Bullen - Means and Their Inequalities

International Journal of Mathematics and Mathematical Sciences
Means and Their Inequalities
Guest Editors: Mowaffaq Hajja, Peter S. Bullen, Janusz Matkowski, Edward Neuman,
and Slavko Simic
International Journal of Mathematics and
Mathematical Sciences
Guest Editors: Mowaffaq Hajja, Peter S. Bullen,

Janusz Matkowski, Edward Neuman, and Slavko Simic
Copyright 2013 Hindawi Publishing Corporation. All rights reserved.
This is a special issue published in International Journal of Mathematics and Mathematical Sciences. All articles are open access articles
distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any
medium, provided the original work is properly cited.
Editorial Board
The editorial board of the journal is organized into sections that correspond to
the subject areas covered by the journal.
Algebra
A. Ballester-Bolinches, Spain Dalibor Froncek, USA Frank C. Sommen, Belgium
P. Basarab-Horwath, Sweden Heinz P. Gumm, Germany Yucai Su, China
Howard E. Bell, Canada P. Haukkanen, Finland Chun-Lei Tang, China
Martin J. Bohner, USA P. E. Jorgensen, USA Ram U. Verma, USA
Tomasz Brzezinski, UK V. R. Khalilov, Russia Dorothy I. Wallace, USA
Stefaan Caenepeel, Belgium Aloys Krieg, Germany Pei Y. Wu, Taiwan
Ral E. Curto, USA Robert H. Redfield, USA Siamak Yassemi, Iran
David E. Dobbs, USA Alexander Rosa, Canada Kaiming Zhao, Canada
Geometry
Teodor Bulboaca, Romania S. M. Gusein-Zade, Russia Misha Rudnev, UK
A. Cavicchioli, Italy Henryk Hudzik, Poland N. Shanmugalingam, USA
Der-Chen Chang, USA R. Lowen, Belgium Zhongmin Shen, China
Christian Corda, Italy Anil Maheshwari, Canada Nistor Victor, USA
M.-E. Craioveanu, Romania Frederic Mynard, USA Luc Vrancken, France
Jerzy Dydak, USA Hernando Quevedo, Mexico
B. Forster-Heinlein, Germany Frdric Robert, France
Logic and Set Theory

R. Diaconescu, Romania Laurent, Romania Andrzej Skowron, Poland
M. A. Efendiev, Germany Radko Mesiar, Slovakia Sergejs Solovjovs, Czech Republic
S. Gottwald, Germany S. Montes-Rodriguez, Spain Richard Wilson, Mexico
B. Jayaram, India Carles Noguera, Czech Republic
Mathematical Analysis
Asao Arai, Japan Xianguo Geng, China Songxiao Li, China
Martino Bardi, Italy Attila Gilanyi, Hungary Noel G. Lloyd, UK
Peter W. Bates, USA Jerome A. Goldstein, USA Raul F. Manasevich, Chile
H. Begehr, Germany Narendra K. Govil, USA B. N. Mandal, India
Oscar Blasco, Spain R. Grimshaw, UK Ram N. Mohapatra, USA

Wolfgang zu Castell, Germany Weimin Han, USA Manfred Moller, South Africa
Shih-sen Chang, China Seppo Hassi, Finland Enrico Obrecht, Italy
Charles E. Chidume, Nigeria Helge Holden, Norway Gelu Popescu, USA
Hi Jun Choe, Korea Nawab Hussain, Saudi Arabia Jean Michel Rakotoson, France
Rodica D. Costin, USA Petru Jebelean, Romania B. E. Rhoades, USA
Prabir Daripa, USA Shyam Kalla, India Paolo Emilio Ricci, Italy
H. S. V. De Snoo, The Netherlands Hark M. Kim, Republic of Korea Naseer Shahzad, Saudi Arabia
Lokenath Debnath, USA Evgeny Korotyaev, Germany Marianna A. Shubov, USA
Andreas Defant, Germany Irena Lasiecka, USA H. S. Sidhu, Australia
Sever Dragomir, Australia Yuri Latushkin, USA Linda R. Sons, USA
Ricardo Estrada, USA Bao Qin Li, USA Ilya M. Spitkovsky, USA
Marco Squassina, Italy Michael M. Tom, USA Yuxi Zheng, USA
H. M. Srivastava, Canada Ingo Witt, Germany
Peter Takac, Germany A. Zayed, USA
Operations Research
Erik J. Balder, The Netherlands Imed Kacem, France Shey-Huei Sheu, Taiwan
Shih-Pin Chen, Taiwan Yan K. Liu, China Theodore E. Simos, Greece
Tamer Eren, Turkey Wen L. Pearn, Taiwan Frank Werner, Germany
Onesimo H. Lerma, Mexico Mihai Putinar, USA Chin-Chia Wu, Taiwan
Probability and Statistics

Kenneth S. Berenhaut, USA Joseph Hilbe, USA Niansheng Tang, China
J. Dshalalow, USA Vladimir Mityushev, Poland Andrei I. Volodin, Canada
Hans Engler, USA Andrew Rosalsky, USA
Serkan N. Erylmaz, Turkey Gideon Schechtman, Israel
Contents
Means and Their Inequalities, Mowaffaq Hajja, Peter S. Bullen, Janusz Matkowski,
Edward Neuman, and Slavko Simic
Volume 2013, Article ID 698906, 1 page
Some Elementary Aspects of Means, Mowaffaq Hajja

Volume 2013, Article ID 689560, 9 pages
On Some Intermediate Mean Values, Slavko Simic

The Monotonicity Results for the Ratio of Certain Mixed Means and Their Applications,
Zhen-Hang Yang
Refinements of Inequalities among Difference of Means, Huan-Nan Shi, Da-Mao Li,

and Jian Zhang
Complete Moment Convergence of Weighted Sums for Arrays of Rowwise -Mixing Random Variables,
Ming Le Guo
On Huygens Inequalities and the Theory of Means, Jozsef Sandor

A Nice Separation of Some Seiffert-Type Means by Power Means, Iulia Costin and
Gheorghe Toader
Hindawi Publishing Corporation
Volume 2013, Article ID 698906, 1 page
http://dx.doi.org/10.1155/2013/698906
Editorial
Mowaffaq Hajja,1 Peter S. Bullen,2 Janusz Matkowski,3

Edward Neuman,4 and Slavko Simic5
1
Yarmouk University, Irbid, Jordan
2
University of British Columbia, 2329 West Mall, Vancouver, BC, Canada V6T 1Z4
3
Faculty of Mathematics, Computer Science and Econometrics, University of Zielona Gora, Szafrana 4a,
65-516 Zielona Gora, Poland
4
Southern Illinois University, 1220 Lincoln Drive, Carbondale, IL 62901, USA
5
Serbian Academy of Sciences and Arts, Belgrade, Serbia
Correspondence should be addressed to Mowaffaq Hajja; mowhajja@yahoo.com
Received 24 June 2013; Accepted 24 June 2013
Copyright 2013 Mowaffaq Hajja et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
The theory of means has its roots in the work of the Py- arising from forms, making this aspect of the subject of
thagoreans who introduced the harmonic, geometric, and interest to algebraists as well. Extensions of Gausss outstand-
arithmetic means with reference to their theories of music ing discoveries that relate the evaluation of certain elliptic
and arithmetic. Later, Pappus introduced seven other means integrals to iterations of the arithmetic and geometric means
and gave the well-known elegant geometric proof of the that led to the beautiful arithmeticogeometric mean resulted
celebrated inequalities among the harmonic, geometric, and in so many interesting results and lines of research. A quick
arithmetic means. look at the table of contents of the book Pi and the AGM by
Nowadays, the families and types of means that are being J. M. Borwein and P. B. Borwein shows how extensive this
investigated by researchers and the variety of questions that line of research is and also shows that the subject is related to
almost everything.
are being asked about them are beyond the scope of any single
The theory of means has applications in so many other
survey, with the voluminous book Handbook of Means and
diverse fields. Quoting from the preface of the aforemen-
Their Inequalities by P. S. Bullen being the best such reference
tioned book of P. S. Bullen, these include electrostatics, heat
in this direction. The theory of means has grown to occupy a
conduction, chemistry, and even medicine.
prominent place in mathematics with hundreds of papers on This issue contains several papers that pertain to some of
the subject appearing every year. the the aforementioned subjects.
The strong relations and interactions of the theory of One of the papers is an exposition of certain elementary
means with the theories of inequalities, functional equations, aspects of the subject, together with several open problems
and probability and statistics add greatly to its importance. that are within the comprehension of a graduate student. It
Continuous versions of some means and inequalities is hoped that such questions will lead to contributions from
among them tie it with real analysis and the theory of experts and amateurs alike.
integration. The fact that centers of triangles and simplices
Mowaffaq Hajja
can be viewed as means of points in the Euclidean spaces Peter S. Bullen
makes the subject of interest to geometers. Janusz Matkowski
Positivity and copositivity tests in the theory of forms Edward Neuman
naturally give rise to questions on internality tests of means Slavko Simic
http://dx.doi.org/10.1155/2013/689560
Research Article
Some Elementary Aspects of Means
Mowaffaq Hajja
Department of Mathematics, Yarmouk University, Irbid, Jordan
Correspondence should be addressed to Mowaffaq Hajja; mowhajja@yahoo.com
Received 24 December 2012; Accepted 31 March 2013
Academic Editor: Peter Bullen
Copyright 2013 Mowaffaq Hajja. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
We raise several elementary questions pertaining to various aspects of means. These questions refer to both known and newly
introduced families of means, and include questions of characterizations of certain families, relations among certain families,
comparability among the members of certain families, and concordance of certain sequences of means. They also include questions
about internality tests for certain mean-looking functions and about certain triangle centers viewed as means of the vertices. The
questions are accessible to people with no background in means, and it is also expected that these people can seriously investigate,
and contribute to the solutions of, these problems. The solutions are expected to require no more than simple tools from analysis,
algebra, functional equations, and geometry.
1. Definitions and Terminology If M and N are two -dimensional means on J, then we

say that M N if M(1 , . . . , ) N(1 , . . . , ) for all
In all that follows, R denotes the set of real numbers and J J. We say that M < N if M(1 , . . . , ) < N(1 , . . . , )
denotes an interval in R. for all J for which 1 , . . . , are not all equal. This
By a data set (or a list) in a set , we mean a finite subset of exception is natural since M(, . . . , ) and N(, . . . , ) must
in which repetition is allowed. Although the order in which
be equal, with each being equal to . We say that M and N
the elements of a data set are written is not significant, we
are comparable if M N or N M.
sometimes find it convenient to represent a data set in of
A distance (or a distance function) on a set is defined to
size by a point in , the cartesian product of copies of .
We will call a data set = (1 , . . . , ) in R ordered if be any function : [0, ) that is symmetric and
1 . Clearly, every data set in R may be assumed positive definite, that is,
ordered.
A mean of variables (or a -dimensional mean) on J is (, ) = (, ) , , ,
defined to be any function M : J J that has the internal- (4)
ity property (, ) = 0 = .
min {1 , . . . , } M (1 , . . . , ) max {1 , . . . , } (1)
for all in J. It follows that a mean M must have the property Thus a metric is a distance that satisfies the triangle inequality
M(, . . . , ) = for all in J.
Most means that we encounter in the literature, and all (, ) + (, ) (, ) , , , , (5)
means considered below, are also symmetric in the sense that
M (1 , . . . , ) = M ((1) , . . . , () ) (2) a condition that we find too restrictive for our purposes.
for all permutations on {1, . . . , }, and 1-homogeneous in the
sense that 2. Examples of Means
M (1 , . . . , ) = M ((1) , . . . , () ) (3) The arithmetic, geometric, and harmonic means of two pos-
for all permissible R. itive numbers were known to the ancient Greeks; see [1,
2 International Journal of Mathematics and Mathematical Sciences
pp. 8490]. They are usually denoted by A, G, and H, since

respectively, and are defined, for , > 0, by
1/
1 + + 1/
+ lim ( ) = (1 . . . ) (14)
A (, ) = , 0
2
for all 1 , . . . , > 0.
G (, ) = ,
The inequalities (7) can be written as P1 < P0 < P1 .
2 2 These inequalities hold for any number of variables and they
H (, ) = = . follow from the more general fact that P (1 , . . . , ), for
1/ + 1/ +
fixed 1 , . . . , > 0, is strictly increasing with . Power means
(6) are studied thoroughly in [3, Chapter III].
The celebrated inequalities
3. Mean-Producing Distances and
H (, ) < G (, ) < A (, ) , > 0 (7)
Distance Means
were also known to the Greeks and can be depicted in the
It is natural to think of the mean of any list of points in any
well-known figure that is usually attributed to Pappus and
set to be the point that is closest to that list. It is also natural
that appears in [2, p. 364]. Several other less well known
to think of a point as closest to a list of points if the sum of its
means were also known to the ancient Greeks; see [1, pp. 84
distances from these points is minimal. This mode of thinking
90].
associates means to distances.
The three means above, and their natural extensions to
If is a distance on , and if = (1 , . . . , ) is a data set
any number of variables, are members of a large two-
in , then a -mean of is defined to be any element of at
parameter family of means, known now as the Gini means and
which the function
defined by

1/()
( , . . . , ) () = (, ) (15)
, (1 , . . . , ) = ( 1 ) , (8) =1
(1 , . . . , )
where are the Newton polynomials defined by attains its minimum. It is conceivable that (15) attains its min-
imum at many points, or nowhere at all. However, we shall be
mainly interested in distances on J for which (15) attains

(1 , . . . , ) = . (9) its minimum at a unique point that, furthermore, has the
=1
property
Means of the type ,1 are known as Lehmers means, and
those of the type ,0 are known as Holder or power means. min { : } max { : } (16)
Other means that have been studied extensively are the
elementary symmetric polynomial and elementary symmetric for every data set . Such a distance is called a mean-produc-
polynomial ratio means defined by ing or a mean-defining distance, and the point is called the
-mean of or the mean of arising from the distance and
1/ will be denoted by (). A mean M is called a distance mean
/
( ) , , (10) if it is of the form for some distance .
1 /1
where is the th elementary symmetric polynomial in Problem Set 1. (1-a) Characterize those distances on J that are
variables, and where mean-producing.
(1-b) Characterize those pairs of mean producing distan-
ces on J that produce the same mean.
= ( ) . (11)
(1-c) Characterize distance means.
These are discussed in full detail in the encyclopedic work [3,
Chapters III and V]. 4. Examples of Mean-Producing Distances
It is obvious that the power means P defined by
If 0 is the discrete metric defined on R by
1/
1 + +
P (1 , . . . , ) = ,0 (1 , . . . , ) = ( ) 1 if =,
0 (, ) = { (17)
(12) 0 if = ,
that correspond to the values = 1 and = 1 are nothing but then the function () in (15) is nothing but the number
the harmonic and arithmetic means H and A, respectively. It of elements in the given data set that are different from
is also natural to set , and therefore every element having maximum frequency
1/ in minimizes (15) and is hence a 0 -mean of . Thus the
P0 (1 , . . . , ) = G (1 , . . . , ) = (1 . . . ) , (13) discrete metric gives rise to what is referred to in statistics as
International Journal of Mathematics and Mathematical Sciences 3
the mode of . Due to the nonuniqueness of the mode, the with equality if and only if 1 = = = . Thus
discrete metric is not a mean-producing distance. is convex and cannot attain its minimum at more than one
Similarly, the usual metric = 1 defined on R by point. That it attains its minimum follows from the continuity
of (), the compactness of [1 , ], and the obvious fact that
1 (, ) = | | (18) () is increasing on [ , ) and is decreasing on (, 1 ].
If we denote the mean that defines by , then () is the
is not a mean-producing distance. In fact, it is not very diffi- unique zero of
cult to see that if = (1 , . . . , ) is an ordered data set of
even size = 2, then any number in the closed interval
1
[ , +1 ] minimizes sign ( ) , (26)
=1

(19) where sign() is defined to be 1 if is nonnegative and 1
=1
otherwise.
Note that no matter what > 1 is, the two-dimensional
and is therefore a 1 -mean of . Similarly, one can show that
mean arising from is the arithmetic mean. Thus when
if is of an odd size = 2 1, then is the unique 1 -
studying , we confine our attention to the case when the
mean of . Thus the usual metric on R gives rise to what is
number of variables is greater than two. For such , it is
referred to in statistics as the median of .
impossible in general to compute () in closed form.
On the other hand, the distance 2 defined on R by
Problem 2. It would be interesting to investigate comparabil-
2 (, ) = ( )2 (20)
ity among { : > 1}.
is a mean-producing distance, although it is not a metric. In It is highly likely that no two means are comparable.
fact, it follows from simple derivative considerations that the
function 5. Deviation and Sparseness
If is a mean-producing distance on , and if is the
2
( ) (21) associated mean, then it is natural to define the -deviation
=1
D () of a data set = (1 , . . . , ) by an expression like
attains its minimum at the unique point
D () = { ( () , ) : 1 } . (27)

