Open Access

On approximating the error function

Journal of Inequalities and Applications20162016:311

https://doi.org/10.1186/s13660-016-1261-3

Received: 12 August 2016

Accepted: 22 November 2016

Published: 29 November 2016

Abstract

In the article, we present the necessary and sufficient condition for the parameter p on the interval \((7/5, \infty)\) such that the function \(x\rightarrow\operatorname{erf}(x)/B_{p}(x)\) is strictly increasing (decreasing) on \((0, \infty)\), and find the best possible parameters p, q on the interval \((7/5, \infty)\) such that the double inequality \(B_{p}(x)<\operatorname{erf}(x)<B_{q}(x)\) holds for all \(x>0\), where \(\operatorname{erf}(x)=2\int_{0}^{x}e^{-t^{2}}\,dt/\sqrt{\pi}\) is the error function, \(B_{p}(x)=\sqrt{1-\lambda(p)e^{-px^{2}}-[1-\lambda(p)]e^{-\mu(p)x^{2}}}\), \(\lambda(p)=16(5p-7)/[(15p^{2}-40p+28)(45p^{2}-60p-4)]\) and \(\mu(p)=4(5p-7)/[5(3p-4)]\).

Keywords

error function monotonicity bound

MSC

33B20 26D15 26A48

1 Introduction

It is well known that the error function
$$ \operatorname{erf}(x)=\frac{2}{\sqrt{\pi}} \int _{0}^{x}e^{-t^{2}}\,dt=\frac {1}{\sqrt{\pi }} \sum_{n=0}^{\infty}\frac{(-1)^{n}}{n! (n+\frac{1}{2} )}x^{2n+1} $$
has numerous applications in probability, statistics, and partial differential equations theory. Recently, the bounds for the error function have attracted the attention of many researchers. In particular, many remarkable inequalities for the error function can be found in the literature [113].
Pólya [14] proved that the inequality
$$ \operatorname{erf}(x)< \sqrt{1-e^{-4x^{2}/\pi}} $$
holds for all \(x>0\).
In [15], Chu proved that the double inequality
$$ \sqrt{1-e^{-px^{2}}}< \operatorname{erf}(x)< \sqrt{1-e^{-qx^{2}}} $$
(1.1)
holds for all \(x>0\) if and only if \(p\in(0, 1]\) and \(q\in[4/\pi, \infty)\).
Alzer [16] presented the double inequality
$$ \bigl(1-e^{-\beta(p) x^{p}} \bigr)^{1/p}< \frac{1}{\Gamma (1+\frac{1}{p} )} \int _{0}^{x}e^{-t^{p}}\,dt< \bigl(1-e^{-\alpha(p) x^{p}} \bigr)^{1/p} $$
for \(x>0\) and \(p>0\) with \(p\neq1\), where \(\Gamma(x)=\int_{0}^{\infty}t^{x-1}e^{-t}\,dt\) is the classical gamma function, and \(\alpha(p)\) and \(\beta(p)\) are, respectively, given by
$$ \alpha(p)=\frac{1}{\Gamma^{p} (1+\frac{1}{p} )} \quad (p>1),\qquad \alpha(p)=1\quad (0< p< 1), $$
and
$$ \beta(p)=\frac{1}{\Gamma^{p} (1+\frac{1}{p} )} \quad (0< p< 1),\qquad \beta(p)=1\quad (p>1). $$
Let \(n\geq2\), and \(\alpha_{n}\), \(\beta_{n}\), \(\alpha^{\ast}_{n}\), \(\beta^{\ast}_{n}\) be, respectively, defined by
$$\begin{aligned}& \alpha_{2}=0.90686\ldots,\qquad \alpha_{n}=1 \quad (n\geq3), \qquad \beta_{n}=n-1, \\& \alpha^{\ast}_{n}=n+1 \quad (n=2k),\qquad \alpha^{\ast}_{n}=n-1\quad (n=2k-1), \qquad \beta^{\ast}_{n}=1. \end{aligned}$$
In [17, 18], Alzer proved that the double inequalities
$$\begin{aligned}& \lambda_{n}\operatorname{erf} \Biggl(\sum _{i=1}^{n}x_{i} \Biggr)\leq \sum _{i=1}^{n}\operatorname{erf}(x_{i})-\prod _{i=1}^{n}\operatorname {erf}(x_{i}) \leq \mu_{n}\operatorname{erf} \Biggl(\sum_{i=1}^{n}x_{i} \Biggr), \\& \lambda\operatorname{erf}\bigl(y+\operatorname{erf}(x)\bigr)< \operatorname {erf} \bigl(x+\operatorname{erf}(y)\bigr)< \mu\operatorname{erf}\bigl(y+\operatorname {erf}(x)\bigr), \\& \lambda^{\ast}\operatorname{erf}\bigl(y\operatorname {erf}(x)\bigr)< \operatorname{erf}\bigl(x\operatorname{erf}(y)\bigr)\leq\mu^{\ast } \operatorname{erf} \bigl(y\operatorname{erf}(x)\bigr), \end{aligned}$$
(1.2)
hold for all \(x_{i}\geq0\) and \(y\geq x>0\) if and only if \(\lambda_{n}\leq\alpha_{n}\), \(\mu_{n}\geq\beta_{n}\), \(\lambda \leq\operatorname{erf}(1)=0.8427\ldots\) , \(\mu\geq2/\sqrt{\pi }=1.1283\ldots\) , \(\lambda^{\ast}\leq0\) and \(\mu^{\ast}\geq1\), and inequality (1.2) holds for all \(x_{i}\leq0\) if and only if \(\lambda_{n}\geq \alpha^{\ast}_{n}\) and \(\mu_{n}\leq\beta^{\ast}_{n}\).
Recently, Neuman [19] proved that the double inequality
$$ \frac{2x}{\sqrt{\pi}}e^{-\frac{x^{2}}{3}}\leq\operatorname {erf}(x)\leq \frac {2x}{\sqrt {\pi}}\frac{e^{-x^{2}}+2}{3} $$
(1.3)
holds for all \(x>0\).
Let \(x\in(0, \infty)\), \(p\in(7/5, \infty)\), \(\lambda(p)\), \(\mu(p)\), and \(B_{p}(x)\) be, respectively, defined by
$$\begin{aligned}& \lambda(p)=\frac{16(5p-7)}{(15p^{2}-40p+28)(45p^{2}-60p-4)},\qquad \mu(p)=\frac{4(5p-7)}{5(3p-4)}, \end{aligned}$$
(1.4)
$$\begin{aligned}& B_{p}(x)=\sqrt{1-\lambda(p)e^{-px^{2}}-\bigl[1-\lambda(p) \bigr]e^{-\mu(p)x^{2}}}. \end{aligned}$$
(1.5)
The main purpose of this paper is to present the best possible parameters p and q on the interval \((7/5, \infty)\) such that the double inequality
$$ B_{p}(x)< \operatorname{erf}(x)< B_{q}(x) $$
holds for all \(x>0\).

2 Lemmas

In order to prove our main results, we need to introduce an auxiliary function at first.

Let \(-\infty\leq a< b\leq\infty\), f and g be differentiable on \((a,b)\), and \(g'\neq0\) on \((a,b)\). Then the function \(H_{f, g}\) [20, 21] is defined by
$$ H_{f,g}\equiv\frac{f'}{g'}g-f. $$
(2.1)
It is not difficult to verify that the auxiliary function \(H_{f,g}\) has the following properties:
$$ \biggl(\frac{f}{g} \biggr)'= \frac{g'}{g^{2}}H_{f,g} $$
(2.2)
if \(g\neq0\) on \((a,b)\), and
$$ H_{f,g}'= \biggl(\frac{f'}{g'} \biggr)'g $$
(2.3)
if both f and g are twice differentiable on \((a,b)\).

Lemma 2.1

[20], Theorem 8

Let \(-\infty\leq a< b\leq\infty\), f and g be differentiable on \((a,b)\) with \(f(a^{+})=g(a^{+})=0\), \(g^{\prime}(x)\neq0\) and \(g^{\prime}(x)H_{f, g}(b^{-})< (>)\, 0\) for all \(x\in(a,b)\). If there exists \(\lambda_{0}\in(a, b)\) such that \(f^{\prime}/g^{\prime}\) is strictly increasing (decreasing) on \((a, \lambda_{0})\) and strictly decreasing (increasing) on \((\lambda_{0}, b)\), then there exists \(\mu_{0}\in(a, b)\) such that \(f/g\) is strictly increasing (decreasing) on \((a, \mu_{0})\) and strictly decreasing (increasing) on \((\mu_{0}, b)\).

Lemma 2.2

Let \(p\in(7/5, \infty)\), \(p^{\ast}_{0}=(50+2\sqrt{30})/35=1.74155\ldots\) , \(\lambda(p)\) and \(\mu(p)\) be defined by (1.4), and \(u_{n}\) be defined by
$$ u_{n}=(5p-6) (5p-8)n-\bigl(15p^{2}-40p+28 \bigr). $$
(2.4)
Then the following statements are true:
  1. (1)

    \(p>\mu(p)\), \(0<\lambda(p)\leq(8+\sqrt{14})/16=0.73385\ldots\) and \(0<\mu(p)<4/3\) for \(p\in(7/5, \infty)\), \(0<\mu(p)\leq1\) for \(p\in(7/5, 8/5]\) and \(1<\mu(p)<4/3\) for \(p\in(8/5, \infty)\);

     
  2. (2)

    \(u_{n}<0\) for all \(n\geq2\) if \(p\in(7/5, 8/5]\);

     
  3. (3)

    \(u_{2}\geq0\) and \(u_{n}> 0\) for all \(n\geq3\) if \(p\in [p^{\ast}, \infty)\);

     
  4. (4)

    there exists \(n_{0}\geq2\) such that \(u_{n_{0}+1}\geq0\), \(u_{n}<0\) for \(2\leq n\leq n_{0}\) and \(u_{n}> 0\) for \(n\geq n_{0}+2\) and \(u_{n}>0\) for \(n>n_{0}\) if \(p\in(8/5, p^{\ast}_{0})\).

