Abstract Generated abstract
This paper extends Sobolev and Nikol’skii type embedding theorems to function classes with weighted integral norms and anisotropic smoothness in several variables. Using results on best approximation by entire functions of finite degree in weighted norms, it formulates embeddings for classes with general weights and logarithmic smoothness factors, including limiting cases expressed through iterated logarithms. The main theorems give conditions under which functions or their mixed partial derivatives belong to corresponding weighted smoothness classes with transformed parameters, and establish norm estimates independent of the particular function. The results reduce to known theorems of Nikol’skii and Besov when the weight is identically one.
Full Text
MATHEMATICS
A. S. DZHAFAROV
EMBEDDING THEOREMS WITH A WEIGHT
(Presented by Academician A. N. Kolmogorov on 28 VIII 1961)
The embedding theorems of S. L. Sobolev \((^{1,2})\), whose fundamental importance in the theory of boundary-value problems of mathematical physics is well known, have very many diverse generalizations, additions, and applications. Among them one should especially note the investigations of S. M. Nikol’skii \((^{3-6})\) on the embedding of the classes
\(H_{p_1,\ldots,p_n}^{(r_1,\ldots,r_n)}[M]\), who was the first to apply, for these purposes, methods of the theory of best approximation of functions of several variables. We also note that embedding theorems with a weight were first proved by L. D. Kudryavtsev \((^{7,8})\). Works \((^{9,10})\) are also devoted to these questions.
In the present paper we give results of the type of S. M. Nikol’skii’s embedding theorems for integral norms containing a general weight, whose character may differ from that of the weight functions used in the above-mentioned works. These results were obtained on the basis of the theorems of \((^{12})\), which relate the behavior of best approximations of functions of several variables in norms containing a weight, by means of entire functions of finite degrees, to the differential properties of the functions under consideration.
Below, in addition to the notation and definitions of \((^{12})\), we shall also use the following definitions. We shall say that a function \(f(x_1,\ldots,x_n)\) belongs to the class \(H_{p,\varphi,x_k}[\psi]\) if \(\|f\|_{p,\varphi}^{(n)}<\infty\) and the inequality
\(\|f(x_1,\ldots,x_{k-1},x_k+h,x_{k+1},\ldots,x_n)-2f(x_1,\ldots,x_n)+f(x_1,\ldots,x_{k-1},x_k-h,x_{k+1},\ldots,x_n)\|_{p,\varphi}^{(n)}\le \psi(|h|)\) holds for all \(h\) for which \(\psi(|h|)\) has meaning, where \(\psi(|h|)\) is a function tending to zero as \(h\to0\). In what follows we shall assume that \(f(x_1,\ldots,x_n)\) belongs to the class \(H_{p,\varphi,n}[\psi]\) if it simultaneously belongs to all the classes \(H_{p,\varphi,x_k}[\psi]\).
Let \(\ln_1 x=\ln x\), \(\ln_j x=\ln(\ln_{j-1}x)\) \((j=2,3,\ldots)\). We shall say that a function \(f(x_1,\ldots,x_n)\) belongs to the class
\(H_{p,\varphi,x_k}^{(r)}\left[M\prod_{j=1}^{N}\ln_j^{s_j}\right]\), where \(r=\bar r+\alpha>0\) (\(\bar r\) is a nonnegative integer, \(0<\alpha\le1\)), and \(s_j\) are arbitrary real numbers, if \(f\in L_{p,\varphi}^{(n)}\) has a generalized partial derivative \(\partial^{\bar r}f/\partial x_k^{\bar r}\) belonging to the class \(H_{p,\varphi,x_k}[\psi]\), where
\[ \psi(|h|)=M|h|^\alpha\prod_{j=1}^{N}\left(\ln_j\frac1{|h|}\right)^{s_j}, \]
and \(M\) does not depend on \(h\) (when \(s_j=0\), \(j=1,\ldots,N\), we shall simply say that \(f\in H_{p,\varphi,x_k}^{(r)}[M]\)).