1 Thus if is defined by
= ( ) . (22)
=1
2
(, ) = ( ) , (28)
Thus 2 is a mean-producing distance, and the corresponding
mean is nothing but the arithmetic mean. then is nothing but the arithmetic mean or ordinary
It is noteworthy that the three distances that come to average defined by
mind most naturally give rise to the three most commonly
used means in statistics. In this respect, it is also worth 1 + +
mentioning that a fourth mean of statistics, the so-called = (1 , . . . , ) = , (29)

midrange, will be encountered below as a very natural limiting
distance mean. and D is the (squared) standard deviation (2) given by
The distances 1 and 2 (and in a sense, 0 also) are mem-
bers of the family of distances defined by 2 2
+ +
(2) (1 , . . . , ) = 1 . (30)
(, ) = | | . (23)
In a sense, this provides an answer to those who are puzzled
It is not difficult to see that if > 1, then is a mean-produc-
and mystified by the choice of the exponent 2 (and not any
ing distance. In fact, if = (1 , . . . , ) is a given data set, and
other exponent) in the standard definition of the standard
if
deviation given in the right-hand side of (30). In fact, distance
means were devised by the author in an attempt to remove

() = , (24) that mystery. Somehow, we are saying that the ordinary
=1
average and the standard deviation (2) must be taken
or discarded together, being both associated with the same
then
distance given in (28). Since few people question the

2 sensibility of the definition of given in (29), accepting the
() = ( 1) 0, (25) standard definition of the standard deviation given in (30) as
=1 is becomes a must.
It is worth mentioning that choosing an exponent other = (1 , . . . , ) is just another way of saying that the point
than 2 in (30) would result in an essentially different notion (, . . . , ) is a best approximant in of the point (1 , . . . , )
of deviations. More precisely, if one defines () by with respect to the -norm given in (34). Here, a point in
a subset of a metric (or distance) space (, ) is said to be a
best approximant in of if (, ) = min{(, ) :
() 1 + + (31)
(1 , . . . , ) = , }. Also, a subset of (, ) is said to be Chebyshev if every

in has exactly one best approximant in ; see [4, p. 21].
then () and (2) would of course be unequal, but more The discussion above motivates the following definition.
importantly, they would not be monotone with respect to each
other, in the sense that there would exist data sets and Definition 1. Let J be an interval in R and let be a distance
with (2) () > () () and (2) () < () (). Thus the choice on J . If the diagonal (J ) of J defined by
of the exponent in defining deviations is not as arbitrary as (J ) = {(1 , . . . , ) J : 1 = = } (35)
some may feel. On the other hand, it is (27) and not (31) that
is the natural generalization of (30). This raises the following, is Chebyshev (with respect to ), then the -dimensional
expectedly hard, problem. mean on J defined by declaring (1 , . . . , ) = if
and only if (, . . . , ) is the best approximant of (1 , . . . , ) in
Problem 3. Let be the distance defined by (, ) = (J ) is called the Chebyshev or best approximation -mean
| | , and let the associated deviation D defined in (27) or the best approximation mean arising from .
be denoted by D . Is D monotone with respect to D2 for any
=2, in the sense that In particular, if one denotes by the best approximation
-dimensional mean on R arising from (the distance on R
D () > D () D2 () > D2 ()? (32) induced by) the norm , then the discussion above says
that exists for all > 1 and that it is equal to defined
We end this section by introducing the notion of sparse-
in Section 4.
ness and by observing its relation with deviation. If is a
In view of this, one may also define to be the best
mean-producing distance on J, and if is the associated
approximation mean arising from the -norm of , that is,
mean, then the -sparseness S () of a data set =
the norm defined on R by
(1 , . . . , ) in J can be defined by

(1 , . . . , ) = max { : 1 } . (36)
S () = { ( , ) : 1 < } . (33)
It is not very difficult to see that () is nothing but what
It is interesting that when is defined by (28), the standard
is referred to in statistics as the mid-range of . Thus if =
deviation coincides, up to a constant multiple, with the
(1 , . . . , ) is an ordered data set, then
sparsenss. One wonders whether this pleasant property char-
acterizes this distance . 1 +
() = . (37)
2
Problem Set 4. (4-a) Characterize those mean-producing dis-
tances whose associated mean is the arithmetic mean. In view of the fact that cannot be defined by anything like
(4-b) If is as defined in (28), and if is another mean- (23) and is thus meaningless, natural question arises as to
producing distance whose associated mean is the arithmetic whether
mean, does it follow that D and D are monotone with
respect to each other? () = lim () (or equivalently = lim ())

(4-c) Characterize those mean-producing distances for
(38)
which the deviation D () is determined by the sparseness
S () for every data set , and vice versa. for every . An affirmative answer is established in [5,
Theorem 1]. In that theorem, it is also established that
6. Best Approximation Means
lim () (or equivalently lim ()) = ()
It is quite transparent that the discussion in the previous sec-
tion regarding the distance mean , > 1, can be written (39)
in terms of best approximation in , the vector space R
endowed with the -norm defined by for all and all . All of this can be expressed by saying that
is continuous in for (1, ] for all .

1/ We remark that there is no obvious reason why (38)

(1 , . . . , ) = ( ) . (34) should immediately follow from the well known fact that
=1
lim = (40)

If we denote by = the line in R consisting of the
points (1 , . . . , ) with 1 = = , then to say that for all points in R .
Problem Set 5. Suppose that is a sequence of distances on (7-a) Explore how the value of () compares with the
a set that converges to a distance (in the sense that common practice of taking the median of to be the
lim (, ) = (, ) for all , in ). Let . midpoint of the median interval (defined in (42) for
various values of .
(5-a) If is Chebyshev with respect to each , is it
necessarily true that is Chebyshev with respect to (7-b) Is continuous on R ? If not, what are its points of
? discontinuity?
(5-b) If is Chebyshev with respect to each and with
(7-c) Given R , is the convergence of () (as
respect to and if is the best approximant in of
with respect to and is the best approximant decreases to 1) to () monotone?
in of with respect to , does it follow that
The convergence of () (as decreases to 1) to ()
converges to ?
is described in [5, Theorem 4], where it is proved that the
We end this section by remarking that if = is convergence is ultimately monotone. It is also proved in
the -dimensional best approximation mean arising from a [5, Theorem 5] that when = 3, then the convergence is
distance on J , then is significant only up to its values of monotone.
the type (, V), where (J ) and V (J ). Other values It is of course legitimate to question the usefulness of
of are not significant. This, together with the fact that defining the median to be , but that can be left to statis-
ticians and workers in relevant disciplines to decide. It is also
every mean is a best approximation mean arising legitimate to question the path that we have taken the limit
(41) along. In other words, it is conceivable that there exists, in
from a metric, addition to , a sequence of distances on R that converges
makes the study of best approximation means less interesting. to 1 such that the limit , as decreases to 1, of their
Fact (41) was proved in an unduly complicated manner in associated distance means is not the same as the limit of
[6], and in a trivial way based on a few-line set-theoretic . In this case, would have as valid a claim as to being
argument in [7]. the median. However, the naturality of may help accepting
as a most legitimate median.
Problem 6. Given a mean M on J, a metric on J is
constructed in [6] so that M is the best approximation Problem Set 8. Suppose that and , N, are sequences
mean arising from . Since the construction is extremely of distances on a set that converge to the distances
complicated in comparison with the construction in [7], it is and
, respectively (in the sense that lim (, ) =
desirable to examine the construction of in [6] and see what (, ) for all , in , etc.).
other nice properties (such as continuity with respect to the
usual metric) has. This would restore merit to the construc- (8-a) If each , N, is mean producing with corre-
tion in [6] and to the proofs therein and provide raison detre sponding mean , does it follow that is mean
for the so-called generalized means introduced there. producing? If so, and if the mean produced by is
, is it necessarily true that converges to ?
7. Towards a Unique Median
(8-b) If and , N {}, are mean producing
As mentioned earlier, the distance 1 on R defined by (23)
does not give rise to a (distance) mean. Equivalently, the 1- distances with corresponding means and , and
norm 1 on R defined by (34) does not give rise to a if = for all N, does it follow that =

(best approximation) mean. These give rise, instead, to the ?
many-valued function known as the median. Thus, following
the statisticians mode of thinking, one may set
8. Examples of Distance Means
1 () = 1 () = the median interval of
(42) It is clear that the arithmetic mean is the distance mean
= the set of all medians of . arising from the the distance 2 given by 2 (, ) = ( )2 .
From a mathematicians point of view, however, this leaves a Similarly, the geometric mean on the set of positive numbers
lot to be desired, to say the least. The feasibility and naturality is the distance mean arising from the distance G given by
of defining as the limit of as approaches gives
us a clue on how the median 1 may be defined. It is a G (, ) = (ln ln )2 . (43)
pleasant fact, proved in [5, Theorem 4], that the limit of ()
(equivalently of ()) as decreases to 1 exists for every In fact, this should not be amazing since the arithmetic mean
R and equals one of the medians described in (42). This A on R and the geometric mean G on (0, ) are equivalent
limit can certainly be used as the definition of the median. in the sense that there is a bijection : (0, ) R, namely
() = ln , for which G(, ) = 1 A((), ()) for all
Problem Set 7. Let be as defined in Section 4, and let be , . Similarly, the harmonic and arithmetic means on (0, )
the limit of as decreases to 1. are equivalent via the bijection () = 1/, and therefore
the harmonic mean is the distance mean arising from the deviation, and if 1 , . . . , are given, then the -deviation
distance H given by mean of 1 , . . . , is defined to be the unique zero of
1 1 2 (1 , ) + + ( , ) . (49)
H (, ) = ( ) . (44)

It is direct to see that (49) has a unique zero and that this zero
The analogous question pertaining to the logarithmic mean does indeed define a mean.
L defined by
Problem 11. Characterize deviation means and explore their
exact relationship with distance means.
L (, ) = , , > 0, (45)
ln ln If is a deviation, then (following [11]), one may define
by
remains open.

Problem 9. Decide whether the mean L (defined in (45)) is (, ) = (, ) . (50)

a distance mean.
Then (, ) 0 and (, ) is a strictly convex function in
9. Quasi-Arithmetic Means for every . The -deviation mean of 1 , . . . , is nothing but
the unique value of at which (1 , ) + + ( , ) attains
A -dimensional mean M on J is called a quasi-arithmetic its minimum. Thus if happens to be symmetric, then
mean if there is a continuous strictly monotone function would be a distance and the -deviation mean would be the
from J to an interval I in R such that distance mean arising from the distance .
M (1 , . . . , ) = 1 (A ( (1 ) , . . . , ( ))) (46)
11. Other Ways of Generating New Means
for all in J. We have seen that the geometric and harmonic If and are differentiable on an open interval J, and if <
means are quasi-arithmetic and concluded that they are are points in J such that () =(), then there exists, by
distance means. To see that L is not quasi-arithmetic, we Cauchys mean value theorem, a point in (, ), such that
observe that the (two-dimensional) arithmetic mean, and
hence any quasi-arithmetic mean M, satisfies the elegant () () ()
functional equation = . (51)
() () ()
M (M (M (, ) , ) , M (M (, ) , )) = M (, ) (47) If and are such that is unique for every , , then we call
the Cauchy mean of and corresponding to the functions
for all , > 0. However, a quick experimentation with a
and , and we denote it by C, (, ).
random pair (, ) shows that (47) is not satisfied by L.
Another natural way of defining means is to take a
This shows that L is not quasi-arithmetic, but does not
continuous function that is strictly monotone on J, and to
tell us whether L is a distance mean, and hence does not
define the mean of , J, =, to be the unique point in
answer Problem 9.
(, ) such that
The functional equation (47) is a weaker form of the
functional equation 1
() = () . (52)
M (M (, ) , M (, )) = M (M (, ) , M (, )) (48)
We call the mean value (mean) of and corresponding to
for all , , , > 0. This condition, together with the
, and we denote it by V(, ).
assumption that M is strictly increasing in each variable,
Clearly, if is an antiderivative of , then (53) can be
characterizes two-dimensional quasi-arithmetic means; see
written as
[8, Theorem 1, pp. 287291]. A thorough discussion of quasi-
arithmetic means can be found in [3, 8]. () ()
() = . (53)

Problem 10. Decide whether a mean M that satisfies the func-
tional equation (47) (together with any necessary smoothness Thus V (, ) = C, (, ), where is the identity function.
conditions) is necessarily a quasi-arithmetic mean. For more on the these two families of means, the reader
is referred to [12] and [13], and to the references therein.
10. Deviation Means In contrast to the attitude of thinking of the mean as the
number that minimizes a certain function, there is what one
Deviation means were introduced in [9] and were further may call the Chisini attitude that we now describe. A function
investigated in [10]. They are defined as follows. on J may be called a Chisini function if and only if the
A real-valued function = (, ) on R2 is called a equation
deviation if (, ) = 0 for all and if (, ) is a strictly
decreasing continuous function of for every . If is a (1 , . . . , ) = (, . . . , ) (54)
has a unique solution = [1 , ] for every ordered data A form = (1 , . . . , ) is said to be copositive if
set (1 , . . . , ) in J. This unique solution is called the Chisini (1 , . . . , ) 0 for all 0. Copositive forms arise
mean associated to . In Chisinis own words, is said to be in the theory of inequalities and are studied in [14] (and in
the mean of numbers 1 , . . . , with respect to a problem, references therein). One of the interesting questions that one
in which a function of them (1 , . . . , ) is of interest, if the may ask about forms pertains to algorithms for deciding
function assumes the same value when all the are replaced whether a given form is copositive. This problem, in full
by the mean value : (1 , . . . , ) = (, . . . , ); see [14, page generality, is still open. However, for quadratic and cubic
256] and [1]. Examples of such Chisini means that arise in forms, we have the following satisfactory answers.
geometric configurations can be found in [15].
Theorem 2. Let = (1 , . . . , ) be a real symmetric form
()
Problem 12. Investigate how the families of distance, devia- in any number 2 of variables. Let v , 1 , be the
tion, Cauchy, mean value, and Chisini means are related. -tuple whose first coordinates are 1s and whose remaining
coordinates are 0 s.
12. Internality Tests
(i) If is quadratic, then is copositive if and only if 0
According to the definition of a mean, all that is required of a at the two test -tuples
function M : J J to be a mean is to satisfy the internality
property k1() = (1, 0, . . . , 0) , k() = (1, 1, . . . , 1) . (60)
min {1 , . . . , } M (1 , . . . , ) max {1 , . . . , } (55)
(ii) If is cubic, then is copositive if and only if 0 at
for all J. However, one may ask whether it is sufficient, the test -tuples
for certain types of functions M, to verify (55) for a finite,
preferably small, number of well-chosen -tuples. This ques- ()
k = (1, . . . , 1, 0, . . . , 0) , 1 . (61)
tion is inspired by certain elegant theorems in the theory of
copositive forms that we summarize below.
Part (i) is a restatement of Theorem 1(a) in [16]. Theo-
12.1. Copositivity Tests for Quadratic and Cubic Forms. By a rem 1(b) there is related and can be restated as
(real) form in variables, we shall always mean a homoge-
neous polynomial = (1 , . . . , ) in the indeterminates (1 , . . . , ) 0, R,
1 , . . . , having coefficients in R. When the degree of a
form is to be emphasized, we call a -form. Forms of 0 at the 3 -tuples (62)
degrees 1, 2, 3, 4, and 5 are referred to as linear, quadratic, (1, 0, . . . , 0) , (1, 1, . . . , 1) , (1, 1, 0, . . . , 0) .
cubic, quartic, and quintic forms, respectively.
The set of all -forms in variables is a vector space (over Part (ii) was proved in [17] for 3 and in [18] for all . Two
R) that we shall denote by F() . It may turn out to be an very short and elementary inductive proofs are given in [19].
interesting exercise to prove that the set It is worth mentioning that the test -tuples in (61)
do not suffice for establishing the copositivity of a quartic
{ } form even when = 3. An example illustrating this that
{ : = } (56)
uses methods from [20] can be found in [19]. However, an
{ =1 =1
} algorithm for deciding whether a symmetric quartic form
is a basis, where is the Newton polynomial defined by in variables is copositive that consists in testing at -tuples

of the type

= . (57)

