Abstract
We study approximation properties of multivariate periodic functions from weighted Wiener spaces by sparse grid methods constructed with the help of quasi-interpolation operators. The class of such operators includes classical interpolation and sampling operators, Kantorovich-type operators, scaling expansions associated with wavelet constructions, and others. We obtain the rate of convergence of the corresponding sparse grid methods in weighted Wiener norms as well as analogues of the Littlewood–Paley-type characterizations in terms of families of quasi-interpolation operators.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
In many applied problems one needs to approximate high-dimensional functions in smooth function spaces. As it is known from previous research, traditional numerical methods such as the interpolation with tensor product grids suffer from the so-called “curse of dimensionality”. In other words, the computation time of such methods grows dramatically with the number of variables and the problem becomes intractable already for mild dimensions. One of the means to overcome these obstacles is to employ different sparse grids approximation methods and to impose additional assumption on smoothness. Typically, one assumes that a function belongs to a certain mixed smoothness Sobolev or Besov space (see, e.g., [6, 17]).
In this paper, we consider approximation methods that are based on generalized sparse grids (see, e.g., [6, 26]). Recall that for given parameters \(T\in [-\infty ,1)\), \(n\in {{\mathbb {N}}}\), and a family of univariate operators \(Y=(Y_j)_{j\in {{\mathbb {Z}}}_+}\), a sparse grid approximation method is defined as follows:
where
and \(Y_j^i\) denotes the univariate operator \(Y_j\) acting on functions in the variable \(x_i\) and \(Y_{-1}=0\). The most well studied case of the family Y is the classical Lagrange interpolation operators \(I=(I_j)_{j\in {{\mathbb {Z}}}_+}\) given by
where \(x_k^j=\frac{\pi k}{2^{j-1}}\) and \(\mathcal {D}_j(x)=\sum _{\ell =-2^{j-1}}^{2^{j-1}-1} e^{\textrm{i}\ell x}\) is the Dirichlet kernel. The corresponding sparse grid for a given level n is then
where \(\mathcal {I}_j=\{x_k^j\,:\,k=-2^j,\dots ,2^j-1\}\), i.e., \(P_{n,T}^I f(y)=f(y)\) for all \(y\in \Gamma (n,T)\) and \(f\in C({{\mathbb {T}}}^d)\). Here, the case \(T=-\infty \) corresponds to the interpolation on the full tensor grid; the case \(T=0\) represents interpolation on the Smolyak grid, which is also called the regular sparse grid (see [46], see also [17, Ch. 5]); and the case \(0<T<1\) resembles the so-called energy-norm based sparse grids (see [6, 24]).
One of the important characteristics of a sparse grid is its cardinality. Note that (see, e.g., [26])
and the same upper bound holds for the number of frequencies of the polynomial \(P_{n,T}^I f\). Thus, the most interesting cases for approximation with the algorithm \(P_{n,T}^I\) is when \(0\le T<1\) since in this case, for the number of elements needed to construct the corresponding algorithm, there is no exponential growth with the dimension d. Let us consider this case in more detail.
Approximation properties of the operators \(P_{n,T}^I\) have been mainly studied in the case \(T=0\), which corresponds to the classical Smolyak grids (see, e.g., [1, 2, 8, 12, 27, 45, 47,48,49]; see also the book [17, Ch. 4 and Ch. 5]). As an example, we mention the following \(L_q\)-error estimates for the Smolyak algorithm \(P_{n,0}^I\) in the case of approximation of functions from the Sobolev space \({\textbf{W}}_{p}^{\alpha }({{\mathbb {T}}}^d)\) of dominating mixed smoothness \({\alpha }>0\) (see, e.g., [45] and [17, Chapters 4 and 5]): if \(1< p,q<\infty \) and \({\alpha }>\max \{1/p,1/2\}\), then
where \(U{\textbf{W}}_{p}^{\alpha }\) denotes the unit ball in the space \({\textbf{W}}_{p}^{\alpha }({{\mathbb {T}}}^d)\). Similar estimates (for \(T=0\)) in the weighted Wiener spaces (or Korobov spaces) were obtained in [11, 27, 47].
In the case \(0<T<1\), the approximations by operators \(P_{n,T}^I\) have been mostly investigated for functions from the so-called generalized mixed smoothness (or hybrid smoothness) Sobolev space
where the parameter \({\beta }\) governs for the isotropic smoothness, whereas \({\alpha }\) reflects the smoothness in the dominating mixed sense. Herewith, the approximation error is estimated in the metric of the classical isotropic Sobolev space \(H^{\gamma }({{\mathbb {T}}}^d)=H^{0,{\gamma }}({{\mathbb {T}}}^d)\), see, e.g., [6, 7, 24, 25]. In particular, we mention a general result obtained in the recent paper [25]: let \({\alpha }\ge 0\), \({\beta }\ge 0\), \({\gamma }-{\beta }<{\alpha }\), \({\alpha }+\frac{{\beta }}{d}>\frac{1}{2}\). Then, for all \(f\in H^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) and \(n\in {{\mathbb {N}}}\), one has
whereFootnote 1
which again shows the importance of the case \(0\le T\le \frac{{\gamma }-{\beta }}{{\alpha }}<1\), cf. (1.2).
Along with the classical interpolation operators \(I=(I_j)_{j\in {{\mathbb {Z}}}_+}\) one also considers the family of the partial sums of Fourier series (see, e.g., [1, 2, 13]), families of convolution type operators (see, e.g., [45, 54]), quasi-interpolation operators based on scaled B-splines with integer knots (see, e.g., [14,15,16]).
In this paper, as a family Y in (1.1), we make use of the general quasi-interpolation operators defined by
where \(({\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\) is a family of univariate trigonometric polynomials and \((\widetilde{\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\) is a family of functions/distributions on \({{\mathbb {T}}}\). Note that in the non-periodic case approximation properties of operators of such type in various function spaces (classical and weighted \(L_p\), Sobolev, Besov, Wiener) have been studied, for example, in the works [21, 29, 30, 38, 39]. It worths noting that the operators (1.5) can be successfully employed in such applied problems, where the data contains some noise and the functional information is provided by other means than point evaluation (averages, divided differences, etc.), see, e.g., [10, 56].
In the periodic case, an analog of estimate (1.3) for any \({\alpha }>0\) has been recently established in [32] for the approximation processes \(P_{n,0}^Q\) with \(Q=(Q_j)_{j\in {{\mathbb {Z}}}_+}\) defined by (1.5). The corresponding proof is essentially based on the results from [35], where under different compatibility conditions on \(({\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\) and \((\widetilde{\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\), \(L_p\)-error of approximation by the operators \(Q_j\) were obtained. Similar results in weighted Wiener spaces and \(L_2({{\mathbb {T}}})\) have been derived in [28, 34], correspondingly.
The goal of the present work is to establish analogues of error estimate (1.4) for sparse grid approximation methods constructed using general quasi-interpolation operators. Comparing our findings with the previously known results, we stress two important differences. Firstly, we build the approximation schemes using quasi-interpolation operators (1.5) rather than the classical interpolation operators \((I_j)_{j\in {{\mathbb {Z}}}_+}\) constructed using the Dirichlet kernel and the values of a function at sets of equidistant interpolation nodes as in [7, 24, 25]. Secondly, we work with a more general scale of spaces, namely with the weighted Wiener spaces \(A_p^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) rather than with the Sobolev spaces \(H^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\), which correspond to the case \(p=2\). We would like to stress that possibility to vary the family \((\widetilde{\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\) allows us to prove the results under essentiality less restrictive conditions on the parameters \({\alpha }\) and \({\beta }\).
