{1-\alpha }d\tau \le CT^{2-\alpha -\beta }\log ^{\delta _{2-\alpha ,\beta }}\left( T\right) . \end{aligned}$$ -
(4)
If \(0\le \beta <1\) and \(\alpha >1\), there exists C such that for every \(-\infty< X<+\infty \) and \(2\le T<+\infty \),
$$\begin{aligned} \int _{T}^{+\infty }\left( 1+| X-\tau | \right) ^{-\beta }\tau ^{1-2\alpha }d\tau \le CT^{2-2\alpha -\beta }. \end{aligned}$$
Proof
(1) If \(X\le 0\), then
If \(0\le X\le 1\), then
If \(X\ge 1\), then
(2) It suffices to observe that there exists C such that for every X,
(3) Let \(X\ge 0\). If \(T\le X/2\), then
If \(X/2\le T\le 2X\), then
If \(T\ge 2X\), then
If \(X<0\) then simply observe that
(4) As before, it suffices to show the result for \(X\ge 0\). If \(T\le X/2\), then
If \(X/2\le T\le 2X\), then
If \(T\ge 2X\), then
\(\square \)
Lemma 4.2
(1) Let \(3/2\le \alpha <2\) and \(\beta >2-\alpha \). Then there exists C such that for every \(k\in {\mathbb {R}}^{2} \) with \(|k|\ge 2\) and for every \(-\infty<Y<+\infty \),
(2) Let \(3/2<\alpha <2\) and \(0\le \beta \le 2-\alpha \). Then there exists C such that for every \(k\in {\mathbb {R}}^{2} \) with \(|k|\ge 2\) and for every \(-\infty<Y<+\infty \),
(3) Let \(\alpha =3/2\) and \(\beta =1/2\). Then there exists C such that for every \(k\in {\mathbb {R}}^{2} \) with \(|k|\ge 2\) and for every \(-\infty<Y<+\infty \),
(4) Let \(3/4<\alpha <3/2\) and \(\beta \ge 0\). Then there exists C such that for every \(k\in {\mathbb {R}}^{2} \) with \(|k|\ge 2\) and for every \(-\infty<Y<+\infty \),
(5) Let \(\alpha >1\). Then for every \(k\in {\mathbb {R}}^2\)
Proof
The symmetry between 0 and k gives
We estimate here the first integral, the second being studied similarly. The integral is invariant under rotations of k, so that one can assume \(k=\left( | k| ,0\right) \). Write in polar coordinates \(y=\left( \rho \cos \left( \vartheta \right) ,\rho \sin \left( \vartheta \right) \right) \), with \(0\le \rho <+\infty \), \(0\le \vartheta \le 2\pi \). In these polar coordinates the ellipse \(\left\{ | x| +| k-x| =\tau \right\} \) has equation
In the variables \(\left( \tau ,\vartheta \right) \), \(|k| \le \tau <+\infty \), \(0\le \vartheta \le 2\pi \), one has
and
Hence,
The term \(1+\left( | k| /\tau \right) \) in the last double integral is bounded between 1 and 2. Hence,
where
When \(0<t<1/2\), \(E(t,\alpha )\le C\). When \(1/2\le t<1,\) the integral over \(\pi /2\le \vartheta \le \pi \) is bounded independently of t, and when \(0\le \vartheta \le \pi /2\) one has \(1-\vartheta ^{2}/2\le \cos \left( \vartheta \right) \le 1-4\vartheta ^{2}/\pi ^{2}\). Hence one ends up with the integral
Hence
Similarly,
Assume \(3/2<\alpha <2\) and \(\alpha >2-\beta \). Then applying Lemma 4.1 (1), we obtain
Assume now \(\alpha =3/2\) and \(\beta >1/2\). Then
Applying Lemma 4.1 (1), we obtain
Applying Lemma 4.1 (1) and (2), we obtain