=1 (
, . . . , , 1, . . . , 1, 0, . . . , 0) ,
The statement above is quite easy to prove in the special case (63)
3, and this is the case we are interested in in this paper. 0 , , +
We also discard the trivial case = 1 and assume always that
2. is established in [21]. It is also proved there that if = 3, then
Linear forms can be written as 1 , and they are not the same algorithm works for quintics but does not work for
worth much investigation. Quadratic forms can be written as forms of higher degrees.
2
= 12 + 2 = ( ) + ( 2 ) . (58) 12.2. Internality Tests for Means Arising from Symmetric
=1 =1 Forms. Let F() be the vector space of all real -forms in
variables, and let , 1 , be the Newton polynomials
Cubic and quartic forms can be written, respectively, as
defined in (57). Means of the type
13 + 1 2 + 3 ,
(59) 1/()
M=( ) , (64)
14 + 12 2 + 1 3 + 22 .
where is a symmetric form of degree , are clearly sym- 1< ( )2 , then is internal if and only
metric and 1-homogeneous, and they abound in the literature. if it is internal at the two test -tuples k() = (1, 1, . . . , 1)
These include the family of Gini means , defined in (8) ()
and k1 = (1, 1, . . . , 1, 0). In the general case, sufficient and
(and hence the Lehmer and Holder means). They also include
the elementary symmetric polynomial and elementary sym- necessary conditions for internality of , in terms of
metric polynomial ratio means defined earlier in (10). the coefficients of and , are found in [23, Theorem 3].
In view of Theorem 2 of the previous section, it is tempt- However, it is not obvious whether these conditions can be
ing to ask whether the internality of a function M of the type rewritten in terms of test -tuples in the manner done in
described in (64) can be established by testing it at a finite Theorem 3.
set of test -tuples. Positive answers for some special cases of
(64), and for other related types, are given in the following 13. Extension of Means, Concordance
theorem. of Means
Theorem 3. Let , , and be real symmetric forms of degrees The two-dimensional arithmetic mean A(2) defined by
1, 2, and 3, respectively, in any number 2 of nonnegative 1 + 2
variables. Let v() , 1 , be as defined in Theorem 2. A(2) (1 , 2 ) = (67)
2
(i) is internal if and only if it is internal at the two test can be extended to any dimension by setting
()
-tuples: k() = (1, 1, . . . , 1) and V1 = (1, 1, . . . , 1, 0). 1 + +
A() (1 , . . . , ) = . (68)
(ii) / is internal if and only if it is internal at the two test
-tuples: k() = (1, 1, . . . , 1) and V1() = (1, 0, . . . , 0). Although very few people would disagree on this, nobody
can possibly give a mathematically sound justification of the
(iii) If 4, then is internal if and only if it is internal
3
feeling that the definition in (68) is the only (or even the best)
at the test -tuples
definition that makes the sequence () of means harmonious
or concordant. This does not seem to be an acceptable defini-
()
k = (1, . . . , 1, 0, . . . , 0) , 1 . (65) tion of the notion of concordance.
In a private communication several years ago, Professor
Zsolt Pales told me that Kolmogorov suggested calling a
Parts (i) and (ii) are restatements of Theorems 3 and 5 in sequence M() of means on J, where M() is -dimensional,
[16]. Part (iii) is proved in [22] in a manner that leaves a lot to concordant if for every and and every , in J, we have
be desired. Besides being rather clumsy, the proof works for
4 only. The problem for 5, together with other open M(+) (1 , . . . , , 1 , . . . , )
problems, is listed in the next problem set. (69)
= M(2) (M() (1 , . . . , ) , M (1 , . . . , )) .
Problem Set 13. Let , , and be real symmetric cubic forms
of degrees 1, 2, and 3, respectively, in non-negative variables. He also told me that such a definition is too restrictive and
seems to confirm concordance in the case of the quasi-arith-
(13-a) Prove or disprove that 3 is internal if and only if it metic means only.
is internal at the test -tuples
Problem 14. Suggest a definition of concordance, and test it
on sequences of means that you feel concordant. In particular,
()
k = (1, . . . , 1, 0, . . . , 0) , 1 . (66) test it on the existing generalizations, to higher dimensions,
of the logarithmic mean L defined in (45).
(13-b) Find, or prove the nonexistence of, a finite set of
test -tuples such that the internality of / at the -
14. Distance Functions in Topology
tuples in gurantees its internality at all nonnegative Distance functions, which are not necessarily metrics, have
-tuples. appeared early in the literature on topology. Given a distance
(13-c) Find, or prove the nonexistence of, a finite set of function on any set , one may define the open ball (, )
test -tuples such that the internality of at in the usual manner, and then one may declare a subset
the -tuples in guarantees its internality at all non- open if it contains, for every , an open ball (, ) with
negative -tuples. > 0. If has the triangle inequality, then one can proceed in
the usual manner to create a topology. However, for a general
Problem (13-b) is open even for = 2. In Section 6 of [15], distance , this need not be the case, and distances that give
it is shown that the two pairs (1, 0) and (1, 1) do not suffice as rise to a coherent topology in the usual manner are called
test pairs. semimetrics and they are investigated and characterized in
As for Problem (13-c), we refer the reader to [23], [2429]. Clearly, these are the distances for which the family
where means of the type were considered. It is {(, ) : > 0} of open balls centered at forms a local
proved in Theorem 2 there that when has the special form base at for every in .
15. Centers and Center-Producing Distances [12] M. E. Mays, Functions which parametrize means, The Ameri-
can Mathematical Monthly, vol. 90, no. 10, pp. 677683, 1983.
A distance may be defined on any set whatsoever. In [13] B. Ebanks, Looking for a few good means, American Mathe-
particular, if is a distance on R2 and if the function () matical Monthly, vol. 119, no. 8, pp. 658669, 2012.
defined by [14] M. Hall, and M. Newman, Copositive and completely positive
quadratic forms, Proceedings of the Cambridge Philosophical

Society, vol. 59, pp. 329339, 1963.
() = (, ) (70)
[15] R. Abu-Saris and M. Hajja, Geometric means of two positive
=1
numbers, Mathematical Inequalities & Applications, vol. 9, no.
attains its minimum at a unique point 0 that lies in the 3, pp. 391406, 2006.
convex hull of { 1 , . . . , } for every choice of 1 , . . . , in [16] M. Hajja, Radical and rational means of degree two, Mathe-
matical Inequalities & Applications, vol. 6, no. 4, pp. 581593,
R2 , then will be called a center-producing distance.
2003.
The Euclidean metric 1 on R2 produces the Fermat-
[17] J. F. Rigby, A method of obtaining related triangle inequalities,
Torricelli center. This is defined to be the point whose distan- with applications, Univerzitet u Beogradu. Publikacije Elek-
ces from the given points have a minimal sum. Its square, trotehnickog Fakulteta. Serija Matematika i Fizika, no. 412460,
2 , which is just a distance but not a metric, produces the pp. 217226, 1973.
centroid. This is the center of mass of equal masses placed at [18] M. D. Choi, T. Y. Lam, and B. Reznick, Even symmetric sextics,
the given points. It would be interesting to explore the centers Mathematische Zeitschrift, vol. 195, no. 4, pp. 559580, 1987.
defined by for other values of . [19] M. Hajja, Copositive symmetric cubic forms, The American
Mathematical Monthly, vol. 112, no. 5, pp. 462466, 2005.
Problem 15. Let , > 1, be the distance defined on R2 by [20] K. B. Stolarsky, The power and generalized logarithmic means,
(, ) = , and let be a triangle. Let = American Mathematical Monthly, vol. 87, pp. 545548, 1980.
(, , ) be the point that minimizes [21] W. R. Harris, Real even symmetric ternary forms, Journal of
Algebra, vol. 222, no. 1, pp. 204245, 1999.
(, ) + (, ) + (, ) [22] R. Abu-Saris and M. Hajja, Internal cubic symmetric forms in
(71) a small number of variables, Mathematical Inequalities & Appli-
= + + . cations, vol. 10, no. 4, pp. 863868, 2007.
[23] R. Abu-Saris and M. Hajja, Quadratic means, Journal of Math-
Investigate how , 1, are related to the known triangle ematical Analysis and Applications, vol. 288, no. 1, pp. 299313,
centers, and study the curve traced by them. 2003.
[24] A. V. Arhangelskii, Mappings and spaces, Russian Mathemat-
The papers [30, 31] may turn out to be relevant to this ical Surveys, vol. 21, no. 4, pp. 115162, 1966.
problem. [25] F. Galvin and S. D. Shore, Completeness in semimetric spaces,
Pacific Journal of Mathematics, vol. 113, no. 1, pp. 6775, 1984.
References [26] F. Galvin and S. D. Shore, Distance functions and topologies,
The American Mathematical Monthly, vol. 98, no. 7, pp. 620623,
[1] T. Heath, A History of Greek Mathematics, vol. 1, Dover, New 1991.
York, NY, USA, 1981. [27] R. Kopperman, All topologies come from generalized metrics,
[2] T. Heath, A History of Greek Mathematics, vol. 2, Dover, New American Mathematical Monthly, vol. 95, no. 2, pp. 8997, 1988.
York, NY, USA, 1981. [28] D. E. Sanderson and B. T. Sims, A characterization and gen-
[3] P. S. Bullen, Handbook of Means and Their Inequalities, vol. eralization of semi-metrizability, The American Mathematical
560, Kluwer Academic Publishers, Dordrecht, The Netherlands, Monthly, vol. 73, pp. 361365, 1966.
2003. [29] W. A. Wilson, On semi-metric spaces, American Journal of
[4] F. Deutsch, Best Approximation in Inner Product Spaces, Mathematics, vol. 53, no. 2, pp. 361373, 1931.
Springer, New York, NY, USA, 2001. [30] D. S. Mitrinovic, J. E. Pecaric, and V. Volenec, The generalized
[5] A. Al-Salman and M. Hajja, Towards a well-defined median, Fermat-Torricelli point and the generalized Lhuilier-Lemoine
Journal of Mathematical Inequalities, vol. 1, no. 1, pp. 2330, 2007. point, Comptes Rendus Mathematiques, vol. 9, no. 2, pp. 95100,
1987.
[6] F. B. Saidi, Generalized deviations and best approximation the-
ory, Numerical Functional Analysis and Optimization, vol. 26, [31] P. Penning, Expoints, Nieuw Archief voor Wiskunde, vol. 4, no.
no. 2, pp. 271283, 2005. 1, pp. 1931, 1986.
[7] M. Hajja, Distance and best approximation means, preprint.
[8] J. Aczel and J. Dhombres, Functional Equations in Several Varia-
bles, vol. 31, Cambridge University Press, Cambridge, UK, 1989.

[9] Z. Daroczy, Uber eine Klasse von Mittelwerten, Publicationes
Mathematicae Debrecen, vol. 19, pp. 211217, 1972.
[10] Z. Pales, On the convergence of means, Journal of Mathemat-
ical Analysis and Applications, vol. 156, no. 1, pp. 5260, 1991.
[11] Z. Pales, Private Communications.
http://dx.doi.org/10.1155/2013/283127
Research Article
On Some Intermediate Mean Values
Slavko Simic
Mathematical Institute SANU, Kneza Mihaila 36, 11000 Belgrade, Serbia
Correspondence should be addressed to Slavko Simic; ssimic@turing.mi.sanu.ac.rs
Received 25 June 2012; Revised 9 December 2012; Accepted 16 December 2012
Academic Editor: Mowaffaq Hajja
Copyright 2013 Slavko Simic. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
We give a necessary and sufficient mean condition for the quotient of two Jensen functionals and define a new class , (, ) of
mean values where , are continuously differentiable convex functions satisfying the relation () = (), R+ . Then we
asked for a characterization of , such that the inequalities (, ) , (, ) (, ) or (, ) , (, ) (, ) hold for
each positive , , where , , , are the harmonic, arithmetic, logarithmic, and identric means, respectively. For a subclass of
with () = , R, this problem is thoroughly solved.
1. Introduction are the harmonic, geometric, logarithmic, identric, arith-

metic, and Gini mean, respectively.
It is said that the mean is intermediate relating to the means An easy task is to construct intermediate means related to
and , if the relation two given means and with . For instance, for an
arbitrary mean , we have that
(, ) (, ) (, ) (1)
(, ) ( (, ) , (, )) (, ). (4)
holds for each two positive numbers , .
It is also well known that The problem is more difficult if we have to decide whether
the given mean is intermediate or not. For example, the
min {, } (, ) (, ) relation
(, ) (, ) (, ) (, ) (2) (, ) (, ) (, ) (5)
max {, } ,
holds for each positive and if and only if 0 1, where
where the Stolarsky mean is defined by (cf [1])
1/(1)
1 1 1
= (, ) := 2( + ) ; (, ) := ( ) . (6)
( )

= (, ) := ; = (, ) := ; Also,
log log
(, ) (, ) (, ) (7)
1/()
( / )
= (, ) := ; holds if and only if 0 1, where the Holder mean of order
is defined by
+
= (, ) := ; = (, ) := /(+) /(+) +
1/
2 (, ) := ( ) . (8)
(3) 2
An inverse problem is to find best possible approximation or, more tightly,

of a given mean by elements of an ordered class of means
(, ) , (, ) (, ), (17)
. A good example for this topic is comparison between the
logarithmic mean and the class of Holder means of order hold for each , R+ ?
. Namely, since 0 = lim 0 = and 1 = , it follows As an illustration, consider the function () defined to
from (2) that be
0 1. (9) + 1
{
{ , ( 1) =
0;
{ ( 1)
Since is monotone increasing in , an improving of the () = { log 1, = 0; (18)
{
{
above is given by Carlson [2]: + 1, = 1.
{ log
0 1/2 . (10) Since
Finally, Lin showed in [3] that { 1 1

{
{ , ( 1) =
0;
{
{ 1
{
{
0 1/3 (11) () = { l
{
{ 1 , = 0;
{
{ (19)
is the best possible approximation of the logarithmic mean by {
{
the means from the class . {log , = 1,
Numerous similar results have been obtained recently.
For example, an approximation of Seifferts mean by the class () = 2 , R, > 0,
is given in [4, 5].
it follows that () is a twice continuously differentiable
In this paper we will give best possible approximations
convex function for R, R+ .
for a whole variety of elementary means (2) by the class
Moreover, it is evident that (+1 , ) .
defined below (see Theorem 5).
We will give in the sequel a complete answer to the above
Let , be twice continuously differentiable (strictly)
question concerning the means
convex functions on R+ . By definition (cf [6], page 5),
+1 (, )
+ := (, ) (20)
(, ) := () + () 2 ( ) > 0, =
, (, )
2 (12)
defined by
(, ) = 0,
(, )
if and only if = .
+1 +1 +1
It turns out that the expression { 1 + 2(( + ) /2)
{
{ + 1 + 2(( + ) /2)
, R/ {1, 0, 1} ;
{
{
{
{
(, ) () + () 2 (( + ) /2) {
{ 2 log (( + ) /2) log log
, (, ) := = {
{ , = 1;
(, ) () + () 2 (( + ) /2) { 1/2 + 1/2 2/ ( + )
={
(13) {
{ log + log ( + ) log (( + ) /2)
{
{ , = 0;
{
{ 2 log (( + ) /2) log log
represents a mean of two positive numbers , ; that is, the {
{
{
{ ( )2
relation { , = 1.
{ 4 ( log + log ( + ) log (( + ) /2))
min {, } , (, ) max {, } (21)
(14)
Those means are obviously symmetric and homogeneous
holds for each , R+ , if and only if the relation of order one.
As a consequence we obtain some new intermediate mean
() = () (15) values; for instance, we show that the inequalities
holds for each R+ . (, ) 1 (, ) (, ) 0 (, ) (, )

Let , (0, ) and denote by the set {(, )} of (22)
1 (, ) (, )
convex functions satisfying the relation (15). There is a natural
question how to improve the bounds in (14); in this sense we hold for arbitrary , R+ . Note that
come upon the following intermediate mean problem.
22 log (/) log (/)
Open Question. Under what additional conditions on , 1 = ; 0 = ;
log (/)
, the inequalities (23)
1
1 = .
(, ) , (, ) (, ), (16) 2 log (/)
2. Results 3. Proofs
We prove firstly the following 3.1. Proof of Theorem 1. We prove firstly the necessity of the
condition (15).
Theorem 1. Let , 2 () with > 0. The expression Since , (, ) is a mean value for arbitrary , ;
, (, ) represents a mean of arbitrary numbers , if = , we have
and only if the relation (15) holds for .
min {, } , (, ) max {, } . (28)
Remark 2. In the same way, for arbitrary , > 0, + = 1,
Hence
it can be deduced that the quotient
lim , (, ) = . (29)
() + () ( + )
, (, ; , ) := (24)
() + () ( + ) From the other hand, due to lHospitals rule we obtain
represents a mean value of numbers , if and only if (15) () (( + ) /2)
holds. lim , (, ) = lim ( )
() (( + ) /2)
A generalization of the above assertion is the next. 2 () (( + ) /2)
= lim ( ) (30)
2 () (( + ) /2)
Theorem 3. Let , : R be twice continuously
differentiable functions with > 0 on and let = { }, ()
= 1, 2, . . . , = 1 be an arbitrary positive weight sequence. = .
Then the quotient of two Jensen functionals ()
Comparing (29) and (30) the desired result follows.
1 ( ) (1 )
, (, ) := , 2, (25) Suppose now that (15) holds and let < . Since () >
1 ( ) (1 ) 0 [, ] by the Cauchy mean value theorem there exists
(( + )/2, ) such that
represents a mean of an arbitrary set of real numbers
1 , 2 , . . . , if and only if the relation () (( + ) /2) ()
= = . (31)
() = () (26) () (( + ) /2) ()
But,
holds for each .
+
Remark 4. It should be noted that the relation () = () < < , (32)
2
determines in terms of in an easy way. Precisely,
and, since is strictly increasing, () ((+)/2) > 0,
() = () 2 () + + , (27) [, ].

Therefore, by (31) we get
where () := 1 () and and are constants.
+ +
( () ( )) () ( )
Our results concerning the means (, ), R are 2 2
(33)
included in the following. +
( () ( )) .
2
Theorem 5. For the class of means (, ) defined above, the
following assertions hold for each , R+ . Finally, integrating (33) over [, ] we obtain the assertion
from Theorem 1.
(1) The means (, ) are monotone increasing in ;
(2) (, ) (, ) for each 4; 3.2. Proof of Theorem 3. We will give a proof of this assertion
(3) (, ) (, ) (, ) for 3 1; by induction on .
By Remark 2, it holds for = 2.
(4) (, ) (, ) (, ) for 1/2 0;
Next, it is not difficult to check the identity
(5) there is a number 0 (1/12, 1/11) such that (, )

(, ) (, ) for 0 1;
( ) ( )
(6) there is a number 1 (1.03, 1.04) such that (, ) 1 1
(, ) (, ) for 1 2;
1 1
(7) (, ) (, ) (, ) for each 2 5; = (1 ) ( ( ) ( ))
(8) there is no finite such that the inequality (, ) 1 1
(, ) holds for each , R+ . + [(1 ) () + ( ) ((1 ) + )] ,
The above estimations are best possible. (34)
where 3.3. Proof of Theorem 5, Part (1). We will prove a general

1
assertion of this type. Namely, for an arbitrary positive
:= ; := , = 1, 2, . . . , 1; sequence x = { } and an associated weight sequence p =
1 (1 ) { }, = 1, 2, . . ., denote
(35)
1
= 1.
1 (p, x)

Therefore, by induction hypothesis and Remark 2, we get
{ ( )
{
{ , R/ {0, 1} ;
{
{ ( 1)
( ) ( ) {
:= {log ( ) log , = 0;
1 1 {
{
{
{
{ log ( ) log ( ) , = 1.
max {1 , 2 , . . . 1 } (1 )
{
1 1 (40)
( ( ) ( ))
1 1
For R, > 0 we have
+ max {, } [(1 ) () + ( )
((1 ) + )] (p, x) ++1 (p, x) +1 (p, x) + (p, x) , (41)

max {1 , 2 , . . . , }
1 1 which is equivalent to
((1 ) ( ( ) ( ))
1 1
Theorem 8. The sequence {+1 (p, x)/ (p, x)} is monotone
+ [(1 ) () + ( ) ((1 ) + )] ) increasing in , R.
This assertion follows applying the result from [7, Theo-
rem 2] which states the following.

