1 Introduction
Laplace–Stieltjes transforms
$$ F(s)= \int_{0}^{+\infty}e^{sx}\,d\alpha(x),\quad s= \sigma+it, $$
(1)
where
\(\alpha(x)\) is a bounded variation on any finite interval
\([0,Y]\) (
\(0< Y<+\infty\)), and
σ and
t are real variables, as we know, if
\(\alpha(t)\) is absolutely continuous, then
\(F(s)\) becomes the classical Laplace integral of the form
$$ F(s)= \int_{0}^{\infty}e^{st}g(t)\,dt. $$
(2)
If
\(\alpha(t)\) is a step-function and satisfies
$$\alpha(x)= \textstyle\begin{cases} a_{1}+a_{2}+\cdots+a_{n}, &\lambda_{n}< x< \lambda_{n+1};\\ 0, &0\leq x< \lambda_{1};\\ \frac{\alpha(x+)+\alpha(x-)}{2},&x>0, \end{cases} $$
where the sequence
\(\{\lambda_{n}\}_{0}^{\infty}\) satisfies
$$ 0=\lambda_{1}< \lambda_{2}< \lambda_{3}< \cdots< \lambda_{n}\uparrow+\infty, $$
(3)
thus
\(F(s)\) becomes a Dirichlet series
$$ F(s)=\sum_{n=1}^{\infty}a_{n}e^{\lambda_{n}s},\quad s=\sigma+it. $$
(4)
(
σ,
t are real variables),
\(a_{n}\) are nonzero complex numbers. Obviously, if
\(\alpha(t)\) is an increasing continuous function which is not absolutely continuous, then the integral (
1) defines a class of functions
\(F(s)\) which cannot be expressed either in the form (
2) or (
4).
Let a sequence
\(\{\lambda_{n}\}_{n=1}^{\infty}\) satisfy (
3), and
$$ \limsup_{n\rightarrow+\infty}(\lambda_{n+1}- \lambda_{n})=h< +\infty ,\qquad \limsup_{n\rightarrow+\infty} \frac{ n}{\lambda_{n}}=D< \infty. $$
(5)
Set
$$A_{n}^{*}=\sup_{\lambda_{n}< x\leq \lambda_{n+1},-\infty< t< +\infty} \biggl\vert \int_{\lambda _{n}}^{x}e^{ity}\,d\alpha(y) \biggr\vert , $$
if
$$ \limsup_{n\rightarrow+\infty}\frac{\log A_{n}^{\ast}}{\lambda_{n}}=0, $$
(6)
it is easy to get
\(\sigma_{u}^{F}=0\), that is,
\(F(s)\) is analytic in the left half plane; if
$$ \limsup_{n\rightarrow+\infty}\frac{\log A_{n}^{\ast}}{\lambda _{n}}=-\infty , $$
(7)
it follows that
\(\sigma_{u}^{F}=+\infty \), that is,
\(F(s)\) is analytic in the whole plane. For convenience, we use
\(\overline{L}_{\beta}\) to be a class of all the functions
\(F(s)\) of the form (
1) which are analytic in the half plane
\(\Re s<\beta\) (
\(-\infty<\beta<\infty\)) and the sequence
\(\{\lambda_{n}\}\) satisfies (
3) and (
5);
\(L_{0}\) to be the class of all the functions
\(F(s)\) of the form (
1) which are analytic in the half plane
\(\Re s<0\) and the sequence
\(\{\lambda_{n}\}\) satisfies (
3), (
5), and (
6); and
\(L_{\infty}\) to be the class of all the functions
\(F(s)\) of the form (
1) which are analytic in the whole plane
\(\Re s<+\infty\) and the sequence
\(\{ \lambda_{n}\}\) satisfies (
3), (
5), and (
7). Thus, if
\(-\infty<\beta<0\) and
\(F(s)\in \overline{L}_{\beta}\), then
\(F(s)\in L_{0}\).
In 1963, Yu [
26] first proved the Valiron–Knopp–Bohr formula of the associated abscissas of bounded convergence, absolute convergence, and uniform convergence of Laplace–Stieltjes transform. Moreover, Yu [
26] also estimated the growth of the maximal molecule
\(M_{u}(\sigma,F)\), the maximal term
\(\mu(\sigma ,F)\), by introducing the concepts of the order of
\(F(s)\), and investigated the singular direction–Borel line of entire functions represented by Laplace–Stieltjes transforms converging in the whole complex plane. After his wonderful works, considerable attention has been paid to the value distribution and the growth of analytic functions represented by Laplace–Stieltjes transforms converging in the whole plane or the half plane (see [
1,
3,
4,
6‐
8,
11‐
15,
18‐
25,
27]).