     

Proof

For part (1), from (1.4) we clearly see that
$$\begin{aligned}& p>\mu(p), \\& \lambda \biggl(\frac{7}{5} \biggr)=\lambda(\infty)=0,\qquad \lambda \biggl(\frac{14+2\sqrt{14}}{15} \biggr)=\frac{8+\sqrt {14}}{16}=0.73385\ldots, \end{aligned}$$
(2.5)
$$\begin{aligned}& \mu \biggl(\frac{7}{5} \biggr)=0, \qquad \mu(\infty)=\frac{4}{3}, \qquad \mu \biggl(\frac{8}{5} \biggr)=1, \end{aligned}$$
(2.6)
$$\begin{aligned}& \lambda^{\prime}(p)=-\frac {80(3p-4)^{2}}{9(15p^{2}-40p+28)^{2}(45p^{2}-60p-4)^{2}} \\& \hphantom{\lambda^{\prime}(p)={}}{}\times\biggl(p-\frac{14-2\sqrt{14}}{15} \biggr) \biggl(p-\frac{14+2\sqrt {14}}{15} \biggr), \end{aligned}$$
(2.7)
$$\begin{aligned}& \mu^{\prime}(p)=\frac{4}{5(3p-4)^{2}}>0 \end{aligned}$$
(2.8)
for \(p>7/5\).

Equation (2.7) implies that \(\lambda(p)\) is strictly increasing on \((7/5, (14+2\sqrt{14})/15]\) and strictly decreasing on \([(14+2\sqrt{14})/15, \infty)\). Therefore, \(0<\lambda(p)\leq(8+\sqrt{14})/16\) for \(p\in(7/5, \infty)\) as follows from (2.5) and the piecewise monotonicity of \(\lambda(p)\) on the interval \((7/5, \infty)\), and the remaining desired results for \(\mu(p)\) follow easily from (2.6) and (2.8).

For parts (2) and (3), let \(x\geq2\), \(p_{1}(x)\) and \(p_{2}(x)\) be defined by
$$ p_{1}(x)=\frac{35x-20-\sqrt{5}\sqrt{5x^{2}+4x-4}}{5(5x-3)},\qquad p_{2}(x)= \frac{35x-20+\sqrt{5}\sqrt{5x^{2}+4x-4}}{5(5x-3)}. $$
Then simple computations lead to
$$\begin{aligned}& p_{1}(2)=\frac{50-2\sqrt{30}}{35}=1.11558\ldots,\qquad p_{1}( \infty)=\frac{6}{5}, \end{aligned}$$
(2.9)
$$\begin{aligned}& p_{2}(2)=p^{\ast}_{0},\qquad p_{2}(3)= \frac{85+\sqrt{265}}{60}=1.68798\ldots,\qquad p_{2}(\infty)= \frac{8}{5}, \end{aligned}$$
(2.10)
$$\begin{aligned}& u_{n}=5(5n-3)\bigl[p-p_{1}(n)\bigr] \bigl[p-p_{2}(n) \bigr], \end{aligned}$$
(2.11)
$$\begin{aligned}& p^{\prime}_{1}(x)=\frac{25x-14-\sqrt{5}\sqrt{5x^{2}+4x-4}}{\sqrt {5}(5x-3)^{2}\sqrt{5x^{2}+4x-4}}>0, \end{aligned}$$
(2.12)
$$\begin{aligned}& p^{\prime}_{2}(x)=-\frac{25x-14+\sqrt{5}\sqrt{5x^{2}+4x-4}}{\sqrt {5}(5x-3)^{2}\sqrt{5x^{2}+4x-4}}< 0 \end{aligned}$$
(2.13)
for \(x\geq2\).
It follows from (2.9)-(2.13) that
$$\begin{aligned}& u_{2}=35\bigl[p-p^{\ast}_{0}\bigr] \biggl[p- \frac{50-2\sqrt{30}}{35} \biggr], \end{aligned}$$
(2.14)
$$\begin{aligned}& \frac{50-2\sqrt{30}}{35}\leq p_{1}(n)< \frac{6}{5},\qquad \frac{8}{5}< p_{2}(n)\leq p^{\ast}_{0} \end{aligned}$$
(2.15)
for \(n\geq2\) and
$$ \frac{8}{5}< p_{2}(n)\leq\frac{85+\sqrt{265}}{60} $$
(2.16)
for \(n\geq3\)

Therefore, parts (2) and (3) follow easily from (2.11) and (2.14)-(2.16).

For part (4), if \(p\in(8/5, p^{\ast}_{0})\), then from (2.4) and (2.14) we clearly see that the sequence \(\{u_{n}\}_{n=2}^{\infty}\) is strictly increasing and
$$ u_{2}< 0,\qquad u_{\infty}=\infty. $$
(2.17)
Therefore, part (4) follows from (2.17) and the monotonicity of the sequence \(\{u_{n}\}_{n=2}^{\infty}\). □

Lemma 2.3

Let \(x\in(0, \infty)\), \(p\in(7/5, \infty)\), \(\lambda(p)\), \(\mu(p)\) and \(B_{p}(x)\) be defined by (1.4) and (1.5), and \(C_{p}(x)\) and \(\alpha(p)\) be defined by
$$ C_{p}(x)=\frac{p\lambda(p)e^{-px^{2}}+\mu(p)(1-\lambda(p))e^{-\mu (p)x^{2}}}{p\lambda(p)+\mu(p)(1-\lambda(p))} $$
and
$$ \alpha(p)=\sqrt{\frac{4}{\pi[p\lambda(p)+\mu(p)(1-\lambda (p))]}}=\sqrt {\frac{45p^{2}-60p-4}{3\pi p(5p-7)}}. $$
(2.18)
Then the following statements are true:
  1. (1)

    the function \(p\rightarrow B_{p}(x)\) is strictly increasing on \((7/5, \infty)\);

     
  2. (2)

    the function \(p\rightarrow C_{p}(x)\) is strictly decreasing on \((7/5, \infty)\);

     
  3. (3)

    the function \(p\rightarrow\alpha(p)B_{p}(x)\) is strictly decreasing on \((7/5, \infty)\).

     

Proof

For part (1), it suffices to show that \(\partial B^{2}_{p}(x)/\partial p>0\) for \(x\in(0, \infty)\) and \(p\in(7/5, \infty)\). Let \(t=(p-\mu(p))x^{2}\) and
$$ F_{1}(t)=-\bigl(p-\mu(p)\bigr)\lambda^{\prime}(p)+ \lambda(p)t+\bigl(p-\mu (p)\bigr)\lambda ^{\prime}(p)e^{t}+ \mu^{\prime}(p) \bigl(1-\lambda(p)\bigr)te^{t}. $$
(2.19)
Then it follows from (1.4), (1.5), (2.19), and Lemma 2.2(1) that
$$\begin{aligned}& \frac{\partial B^{2}_{p}(x)}{\partial p}=\frac{e^{-px^{2}}}{p-\mu(p)}F_{1}(t), \end{aligned}$$
(2.20)
$$\begin{aligned}& p-\mu(p)>0, \quad t>0, \end{aligned}$$
(2.21)
$$\begin{aligned}& F_{1}(0)=0, \end{aligned}$$
(2.22)
$$\begin{aligned}& F^{\prime}_{1}(t)=\lambda(p)+\bigl(p-\mu(p)\bigr) \lambda^{\prime }(p)e^{t}+\mu ^{\prime}(p) \bigl(1- \lambda(p)\bigr)e^{t}+\mu^{\prime}(p) \bigl(1-\lambda(p) \bigr)te^{t}, \\& F^{\prime}_{1}(0)=\frac{12(15p^{2}-40p+28)}{(45p^{2}-60p-4)^{2}}>0, \end{aligned}$$
(2.23)
$$\begin{aligned}& F^{\prime\prime}_{1}(t)=\frac {40(3p-2)(3p-4)}{(45p^{2}-60p-4)^{2}}e^{t}+ \frac {20p(3p-4)}{(15p^{2}-40p+28)(45p^{2}-60p-4)}te^{t}>0 \end{aligned}$$
(2.24)
for \(p\in(7/5, \infty)\) and \(t>0\).
From (2.22)-(2.24) we clearly see that
$$ F_{1}(t)>0 $$
(2.25)
for \(p\in(7/5, \infty)\) and \(t>0\).

Therefore, part (1) follows from (2.20), (2.21), and (2.25).