The paper is organized as follows. In Sect. 2 we introduce basic notations, define isotropic, mixed, and hybrid weighted Wiener spaces and general quasi-interpolation operators. Section 3 is devoted to auxiliary results. In particular, we prove the following useful estimate
see Lemma 3.1. In Sect. 4 we establish our main tools, the so-called “discrete” Littlewood-Paley type characterizations. In Sect. 5 we prove our main results: we consider approximation in the isotropic Wiener space \(A_q^{\gamma }({{\mathbb {T}}}^d)\) (in Subsect. 5.1) and in the mixed Wiener spaces \(A_{q,{\text {mix}}}^{\gamma }({{\mathbb {T}}}^d)\) (in Subsect. 5.2). In Subsect. 5.3 we discuss the sharpness of the obtained results. In Sect. 6 we consider the specific sets of parameters, where our main results (Theorems 5.1 and 5.3) provide the most effective error estimates with respect to the approximation rate and the number of degrees of freedom.
2 Weighted Wiener Spaces and Quasi-interpolation Operators
2.1 Basic Notation
In what follows, \({{\mathbb {Z}}}_+^d=\{{\varvec{x}}\in {{\mathbb {Z}}}^d:~x_i\ge ~{0}, i=1,\dots , d\}\) and \({{\mathbb {T}}}^d={\mathbb {R}}^d/2\pi {{\mathbb {Z}}}^d\) is the d-dimensional torus. Further, for vectors \({\varvec{x}}= (x_1,\dots , x_d)\) and \({\varvec{k}}=(k_1,\dots , k_d)\) in \({\mathbb {R}}^d\), we denote \(({\varvec{x}}, {\varvec{k}})=x_1k_1+\dots +x_dk_d\).
If \({\varvec{j}}\in {{\mathbb {Z}}}^d_+\), we set \(|\varvec{j}|_1=\sum _{k=1}^d j_k\), \(|\varvec{j}|_\infty =\max _{k=1}^d j_k\), and \(2^{{\varvec{j}}}=(2^{j_1},\dots ,2^{j_d})\). For \(1\le p\le \infty \), \(p'\) is given by \(\frac{1}{p}+\frac{1}{p'}=1.\) For \(1\le p,q\le \infty \), we set \({\sigma }_{p,q}=\left( \frac{1}{q}-\frac{1}{p}\right) _+\).
If \(f\in L_1({{\mathbb {T}}}^d)\), then
denotes the k-th Fourier coefficient of f. As usual, the convolution of integrable functions f and g is given by
By \(\mathcal {T}_{\varvec{j}}^d\), \({\varvec{j}}\in {{\mathbb {Z}}}_+^d\), we denote the following set of trigonometric polynomials:
where
Let \({\mathcal {D}} = C^{\infty }({{\mathbb {T}}})\) be the space of infinitely differentiable functions on \({{\mathbb {T}}}\). The linear space of periodic distributions (continuous linear functionals on \({\mathcal {D}}\)) is denoted by \({\mathcal {D}}'\). It is known (see, e.g., [44, p. 144]) that any periodic distribution \({\widetilde{\varphi }}\) can be expanded in a weakly convergent (in \({\mathcal {D}}'\)) Fourier series
where the sequence \((\widehat{\widetilde{\varphi }}(k))_k\) has at most polynomial growth.
Throughout the paper, we use the notation \( \, A \lesssim B, \) with \(A,B\ge 0\), for the estimate \(\, A \le C\, B,\) where \(\, C\) is a positive constant independent of the essential variables in \(\, A\) and \(\, B\) (usually, f, j, and n). If \(\, A \lesssim B\) and \(\, B \lesssim A\) simultaneously, we write \(\, A \asymp B\) and say that \(\, A\) is equivalent to \(\, B\). For two function spaces \(\, X\) and \(\, Y,\) we will use the notation \( \, Y \hookrightarrow X \) if \(\, Y \subset X\) and \(\, \Vert f\Vert _X \lesssim \Vert f\Vert _Y\) for all \(\, f \in Y.\) The unit ball in some normed vector space X is denoted by UX.
2.2 Weighted Wiener Spaces
We will employ the following function spaces with the parameters \({\alpha }, {\beta }\in {\mathbb {R}}\), and \(0<q\le \infty \).
\(\bullet \) The periodic (isotropic) Wiener space \(A_{q}^{\alpha }({{\mathbb {T}}}^d)\) is the collection of all \(f\in L_1({{\mathbb {T}}}^d)\) such that
In the case \({\alpha }=0\), we use the following standard notation \( A_q({{\mathbb {T}}}^d)=A_q^0({{\mathbb {T}}}^d). \)
\(\bullet \) The periodic (mixed) Wiener space \(A_{q,{\text {mix}}}^{\alpha }({{\mathbb {T}}}^d)\) is the collection of all \(f\in L_1({{\mathbb {T}}}^d)\) such that
\(\bullet \) The periodic (hybrid) Wiener space \(A_q^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) is the collection of all \(f\in L_1({{\mathbb {T}}}^d)\) such that
Remark 2.1
-
(i)
It is easy to see that for any \({\alpha }>0\) and \(1\le q\le \infty \), the following embeddings hold:
$$\begin{aligned} A^{d{\alpha }}_q({{\mathbb {T}}}^d) \hookrightarrow A^{{\alpha }}_{q,{\text {mix}}}({{\mathbb {T}}}^d) \hookrightarrow A^{{\alpha }}_q({{\mathbb {T}}}^d). \end{aligned}$$ -
(ii)
Note that the isotropic Wiener space \(A_q^{\alpha }({{\mathbb {T}}}^d)\) coincides with the corresponding periodic Sobolev space \(H^{\alpha }({{\mathbb {T}}}^d)\) in the case \(q=2\). The same holds for the mixed or hybrid Wiener spaces.
-
(iii)
For more information about the weighted Wiener spaces \(A_q^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\), see papers [5, 19, 20, 40, 41, 43].
-
(iv)
Note that according to the terminology suggested by H. Feichtinger, the spaces \(A_q^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) can be also called Fourier-Wermer spaces, see [55] for the motivation.
As usual, for \(f\in L_1({{\mathbb {T}}}^d)\), we define the diadic blocks \({\delta }_{\varvec{k}}(f)\), \({\varvec{k}}\in {{\mathbb {Z}}}_+^d\), by
where
\(P_j=\{\ell \in {{\mathbb {Z}}}\,:\, 2^{j-1}\le |\ell |<2^j\}\) for \(j>0\), and \(P_0=\{0\}\).
Recall that for all \(f\in L_p({{\mathbb {T}}}^d)\), \(1<p<\infty \), the Littlewood-Paley decomposition reads as follows
The next lemma is a simple consequence of the definition of the space \(A_q^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\).
Lemma 2.2
Let \(0< q\le \infty \), \({\alpha }\ge 0\), and let \({\beta }\in {\mathbb {R}}\) be such that \({\alpha }+{\beta }\ge 0\). Then
with the usual modification in the case \(q=\infty \) in the sense of equivalent norms.
2.3 Quasi-interpolation Operators
Consider a family of general univariate quasi-interpolation operators \(Q=(Q_j)_{j\in {{\mathbb {Z}}}_+}\) given by
where \(({\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\) is a family of univariate trigonometric polynomials in \(\mathcal {T}_j^1\), \((\widetilde{\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\) is a family of functions/distributions, and the convolution \(f*\widetilde{\varphi }_j\) is defined in some suitable way for any \(j\in {{\mathbb {Z}}}_+\).
