Estimates for the approximation characteristics of the Nikol’skii-Besov classes of functions with mixed smoothness in the space B_{q,1}

To get the upper estimates in (?), it is sufficient, due to (?), to respectively estimate the quantity \mathscr{E}_{Q^{{\boldsymbol{\gamma}}}_n} ( B^{\boldsymbol{r}}_{p,\theta})_{B_{p,1}}.

Let f be a function from the class B^{\boldsymbol{r}}_{p,\theta}. Then, from the equivalent representation (?) of the norm in the space B_{q,1}, we have

To further estimate the quantity J_1, let us distinguish three cases depending on the value of the parameter \theta.

Case \theta = 1. Then, taking into account Theorem A and the fact that ({\boldsymbol{s}},{\boldsymbol{1}})\leq ({\boldsymbol{s}},{\boldsymbol{\gamma}}), we can write

Case 1 < \theta < \infty. Using the Hölder’s inequality with the power \theta to J_1, we get

where

\widetilde{{\boldsymbol{\gamma}}} \in \mathbb{R}^d is such that \widetilde{\gamma}_j = (r_j - 1/p + 1/q)/(r_1 - 1/p + 1/q), j=1, \dots, d.

One can easily check that \widetilde{\gamma}_j = {\gamma}_j=1 for j =1, \dots, \nu and 1 < {\gamma}_j < \widetilde{{\gamma}}_j for j = \nu +1, \dots, d. From this, using Lemma A for the last sum in (?), we obtain

Case \theta = \infty. Taking into account, that for f \in B^{\boldsymbol{r}}_{p,\infty}, 1 < p < \infty, it holds (see (?))

using Lemma A, we have that

Combining (?) – (?), we obtain the upper estimate for the quantity \mathscr{E}_{Q^{{\boldsymbol{\gamma}}}_n} ( B^{\boldsymbol{r}}_{p,\theta})_{B_{q,1}}, and hence for E_{Q^{{\boldsymbol{\gamma}}}_n} ( B^{\boldsymbol{r}}_{p,\theta})_{B_{q,1}}.

It is sufficient to get the lower estimate in (?) for the case \nu = d. Below we construct a function that realizes the obtained upper order estimate.

Let

where

Let us first show that for certain choice of the constant C_4, g \in B^{\boldsymbol{r}}_{p,\theta}, {\boldsymbol{r}}=(r_1, \dots, r_1)\in \mathbb{R}^d, r_1>0, 1<p<\infty, 1 \leq \theta \leq \infty.

Let first 1\leq \theta <\infty. Then we can write

Then we use the known relation

(see, e.g., [Temlyakov1993, Ch. 1, Paragraph 1]), and get

Substituting (?) into (?), we continue the estimation of

\|g\|_{B^{\boldsymbol{r}}_{p,\theta}} as follows

Let further \theta = \infty. Then the function g takes the form

and therefore, in view of (?) and (?), we get

We make a conclusion that for an appropriate choice of the constant C_4>0, the function g \in B^{\boldsymbol{r}}_{p,\theta}.

Then, taking into account that S_{Q^{\boldsymbol{1}}_n}(g) = 0 and the relation (?), we can write

Theorem 3.1 is proved

We will argue similarly to that in Theorem 3.1. For f\in B^{\boldsymbol{r}}_{p,\theta}, we obtain

Further, as above, we distinguish three cases.

For the case \theta = 1, we note that 2^{-({\boldsymbol{s}},{\boldsymbol{\gamma}})r_1}\leq 2^{-({\boldsymbol{s}},{\boldsymbol{\gamma}}')r_1}.

In the case 1 <\theta < \infty, by Lemma A, in view of {\boldsymbol{\gamma}} = {\boldsymbol{r}}/r_1, we obtain

Case \theta = \infty. In view of (?),

Lemma A yields

For the respective lower estimate, it is sufficient to consider the function g\in B^{\boldsymbol{r}}_{p,\theta} from the proof of Theorem 3.1.

Theorem 3.2 is proved.

We begin with the upper estimate.

As an approximation aggregate for a function f\in B^{\boldsymbol{r}}_{p,\theta}, we take the polynomial

One can easily check that t_n \in T(Q^{{\boldsymbol{\gamma}}'}_n).

Then, by the definition (?) of the norm in the space B_{p,1}, p \in \{ 1,\infty \}, we can write

Then we use the fact that, in view of the relation \|V_{2^s}\|_1 \leq C_5, C_5 > 0 (see, e.g., [Temlyakov1993, Ch. 1, Paragraph 1]), it holds

Hence, from (?) we obtain

Let first 1 \leq \theta < \infty. Using the Hölder’s inequality with the power \theta

(with an appropriate modification

in the case \theta = 1), and then in view of (?), {\boldsymbol{\gamma}} = {\boldsymbol{r}}/r_1 and Lemma A, from (?) we get

If \theta = \infty, then, in view of the relation \|A_{\boldsymbol{s}}(f)\|_p \ll 2^{-({\boldsymbol{s}},{\boldsymbol{r}})}

(see (?)), from (?) for p \in \{1, \infty\} we get

The upper estimate is now proved.

Concerning the lower estimate in (3.5), we note the following.