For part (2), it is enough to prove that \(\partial C_{p}(x)/\partial p<0\) for \(x\in(0, \infty)\) and \(p\in(7/5, \infty)\). Let \(t=(p-\mu(p))x^{2}\) and
$$ F_{2}(t)=-\frac{15p^{2}-40p+28}{p-\mu(p)}t-10(3p-4)+ \biggl[-\frac {15p^{2}-40p+28}{p-\mu(p)}t+10(3p-4) \biggr]e^{t}. $$
Then elaborated computations lead to
$$\begin{aligned}& C_{p}(x)=\frac{1}{3} \biggl[\frac{4e^{-px^{2}}}{15p^{2}-40p+28}+ \frac {5(3p-4)^{2}e^{-\mu(p)x^{2}}}{15p^{2}-40p+28} \biggr], \\& \frac{\partial C_{p}(x)}{\partial p}= -\frac{1}{3} \biggl[\frac{4x^{2}}{15p^{2}-40p+28}+ \frac {40(3p-4)}{(15p^{2}-40p+28)^{2}} \biggr]e^{-px^{2}} \\& \hphantom{\frac{\partial C_{p}(x)}{\partial p}={}}{}+\frac{1}{3} \biggl[-\frac{4x^{2}}{15p^{2}-40p+28}+ \frac {40(3p-4)}{(15p^{2}-40p+28)^{2}} \biggr]e^{-\mu(p)x^{2}} \\& \hphantom{\frac{\partial C_{p}(x)}{\partial p}}= \frac{4e^{-px^{2}}}{3(15p^{2}-40p+28)^{2}}F_{2}(t), \end{aligned}$$
(2.26)
$$\begin{aligned}& F_{2}(t)= -5(3p-4)t-10(3p-4)+\bigl[-5(3p-4)t+10(3p-4) \bigr]e^{t} \\& \hphantom{F_{2}(t)}= -5(3p-4)\bigl[(t+2)+(t-2)e^{t}\bigr] \\& \hphantom{F_{2}(t)}= -5(3p-4)\sum_{n=2}^{\infty} \frac{(n-2)}{n!}t^{n}< 0 \end{aligned}$$
(2.27)
for \(p\in(7/5,\infty)\) and \(t>0\).

Therefore, part (2) follows from (2.21), (2.26), and (2.27).

For part (3), let \(G_{p}(x)\) be defined by
$$ G_{p}(x)=\frac{\pi}{4}\alpha^{2}(p)B^{2}_{p}(x)= \frac{1-\lambda (p)e^{-px^{2}}-(1-\lambda(p))e^{-\mu(p)x^{2}}}{p\lambda(p)+\mu (p)(1-\lambda(p))}. $$
(2.28)
Then elaborated computations lead to
$$ \frac{\partial G_{p}(x)}{\partial x}=\frac{2x [p\lambda(p)e^{-px^{2}}+\mu(p)(1-\lambda (p))e^{-\mu (p)x^{2}} ]}{p\lambda(p)+\mu(p)(1-\lambda(p))} =2xC_{p}(x). $$
(2.29)
It follows from Lemma 2.3(2) and (2.29) that
$$ \frac{\partial}{\partial x} \biggl(\frac{\partial G_{p}(x)}{\partial p} \biggr) = \frac{\partial}{\partial p} \biggl(\frac{\partial G_{p}(x)}{\partial x} \biggr)=2x\frac{\partial C_{p}(x)}{\partial p}< 0 $$
(2.30)
for \(x>0\) and \(p\in(7/5, \infty)\).
Inequality (2.30) implies that the function \(x\rightarrow\partial G_{p}(x)/\partial p\) is strictly decreasing on \((0, \infty)\) and
$$\begin{aligned} \frac{\partial G_{p}(x)}{\partial p} < &\frac{\partial G_{p}(x)}{\partial p}\bigg|_{x=0} \\ =& \biggl[-\frac{1-\lambda(p)e^{-px^{2}}-(1-\lambda(p))e^{-\mu (p)x^{2}}}{(p\lambda(p)+\mu(p)(1-\lambda(p)))^{2}} \frac{d(p\lambda(p)+\mu(p)(1-\lambda(p)))}{dp} \biggr]_{x=0} \\ &{}+ \biggl[\frac{-\lambda^{\prime}(p)e^{-px^{2}}+\lambda (p)x^{2}e^{-px^{2}}+\lambda^{\prime}(p)e^{-\mu(p)x^{2}} +\mu^{\prime}(p)x^{2}(1-\lambda(p))e^{-\mu(p)x^{2}}}{p\lambda (p)+\mu (p)(1-\lambda(p))} \biggr]_{x=0} \\ =& 0 \end{aligned}$$
(2.31)
for \(x>0\) and \(p\in(7/5, \infty)\).

Therefore, part (3) follows from (2.28) and (2.31). □

Lemma 2.4

Let \(p\in(7/5, \infty)\), \(x\in(0, \infty)\), \(\lambda(p)\), \(\mu(p)\), \(B_{p}(x)\) and \(H_{f, g}(x)\) be, respectively, defined by (1.4), (1.5) and (2.1), and \(f_{1}(x)\), \(g_{1}(x)\), \(f_{2}(x)\) and \(g_{2}(x)\) be, respectively, defined by
$$\begin{aligned}& f_{1}(x)=B^{2}_{p}(x)=1-\lambda(p)e^{-px^{2}}- \bigl(1-\lambda(p)\bigr)e^{-\mu(p)x^{2}},\qquad g_{1}(x)= \operatorname{erf}^{2}(x), \end{aligned}$$
(2.32)
$$\begin{aligned}& f_{2}(x)= \bigl[p\lambda(p)e^{(1-p)x^{2}}+\mu(p) \bigl(1-\lambda (p) \bigr)e^{(1-\mu (p))x^{2}} \bigr]x,\qquad g_{2}(x)=\frac{2}{\sqrt{\pi}} \operatorname{erf}(x). \end{aligned}$$
(2.33)
Then
$$\begin{aligned}& H_{f_{2}, g_{2}}(\infty)=\lim_{x\rightarrow \infty} \biggl( \frac{f^{\prime}_{2}(x)}{g^{\prime }_{2}(x)}g_{2}(x)-f_{2}(x) \biggr)= \textstyle\begin{cases} \infty, & p\in (\frac{7}{5}, \frac{8}{5} ], \\ -\infty, & p\in (\frac{8}{5}, \infty ), \end{cases}\displaystyle \end{aligned}$$
(2.34)
$$\begin{aligned}& H_{f_{1}, g_{1}}(\infty)=\lim_{x\rightarrow \infty} \biggl( \frac{f^{\prime}_{1}(x)}{g^{\prime }_{1}(x)}g_{1}(x)-f_{1}(x) \biggr)= \textstyle\begin{cases} \infty, & p\in (\frac{7}{5}, \frac{8}{5} ], \\ -1, & p\in (\frac{8}{5}, \infty ). \end{cases}\displaystyle \end{aligned}$$
(2.35)

Proof

Let \(t=(p-\mu(p))x^{2}\) and
$$\begin{aligned} h_{1}(t) =&p\lambda(p) \bigl(p-\mu(p)\bigr)-2p \lambda(p) (p-1)t \\ &{}+\mu(p) \bigl(p-\mu(p)\bigr) \bigl(1-\lambda(p)\bigr)e^{t}-2 \mu(p) \bigl(\mu(p)-1\bigr) \bigl(1-\lambda (p)\bigr)te^{t}. \end{aligned}$$
(2.36)
Then (2.1) and (2.33) lead to
$$\begin{aligned}& f_{2}(x)=\frac{\sqrt{t}}{\sqrt{p-\mu(p)}} \bigl(p\lambda (p)e^{\frac {1-p}{p-\mu(p)}t}+\mu(p) \bigl(1-\lambda(p)\bigr)e^{\frac{1-\mu(p)}{p-\mu (p)}t} \bigr), \end{aligned}$$
(2.37)
$$\begin{aligned}& \frac{f^{\prime}_{2}(x)}{g^{\prime}_{2}(x)}=\frac{\pi}{4(p-\mu (p))}e^{\frac{2-p}{p-\mu(p)}t}h_{1}(t), \\& H_{f_{2},g_{2}}(x)=\frac{f^{\prime}_{2}(x)}{g^{\prime }_{2}(x)}g_{2}(x)-f_{2}(x) \\& \hphantom{H_{f_{2},g_{2}}(x)}=\frac{\sqrt{\pi}}{2(p-\mu(p))}\operatorname{erf}(x)e^{\frac {2-p}{p-\mu (p)}t}h_{1}(t) \\& \hphantom{H_{f_{2},g_{2}}(x)={}}{}-\frac{\sqrt{t}}{\sqrt{p-\mu(p)}} \bigl(p\lambda(p)e^{\frac {1-p}{p-\mu (p)}t}+\mu(p) \bigl(1-\lambda(p)\bigr)e^{\frac{1-\mu(p)}{p-\mu(p)}t} \bigr). \end{aligned}$$
(2.38)
If \(p\in(8/5, \infty)\), then Lemma 2.2(1), (2.36), and (2.37) lead to
$$\begin{aligned}& p>\mu(p),\qquad 0< \lambda(p)< 1,\qquad 1< \mu(p)< \frac{4}{3}, \end{aligned}$$
(2.39)
$$\begin{aligned}& f_{2}(\infty)=0, \end{aligned}$$
(2.40)
$$\begin{aligned}& \lim_{t\rightarrow \infty}\frac{h_{1}(t)}{te^{t}}=-2\mu(p) \bigl(\mu(p)-1\bigr) \bigl(1-\lambda(p)\bigr)< 0, \\& \lim_{t\rightarrow \infty}e^{\frac{2-p}{p-\mu(p)}t}h_{1}(t)=\lim _{t\rightarrow \infty}te^{\frac{2-\mu(p)}{p-\mu(p)}t}\lim_{t\rightarrow \infty} \frac{h_{1}(t)}{te^{t}}=-\infty. \end{aligned}$$
(2.41)

Therefore, \(H_{f_{2}, g_{2}}(\infty)=-\infty\) for \(p\in(8/5, \infty)\) follows from (2.38), (2.39), and (2.41).