Further we shall assume that \(f(x_1,\ldots,x_n)\) belongs to the class
\[ H_{p,\varphi}^{(r_1,\ldots,r_n)} \left[ M_1\prod_{j=1}^{N}\ln_j^{s_j^{(1)}},\ldots, M_n\prod_{j=1}^{N}\ln_j^{s_j^{(n)}} \right], \tag{1} \]
if it belongs simultaneously to the classes
\(H_{p,\varphi,x_k}^{(r_k)}\left[M_k\prod_{j=1}^{N}\ln_j^{s_j^{(k)}}\right]\)
\((k=1,2,\ldots,n)\). For \(\varphi\equiv1\), \(s_j^{(i)}=0\), \((i=1,2,\ldots,n;\ j=1,2,\ldots,N)\)
the class (1) coincides with the class \(H_p^{(r_1,\ldots,r_n)}[M_1,\ldots,M_n]\) of S. M. Nikol’skii. Let \(r_i>0,\ 1\le p_i\le \infty\) \((i=1,2,\ldots,n)\); we shall say that \(f(x_1,\ldots,x_n)\) belongs to the class
\[ H_{p_1,\ldots,p_n}^{(r_1,\ldots,r_n)}[M], \tag{2} \]
if, in each of the variables \(x_i\), it belongs to the class \(H_{p_i,\varphi,x_i}^{(r_i)}[M]\) \((i=1,\ldots,n)\). For \(\varphi\equiv1\) the class (2) was first introduced by S. M. Nikol’skii \({}^{(6)}\).
Theorem 1. Let \(1\le p\le p'\le \infty,\ r_i>0\) \((i=1,2,\ldots,n)\), \(1\le m\le n\),
\[ \varkappa_m = 1-\left(\frac1p-\frac1{p'}\right)\sum_{i=1}^{m}\frac1{r_i} -\frac1p\sum_{i=m+1}^{n}\frac1{r_i}, \]
\[ \theta_m^{(j)} = \left(\frac1p-\frac1{p'}\right)\sum_{i=1}^{m}\frac{s_i^{(j)}}{r_i} -\frac1p\sum_{i=m+1}^{n}\frac{s_i^{(j)}}{r_i}, \]
and let \(f\) belong to the class (1).
Then:
1) If \(\varkappa_m>0\), then, for fixed \(x_{m+1},\ldots,x_n\), the function \(f\), in the variables \(x_1,\ldots,x_m\), belongs to the class
\[ H_{p',\varphi}^{(\rho_1,\ldots,\rho_m)} \left[ M^*\prod_{j=1}^{N}\ln_j^{\,\varkappa_m s_j^{(1)}+\theta_m^{(j)}}, \ldots, M^*\prod_{j=1}^{N}\ln_j^{\,\varkappa_m s_j^{(m)}+\theta_m^{(j)}} \right], \]
where \(\rho_i=\varkappa_m r_i,\ M^*\le c_1\left(\|f\|_{p,\varphi}^{(n)}+\sum_{1}^{n}M_k\right)\), and \(C_1\) does not depend on \(f,\ M_k,\ x_{m+1},\ldots,x_n,\ h\).
2) If \(\varkappa_m=0\), but there exists a natural number \(l\) such that for \(j=1,2,\ldots,l-1\) \(\theta_m^{(j)}=-1\) and \(\theta_m^{(l)}<-1\), then, for fixed \(x_{m+1},\ldots,x_n\), the function \(f\), with respect to \(x_1,\ldots,x_m\), belongs to the class
\[ H_{p',\varphi,m} \left[ M^*\ln_l^{\,\theta_m^{(l)}+1} \prod_{j=l+1}^{N}\ln_j^{\,\theta_m^{(j)}} \right], \]
where \(M^*\) is defined above.
