A separation of some Seiffert-type means by power means

Iulia Costin\(^\ast \) Gheorghe Toader\(^\S \)

June 12, 2012

\(^\ast \)Department of Computer Science, Technical University of Cluj-Napoca, Baritiu st. no. 28, Cluj-Napoca, Romania, e-mail: Iulia.Costin@cs.utcluj.ro.

\(^\S \)Department of Mathematics, Technical University of Cluj-Napoca, Baritiu st. no. 25, Cluj-Napoca, Romania, e-mail: Gheorghe.Toader@math.utcluj.ro.

Consider the identric mean \(\mathcal{I}\), the logarithmic mean \(\mathcal{L,}\) two trigonometric means defined by H. J. Seiffert and denoted by \(\mathcal{P} \) and \(\mathcal{T,}\) and the hyperbolic mean \(\mathcal{M}\) defined by E. Neuman and J. Sándor\(.\) There are a number of known inequalities between these means and some power means \(\mathcal{A}_{p}.\) We add to these inequalities some new results obtaining the following chain of inequalities

\[ \mathcal{A}_{0}{\lt}\mathcal{L}{\lt}\mathcal{A}_{1/3}{\lt}\mathcal{P{\lt}A}_{2/3}{\lt}\mathcal{I}{\lt}\mathcal{A}_{3/3}{\lt}\mathcal{M}{\lt}\mathcal{A}_{4/3}{\lt}\mathcal{T}{\lt}\mathcal{A}_{5/3}. \]

MSC. 26E60

Keywords. Seiffert type means; power means; logarithmic mean; identric mean; inequalities of means.

1 Introduction

A mean is a function \(M:\mathbb {R}_{+}^{2}\rightarrow \mathbb {R}_{+},\) with the property

\[ \min (a,b)\leq M(a,b)\leq \max (a,b),\quad \forall a,b{\gt}0. \]

Each mean is reflexive, that is

\[ M(a,a)=a,\quad \forall a{\gt}0. \]

This is also used as the definition of \(M(a,a)\).

A mean is symmetric if

\[ M(b,a)=M(a,b),\quad \forall a,b{\gt}0; \]

it is homogeneous (of degree 1) if

\[ M(ta,tb)=t\cdot M(a,b),\quad \forall a,b,t{\gt}0. \]

We shall refer here to the following symmetric and homogeneous means:

- the power means \(\mathcal{A}_{p}\), defined by

\[ \mathcal{A}_{p}(a,b)=\left[ \tfrac {a^{p}+b^{p}}{2}\right] ^{\tfrac {1}{p}},\quad p\neq 0; \]

- the geometric mean \(\mathcal{G}\), defined as \(\mathcal{G}(a,b)=\sqrt{ab},\) but verifying also the property

\[ \lim _{p\rightarrow 0}\mathcal{A}_{p}(a,b)=\mathcal{A}_{0}(a,b)=\mathcal{G}(a,b); \]

- the identric mean \(\mathcal{I}\) defined by

\[ \mathcal{I}(a,b)=\tfrac {1}{e}\left( \tfrac {a^{a}}{b^{b}}\right) ^{\tfrac {1}{a-b}},\quad a\neq b; \]

- the Gini mean \(\mathcal{S}\) defined by

\[ \mathcal{S}(a,b)=\left( a^{a}b^{b}\right) ^{\tfrac {1}{a+b}}; \]

- the first Seiffert mean \(\mathcal{P}\), defined in [ 9 ] by

\[ \mathcal{P}(a,b)=\tfrac {a-b}{2\sin ^{-1}\tfrac {a-b}{a+b}},\quad a\neq b; \]

- the second Seiffert mean \(\mathcal{T}\), defined in [ 10 ] by

\[ \mathcal{T}(a,b)=\tfrac {a-b}{2\tan ^{-1}\tfrac {a-b}{a+b}},\quad a\neq b; \]

- the Neuman-Sándor mean \(\mathcal{M}\), defined in [ 6 ] by

\[ \mathcal{M}(a,b)=\tfrac {a-b}{2\sinh ^{-1}\tfrac {a-b}{a+b}},\quad a\neq b; \]

- the logarithmic mean \(\mathcal{L}\) defined by

\[ \mathcal{L}(a,b)=\tfrac {a-b}{\ln a-\ln b},\quad a\neq b. \]

As remarked B.C. Carlson in [ 1 ] , the logarithmic mean can be represented also by

\[ \mathcal{L}(a,b)=\tfrac {a-b}{2\tanh ^{-1}\tfrac {a-b}{a+b}},\quad a\neq b, \]

thus the last four means are very similar.