= max {1 , 2 , . . . , } ( ( ) ( )) . Lemma 9. For < < < < +, the inequality
1 1
(36)
The inequality
( (p, x)) ( (p, x)) ( (p, x)) (42)
min {1 , 2 , . . . , } , (, ) (37)
can be proved analogously. holds for arbitrary sequences p, x.
For the proof of necessity, put 2 = 3 = = and
proceed as in Theorem 1. Putting there = , = + 1, = + + 1 and = ,
= + , = + + 1, we successively obtain
Remark 6. It is evident from (15) that if R+ then has to
be also convex on . Otherwise, it shouldnt be the case. For
example, the conditions of Theorem 3 are satisfied with () = (+1 (p, x))
+1
( (p, x)) ++1 (p, x),
3 /3, () = 2 , R. Hence, for an arbitrary sequence { }1 (43)
of real numbers, we obtain +1
(+ (p, x)) (p, x) (++1 (p, x)) .
3
1 3 (1 )
min {1 , 2 , . . . , } 2
3 (1 2 (1 ) ) (38) Since > 0, multiplying those inequalities we get the
relation (41), that is, the proof of Theorem 8.
max {1 , 2 , . . . , } . The part (1) of Theorem 5 follows for 1 = 2 = 1/2.
Because the above inequality does not depend on , a A general way to prove the rest of Theorem 5 is to use an
probabilistic interpretation of the above result is contained in easy-checkable identity
the following.
Theorem 7. For an arbitrary probability law of random (, )

= (1 + , 1 ) , (44)
variable with support on (, +), one has (, )
()3 + 3 (min )
2
3 ()3 + 3 (max )
2
.
(39) with := ( )/( + ).
Since 0 < < , we get 0 < < 1. Also, we get

(, ) (, ) ()
= 1 2 ; = 1 2 ;
(, ) (, ) 2
(, ) 2
1
1
= ; = ( + ) 2
(, ) log (1 + ) log (1 ) + 1 =0 (2 2 + 1) ( + 1) (2 + 1)
=0
(, )
(, ) = 2 .
0
(1 + ) log (1 + ) (1 ) log (1 ) (51)
= exp ( 1) ;
2
Hence,
(, )
1
(, ) 0 = 1 = 0; 2 = , (52)
90
1
= exp ( ((1 + ) log (1 + ) + (1 ) log (1 ))) . and, after some calculation, we get
2
(45)
2 1 1
= (( + 2) ( + 1) ) ,
Therefore, we have to compare some one-variable ( + 1) (2 + 3) 1 2 + 1 1 2
inequalities and to check their validness for each (0, 1).
For example, we will prove that the inequality > 1.
(53)
(, ) (, ) (46)
Now, one can easily prove (by induction, e.g.) that
holds for each positive , if and only if 0.
Since (, ) is monotone increasing in , it is enough to
1 1
prove that := ( + 2) ( + 1) (54)
1 2 + 1 1 2
0 (, )
1. (47) is a negative real number for 2. Therefore 0, and the
(, ) proof of the first part is done. For 0 < < 1 we have
By the above formulae, this is equivalent to the assertion (, )
that the inequality 1
(, )
() 0 (48)
(1 ) ((1 + )+1 + (1 )+1 2) log ((1 + ) / (1 ))
holds for each (0, 1), with = 1
2 (1 + ) (2 (1 + ) (1 ) )
log (1 + ) log (1 ) 1
() := = 2 + (4 ) ( 0) .
2 6
(49) (55)
((1 + ) log (1 + ) + (1 ) log (1 ))
+ log (1 + ) + log (1 ). Therefore, (, ) > (, ) for > 0 and sufficiently
small := ( )/( + ).
We will prove that the power series expansion of () Similarly, we will prove that the inequality
have non-positive coefficients. Thus the relation (48) will be
proved. (, ) (, ) (56)
Since
holds for each , ; 0 < < if and only if 1.
2 As before, it is enough to consider the expression
log (1 + ) log (1 )
= ;
2 0 2 +1 (, )
= () () := () , (57)
1 (, )
2
2
log (1 + ) + log (1 ) = ;
0 +1 (50) with
(1 + ) log (1 + ) + (1 ) log (1 ) (1 + ) log (1 + ) (1 ) log (1 )

() = 1;
2
(58)
2 (1 + ) log (1 + ) + (1 ) log (1 )
= 2 , () = .
0 ( + 1) (2 + 1) 2
It is not difficult to check the identity Therefore, by the transformation given above, we get
() () 5
() = . (59) log
3
Hence by (48), we get () > 0, that is, () is monotone 2 (1 + )6 + (1 )6 2

= log [ ]
increasing for (0, 1). 3 (1 + )5 + (1 )5 2
Therefore
2 15 + 152 + 4
= log [ ]
(, ) 15 2 + 2
lim () = 1. (60)
1 (, ) 0+
1 + 2 + 4 /4 2
log [ ] = log (1 + )
By monotonicity it follows that (, ) (, ) for 1. 1 + 2 /2 2 (68)
For > 1, ( )/( + ) = , we have
2 4 6

1 = +
(, ) (, ) = ( ( 1) 2 + (4 )) (, ) 2 8 24
6 (61)
2 4 6
( 0+ ) . + + +
2 12 30
Hence, (, ) > (, ) for > 1 and sufficiently small. 1
= ((1 + ) log (1 + ) + (1 ) log (1 ))
From the other hand, 2

(, ) ( 1) (2+1 2) = log ,
lim [ 1] = 1 := () .
1 (, ) 2 ( + 1) (2 2)
(62) and the proof is done.
Further, we have to show that (, ) > (, ) for some
Examining the function (), we find out that it has the positive , whenever > 5.
only real zero at 0 1.0376 and is negative for (1, 0 ). Indeed, since
Remark 10. Since () is monotone increasing, we also get

(1 + ) + (1 ) 2 = ( ) 2 + ( ) 4 + (6 ) , (69)
2 4
(, ) 4 log 2
lim () = . (63)
1 (, ) 1 for > 5 and sufficiently small , we get
Hence +1 2 +1 4 6
1 ( 2 ) + ( 4 ) + ( )
=
(, ) 4 log 2 + 1 ( 2 ) 2 + ( 4 ) 4 + (6 )
1 . (64)
1 (, )
1 + ( 1) ( 2) 2 /12 + (4 ) (70)
A calculation gives 4 log 2/ 1.0200. =
1 + ( 2) ( 3) 2 /12 + (4 )
Note also that 1
= 1 + ( ) 2 + (4 ) .
6 3
2 (, ) (, ) . (65)
Similarly,
Therefore, applying the assertion from the part 1, we get
1
= exp ( ((1 + ) log (1 + ) + (1 ) log (1 )))
(, ) (, ), 2; 2
(66) (71)
(, ) (, ), 2. 2 2
= exp ( + (4 )) = 1 + + (4 ) .
2 2
Finally, we give a detailed proof of the part 7.
We have to prove that (, ) (, ) for 5. Since Hence,
(, ) is monotone increasing in , it is sufficient to prove
that the inequality 1 1
( ) = ( 5) 2 + (4 ) , (72)
6
5 (, ) (, ) (67)
and this expression is positive for > 5 and sufficiently
holds for each , R+ . small, that is, sufficiently close to .
As for the part 8, applying the above transformation we

obtain
(, )
(, )
1 (1 + )+1 + (1 )+1 2
=
+ 1 (1 + ) + (1 ) 2
1
exp ( ((1 + ) log (1 + ) + (1 ) log (1 ))) ,
2
(73)
where 0 < < , = ( )/( + ).

Since for > 5,
1 2 1
lim = , (74)
1 + 1 2 2
and the last expression is less than one, it follows that the
inequality (, ) < (, ) cannot hold whenever / is
sufficiently large.
The rest of the proof is straightforward.
Acknowledgment
The author is indebted to the referees for valuable sugges-
tions.
References
[1] K. B. Stolarsky, Generalizations of the logarithmic mean,
Mathematics Magazine, vol. 48, pp. 8792, 1975.
[2] B. C. Carlson, The logarithmic mean, The American Mathe-
matical Monthly, vol. 79, pp. 615618, 1972.
[3] T. P. Lin, The power mean and the logarithmic mean, The
American Mathematical Monthly, vol. 81, pp. 879883, 1974.
[4] P. A. Hasto, Optimal inequalities between Seifferts mean and
power means, Mathematical Inequalities & Applications, vol. 7,
no. 1, pp. 4753, 2004.
[5] Z.-H. Yang, Sharp bounds for the second Seiert mean in terms
of power mean, http://arxiv.org/abs/1206.5494.
[6] G. H. Hardy, J. E. Littlewood, and G. Polya, Inequalities,
Cambridge University Press, Cambridge, UK, 1978.
[7] S. Simic, On logarithmic convexity for differences of power
means, Journal of Inequalities and Applications, vol. 2007,
Article ID 37359, 8 pages, 2007.
doi:10.1155/2012/540710
Research Article
The Monotonicity Results for the Ratio of
Certain Mixed Means and Their Applications
Zhen-Hang Yang
Power Supply Service Center, Zhejiang Electric Power Company, Electric Power Research Institute,
Hangzhou 310014, China
Correspondence should be addressed to Zhen-Hang Yang, yzhkm@163.com
Received 22 July 2012; Accepted 26 September 2012
Academic Editor: Edward Neuman
Copyright q 2012 Zhen-Hang Yang. This is an open access article distributed under the Creative
Commons Attribution License, which permits unrestricted use, distribution, and reproduction in
any medium, provided the original work is properly cited.
We continue to adopt notations and methods used in the papers illustrated by Yang 2009, 2010
to investigate the monotonicity properties of the ratio of mixed two-parameter homogeneous
means. As consequences of our results, the monotonicity properties of four ratios of mixed
Stolarsky means are presented, which generalize certain known results, and some known and new
inequalities of ratios of means are established.
1. Introduction
Since the Ky Fan 1 inequality was presented, inequalities of ratio of means have attracted
attentions of many scholars. Some known results can be found in 214. Research for the
properties of ratio of bivariate means was also a hotspot at one time.
In this paper, we continue to adopt notations and methods used in the paper 13, 14
to investigate the monotonicity properties of the functions Qif i 1, 2, 3, 4 defined by

g1f p; a, b
Q1f p : ,
g1f p; c, d

g2f p; a, b
Q2f p : ,
g2f p; c, d
1.1
g3f p; a, b
Q3f p : ,
g3f p; c, d

g4f p; a, b
Q4f p : ,
g4f p; c, d
where

g1f p g1f p; a, b : Hf p, q Hf 2k p, q , 1.2

g2f p g2f p; a, b : Hf p, p m Hf 2k p, 2k p m , 1.3

g3f p g3f p; a, b : Hf p, 2m p Hf 2k p, 2m 2k p , 1.4

g4f p g4f p; a, b : Hf pr, ps Hf 2k p r, 2k p s , 1.5
the q, r, s, k, m R, a, b, c, d R with b/a > d/c 1, Hf p, q is the so-called two-parameter

homogeneous functions defined by 15, 16. For conveniences, we record it as follows.
Definition 1.1. Let f: R2 \ {x, x, x R } R be a first-order homogeneous continuous

function which has first partial derivatives. Then, Hf : R2 R2 R is called a homogeneous
function generated by f with parameters p and q if Hf is defined by for a / b
1pq
fap , bp
Hf p, q; a, b , if pq p q / 0,
faq , bq
1.6
ap fx ap , bp ln a bp fy ap , bp ln b
Hf p, p; a, b exp , if p q /
0,
fap , bp
where fx x, y and fy x, y denote first-order partial derivatives with respect to first and
second component of fx, y, respectively.
If limy x fx, y exits and is positive for all x R , then further define
1/p
fap , bp
Hf p, 0; a, b , 0, q 0,
if p /
f1, 1
1/q
faq , bq 1.7
Hf 0, q; a, b , if p 0, q /
0,
f1, 1
Hf 0, 0; a, b afx 1,1/f1,1 bfy 1,1/f1,1 , if p q 0,
and Hf p, q; a, a a.
Remark 1.2. Witkowski 17 proved that if the function x, y fx, y is a symmetric and
first-order homogeneous function, then for all p, q Hf p, q; a, b is a mean of positive numbers
a and b if and only if f is increasing in both variables on R . In fact, it is easy to see that the
condition fx, y is symmetric can be removed.
If Hf p, q; a, b is a mean of positive numbers a and b, then it is called two-parameter
homogeneous mean generated by f.
For simpleness, Hf p, q; a, b is also denoted by Hf p, q or Hf a, b.

The two-parameter homogeneous function Hf p, q; a, b generated by f is very
important because it can generates many well-known means. For example, substituting
L Lx, y x y/ln x ln y if x, y > 0 with x /

y and Lx, x x for f yields Stolarsky
means HL p, q; a, b Sp,q a, b defined by
p

q a bp 1/pq

, if pq p q / 0,

p aq bq

L1/p ap , bp , if p
/ 0, q 0,
Sp,q a, b 1.8

L1/q aq , bq , 0, p 0,
if q /

I 1/p ap , bp , if p q /
0,

ab, if p q 0,
where Ix, y e1 xx /yy 1/xy if x, y > 0, with x /

y, and Ix, x x is the identric
exponential mean see 18. Substituting A Ax, y x y/2 for f yields Gini means
HA p, q; a, b Gp,q a, b defined by
p p 1/pq
a b

q,
, if p /
Gp,q a, b aq bq 1.9

1/p p p
Z a , b , if p q,
where Za, b aa/ab bb/ab see 19.

As consequences of our results, the monotonicity properties of four ratios of mixed
Stolarsky means are presented, which generalize certain known results, and some known and
new inequalities of ratios of means are established.
2. Main Results and Proofs

In 15, 16, 20, two decision functions play an important role, that are,

2 ln f x, y
I I x, y ln f x, y xy ln f xy ,
xy
2.1
xI
J J x, y x y x y xIx .
x
In 14, it is important to another key decision function defined by

x 3
T3 x, y : xyxIx ln , where I ln f xy , x at , y bt . 2.2
y
Note that the function T defined by

T t : ln f at , bt , 0
t/ 2.3
has well properties see 15, 16. And it has shown in 14, 3.4, 16, Lemma 4 the relation
among T t, Jx, y and T3 x, y:

T t t3 T3 x, y , where x at , y bt , 2.4
1 3
T t Ct3 J x, y , where C xy x y ln x ln y > 0. 2.5
Moreover, it has revealed in 14, 3.5 that
y
x
T3 x, y T3 , 1 T3 1, . 2.6
y x
Now, we observe the monotonicities of ratio of certain mixed means defined by 1.1.
Theorem 2.1. Suppose that f: R R R is a symmetric, first-order homogenous, and three-

time dierentiable function, and T3 1, u strictly increase (decrease) with u > 1 and decrease (increase)
with 0 0 with b/a > d/c 1 and fixed q 0, k 0, but q, k are
not equal to zero at the same time, Q1f is strictly increasing (decreasing) in p on k, and decreasing
(increasing) on , k.
The monotonicity of Q1f is converse if q 0, k 0, but q, k are not equal to zero at the same
time.
Proof. Since fx, y > 0 for x, y R R , so T t is continuous on p, q or q, p for
p, q R, then 2.13 in 13 holds. Thus we have
1 1
1 1 1 1
ln g1f p ln Hf p, q ln Hf 2k p, q T t11 dt T t12 dt, 2.7
2 2 2 0 2 0
where

t12 tp 1 tq, t11 t 2k p 1 tq. 2.8
Partial derivative leads to
1 1
1 1
ln g1f p tT t12 dt tT t11 dt
2 0 2 0
1
1 1 1
tT |t12 |dt tT |t11 |dt by13, 2.7 2.9
2 0 2 0
1 |t12 |
1
t T vdv dt,
2 0 |t11 |
and then

ln Q1f p ln g1f p; a, b ln g1f p; c, d
1 |t12 |
1 1 1 |t12 |
t T vdv dt t T v; c, ddv dt
2 0 |t11 | 2 0 |t11 |
1 |t12 | 2.10
|t11 |
T v; a, b T v; c, ddv
t|t12 | |t11 | dt
0 |t12 | |t11 |
1
: t|t12 | |t11 |h|t11 |, |t12 |dt,
0
where
y
T v; a, b T v; c, ddv
x , if x
/ y,
h x, y : yx 2.11

T x; a, b T x; c, d, if x y.
Since T3 1, u strictly increase decrease with u > 1 and decrease increase with 0 d/c 1 yield
T v; a, b T v; c, d v3 T3 av , bv T3 cv , dv

v v
3 b d
v T3 1, T3 1, > <0, for v > 0,
a c
2.12
and therefore hx, y > <0 for x, y > 0. Thus, in order to prove desired result, it suces to
determine the sign of |t12 | |t11 |. In fact, if q 0, k 0, then for t 0, 1

t2 t2 q1 t kt > 0, if p > k,
|t12 | |t11 | 12 11 4t pk 2.13
|t12 | |t11 | t12 t11 < 0, if p < k.
It follows that

> <0, if p > k,
ln Q1f p 2.14
< >0, if p < k.
Clearly, the monotonicity of Q1f is converse if q 0, k 0.