If \(3/2<\alpha <2\) and \(\alpha \le 2-\beta \), then applying Lemma 4.1 (3), we obtain
and, by Lemma 4.1 (4),
Assume \(\alpha =3/2\) and \(\beta =1/2\). Then, by Lemma 4.1 (2), (3) and (4),
Assume \(3/4<\alpha < 3/2\) and \(\beta >1/2\). Then applying Lemma 4.1 (1), we obtain
If \(3/4<\alpha <3/2\) and \(0\le \beta \le 1/2\), then applying Lemma 4.1 (3), we obtain
and, by Lemma 4.1 (4),
Finally, (5) follows by a simple rearrangement inequality (see [13, Theorem 3.4]),
\(\square \)
The next two lemmas are the counterpart of Lemmas 3.7 and 3.8 in the case of the ellipse. The minimal value \(z_4\) of \({\mathrm {Re}}(z)\) for which the \(L^4\) norm of \(\Phi (\delta ,z,\cdot ,\cdot )\) is proved to be bounded is lowered from \(\max \{(6-\beta )/4,11/8\}\) to \(\max \{(6-\beta )/4,5/4\}\). Similarly, \(z_6\) is lowered from \(\max \{(10-\beta )/6,8/5\}\) to \(\max \{(10-\beta )/6,3/2\}\) (Figs. 3, 4).
Lemma 4.3
Let \(d=2\) and \(\beta \ge 0\). Set \(z_4=\max \{(6-\beta )/4,5/4\}\). If \({\mathrm {Re}}(z) \ge z_4\), then there exists \(C>0\) such that for every \(R\ge 1\) and \(0<\delta <1/2\),
Proof
Call \(\alpha ={\mathrm {Re}}\left( z\right) \). By Lemma 2.5 with \(N=2\), it suffices to estimate
Split \({\mathbb {R}}^{2}\) as \(\left\{ |k| \le 2\right\} \cup \left\{ | k| \ge 2\right\} \). By Lemma 4.2 (5), the integral over the disc \(\left\{ | k| \le 2\right\} \) is bounded by
Consider now the case \(\left\{ | k| \ge 2\right\} \). Assume \(\beta >1/2\). Apply Lemma 4.2 (4) to the integral with respect to n with Y replaced with \(|m|+|k-m| \).
Thus we obtain the integral
If \(\beta =1\) and \(\alpha >5/4\), the logarithm can be removed as long as one replaces \(\alpha \) with a slightly smaller value greater than 5 / 4. Therefore, if \((\alpha ,\beta )\ne (5/4,1)\), then we only need to estimate
where we have applied Lemma 4.2 (4). Finally, the integral over \(\left\{ | k| \ge 2\right\} \) gives
Assume \(0\le \beta \le 1/2\). Apply Lemma 4.2(4) to the integral with respect to n with Y replaced with \(|m|+|k-m| \).
Thus we obtain the integral
Finally, the integral over \(\left\{ | k| \ge 2\right\} \) gives
It remains to study the case \(\alpha =5/4\) and \(\beta =1\). By Lemma 4.2 (4),
Since there is a \(C>0\) such that for \(0<\delta <1/2\)
and \(2+||m|+|k-m|-|k||\le 2+2|m|\), we obtain
by Lemma 4.2 (4). \(\square \)
Lemma 4.4
Let \(d=2\) and \(\beta \ge 0\). Set \(z_6=\max \{(10-\beta )/6,3/2\}\). If \({\mathrm {Re}}(z) \ge z_6\), then there exists \(C>0\) such that for every \(R\ge 1\) and \(0<\delta <1/2\),
Proof
Call \(\alpha ={\mathrm {Re}}\left( z\right) \). By Lemma 2.5 with \(N=3\), it suffices to estimate
Split \({\mathbb {R}}^{2}\) as \(\left\{ |k| \le 2\right\} \cup \left\{ | k| \ge 2\right\} \). The integral over the disc \(\left\{ | k| \le 2\right\} \) is bounded by
since \(\alpha \ge z_6>4/3\).