Being rather complicated, these means were evaluated by simpler means, first of all by power means. For two means \(M\) and \(N\) we write \(M{\lt}N\) if \(M(a,b){\lt}N(a,b)\) for \(a\neq b.\) It is known that the family of power means is an increasing family of means, thus

\[ \mathcal{A}_{p}{\lt}\mathcal{A}_{q}\text{ if }p{\lt}q. \]

The evaluation of a given mean \(M\) by power means assumes the determination of some real indices \(p\) and \(q\) such that \(\mathcal{A}_{p}{\lt}M{\lt}\mathcal{A}_{q}.\) The evaluation is optimal if \(p\) is the the greatest and \(q\) is the smallest index with this property. This means that \(M \) cannot be compared with \(\mathcal{A}_{r}\) if \(p{\lt}r{\lt}q.\)

Optimal evaluation were given for the logarithmic mean in [ 5 ]

\[ \mathcal{A}_{0}{\lt}\mathcal{L}{\lt}\mathcal{A}_{1/3}, \]

for the identric mean in [ 8 ]

\[ \mathcal{A}_{2/3}{\lt}\mathcal{I}{\lt}\mathcal{A}_{\ln 2}, \]

and for the first Seiffert mean in [ 3 ]

\[ \mathcal{A}_{\ln 2/\ln \pi }{\lt}\mathcal{P}{\lt}\mathcal{A}_{2/3}. \]

Following evaluations are also known:

\[ \mathcal{A}_{1/3}{\lt}\mathcal{P}{\lt}\mathcal{A}_{2/3}, \]

proved in [ 4 ] ,

\[ \mathcal{A}_{1}{\lt}\mathcal{T}{\lt}\mathcal{A}_{2}, \]

given in [ 10 ] ,

\[ \mathcal{A}_{1}{\lt}\mathcal{M}{\lt}\mathcal{T}, \]

as it was shown in [ 6 ] and

\[ \mathcal{S{\gt}A}_{2} \]

as it is proved in [ 7 ] . In [ 2 ] it is proven that

\begin{equation} \mathcal{M}<\mathcal{A}_{3/2}<\mathcal{T}\label{u}\end{equation}
1

and using some of the above results, it is obtained the following chain of inequalities

\[ \mathcal{A}_{0}{\lt}\mathcal{L}{\lt}\mathcal{A}_{1/2}{\lt}\mathcal{P}{\lt}\mathcal{A}_{1}{\lt}\mathcal{M}{\lt}\mathcal{A}_{3/2}{\lt}\mathcal{T{\lt}\mathcal{A}}_{2}. \]

Here we retain another chain of inequalities

\begin{equation} \mathcal{A}_{0}<\mathcal{L}<\mathcal{A}_{1/3}<\mathcal{P}<\mathcal{A}_{2/3}<\mathcal{I}<\mathcal{A}_{1}<\mathcal{M}<\mathcal{T<\mathcal{A}}_{2}<\mathcal{S}.\label{f}\end{equation}
2

Our aim is to prove that \(\mathcal{A}_{4/3}\) can be put between \(\mathcal{M}\) and \(\mathcal{T}\) and \(\mathcal{A}_{2}\) can be replaced by \(\mathcal{A}_{5/3}\mathcal{.}\) We obtain so another nice separation of these means by “equidistant" power means.

2 Main results

We add to the inequalities (2) the next results.

Theorem 1

The following inequalities

\[ \mathcal{M{\lt}\mathcal{A}}_{4/3}{\lt}\mathcal{T{\lt}\mathcal{A}}_{5/3} \]

hold.

