Abstract Generated abstract
This paper establishes criteria under which systems of analytic functions of the form \(z^n\varphi_n(z)\) constitute a basis for holomorphic functions in a disk or related simply connected domain. The authors express the factors through analytic logarithms, derive coefficient recursions for formal expansions, and use majorization estimates to prove convergence and uniqueness under a sublinear growth condition on cumulative logarithmic differences. The main theorem is extended to systems \(u^n(z)\psi_n(z)\) associated with a univalent mapping, and to cases with weaker growth bounds yielding a smaller basis domain. Several examples show how the criterion applies to exponential perturbations and powers of analytic factors.
Full Text
MATHEMATICS
M. A. EVGRAFOV and A. D. SOLOV’EV
ON A GENERAL CRITERION FOR A BASIS
(Presented by Academician A. N. Kolmogorov on 12 X 1956)
We shall say that a system of functions
\[ u_n(z)=z^n\varphi_n(z),\qquad \varphi_n(0)=1,\qquad n=0,1,2,\ldots, \tag{1} \]
regular in a domain \(G\), forms a basis in a domain \(G_1\subset G\), if every function \(f(z)\), regular in \(G_1\), is represented in this domain by a uniformly convergent series
\[ f(z)=\sum_{n=0}^{\infty} a_n u_n(z), \]
and this representation is unique.
Theorem 1. Let a system (1) be given, where the functions \(\varphi_n(z)\) are regular in the disk \(|z|<R\) and do not vanish inside this disk.
Write the system (1) in the form
\[ u_n(z)=z^n e^{-\lambda_n(z)}, \]
where the functions \(\lambda_n(z)\) are regular in the disk \(|z|<R\).
Introduce the notation
\[ \lambda_n(z)-\lambda_{n-1}(z)=\Delta_n(z)=\sum_{k=1}^{\infty}\Delta_{nk}z^k,\qquad \Delta_0(z)=\lambda_0(z); \]
\[ \Delta_n^0(r)=\sum_{k=1}^{\infty}|\Delta_{nk}|r^k,\qquad l_n(r)=\sum_{k=0}^{n}\Delta_k^0(r). \tag{2} \]
If the functions \(\lambda_n(z)\) satisfy the condition
\[ \lim_{n\to\infty}\frac{l_n(r)}{n}=0 \quad\text{for every } r<R, \tag{3} \]
then the system (1) forms a basis in the disk \(|z|<R\).
Proof. Let \(f(z)=\sum_{n=0}^{\infty}c_n z^n\) be an arbitrary function regular in the disk \(|z|<R\). Write the expansion
\[ f(z)=\sum_{n=0}^{\infty} a_n z^n e^{-\lambda_n(z)}. \]
From this formal identity, by comparing coefficients of like powers of \(z\), one can obtain finite recurrence relations from which the numbers \(a_n\) are determined successively (and moreover uniquely).
By virtue of condition (3),
\[ \lim_{n\to\infty}\sqrt[n]{\,|z^n\varphi_n(z)|\,}=|z|,\qquad |z|<R, \]
therefore, in order to prove the theorem it is enough to show that
\[ \lim_{n\to\infty}\sqrt[n]{|a_n|}\leq \frac1R . \]
Introduce the numbers \(a_{nk}\) as the coefficients of the series
\[ f(z)=\sum_{n=0}^{k-1} a_{nk}z^n e^{-\lambda_n(z)} +\sum_{n=k}^{\infty} a_{nk}z^n e^{-\lambda_k(z)} . \]
It is not difficult to verify that the property of the system (1) of being a basis in the disk \(|z|<R\) will not be violated if we change a finite number of functions \(\varphi_n(z)\) (while, of course, preserving the conditions: \(\varphi_n(z)\) are regular for \(|z|<R\) and \(\varphi_n(0)=1\)). Therefore our series converges in the same disk in which the series
\[ \sum_{n=0}^{\infty} \widetilde a_{nk}z^n=f(z)e^{\lambda_k(z)}, \]
does, i.e. for \(|z|<R\).
We note that \(a_{nk}=a_n\) for \(n<k+1\), since the coefficients \(a_n\) depend only on the first \(n\) functions \(\lambda_0(z),\lambda_1(z),\ldots,\lambda_{n-1}(z)\).
Write two identities:
\[ f(z)=\sum_{n=0}^{k-1} a_n z^n e^{-\lambda_n(z)} +\left(\sum_{n=k}^{\infty} a_{nk}z^n\right)e^{-\lambda_k(z)}, \]
\[ f(z)=\sum_{n=0}^{k-1} a_n z^n e^{-\lambda_n(z)} +\left(\sum_{n=k}^{\infty} a_{n,k-1}z^n\right)e^{-\lambda_{k-1}(z)} . \]
Comparing them, we obtain
\[ \sum_{n=k}^{\infty} a_{nk}z^n = \left(\sum_{n=k}^{\infty} a_{n,k-1}z^n\right)e^{\Delta_k(z)} . \tag{4} \]
We shall prove that every \(a_{nk}=a_{nk}(\Delta_{ij};c_m)\) is a polynomial in the \(\Delta_{ij}\) and \(c_m\) with positive coefficients. We carry out the proof by induction on the index \(k\).
\[ \sum_{n=0}^{\infty} a_{n0}z^n=f(z)e^{\Delta_0(z)}, \]
so that for the numbers \(a_{n0}\) this assertion is true. Suppose it is true for the numbers \(a_{n,k-1}\). Then from identity (4) it follows that it is true also for the numbers \(a_{nk}\).
Now assume that all coefficients \(c_m\) and \(\Delta_{ij}\) are nonnegative. Then all \(a_{nk}\geq 0\).