If \(p\in(7/5, 8/5]\), then it follows from Lemma 2.2(1) and (2.36) together with (2.37) that
$$\begin{aligned}& p>\mu(p),\qquad 0< \lambda(p)< 1,\qquad 0< \mu(p)\leq1, \end{aligned}$$
(2.42)
$$\begin{aligned}& \lim_{t\rightarrow \infty}\frac{f_{2}(x)}{te^{\frac{1-\mu(p)}{p-\mu(p)}t}}=\lim_{t\rightarrow \infty} \frac{p\lambda(p)e^{-t}+\mu(p)(1-\lambda(p))}{\sqrt{(p-\mu (p))t}}=0, \end{aligned}$$
(2.43)
$$\begin{aligned}& \lim_{t\rightarrow \infty}\frac{h_{1}(t)}{te^{\frac{1-\mu(p)}{p-\mu(p)}t}} \\& \quad =\lim_{t\rightarrow \infty} \biggl[\frac{p\lambda(p)(p-\mu(p))}{t}e^{\frac{\mu (p)-1}{p-\mu (p)}t}-2p \lambda(p) (p-1)e^{\frac{\mu(p)-1}{p-\mu(p)}t} \biggr] \\& \qquad {}+\lim_{t\rightarrow \infty} \biggl[\frac{\mu(p)(1-\lambda(p))(p-\mu(p))}{t}e^{\frac {p-1}{p-\mu (p)}t}+2 \mu(p) \bigl(1-\lambda(p)\bigr) \bigl(1-\mu(p)\bigr)e^{\frac{p-1}{p-\mu (p)}t} \biggr] \\& \quad =\infty. \end{aligned}$$
(2.44)
Therefore,
$$ H_{f_{2}, g_{2}}(\infty)=\lim_{t\rightarrow \infty}te^{\frac{1-\mu(p)}{p-\mu(p)}t} \biggl[ \frac{\sqrt{\pi}e^{\frac{2-p}{p-\mu(p)}t}}{2(p-\mu (p))}\operatorname{erf} (x)\frac{h_{1}(t)}{te^{\frac{1-\mu(p)}{p-\mu(p)}t}} -\frac{f_{2}(x)}{te^{\frac{1-\mu(p)}{p-\mu(p)}t}} \biggr]=\infty $$
for \(p\in(7/5, 8/5]\) as follows from (2.38) and (2.42)-(2.44).
Similarly, from (2.1) and (2.32) we have
$$\begin{aligned} H_{f_{1}, g_{1}}(x) =&\frac{\sqrt{\pi}}{2}x\operatorname{erf}(x) \bigl[p\lambda (p)e^{(1-p)x^{2}}+\mu(p) \bigl(1-\lambda(p)\bigr)e^{(1-\mu(p))x^{2}} \bigr] \\ &{}- \bigl[1-\lambda(p)e^{-px^{2}}-\bigl(1-\lambda(p)\bigr)e^{-\mu (p)x^{2}} \bigr]. \end{aligned}$$
(2.45)
If \(p\in(7/5, 8/5]\), then Lemma 2.2(1) gives
$$ 0< \lambda(p)< 1,\qquad 0< \mu(p)\leq1. $$
(2.46)
Therefore, \(H_{f_{1}, g_{1}}(\infty)=\infty\) for \(p\in(7/5, 8/5]\) as follows from (2.45) and (2.46).
If \(p\in(8/5, \infty)\), then Lemma 2.2(1) leads to
$$ 0< \lambda(p)< 1, \qquad 1< \mu(p)\leq\frac{4}{3}. $$
(2.47)
Therefore, \(H_{f_{1}, g_{1}}(\infty)=-1\) for \(p\in(8/5, \infty)\) as follows from (2.45) and (2.47). □

Lemma 2.5

Let \(p\in(7/5, \infty)\), \(p^{\ast}_{0}=(50+2\sqrt{30})/35=1.74155\ldots\) , \(x\in(0, \infty)\), \(\lambda(p)\), \(\mu(p)\), \(B_{p}(x)\), \(H_{f, g}(x)\), \(f_{1}(x)\), \(g_{1}(x)\), \(f_{2}(x)\) and \(g_{2}(x)\) be, respectively, defined by (1.4), (1.5), (2.1), (2.32) and (2.33). Then the following statements are true:
  1. (1)

    if \(f_{1}(x)/g_{1}(x)\) is strictly increasing on \((0, \infty)\), then \(p\in(7/5, 8/5]\);

     
  2. (2)

    if \(f_{1}(x)/g_{1}(x)\) is strictly decreasing on \((0, \infty)\), then \(p\in[p^{\ast}_{0}, \infty)\).

     

