On Estimating the Norm of a Linear Operator in the Class of Entire Functions of Finite Degree
Unknown
Submitted 1963-01-01 | SovietRxiv: ru-196301.90717 | Translated from Russian

Abstract Generated abstract

This note studies norm estimates for linear operators acting on classes of entire functions of finite degree in several variables, with particular attention to relations between uniform and Lebesgue norms. Using integral representations of entire functions of exponential type and shift invariance assumptions on operators, it derives inequalities transferring bounds in the supremum norm to bounds in Lp norms and comparing the norms of two such operators. The results include Bernstein and Bernstein, Nikol’skii type inequalities for derivatives, refinements of Nikol’skii inequalities between Lp and Lq norms, and extensions to mixed norm generalized Lebesgue classes. Constants are given explicitly in terms of the degrees of the entire functions and the relevant exponents.

Full Text

MATHEMATICS

I. I. IBRAGIMOV

ON ESTIMATING THE NORM OF A LINEAR OPERATOR IN A CLASS OF ENTIRE FUNCTIONS OF FINITE DEGREE

(Presented by Academician V. I. Smirnov on May 6, 1963)

1. Let \(W_{\nu_1,\ldots,\nu_n}^{(p)}\) \((p \geqslant 1)\) denote the class of entire functions \(g(z_1,\ldots,z_n)\) of degree \(\leqslant (\nu_1,\ldots,\nu_n)\) and satisfying the condition

\[ \bigl(\|g\|_{p}^{(n)}\bigr)^p = \int_{-\infty}^{\infty}\cdots\int_{-\infty}^{\infty} |g(x_1,\ldots,x_n)|^p\,dx_1\cdots dx_n < +\infty . \]

Let, further, \(\mathfrak M\) be the set of linear operators \(T\) possessing the following properties:

\(1^\circ.\) \(T\) is defined on the set \(W_{\nu_1,\ldots,\nu_n}^{(p)}\), and its norm
\(\|T[g]\|_{p}^{(n)}\) is invariant with respect to any real shift in each argument.

\(2^\circ.\) There exists a constant \(A(\nu_1,\ldots,\nu_n)\) such that

\[ \|T[g]\|_{\infty}^{(n)} \leqslant A(\nu_1,\ldots,\nu_n)\,\|g\|_{p}^{(n)}, \]

where \(\|f\|_{\infty}^{(n)}\) denotes the norm of the function \(f(x_1,\ldots,x_n)\) in the metric of the space \(C^{(n)}(-\infty,\infty)\).

The first problem of the present note* is to establish, in the form of an inequality, a dependence between the various norms of different linear operators \(T\) and \(S\) from the set \(\mathfrak M\) in the class of entire functions \(W_{\nu_1,\ldots,\nu_n}^{(p)}\) \((p \geqslant 1)\). The solution of this problem is based on the fact that for any entire function \(g(z)=g(x+iy)\) from the class \(W_{\nu}^{(p)}\) \((p>1)\), for any \(z=x+iy\), the identity holds (see (1), p. 59)

\[ g(x+iy)=\frac{1}{\pi}\int_{-\infty}^{\infty} g(t-x)\,\frac{\sin \nu(t+iy)}{t+iy}\,dt . \tag{1} \]

By successive application of identity (1) to the function \(g(z_1,\ldots,z_n)\) with respect to each argument, we find that for an entire function \(g(z_1,\ldots,z_n)\) from the class \(W_{\nu_1,\ldots,\nu_n}^{(p)}\) \((p \geqslant 1)\) we have:

\[ g(x_1+iy_1,\ldots,x_n+iy_n)= \]

\[ = \left(\frac{1}{\pi}\right)^n \int_{-\infty}^{\infty}\cdots\int_{-\infty}^{\infty} f(t_1-x_1,\ldots,t_n-x_n) \prod_{k=1}^{n} \frac{\sin \nu_k(t_k+iy_k)}{t_k+iy_k} \,dt_1\cdots dt_n . \tag{2} \]