Set
$$\begin{aligned} &\mu(\sigma,F)=\max_{n\in \mathbb{N}}\bigl\{ A_{n}^{*}e^{\lambda_{n}\sigma} \bigr\} \quad(\sigma< 0) , \\ & M_{u}(\sigma,F)=\sup _{0< x< +\infty,-\infty< t< +\infty} \biggl\vert \int _{0}^{x}e^{(\sigma+it)y}\,d\alpha(y) \biggr\vert \quad(\sigma< 0). \end{aligned} $$
For
\(F(s)\in L_{0}\), in view of
\(M_{u}(\sigma,F)\rightarrow+\infty\) as
\(\sigma\rightarrow0^{-}\), the concepts of order and type can be usually used in estimating the growth of
\(F(s)\) precisely.
In 2012 and 2014, Luo and Kong [
9,
10] investigated the growth of Laplace–Stieltjes transform converging on the whole plane and obtained the following.
In this paper, the first aim is to investigate the growth of analytic functions represented by Laplace–Stieltjes transforms with generalized order converging in the half plane, and we obtain some theorems about the generalized order
\(A_{n}^{*}\) and
\(\lambda_{n}\), which are improvements of the previous results given by Luo and Kong [
9,
10]. To state our results, we first introduce the following notations and definitions.
Let Γ be a class of continuous increasing functions \(\mathscr {A}\) such that \(\mathscr{A}(x)\geq0\) for \(x\geq x_{0}\), \(\mathscr {A}(x)=\mathscr{A}(x_{0})\) for \(x\leq x_{0}\) and on \([x_{0},+\infty)\) the function \(\mathscr{A}\) increases to +∞; and \(\Gamma^{0}\) be a class such that \(\Gamma^{0}\subset\Gamma\) and \(\mathscr {A}(x(1+o(1))=(1+o(1))\mathscr{A}(x)\) as \(x\rightarrow+\infty\), for \(\mathscr{A}\in\Gamma^{0}\); further, \(\mathscr{A}\in\Gamma^{0i}\) if \(\mathscr{A}\in\Gamma\) and for any \(\eta>0\), \(\mathscr{A}(\eta x)=(1+o(1))\mathscr{A}(x)\) as \(x\rightarrow+\infty\). Obviously, it follows \(\Gamma^{0i}\subset\Gamma^{0}\) and \(h(x)\in\Gamma\).
2 Results and discussion
For generalized order of Laplace–Stieltjes transform (
1), we obtain the following.
If Laplace–Stieltjes transform (
1) satisfies
\(A^{*}_{n}=0\) for
\(n\geq k+1\) and
\(A^{*}_{k}\neq0\), then
\(F(s)\) will be said to be an exponential polynomial of degree
k usually denoted by
\(p_{k}\),
i.e.,
\(p_{k}(s)=\int^{\lambda_{k}}_{0}\exp(sy)\,d\alpha(y)\). If we choose a suitable function
\(\alpha(y)\), the function
\(p_{k}(s)\) may be reduced to a polynomial in terms of
\(\exp(s\lambda_{i})\), that is,
\(\sum_{i=1}^{k}b_{i}\exp(s\lambda_{i})\). We denote
\(\Pi_{k}\) to be the class of all exponential polynomials of degree almost
k, that is,
$$\Pi_{k}= \Biggl\{ \sum_{i=1}^{k} b_{i}\exp(s\lambda_{i}): (b_{1},b_{2}, \ldots ,b_{k})\in \mathbb{C}^{k} \Biggr\} . $$
For
\(F(s)\in\overline{L}_{\beta}\),
\(-\infty<\beta<0\), we denote by
\(E_{n}(F,\beta)\) the error in approximating the function
\(F(s)\) by exponential polynomials of degree
n in uniform norm as
$$E_{n}(F,\beta)=\inf_{p\in\Pi_{n}}\Vert F-p \Vert_{\beta}, \quad n=1,2,\ldots, $$
where
$$\Vert F-p\Vert_{\beta}=\max_{-\infty< t< +\infty} \bigl\vert F(\beta +it)-p(\beta+it) \bigr\vert . $$
In 2017, Singhal and Srivastava [
17] studied the approximation of Laplace–Stieltjes transforms of finite order converging on the whole plane and obtained the following theorem.
In the same year, the author and Kong [
20] investigated the approximation of Laplace–Stieltjes transform
\(F(s)\in L_{0}\) with infinite order and obtained the following.
The second purpose of this paper is to study the approximation of Laplace–Stieltjes transform \(F(s)\in L_{0}\) with generalized order, and our results are listed as follows.
4 Methods
To prove our results, we also need to give the following lemmas (see [
16]).