Proof

(1) It follows from (2.1) and (2.32) that
$$ \lim_{x\rightarrow\infty}e^{x^{2}} \biggl( \frac {f_{1}(x)}{g_{1}(x)} \biggr)^{\prime} =\lim_{x\rightarrow\infty}e^{x^{2}} \frac{g^{\prime }_{1}(x)}{g^{2}_{1}(x)}H_{f_{1}, g_{1}}(x) =\frac{4}{\sqrt{\pi}}\lim _{x\rightarrow\infty}H_{f_{1}, g_{1}}(x). $$
(2.48)
If \(f_{1}(x)/g_{1}(x)\) is strictly increasing on \((0, \infty)\), then (2.48) leads to
$$ \lim_{x\rightarrow\infty}H_{f_{1}, g_{1}}(x)\geq0 $$
(2.49)
and we assert that \(p\in(7/5, 8/5]\). Otherwise, \(p>8/5\) and (2.35) lead to the conclusion \(H_{f_{1}, g_{1}}(\infty)=-1\), which contradicts with (2.49).
(2) Let \(t=(p-\mu(p))x^{2}\), \(u_{n}\) and \(h_{1}(t)\) be, respectively, defined by (2.4) and (2.36), and \(h_{2}(t)\) and \(v_{n}\) be, respectively, defined by
$$\begin{aligned}& h_{2}(t)= 2\mu(p) \bigl(1-\lambda(p)\bigr) \bigl(\mu(p)-1\bigr) \bigl( \mu(p)-2\bigr)te^{t}-\mu (p) \bigl(1-\lambda(p)\bigr) \bigl(3\mu(p)-4 \bigr) \\& \hphantom{h_{2}(t)={}}{}\times \bigl(p-\mu(p)\bigr)e^{t}+2p\lambda(p) (p-1) (p-2)t \\& \hphantom{h_{2}(t)={}}{}-p\lambda(p) (3p-4) \bigl(p-\mu(p)\bigr), \end{aligned}$$
(2.50)
$$\begin{aligned}& v_{n}=-\frac{4\mu(p)(1-\lambda(p))}{25(3p-4)^{2}}u_{n}. \end{aligned}$$
(2.51)
Then from (2.1)-(2.3), (2.32), (2.33), (2.36), and (2.50) we have
$$\begin{aligned}& \biggl(\frac{f^{\prime}_{2}(x)}{g^{\prime}_{2}(x)} \biggr)^{\prime }= \frac {\pi}{4(p-\mu(p))} \frac{d}{dt} \bigl[e^{\frac{2-p}{p-\mu (p)}t}h_{1}(t) \bigr] \frac{dt}{dx} \\& \hphantom{\biggl(\frac{f^{\prime}_{2}(x)}{g^{\prime}_{2}(x)} \biggr)^{\prime }}= \frac{\pi x}{2(p-\mu(p))}e^{\frac{2-p}{p-\mu(p)}t}h_{2}(t), \end{aligned}$$
(2.52)
$$\begin{aligned}& h_{2}(t)= 2\mu(p) \bigl(1-\lambda(p)\bigr) \bigl(\mu(p)-1\bigr) \bigl( \mu(p)-2\bigr)\sum_{n=1}^{\infty } \frac{t^{n}}{(n-1)!} \\& \hphantom{h_{2}(t)={}}{}-\mu(p) \bigl(1-\lambda(p)\bigr) \bigl(3\mu(p)-4\bigr) \bigl(p- \mu(p)\bigr)\sum_{n=0}^{\infty }\frac {t^{n}}{n!} \\& \hphantom{h_{2}(t)={}}{}+2p\lambda(p) (p-1) (p-2)t-p\lambda(p) (3p-4) \bigl(p-\mu(p) \bigr) \\& \hphantom{h_{2}(t)}= -\bigl(p-\mu(p)\bigr)\bigl[\bigl(p-\mu(p)\bigr) \bigl(3p+3 \mu(p)-4\bigr)\lambda(p)+\mu(p) \bigl(3\mu (p)-4\bigr)\bigr] \\& \hphantom{h_{2}(t)={}}{}+\bigl(p-\mu(p)\bigr) \bigl(2p^{2}+5 \mu^{2}(p)+2p\mu(p)-6p-10\mu(p)+4\bigr)\lambda (p) \\& \hphantom{h_{2}(t)={}}{}+\mu(p) \bigl(4p-3p\mu(p)+5\mu^{2}(p)-10\mu(p)+4 \bigr)+\sum_{n=2}^{\infty }\frac {v_{n}t^{n}}{n!} \\& \hphantom{h_{2}(t)}= \sum_{n=2}^{\infty} \frac{v_{n}t^{n}}{n!}, \\& \biggl(\frac{f^{\prime}_{1}(x)}{g^{\prime}_{1}(x)} \biggr)^{\prime} =\frac{g^{\prime}_{1}(x)}{g^{2}_{1}(x)} \biggl[ \frac{f^{\prime }_{1}(x)}{g^{\prime}_{1}(x)}g_{1}(x)-f_{1}(x) \biggr] = \frac{g^{\prime}_{1}(x)}{g^{2}_{1}(x)}H_{f_{1}, g_{1}}(x), \\& H^{\prime}_{f_{1}, g_{1}}(x)= \biggl(\frac{f^{\prime}_{1}(x)}{g^{\prime}_{1}(x)} \biggr)^{\prime}g_{1}(x) = \biggl(\frac{f_{2}(x)}{g_{2}(x)} \biggr)^{\prime}g_{1}(x) =\frac{g^{\prime}_{2}(x)}{g^{2}_{2}(x)}H_{f_{2}, g_{2}}(x)g_{1}(x), \\& H^{\prime}_{f_{2}, g_{2}}(x)= \biggl(\frac{f^{\prime}_{2}(x)}{g^{\prime}_{2}(x)} \biggr)^{\prime}g_{2}(x) =\frac{\pi x}{2(p-\mu(p))}e^{\frac{2-p}{p-\mu(p)}t}h_{2}(t)g_{2}(x), \\& \frac{g^{\prime}_{1}(x)}{g^{2}_{1}(x)}=\frac{4}{\sqrt{\pi}}\frac {e^{-x^{2}}}{\operatorname{erf}^{3}(x)}\sim \frac{\pi}{2x^{3}}, \qquad \frac{g^{\prime}_{2}(x)}{g^{2}_{2}(x)}g_{1}(x)=e^{-x^{2}}\sim1\quad \bigl(x \rightarrow0^{+}\bigr), \\& g_{2}(x)=\frac{2}{\sqrt{\pi}}\operatorname{erf}(x)\sim\frac {4x}{\pi}, \qquad h_{2}(t)\sim\frac{v_{2}}{2}t^{2}= \frac{(p-\mu(p))^{2}}{2}v_{2}x^{4} \quad \bigl(x \rightarrow0^{+}\bigr), \\& \biggl(\frac{f_{1}(x)}{g_{1}(x)} \biggr)^{\prime}\sim\frac{\pi }{2x^{3}}H_{f_{1}, g_{1}}(x) \quad \bigl(x\rightarrow0^{+}\bigr), \\& H^{\prime}_{f_{1}, g_{1}}(x)\sim H_{f_{2}, g_{2}}(x), \\& H^{\prime}_{f_{2}, g_{2}}(x)\sim\bigl(p-\mu(p)\bigr)v_{2}x^{6} \quad \bigl(x\rightarrow0^{+}\bigr). \end{aligned}$$
(2.53)
Note that \(H_{f_{1}, g_{1}}(0^{+})=H_{f_{2}, g_{2}}(0^{+})=0\). Making use of the L’Hôspital rule we get
$$\begin{aligned}& \lim_{x\rightarrow 0^{+}}x^{-5} \biggl(\frac{f_{1}(x)}{g_{1}(x)} \biggr)^{\prime } \\& \quad = \frac{\pi}{2}\lim_{x\rightarrow0^{+}}\frac{H_{f_{1}, g_{1}(x)}(x)}{x^{8}}= \frac{\pi}{16}\lim_{x\rightarrow 0^{+}}\frac{H_{f_{2}, g_{2}(x)}(x)}{x^{7}} \\& \quad = \frac{\pi}{112}\lim_{x\rightarrow0^{+}}\frac{H^{\prime}_{f_{2}, g_{2}(x)}(x)}{x^{6}}= \frac{\pi(p-\mu(p))}{112}v_{2} \\& \quad = -\frac{\pi\mu(p)(p-\mu(p))(1-\lambda(p))}{700} \bigl(35p^{2}-100p+68 \bigr). \end{aligned}$$
(2.54)
If \(p\in(7/5, \infty)\) and \(f_{1}(x)/g_{1}(x)\) is strictly decreasing on \((0, \infty)\), then it follows from Lemma 2.2(1) and (2.54) that
$$ 35p^{2}-100p+68\geq0, $$
which leads to \(p\geq(50+2\sqrt{30})/35=p^{\ast}_{0}\). □

3 Main results

Theorem 3.1

Let \(p\in(7/5, \infty)\), \(x>0\), \(p^{\ast}_{0}=(50+2\sqrt{30})/35\), \(\lambda(p)\), \(\mu(p)\), \(B_{p}(x)\) and \(\alpha(p)\) be, respectively, defined by (1.4), (1.5), and (2.18), \(x_{0}\) be the unique solution of the equation
$$ \frac{d}{dx} \biggl(\frac{B^{2}_{p}(x)}{\operatorname {erf}^{2}(x)} \biggr)=0 $$
on the interval \((0, \infty)\) and \(\beta(p)=\operatorname{erf}(x_{0})/B_{p}(x_{0})\). Then the following statements are true:
  1. (1)
    the function \(x\rightarrow Q_{p}(x)=\operatorname {erf}(x)/B_{p}(x)\) is strictly decreasing on \((0, \infty)\) if and only if \(p\in(7/5, 8/5]\), and the double inequality
    $$ 1< \frac{\operatorname{erf}(x)}{B_{p}(x)}< \alpha(p) $$
    (3.1)
    holds for all \(x>0\) with the best possible parameters 1 and \(\alpha(p)\) if \(p\in(7/5, 8/5]\);
     
  2. (2)
    the function \(x\rightarrow Q_{p}(x)=\operatorname {erf}(x)/B_{p}(x)\) is strictly increasing on \((0, \infty)\) if and only if \(p\in [p^{\ast}_{0}, \infty)\), and the double inequality
    $$ \alpha(p)< \frac{\operatorname{erf}(x)}{B_{p}(x)}< 1 $$
    (3.2)
    holds for all \(x>0\) with the best possible parameters 1 and \(\alpha(p)\) if \(p\in[p^{\ast}_{0}, \infty)\);
     
  3. (3)
    if \(p\in(8/5, p^{\ast}_{0})\), then \(Q_{p}(x)\) is strictly decreasing on \((0, x_{0}]\) and strictly increasing on \([x_{0}, \infty)\), and the double inequality
    $$ \beta(p)\leq\frac{\operatorname{erf}(x)}{B_{p}(x)}< \max\bigl\{ 1, \alpha (p)\bigr\} $$
    (3.3)
    for all \(x>0\).
     

Proof

Let \(t=(p-\mu(p))x^{2}\), \(f_{1}(x)\), \(g_{1}(x)\), \(f_{2}(x)\), \(g_{2}(x)\), \(u_{n}\), \(v_{n}\), and \(h_{2}(t)\) be defined by (2.32) and (2.33), (2.4), (2.51), and (2.53). Then
$$\begin{aligned}& Q^{-2}_{p}(x)=\frac{f_{1}(x)}{g_{1}(x)}, \\& f_{1}\bigl(0^{+}\bigr)=g_{1} \bigl(0^{+}\bigr)=0,\qquad g^{\prime}_{1}(x)>0, \end{aligned}$$
(3.4)
$$\begin{aligned}& f_{2}\bigl(0^{+}\bigr)=g_{2}\bigl(0^{+} \bigr)=0,\qquad g^{\prime}_{2}(x)>0, \end{aligned}$$
(3.5)
$$\begin{aligned}& \frac{f^{\prime}_{1}(x)}{g^{\prime}_{1}(x)}=\frac{f_{2}(x)}{g_{2}(x)}. \end{aligned}$$
(3.6)

(1) If \(Q_{p}(x)=\operatorname{erf}(x)/B_{p}(x)\) is strictly decreasing on \((0, \infty)\), then \(f_{1}(x)/g_{1}(x)\) is strictly increasing on \((0, \infty)\) and \(p\in(7/5, 8/5]\) by Lemma 2.5(1).

If \(p\in(7/5, 8/5]\), then it follows from Lemma 2.2(1) and (2) together with (2.51)-(2.53) that the function \(f^{\prime}_{2}(x)/g^{\prime}_{2}(x)\) is strictly increasing on \((0, \infty)\). Then from the monotone form of L’Hôpital’s rule [22], Theorem 1.25, and (3.5) together with (3.6) we know that the function \(f^{\prime}_{1}(x)/g^{\prime}_{1}(x)\) is strictly increasing on \((0, \infty)\). Therefore, \(Q_{p}(x)\) is strictly decreasing or \(f_{1}(x)/g_{1}(x)\) is strictly increasing on \((0, \infty)\) as follows from the monotone form of L’Hôpital’s rule [22], Theorem 1.25, and (3.4) together with the monotonicity of the function \(f^{\prime}_{1}(x)/g^{\prime}_{1}(x)\) on the interval \((0, \infty)\).