Below, we assume that the following conditions on \(({\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\) and \((\widetilde{\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\) hold:
\(\bullet \) The growth condition of order \(N\ge 0\) for the Fourier coefficients of \(\widetilde{\varphi }_j\):
\(\bullet \) The uniform boundedness condition for the Fourier coefficients of \({\varphi }_j\):
\(\bullet \) The compatibility condition of order \(s> 0\) for \({\varphi }_j\) and \(\widetilde{\varphi }_j\):
Let us consider two important classes of quasi-interpolation operators and examine the above conditions.
Example 2.3
Quasi-interpolation sampling operators are defined by
where \(a_{\nu ,j}\in {\mathbb {C}}\), \(r_\nu \in {{\mathbb {Z}}}_+\), and \({\varphi }_j\in \mathcal {T}_j^1\). Note that \(S_j(f,{\varphi }_j)=Q_j(f,{\varphi }_j,\widetilde{\varphi }_j)\) with
One can see that condition (2.1) with \(N=\max _{|\nu |\le m} r_\nu \) is satisfied if
and condition (2.3) with \(s>0\) is fulfilled if
(i) A particular example of (2.4) is the classical Lagrange interpolation operator
where
is the Dirichlet kernel. Note that \(I_j(f)=S_j(f,{\varphi }_j)\) with \({\varphi }_j=\mathcal {D}_j\), \(m=0\), \(a_{0,j}=1\), and \(r_0=0\). In this case it is easy to see that conditions (2.1)–(2.3) are fulfilled for \(N=0\) and any \(s>0\).
(ii) As an example of quasi-interpolation operators that are generated by an average sampling instead of the exact samples of f, we consider
where
We have that \(A_j(f)=Q_j(f,{\varphi }_j,\widetilde{\varphi }_j)\) with \({\varphi }_j(x)=\mathcal {D}_j(x)\) and \(\widetilde{\varphi }_j(x)\sim \sum _{\ell \in {{\mathbb {Z}}}}\cos ^2\left( \frac{\pi \ell }{2^{j+1}}\right) e^{\textrm{i}\ell x}\) and conditions (2.1)–(2.3) are fulfilled with \(N=0\) and \(s=2\). Note that the operators of such type are used in applications, for example, in order to reduce noise contribution (see, e.g., [56]).
(iii) At the same time if in (2.6), we replace the Dirichlet kernel \(\mathcal {D}_j\) by
then condition (2.3) will hold for arbitrary \(s>0\).
(iv) We also consider the following type of operators:
We have that \(B_j(f,{\varphi }_j)=Q_j(f,{\varphi }_j,\widetilde{\varphi }_j)\) if \({\varphi }_j=\mathcal {D}_j\) and
It is not difficult to see that if \(b\ne 0\), then condition (2.1) holds with \(N=2\). At the same time if \(a\ne 0\), then compatibility condition (2.3) holds with \(s=1\). Note that in the non-periodic case operators of type (2.7) have been studied, e.g., in [4, 33].
Example 2.4
Kantorovich-type operators are defined by
where \({\sigma }\ge 1\) and as above \({\varphi }_j\in \mathcal {T}_j\). It is clear that by taking \(\widetilde{\varphi }_j(x)=2^{j+{\sigma }} \chi _{[-\pi 2^{-j-{\sigma }}, \pi 2^{-j-{\sigma }}]}(x)\), i.e., the normalized characteristic function of \([-\pi 2^{-j-{\sigma }},\pi 2^{-j-{\sigma }}]\), we have that \(K_j(f,{\varphi }_j)=Q_j(f,{\varphi }_j,\widetilde{\varphi }_j)\). Next, since
it is not difficult to see that (2.1) holds for \(N=0\). Concerning condition (2.3), we have that in this case it has the following form:
(i) If \({\varphi }_j=\mathcal {D}_j\), we have that (2.9) holds for \(s=2\).
(ii) At the same time, for
condition (2.9) holds for any \(s>0\).
Note that in recent years, the Kantorovich type operators (2.8) have been intensively studied in many works, see, e.g., [3, 9, 10, 33, 36, 38] in the non-periodic case and [28, 34, 35] in the periodic case. It is worth noting that operators of this type have several advantages over the interpolation and sampling operators. Particularly, using the averages of a function instead of the sampled values \(f(x_k^j)\) allows to deal with discontinues signals and to reduce the so-called time-jitter errors, which is an important issue in digital image processing.
3 Auxiliary Results
For \(j\in {{\mathbb {Z}}}_+\) and \(\psi \in L_1({{\mathbb {T}}})\), we define the following amalgam-type norm:
and
Lemma 3.1
Let \(1\le q\le \infty \), \(0\le {\gamma }<{\alpha }\), and let \(({\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\) and \((\widetilde{\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\) be such that \(\widetilde{\varphi }_j\in \mathcal {D}'({{\mathbb {T}}})\) and \({\varphi }_j\in \mathcal {T}_j^1\) for each \(j\in {{\mathbb {Z}}}_+\). Suppose conditions (2.1), (2.2), and (2.3) are fulfilled with \(N\ge 0\) and \(s>0\). Further suppose that
- (i):
-
\({\alpha }>N+1/q'\) if \(q\ne 1\) and \({\alpha }\ge N\) if \(q=1\) or
- (ii):
-
\(N=0\) and \(\sup _{j\in {{\mathbb {Z}}}_+}\Vert \widetilde{\varphi }_j\Vert _{\widetilde{A}_{q',j}({{\mathbb {T}}})}<\infty \).
Then, for all \(f\in A_q^{\alpha }({{\mathbb {T}}})\) and \(j\in {{\mathbb {Z}}}_+\), we have
Proof
The proof of the lemma under conditions in (i) can be found in [34, Remark 7]. In what follows, we prove (3.1) assuming that condition (ii) holds. We consider only the case \(1\le q<\infty \). The case \(q=\infty \) can be treated similarly. First we show that
Indeed, using the representation
condition (2.2), and Hölder’s inequality, we derive
which gives (3.2).
Now, we prove inequality (3.1). Let
Applying (3.2), we obtain
Next, using inequality (2.3), we get
Finally, combining (3.4) and (3.3), we arrive at (3.1). \(\square \)
Remark 3.2
With regard to (i) and (ii) in Lemma 3.1, we note that the condition (i) can be applied to the operators \(S_j(f,{\varphi }_j)\) in Example 2.3. Condition (ii), unlike condition (i), allows any parameter \({\alpha }>{\gamma }\) and is especially beneficial for the Kantorovich type operators \(K_j(f,{\varphi }_j)\), see Example 2.4. Indeed, for \(\widetilde{\varphi }_j(x)=2^{j+{\sigma }} \chi _{[-\pi 2^{-j-{\sigma }}, \pi 2^{-j-{\sigma }}]}(x)\sim \sum _{\ell \in {{\mathbb {Z}}}}\frac{\sin \pi 2^{-j-{\sigma }}\ell }{\pi 2^{-j-{\sigma }}\ell }e^{\textrm{i}\ell x}\), we have
in the case \(1<q'<\infty \). The case \(q'=\infty \) is clear.
We will need the following Bernstein inequality.