Consider now the case \(\left\{ | k| \ge 2\right\} \). Assume \(\alpha >2-\beta \) and \(\alpha >3/2\). When \(|k-n_1|\ge 2\), apply Lemma 4.2 (1) to the integral with respect to \(n_2\) with k replaced with \(k-n_1\) and Y replaced with \(|m_1|+|m_2|+|k-m_1-m_2|-|n_1|\), and when \(|k-n_1|\le 2\) apply Lemma 4.2 (5) to the same integral,
The last integral is bounded by a constant times \(|k|^{-\alpha }\). If \(\beta =1\), the logarithm can be removed as long as one replaces \(\alpha \) with a slightly smaller value greater than 3 / 2. Thus we only need to estimate
where we have applied Lemma 4.2 (1) and (2). Moreover,
Finally, the integral over \(\left\{ | k| \ge 2\right\} \) gives
Assume now \(3/2<\alpha \le 2-\beta \) so that \(0\le \beta <1/2\). Apply Lemma 4.2 (2) and (5) to the integral with respect to \(n_2\) with k replaced with \(k-n_1\),
Moreover,
Finally, the integral over \(\left\{ | k| \ge 2\right\} \) gives
It remains to study the case \(\alpha =3/2\) and \(\beta \ge 1\). Apply Lemma 4.2 (1) and (5) to the integral with respect to \(n_2\) with k replaced with \(k-n_1\),
Since there is a \(C>0\) such that for \(0<\delta <1/2\)
we obtain
By Lemma 4.2(3), the above integrals are bounded by
Moreover,
Finally, the integral over \(\left\{ | k| \ge 2\right\} \) gives
\(\square \)
Lemma 4.5
The notation is as in the previous lemmas and let \(d=2\).
If \(0\le \beta <2/5\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge 3/2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
If \(\beta =2/5\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge 3/2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
If \(2/5<\beta <1\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge 3/2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
If \(\beta =1\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge 3/2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
If \(\beta >1\) then there exists a constant C such that for every \({\mathrm {Re}}\left( z\right) \ge 3/2\), for every \(R\ge 1\) and \(0<\delta <1/2\),
Proof
Again, it is enough to prove the result for \(z=3/2\). The case \(\beta =0\) is contained in Lemma 4.3. If \(\beta >0\) the proof follows by complex interpolation with \(a=4\), \(b=6\), \(A=z_4+\varepsilon \), \(B=z_6+\varepsilon \), with \(\varepsilon \ge 0\) (Fig. 4). The norms H and K are given in Lemmas 4.3 and 4.4. Set
This gives
and
When \(\varepsilon >0\) and \(p<\min \{6,4+2\beta \} \),
When \(\varepsilon =0\) and \(p=\min \{6,4+2\beta \} \),
where \(\omega =4\) if \(\beta =1\), \(\omega =3\) if \(\beta >1\), \(\omega =1\) if \(\beta =2/5\) and \(\omega =0\) in the other cases, while \(\eta =1\) if \(\beta =1\) and \(\eta =0\) in the other cases. \(\square \)
The proof of Theorem 1.3 can now be concluded as in the case of the general convex set with smooth boundary with strictly positive curvature.
5 Concluding Remarks
As already observed in the Introduction, the normalization \(r^{-(d-1)/2}{\mathcal {D}}(\Omega , r,x)\) is optimal in the sense that the exponent \((d-1)/2\) cannot be replaced by a smaller exponent. For the case \(p=2\) see [11], the case \(p>2\) immediately follows from the embedding of \(L^p\) in \(L^2\).
On the other hand, we do not know to which extent the range of indices \(p=p(\beta )\) for which the above theorems hold is optimal, even when \(\beta \ge 1\), which contains the case of a uniformly distributed measure, or when \(\beta =0\), which contains the case of a measure concentrated at a single point.