This completes the proof.
Theorem 2.2. The conditions are the same as those of Theorem 2.1. Then, for any a, b, c, d > 0 with
b/a > d/c 1 and fixed m, k with k 0, k m 0, but m, k are not equal to zero at the same time,
Q2f is strictly increasing (decreasing) in p on k, and decreasing (increasing) on , k.
The monotonicity of Q2f is converse if k 0 and k m 0, but m, k are not equal to zero at
the same time.
Proof. By 2.13 in 13 we have
1 1
ln g2f p ln Hf p, p m ln Hf 2k p, 2k p m
2 2
1 1 2.15
1 1
T t22 dt T t21 dt,
2 0 2 0
where

t22 tp 1 t p m , t21 t 2k p 1 t 2k p m . 2.16
Direct calculation leads to

1 1 1 |t22 |
1 1 1
ln g2f p T t22 dt T t21 dt T vdv dt, 2.17
2 0 2 0 2 0 |t21 |
and then

1 |t22 | 1 |t22 |
1 1
T v; a, bdv dt T v; c, ddv dt
2 0 |t21 | 2 0 |t21 | 2.18
1
1
|t22 | |t21 |h|t21 |, |t22 |dt,
2 0
where hx, y is defined by 2.11. As shown previously, hx, y > <0 for x, y > 0 if T3 1, u
strictly increase decrease with u > 1 and decrease increase with 0 0, if p > k,
|t22 | |t21 | 22 21 4 pk 2.19
|t22 | |t21 | |t22 | |t21 | < 0, if p < k.
Thus, we have

> <0, if p > k,
ln Q2f p 2.20
< >0, if p < k.
Clearly, the monotonicity of Q2f is converse if k 0 and k m 0.

The proof ends.
b/a > d/c 1 and fixed m > 0, 0 k 2m, Q3f is strictly increasing (decreasing) in p on k,
and decreasing (increasing) on , k.
The monotonicity of Q2f is converse if m < 0, 2m k 0.
Proof. From 2.13 in 13, it is derived that
1 1
ln g3f p ln Hf p, 2m p ln Hf 2k p, 2m 2k p
2 2
1 1 2.21
1 1
T t32 dt T t31 dt,
2 0 2 0
where

t32 tp 1 t 2m p , t31 t 2k p 1 t 2m 2k p . 2.22
Simple calculation yields
1 1 |t32 |
1
1

ln g3f p 2t 1 T t32 T t31 dt 2t 1 T v; a, bdv dt.
2 0 2 0 |t31 |
2.23
Hence,

1 |t32 |
1
2t 1 T v; a, b T v; c, d dv dt
2 0 |t31 | 2.24
1
1
2t 1|t32 | |t31 |h|t31 |, |t32 |dt,
2 0
where hx, y is defined by 2.11. It has shown that hx, y > <0 for x, y > 0 if T3 1, u
strictly increase decrease with u > 1 and decrease increase with 0 0, 0 k 2m,
then

t232 t231
2t 1|t32 | |t31 | 2t 1
|t32 | |t31 |
tk 1 t2m k
42t 12 pk 2.25
|t32 | |t31 |

> 0, if p > k,

< 0, if p < k,
which yields

> <0, if p > k,
ln Q3f p 2.26
< >0, if p < k.
It is evident that the monotonicity of Q3f is converse if m < 0, 2m k 0.

Thus the proof is complete.
b/a > d/c 1 and fixed k, r, s R with r s / 0, Q4f is strictly increasing (decreasing) in p on
k, and decreasing (increasing) on , k if kr s > 0.
The monotonicity of Q4f is converse if kr s < 0.
Proof. By 2.13 in 13, ln Hf pr, ps can be expressed in integral form

1 r
s,
T pt dt, if r /
ln Hf pr, ps r s s 2.27
T pr , if r s.
The case r s /
0 has no interest since it can come down to the case of m 0 in Theorem 2.2.
Therefore, we may assume that r / s. We have

ln g4f p ln Hf pr, ps Hf 2k p r, 2k p s
r r 2.28
1 1 1 1
T pt dt T 2k p t dt,
2rs s 2rs s
and then
r r
1 1
1 1
ln g4f p tT pt dt tT 2k p t dt
2rs s 2rs s
r 2.29
1 1
t T pt T 2k p t .

2rs s
Note that T t is even see 13, 2.7 and so tT pt T 2k pt is odd, then make use
of Lemma 3.3 in 13, ln g4f p can be expressed as
|r|
1 r s
ln g4f p t T pt T 2k p t dt
2 |r| |s| |s|
2.30
1 r s |r| |t42 |
t T vdv dt,
2 |r| |s| |s| |t41 |
where

t42 pt, t41 2k p t. 2.31
Hence,

|r| |t42 |
1 rs
t T v; a, b T v; c, d dv dt
2 |r| |s| |s| |t41 | 2.32
|r|
1 rs
t|t42 | |t41 |h|t41 |, |t42 |dt,
2 |r| |s| |s|
where hx, y is defined by 2.11. We have shown that hx, y > <0 for x, y > 0 if T3 1, u
strictly increase decrease with u > 1 and decrease increase with 0 <0, if kr s > 0, p > k,

< >0, if kr s > 0, p < k, 2.34

< >0, if kr s < 0, p > k,

> <0, if kr s < 0, p < k.
This proof is accomplished.
3. Applications
As shown previously, Sp,q a, b HL p, q; a, b, where L Lx, y is the logarithmic mean.
Also, it has been proven in 14 that T3 1, u < 0 if u > 1 and T3 1, u > 0 if 0 0 with b/a > d/c 1. Then, the following four functions are all
strictly decreasing (increasing) on k, and increasing (decreasing) on , k:
i Q1L is defined by

Sp,q a, bS2kp,q a, b
Q1L p , 3.1
Sp,q c, dS2kp,q c, d
for fixed q 0, k 0, but q, k are not equal to zero at the same time,
ii Q2L is defined by

Sp,pm a, bS2kp,2kpm a, b
Q2L p , 3.2
Sp,pm c, dS2kp,2kpm c, d
for fixed m, k with k 0 and k m 0, but m, k are not equal to zero at the same
time,
iii Q3L is defined by

Sp,2mp a, bS2kp,2m2kp a, b
Q3L p , 3.3
Sp,2mp c, dS2kp,2m2kp c, d
for fixed m > <0, k 0, 2 m 2m, 0.

iv Q4L is defined by

Spr,ps a, bS2kpr,2kps a, b
Q4L p , 3.4
Spr,ps c, dS2kpr,2kps c, d
for fixed k, r, s R with kr s > <0.
Remark 3.2. Letting in the first result of Corollary 3.1, q k yields Theorem 3.4 in 13 since

Sp,k S2kp,k Sp,2kp . Letting q 1, k 0 yields

Ga, b Sp,1 a, bSp,1 a, b La, b
Q1L < < Q1L 0 . 3.5
Gc, d Sp,1 c, dSp,1 c, d Lc, d
Inequalities 3.5 in the case of d c were proved by Alzer in 21. By letting q 1, k 1/2
from Q1L 1/2 > Q1L 1 > Q1L 2, we have

Aa, b Ga, b La, bIa, b Aa, bGa, b
> > . 3.6
Ac, d Gc, d Lc, dIc, d Ac, dGc, d
Inequalities 3.6 in the case of d c are due to Alzer 22.
Remark 3.3. Letting in the second result of Corollary 3.1, m 1, k 0 yields Cheung and Qis
result see 23, Theorem 2. And we have

Ga, b Sp,p1 a, bSp,p1 a, b La, b
Q2L < < Q2L 0 . 3.7
Gc, d Sp,p1 c, dSp,p1 c, d Lc, d
When d c, inequalities 3.7 are changed as Alzers ones given in 24.

Remark 3.4. In the third result of Corollary 3.1, letting k m also leads to Theorem 3.4 in 13.
Put m 1/2, k 1/4. Then from Q3L 1/4 > Q3L 1/2, we obtain a new inequality

He1/2 a, b La, bI1/2 a, b
> . 3.8
He1/2 c, d Lc, dI1/2 c, d
Putting m 1/2, k 1/3 leads to another new inequality

A1/3 a, b S1/6,5/6 a, bI1/2 a, b
> . 3.9
A1/3 c, d S1/6,5/6 c, dI1/2 c, d
Remark 3.5. Letting in the third result of Corollary 3.1, k 1/2 and r, s 1, 0, 1, 1, 2, 1,
and we deduce that all the following three functions

Lp a, bL1p a, b Ip a, bI1p a, b Ap a, bA1p a, b
p , p , p ,
Lp c, dL1p c, d Ip c, dI1p c, d Ap c, dA1p c, d
3.10
are strictly decreasing on 1/2, and increasing on , 1/2, where Lp L1/p ap , bp , Ip
I 1/p ap , bp , and Ap A1/p ap , bp are the p-order logarithmic, identric exponential, and

power mean, respectively, particularly, so are the functions Lp L1p , Ip I1p , Ap A1p .
4. Other Results
Let d c in Theorems 2.12.4. Then, Hf p, q; c, d c and T t; c, c 0. From the their
proofs, it is seen that the condition T3 1, u strictly increases decreases with u > 1 and
decreases increases with 0 <0 for v > 0, which is
equivalent with J x yxIx < >0, where I ln fxy , by 2.4. Thus, we obtain critical
theorems for the monotonicities of gif , i 1 4, defined as 1.21.5.
Theorem 4.1. Suppose that f: R R R is a symmetric, first-order homogenous, and three-time

dierentiable function and J x yxIx < >0, where I ln fxy . Then, for a, b > 0 with
a/ b, the following four functions are strictly increasing (decreasing) in p on k, and decreasing
(increasing) on , k:
i g1f is defined by 1.2, for fixed q, k 0, but q, k are not equal to zero at the same time;
ii g2f is defined by 1.3, for fixed m, k with k 0 and k m 0, but m, k are not equal to
zero at the same time;
iii g3f is defined by 1.4, for fixed m > 0 and 0 k 2m;
iv g4f is defined by 1.5, for fixed k, r, s R with kr s > 0.
If f is defined on R2 \ {x, x, x R }, then T t may be not continuous at t 0, and
2.13 in 13 may not hold for p, q R but must be hold for p, q R . And then, we easily
derive the following from the proofs of Theorems 2.12.4.
Theorem 4.2. Suppose that f: R2 \ {x, x, x R } R is a symmetric, first-order homogenous
and three-time dierentiable function and J x yxIx < >0, where I ln fxy . Then for
b the following four functions are strictly increasing (decreasing) in p on k, 2k and
a, b > 0 with a /
decreasing (increasing) on 0, k:
i g1f is defined by 1.2, for fixed q, k > 0;
ii g2f is defined by 1.3, for fixed m, k with k > 0 and k m > 0;
iii g3f is defined by 1.4, for fixed m > 0 and 0 k 2m;
iv g4f is defined by 1.5, for fixed k, r, s > 0.
If we substitute L, A, and I for f, where L, A, and I denote the logarithmic, arithmetic,

and identric exponential mean, respectively, then from Theorem 4.1, we will deduce some
known and new inequalities for means. Similarly, letting in Theorem 4.2 fx, y Dx, y
|xy|, Kx, y xy| lnx/y|, where x, y > 0 with x / y, we will obtain certain companion
ones of those known and new ones. Here no longer list them.
Disclosure
This paper is in final form and no version of it will be submitted for publication elsewhere.
References
1 E. F. Beckenbach and R. Bellman, Inequalities, Springer, Berlin, Germany, 1961.
2 W. L. Wang, G. X. Li, and J. Chen, Some inequalities of ratio of means, Journal of Chendu University
of Science and Technology, vol. 1988, no. 6, pp. 8388, 1988.
3 J. Chen and Z. Wang, The Heron mean and the power mean inequalities, Hunan Bulletin of Mathe-
matics, vol. 1988, no. 2, pp. 1516, 1988 Chinese.

4 C. E. M. Pearce and J. Pecaric, On the ration of Logarithmic means, Anzeiger der Osterreichischen
Akademie der Wissenschaften. Mathematisch-Naturwissenschaftliche, vol. 131, pp. 3944, 1994.
5 C. P. Chen and F. Qi, Monotonicity properties for generalized logarithmic means, Australian Journal
of Mathematical Analysis and Applications, vol. 1, no. 2, article 2, 2004.
6 F. Qi, S. X. Chen, and C. P. Chen, Monotonicity of ratio between the generalized logarithmic means,
Mathematical Inequalities & Applications, vol. 10, no. 3, pp. 559564, 2007.
7 F. Qi and S. X. Chen, Complete monotonicity of the logarithmic mean, Mathematical Inequalities and
Applications, vol. 10, no. 4, pp. 799804, 2007.
8 E. Neuman and J. Sandor, Inequalities for the ratios of certain bivariate means, Journal of Mathe-
matical Inequalities, vol. 2, no. 3, pp. 383396, 2008.
9 C. P. Chen, The monotonicity of the ratio between generalized logarithmic means, Journal of Mathe-
matical Analysis and Applications, vol. 345, no. 1, pp. 8689, 2008.
10 C. P. Chen, Stolarsky and Gini means, RGMIA Research Report Collection, vol. 11, no. 4, article 11,
2008.
11 C. P. Chen, The monotonicity of the ratio between Stolarsky means, RGMIA Research Report Collec-
tion, vol. 11, no. 4, article 15, 2008.
12 L. Losonczi, Ratio of Stolarsky means: Monotonicity and comparison, Publicationes Mathematicae,
vol. 75, no. 1-2, article 18, pp. 221238, 2009.
13 Z. H. Yang, Some monotonictiy results for the ratio of two-parameter symmetric homogeneous func-
tions, International Journal of Mathematics and Mathematical Sciences, vol. 2009, Article ID 591382, 12
pages, 2009.
14 Z. H. Yang, Log-convexity of ratio of the two-parameter symmetric homogeneous functions and an
application, Journal of Inequalities and Special Functions, no. 11, pp. 1629, 2010.
15 Z. H. Yang, ON the homogeneous functions with two parameters and its monotonicity, Journal of
Inequalities in Pure and Applied Mathematics, vol. 6, no. 4, article 101, 2005.
16 Z. H. Yang, On the log-convexity of two-parameter homogeneous functions, Mathematical Inequa-

lities and Applications, vol. 10, no. 3, pp. 499516, 2007.
17 A. Witkowski, On two- and four-parameter families, RGMIA Research Report Collection, vol. 12, no.
1, article 3, 2009.
18 K. B. Stolarsky, Generalizations of the Logarithmic Mean, Mathematics Magazine, vol. 48, pp. 8792,
1975.
19 C. Gini, Diuna formula comprensiva delle media, Metron, vol. 13, pp. 322, 1938.
20 Z. H. Yang, On the monotonicity and log-convexity of a four-parameter homogeneous mean,
Journal of Inequalities and Applications, vol. 2008, Article ID 149286, 12 pages, 2008.
21 H. Alzer, Uber Mittelwerte, die zwischen dem geometrischen und dem logarithmischen, Mittel

zweier Zahlen liegen, Anzeiger der Osterreichischen Akademie der Wissenschaften. Mathematisch-Nat-
urwissenschaftliche, vol. 1986, pp. 59, 1986 German.
22 H. Alzer, Ungleichungen fur Mittelwerte, Archiv der Mathematik, vol. 47, no. 5, pp. 422426, 1986.
23 W.-S. Cheung and F. Qi, Logarithmic convexity of the one-parameter mean values, Taiwanese Journal
of Mathematics, vol. 11, no. 1, pp. 231237, 2007.
24 H. Alzer, Uer eine einparametrige familie von Mitlewerten, II, Bayerische Akademie der Wissen-
schaften. Mathematisch-Naturwissenschaftliche Klasse. Sitzungsberichte, vol. 1988, pp. 2329, 1989 Ger-
man.
doi:10.1155/2012/315697
Research Article
Refinements of Inequalities among
Difference of Means
Huan-Nan Shi, Da-Mao Li, and Jian Zhang

Department of Electronic Information, Teachers College, Beijing Union University, Beijing 100011, China
Correspondence should be addressed to Huan-Nan Shi, shihuannan@yahoo.com.cn
Received 21 June 2012; Accepted 10 September 2012
Academic Editor: Janusz Matkowski
Copyright q 2012 Huan-Nan Shi et al. This is an open access article distributed under the Creative
In this paper, for the dierence of famous means discussed by Taneja in 2005, we study the Schur-
geometric convexity in 0, 0, of the dierence between them. Moreover some inequalities
related to the dierence of those means are obtained.
1. Introduction
In 2005, Taneja 1 proved the following chain of inequalities for the binary means for a, b
R2 0, 0, :
Ha, b Ga, b N1 a, b N3 a, b N2 a, b Aa, b Sa, b, 1.1
where
ab
Aa, b ,
2