Lemma 3.3
Let \(1\le q\le \infty \), \(\min \{{\alpha },{\alpha }+{\beta }-{\gamma }\}>0\), and \({\varvec{\ell }}\in {{\mathbb {Z}}}_+^d\). Then, for any \(f\in \mathcal {T}_{\varvec{\ell }}^d\), we have
Proof
The proof is similar to the proof of Lemma 2.10 in [7]. \(\square \)
Lemma 3.4
(See [7]) Let \({\alpha }>0\), \({\beta }\in {\mathbb {R}}\), \({\varepsilon }=\min ({\alpha },{\alpha }+{\beta })>0\), and
Then the inequality
holds for all \({\varvec{k}},{\varvec{k}}'\in {{\mathbb {Z}}}_+^d\) with \({\varvec{k}}'\ge {\varvec{k}}\) componentwise.
Lemma 3.5
Let \(T<1\), \(r<t\), and \(t\ge 0\). Then, for all \(n\in {{\mathbb {N}}}\),
and
Proof
Estimate (3.5) can be found in the proof of [31, Theorem 4]. Estimate (3.6) can be proved by standard arguments using the method of Lagrange multipliers and Kuhn-Tacker conditions. \(\square \)
4 Littlewood–Paley Type Characterizations
Proposition 4.1
Let \(1\le q\le \infty \), \({\alpha }>0\), \({\beta }\in {\mathbb {R}}\), \({\alpha }+{\beta }>0\), and let \(Q=(Q_j(\cdot ,{\varphi }_j,\widetilde{\varphi }_j))_{j\in {{\mathbb {Z}}}_+}\), where \(({\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\) and \((\widetilde{\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\) be such that \(\widetilde{\varphi }_j\in \mathcal {D}'({{\mathbb {T}}})\) and \({\varphi }_j\in \mathcal {T}_j^1\) for each \(j\in {{\mathbb {Z}}}_+\). Suppose conditions (2.1), (2.2), and (2.3) are satisfied with the parameters \(N\ge 0\) and \(s>\max ({\alpha }+{\beta },{\alpha })\). Assume also that
- (i):
-
\(\min ({\alpha }+{\beta },{\alpha })>N+1/q'\) or
- (ii):
-
\(N=0\) and \(\sup _{j\in {{\mathbb {Z}}}_+}\Vert \widetilde{\varphi }_j\Vert _{\widetilde{A}_{q',j}({{\mathbb {T}}})}<\infty \).
Then every function \(f\in A_q^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) can be represented by the series
which converges unconditionally in \(A_q^{\widetilde{\alpha },{\beta }}({{\mathbb {T}}}^d)\) with \(0\le \widetilde{\alpha }<{\alpha }\) and satisfies
Proof
Step 1 First we prove the proposition assuming that the condition in (i) holds. Let \(f\in A_q^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) and \({\varvec{j}}\in {{\mathbb {Z}}}_+^d\). We have
where we set \({\delta }_{{\varvec{j}}+{\varvec{\ell }}}(f)=0\) for \({\varvec{j}}+{\varvec{\ell }}\in {{\mathbb {Z}}}^d \setminus {{\mathbb {Z}}}_+^d\). In light of (4.3), we get
and, therefore,
In what follows, for simplicity we consider only the case \(q<\infty \). The case \(q=\infty \) can be treated similarly. Multiplying by \(2^{{\alpha }|{\varvec{j}}|_1+{\beta }|{\varvec{j}}|_\infty }\) and taking \(\ell _q\)-norm on both sides of (4.4), we obtain
where \({\varvec{\ell }}_k^d=(\ell _k,\dots ,\ell _d)\), \(k=2,\dots ,d\).
Consider the sum \(S_1\). Denoting
where \(Q_{j_i}^i\) is the univariate operator \(Q_{j_i}(\cdot ,{\varphi }_{j_i},\widetilde{\varphi }_{j_i})\) acting on functions in the variable \(x_i\), we obtain
where I is the identity operator. Taking into account that \( Q_j(t,{\varphi },\widetilde{\varphi })=\widetilde{\varphi }_j*{\varphi }_j*t \) for any trigonometric polynomial \(t\in \mathcal {T}_{j-1}^1\) and using condition (2.3) and Bernstein’s inequality, we derive that
where in the last inequality we use the estimates
and
Next, combining (4.6) and (4.7) and using the fact that \(\sum _{\ell _{1}<-1} 2^{(s-\max ({\alpha }+{\beta },{\alpha }))\ell _1} <\infty \), we obtain
Now, we consider the sum \(S_2\). Similar to (4.6), we have
Choosing \(\zeta \) such that \(N+1/q'<\zeta <\min ({\alpha },{\alpha }+{\beta })\) and applying Lemma 3.1(i), Bernstein’s inequality, and Lemma 3.4, we obtain for \(\ell _1\ge 0\) that
A similar estimate clearly holds for \(\ell _1=-1\). Thus, combining (4.9) and (4.10) and taking into account that \(\sum _{\ell _1\ge -1} 2^{-(\min ({\alpha },{\alpha }+{\beta })-\zeta )\ell _1}<\infty \), we get
In the next step, collecting (4.5), (4.8), and (4.11) implies
Then, repeating the above procedure for the parameters \(\ell _2,\dots ,\ell _d\), we prove (4.2) by Lemma 2.2.
Step 2 Let us prove representation (4.1). Applying Lemma 3.3 (here without loss of generality, we can assume that \(\min (\widetilde{\alpha }+{\beta },\widetilde{\alpha })>0\)), Hölder’s inequality, and (4.2), we obtain
Therefore, \(\sum _{{\varvec{k}}\in {{\mathbb {Z}}}_+^d}\eta _{\varvec{k}}^Q(f)\) converges unconditionally in \(A_q^{\widetilde{\alpha },{\beta }}({{\mathbb {T}}}^d)\).
Now we show that for any trigonometric polynomial g,
It is clear that it suffices to verify (4.13) for \(t_{\varvec{j}}({\varvec{x}})=e^{\textrm{i}({\varvec{j}},{\varvec{x}})}\) with arbitrary \({\varvec{j}}\in {{\mathbb {Z}}}^d\). By the triangle inequality, we have
We obviously have that
Since
we obtain
Thus, for any \({\varvec{j}}\in (-2^{m-1},2^{m-1})^d\cap {{\mathbb {Z}}}^d\), we have
Using this and conditions (2.1), (2.2), (2.3), we find
Therefore, combining (4.14), (4.15), and (4.16), we arrive at (4.13).
The rest of the proof is quite standard. Denote \(F:=\sum _{{\varvec{k}}\in {{\mathbb {Z}}}_+^d}\eta _{\varvec{k}}^Q(f)\). Using (4.13), we have for every trigonometric polynomial g that
with convergence in \(A_q^{\widetilde{\alpha },{\beta }}({{\mathbb {T}}}^d)\). Hence, by (4.12), we derive
Choosing g close enough to f yields \(\Vert F-f\Vert _{A_q^{\widetilde{\alpha },{\beta }}({{\mathbb {T}}}^d)}<{\varepsilon }\) for all \({\varepsilon }>0\) and hence \(\Vert F-f\Vert _{A_q^{\widetilde{\alpha },{\beta }}({{\mathbb {T}}}^d)}=0\), which implies (4.1).
By the same scheme, using Lemma 3.1 (ii), the proof of the proposition under condition (ii) also follows. \(\square \)
We will also need the following modification of inequality (4.2).
Lemma 4.2
Let \(f\in A_q({{\mathbb {T}}}^d)\), \(1\le q<p\le \infty \), and \(1\le {\theta }\le \infty \). Under conditions of Proposition 4.1, there exists a constant \(C=C({\alpha },{\beta },q,{\theta },d)>0\) such that
whenever the sum in the right-hand side is finite.