One can expect that a different normalization \(r^{-\gamma }{\mathcal {D}}(\Omega , r,x)\) with \(\gamma >(d-1)/2\) gives a larger range of indices p and \(\beta \). In any case the optimal range on indices is likely a quite hard problem. See for example the survey paper [10] for results and open problems in the cases \(p=+\infty \) and \(\mu \) a point mass measure.
References
Bergh, J., Löfström, J.: Interpolation Spaces, an Introduction. Springer, Berlin (1976)
Brandolini, L., Colzani, L., Gigante, G., Travaglini, G.: \(L^{p}\) and weak-\(L^{p}\) estimates for the number of integer points in translated domains. Math. Proc. Camb. Philos. Soc. 159, 471–480 (2015)
Colzani, L., Gariboldi, B., Gigante, G.: Mixed \(L^{p}(L^{2})\) norms of the lattice point discrepancy. Trans. Am. Math. Soc. (to appear). https://doi.org/10.1090/tran/7624
Falconer, K.: Fractal Geometry, 3rd edn. Wiley, New York (2014)
Gariboldi, B.: Norms of the lattice point discrepancy, Ph. D. Dissertation, Università degli Studi di Milano-Bicocca (2017)
Gelfand, I.M., Graev, M.I., Vilenkin, N.Ya.: Generalized Functions, vol. 5. Integral Geometry and Representation Theory, Translated from the Russian by Eugene Saletan, Academic Press, New York-London (1966)
Herz, C.S.: Fourier transforms related to convex sets. Ann. Math. 75, 81–92 (1962)
Hlawka, E.: Über Integrale auf convexen Körpern, I, II. Monatsh. Math. 54(1–36), 81–99 (1950)
Huxley, M.: A fourth power discrepancy mean. Monatsh. Math. 73, 231–238 (2014)
Ivić, A., Krätzel, E., Kühleitner, M., Nowak, W.G.: Lattice points in large regions and related arithmetic functions: recent developments in a very classic topic, Elementare und analytische Zahlentheorie, Schr. Wiss. Ges. Johann Wolfgang Goethe Univ. Frankfurt am Main, 20, Franz Steiner Verlag Stuttgart, Stuttgart, pp. 89–128 (2006)
Kendall, D.G.: On the number of lattice points inside a random oval. Q. J. Math. 19, 1–26 (1948)
Krätzel, E.: Lattice Points. Kluwer Academic Publisher, Dordrecht (1988)
Lieb, E.H., Loss, M.: Analysis. Graduate Studies in Mathematics, vol. 14, 2nd edn. American Mathematical Society, Providence, RI (2001)
Mattila, P.: Geometry of Sets and Measures in Euclidean Spaces. Fractals and Rectifiability. Cambridge Studies in Advanced Mathematics, vol. 44. Cambridge University Press, Cambridge (1995)
Salem, R.: On singular monotonic functions whose spectrum has a given Hausdorff dimension. Ark. Mat. 1, 353–365 (1951)
Schneider, R.: Convex Bodies: The Brunn-Minkowski Theory. Cambridge University Press, Cambridge (2013)
Stein, E.M.: Harmonic Analysis: Real-Variable Methods, Orthogonality, and Oscillatory Integrals. Princeton University Press, Princeton, NJ (1993)
Stein, E.M., Weiss, G.: Introduction to Fourier Analysis on Euclidean Spaces. Princeton University Press, Princeton, NJ (1971)
Travaglini, G.: Number Theory, Fourier Analysis and Geometric Discrepancy. London Mathematical Society Student Texts, vol. 81. Cambridge University Press, Cambridge (2014)
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by A. Iosevich.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Colzani, L., Gariboldi, B. & Gigante, G. \(L^{p}\) Norms of the Lattice Point Discrepancy. J Fourier Anal Appl 25, 2150–2195 (2019). https://doi.org/10.1007/s00041-019-09665-1
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00041-019-09665-1