Let
\(\Xi_{0}\) denote the set of positive unbounded functions
ϕ on
\((-\infty,0)\) such that the derivative
\(\phi'\) is positive, continuous, and increasing to +∞ on
\((-\infty,0)\). Thus, if
\(\phi\in\Xi_{0}\), then
\(\phi(x)\rightarrow\zeta\geq0\) and
\(\phi '(x)\rightarrow0\) as
\(x\rightarrow-\infty\). Let
φ be the inverse function of
\(\phi'\), then
φ is continuous on
\((0,+\infty)\) and increases to 0. Set
\(\phi\in\Xi_{0}\) and
\(\psi (x)=x-\frac{\phi(x)}{\phi'(x)}\). For
\(-\infty< x< x+\iota<0\), since
\(\phi'\) is increasing on
\((-\infty,0)\), we have
$$\begin{aligned} \phi'(x)\phi(x+\iota)-\phi'(x+\iota)\phi(x)&< \phi'(x)\bigl[\phi (x+\iota)-\phi(x)\bigr] =\phi'(x) \int_{x}^{x+\iota}\phi'(t)\,dt \\ &< (x+\iota-x)\phi'(x)\phi'(x+\iota), \end{aligned}$$
that is,
$$\psi(x)=x-\frac{\phi(x)}{\phi'(x)}< x+\iota-\frac{\phi(x+\iota )}{\phi'(x+\iota)}=\psi(x+\iota). $$
Thus, it means that
ψ is an increasing function on
\((-\infty,0)\).
Next, we will prove that
\(\psi(x)\rightarrow0\) as
\(x\rightarrow0\), that is, there is no constant
\(\eta<0\) such that
\(\psi(x)\leq\eta\) for all
\(x\in(-\infty,0)\). Assume that there exist two constants
η,
\(K_{1}\) such that
\(\psi(x)\leq\eta\) for all
\(x\in(-\infty,0)\) and
\(\eta< K_{1}<0\). Since
ψ is an increasing function and
\(\psi (x)< x<0\), then it follows
\(\frac{\phi'(x)}{\phi(x)}\leq\frac{1}{x-\eta}\) for
\(K_{1}\leq x<0\). Thus, it follows
$$\begin{aligned} \log\phi(x)&=\log\phi(K_{1})+ \int_{K_{1}}^{x}\frac{\phi'(t)}{\phi (t)}\,dt\leq\log \phi(K_{1})+ \int_{K_{1}}^{x}\frac{1}{t-\eta}\,dt \\ &=\log\phi(K_{1})+\log\frac{x-\eta}{K_{1}-\eta}. \end{aligned}$$
Hence
\(\phi(x)\leq\phi(K_{1})\frac{x-\eta}{K_{1}-\eta}\). In view of
\(\phi'(x)\rightarrow+\infty\) (
\(x\rightarrow0\)), we get a contradiction. Thus, it follows
\(\psi(x)\rightarrow0\) as
\(x\rightarrow0\).
Besides, let \(\psi^{-1}\) be the inverse function of ψ. Then \(\psi^{-1}\) is an increasing function on \((-\infty, 0)\) and \(\phi '(\psi^{-1}(\sigma))\) increases to +∞ on \((-\infty,0)\).
4.1 Proofs of Theorems 2.1 and 2.2
4.1.1 The proof of Theorem 2.1
Suppose that
\(\rho:=\rho_{\mathscr{A}\mathscr{B}}(F)<+\infty \) and
$$\vartheta=\limsup_{n\rightarrow+\infty}\frac{\mathscr{A}(\lambda _{n})}{\mathscr{B} (\frac{\lambda_{n}}{\log A_{n}^{*}} )}. $$
In view of the definition of generalized order and Lemma
4.2, for any
\(\varepsilon>0\), there exists a constant
\(\sigma_{0}<0\) such that, for all
\(0>\sigma>\sigma_{0}\),
$$ \log\mu(\sigma,F)\leq\mathscr{A}^{-1} \biggl((\rho+\varepsilon ) \mathscr{B} \biggl(-\frac{1}{\sigma} \biggr) \biggr)+\log p, $$
that is,
$$ \log A_{n}^{*}\leq\mathscr{A}^{-1} \biggl(( \rho+\varepsilon)\mathscr {B} \biggl(-\frac{1}{\sigma} \biggr) \biggr)- \lambda_{n}\sigma+\log p, \quad n\geq0. $$
(16)
Choosing
$$-\frac{1}{\sigma}=\mathscr{B}^{-1} \biggl(\frac{1}{\rho+\varepsilon } \mathscr{A} \biggl(\frac{\lambda_{n}}{\mathscr{B}^{-1} (\frac {\mathscr{A}(\lambda_{n})}{\rho+\varepsilon} )} \biggr) \biggr), $$
we conclude from (
9) and (
16) that
$$\begin{aligned} \log A_{n}^{*}&\leq\frac{\lambda_{n}}{\mathscr{B}^{-1} (\frac{\mathscr{A}(\lambda_{n})}{\rho+\varepsilon} )}+\frac {\lambda_{n}}{\mathscr{B}^{-1} (\frac{1}{\rho+\varepsilon} \mathscr{A} (\frac{\lambda_{n}}{\mathscr{B}^{-1} (\frac {\mathscr{A}(\lambda_{n})}{\rho+\varepsilon} )} ) )}+\log p \\ &=\frac{(1+o(1))\lambda_{n}}{\mathscr{B}^{-1} ((1+o(1))\frac {\mathscr{A}(\lambda_{n})}{\rho+\varepsilon} )},\quad \mbox{as } n\rightarrow+\infty , \end{aligned}$$
which implies
$$ \mathscr{A}(\lambda_{n})\leq(\rho+\varepsilon) \mathscr{B} \biggl(\frac{(1+o(1))\lambda_{n}}{\log A_{n}^{*}} \biggr), \quad\mbox{as } n\rightarrow+\infty . $$
(17)
Since
\(\mathscr{A}\in\Gamma^{0i}\),
\(\mathscr{B}\in\Gamma^{0i}\) and let
\(\varepsilon\rightarrow0^{+}\), we can conclude from (
17) that
\(\vartheta\leq\rho\).