Note that
$$ \lim_{x\rightarrow 0^{+}}\frac{f_{1}(x)}{g_{1}(x)}= \frac{\pi}{4}\bigl[p\lambda(p)+\mu (p) \bigl(1-\lambda(p)\bigr)\bigr]= \frac{1}{\alpha^{2}(p)},\qquad \lim_{x\rightarrow\infty}\frac{f_{1}(x)}{g_{1}(x)}=1. $$
(3.7)
Therefore, the double inequality (3.1) holds for all \(x>0\) and \(p\in (7/5, 8/5]\) with the best possible parameters 1 and \(\alpha(p)\) as follows from (3.7) and the monotonicity of \(f_{1}(x)/g_{1}(x)\) on the interval \((0, \infty)\).

(2) If \(Q_{p}(x)=\operatorname{erf}(x)/B_{p}(x)\) is strictly increasing on \((0, \infty)\), then \(f_{1}(x)/g_{1}(x)\) is strictly decreasing on \((0, \infty)\) and \(p\in[p^{\ast}_{0}, \infty)\) by Lemma 2.5(2).

If \(p\in[p^{\ast}_{0}, \infty)\), then it follows from Lemma 2.2(1) and (3) together with (2.51)-(2.53) that the function \(f^{\prime}_{2}(x)/g^{\prime}_{2}(x)\) is strictly decreasing on \((0, \infty)\). Therefore, \(Q_{p}(x)\) is strictly increasing or \(f_{1}(x)/g_{1}(x)\) is strictly decreasing on \((0, \infty)\) as follows from the monotone form of L’Hôpital’s rule and (3.4)-(3.6) together with the monotonicity of the function \(f^{\prime}_{2}(x)/g^{\prime}_{2}(x)\) on the interval \((0, \infty)\), and the double inequality (3.2) holds for all \(x>0\) and \(p\in [p^{\ast}_{0}, \infty)\) with the best possible parameters 1 and \(\alpha(p)\) as follows from (3.7) and the monotonicity of \(f_{1}(x)/g_{1}(x)\) on the interval \((0, \infty)\).

(3) If \(p\in(8/5, p^{\ast}_{0})\), then it follows from [23], Lemma 6.4, or [24], Lemma 7, Lemma 2.2(1) and (4), (2.34), (2.35), and (2.51)-(2.53) that there exists \(x_{1}\in(0, \infty)\) such that \(f^{\prime}_{2}(x)/g^{\prime}_{2}(x)\) is strictly increasing on \((0, x_{1})\) and strictly decreasing on \((x_{1}, \infty)\), and
$$\begin{aligned}& H_{f_{1}, g_{1}}(\infty)=-1, \end{aligned}$$
(3.8)
$$\begin{aligned}& H_{f_{2}, g_{2}}(\infty)=-\infty. \end{aligned}$$
(3.9)
From Lemma 2.1, (3.5), (3.6), (3.9) and the piecewise monotonicity of \(f^{\prime}_{2}(x)/g^{\prime}_{2}(x)\) on the interval \((0, \infty)\) we known that there exists \(x_{2}\in(0, \infty)\) such that \(f^{\prime}_{1}(x)/g^{\prime}_{1}(x)\) is strictly increasing on \((0, x_{2})\) and strictly decreasing on \((x_{2}, \infty)\). Then (3.4), (3.8) and Lemma 2.1 lead to the conclusion that there exists \(x_{0}\in(0, \infty)\) such that the function \(f_{1}(x)/g_{1}(x)=B^{2}_{p}(x)/\operatorname{erf}^{2}(x)\) is strictly increasing on \((0, x_{0})\) and strictly decreasing on \((x_{0}, \infty)\). We clearly see that \(x_{0}\) is the unique solution of the equation
$$ \frac{d}{dx} \biggl(\frac{B^{2}_{p}(x)}{\operatorname {erf}^{2}(x)} \biggr)=0 $$
on the interval \((0, \infty)\). Therefore, \(Q_{p}(x)=\operatorname{erf}(x)/B_{p}(x)=(f_{1}(x)/g_{1}(x))^{-1/2}\) is strictly decreasing on \((0, x_{0}]\) and strictly increasing on \([x_{0}, \infty)\), and inequality (3.3) holds for all \(x>0\) as follows easily from (3.7). □
Let \(p\in(7/5, \infty)\) and \(\alpha(p)=\sqrt{ (45p^{2}-60p-4 )/[3p\pi(5p-7)]}=1\), then \(p=p_{0}=(21\pi-60+\sqrt{3}\sqrt{147\pi^{2}-920\pi+1440})/[30(\pi -3)]=1.71318\ldots\in (7/5, p^{\ast}_{0})\). Numerical computations show that \(x_{0}=1.68913\ldots\) is the unique solution of the equation
$$ \frac{d}{dx} \biggl(\frac{B^{2}_{p_{0}}(x)}{\operatorname {erf}^{2}(x)} \biggr)=0 $$
(3.10)
on the interval \((0, \infty)\), \(\beta(p_{0})=\operatorname{erf}(x_{0})/B_{p_{0}}(x_{0})=0.9998\ldots \) . Therefore, Theorem 3.1(3) leads to Corollary 3.1 immediately.

Corollary 3.1

Let \(p_{0}=(21\pi-60+\sqrt{3}\sqrt{147\pi^{2}-920\pi+1440})/[30(\pi-3)]\), \(B_{p}(x)\) be defined by (1.5) and \(x_{0}=1.68913\ldots\) be the unique solution of equation (3.10) on the interval \((0, \infty)\). Then the double inequality
$$ 0.9998< \frac{\operatorname{erf}(x_{0})}{B_{p_{0}}(x_{0})}\leq\frac {\operatorname{erf} (x)}{B_{p_{0}}(x)}< 1 $$
(3.11)
holds for all \(x>0\).

Theorem 3.2

Let \(p_{0}=(21\pi-60+\sqrt{3}\sqrt{147\pi^{2}-920\pi+1440})/[30(\pi-3)]\), \(p\in(7/5, \infty)\), \(x>0\), \(\lambda(p)\), \(\mu(p)\) and \(B_{p}(x)\) be, respectively, defined by (1.4) and (1.5). Then the inequality
$$ \operatorname{erf}(x)>B_{p}(x) $$
(3.12)
holds for all \(x>0\) if and only if \(p\in(7/5, 8/5]\), and inequality (3.12) is reversed if and only if \(p\in[p_{0}, \infty)\).

Proof

Making use of the L’Hôspital rule and Lemma 2.2(1) we have
$$\begin{aligned}& \lim_{x\rightarrow \infty}\frac{\operatorname{erf}^{2}(x)-B^{2}_{p}(x)}{e^{-\mu (p)x^{2}}} \\& \quad = \lim_{x\rightarrow \infty} \biggl[-\frac{2\operatorname{erf}(x)}{\sqrt{\pi}\mu (p)} \frac{e^{(\mu (p)-1)x^{2}}}{x}+\frac{p\lambda(p)}{ \mu(p)}e^{-(p-\mu(p))x^{2}}+1-\lambda(p) \biggr], \end{aligned}$$
(3.13)
$$\begin{aligned}& p>\mu(p),\qquad 0< \lambda(p)\leq\frac{8+\sqrt{14}}{16}=0.73385\ldots, \end{aligned}$$
(3.14)
$$\begin{aligned}& 0< \mu(p)\leq1\quad (7/5< p\leq8/5), \end{aligned}$$
(3.15)
$$\begin{aligned}& 1< \mu(p)< \frac{4}{3} \quad (8/5< p< \infty). \end{aligned}$$
(3.16)
It follows from (3.13)-(3.16) that
$$ \lim_{x\rightarrow \infty}\frac{\operatorname{erf}^{2}(x)-B^{2}_{p}(x)}{e^{-\mu(p)x^{2}}}= \textstyle\begin{cases} 1-\lambda(p)>0, & p\in (\frac{7}{5}, \frac{8}{5} ], \\ -\infty, & p\in (\frac{8}{5}, \infty ). \end{cases} $$
(3.17)
If inequality (3.12) holds for all \(x>0\), then
$$ \lim_{x\rightarrow \infty}\frac{\operatorname{erf}^{2}(x)-B^{2}_{p}(x)}{e^{-\mu (p)x^{2}}}\geq0, $$
(3.18)
and \(p\in(7/5, 8/5]\) as follows easily from (3.17) and (3.18).

If \(p\in(7/5, 8/5]\), then inequality (3.12) holds for all \(x>0\) as follows directly from Theorem 3.1(1).

If \(\operatorname{erf}(x)< B_{p}(x)\) for all \(x>0\), then \(p\geq p_{0}\) as follows easily from
$$ \lim_{x\rightarrow 0^{+}}\frac{\operatorname{erf}(x)}{B_{p}(x)}=\alpha(p)=\sqrt{ \frac {45p^{2}-60p-4}{3\pi p(5p-7)}}\leq1. $$

If \(p\in[p_{0}, \infty)\), then we divide the proof into two cases.

Case 1. \(p\in[p^{\ast}_{0}, \infty)\). Then \(\operatorname {erf}(x)< B_{p}(x)\) for all \(x>0\) as follows from Theorem 3.1(2).