Ga, b ab,
2ab 1.2
Ha, b ,
ab
2
a b Aa, b Ga, b
N1 a, b ,
2 2

aab b 2Aa, b Ga, b
N3 a, b ,
3 3

a b a b 1.3
N2 a, b ,
2 2

a2 b2
Sa, b .
2
The means A, G, H, S, N1 and N3 are called, respectively, the arithmetic mean, the geometric
mean, the harmonic mean, the root-square mean, the square-root mean, and Herons mean.
The N2 one can be found in Taneja 2, 3.
Furthermore Taneja considered the following dierence of means:
MSA a, b Sa, b Aa, b,
MSN2 a, b Sa, b N2 a, b,
MSG a, b Sa, b Ga, b,
MSH a, b Sa, b Ha, b, 1.4
MAN2 a, b Aa, b N2 a, b,
MAG a, b Aa, b Ga, b,
MAH a, b Aa, b Ha, b,
MN2 N1 a, b N2 a, b N1 a, b,
MN2 G a, b N2 a, b Ga, b
and established the following.
Theorem A. The dierence of means given by 1.4 is nonnegative and convex in R2 0,
0, .
Further, using Theorem A, Taneja proved several chains of inequalities; they are
refinements of inequalities in 1.1.
Theorem B. The following inequalities among the mean dierences hold:
1 1 1
MSA a, b MSH a, b MAH a, b MSG a, b MAG a, b, 1.5
3 2 2
1 1 1
MAH a, b MN2 N1 a, b MN2 G a, b MAG a, b MAN2 a, b, 1.6
8 3 4
4
MSA a, b MSN2 a, b 4MAN2 a, b, 1.7
5
3
MSH a, b 2MSN1 a, b MSG a, b, 1.8
2
3 2
MSA a, b MSN3 a, b MSN1 a, b. 1.9
4 3
For the dierence of means given by 1.4, we study the Schur-geometric convexity of
dierence between these dierences in order to further improve the inequalities in 1.1. The
main result of this paper reads as follows.
Theorem I. The following dierences are Schur-geometrically convex in R2 0, 0, :
1
DSHSA a, b MSH a, b MSA a, b,
3
1 1
DAHSH a, b MAH a, b MSH a, b,
2 3
DSGAH a, b MSG a, b MAH a, b,
1
DAGSG a, b MAG a, b MSG a, b,
2
1
DN2 N1 AH a, b MN2 N1 a, b MAH a, b,
8
1
DN2 GN2 N1 a, b MN2 G a, b MN2 N1 a, b,
3
1.10
1 1
DAGN2 G a, b MAG a, b MN2 G a, b,
4 3
1
DAN2 AG a, b MAN2 a, b MAG a, b,
4
4
DSN2 SA a, b MSN2 a, b MSA a, b,
5
4
DAN2 SN2 a, b 4MAN2 a, b MSN2 a, b,
5
DSN1 SH a, b 2MSN1 a, b MSH a, b,
3
DSGSN1 a, b MSG a, b 2MSN1 a, b,
2
3
DSN3 SA a, b MSN3 a, b MSA a, b, 1.11
4
2 3
DSN1 SN3 a, b MSN1 a, b MSN3 a, b.
3 4
The proof of this theorem will be given in Section 3. Applying this result, in Section 4,
we prove some inequalities related to the considered dierences of means. Obtained
inequalities are refinements of inequalities 1.51.9.
2. Definitions and Auxiliary Lemmas

The Schur-convex function was introduced by Schur in 1923, and it has many important
applications in analytic inequalities, linear regression, graphs and matrices, combinatorial
optimization, information-theoretic topics, Gamma functions, stochastic orderings, reliability,
and other related fields cf. 414.
In 2003, Zhang first proposed concepts of Schur-geometrically convex function
which is extension of Schur-convex function and established corresponding decision
theorem 15. Since then, Schur-geometric convexity has evoked the interest of many
researchers and numerous applications and extensions have appeared in the literature cf.
1619.
In order to prove the main result of this paper we need the following definitions and
auxiliary lemmas.
Definition 2.1 see 4, 20. Let x x1 , . . . , xn Rn and y y1 , . . . , yn Rn .

i x is said to be majorized by y in symbols x y if ki1 xi ki1 yi for k

n
n
1, 2, . . . , n 1 and i1 xi i1 yi , where x1 xn and y1 yn are
rearrangements of x and y in a descending order.
ii Rn is called a convex set if x1 y1 , . . . , xn yn for every x and y ,
where and 0, 1 with 1.
iii Let Rn . The function : R is said to be a Schur-convex function on if
x y on implies x y. is said to be a Schur-concave function on if and
only if is Schur-convex.
Definition 2.2 see 15. Let x x1 , . . . , xn Rn and y y1 , . . . , yn Rn .

i Rn is called a geometrically convex set if x1 y1 , . . . , xn yn for all x,y
and , 0, 1 such that 1.
ii Let Rn . The function : R is said to be Schur-geometrically convex
function on if ln x1 , . . . , ln xn ln y1 , . . . , ln yn on implies x y. The
function is said to be a Schur-geometrically concave on if and only if is
Schur-geometrically convex.
Definition 2.3 see 4, 20. i The set Rn is called symmetric set, if x implies P x
for every n n permutation matrix P .
ii The function : R is called symmetric if, for every permutation matrix P ,
P x x for all x .
Lemma 2.4 see 15. Let Rn be a symmetric and geometrically convex set with a nonempty
interior 0 . Let : R be continuous on and dierentiable in 0 . If is symmetric on and

ln x1 ln x2 x1 x2 0 0 2.1
x1 x2
holds for any x x1 , . . . , xn 0 , then is a Schur-geometrically convex Schur-geometrically

concave function.
Lemma 2.5. For a, b R2 0, 0, one has
ab 1 2ab
1 , 2.2
2a2 b2 2 a b2
ab ab 3
, 2.3
2a2 b2 a b2 4

3 ab a b 5 ab

2 2 a b
. 2.4
2 a b 4 a b2
Proof. It is easy to see that the left-hand inequality in 2.2 is equivalent to a b2 0, and
the right-hand inequality in 2.2 is equivalent to

2a2 b2 a b a b2 4ab
, 2.5
2a2 b2 2a b2
that is,
a b2 a b2
. 2.6
2a2 b2 2a2 b2 a b 2a b2
Indeed, from the left-hand inequality in 2.2 we have

2 a2 b2 2a2 b2 a b 2 a2 b2 a b2 2a b2 , 2.7
so the right-hand inequality in 2.2 holds.

The inequality in 2.3 is equivalent to

2a2 b2 a b a b2
. 2.8
2a2 b2 4a b2
Since

2a2 b2 a b 2 a2 b2 a b2

2a2 b2 2a2 b2 2a2 b2 a b
2.9
a b2
,
2a2 b2 a b 2a2 b2
so it is sucient prove that

2 a2 b2 a b 2a2 b2 4a b2 , 2.10
that is,

a b 2a2 b2 2 a2 b2 4ab , 2.11
and, from the left-hand inequalities in 2.2, we have

a b 2a2 b2 2 a2 b2 2 a2 b2 4ab , 2.12
so the inequality in 2.3 holds.

Notice that the functions inthe inequalities 2.4
are homogeneous. So, without loss of

generality, we may assume a b 1, and set t ab. Then 0 < t 1/4 and 2.4 reduces
to

3 1 2t 1 5 t2
. 2.13
2 2 2 1 2t 4 1 2t2
Squaring every side in the above inequalities yields
9 1 2t 1 25 t4 5t2
1 . 2.14
4 2 2 4t 16 1 2t4 21 2t2
Reducing to common denominator and rearranging, the right-hand inequality in 2.14

reduces to

1 2t 16t2 2t 12 1/816t 72 7/8
0, 2.15
162t 14
and the left-hand inequality in 2.14 reduces to

21 2t2 2 51 2t 1 2t

0, 2.16
21 2t 2
so two inequalities in 2.4 hold.
Lemma 2.6 see 16. Let a b, ut ta 1 tb, vt tb 1 ta. If 1/2 t2 t1 1 or
0 t1 t2 1/2, then

ab ab
, ut2 , vt2 ut1 , vt1 a, b. 2.17
2 2
3. Proof of Main Result

Proof of Theorem I. Let a, b R2 .
1 For

1 ab 2ab 2 a2 b2 3.1
DSHSA a, b MSH a, b MSA a, b ,
3 2 3a b 3 2
we have
DSHSA a, b 1 2b2 2 a

2
,
a 2 3a b 3 2a2 b2
3.2
DSHSA a, b 1 2a2 2 b
,
b 2 3a b2 3 2a2 b2
whence

DSHSA a, b DSHSA a, b
: ln a ln b a b
a b
3.3
1 2ab 2 ab
a bln a ln b .
2 3a b2 3 2a2 b2
From 2.3 we have
1 2ab 2 ab
0, 3.4
2 3a b2 3 2a2 b2
which implies 0 and, by Lemma 2.4, it follows that DSHSA is Schur-geometrically convex
in R2 .
2 For

1 1 ab ab 1 a2 b2 3.5
DAHSH a, b MAH a, b MSH a, b .
2 3 4 3a b 3 2
To prove that the function DAHSH is Schur-geometrically convex in R2 it is enough to

notice that DAHSH a, b 1/2DSHSA a, b.
3 For

a2 b2 ab 2ab 3.6
DSGAH a, b MSG a, b MAH a, b ab ,
2 2 ab
we have
DSGAH a, b a b 1 2b2

,
a 2a2 b2 2 ab 2 a b2
3.7
DSGAH a, b b a 1 2a2
,
b 2a2 b2 2 ab 2 a b2
and then

: ln a ln b a b
a b
3.8
ab 1 2ab
a bln a ln b .
2a2 b2 2 a b2
From 2.2 we have 0, so by Lemma 2.4, it follows that DSHSA is Schur-

geometrically convex in R2 .
4 For

1 1 a2 b2
DAGSG a, b MAG a, b MSG a, b a b ab , 3.9
2 2 2
we have

DAGSG a, b 1 b a
1 ,
a 2 2 ab 2a2 b2
3.10
DAGSG a, b 1 a b
1 ,
b 2 2 ab 2a2 b2
and then

: ln a ln b a b
a b
3.11
ab
a bln a ln b 1 .
2a2 b2
By 2.2 we infer that
ab
1 0, 3.12
2a2 b2
so 0. By Lemma 2.4, we get that DAGSG is Schur-geometrically convex in R2 .

5 For
1
DN2 N1 AH a, b MN2 N1 a, b MAH a, b
8

a b a b 1 1 1 ab 2ab
a b ab ,
2 2 4 2 8 2 ab
3.13
we have

DN2 N1 AH a, b 1 ab 1 b a b 1/2
a

a 4 a 2 4 22

1 b 1 1 2b2
,
4 4 ab 8 2 a b2
3.14

DN2 N1 AH a, b 1 ab 1 a b a b 1/2

b 4 b 2 4 2 2

1 a 1 1 2a2
,
4 4 ab 8 2 a b2
and then

DN2 N1 AH a, b DN2 N1 AH a, b
ln a ln b a b
a b

3.15
1 ab a b 5 ab
a bln a ln b .
4 2 a b 2 ab 4 a b2
From 2.4 we have

ab a b 5 ab
0, 3.16
2 a b 2 a b 4 a b2
so 0; it follows that DN2 N1 AH is Schur-geometrically convex in R2 .

6 For
1
DN2 GN2 N1 a, b MN2 G a, b MN2 N1 a, b
3
3.17
ab ab 2 a b a b
,
4 6 3 2 2
we have

DN2 GN2 N1 a, b 1 b 1 ab 1 a b a b 1/2
,
a 4 12 ab 6 a 2 6 2 2
3.18

DN2 GN2 N1 a, b 1 a 1 ab 1 a b a b 1/2
,
b 4 12 ab 6 b 2 6 2 2
and then

DN2 GN2 N1 a, b DN2 GN2 N1 a, b
ln a ln b a b
a b

a b a b 1/2
1 a b ab ab
ln a ln b a b
4 6 2 12 2 3.19

1 3 ab a b
a bln a ln b .
6 2 2 a b 2 ab
By 2.4 we infer that 0, which proves that DN2 GN2 N1 is Schur-geometrically convex in
R2 .
7 For
1 1
DAGN2 G a, b MAG a, b MN2 G a, b
4 3
3.20
ab 1 1 a b a b
ab ,
8 12 3 2 2
we have

DAGN2 G a, b 1 b ab a b
,
a 8 24 ab 12 2a 12 2a b
3.21
DAGN2 G a, b 1 a ab a b
,
b 8 24 ab 12 2b 12 2a b
and then

DAGN2 G a, b DAGN2 G a, b
ln a ln b a b
a b

a b a b a b a b
a b
ln a ln b
8 12 2 12 2a b 3.22

a bln a ln b 2 ab a b
1 .
8 3 2 a b 2 ab
From 2.4 we have 0, and, consequently, by Lemma 2.4, we obtain that DAGN2 G is
Schur-geometrically convex in R2 .
8 In order to prove that the function DAN2 AG a, b is Schur-geometrically convex in
2
R it is enough to notice that
1
DAN2 AG a, b MAN2 a, b MAG a, b 3DAGN2 G a, b. 3.23
4
9 For
4
DSN2 SA a, b MSN2 a, b MSA a, b
5
3.24
a b 1 a2 b2 1
a b 2a b,
2 5 2 5
we have

DSN2 SA a, b 1 a 1 ab a b
,
a 2 5 2a2 b2 5 2a 5 2a b
3.25

DSN2 SA a, b 1 b 1 ab a b
,
b 2 5 2a2 b2 5 2b 5 2a b
and then

DSN2 SA a, b DSN2 SA a, b
ln a ln b
a b

a b a b
a b a b aa b ba b
2 2
1
ln a ln b
2 5 2a b 5
2 2 2 2 5 2a b

a bln a ln b 5 ab ab a b
.
5 2 2 a2 b2 a b ab
3.26
From 2.2 and 2.4 we obtain that

5 ab ab a b 5 3
2 0, 3.27
2 a b
2 2 a b ab 2 2
so 0, which proves that the function DSN2 SA a, b is Schur-geometrically convex in R2 .
10 One can easily check that
DANA N2 SN2 a, b 4DSN2 SA a, b, 3.28
and, consequently, the function DAN2 SN2 is Schur-geometrically convex in R2 .

11 To prove that the function

a2 b2 a b 2ab 3.29
DSN1 SH a, b 2MSN1 a, b MSH a, b ab
2 2 ab
is Schur-geometrically convex in R2 it is enough to notice that
DSN1 SH a, b DSGAH a, b. 3.30
12 For
3
DSGSN1 a, b MSG a, b 2MSN1 a, b
2
3.31
1 a b
2 2
a b ab ,
2 2
we have

DSGSN1 a, b 1 b a
1 ,
a 2 2 ab 2a2 b2
3.32
DSGSN1 a, b 1 a b
1 ,
b 2 2 ab 2a2 b2
and then

DSGSN1 a, b DSGSN1 a, b
ln a ln b a b
a b
3.33
a bln a ln b ab
1 .
2 2a2 b2
By the inequality 2.2 we get that 0, which proves that DSGSN1 is Schur-
geometrically convex in R2 .
13 It is easy to check that
1
DSN3 SA a, b DAGSG a, b, 3.34
2
which means that the function DSN3 SA is Schur-geometrically convex in R2 .