Assume
\(\vartheta< \rho\), then we can choose a constant
\(\rho_{1}\) such that
\(\vartheta<\rho_{1}<\rho\). Since
\(\mathscr{B}^{-1} (\frac {\mathscr{A}(x)}{\rho_{1}} )\) is an increasing function, then there exists a positive integer
\(n_{0}\) such that, for
\(n\geq n_{0}\),
$$ \log A_{n}^{*}\leq\frac{\lambda_{n}}{\mathscr{B}^{-1} (\frac {\mathscr{A}(\lambda_{n})}{\rho_{1}} )}\leq \int_{\lambda _{n_{0}}}^{\lambda_{n}}\frac{1}{\mathscr{B}^{-1} (\frac{\mathscr {A}(t)}{\rho_{1}} )}\,dt+K_{1}, $$
(18)
where here and further
\(K_{j}\) is a constant.
Since
\(\phi\in\Xi_{0}\), and let
$$\phi(\sigma)= \int_{-\frac{1}{\sigma_{0}}}^{-\frac{1}{\sigma}}\frac {\mathscr{A}^{-1}(\rho_{1}\mathscr{B}(t))}{t^{2}}\,dt+K_{2} \quad \mbox{for } 0>\sigma\geq\sigma_{0}. $$
Then it follows
$$\begin{aligned} &\phi'(\sigma)=\mathscr{A}^{-1}\biggl(\rho_{1} \mathscr{B}\biggl(-\frac{1}{\sigma }\biggr)\biggr), \qquad \varphi( \lambda_{n})=-\frac{1}{\mathscr{B}^{-1} (\frac{\mathscr{A}(\lambda_{n})}{\rho_{1}} )}, \\ &\bigl[\lambda_{n}\psi\bigl(\varphi(\lambda_{n})\bigr) \bigr]'=\bigl[\lambda_{n}\varphi(\lambda _{n})-\phi\bigl(\varphi(\lambda_{n})\bigr) \bigr]'=\varphi(\lambda_{n}), \end{aligned} $$
and
$$ -\lambda_{n}\psi\bigl(\varphi(\lambda_{n}) \bigr)=- \int_{\lambda_{n_{0}}}^{\lambda _{n}}\varphi(t)\,dt+K_{2}= \int_{\lambda_{n_{0}}}^{\lambda_{n}}\frac {1}{\mathscr{B}^{-1} (\frac{\mathscr{A}(t)}{\rho_{1}} )}\,dt+K_{2}. $$
(19)
Thus, in view of (
18) and (
19), it follows
$$\begin{aligned} \log\mu(\sigma,F)&\leq\phi(\sigma)= \int_{-\frac {1}{\sigma_{0}}}^{-\frac{1}{\sigma}}\frac{\mathscr{A}^{-1}(\rho _{1}\mathscr{B}(t))}{t^{2}}\,dt+K_{2} \\ &\leq\mathscr{A}^{-1} \biggl(\rho_{1}\mathscr{B} \biggl(-\frac {1}{\sigma} \biggr) \biggr) \int_{-\frac{1}{\sigma_{1}}}^{-\frac {1}{\sigma}}\frac{dt}{t^{2}}+K_{3} \\ &\leq-\sigma_{1} \mathscr{A}^{-1} \biggl( \rho_{1}\mathscr{B} \biggl(-\frac{1}{\sigma} \biggr) \biggr)+K_{3}. \end{aligned}$$
(20)
Since
\(\mathscr{A}\in\Gamma^{0i}\), in view of (
10), (
20) and by applying Lemma
4.2, we can deduce
\(\rho_{\mathscr {A}\mathscr{B}}(F)\leq\rho_{1}\), which implies a contradiction with
\(\rho_{\mathscr{A}\mathscr {B}}(F)>\rho_{1}\). Hence
\(\vartheta=\rho_{\mathscr{A}\mathscr{B}}(F)\).