In both cases
\[ \|f\|_{p',\varphi}^{(m)} \le C_2\left(\|f\|_{p,\varphi}^{(n)}+\sum_{i=1}^{n}M_i\right), \]
where \(C_2\) does not depend on \(f,\ x_{m+1},\ldots,x_n\).
Theorem 2. Let the function \(f\) belong to the class (1), and let \(\lambda_1,\ldots,\lambda_n\) be nonnegative integers,
\[ \sigma=1-\sum_{k=1}^{n}\frac{\lambda_k}{r_k}, \qquad \Delta_j=\sum_{i=1}^{n}\frac{s_j^{(i)}\lambda_i}{r_i}, \]
\[ r_i'=\sigma r_i,\qquad \alpha_j^{(i)}=\sigma s_j^{(i)}+\Delta_j. \]
Then, if: 1) \(\sigma>0\), or else 2) \(\sigma=0\), but there exists a natural number \(l\) such that for \(j=1,2,\ldots,l-1\) \(\Delta_j=-1\) and \(\Delta_l<-1\), then on \(E_n\) there exists the partial derivative \(\partial^{\lambda_1+\cdots+\lambda_n}f/\partial x_1^{\lambda_1}\cdots\partial x_n^{\lambda_n}\), belonging in case 1) to the class
\[ H_{p,\varphi}^{(r_1',\ldots,r_n')} \left[ \overline M\prod_{j=1}^{N}\ln_j^{\alpha_j^{(1)}}, \ldots, \overline M\prod_{j=1}^{N}\ln_j^{\alpha_j^{(n)}} \right]. \]
and in case 2) to the class
\[ H_{p,\varphi,n}\left[\overline M \ln_l^{\Delta_l+1}\prod_{j=l+1}^N \ln_j^{\Delta_j}\right]. \]
In both cases
\[ M+\left\|\frac{\partial^{\lambda_1+\cdots+\lambda_n} f}{\partial x_1^{\lambda_1}\cdots \partial x_n^{\lambda_n}}\right\|_{p,\varphi}^{(n)} \leqslant C_3\left(\|f\|_{p,\varphi}^{(n)}+\sum_{k=1}^n M_k\right), \]
where \(C_3\) does not depend on \(f, M_k, h\).
Theorem 3. Let \(r_i>0;\ 1\leqslant p_i\leqslant q\leqslant\infty;\ n, m\) be natural numbers for which \(1\leqslant m\leqslant n\),
\[ \rho^{(i)}=\frac{r_i\varkappa}{\varkappa^{(i)}}>0\qquad (i=1,\ldots,n), \]
where
\[ \varkappa= \left| \begin{matrix} 1-\displaystyle\sum_{j=1}^n \frac{\frac1{p_j}-\frac1q}{r_j} & -\displaystyle\frac1q\sum_{j=1}^n \frac1{r_j} \\[1.2em] -\displaystyle\sum_{j=m+1}^n \frac{\frac1{p_j}-\frac1q}{r_j} & 1-\displaystyle\frac1q\sum_{j=m+1}^n \frac1{r_j} \end{matrix} \right|, \]
\[ \varkappa^{(i)}=1-\sum_{j=1}^n \frac{\frac1{p_j}-\frac1{p_i}}{r_j}\qquad (i=1,\ldots,n). \]
If the function \(f(x_1,\ldots,x_n)\) belongs to the class \(H_{p_1,\ldots,p_n;\varphi}^{(r_1,\ldots,r_n)}[M]\), then, for any fixed \(x_{m+1},\ldots,x_n\), the function \(f\), as a function of \(x_1,\ldots,x_m\), belongs to the class \(H_{q,\varphi}^{(\rho^{(1)},\ldots,\rho^{(m)})}[\overline M]\), and the inequality
\[ \|f\|_{q,\varphi}^{(m)}+\overline M < C_4\left(\min_{1\leqslant i\leqslant n}\|f\|_{p_i,\varphi}^{(n)}+M\right) \]
holds, where the constant \(C_4\) does not depend on \(f, M, x_{m+1},\ldots,x_n\).