Denote
\[ f_k(z)=\sum_{n=k}^{\infty} a_{nk}z^n . \]
From (4) it follows that
\[ f_k(z)\ll f_{k-1}(z)e^{\Delta_k(z)}, \tag{5} \]
where the symbol \(\ll\) means that the series on the left is majorized by the series on the right, i.e. the coefficients of the left-hand series do not exceed the corresponding coefficients of the right-hand series.
From (5) it follows immediately that
\[ f_k(z)\ll e^{\lambda_k(z)}f(z), \]
whence we obtain
\[ a_{nk}\ll A_{nk}=\frac{1}{2\pi i}\int_{|z|=r<R} \frac{f(z)e^{\lambda_k(z)}}{z^{n+1}}\,dz \ll \frac{e^{\lambda_k(r)}}{r^n}f(r). \]
If now \(c_m\) and \(\Delta_{ij}\) are arbitrary complex numbers, then
\[ |a_{nk}(\Delta_{ij},c_m)|\ll a_{nk}(|\Delta_{ij}|;|c_m|) \ll \frac{e^{l_k(r)}}{r^n}M_0(r), \]
where
\[ M_0(r)=\sum_{m=0}^{\infty}|c_m|r^m. \]
In particular,
\[ |a_n|\ll M_0(r)\frac{e^{l_n(r)}}{r^n},\qquad r<R. \tag{6} \]
From this inequality it follows that
\[ \varlimsup_{n\to\infty}\sqrt[n]{|a_n|}\ll \frac1r, \]
whence, by virtue of the arbitrariness of \(r<R\),
\[ \varlimsup_{n\to\infty}\sqrt[n]{|a_n|}\ll \frac1R. \]
The theorem is proved.
Remark 1. Theorem 1 will also be valid in the case when, in each disk \(|z|\le r<R\), only a finite number of the functions \(\varphi_n(z)\) have zeros, since we can always change the functions having zeros for \(|z|\le r\) (putting, for example, \(\varphi_n(z)=1\)) without violating the property of system (1) of being a basis in the disk \(|z|<R\).
Remark 2. It can be shown that if \(|z|<R\) is the largest disk in which the conditions of Theorem 1 are fulfilled (weakened by Remark 1), then for the given functions \(\Delta_n^0(r)\) there exists a system (1) which in the disk \(|z|<R+\varepsilon\), for any \(\varepsilon>0\), no longer forms a basis.
Remark 3. Inequality (6) does not depend on condition (3) and has a universal character.
From Theorem 1 the following follows without difficulty:
Theorem 2. Suppose a system is given
\[ u_n(z)=u^n(z)\psi_n(z),\qquad \psi_n(0)=1,\qquad n=0,1,\ldots, \tag{7} \]
where \(u(z)=z+\ldots\) is a function regular and univalent in a simply connected domain \(G\), which is mapped by the function \(u(\zeta)\) onto a disk with center at the origin. The functions \(\psi_n(z)\) are regular in the domain \(G\), and on each closed set \(E\subset G\) only a finite number of these functions have zeros. Denote
\[ l_n(E)=\sum_{k=k_0}^{n}\max_{z\in E}\left|\ln\left[\frac{\psi_{k-1}(z)}{\psi_k(z)}\right]\right| \]
\[ (E\subset G\text{ is a closed set},\quad k_0=k_0(E)). \]
If
\[ \lim_{n\to\infty}\frac{l_n(E)}{n}=0 \quad \text{for every } E\subset G, \]
then the system (7) forms a basis in the domain \(G\).
Theorem 3. If in Theorem 1 condition (3) is replaced by the condition
\[ \overline{\lim_{n\to\infty}}\frac{l_n(r)}{n}=l(r), \]
then the system (1) will form a basis in the disk
\[ |z|\,e^{l_1(|z|)}<\rho, \]
where
\[ \rho=\sup_{r<R} r e^{-l(r)}, \qquad l_1(r)=\overline{\lim_{n\to\infty}}\frac{\max_{|z|=r}|\lambda_n(z)|}{n}\ll l(r). \]
Let us consider several examples of applications of these theorems.
1.
\[ u_n(z)=z^n\varphi(z)e^{\lambda_n z}. \tag{8} \]
Suppose that
\[ \sum_{k=0}^{n}|\lambda_k-\lambda_{k-1}|=o(n). \]
In this case the system forms a basis in the disk \(|z|<|\alpha_1|\), where \(\alpha_1\) is the zero of the function \(\varphi(z)\) nearest to the origin.
2.
\[ u_n(z)=u^n(z)e^{\lambda_n z}, \qquad u(z)=z+az^2+\cdots . \tag{9} \]
Again require that
\[ \sum_{k=0}^{n}|\lambda_k-\lambda_{k-1}|=o(n). \]
Then the system (9) forms a basis in the domain \(G\) in which the function \(\mu(z)\) is univalent and which is mapped by the function \(u(z)\) onto a disk with center at the origin.
3.
\[ u_n(z)=z^n\varphi^{\lambda_n}(z), \qquad \varphi(0)=1. \tag{10} \]
Let, again,
\[ \sum_{k=0}^{n}|\lambda_k-\lambda_{k-1}|=o(n). \]
Then the system will form a basis in the disk \(|z|<|\alpha_1|\) (\(\alpha_1\) is the zero of the function \(\varphi(z)\) nearest to the origin).
Received
12 X 1956
REFERENCES
- A. O. Gel’fond, Calculus of Finite Differences, 1952.
- M. A. Evgrafov, Izv. AN SSSR, Ser. Mat., 18, 449 (1954).
- A. I. Markushevich, Matem. sborn., 17 (59), no. 2, 211 (1945).