14 To prove that the function DSN1 SN3 is Schur-geometrically convex in R2 it is
enough to notice that
1
DSN1 SN3 a, b DAGSG a, b. 3.35
6
The proof of Theorem I is complete.
4. Applications
Applying Theorem I, Lemma 2.6, and Definition 2.2 one can easily prove the following.
Theorem II. Let 0 < a b. 1/2 t 1 or 0 t 1/2, u at b1t and v bt a1t . Then

1 1 1
MSA a, b MSH a, b MSH u, v MSA u, v MSH a, b
3 3 3

1 1 1 1
MAH a, b MAH u, v MSH u, v MAH a, b
2 2 3 2
4.1
1 1 1 1
MSG a, b MSG u, v MAH u, v MSG a, b
2 2 2 2

1
MAG a, b MAG u, v MSG u, v MAG a, b,
2

1 1
MAH a, b MN2 N1 a, b MN2 N1 u, v MAH u, v MN2 N1 a, b
8 8

1 1 1
MN2 G a, b MN2 G u, v MN2 N1 u, v MN2 G a, b
3 3 3
4.2
1 1 1 1
MAG a, b MAG u, v MN2 G u, v MAG a, b
4 4 3 4

1
MAN2 a, b MAN2 u, v MAG u, v MAN2 a, b,
4

4 4 4 4
MSA a, b MSN2 a, b MSN2 u, v MSN2 u, v MSN2 a, b
5 5 5 5
4.3
4
4MAN2 a, b 4MAN2 u, v MSN2 u, v 4MAN2 a, b,
5
MSH a, b 2MSN1 a, b 2MSN1 u, v MSH u, v 2MSN1 a, b
4.4
3 3 3 3
MSG a, b MSG u, v MSG u, v MSG a, b,
2 2 2 2

3 3 3
MSA a, b MSN3 a, b MSN3 u, v MSA u, v MSN3 a, b
4 4 4
4.5
2 2 3 2
MSN1 a, b MSN1 u, v MSN3 u, v MSN1 a, b.
3 3 4 3
Remark 4.1. Equation 4.1, 4.2, 4.3, 4.4, and 4.5 are a refinement of 1.5, 1.6, 1.7,
1.8, and 1.9, respectively.
Acknowledgments
The authors are grateful to the referees for their helpful comments and suggestions. The
first author was supported in part by the Scientific Research Common Program of Beijing
Municipal Commission of Education KM201011417013.
References
1 I. J. Taneja, Refinement of inequalities among means, Journal of Combinatorics, Information & System
Sciences, vol. 31, no. 14, pp. 343364, 2006.
2 I. J. Taneja, On a Dierence of Jensen Inequality and its Applications to Mean Divergence Measures,
RGMIA Research Report Collection, vol. 7, article 16, no. 4, 2004, http://rgmia.vu.edu.au/.
3 I. J. Taneja, On symmetric and nonsymmetric divergence measures and their generalizations,
Advances in Imaging and Electron Physics, vol. 138, pp. 177250, 2005.
4 A. W. Marshall and I. Olkin, Inequalities: Theory of Majorization and Its Applications, vol. 143 of
Mathematics in Science and Engineering, Academic Press, New York, NY, USA, 1979.
5 X. Zhang and Y. Chu, The Schur geometrical convexity of integral arithmetic mean, International
Journal of Pure and Applied Mathematics, vol. 41, no. 7, pp. 919925, 2007.
6 K. Guan, Schur-convexity of the complete symmetric function, Mathematical Inequalities &
7 K. Guan, Some properties of a class of symmetric functions, Journal of Mathematical Analysis and
8 C. Stepniak, An eective characterization of Schur-convex functions with applications, Journal of
Convex Analysis, vol. 14, no. 1, pp. 103108, 2007.
9 H.-N. Shi, Schur-convex functions related to Hadamard-type inequalities, Journal of Mathematical
Inequalities, vol. 1, no. 1, pp. 127136, 2007.
10 H.-N. Shi, D.-M. Li, and C. Gu, The Schur-convexity of the mean of a convex function, Applied
Mathematics Letters, vol. 22, no. 6, pp. 932937, 2009.
11 Y. Chu and X. Zhang, Necessary and sucient conditions such that extended mean values are Schur-
convex or Schur-concave, Journal of Mathematics of Kyoto University, vol. 48, no. 1, pp. 229238, 2008.
12 N. Elezovic and J. Pecaric, A note on Schur-convex functions, The Rocky Mountain Journal of
Mathematics, vol. 30, no. 3, pp. 853856, 2000.
13 J. Sandor, The Schur-convexity of Stolarsky and Gini means, Banach Journal of Mathematical Analysis,
vol. 1, no. 2, pp. 212215, 2007.
14 H.-N. Shi, S.-H. Wu, and F. Qi, An alternative note on the Schur-convexity of the extended mean
values, Mathematical Inequalities & Applications, vol. 9, no. 2, pp. 219224, 2006.
15 X. M. Zhang, Geometrically Convex Functions, Anhui University Press, Hefei, China, 2004.
16 H.-N. Shi, Y.-M. Jiang, and W.-D. Jiang, Schur-convexity and Schur-geometrically concavity of Gini
means, Computers & Mathematics with Applications, vol. 57, no. 2, pp. 266274, 2009.
17 Y. Chu, X. Zhang, and G. Wang, The Schur geometrical convexity of the extended mean values,
Journal of Convex Analysis, vol. 15, no. 4, pp. 707718, 2008.
18 K. Guan, A class of symmetric functions for multiplicatively convex function, Mathematical
Inequalities & Applications, vol. 10, no. 4, pp. 745753, 2007.
19 H.-N. Shi, M. Bencze, S.-H. Wu, and D.-M. Li, Schur convexity of generalized Heronian means
involving two parameters, Journal of Inequalities and Applications, vol. 2008, Article ID 879273, 9 pages,
2008.
20 B. Y. Wang, Foundations of Majorization Inequalities, Beijing Normal University Press, Beijing, China,
1990.
doi:10.1155/2012/730962
Research Article
Complete Moment Convergence of Weighted Sums
for Arrays of Rowwise -Mixing Random Variables
Ming Le Guo
School of Mathematics and Computer Science, Anhui Normal University, Wuhu 241003, China
Correspondence should be addressed to Ming Le Guo, mleguo@163.com
Received 5 June 2012; Accepted 20 August 2012
Academic Editor: Mowaaq Hajja
Copyright q 2012 Ming Le Guo. This is an open access article distributed under the Creative
The complete moment convergence of weighted sums for arrays of rowwise -mixing random
variables is investigated. By using moment inequality and truncation method, the sucient
conditions for complete moment convergence of weighted sums for arrays of rowwise -mixing
random variables are obtained. The results of Ahmed et al. 2002 are complemented. As an
application, the complete moment convergence of moving average processes based on a -mixing
random sequence is obtained, which improves the result of Kim et al. 2008.
1. Introduction
Hsu and Robbins 1 introduced the concept of complete convergence of {Xn }. A sequence
{Xn , n 1, 2, . . .} is said to converge completely to a constant C if

P |Xn C| > < , > 0. 1.1
n1
Moreover, they proved that the sequence of arithmetic means of independent identically
distributed i.i.d. random variables converge completely to the expected value if the variance
of the summands is finite. The converse theorem was proved by Erdos 2. This result has
been generalized and extended in several directions, see Baum and Katz 3, Chow 4, Gut
5, Taylor et al. 6, and Cai and Xu 7. In particular, Ahmed et al. 8 obtained the following
result in Banach space.
Theorem A. Let {Xni ; i 1, n 1} be an array of rowwise independent random elements in a

separable real Banach space B, . Let P Xni > x CP |X| > x for some random variable X,
constant C and all n, i and x > 0. Suppose that {ani , i 1, n 1} is an array of constants such that

sup|ani | O nr , for some r > 0,
i1

1.2
|ani | On , for some 0, r.
i1
Let be such that /

1 and fix > 0 such that 1 /r < 2. Denote s max1

1/r, . If E|X|s < and Sn i1 ani Xni 0 in probability, then n1 n P Sn > <
for all > 0.
Chow 4 established the following refinement which is a complete moment

convergence result for sums of i.i.d. random variables.
Theorem B. Let EX1 0, 1 p < 2 and r p. Suppose that E|X1 |r |X1 | log1 |X1 | < .
Then

n
r/p21/p 1.3
n E Xi n1/p
< , > 0.
n1
i1
The main purpose of this paper is to discuss again the above results for arrays of
rowwise -mixing random variables. The author takes the inspiration in 8 and discusses
the complete moment convergence of weighted sums for arrays of rowwise -mixing random
variables by applying truncation methods. The results of Ahmed et al. 8 are extended to -
mixing case. As an application, the corresponding results of moving average processes based
on a -mixing random sequence are obtained, which extend and improve the result of Kim
and Ko 9.
For the proof of the main results, we need to restate a few definitions and lemmas for
easy reference. Throughout this paper, C will represent positive constants, the value of which
may change from one place to another. The symbol IA denotes the indicator function of A;
x indicates the maximum integer not larger than x. For a finite set B, the symbol B denotes
the number of elements in the set B.
Definition 1.1. A sequence of random variables {Xi , 1 i n} is said to be a sequence of

-mixing random variables, if
m sup |P B | A P B| ; A k1 , B

k m , P A > 0 0, as m , 1.4
k1
where kj {Xi ; j i k}, 1 j k .
Definition 1.2. A sequence {Xn , n 1} of random variables is said to be stochastically

dominated by a random variable X write {Xi } X if there exists a constant C, such that
P {|Xn | > x} CP {|X| > x} for all x 0 and n 1.
The following lemma is a well-known result.
Lemma 1.3. Let the sequence {Xn , n 1} of random variables be stochastically dominated by a
random variable X. Then for any p > 0, x > 0

E|Xn |p I|Xn | x C E|X|p I|X| x xp P {|X| > x} , 1.5
E|Xn |p I|Xn | > x CE|X|p I|X| > x. 1.6
Definition 1.4. A real-valued function lx, positive and measurable on A, for some A > 0,
is said to be slowly varying if limx lx/lx 1 for each > 0.
By the properties of slowly varying function, we can easily prove the following lemma.
Here we omit the details of the proof.
Lemma 1.5. Let lx > 0 be a slowly varying function as x , then there exists C (depends only
on r) such that
k
i Ckr 1 lk n1 nr ln Ckr 1 lk for any r > 1 and positive integer k,

ii Ckr 1 lk nk nr ln Ckr 1 lk for any r < 1 and positive integer k.
The following lemma will play an important role in the proof of our main results. The
proof is due to Shao 10.
Lemma 1.6. Let {Xi , 1 i n} be a sequence of -mixing random variables with mean zero. Suppose
2
that there exists a sequence {Cn } of positive numbers such that E k m
ik 1 Xi Cn for any k 0, n
1, m n. Then for any q 2, there exists C Cq, such that
q

k j
E max q/2 q
Xi C Cn E max |Xi | . 1.7
1jn k 1ik n
ik 1

Lemma 1.7. Let {Xi , 1 i n} be a sequence of -mixing random variables with i1 1/2 i < ,
then there exists C such that for any k 0 and n 1
2

k n
k n
E Xi C EXi2 . 1.8
ik 1 ik 1
Proof. By Lemma 5.4.4 in 11 and Holders

inequality, we have
2

k n
k n
E Xi EXi2 2 EXi Xj
ik 1 ik 1 k 1i<jk n

k n 1/2 1/2
EXi2 4 1/2 j i EXi2 EXj2
ik 1 k 1i<jk n
1.9

k n
k n1 k n

EXi2 2 1/2 j i EXi2 EXj2
ik 1 ik 1 ji 1

n
k n
1 4 1/2
i EXi2 .
i1 ik 1
Therefore, 1.8 holds.
2. Main Results
Now we state our main results. The proofs will be given in Section 3.
Theorem 2.1. Let {Xni , i 1, n 1} be an array of rowwise -mixing random variables with EXni

0, {Xni } X and m1
1/2
m < . Let lx > 0 be a slowing varying function, and {ani , i 1, n
1} be an array of constants such that

sup|ani | O nr , for some r > 0,
i1

2.1
|ani | On , for some 0, r.
i1
a If 1 > 0 and there exists some > 0 such that /r 1 < 2, and s
max1 1/r, , then E|X|s l|X|1/r < implies

k
2.2
n lnE sup ani Xni < ,

> 0.
n1

k1 i1

b If 1, > 0, then E|X|1 a/r 1 l|X|1/r < implies

k
1 2.3
n lnE sup ani Xni < , > 0.
n1

k1 i1

Remark 2.2. If 1 < 0, then E|X| < implies that 2.2 holds. In fact,

k

n lnE sup ani Xni

n ln |ani |E|Xni | n ln
n1

k1 i1
n1 i1 n1
2.4

C
n lnE|X| n ln < .

n1 n1
Remark 2.3. Note that

k k

> n lnE sup ani Xni

n ln P sup ani Xni > x dx
n1

k1 i1
n1 0
k1 i1

2.5

k

n lnP sup ani Xni > x dx.
0 n1 k1
i1

Therefore, from 2.5, we obtain that the complete moment convergence implies the complete
convergence, that is, under the conditions of Theorem 2.1, result 2.2 implies

k

n lnP sup ani Xni > < ,

2.6
n1

k1 i1

and 2.3 implies

k
1
n lnP sup ani Xni > < . 2.7
n1

k1 i1

Corollary 2.4. Under the conditions of Theorem 2.1,
1 if 1 > 0 and there exists some > 0 such that /r 1 < 2, and s
max1 1/r, , then E|X|s l|X|1/r < implies

2.8
n lnE ani Xni < ,

> 0,
n1
i1
2 if 1, > 0, then E|X|1 /r 1 l|X|1/r < implies

1 2.9
n lnE ani Xni < , > 0.
n1
i1
Corollary 2.5. Let {Xni , i 1, n 1} be an array of rowwise -mixing random variables with

EXni 0,{Xni } X and m1
1/2
m < . Suppose that lx > 0 is a slowly varying function.
1 Let p > 1 and 1 t < 2. If E|X|pt l|X|t < , then

k
2.10
np21/t
lnE max Xni n1/t
< , > 0.
1kn
n1 i1
2 Let 1 < t < 2. If E|X|t 1 l|X|t < , then

k
11/t 2.11
n lnE max Xni n1/t
< , > 0.
1kn
n1 i1

Corollary 2.6. Suppose that Xn i ai n Yi , n 1, where {ai , < i < } is a sequence of

real numbers with |ai | < , and {Yi , < i < } is a sequence of -mixing random variables

with EYi 0, {Yi } Y and m1
1/2
m < . Let lx be a slowly varying function.
1 Let 1 t < 2, r 1 t/2. If E|Y |r l|Y |t < , then

n
r/t21/t 2.12
n lnE Xi n1/t
< , > 0.
n1
i1
2 Let 1 < t < 2. If E|Y |t 1 l|Y |t < , then

n
11/t 2.13
n lnE Xi n1/t
< , > 0.
n1
i1
Remark 2.7. Corollary 2.6 obtains the result about the complete moment convergence of
moving average processes based on a -mixing random sequence with dierent distributions.
We extend the results of Chen et al. 12 from the complete convergence to the complete
moment convergence. The result of Kim and Ko 9 is a special case of Corollary 2.6 1.
Moreover, our result covers the case of r t, which was not considered by Kim and Ko.
3. Proofs of the Main Results

Proof of Theorem 2.1. Without loss of generality, we can assume

sup|ani | nr , |ani | n . 3.1
i1 i1

Let Snk x ki1 ani Xni I|ani Xni | nr x for any k 1, n 1, and x 0. First note that
E|X|s l|X|1/r < implies E|X|t < for any 0 < t < s. Therefore, for x > nr ,

k

1 r 1 r r
x n supE|Snk x| x n supE ani Xni I |ani Xni | > n x EXni 0
k1 k1
i1

E|ani Xni |I |ani Xni | > nr x E|ani X|I |ani X| > nr x
i1 i1 3.2

|ani |E|X|I|X| > x n E|X|I|X| > x
i1
x/r E|X|I|X| > x E|X|1 /r I|X| > nr 0 as n .
Hence, for n large enough we have supk1 E|Snk x| < /2nr x. Then

k

n lnE sup ani Xni

n1

k1 i1

k

n ln
P sup ani Xni x dx
n1
k1 i1

k
r 3.3
n r
ln P sup ani Xni n x dx
n1 nr
k1 i1

r
C n r
ln P sup|ani Xni | > n x dx
n1 nr i

r
C r
n ln P sup|Snk x ESnk x| n x dx : I1 I2 .
n1 nr k1 2
Noting that > 1, by Lemma 1.5, Markov inequality, 1.6, and 3.1, we have

I1 C nr ln P |ani Xni | > nr x dx
n1 nr i1

C nr ln nr x1 E|ani Xni |I |ani Xni | > nr x dx
n1 nr i1

C n ln x1 E|X|I|X| > xdx
n1 nr

kr 1

C n
ln x1 E|X|I|X| > xdx
n1 kn kr

k
C n ln k1 E|X|I|X| > kr C k1 E|X|I|X| > kr n ln
n1 kn k1 n1

C k lkE|X|I|X| > kr CE|X|1 1 /r l |X|1/r < .
k1
3.4

Now we estimate I2 , noting that m1 1/2 m < , by Lemma 1.7, we have
2

m

m

r r
sup E ani Xni I |ani Xni | n x E ani Xni I |ani Xni | n x
1m< i1 i1
3.5

r
C Ea2ni Xni
2
I |ani Xni | n x .
i1
By Lemma 1.6, Markov inequality, Cr inequality, and 1.5, for any q 2, we have

r
P sup|Snk x ESnk x| n x Cnrq xq Esup|Snk x ESnk x|q
k1 2 k1
q/2

Cnrq xq Ea2ni Xni
2
I |ani Xni | nr x E|ani Xni |q I |ani Xni | nr x
i1 i1
q/2
rq q

r

3.6
Cn x Ea2ni X 2 I |ani X| n x Cnrq xq E|ani X|q I |ani X| nr x
i1 i1
q/2

r
C P |ani X| > n x C P |ani X| > nr x
i1 i1
: J1 J2 J3 J4 .
So,

I2 nr ln J1 J2 J3 J4 dx. 3.7
n1 nr

From 3.4, we have n1 n
r
ln nr J4 dx < .
For J1 , we consider the following two cases.
If s > 2, then EX 2 < . Taking q 2 such that q r/2 < 1, we have

r
n ln J1 dx
n1 nr
q/2

q 3.8
C n r rq
ln x a2ni dx
n1 nr i1

C nr rq lnnqr/2 nrq 1 C n qr/2 ln < .
n1 n1
If s 2, we choose s such that 1 /r < s < s. Taking q 2 such that qr/21 /r
s < 1, we have

r
n ln J1 dx
n1 nr
q/2

s 1 2s s
q r
C n r rq
ln x |ani ||ani | E|ani X| |X| I |ani X| n x dx
n1 nr i1
3.9

q/22s
C nr rq lnnq/2 nqr/2s 1 x q
nr x dx
n1 nr

C n qr/21 /rs ln < .
n1

So, n1nr ln nr J1 dx < .
Now, we estimate J2 . Set Inj {i 1 | nj 1r < |ani | njr }, j 1, 2, . . .. Then
j1 Inj N, where N is the set of positive integers. Note also that for all k 1, n 1,

n |ani | |ani |
i1 j1 iInj
3.10

r

rq
Inj n j 1 nr Inj j 1 k 1rqr .
j1 jk
Hence, we have

Inj j rq Cn r krrq . 3.11
jk
Note that

r
n ln J2 dx
n1 nr

C n r rq
ln xq E|ani X|q I |ani X| nr x dx
n1 nr j1 iInj
k 1r

rq

r
C n r rq
ln Inj nj xq E|X|q I |X| x j 1 dx
n1 j1 kn kr

rq

r
C nr rq ln Inj nj krq 11 E|X|q I |X| k 1r j 1
n1 j1 kn
3.12

k 1j 11

C nr ln krq 11 Inj j rq E|X|q I ir < |X| i 1r
n1 kn j1 i0

2k 11

C nr ln krq 11 Inj j rq E|X|q I ir < |X| i 1r
n1 kn j1 i0

k 1j 1

rq

C nr
ln k rq 11
Inj j E|X|q I ir < |X| i 1r
n1 kn j1 i2k 1
: J2 J2 .
Taking q 2 large enough such that rq r < 1, for J2 , by Lemma 1.6 and 3.11, we
get