Remark. The theorems stated above, for \(\varphi\equiv 1,\ s_j^{(i)}=0\ (i=1,\ldots,n;\ j=1,\ldots,N)\), coincide completely with the corresponding theorems of S. M. Nikol’skii \((^{3\text{–}6})\). For \(\varphi\equiv 1\), Theorems 1 and 2 were obtained earlier in \((^{11})\).
Let us note that if, by \(B_{p,\theta,\varphi}^{(r_1,\ldots,r_n)}\), we denote the space of functions \(f\in L_{p,\varphi}^{(n)}\) having on \(E_n\) partial generalized unmixed derivatives in the sense of S. L. Sobolev
\[
\partial^k f/\partial x_i^k\in L_{p,\varphi}^{(n)}\quad (k=0,1,\ldots,r_i;\ i=1,2,\ldots,n)
\]
with norm
\[ \|f\|_{B_{p,\theta,\varphi}^{(r_1,\ldots,r_n)}}= \|f\|_{p,\varphi}^{(n)} + \sum_{i=1}^n \left\{ \int_0^1 \frac{ \omega_{1+[\alpha_i],x_i}^{\theta} \left(t;\frac{\partial^{r_i}f}{\partial x_i^{r_i}}\right)_{p,\varphi}^{(n)} }{ t^{\theta\alpha_i+1} }\,dt \right\}^{1/\theta} <\infty, \]
where \(r_i=\bar r_i+\alpha_i>0\), \(\bar r_i\) are nonnegative integers, \(0<\alpha_i\leqslant 1\), then
Theorems 1–3 of O. V. Besov\(^{13}\) remain valid if, in their formulations, \(B_{p,\theta}^{(r_1,\ldots,r_n)}\) is replaced in the corresponding way by \(B_{p,\theta,\varphi}^{(r_1,\ldots,r_n)}\).
For \(\varphi \equiv 1\), the space \(B_{p,\theta,\varphi}^{(r_1,\ldots,r_n)}\) coincides with O. V. Besov’s space \(B_{p,\theta}^{(r_1,\ldots,r_n)}\).
Institute of Mathematics and Mechanics
Academy of Sciences of the Azerbaijan SSR
Received
21 VII 1961
REFERENCES
\(^{1}\) S. L. Sobolev, Matem. sborn., 4 (46), 3, 471 (1938).
\(^{2}\) S. L. Sobolev, Some Applications of Functional Analysis in Mathematical Physics, 1950.
\(^{3}\) S. M. Nikol’skii, DAN, 76, 785 (1951).
\(^{4}\) S. M. Nikol’skii, Tr. Matem. inst. im. V. A. Steklova AN SSSR, 38, 244 (1951).
\(^{5}\) S. M. Nikol’skii, Matem. sborn., 33 (75), 2, 261 (1953).
\(^{6}\) S. M. Nikol’skii, Izv. AN SSSR, ser. matem., 22, 321 (1958).
\(^{7}\) L. D. Kudryavtsev, DAN, 107, No. 4, 501 (1956).
\(^{8}\) L. D. Kudryavtsev, Tr. Matem. inst. im. V. A. Steklova AN SSSR, 60 (1959).
\(^{9}\) V. P. Glushko, DAN, 126, No. 3, 467 (1959).
\(^{10}\) V. P. Glushko, S. T. Krein, Sibirsk. matem. zhurn., 1, 3, 343 (1960).
\(^{11}\) A. S. Dzhafarov, Dokl. AN AzerbSSR, 14, No. 7, 499 (1958).
\(^{12}\) A. S. Dzhafarov, DAN, 142, No. 2 (1961).
\(^{13}\) O. V. Besov, DAN, 126, No. 6, 1163 (1959).