If
\(\rho_{\mathscr{A}\mathscr{B}}(F)=+\infty \), by using the same argument as above, it is easy to prove that the conclusion is true. Therefore, this completes the proof of Theorem
2.1.
4.1.2 The proof of Theorem 2.2
Suppose that
\(\rho:=\rho_{\mathscr{A}\mathscr{B}}(F)<+\infty \) and
$$\vartheta_{1}=\limsup_{n\rightarrow+\infty}\frac{\mathscr{A}(\log A_{n}^{*})}{\mathscr{B} (\lambda_{n} )}. $$
In view of the definition of generalized order and Lemma
4.2, for any
\(\varepsilon>0\), there exists a constant
\(\sigma_{0}<0\) such that, for all
\(0>\sigma>\sigma_{0}\),
$$ \log\mu(\sigma,F)\leq\mathscr{A}^{-1} \biggl((\rho+\varepsilon ) \mathscr{B} \biggl(-\frac{1}{\sigma} \biggr) \biggr)+\log p, $$
that is,
$$ \log A_{n}^{*}\leq\mathscr{A}^{-1} \biggl(( \rho+\varepsilon)\mathscr {B} \biggl(-\frac{1}{\sigma} \biggr) \biggr)- \lambda_{n}\sigma+\log p, \quad n\geq0. $$
(21)
Choosing
\(-\frac{1}{\sigma}=\lambda_{n}\), we conclude from (
21) that
$$\begin{aligned} \log A_{n}^{*}&\leq\mathscr{A}^{-1}\bigl(( \rho+\varepsilon)\mathscr {B}(\lambda_{n})\bigr)+1+\log p \\ &\leq \bigl(1+o(1)\bigr)\mathscr{A}^{-1}\bigl((\rho +\varepsilon)\mathscr{B}( \lambda_{n})\bigr),\quad \mbox{as } n\rightarrow +\infty . \end{aligned}$$
(22)
Since
\(\mathscr{A}\in\Gamma^{0i}\),
\(\mathscr{B}\in\Gamma^{0i}\) and let
\(\varepsilon\rightarrow0^{+}\), we can conclude from (
22) that
\(\vartheta_{1}\leq\rho\).
Assume
\(\vartheta_{1}< \rho\), then we can choose a constant
\(\rho_{2}\) such that
\(\vartheta_{1}<\rho_{2}<\rho\). It means that there exists a positive integer
\(n_{0}\) such that, for
\(n\geq n_{0}\),
$$\log A_{n}^{*}\leq\mathscr{A}^{-1}\bigl(\rho_{2} \mathscr{B}(\lambda_{n})\bigr), $$
that is,
$$ \log\mu(\sigma,F)\leq\max\bigl\{ \mathscr{A}^{-1}\bigl( \rho_{2}\mathscr {B}(\lambda_{n})\bigr)+\lambda_{n} \sigma: n\geq n_{0}\bigr\} +K_{5}. $$
(23)
In view of (
10), the following equation
$$\mathscr{A}^{-1}\bigl(\rho_{2}\mathscr{B}(t)\bigr)+t\sigma=0 $$
has a unique solution
\(t_{1}:=t(\sigma)\) such that
\(t_{1}\uparrow+\infty \) as
\(\sigma\rightarrow0^{-}\), and for
\(t\geq t_{1}\) we can deduce that
\(\mathscr{A}^{-1}(\rho_{2}\mathscr{B}(t))+t\sigma\leq0\). Hence, it follows
$$\begin{aligned} \log\mu(\sigma,F)&\leq\max\bigl\{ \mathscr{A}^{-1}\bigl(\rho _{2}\mathscr{B}(t)\bigr)+t\sigma: t_{0}\leq t\leq t_{1}\bigr\} +K_{6} \\ &\leq\mathscr{A}^{-1}\bigl(\rho_{2}\mathscr{B}(t_{1}) \bigr)+K_{6}. \end{aligned}$$
(24)
In view of
$$-\frac{1}{\sigma}=\frac{t_{1}}{\mathscr{A}^{-1}(\rho_{2}\mathscr{B}(t_{1}))}, $$
it follows from (
12) that
$$ \mathscr{B} \biggl(-\frac{1}{\sigma} \biggr)=\mathscr{B} \biggl(\frac {t_{1}}{\mathscr{A}^{-1}(\rho_{2}\mathscr{B}(t_{1}))} \biggr)=\bigl(1+o(1)\bigr)\mathscr{B}(t_{1}), \quad\sigma\rightarrow0^{-}. $$
(25)
Hence, we can deduce from (
24) and (
25) that
$$\log\mu(\sigma,F)\leq\mathscr{A}^{-1} \biggl(\rho _{2} \bigl(1+o(1)\bigr)\mathscr{B} \biggl(-\frac{1}{\sigma} \biggr) \biggr),\quad \mbox{as } \sigma\rightarrow0^{-}, $$
which implies
\(\rho_{\mathscr{A}\mathscr{B}}(F)\leq\rho_{2}<\rho _{\mathscr{A}\mathscr{B}}(F)\) by combining Lemma
4.2 and (
10), a contradiction. Therefore,
\(\vartheta_{1}=\rho_{\mathscr {A}\mathscr{B}}(F)\).