Proof â–¼
As the means are symmetric and homogenous, for the first inequality
\[ \tfrac {a-b}{2\sinh ^{-1}\tfrac {a-b}{a+b}}{\lt}\left( \tfrac {a^{4/3}+b^{4/3}}{2}\right) ^{\tfrac {3}{4}},\quad a\neq b, \]

we can assume that \(a{\gt}b\) and denote \(a/b=t^{3}{\gt}1.\) The inequality becomes

\[ \tfrac {t^{3}-1}{2\sinh ^{-1}\tfrac {t^{3}-1}{t^{3}+1}}{\lt}\left( \tfrac {t^{4}+1}{2}\right) ^{\tfrac {3}{4}},\quad t{\gt}1, \]

or

\[ \tfrac {2^{\tfrac {3}{4}}\left( t^{3}-1\right) }{2\left( t^{4}+1\right) ^{\tfrac {3}{4}}}{\lt}\sinh ^{-1}\tfrac {t^{3}-1}{t^{3}+1},\quad t{\gt}1. \]

Denoting

\[ f(t)=\sinh ^{-1}\tfrac {t^{3}-1}{t^{3}+1}-2^{-\tfrac {1}{4}}\left( t^{3}-1\right) \left( t^{4}+1\right) ^{-\tfrac {3}{4}} \]

we have to prove that \(f(t){\gt}0\) for \(t{\gt}1.\) As \(f(1)=0,\) we want to prove that \(f^{\prime }(t){\gt}0\) for \(t{\gt}1.\) We have

\begin{align*} f^{\prime }(t) & =\tfrac {6t^{2}}{\left( t^{3}+1\right) \sqrt{2\left( t^{6}+1\right) }}-2^{-\tfrac {1}{4}}\tfrac {3t^{2}\left( t+1\right) }{\left( t^{4}+1\right) ^{\tfrac {7}{4}}}\\ & =\tfrac {3t^{2}\left[ 2^{\tfrac {3}{4}}\left( t^{4}+1\right) ^{\tfrac {7}{4}}-\left( t+1\right) \left( t^{3}+1\right) \sqrt{t^{6}+1}\right] }{2^{\tfrac {1}{4}}\left( t^{3}+1\right) \sqrt{t^{6}+1}\left( t^{4}+1\right) ^{\tfrac {7}{4}}}\end{align*}

and so it is positive if

\[ g(t)=\left[ 2^{\tfrac {3}{4}}\left( t^{4}+1\right) ^{\tfrac {7}{4}}\right] ^{4}-\left[ \left( t+1\right) \left( t^{3}+1\right) \sqrt{t^{6}+1}\right] ^{4} \]

is positive. Or

\begin{align*} g(t) & =\left( t-1\right) ^{4}(7t^{24}+24t^{23}+48t^{22}+68t^{21}+112t^{20}\\ & \quad +184t^{19}+264t^{18}+296t^{17}+344t^{16}+428t^{15}\\ & \quad +512t^{14}+488t^{13}+466t^{12}+488t^{11}+512t^{10}\\ & \quad +428t^{9}+344t^{8}+296t^{7}+184t^{5}+112t^{4}\\ & \quad +68t^{3}+48t^{2}+24t+7) \end{align*}

so that the property is certainly true. The second inequality is a simple consequence of (1) because \(\mathcal{\mathcal{A}}_{4/3}{\lt}\mathcal{\mathcal{A}}_{3/2}\mathcal{.}\) For the last inequality

\[ \tfrac {a-b}{2\tan ^{-1}\tfrac {a-b}{a+b}}{\lt}\left( \tfrac {a^{5/3}+b^{5/3}}{2}\right) ^{\tfrac {3}{5}},\quad a\neq b, \]

we can again consider \(\tfrac {a}{b}=t^{3}{\gt}1\) and we have to prove that

\[ \tfrac {t^{3}-1}{2\tan ^{-1}\tfrac {t^{3}-1}{t^{3}+1}}{\lt}\left( \tfrac {t^{5}+1}{2}\right) ^{\tfrac {3}{5}},\quad t{\gt}1. \]