Hence, when \(y_1=y_2=\cdots=y_n=0\), it follows that for an entire function
\(g(z_1,\ldots,z_n)\in W_{\nu_1,\ldots,\nu_n}^{(p)}\) \((p \geqslant 1)\), for any real \(x_1,\ldots,x_n\), the identity holds

\[ g(x_1,\ldots,x_n)= \]

\[ = \left(\frac{1}{\pi}\right)^n \int_{-\infty}^{\infty}\cdots\int_{-\infty}^{\infty} g(t_1-x_1,\ldots,t_n-x_n) \prod_{k=1}^{n} \frac{\sin \nu_k t_k}{t_k} \,dt_1\cdots dt_k . \tag{3} \]

* The results of the present note were reported at the International Congress of Mathematicians in August 1962 in Stockholm and at the Second All-Union Conference on Constructive Function Theory in October 1962 in Baku.

Moreover, we use the fact that if \(g(z_1,\ldots,z_n)\in W^{(p)}_{\nu_1,\ldots,\nu_n}\) \((p\geqslant 1)\) and \(\varphi(x_1,\ldots,x_n)\in \mathscr{L}^{(n)}_{p'}\left(\dfrac1p+\dfrac1{p'}=1\right)\), then

\[ F(z_1,\ldots,z_n)=\int_{-\infty}^{\infty}\cdots\int_{-\infty}^{\infty} g(z_1+t_1,\ldots,z_n+t_n)\varphi(t_1,\ldots,t_n)\,dt_1\cdots dt_n \tag{4} \]

is an entire function of the class \(B_{\nu_1,\ldots,\nu_n}\) (see \((1)\), p. 38), and the function \(\varphi(t_1,\ldots,t_n)\) can be chosen so that the equality

\[ F(0,0,\ldots,0)=\|g\|^{(n)}_p \tag{5} \]

holds.

This leads to the following assertions:

Theorem 1. If \(g(z_1,\ldots,z_n)\) is an entire function of the class \(W^{(p)}_{\nu_1,\ldots,\nu_n}\) \((p\geqslant 1)\); \(T,S\) are linear operators from the set \(\mathfrak{M}\), then from the validity of the inequality*

\[ \|T[g]\|^{(n)}_\infty \leqslant A(\nu_1,\ldots,\nu_n)\|g\|^{(n)}_\infty \]

it follows that

\[ \|T[g]\|^{(n)}_p \leqslant A(\nu_1,\ldots,\nu_n)\|g\|^{(n)}_p \]

for every \(p\geqslant 1\). Moreover, from the validity of the inequality

\[ \|T[g]\|^{(n)}_\infty \leqslant \lambda(\nu_1,\ldots,\nu_n)\|S[g]\|^{(n)}_\infty \]

it follows that

\[ \|T[g]\|^{(n)}_p \leqslant \lambda(\nu_1,\ldots,\nu_n)\|S[g]\|^{(n)}_p \]

for every \(p\geqslant 1\).

In particular, if \(T\) is the differentiation operator, then from the validity of the classical inequality of S. N. Bernstein

\[ \left|\frac{\partial g(x_1,\ldots,x_n)}{\partial x_k}\right|\leqslant \nu_k\|g\|^{(n)}_\infty \]

there follows the Bernstein–Nikol’skii inequality \((3)\)

\[ \left\|\frac{\partial g(x_1,\ldots,x_n)}{\partial x_k}\right\|^{(n)}_p \leqslant \nu_k\|g\|^{(n)}_p \]

for every \(p\geqslant 1\).

Theorem 2. Let \(T,S\) be linear operators from the set \(\mathfrak{M}\), and let \(g(z_1,\ldots,z_n)\) be an entire function of the class \(W_{\nu_1,\ldots,\nu_n}\) \((p\geqslant 1)\). Then from the validity of the inequality

\[ \|T[g]\|^{(n)}_\infty \leqslant \lambda(\nu_1,\ldots,\nu_n)\|S[g]\|^{(n)}_\infty \]

it follows that

\[ \|T[g]\|^{(n)}_\infty \leqslant \prod_{k=1}^{n}\left(\frac{\nu_k}{\pi}\right)^{1/p} \lambda(\nu_1,\ldots,\nu_n)\|S[g]\|^{(n)}_p \]

for \(1\leqslant p\leqslant 2\), and, moreover,

\[ \|T[g]\|^{(n)}_q \leqslant \left(\prod_{k=1}^{n}\frac{\nu_k}{\pi}\right)^{1/p-1/q} \lambda(\nu_1,\ldots,\nu_n)\|S[g]\|^{(n)}_p \]

for \(1\leqslant p<q\leqslant +\infty\).