If
\(\rho_{\mathscr{A}\mathscr{B}}(F)=+\infty \), by using the same argument as above, it is easy to prove that the conclusion is true. Therefore, this completes the proof of Theorem
2.2.
4.2 Proofs of Theorems 2.5 and 2.6
4.2.1 The proof of Theorem 2.5
Suppose that
\(\rho:=\rho_{\mathscr{A}\mathscr{B}}(F)<+\infty \) and
$$\vartheta_{3}=\limsup_{n\rightarrow+\infty}\frac{\mathscr {A}(\lambda_{n})}{\mathscr{B} (\frac{\lambda_{n}}{\log [E_{n-1}(F,\beta)\exp\{-\beta\lambda_{n}\} ]} )}. $$
In view of the definition of generalized order and Lemma
4.2, for any
\(\varepsilon>0\), there exists a constant
\(\sigma_{0}<0\) such that, for all
\(0>\sigma>\sigma_{0}\),
$$ \log M_{u}(\sigma,F)\leq\mathscr{A}^{-1} \biggl((\rho+\varepsilon )\mathscr{B} \biggl(-\frac{1}{\sigma} \biggr) \biggr). $$
(26)
Since
\(F(s)\in L_{0}\), and for any constant
β (
\(-\infty<\beta<0\)), then
\(F(s)\in \overline{L}_{\beta}\). Hence, for
\(\beta<\sigma<0\) and
\(p_{k}\in\Pi _{k}\), it follows
$$\begin{aligned} E_{k}(F,\beta) &\leq\Vert F-p_{k} \Vert_{\beta}\leq \bigl\vert F(\beta +it)-p_{k}(\beta+it) \bigr\vert \\ &\leq \biggl\vert \int_{0}^{+\infty}\exp\{sy\}\,d\alpha(y)- \int _{0}^{\lambda_{k}}\exp\{sy\}\,d\alpha(y) \biggr\vert = \biggl\vert \int_{\lambda_{k}}^{\infty}\exp\{sy\}\,d\alpha(y) \biggr\vert . \end{aligned}$$
(27)
Let
$$I_{j+k}(b;it)= \int_{\lambda_{j+k}}^{b}\exp\{ity\}\,d\alpha(y) \quad ( \lambda_{j+k}< b\leq\lambda_{j+k+1}), $$
then
\(|I_{j+k}(b;it)|\leq A_{j+k}^{*}\). In view of
$$\biggl\vert \int_{\lambda_{k}}^{\infty}\exp\bigl\{ (\beta+it)y\bigr\} \,d\alpha (y) \biggr\vert =\lim_{b\rightarrow+\infty} \biggl\vert \int_{\lambda _{k}}^{b}\exp\bigl\{ (\beta+it)y\bigr\} \,d \alpha(y) \biggr\vert , $$
where
\(-\infty <\beta<0\), hence
$$\begin{aligned} & \biggl\vert \int_{\lambda_{k}}^{b}\exp\bigl\{ (\beta+it)y\bigr\} \,d\alpha (y) \biggr\vert \\ &\quad = \Biggl\vert \sum_{j=k}^{n+k-1} \int_{\lambda_{j}}^{\lambda _{j+1}}\exp\{\beta y\}d_{y}I_{j}(y;it)+ \int_{\lambda_{n+k}}^{b}\exp\{\beta y\}d_{y}I_{n+k}(y;it) \Biggr\vert \\ &\quad = \Biggl\vert \Biggl[\sum_{j=k}^{n+k-1}e^{\lambda_{j+1}\beta }I_{j}( \lambda_{j+1};it)-\beta \int_{\lambda_{j}}^{\lambda _{j+1}}e^{\beta y}I_{j}(y;it) \,dy \Biggr] \\ &\qquad{}+e^{\beta b}I_{n+k}(b;it)-\beta \int_{\lambda _{n+k}}^{b}e^{\beta y}I_{j}(y;it) \,dy \Biggr\vert \\ &\quad \leq \sum_{j=k}^{n+k-1} \bigl[A_{j}^{*}e^{\lambda_{j+1}\beta }+A_{j}^{*} \bigl(e^{\lambda_{j+1}\beta}-e^{\lambda_{j}\beta}\bigr) \bigr] +2e^{\beta\lambda_{n+k+1}}A_{n+k}^{*}-e^{\beta\lambda _{n+k}}A_{n+k}^{*} \\ &\quad \leq 2\sum_{j=k}^{n+k} A_{n}^{*}e^{\lambda_{n+1}\beta}. \end{aligned}$$
Therefore, we conclude