Proof
First, repeating the same procedure as in the proof of Step 1 of Proposition 4.1, we obtain
Then, applying the inequality
which easily follows from Hölder’s inequality and the fact that \({\text {spec}}{\delta }_{\varvec{j}}(f)\subset P_{j_1}\times \dots \times P_{j_d}\), \(P_j=\{\ell \in {{\mathbb {Z}}}\,:\, 2^{j-1}\le |\ell |<2^j\}\), we arrive at (4.17). \(\square \)
A reverse statement to Proposition 4.1 is written as follows.
Proposition 4.3
Let \({\alpha }>0\), \({\beta }\in {\mathbb {R}}\), \({\alpha }+{\beta }>0\), \(1\le q\le \infty \), and let \((f_{\varvec{j}})_{{\varvec{j}}\in {{\mathbb {Z}}}_+^d}\) be such that \(f_{\varvec{j}}\in \mathcal {T}_{\varvec{j}}^d\) and
Suppose that the series \(\sum _{{\varvec{j}}\in {{\mathbb {Z}}}_+^d}f_{\varvec{j}}\) converges to a function f in \(A_q({{\mathbb {T}}}^d)\). Then \(f\in A_q^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) and moreover, there is a constant \(C=C({\alpha },{\beta },q,d)\) such that
Proof
The proposition can be proved repeating step by step the proof of Proposition 3.4 in [7]. For completeness we present a detailed proof.
For \({\varvec{\ell }}\in {{\mathbb {Z}}}_+^d\), we write f as the series
with \(f_{{\varvec{\ell }}+{\varvec{j}}}:=0\) for \({\varvec{j}}+{\varvec{\ell }}\in {{\mathbb {Z}}}^d\setminus {{\mathbb {Z}}}_+^d\). Using the triangle inequality and taking into account that \({\delta }_{\varvec{\ell }}(f_{{\varvec{\ell }}+{\varvec{j}}})=0\) for \({\varvec{j}}\not \in {{\mathbb {Z}}}_+^d\), we obtain
This inequality together with Lemma 3.4 yields
Then, by Minkowski’s inequality, we obtain
Thus, Lemma 2.2 concludes the proof. \(\square \)
Propositions 4.1 and 4.3 suggest the following useful necessary and sufficient conditions for \(f\in A_q^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) to be represented as \(f=\sum _{{\varvec{j}}\in {{\mathbb {Z}}}_+^d}\eta _j^Q(f)\). This generalizes Theorem 3.6 in [7].
Theorem 4.4
Let \(1\le q\le \infty \), \({\alpha }>0\), \({\beta }\in {\mathbb {R}}\), \({\alpha }+{\beta }>0\), and let \(Q=(Q_j(\cdot ,{\varphi }_j,\widetilde{\varphi }_j))_{j\in {{\mathbb {Z}}}_+}\), where \(({\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\) and \((\widetilde{\varphi }_j)_{j\in {{\mathbb {Z}}}_+}\) be such that \(\widetilde{\varphi }_j\in \mathcal {D}'({{\mathbb {T}}})\) and \({\varphi }_j\in \mathcal {T}_j^1\) for each \(j\in {{\mathbb {Z}}}_+\). Suppose conditions (2.1), (2.2), and (2.3) are satisfied with parameters \(N\ge 0\) and \(s>\max ({\alpha }+{\beta },{\alpha })\). Assume also that
- (i):
-
\(\min ({\alpha }+{\beta },{\alpha })>N+1/q'\) or
- (ii):
-
\(N=0\) and \(\sup _{j\in {{\mathbb {Z}}}_+}\Vert \widetilde{\varphi }_j\Vert _{\widetilde{A}_{q',j}({{\mathbb {T}}})}<\infty \).
Then a function f belongs to \(A_q^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) if and only if it can be represented by the series (4.1) converging unconditionally in \(A_q^{\widetilde{\alpha },{\beta }}({{\mathbb {T}}}^d)\) with \(\widetilde{\alpha }<{\alpha }\) and satisfying \(\sum _{{\varvec{j}}\in {{\mathbb {Z}}}_+^d} 2^{q({\alpha }|{\varvec{j}}|_1+{\beta }|{\varvec{j}}|_\infty )}\Vert \eta _{\varvec{j}}^Q(f)\Vert _{A_q({{\mathbb {T}}}^d)}^q<\infty \). Moreover, the norm \(\Vert f\Vert _{A_q^{{\alpha },{\beta }}({{\mathbb {T}}}^d)}\) is equivalent to the norm
5 Error Estimates
In this section, we obtain estimates for the error of approximation by quasi-interpolation operators
where
In what follows, we distinguish between approximation of a function \(f\in A_p^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) in the isotropic space \(A_q^{\gamma }({{\mathbb {T}}}^d)\) (Theorem 5.1) and in the mixed space \(A_{q,{\text {mix}}}^{\gamma }({{\mathbb {T}}}^d)\) (Theorem 5.3) since we use slightly different ingredients in the corresponding proofs.
Recall that
5.1 Error Estimates in \(A_{q}^{\gamma }({{\mathbb {T}}}^d)\)
Theorem 5.1
Let \(1\le p,q\le \infty \), \({\alpha }>{\sigma }_{p,q}\), \({\beta }\in {\mathbb {R}}\), \({\gamma }\ge 0\), \({\gamma }-{\beta }<{\alpha }-{\sigma }_{p,q}\), and let \(Q=(Q_j(\cdot ,{\varphi }_j,\widetilde{\varphi }_j))_{j\in {{\mathbb {Z}}}_+}\), where \(\widetilde{\varphi }_j\in \mathcal {D}'({{\mathbb {T}}})\) and \({\varphi }_j\in \mathcal {T}_j^1\) for each \(j\in {{\mathbb {Z}}}_+\). Suppose conditions (2.1), (2.2), and (2.3) are satisfied with parameters \(N\ge 0\) and \(s>\max ({\alpha }+{\beta },{\alpha })\). Assume also that
- (i):
-
\(\min ({\alpha }+{\beta },{\alpha })>N+1/p'\) or
- (ii):
-
\(N=0\) and \(\sup _{j\in {{\mathbb {Z}}}_+}\Vert \widetilde{\varphi }_j\Vert _{\widetilde{A}_{q',j}({{\mathbb {T}}})}<\infty \).
Then, for all \(f\in A_p^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) and \(n\in {{\mathbb {N}}}\), we have
where
and the constant C does not depend on f and n.
Remark 5.2
(i) In the case \(p=q=2\) and \(Q=(I_j)_{j\in {{\mathbb {Z}}}_+}\), where \(I_j\) is the Lagrange interpolation operator defined in (1.3), Theorem 5.1 was proved in [25], see also [7] and [24]. For similar results in the case \(p=q=2\), \({\gamma }=T=0\), and \(Q=(K_j)_{j\in {{\mathbb {Z}}}_+}\), where \(K_j\) is defined in (2.8), see [32]. (ii) Under conditions of Theorem 5.1 with \(1\le q\le 2\) and \({\gamma }=0\), by the Hausdorff-Young inequality, estimate (5.1) implies that
We can further extend this result considering a more general Pitt’s inequality [23, 52, 53]
where \(\Vert f\Vert _{L_{\xi }^\eta ({{\mathbb {T}}}^d)}=(\int _{{{\mathbb {T}}}^d}|f(x)|^\xi |x|^\eta dx)^{1/\xi }\) under the suitable conditions on \(\xi \) and \(\eta \). In particular, using [22, (5.4)] we have
for \(\xi \ge 2\), \(\max (q,q')\le \xi \), and \(\gamma =d (1-\frac{1}{\xi }-\frac{1}{q})\). Taking \(\xi =q'\) we see that (5.4) coincides with (5.2).