Case 2. \(p\in[p_{0}, p^{\ast}_{0})\). Then
$$ \alpha(p)=\sqrt{\frac{45p^{2}-60p-4}{3\pi p(5p-7)}}\leq \alpha(p_{0})=1, $$
(3.19)
and \(\operatorname{erf}(x)< B_{p}(x)\) for all \(x>0\) as follows from Theorem 3.1(3) and (3.19). □

Remark 3.1

Let \(p^{\ast}_{0}=(50+2\sqrt{30})/35\), and \(f_{2}(x)\) and \(g_{2}(x)\) be defined by (2.33). Then from (3.6) and the proof of Theorem 3.1 we know that the function \(f_{2}(x)/g_{2}(x)\) is strictly increasing on \((0, \infty)\) if \(p\in(7/5, 8/5]\) and strictly decreasing on \((0, \infty)\) if \(p\in[p^{\ast}_{0}, \infty)\). Therefore, we have
$$ \frac{\pi}{4}\bigl[p\lambda(p)+\mu(p) \bigl(1-\lambda(p)\bigr)\bigr]=\lim _{x\rightarrow 0^{+}}\frac{f_{2}(x)}{g_{2}(x)}< \frac{f_{2}(x)}{g_{2}(x)}< \lim _{x\rightarrow \infty}\frac{f_{2}(x)}{g_{2}(x)}=\infty $$
for all \(x\in(0, \infty)\) and \(p\in(7/5, 8/5]\), and
$$ 0=\lim_{x\rightarrow \infty}\frac{f_{2}(x)}{g_{2}(x)}< \frac{f_{2}(x)}{g_{2}(x)}< \lim _{x\rightarrow 0^{+}}\frac{f_{2}(x)}{g_{2}(x)}=\frac{\pi}{4}\bigl[p\lambda(p)+ \mu (p) \bigl(1-\lambda(p)\bigr)\bigr] $$
for all \(x\in(0, \infty)\) and \(p\in[p^{\ast}_{0}, \infty)\).

Remark 3.1 can be restated as Theorem 3.3.

Theorem 3.3

Let \(p^{\ast}_{0}=(50+2\sqrt{30})/35\) and \(C_{p}(x)\) be defined by Lemma  2.3. Then the inequality
$$ \operatorname{erf}(x)< \frac{2xe^{x^{2}}}{\sqrt{\pi}}C_{p}(x) $$
(3.20)
holds for all \(x>0\) if \(p\in(7/5, 8/5]\), and inequality (3.20) is reversed for all \(x>0\) if \(p\in[p^{\ast}_{0}, \infty)\).

Remark 3.2

Let \(p_{0}=(21\pi-60+\sqrt{3}\sqrt{147\pi^{2}-920\pi+1440})/[30(\pi-3)]\), \(p, q\in(7/5, \infty)\), \(x>0\), \(\lambda(p)\), \(\mu(p)\) and \(B_{p}(x)\) be, respectively, defined by (1.4) and (1.5). Then it follows from Lemma 2.3(1) and Theorem 3.2 that the double inequality
$$\begin{aligned}& \sqrt{1-\lambda(p)e^{-px^{2}}-\bigl[1-\lambda(p)\bigr]e^{-\mu (p)x^{2}}} \\& \quad =B_{p}(x)< \operatorname{erf}(x)< B_{q}(x)=\sqrt{1- \lambda(q)e^{-qx^{2}}-\bigl[1-\lambda(q)\bigr]e^{-\mu(q)x^{2}}} \end{aligned}$$
holds for all \(x>0\) with the best possible parameters \(p=8/5\) and \(q=p_{0}\).

Corollary 3.2

Let \(p_{0}=(21\pi-60+\sqrt{3}\sqrt{147\pi^{2}-920\pi+1440})/[30(\pi-3)]\), and \(\lambda(p)\) and \(\mu(p)\) be defined by (1.4). Then the inequalities
$$\begin{aligned} \sqrt{1-e^{-x^{2}}}&< \sqrt{1-\frac{25}{57}e^{-8x^{2}/5}- \frac {32}{57}e^{-x^{2}}}< \operatorname{erf}(x) \\ &< \sqrt{1-\lambda(p_{0})e^{-p_{0}x^{2}}-\bigl[1- \lambda(p_{0})\bigr]e^{-\mu (p_{0})x^{2}}}< \sqrt{1-e^{4x^{2}/\pi}} \end{aligned}$$
hold for all \(x>0\).

Proof

From (1.4) one has
$$ \lambda \biggl(\frac{8}{5} \biggr)=\frac{25}{57}, \qquad \mu \biggl(\frac{8}{5} \biggr)=1. $$
(3.21)
Note that \(p_{0}\) satisfies the identity
$$ \frac{45p^{2}_{0}-60p_{0}-4}{p_{0}(5p_{0}-7)}=3\pi. $$
(3.22)
It follows from Remark 3.2 and (3.21) that
$$ \sqrt{1-\frac{25}{57}e^{-8x^{2}/5}-\frac {32}{57}e^{-x^{2}}}< \operatorname{erf} (x)< \sqrt{1-\lambda(p_{0})e^{-p_{0}x^{2}}-\bigl[1- \lambda(p_{0})\bigr]e^{-\mu (p_{0})x^{2}}} $$
for all \(x>0\). Therefore, it suffices to prove that
$$\begin{aligned}& \frac{25}{57}e^{-8x^{2}/5}+\frac{32}{57}e^{-x^{2}}< e^{-x^{2}}, \end{aligned}$$
(3.23)
$$\begin{aligned}& \lambda(p_{0})e^{-p_{0}x^{2}}+\bigl[1-\lambda(p_{0}) \bigr]e^{-\mu (p_{0})x^{2}}>e^{-4x^{2}/\pi} \end{aligned}$$
(3.24)
for all \(x>0\).
Inequality (3.23) follows easily from
$$ \frac{25}{57}e^{-8x^{2}/5}+\frac{32}{57}e^{-x^{2}}-e^{-x^{2}}= \frac {25}{57}e^{-8x^{2}/5} \bigl(1-e^{3x^{2}/5} \bigr). $$
Making use of (1.4) and (3.22) together with the arithmetic-geometric mean inequality one has
$$\begin{aligned} \lambda(p_{0})e^{-p_{0}x^{2}}+\bigl[1-\lambda(p_{0}) \bigr]e^{-\mu (p_{0})x^{2}}&>e^{- [p_{0}\lambda(p_{0})+\mu(p_{0})(1-\lambda (p_{0})) ]x^{2}} \\ &=e^{-\frac {12p_{0}(5p_{0}-7)}{45p^{2}_{0}-60p_{0}-4}x^{2}}=e^{-4x^{2}/\pi} \end{aligned}$$
for all \(x>0\). □

Remark 3.3

We clearly see that the results given in Theorem 3.2, Remark 3.2, and Corollary 3.2 are improvements and refinements of inequality (1.1).

Let \(p^{\ast}_{0}=(50+2\sqrt{30})/35\) and \(\alpha(p)\) be defined by (2.18). Then
$$\begin{aligned}& \alpha \biggl(\frac{8}{5} \biggr)=\sqrt{\frac{19}{6\pi }}=1.00398 \ldots,\qquad \alpha \biggl(\frac{3}{2} \biggr)=\sqrt{ \frac{29}{9\pi }}=1.01275\ldots, \end{aligned}$$
(3.25)
$$\begin{aligned}& \alpha(\infty)=\sqrt{\frac{3}{\pi}}=0.97720\ldots,\qquad \alpha(2)=\sqrt{ \frac{28}{9\pi}}=0.99513\ldots, \\& \alpha\bigl(p^{\ast}_{0}\bigr)=\sqrt{\frac{160}{51\pi}}=0.99930 \ldots. \end{aligned}$$
(3.26)

Corollary 3.3

Let \(p=3/2, 8/5\) in Theorem  3.1(1) and \(p=p^{\ast}_{0}, 2, \infty\) in Theorem  3.1(2). Then Lemma  2.3(1) and (3) together with (3.25) and (3.26) leads to
$$\begin{aligned}& \sqrt{1-\frac{128}{203}e^{-3x^{2}/2}-\frac {75}{203}e^{-4x^{2}/5}} \\& \quad < \sqrt {1-\frac{25}{57}e^{-8x^{2}/5}-\frac{32}{57}e^{-x^{2}}}< \operatorname{erf} (x) \\& \quad < \sqrt{\frac{19}{6\pi}}\sqrt{1-\frac{25}{57}e^{-8x^{2}/5}- \frac {32}{57}e^{-x^{2}}} \\& \quad < \sqrt{\frac{29}{9\pi}}\sqrt{1- \frac{128}{203}e^{-3x^{2}/2}-\frac {75}{203}e^{-4x^{2}/5}}, \end{aligned}$$
(3.27)
$$\begin{aligned}& \sqrt{\frac{3}{\pi}}\sqrt{1-e^{-4x^{2}/3}} \\& \quad < \sqrt{\frac{28}{9\pi }}\sqrt {1-\frac{3}{28}e^{-2x^{2}}- \frac{25}{28}e^{-6x^{2}/5}} \\& \quad < \sqrt{\frac{160}{51\pi}}\sqrt{1-\frac{480-43\sqrt {30}}{960}e^{-(50+2\sqrt{30})x^{2}/35}- \frac{480+43\sqrt {30}}{960}e^{-(50-2\sqrt{30})x^{2}/35}} \\& \quad < \operatorname{erf}(x)< \sqrt{1-\frac{480-43\sqrt {30}}{960}e^{-(50+2\sqrt {30})x^{2}/35}- \frac{480+43\sqrt{30}}{960}e^{-(50-2\sqrt {30})x^{2}/35}} \\& \quad < \sqrt{1-\frac{3}{28}e^{-2x^{2}}-\frac{25}{28}e^{-6x^{2}/5}} \\& \quad < \sqrt {1-e^{-4x^{2}/3}}. \end{aligned}$$
(3.28)

Remark 3.4

From inequality (3.27) we clearly see that the double inequalities
$$\begin{aligned}& 0< \frac{\operatorname{erf}(x)-B_{3/2}(x)}{\operatorname {erf}(x)}< \sqrt{\frac{29}{9\pi }}-1=0.01275\ldots, \\& 0< \frac{\operatorname{erf}(x)-B_{3/2}(x)}{\operatorname {erf}(x)}< \sqrt{\frac{19}{6\pi }}-1=0.00398\ldots , \end{aligned}$$
hold for all \(x>0\).