$$ \biggl\vert \int_{\lambda_{k}}^{\infty}\exp\bigl\{ (\beta+it)y\bigr\} \,d\alpha (y) \biggr\vert \leq2\sum_{n=k}^{+\infty}A_{n}^{*} \exp\{\beta\lambda _{n+1}\}, \quad \mbox{as } n\rightarrow+\infty . $$
(28)
In view of Lemma
4.2, it follows
\(A_{n}^{*}\leq p M_{u}(\sigma ,F)e^{-\sigma\lambda_{n}}\). So, for any
σ (
\(\beta<\sigma<0\)), it yields from (
27) and (
28) that
$$ E_{n}(F,\beta)\leq2\sum_{k=n+1}^{\infty}A_{k-1}^{*}\exp\{\beta\lambda _{k}\}\leq2pM_{u}( \sigma,F)\sum_{k=n+1}^{\infty}\exp\bigl\{ (\beta- \sigma )\lambda_{k}\bigr\} . $$
(29)
In view of (
5), we can choose
\(h'\) (
\(0< h'< h\)) such that
\((\lambda_{n+1} -\lambda_{n})\geq h'\) for
\(n\geq0\). Then, for
\(\sigma\geq \frac{\beta}{2}\), it follows from (
29) that
$$\begin{aligned} E_{n}(F,\beta) &\leq M_{u}(\sigma,F)\exp\bigl\{ \lambda_{n+1}(\beta-\sigma)\bigr\} \sum_{k=n+1}^{\infty}\exp\bigl\{ (\lambda_{k}-\lambda_{n+1}) (\beta-\sigma)\bigr\} \\ &\leq M_{u}(\sigma,F)\exp\bigl\{ \lambda_{n+1}(\beta-\sigma) \bigr\} \exp \biggl\{ -\frac{\beta}{2}h'(n+1) \biggr\} \sum _{k=n+1}^{\infty}\exp \biggl\{ \frac{\beta}{2}h'k \biggr\} \\ &=M_{u}(\sigma,F)\exp\bigl\{ \lambda_{n+1}(\beta-\sigma)\bigr\} \biggl(1-\exp\biggl\{ \frac{\beta}{2}h'\biggr\} \biggr)^{-1}, \end{aligned}$$
that is,
$$ E_{n-1}(F,\beta)\leq KM_{u}(\sigma,F)\exp\bigl\{ \lambda_{n}(\beta-\sigma )\bigr\} , $$
(30)
where
K is a constant. Hence, it follows from (
26) and (
30) that
$$ \log\bigl[E_{n-1}(F,\beta)\exp\{-\beta \lambda_{n}\}\bigr]\leq\mathscr {A}^{-1} \biggl((\rho+ \varepsilon)\mathscr{B} \biggl(-\frac{1}{\sigma } \biggr) \biggr)- \lambda_{n}\sigma+\log K, \quad n\geq0. $$
(31)
Let
$$-\frac{1}{\sigma}=\mathscr{B}^{-1} \biggl(\frac{1}{\rho+\varepsilon } \mathscr{A} \biggl(\frac{\lambda_{n}}{\mathscr{B}^{-1} (\frac {\mathscr{A}(\lambda_{n})}{\rho+\varepsilon} )} \biggr) \biggr), $$
we conclude from (
9) and (
31) that
$$\begin{aligned} \log\bigl[E_{n-1}(F,\beta)\exp\{-\beta\lambda_{n}\} \bigr]&\leq\frac {\lambda_{n}}{\mathscr{B}^{-1} (\frac{\mathscr{A}(\lambda _{n})}{\rho+\varepsilon} )}+\frac{\lambda_{n}}{\mathscr {B}^{-1} (\frac{1}{\rho+\varepsilon} \mathscr{A} (\frac{\lambda_{n}}{\mathscr{B}^{-1} (\frac {\mathscr{A}(\lambda_{n})}{\rho+\varepsilon} )} ) )}+\log K \\ &=\frac{(1+o(1))\lambda_{n}}{\mathscr{B}^{-1} ((1+o(1))\frac {\mathscr{A}(\lambda_{n})}{\rho+\varepsilon} )}, \quad\mbox{as } n\rightarrow+\infty , \end{aligned}$$
which implies
$$ \mathscr{A}(\lambda_{n})\leq(\rho+\varepsilon) \mathscr{B} \biggl(\frac{(1+o(1))\lambda_{n}}{\log[E_{n-1}(F,\beta)\exp\{-\beta \lambda_{n}\}]} \biggr), \quad \mbox{as } n\rightarrow+ \infty . $$
(32)
Since
\(\mathscr{A}\in\Gamma^{0i}\),
\(\mathscr{B}\in\Gamma^{0i}\) and let
\(\varepsilon\rightarrow0^{+}\), we can conclude from (
32) that
\(\vartheta_{3}\leq\rho\).