(iii) It is clear that the case \({\gamma }-{\beta }={\alpha }-{\sigma }_{p,q}\) is degenerate since then we have \(\Omega (n)=1\) and hence the right-hand side of (5.1) does not tend to zero.
Proof of Theorem 5.1
First, we consider the case \(1<p\le q\le \infty \).
Using the estimate \(\Vert \cdot \Vert _{\ell _q}\le \Vert \cdot \Vert _{\ell _p}\), Proposition 4.1, Lemma 3.3, and Hölder’s inequality, we obtain
Thus, Proposition 4.1 implies
Next, combining (5.6) and (3.5), we derive (5.1) in the case \(p>1\). The case \(p=1\) is treated similarly using (3.6).
Now, we consider the case \(1\le q<p\le \infty \). Since \(1/p'>1/q'\), we can apply the intermediate estimate in (5.5) with \(p=q\) given by
Then, using Hölder’s inequality, Lemmas 3.5, and 4.2 (note that condition (i) implies that \(\min ({\alpha }-1/q+1/p+{\beta },{\alpha }-1/q+1/p)>N+1/q'\)), we get
where in the last estimate we have taken into account Lemma 2.2. \(\square \)
5.2 Error Estimates in \(A_{q,{\text {mix}}}^{\gamma }({{\mathbb {T}}}^d)\)
Theorem 5.3
Let \(1\le p,q\le \infty \), \({\beta }\in {\mathbb {R}}\), \({\gamma }> 0\), \({\gamma }-{\beta }+{\sigma }_{p,q}<{\alpha }\), \({\gamma }+{\sigma }_{p,q}\le {\alpha }\), and let \(Q=(Q_j(\cdot ,{\varphi }_j,\widetilde{\varphi }_j))_{j\in {{\mathbb {Z}}}_+}\), where \(\widetilde{\varphi }_j\in \mathcal {D}'({{\mathbb {T}}})\) and \({\varphi }_j\in \mathcal {T}_j^1\) for each \(j\in {{\mathbb {Z}}}_+\). Suppose conditions (2.1), (2.2), and (2.3) are satisfied with parameters \(N\ge 0\) and \(s>\max ({\alpha }+{\beta },{\alpha })\). Assume also that
-
(i)
\(\min ({\alpha }+{\beta },{\alpha })>N+1/p'\) or
-
(ii)
\(N=0\) and \(\sup _{j\in {{\mathbb {Z}}}_+}\Vert \widetilde{\varphi }_j\Vert _{\widetilde{A}_{q',j}({{\mathbb {T}}})}<\infty \).
Then, for all \(f\in A_p^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) and \(n\in {{\mathbb {N}}}\), we have
where
where the constant C does not depend on f and n.
Remark 5.4
(i) This result generalizes Theorem 5.1 in [7], which corresponds to the case \(p=q=2\), \(T={\beta }=0\), and \(Q=(I_j)_{j\in {{\mathbb {Z}}}_+}\), where \(I_j\) is defined in (1.3).
(ii) Using the inequality (see, e.g., [7, Lemma 5.7])
we easily obtain that under conditions of Theorem 5.3 with \(q=2\) and \({\gamma }=\frac{1}{2}-\frac{1}{r}\), inequality (5.8) implies the following error estimate:
where
and
Comparing inequalities (5.2) and (5.9) with \(r=q'\) and \(1<q<2\), we see that (5.9) provides better approximation order in the case \(1<q<2\le p\le \infty \).
Proof of Theorem 5.3
First, we consider the case \(1\le p\le q\le \infty \). Using Proposition 4.3 with
and taking into account that \(A_{q,{\text {mix}}}^{\gamma }({{\mathbb {T}}}^d)=A_q^{{\gamma },0}({{\mathbb {T}}}^d)\) and \(f-P_{n,T}^Q f=\sum _{{\varvec{j}}\in {{\mathbb {Z}}}_+^d} f_{\varvec{j}}\), we obtain
where the last inequality follows from Proposition 4.1.
Second, let \(1\le q<p\le \infty \). Similarly to the proof of (5.7), using estimates (5.10) with \(p=q\), Hölder’s inequality, and Lemmas 3.5 and 4.2, we have
which proves (5.8) for \(1\le q<p\le \infty \) by Lemma 2.2. \(\square \)
It is not difficult to see that Theorems 5.1 and 5.3 can also be established for more general operators
where \(\Gamma \) is some arbitrary set of indices in \({{\mathbb {Z}}}_+^d\). More precisely, we obtain the following remarks.
Remark 5.5
Suppose that conditions of Theorem 5.1 hold with \(\Gamma \) instead of \({\Delta }(n,T)\). Then, for all \(f\in A_p^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\), we have
where the constant C does not depend on f and \(\Gamma \).
Remark 5.6
Suppose that conditions of Theorem 5.3 hold with \(\Gamma \) instead of \({\Delta }(n,T)\). Then, for all \(f\in A_p^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\), we have
where
and the constant C does not depend on f and \(\Gamma \).
5.3 Sharpness
Here, we show that inequalities (5.1) and (5.8) given in Theorems 5.1 and 5.3 are sharp for specific cases of parameters. Those cases are of special interest since they provide the best order of approximation and simultaneously are optimal with respect to the computation time (cf. (1.2)).
Theorem 5.7
\(\mathrm{(i)}\) Under conditions of Theorem 5.1, if \(0<T<\frac{{\gamma }-{\beta }}{{\alpha }}\), we have
for sufficiently large n.
\(\mathrm{(ii)}\) Under conditions of Theorem 5.3, if \(0<T<\frac{-{\beta }}{{\alpha }-{\gamma }}\), we have
for sufficiently large n.
Proof
In view of Theorems 5.1 and 5.3, it is enough to consider only estimates from below. We prove only (i). The assertion (ii) can be treated similarly. We follow the idea of the proof of [7, Theorem 6.7] (see also [18]) taking into account the following lemma on lower estimates for linear widths (see, e.g., [Theorem 1][50, 51]).
Lemma 5.8
Let \(L_{m+1}\) be \((m+1)\)-dimensional subspace in a Banach space X, and let \(B_{m+1}(r){:}{=}\{f\in L_{m+1}\,:\,\Vert f\Vert _X\le r\}\). Then
where infimum is taken over all continuous linear operators \(A_m\) in X with rank at most m.
We use this lemma with \(X=A_p^{\gamma }({{\mathbb {T}}}^d)\) and \(L_{2^n+1}=\textrm{span}\,\{e^{\textrm{i}kx_1}\,:\,k=0,\dots ,2^n\}\). Let also \(n_0\in {{\mathbb {N}}}\) be such that \(\textrm{rank}\, P_{n-n_0,T}^Q \le 2^n\) (we can always find such \(n_0\) in view of (1.2)). For any \(f\in L_{2^n+1}\), we have
Thus, by choosing \(r=2^{-({\alpha }+{\beta }-{\gamma })n}\), we get that \(B_{2^n+1}(r)\subset UA_p^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\). Using this embedding and Lemma 5.8, we obtain
which implies (5.11). \(\square \)
Remark 5.9
Note that the sharpness of Theorem 5.1 under certain natural restrictions on distributions \(\widetilde{\varphi }_j\) in the case \(T=0\), \({\beta }={\gamma }=0\), and \(p=2\le q\le \infty \) follows from the proof of [32, Theorem 4]. For particular cases of the parameters (mainly for the cases \(T=0\), \(p,q\in \{1,2\}\), \(p\le q\), \({\gamma }\in \{0,1\}\), \({\beta }=0\)), the sharpness of Theorem 5.1 can be also established using general estimates of linear widths (see, e.g., [7, 42]).