Theorem 3. If \(g(z_1,\ldots,z_n)\) is an entire function of the class \(W^{(p)}_{\nu_1,\ldots,\nu_n}\) \((p\geqslant 1)\), \(T\) is a linear operator from the set \(\mathfrak{M}\), \(1\leqslant p\leqslant 2\),

* Theorem 1 in the one-dimensional case was proved by another method in work \((2)\).

\(1 \leqslant p < q \leqslant +\infty\), then

\[ \|T[g]\|_{q}^{(n)} \leqslant \left[\left(\frac{1}{\pi}\right)^n B_q^{n/q} \left(\prod_{k=1}^{n} \nu_k\right)^{1/p}\right]^{1-p/q} \|T[g]\|_{p}^{(n)}, \]

where

\[ B_{\infty}=1,\qquad B_q=\int_{-\infty}^{\infty}\left|\frac{\sin u}{u}\right|^q\,du. \]

Hence, in the case when \(T\) is the identity transformation, the inequality

\[ \|g\|_{q}^{(n)} \leqslant \prod_{k=1}^{n}\left(\frac{s\nu_k}{\pi}\right)^{1/p-1/q} \|g\|_{p}^{(n)} \]

holds for all \(p\) and \(q\) satisfying the condition \(1 \leqslant p<q\leqslant+\infty\), where \(s=[[{-p/2}]]\) is the least integer not less than \(p/2\). The last inequality is a refinement of S. M. Nikol’skii’s inequality \((^3)\), which was generalized and sharpened in works \((^{4-6})\) for more general differential operators.

II. Let \(\varphi(x_1,\ldots,x_n)\geqslant 1\) be a continuous function in \(n\)-dimensional Euclidean space \((R_n)\), let \(p\geqslant 1\) be any number, and let \(\Lambda_{P,\varphi}\) be the class of functions \(f(x_1,\ldots,x_n)\) possessing the property

\[ \|f\|_{P,\varphi}^{(n)} = \left\|\ldots \left\{ \left\|\ldots \left(\left\|\frac{f}{\varphi}\right\|_{p_1}\right) \ldots\right\|_{p_k} \right\} \ldots\right\|_{p_n} <+\infty, \]

where \(P=(p_1,p_2,\ldots,p_n)\) and \(p_1,p_2,\ldots,p_n\) are various numbers not less than unity. In particular, for \(n=3\) the norm \(\|f\|_{P,\varphi}^{(3)}\) has the form:

\[ \|f\|_{P,\varphi}^{(3)} = \left\{ \int_{-\infty}^{\infty} \left[ \int_{-\infty}^{\infty} \left( \int_{-\infty}^{\infty} \left| \frac{f(x_1,x_2,x_3)}{\varphi(x_1,x_2,x_3)} \right|^{p_1} dx_1 \right)^{p_2/p_1} dx_2 \right]^{p_3/p_2} dx_3 \right\}^{1/p_3} <+\infty. \]

Obviously, the class \(\Lambda_{P,\varphi}^{(n)}\), called the generalized Lebesgue class, coincides with the ordinary Lebesgue class \(\mathcal L_p^{(n)}(-\infty,\infty)\) when

\[ \varphi(x_1,\ldots,x_n)\equiv 1,\qquad p_1=\cdots=p_n=p. \]

Further, let \(W_{\nu_1,\ldots,\nu_n}^{(P,\varphi)}\) denote the class of entire functions \(g(z_1,\ldots,z_n)\) of finite degree \((\nu_1,\ldots,\nu_n)\) that belong to the space \(\Lambda_{P,\varphi}^{(n)}\). Obviously, the class \(W_{\nu_1,\ldots,\nu_n}^{(P,\varphi)}\) in the case \(\varphi\equiv 1\) and \(p_1=p_2=\cdots=p_n=p\) coincides with the class \(W_{\nu_1,\ldots,\nu_n}^{(n)}\).