Remark 3.5

Let \(p_{0}=(21\pi-60+\sqrt{3}\sqrt{147\pi^{2}-920\pi+1440})/[30(\pi-3)]\), \(p^{\ast}_{0}=(50+2\sqrt{30})/35\), \(B_{p}(x)\) be defined by (1.5) and \(x_{0}=1.68913\ldots\) be the unique solution of equation (3.10) on the interval \((0, \infty)\), \(\beta(p_{0})=\operatorname{erf}(x_{0})/B_{p_{0}}(x_{0})=0.9998\ldots \) . Then Corollary 3.1 and (3.28) lead to
$$\begin{aligned}& -0.00013\ldots=1-\frac{1}{\beta(p_{0})}< \frac{\operatorname{erf} (x)-B_{p_{0}}(x)}{\operatorname{erf} (x)}\leq 0, \\& -0.00069\ldots=1-\sqrt{\frac{51\pi}{160}}< \frac{\operatorname{erf} (x)-B_{p^{\ast }_{0}}(x)}{\operatorname{erf}(x)}< 0, \\& -0.00488\ldots=1-\sqrt{\frac{9\pi}{28}}< \frac{\operatorname{erf} (x)-B_{2}(x)}{\operatorname{erf}(x)}< 0, \\& -0.02332\ldots=1-\sqrt{\frac{\pi}{3}}< \frac{\operatorname {erf}(x)-\sqrt {1-e^{-4x^{2}/3}}}{\operatorname{erf}(x)}< 0 \end{aligned}$$
for all \(x>0\).

Corollary 3.4

Let \(p=(7/5)^{+}, 3/2, 8/5\) and \(p=2, \infty\) in Theorem  3.3. Then it follows from Lemma  2.3(2) that the inequalities
$$\begin{aligned} \frac{2x}{\sqrt{\pi}}e^{-x^{2}/3}&< \frac{2x}{\sqrt{\pi}}\frac {e^{-x^{2}}+5e^{-x^{2}/5}}{6}< \operatorname{erf}(x) < \frac{2x}{\sqrt{\pi}}\frac{5e^{-3x^{2}/5}+4}{9} \\ &< \frac{2x}{\sqrt{\pi}}\frac{16e^{-x^{2}/2}+5e^{x^{2}/5}}{21}< \frac {2x}{\sqrt{\pi}}\frac{20e^{-2x^{2}/5}+e^{x^{2}}}{21} \end{aligned}$$
hold for all \(x>0\).

Remark 3.6

From the identities
$$\begin{aligned}& \frac{e^{-x^{2}}+5e^{-x^{2}/5}}{6}-e^{-x^{2}/3} \\& \quad =\frac {1}{6}e^{-x^{2}} \bigl(e^{2x^{2}/15}-1 \bigr)^{2} \bigl(5e^{8x^{2}/15}+4e^{2x^{2}/5}+3e^{4x^{2}/15}+2e^{2x^{2}/15}+1 \bigr) \end{aligned}$$
and
$$\begin{aligned}& \frac{5e^{-3x^{2}/5}+4}{9}-\frac{e^{-x^{2}}+2}{3} \\& \quad = -\frac{e^{-x^{2}}}{9} \bigl(e^{x^{2}/5}-1 \bigr)^{2} \bigl(2e^{3x^{2}/5}+4e^{2x^{2}/5}+6e^{x^{2}/5}+3 \bigr) \end{aligned}$$
we know that the results given in Theorem 3.3 or Corollary 3.4 are better than that in (1.3).

Declarations

Acknowledgements

The research was supported by the Natural Science Foundation of China under Grants 61374086, 11371125 and 11401191.

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.

Authors’ Affiliations

(1)
School of Mathematics and Computation Sciences, Hunan City University
(2)
Customer Service Center, State Grid Zhejiang Electric Power Research Institute

References

  1. Laforgia, A, Sismondi, S: Monotonicity results and inequalities for the gamma and error functions. J. Comput. Appl. Math. 23(1), 25-33 (1988) MathSciNetView ArticleMATHGoogle Scholar
  2. Alzer, H: Error function inequalities. Adv. Comput. Math. 33(3), 349-379 (1988) MathSciNetView ArticleMATHGoogle Scholar
  3. Qi, F, Guo, S-L: Inequalities for the incomplete gamma and related functions. Math. Inequal. Appl. 2(1), 47-53 (1999) MathSciNetMATHGoogle Scholar
  4. Qi, F, Cui, L-H, Xu, S-L: Some inequalities constructed by Tchebysheff’s integral inequality. Math. Inequal. Appl. 2(4), 517-528 (1999) MathSciNetMATHGoogle Scholar
  5. Qi, F, Mei, J-Q: Some inequalities of the incomplete gamma and related functions. Z. Anal. Anwend. 18(3), 793-799 (1999) MathSciNetView ArticleMATHGoogle Scholar
  6. Qi, F: Monotonicity results and inequalities for the gamma and incomplete gamma functions. Math. Inequal. Appl. 5(1), 61-67 (2002) MathSciNetMATHGoogle Scholar
  7. Guo, B-N, Qi, F: A class of completely monotonic functions involving divided differences of the psi and tri-gamma functions and some applications. J. Korean Math. Soc. 48(3), 655-667 (2011) MathSciNetView ArticleMATHGoogle Scholar
  8. Gasull, A, Utzet, F: Approximating Mills ratio. J. Math. Anal. Appl. 420(2), 1832-1853 (2014) MathSciNetView ArticleMATHGoogle Scholar
  9. Chu, Y-M, Li, Y-M, Xia, W-F: Best possible inequalities for the harmonic mean of error function. J. Inequal. Appl. 2014, Article ID 525 (2014) MathSciNetView ArticleGoogle Scholar
  10. Xia, W-F, Chu, Y-M: Optimal inequalities for the convex combination of error function. J. Math. Inequal. 9(1), 85-99 (2015) MathSciNetView ArticleMATHGoogle Scholar
  11. Yang, Z-H, Chu, Y-M: On approximating Mills ratio. J. Inequal. Appl. 2015, Article ID 273 (2015) MathSciNetView ArticleMATHGoogle Scholar
  12. Li, Y-M, Xia, W-F, Chu, Y-M, Zhang, X-H: Optimal lower and upper bounds for the geometric convex combination of the error function. J. Inequal. Appl. 2015, Article ID 382 (2015) MathSciNetView ArticleMATHGoogle Scholar
  13. Chu, Y-M, Zhao, T-H: Concavity of the error function with respect to Hölder means. Math. Inequal. Appl. 19(2), 589-595 (2016) MathSciNetMATHGoogle Scholar
  14. Pólya, G: Remarks on computing the probability integral in one and two dimensions. In: Proceedings of the Berkeley Symposium on Mathematical Statistics and Probability. University of California Press, Berkely (1949) Google Scholar
  15. Chu, J-T: On bounds for the normal integral. Biometrika 42, 263-265 (1955) MathSciNetView ArticleMATHGoogle Scholar
  16. Alzer, H: On some inequalities for the incomplete gamma function. Math. Comput. 66(218), 771-778 (1997) MathSciNetView ArticleMATHGoogle Scholar
  17. Alzer, H: Functional inequalities for the error function. Aequ. Math. 66(1-2), 119-127 (2003) MathSciNetView ArticleMATHGoogle Scholar
  18. Alzer, H: Functional inequalities for the error function II. Aequ. Math. 78(1-2), 113-121 (2009) MathSciNetView ArticleMATHGoogle Scholar
  19. Neuman, E: Inequalities and bounds for the incomplete gamma function. Results Math. 2013(3-4), 1209-1214 (2009) MathSciNetMATHGoogle Scholar
  20. Yang, Z-H: A new way to prove L’Hospital monotone rules with applications. arXiv:1409.6408 [math.CA]
  21. Yang, Z-H, Chu, Y-M, Wang, M-K: Monotonicity criterion for the quotient of power series with applications. J. Math. Anal. Appl. 428(1), 587-604 (2015) MathSciNetView ArticleMATHGoogle Scholar
  22. Anderson, GD, Vamanamurthy, MK, Vuorinen, M: Conformal Invariants, Inequalities, and Quasiconformal Maps. Wiley, New York (1997) MATHGoogle Scholar
  23. Belzunce, F, Ortega, E-M, Ruiz, JM: On non-monotonic ageing properties from the Laplace transform, with actuarial applications. Insur. Math. Econ. 40(1), 1-14 (2007) MathSciNetView ArticleMATHGoogle Scholar
  24. Yang, Z-H, Chu, Y-M, Tao, X-J: A double inequality for the trigamma function and its applications. Abstr. Appl. Anal. 2014, Article ID 702718 (2014) MathSciNetGoogle Scholar

Copyright

© The Author(s) 2016