Assume
\(\vartheta_{3}< \rho\), then we can choose a constant
\(\rho_{3}\) such that
\(\vartheta_{3}<\rho_{3}<\rho\). Since
\(\mathscr{B}^{-1} (\frac {\mathscr{A}(x)}{\rho_{3}} )\) is an increasing function, then there exists a positive integer
\(n_{0}\) such that, for
\(n\geq n_{0}\),
$$ \log\bigl[E_{n-1}(F,\beta)\exp\{-\beta \lambda_{n}\}\bigr]\leq\frac{\lambda _{n}}{\mathscr{B}^{-1} (\frac{\mathscr{A}(\lambda_{n})}{\rho _{3}} )}\leq \int_{\lambda_{n_{0}}}^{\lambda_{n}}\frac {1}{\mathscr{B}^{-1} (\frac{\mathscr{A}(t)}{\rho_{3}} )}\,dt+K_{1}. $$
(33)
For any
\(\beta<0\), then there exists
\(p_{1}\in\Pi_{n-1}\) such that
$$ \Vert F-p_{1} \Vert \leq2E_{n-1}(F,\beta). $$
(34)
And since
$$\begin{aligned} A_{n}^{*}\exp\{\beta\lambda_{n}\}& =\sup _{\lambda_{n}< x\leq\lambda_{n+1},-\infty< t< +\infty } \biggl\vert \int_{\lambda_{n}}^{x}\exp\{ity\}\,d\alpha(y) \biggr\vert \exp\{\beta\lambda_{n}\} \\ &\leq\sup_{\lambda_{n}< x\leq\lambda_{n+1},-\infty< t< +\infty } \biggl\vert \int_{\lambda_{n}}^{x}\exp\bigl\{ (\beta+it)y\bigr\} \,d \alpha(y) \biggr\vert \\ &\leq\sup_{-\infty< t< +\infty} \biggl\vert \int_{\lambda _{n}}^{\infty}\exp\bigl\{ (\beta+it)y\bigr\} \,d \alpha(y) \biggr\vert , \end{aligned}$$
thus, for any
\(p\in\Pi_{n-1}\), it follows
$$ A_{n}^{*}\exp\{\beta\lambda_{n}\}\leq \bigl\vert F(\beta+it)-p(\beta +it) \bigr\vert \leq \Vert F-p \Vert _{\beta}. $$
(35)
Hence, for any
\(\beta<0\) and
\(F(s)\in L_{0}\), it follows from (
34) and (
35) that
$$ A_{n}^{*}\leq2E_{n-1}(F,\beta)\exp\{-\beta \lambda_{n}\}. $$
(36)
Hence, (
18) follows from (
33) and (
36).
Since
\(\phi\in\Xi_{0}\), and let
$$\phi(\sigma)= \int_{-\frac{1}{\sigma_{0}}}^{-\frac{1}{\sigma}}\frac {\mathscr{A}^{-1}(\rho_{1}\mathscr{B}(t))}{t^{2}}\,dt+K_{2} \quad \mbox{for } 0>\sigma\geq\sigma_{0}, $$
and
$$\varphi(\lambda_{n})=-\frac{1}{\mathscr{B}^{-1} (\frac{\mathscr {A}(\lambda_{n})}{\rho_{1}} )}. $$
By using the same argument as in the proof of Theorem
2.1, we conclude
$$\begin{aligned} \log\mu(\sigma,F)&\leq\phi(\sigma)= \int_{-\frac {1}{\sigma_{0}}}^{-\frac{1}{\sigma}}\frac{\mathscr{A}^{-1}(\rho _{1}\mathscr{B}(t))}{t^{2}}\,dt+K_{2} \\ &\leq-\sigma_{1} \mathscr{A}^{-1} \biggl( \rho_{1}\mathscr{B} \biggl(-\frac{1}{\sigma} \biggr) \biggr)+K_{3}. \end{aligned}$$
(37)
Since
\(\mathscr{A}\in\Gamma^{0i}\), in view of (
10), (
37) and by applying Lemma
4.2, we can deduce
\(\rho_{\mathscr {A}\mathscr{B}}(F)\leq\rho_{3}\), which implies a contradiction with
\(\rho_{\mathscr{A}\mathscr {B}}(F)>\rho_{1}\). Hence
\(\vartheta_{3}=\rho_{\mathscr{A}\mathscr{B}}(F)\).
If
\(\rho_{\mathscr{A}\mathscr{B}}(F)=+\infty \), by using the same argument as above, it is easy to prove that the conclusion is true. Therefore, this completes the proof of Theorem
2.5.
4.2.2 The proof of Theorem 2.6
By combining the arguments as in the proofs of Theorems
2.2 and
2.5, we can easily prove the conclusion of Theorem
2.6.
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.