In the case when \(p_1,p_2,\ldots,p_n\) are various numbers not less than unity, and \(\varphi(x_1,\ldots,x_n)\equiv 1\), the notations used are

\[ W_{\nu_1,\ldots,\nu_n}^{(p,1)} \equiv W_{\nu_1,\ldots,\nu_n}^{(p_1,\ldots,p_n)}, \qquad \|f\|_{P,1}^{(n)}=\|f\|_{p_1,\ldots,p_n}. \]

The second problem* of the present note consists in establishing a connection between the different norms \(\|g\|_{p_1,\ldots,p_n}\) and \(\|g\|_{p'_1,\ldots,p'_n}\) of an entire function \(g(z_1,\ldots,z_n)\) from the class \(W_{\nu_1,\ldots,\nu_n}^{(p_1,\ldots,p_n)}\), where \(1\leqslant p_i<p'_i\leqslant\infty\) \((i=1,2,\ldots,n)\).

\[ \text{*} \]

  • A less precise result with respect to the constant was obtained by the author, by another method, in the work \((^6)\), carried out jointly with A. S. Dzhafarov, where a connection was established between the different norms \(\|g\|_{P',\varphi}^{(n)}\) and \(\|g\|_{P,\varphi}^{(n)}\) in the class \(W_{\nu_1,\ldots,\nu_n}^{(P,\varphi)}\), with \(P=(p_1,\ldots,p_n)\), \(P'=(p'_1,\ldots,p'_n)\), and \(1\leqslant p_i<p'_i\leqslant\infty\) \((i=1,2,\ldots,n)\).
  1. If \(g(z_1,\ldots,z_n)\in W_{\nu_1,\ldots,\nu_n}^{(p_1,\ldots,p_n)}\) and \(p_1,p_2,\ldots,p_n\) are distinct numbers not less than one, then

\[ \max_{-\infty<x_1,\ldots,x_n<\infty}|g(x_1,\ldots,x_n)| \le \prod_{k=1}^{n}\left(\frac{s_k\nu_k}{\pi}\right)^{1/p_k} \|g\|_{p_1,\ldots,p_n}, \]

where \(s_k=|[-p_k/2]|\) is the least integer not less than \(p_k/2\) \((k=1,2,\ldots,n)\).

  1. If \(g(z_1,\ldots,z_n)\in W_{\nu_1,\ldots,\nu_n}^{(p_1,\ldots,p_n)}\); \(p_1,p_2,\ldots,p_n,\ p'_1,p'_2,\ldots,p'_n\) are distinct numbers not less than one, and \(1\le p_i\le p'_i\le\infty\) \((i=1,2,\ldots,n)\), then we have*

\[ \|g\|_{p'_1,\ldots,p'_n} \le \prod_{k=1}^{n}\left(\frac{s_k\nu_k}{\pi}\right)^{1/p_k-1/p'_k} \|g\|_{p_1,\ldots,p_n}, \tag{7} \]

where

\[ s_k=|[-p_k/2]| \quad (k=1,2,\ldots,n). \]

Institute of Mathematics and Mechanics
Academy of Sciences of the Azerbaijan SSR

Received
3 V 1963

REFERENCES

¹ I. I. Ibragimov, Extremal Properties of Entire Functions of Finite Degree, Baku, 1962.
² E. Stein, Ann. Math., 65, No. 3 (1957).
³ S. M. Nikol’skii, Tr. Mat. Inst. im. V. A. Steklova AN SSSR, 38 (1951).
⁴ I. I. Ibragimov, Izv. AN SSSR, ser. matem., 23, 243 (1959).
⁵ I. I. Ibragimov, A. S. Dzhafarov, DAN, 138, No. 4 (1961).
⁶ I. I. Ibragimov, A. S. Dzhafarov, Izv. AN AzerbSSR, ser. phys.-math. and techn. sciences, No. 5 (1962).
⁷ S. M. Nikol’skii, Siberian Math. Journal, 3, No. 6 (1962).

* A special case of inequality (7) with a nonsharp constant was considered independently of us by S. M. Nikol’skii (⁷).

Submission history

On Estimating the Norm of a Linear Operator in the Class of Entire Functions of Finite Degree