This is equivalent with the condition that the function

\[ h(t)=\tan ^{-1}\tfrac {t^{3}-1}{t^{3}+1}-\tfrac {t^{3}-1}{2^{\tfrac {2}{5}}\left( t^{5}+1\right) ^{\tfrac {3}{5}}} \]

is positive for \(t{\gt}1.\) As \(h(1)=0\) and

\begin{align*} h^{\prime }(t) & =\tfrac {3t^{2}}{t^{6}+1}-\tfrac {3t^{2}\left( t^{2}+1\right) }{2^{\tfrac {2}{5}}\left( t^{5}+1\right) ^{\tfrac {8}{5}}}\\ & =\tfrac {3t^{2}\left[ 2^{\tfrac {2}{5}}\left( t^{5}+1\right) ^{\tfrac {8}{5}}-\left( t^{2}+1\right) \left( t^{6}+1\right) \right] }{2^{\tfrac {2}{5}}\left( t^{5}+1\right) ^{\tfrac {8}{5}}\left( t^{6}+1\right) }, \end{align*}

we have \(h(t){\gt}0\) for \(t{\gt}1\) if \(h^{\prime }(t){\gt}0\) for \(t{\gt}1,\) thus if the function

\[ k(t)=\left[ 2^{\tfrac {2}{5}}\left( t^{5}+1\right) ^{\tfrac {8}{5}}\right] ^{5}-\left[ \left( t^{2}+1\right) \left( t^{6}+1\right) \right] ^{5} \]

is positive for \(t{\gt}1.\) Or this is obvious because

\begin{align*} k(t) & =\left( t-1\right) ^{4}(185t^{28}+200t^{27}+221t^{26}+365t^{24}\\ & \quad +410t^{22}+520t^{19}+580t^{18}+520t^{17}+430t^{16}\\ & \quad +400t^{15}+410t^{14}+440t^{13}+365t^{12}+284t^{11}\\ & \quad +221t^{10}+200t^{9}+185t^{8}+140t^{7}+90t^{6}\\ & \quad +60t^{5}+45t^{4}+25t^{2}+40t^{3}+12t+3). \end{align*}

Remark 2

For the factorization of the polynomials \(g\) and \(k\ \)we have used the computer algebra Maple.â–¡

Remark 3

It is an open problem for us to find a mean \(N\), related to the above mentioned means, with the property that

\[ \mathcal{A}_{5/3}{\lt}N{\lt}\mathcal{A}_{2}. \]

For instance, the mean \(\mathcal{S},\) which is similar to \(\mathcal{I},\) is not convenient as follows from (2).â–¡

Corollary 4

For each \(x\in \left( 0,1\right) \) we have the following evaluations

\[ 1{\lt}\tfrac {x}{\sinh ^{-1}x}{\lt}\mathcal{\mathcal{A}}_{4/3}\left( 1-x,1+x\right) {\lt}\tfrac {x}{\tan ^{-1}x}{\lt}\mathcal{\mathcal{A}}_{5/3}\left( 1-x,1+x\right) . \]

Bibliography

1

B.C. Carlson, The logarithmic mean, Amer. Math. Monthly, 79 (1972), pp. 615–618.

2

I. Costin and G. Toader, A nice evaluation of some Seiffert type means by power means, Int. J. Math. Math. Sc., 2012, Article ID 430692, 6 pages, doi:10.1155/2012/430692.

3

P.A. Hästö, Optimal inequalities between Seiffert’s means and power means, Math. Inequal. Appl., 7 (2004) no. 1, pp. 47–53.

4

A.A. Jagers, Solution of problem 887, Niew Arch. Wisk. (Ser. 4), 12 (1994), pp. 230–231.

5

T.P. Lin, The power and the logarithmic mean, Amer. Math. Monthly, 81 (1974), pp. 879–883.

6

E. Neuman and J. Sándor, On the Schwab-Borchardt mean, Math. Panon., 14 (2003) no. 2, pp. 253–266.

7

E. Neuman and J. Sándor, Comparison inequalities for certain bivariate means, Appl. Anal. Discrete Math., 3 (2009), pp. 46–51.

8

A.O. Pittenger, Inequalities between arithmetic and logarithmic means, Univ. Beograd. Publ. Elektrotehn. Fak. Ser. Mat. Fiz., 678-715 (1980), pp. 15–18.

9

H.-J. Seiffert, Problem 887, Niew Arch. Wisk. (Ser. 4), 11 (1993), pp. 176–176.

10

H.-J. Seiffert, Aufgabe \(\beta 16\), Die Wurzel, 29 (1995), pp. 221–222.