6 Effective Error Estimates
6.1 Energy-Norm Based Sparse Grids
Along with the general operators \(P_{n,T}^Q\), in [7] and [15] the authors studied quasi-interpolation operators
with specific choice of the family Q, where
For the reader’s convenience, we reformulate Theorem 5.1 for \(P_{{\Delta }(\xi )}^Q\) noting that \({\Delta }(\xi )\) corresponds to the set \({\Delta }(n,T)\) with \(T=\frac{{\gamma }-{\beta }-{\varepsilon }}{{\alpha }-{\sigma }_{p,q}-{\varepsilon }}\) and \(n=\frac{\xi }{{\alpha }-{\sigma }_{p,q}-{\gamma }+{\beta }}\).
Corollary 6.1
Under conditions of Theorem 5.1, if \(0<{\varepsilon }<{\gamma }-{\beta }<{\alpha }-{\sigma }_{p,q}\), then, for all \(f\in A_p^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) and \(\xi \in {{\mathbb {N}}}\), we have
where the constant C does not depend on f and \(\xi \). In particular, in the case \(p=q=2\) and \({\gamma }=0\), for all \(f\in H^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) and \(\xi \in {{\mathbb {N}}}\), we have
Proof
The proof directly follows from Theorem 5.1 by taking \(T=\frac{{\gamma }-{\beta }-{\varepsilon }}{{\alpha }-{\sigma }_{p,q}-{\varepsilon }}\). \(\square \)
Remark 6.2
(i) It follows from the proof of Theorem 5.1 that in the case \(p=1\), the assertion of Corollary 6.1 remains true for \({\varepsilon }=0\).
(ii) Corollary 6.1 extends Theorem 4.1 in [7], cf. [7, Remark 4.4], which corresponds to the case \(p=q=2\) and \(Q=(I_j)_{j\in {{\mathbb {Z}}}_+}\), where \(I_j\) is defined in (2.5).
6.2 Smolyak Grids
In some special cases of parameters in Theorems 5.1 and 5.3, the Smolyak algorithm, i.e., the operators \(P_{n,T}^Q\) with \(T=0\), provides more effective error estimates with respect to the number of frequencies than the operators \(P_{n,T}^Q\), \(0<T<1\), which correspond to the energy norm based grids. In particular, applying Theorem 5.1 with \(T=0\) and \({\beta }={\gamma }\), we obtain the following corollary about approximation in the space \(A_q^{\beta }({{\mathbb {T}}}^d)\).
Corollary 6.3
Under conditions of Theorem 5.1, for all \(f\in A_{p}^{{\alpha },{\beta }}({{\mathbb {T}}}^d)\) and \(n\in {{\mathbb {N}}}\), we have
where the constant C does not depend on f and n. In particular, if \(p=q=2\), we have
Remark 6.4
(i) By the same arguments as in Remark 5.2(ii), we have that inequality (6.1) with \(1\le q\le 2\) and \({\beta }=0\) implies that
In particular, if \(q=1\) and \(Q=(I_j)_{j\in {{\mathbb {Z}}}_+}\), the above inequality generalizes [7, Theorem 5.6] (the case \(p=2\)) and the main results of [27] (the case \(p=\infty \), which corresponds to the Korobov space).
(ii) It follows from (6.1) and [42, Corollary 4.3 and (1.1)] that in the cases \(p=q=1\) and \(p=1, q=2\) the Smolyak algorithm \(P_{n,0}^Q\) provides optimal in order approximation among all continuous linear operator of finite rank.
In a similar way, applying Theorem 5.3 with \(T={\beta }=0\), we get the following result concerning approximation by \(P_{n,0}^Q f\) in the space \(A_{q, {\text {mix}}}^{\gamma }({{\mathbb {T}}}^d)\).
Corollary 6.5
Under the conditions of Theorem 5.3, for all \(f\in A_{p,{\text {mix}}}^{{\alpha }}({{\mathbb {T}}}^d)\) and \(n\in {{\mathbb {N}}}\), we have
where the constant C does not depend on f and n. In particular, if \(p=q=2\), we have
References
Andrianov, A.V., Temlyakov, V.N.: On two methods of generalization of properties of univariate function systems to their tensor product. Trudy MIAN 219, 32–43 (1997)
Andrianov, A.V., Temlyakov, V.N.: Proc. Steklov Inst. Math. 219, 25–35 (1997)
Bardaro, C., Butzer, P.L., Stens, R.L., Vinti, G.: Kantorovich-type generalized sampling series in the setting of Orlicz spaces. Sampl. Theor. Signal Image Process. 6, 29–52 (2007)
de Boor, C., Höllig, K., Riemenschneider, S.: Box Splines. In: Applied Mathematical Sciences, Vol. 98. Springer-Verlag, New York (1993)
Bugrov, Ya.. S.: Summability of Fourier transforms and absolute convergence of multiple Fourier series. Trudy Mat. Inst. Steklov. 187, 22–30 (1989)
Bungartz, H.-J., Griebel, M.: Sparse grids. Acta Numer. 13, 147–269 (2004)
Byrenheid, G., Dũng, D., Sickel, W., Ullrich, T.: Sampling on energy-norm based sparse grids for the optimal recovery of Sobolev type functions in \(H^\gamma \). J. Approx. Theory 207, 207–231 (2016)
Byrenheid, G., Ullrich, T.: Optimal sampling recovery of mixed order Sobolev embeddings via discrete Littlewood–Paley type characterizations. Anal. Math. 43(2), 133–191 (2017)
Costarelli, D., Vinti, G.: Inverse results of approximation and the saturation order for the sampling Kantorovich series. J. Approx. Theory 242, 64–82 (2019)
Costarelli, D., Seracini, M., Vinti, G.: A comparison between the sampling Kantorovich algorithm for digital image processing with some interpolation and quasi-interpolation methods. Appl. Math. Comp. 374, 125046 (2020)
Delvos, F.-J., Schempp, W.: Boolean Methods in Interpolation and Approximation. Longman Scientific & Technical, Harlow (1989)
Dũng, D.: Approximation of multivariate functions by means of harmonic analysis, Hab. Dissertation, Moscow, MGU (1985)
Dũng, D.: Approximation of functions of several variables on a torus by trigonometric polynomials. Mat. Sb. 131, 251–271 (1986)
Dũng, D.: \(B\)-spline quasi-interpolant representations and sampling recovery of functions with mixed smoothness. J. Complex. 27, 541–567 (2011)
Dũng, D.: Sampling and cubature on sparse grids based on a B-spline quasi-interpolation. Found. Comput. Math. 16, 1193–1240 (2016)
Dũng, D.: \(B\)-spline quasi-interpolation sampling representation and sampling recovery in Sobolev spaces of mixed smoothness. Acta Math. Vietnamica 43, 83–110 (2018)
Dũng, D., Temlyakov, V.N., Ullrich, T.: Hyperbolic Cross Approximation. Advanced Courses in Mathematics, CRM Barcelona. Birkhäuser/Springer, Basel (2018)
Dũng, D., Ullrich, T.: \(N\)-widths and \(\varepsilon \)-dimensions for high-dimensional approximations. Found. Comput. Math. 13, 965–1003 (2013)
Dyachenko, M.I.: Some problems in the theory of multiple trigonometric series. Uspekhi Mat. Nauk 47(5), 97–162 (1992). ((in Russian))