The estimate of the quantity E_{Q^{{\boldsymbol{\gamma}}'}_n} ( B^{\boldsymbol{r}}_{1,\theta})_{B_{1,1}} is a corollary from Theorem D (see below) and the relation \|\cdot\|_{B_{1,1}} \gg \|\cdot\|_1.

The lower estimate of E_{Q^{{\boldsymbol{\gamma}}'}_n} ( B^{\boldsymbol{r}}_{\infty,\theta})_{B_{\infty,1}} follows from the corresponding estimate of the Kolmogorov widths (see Theorem 4.2 and Remark 4.3), that will be obtained in

Section ?.

Theorem 3.5 is proved.

The upper estimates follow from Theorem 3.2. Indeed, taking into account that \|\cdot\|_{B_{1,1}} \ll \|\cdot\|_{B_{q,1}}, 1 \leq q < \infty, it is sufficient to prove these estimates in the case 1 < q < p \leq \infty. From the other side, due to the embeddings B^{\boldsymbol{r}}_{\infty,\theta} \subset B^{\boldsymbol{r}}_{p,\theta} \subset B^{\boldsymbol{r}}_{q, \theta}, we can restrict ourself in considering the case 1 < p =q < \infty, i.e., use the corresponding estimates obtained in Theorem 3.2.

As for the lower estimates in (3.10), we note that they follow from the corresponding estimates for the Kolmogorov width

(see Remark 4.3 to Theorem 4.2), that will be obtained in

Section ?.

Theorem 3.10 is proved.

In the case \theta = \infty, as it was noted above, the estimate was obtained in [Temlyakov1990]. Therefore, we assume that 1\leq \theta <\infty. Besides, it is sufficient to consider the case \nu = d.

We will use some relations from the proof of Theorem 2.2 in [Temlyakov1990], in particular, the estimate

where M = |Q^{'}_n| \asymp 2^n n^{d - 1}.

Since the embedding

is true, for f \in T'(Q'_n)_{\infty} \, 2^{-n r_1} it holds \|A_{\boldsymbol{s}}(f)\|_{\infty} \ll 2^{-({\boldsymbol{s}},r)}, s \in \Omega_n (see (?)).

Therefore, for f \in T'(Q'_n)_{\infty} \, 2^{-n r_1} from (?) and (?) we get

and thus

Hence, according to (Proof), in view of M \asymp 2^n n^{d - 1}, i.e., n \asymp \log M, we can write

Let A be a compact set and M_{\varepsilon}(A, \mathscr{X}) be the maximum number of points x^j \in A such that \|x^i - x^j\|_{\mathscr{X}} > \varepsilon, i \neq j; N_{\varepsilon}(A, \mathscr{X}) be a number defined by the formula (?).

It is well known that

The relations (?) and (?) yield that in T'(Q'_n)_{\infty}\, 2^{-n r_1} n^{ - \frac{d -1}{\theta}} there exist 2^M functions \{ f_j \}^{2^M}_{j=1} such that for i \neq j it holds

Further

we

show that

Indeed, taking into account (?) and the definition (?) of the norm in the space B_{1,1}, for f \in T'(Q'_n)_{\infty} we obtain

Combining (?) with (?), we get (?).

Hence, from (?) and (?), in view of M \asymp 2^n n^{d - 1}, we obtain the estimate

Theorem 4.1 is proved.

First we get the right order for d_M ( B^{\boldsymbol{r}}_{p, \theta}, B_{q,1}) and \varepsilon_M ( B^{\boldsymbol{r}}_{p,\theta}, B_{q,1}) by using Lemma B.

Indeed, on the one hand, the upper estimate for the Kolmogorov width follows from Theorems 3.2, 3.5 and 3.10 under the condition M \asymp 2^n n^{\nu - 1}:

On the other hand, the lower estimate for the entropy numbers is a corollary from Theorem 4.1:

As for \lambda_M ( B^{\boldsymbol{r}}_{p, \theta}, B_{q,1}), here we note, that for (p, q)\in \{(1, 1), (\infty, \infty)\}

the upper estimate for the quantity of the best approximation E_{Q^{{\boldsymbol{\gamma}}'}_n} ( B^{\boldsymbol{r}}_{p,\theta})_{B_{p,1}} from Theorem 3.5 is realized by a linear method, based on the polynomials (Proof). The corresponding lower estimate is due to the relation (?).

Theorem 4.2 is proved.

As in Theorem I, the upper estimate in Corollary 4.5

is realized by approximating functions from the respective classes

by trigonometric polynomials with harmonics from the hyperbolic crosses (see [Temlyakov1989_121p, p. 55] for the case \theta = \infty and [Romanyuk2011] for 1 \leq \theta < \infty).

The lower estimate in (4.5) follows from (H) and (I).

Let first 2\leq \theta < \infty. Then, on the one hand, Theorem B as M \asymp 2^n n^{\nu - 1} yields

On the other hand, according to [Romanyuk2015, Thm. 3], it holds

Using Lemma B with respect to the estimates (Proof) and (?), we get

Let \theta = \infty. Then we

use Lemma B with respect to the estimates

and

(see, e.g., [Temlyakov1989] and [Temlyakov1989_121p, Ch. 3, \S 2], respectively).

As for the linear widths, the respective estimate is due to the relation

under the condition M \asymp 2^n n^{\nu - 1}.

Theorem 4.7 is proved.