Dyachenko, M.I.: Russ. Math. Surv. 47(5), 103–171 (1992)
Feichtinger, H.G., Kaiblinger, N.: Quasi-interpolation in the Fourier algebra. J. Approx. Theory 144(1), 103–118 (2007)
Gorbachev, D., Tikhonov, S.: Moduli of smoothness and growth properties of Fourier transforms: two-sided estimates. J. Approx. Theory 164(9), 1283–1312 (2012)
Gorbachev, D., Liflyand, E., Tikhonov, S.: Weighted Fourier Inequalities: Boas conjecture in \({\mathbb{R} }^n\). J. d’Analyse Math. 114, 99–120 (2011)
Griebel, M., Hamaekers, J.: Fast Discrete Fourier Transform on Generalized Sparse Grids, Sparse grids and Applications. In: Lecture Notes in Computational Science and Engineering, pp. 75–108. Springer, Berlin (2014)
Griebel, M., Hamaekers, J.: Generalized sparse grid interpolation based on the fast discrete Fourier transform. INS Preprint No. 1902 (2020)
Griebel, M., Knapek, S.: Optimized tensor-product approximation spaces. Constr. Approx. 16(4), 525–540 (2000)
Hallatschek, K.: Fourier-transform on sparse grids with hierarchical bases. Numer. Math. 63(1), 83–97 (1992)
Jacob, M., Blu, T., Unser, M.: Sampling of periodic signals: a quantitative error analysis. IEEE Trans. Signal Process. 50(5), 1153–1159 (2002)
Jetter, K., Zhou, D.X.: Order of linear approximation from shift invariant spaces. Constr. Approx. 11(4), 423–438 (1995)
Jia, R.-Q.: Approximation by quasi-projection operators in Besov spaces. J. Approx. Theory 162(1), 186–200 (2010)
Knapek, S.: Approximation und Kompression mit Tensorprodukt-Multiskalenräumen. Dissertation, University of Bonn (2000)
Kolomoitsev, Yu.: Approximation by quasi-interpolation operators and Smolyak’s algorithm. J. Complexity 69, Paper No. 101601 (2022)
Kolomoitsev, Yu., Krivoshein, A., Skopina, M.: Differential and falsified sampling expansions. J. Fourier Anal. Appl. 24(5), 1276–1305 (2018)
Kolomoitsev, Yu., Krivoshein, A., Skopina, M.: Approximation by periodic multivariate quasi-projection operators. J. Math. Anal. Appl. 489(2), 124192 (2020)
Kolomoitsev, Yu., Prestin, J.: Approximation properties of periodic multivariate quasi-interpolation operators. J. Approx. Theory 270, 105631 (2021)
Kolomoitsev, Yu., Skopina, M.: Approximation by multivariate Kantorovich–Kotelnikov operators. J. Math. Anal. Appl. 456(1), 195–213 (2017)
Kolomoitsev, Yu., Skopina, M.: Approximation by sampling-type operators in \(L_p\)-spaces. Math. Methods Appl. Sciences 43(16), 9358–9374 (2020)
Kolomoitsev, Yu., Skopina, M.: Quasi-projection operators in the weighted \(L_p\) spaces. Appl. Comput. Harmon. Anal. 52, 165–197 (2021)
Kyriazis, G.C.: Approximation of distribution spaces by means of kernel operators. J. Fourier Anal. Appl. 3, 261–286 (1996)
Liflyand, E., Samko, S., Trigub, R.: The Wiener algebra of absolutely convergent Fourier integrals: an overview. Anal. Math. Phys. 2, 1–68 (2012)
Musielak, J.: On absolute convergence of multiple Fourier series. Ann. Pol. Math. 5, 107–120 (1958)
Nguyen, V.D., Nguyen, V.K., Sickel, W.: s-numbers of embeddings of weighted Wiener algebras. J. Approx. Theory 279, Paper No. 105745 (2022)
Nursultanov, E., Tikhonov, S.: Wiener–Beurling spaces and their properties. Bull. Sci. Math. 159, 102825 (2020)
Schmeisser, H.-J., Triebel, H.: Topics in Fourier Analysis and Function Spaces. Mathematik und ihre Anwendungen in Physik und Technik [Mathematics and its Applications in Physics and Technology], vol. 42. Akademische Verlagsgesellschaft Geest & Portig K.-G, Leipzig (1987)
Sickel, W., Ullrich, T.: The Smolyak algorithm, sampling on sparse grids and function spaces of dominating mixed smoothness. East J. Approx. 13(4), 387–425 (2007)
Smolyak, S.A.: Quadrature and interpolation formulas for tensor products of certain classes of functions. Dokl. Akad. Nauk SSSR 148(5), 1042–1045 (1963)
Sprengel, F.: A class of periodic function spaces and interpolation on sparse grids. Numer. Funct. Anal. Optim. 21(1–2), 273–293 (2000)
Temlyakov, V.N.: Approximation of functions with bounded mixed derivative. Trudy MIAN 178, 1–112 (1986)
Temlyakov, V. N.: English transl. Proc. Steklov Inst. Math. 1 (1989)
Tikhomirov, V.M.: Diameters of sets in function spaces and the theory of best approximations. Uspekhi Mat. Nauk 3(93), 81–120 (1960)
Tikhomirov, V.M.: Russ. Math. Surv. 15(3), 75–111 (1960)
Tikhonov, S.: Weighted Fourier inequalities and boundedness of variations. Proc. Steklov Inst. Math. 312, 282–300 (2021)
Tikhonov, S.: Trudy MIAN 312, 294–312 (2021)
Ullrich, T.: Smolyak’s algorithm, sampling on sparse grids and Sobolev spaces of dominating mixed smoothness. East J. Approx. 14(1), 1–38 (2008)
Wermer, J.: On a class of normed rings. Ark. Mat. 2(6), 537–551 (1954)
Zhang, Q., Wang, L., Sun, W.: Signal denoising with average sampling. Digit. Signal Process. 22(2), 226–232 (2012)
Acknowledgements
We would like to thank Kristina Oganesyan for useful remarks.
Funding
Open Access funding enabled and organized by Projekt DEAL.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by E.Liflyand.
Dedicated to the 80th anniversary of Professor Stefan Samko.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supported by the German Research Foundation, project KO 5804/1-2
Support by the German Research Foundation in the framework of the RTG 2088
Supported by PID2020-114948GB-I00, 2017 SGR 358, AP14870758 and the CERCA Programme of the Generalitat de Catalunya. Also, supported by the Spanish State Research Agency, through the Severo Ochoa and María de Maeztu Program for Centers and Units of Excellence in R &D (CEX2020-001084-M). The author thanks CERCA Programme/Generalitat de Catalunya for institutional support.
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
Kolomoitsev, Y., Lomako, T. & Tikhonov, S. Sparse Grid Approximation in Weighted Wiener Spaces. J Fourier Anal Appl 29, 19 (2023). https://doi.org/10.1007/s00041-023-09994-2
Received:
Revised:
Accepted:
Published:
DOI: https://doi.org/10.1007/s00041-023-09994-2
Keywords
- Sparse grid
- Weighted Wiener spaces
- Quasi-interpolation operators
- Kantorovich operators
- Smolyak algorithm
- Littlewood–Paley-type characterizations