2k 11

J2 C nr ln krq 11 n r E|X|q I ir < |X| i 1r
n1 kn i0

2k 11

k
C krq 11 E|X|q I ir < |X| i 1r n ln
k1 i0 n1

2k 11

C k rq r lk E|X|q I ir < |X| i 1r
k1 i0

C C E|X|q I ir < |X| i 1r k rq r lk
i3 ki/2

C C i rq r 1 liE|X|q I ir < |X| i 1r C CE|X|1 1/r l |X|1/r < .
i3
3.13
For J2 , we obtain

j 1k 1

J2 C nr ln krq 11 Inj j rq E|X|q I ir < |X| i 1r
n1 kn j1 i2k 1

C nr ln krq 11 E|X|q I ir < |X| i 1r Inj j rq
n1 kn i2k 1 jik 11 1

C nr ln krq 11 nr ir1q kr1q E|X|q I ir < |X| i 1r
n1 kn i2k 1
3.14

r

k
C k1 ir1q E|X|q I ir < |X| i 1 n ln
k1 i2k 1 n1

C k lk ir1q E|X|q I ir < |X| i 1r
k1 i2k 1

C i 1 rrq E|X|q I ir < |X| i 1r CE|X|1 1/r l |X|1/r < .
i4

So n1 n
r
ln nr J2 dx < . Finally, we prove n1 n
r
ln nr J3 dx < . In fact, noting
1 a/r < s < s and qr/21 /r s < 1, using Markov inequality and 3.1, we
get

q/2
rs s s
r
n ln J3 dx C nr
ln n x E|ani X| dx
n1 nr n1 nr i1

qrs /2 rs 1q/2 q/2
C nr
lnn n n xs q/2 dx
n1 nr

C nr rq/2 q/2 lnnrs q/2 1 C n qr/21 /2s ln < .
n1 n1
3.15
Thus, we complete the proof in a. Next, we prove b. Note that E|X|1 /r < implies that
3.2 holds. Therefore, from the proof in a, to complete the proof of b, we only need to
prove

1r r
I2 C n ln P sup|Snk x ESnk x| n x dx < . 3.16
n1 nr k1 2
In fact, noting 1, 1 > 0, r < 1 and E|X|1 /r l|X|1/r < . By taking q 2

in the proof of 3.12, 3.13, and 3.14, we get

1 r

C n ln x2 Ea2ni X 2 I |ani X| nr x dx C CE|X|1 /r l |X|1/r < . 3.17
n1 nr i1
Then, by 3.17, we have

I2 C n1r ln n2r x2 E|Sxn ESxn |2 dx
n1 nr

C n1 r ln x2 Ea2ni Xni
2
I |ani Xni | nr x dx
n1 nr i1

C n1 r ln x2 Ea2ni X 2 I |ani X| nr x dx 3.18
n1 nr i1

C n1r ln P |ani X| > nr x dx
n1 nr i1

C n1 r ln x2 Ea2ni X 2 I |ani X| nr x dx C < .
n1 nr i1
The proof of Theorem 2.1 is completed.
Proof of Corollary 2.4. Note that

k
3.19
ani Xni sup ani Xni .
i1
k1 i1

Therefore, 2.8 and 2.9 hold by Theorem 2.1.
Proof of Corollary 2.5. By applying Theorem 2.1, taking p 2, ani n1/t for 1 i n,
and ani 0 for i > n, then we obtain 2.10. Similarly, taking 1, ani n1/t for 1 i n,
and ani 0 for i > n, we obtain 2.11 by Theorem 2.1.

Proof of Corollary 2.6. Let Xni Yi and ani n1/t nj1 ai j for all n 1, 0.
n1
i1 n1
i ni ni
3.20
Therefore, 2.12 and 2.13 hold.

Acknowledgment
The paper is supported by the National Natural Science Foundation of China no. 11271020
and 11201004, the Key Project of Chinese Ministry of Education no. 211077, the Natural
Science Foundation of Education Department of Anhui Province KJ2012ZD01, and the
Anhui Provincial Natural Science Foundation no. 10040606Q30 and 1208085MA11.
References
1 P. L. Hsu and H. Robbins, Complete convergence and the law of large numbers, Proceedings of the
National Academy of Sciences of the United States of America, vol. 33, pp. 2531, 1947.
2 P. Erdos,
On a theorem of hsu and robbins, Annals of Mathematical Statistics, vol. 20, pp. 286291,
1949.
3 L. E. Baum and M. Katz, Convergence rates in the law of large numbers, Transactions of the American
Mathematical Society, vol. 120, pp. 108123, 1965.
4 Y. S. Chow, On the rate of moment convergence of sample sums and extremes, Bulletin of the Institute
of Mathematics, vol. 16, no. 3, pp. 177201, 1988.
5 A. Gut, Complete convergence and ces`aro summation for i.i.d. random variables, Probability Theory
and Related Fields, vol. 97, no. 1-2, pp. 169178, 1993.
6 R. L. Taylor, R. F. Patterson, and A. Bozorgnia, A strong law of large numbers for arrays of rowwise
negatively dependent random variables, Stochastic Analysis and Applications, vol. 20, no. 3, pp. 643
656, 2002.
7 G. H. Cai and B. Xu, Complete convergence for weighted sums of -mixing sequences and its
application, Journal of Mathematics, vol. 26, no. 4, pp. 419422, 2006.
8 S. E. Ahmed, R. G. Antonini, and A. Volodin, On the rate of complete convergence for weighted sums
of arrays of banach space valued random elements with application to moving average processes,
Statistics & Probability Letters, vol. 58, no. 2, pp. 185194, 2002.
9 T. S. Kim and M. H. Ko, Complete moment convergence of moving average processes under
dependence assumptions, Statistics & Probability Letters, vol. 78, no. 7, pp. 839846, 2008.
10 Q. M. Shao, A moment inequality and its applications, Acta Mathematica Sinica, vol. 31, no. 6, pp.
736747, 1988.
11 W. F. Stout, Almost Sure Convergence, Academic Press, New York, NY, USA, 1974.
12 P. Y. Chen, T. C. Hu, and A. Volodin, Limiting behaviour of moving average processes under -
mixing assumption, Statistics & Probability Letters, vol. 79, no. 1, pp. 105111, 2009.
doi:10.1155/2012/597490
Research Article
On Huygens Inequalities and the Theory of Means

Jozsef
Sandor
Department of Mathematics, Babes-Bolyai University, Strada Kogalniceanu no. 1,
400084 Cluj-Napoca, Romania
Correspondence should be addressed to Jozsef

Sandor, jsandor@math.ubbcluj.ro
Received 24 March 2012; Accepted 20 August 2012
Academic Editor: Mowaaq Hajja
Copyright q 2012 Jozsef

Sandor. This is an open access article distributed under the Creative
By using the theory of means, various refinements of Huygens trigonometric and hyperbolic
inequalities will be proved. New Huygens type inequalities will be provided, too.
1. Introduction
The famous Huygens trigonometric inequality see e.g., 13 states that for all x 0, /2
one has
2 sin x tan x > 3x. 1.1
The hyperbolic version of inequality 1.1 has been established recently by Neuman
and Sandor 3:
2 sinh x tanh x > 3x, for x > 0. 1.2
Let a, b > 0 be two positive real numbers. The logarithmic and identric means of a and
b are defined by
ba
L La, b : b; La, a a,
for a /
ln b ln a
1/ba 1.3
1 bb
I Ia, b : b; Ia, a a,
for a /
e aa
respectively. Seierts mean P is defined by
ab
P P a, b : for a
/ b, P a, a a. 1.4
2 arcsina b/a b
Let
ab
A Aa, b : , G Ga, b ab,
2
1.5
1 1
H Ha, b 2
a b
denote the arithmetic, geometric, and harmonic means of a and b, respectively. These means
have been also in the focus of many research papers in the last decades. For a survey of
results, see, for example, 46. In what follows, we will assume a /
b.
Now, by remarking that letting a 1 sin x, b 1 sin x, where x 0, /2, in
P, G, and A, we find that
sin x
P , G cos x, A 1, 1.6
x
so Huygens inequality 1.1 may be written also as

3AG 2 1
P> 3 HA, A, G. 1.7
2G A A G
Here Ha, b, c denotes the harmonic mean of the numbers a, b, c:

1 1 1
Ha, b, c 3 . 1.8
a b c
On the other hand, by letting a ex , b ex in L, G, and A, we find that
sinh x
L , G 1, A cosh x, 1.9
x
so Huygens hyperbolic inequality 1.2 may be written also as

3AG 2 1
L> 3 HG, G, A. 1.10
2A G G A
2. First Improvements
b.
Suppose a, b > 0, a /
Theorem 2.1. One has
3AG
P > HL, A > HA, A, G, 2.1
2G A
3AG
L > HP, G > HG, G, A. 2.2
2A G
Proof. The inequalities P > HL, A and L > HP, G have been proved in paper 7 see
Corollary 3.2. In fact, stronger relations are valid, as we will see in what follows.
Now, the interesting fact is that the second inequality of 2.1, that is, 2LA/L A >
3AG/2GA becomes, after elementary transformations, exactly inequality 1.10, while the
second inequality of 2.2, that is, 2P G/P G > 3AG/2AG becomes inequality 1.7.
Another improvements of 1.7, respectively, 1.10 are provided by
Theorem 2.2. One has the inequalities:
3AG
2.3
3
P> A2 G > ,
2G A
3AG
2.4
3
L > G2 A > .
2A G
Proof. The first inequality of 2.3 is proved in 6, while the first inequality of 2.8 is a well-
known inequality due to Leach and Sholander 8 see 4 for many related references. The
second inequalities of 2.3 and 2.4 are immediate consequences of the arithmetic-geometric
inequality applied for A, A, G and A, G, G, respectively.
Remark 2.3. By 2.3 and 1.6, we can deduce the following improvement of the Huygens
inequality 1.1:
sin x 3 cos x
> 3 cos x > , x 0, . 2.5
x 2 cos x 1 2
From 2.1 and 1.6, we get

sin x 2L 3 cos x

> > , x 0, . 2.5
x L 1 2 cos x 1 2
Similarly, by 2.4 and 1.9, we get
sinh x 3 cosh x
3
> cosh x > , x > 0. 2.6
x 2 cosh x 1
From 2.2 and 1.9, we get
sinh x 2P 3 cosh x

> > , x > 0. 2.6
x P 1 2 cosh x 1
Here, L L1 sin x, 1 sin x, P P ex , ex .

We note that the first inequality of 2.5 has been discovered by Adamovic and
Mitrinovic see 3, while the first inequality of 2.6 by Lazarevic see 3.
Now, we will prove that inequalities 2.2 of Theorem 2.1 and 2.4 of Theorem 2.2
may be compared in the following way.

3AG
2.7
3
L> G2 A > HP, G > .
2A G
Proof. We must prove the second inequality of 2.7. For this purpose, we will use the inequal-
ity see 6:
2A G
P< . 2.8
3
This implies G/P > 3G/G 2A, so 1/21 G/P > 2G A/G 2A.
Now, we will prove that

2G A 3 G 2.9
> .
G 2A A
By letting x G/A 0, 1, inequality 2.9 becomes
2x 1
> 3 x. 2.10
x2
Put x a3 , where a 0, 1. After elementary transformations, inequality 2.10

becomes a 1a 13 < 0, which is true.
Note. The Referee suggested the following alternative proof: since P < 2A G/3and the
3
harmonic mean increases in both variables, it suces to prove stronger inequality A2 G >
H2A G/3, G which can be written as 2.9.

Remark 2.5. The following refinement of inequalities 2.6 is true:
sinh x 2P 3 cosh x
3
> cosh x > > , x > 0. 2.11
x P 1 2 cosh x 1
Unfortunately, a similar refinement to 2.7 for the mean

P is not possible, as by numeri-
3
cal examples one can deduce that generally HL, A and A2 G are not comparable. However,
in a particular case, the following result holds true.
Theorem 2.6. Assume that A/G 4. Then one has

3AG
2.12
3
P > HL, A > A2 G > .
2G A
First, prove one the following auxiliary results.

Lemma 2.7. For any x 4, one has

3
x 12 2 3 x 1 > x 4.
3
2.13
Proof. A computer computation shows that 2.13 is true for x 4. Now put x a3 in 2.13.
By taking logarithms, the inequality becomes

a3 1
fa 2 ln 9 ln a 3 ln2a 1 > 0. 2.14
2
An easy computation implies

a2a 1 a3 1 f a 3a 1 a2 a 3 . 2.15

As 42 4 3 2 3 2 22 3 3 2 1 3 2 3 > 0, we getthat f a > 0 for
3 3 3

a 4. This means that fa > f 4 > 0, as the inequality is true for a 4.
3 3 3
Proof of the theorem. We will apply the inequality:

3 AG 2 2.16
L> G ,
2
due to the author 9. This implies

1 A 1 4A3
1 < 1 3
N. 2.17
2 L 2 GA G2
By letting x A/G in 2.13, we can deduce

3 A 2.18
N< .
G
So

1 A 3 A
2.19
1 < .
2 L G

3
This immediately gives HL, A > A2 G.
Remark 2.8. If cos x 1/4, x 0, /2, then
sin x 2L 3 cos x
> > 3 cos x > , 2.20
x L 1 2 cos x 1

which is a refinement, in this case, of inequality 2.5 .
3. Further Improvements
AG 3AG
3.1
3
P> A2 G >
LA > > ,
L 2G A
AG 3AG
3.2
3
L > GP > G2 A > > .
P 2A G

Proof. The inequalities P > LA and L > GP are proved in 10. We will see, that further
refinements of these inequalities are true. Now, the second inequality of 3.1 follows by the
first inequality of 2.3, while the second inequality of 3.2 follows by the first inequality of
2.4. The last inequality is in fact an inequality by Carlson 11. For the inequalities on AG/P,
we use 2.3 and 2.8.
Remark 3.2. One has

sin x cos x 3 cos x
> L > 3 cos x > > , x 0, , 3.3
x L 2 cos x 1 2
sinh x cosh x 3 cosh x
3
> P > cosh x >
> , x > 0, 3.4
x P 2 cosh x 1

where L and P are the same as in 2.6 and 2.5 .

AL AG 3AG
P> LA > HA, L >
> > , 3.5
I L 2G A
I G 3AG
3.6
3
L>L > IG > P G > G2 A > .
AL 2A G
Proof. The first two inequalities of 3.5 one followed by the first inequality of 3.1 and the
fact that Gx, y > Hx, y with x L, y A.
Now, the inequality HA, L > AL/I may be written also as
AL
I> , 3.7
2
which has been proved in 4 see also 12.

Further, by Alzers inequality L2 > GI see 13 one has
L G
> 3.8
I L
and by Carlsons inequality L < 2G A/3 see 11, we get

AL AG 3AG
> > , 3.9
I L 2G A
so 3.5 is proved.
The first two inequalities of 3.6 have been proved by the author in 5. Since I > P
see 14 and by 3.2, inequalities 3.6 are completely proved.
Remark 3.4. One has the following inequalities:
sin x 2L L cos x 3 cos x

> L > > >
> , x 0, , 3.10
x L 1 I L 2 cos x 1 2
where I I1 sin x, 1 sin x;

sinh x sinh x ex coth x1 1 3 cosh x
> ex coth x1/2 > P >
3
> cosh x > .
x x cosh x sinh x/x 2 cosh x 1
3.11

3 AG 2 A 2G AL 3AG 3.12
P> A > A > AL > HA, L > > ,
2 3 I 2G A

3 AG 2 2A G 3 3AG 3.13
L> G > IG > G > P G > G2 A > .
2 3 2A G
Proof. In 3.12, we have to prove the first three inequalities, the rest are contained in 3.5.
The first inequality of 3.12 is proved in 6. For the second inequality, put A/G t > 1
By taking logarithms, we have to prove that

t1 t2
gt 4 ln 3 ln ln t > 0. 3.14
2 3
As g ttt 1t 2 2t 1 > 0, gt is strictly increasing, so
gt > g1 0. 3.15
The third inequality of 3.12 follows by Carlsons relation L < 2G A/3 see 11.
The first inequality of 3.13 is proved in 9, while the second one in 15. The third
inequality follows by I > 2A G/3 see 12, while the fourth one by relation 2.9. The
fifth one is followed by 2.3.
Remark 3.6. The first three inequalities of 3.12 oer a strong improvement of the first
inequality of 3.1; the same is true for 3.13 and 3.2.
4. New Huygens Type Inequalities

The main result of this section is contained in the following:

A G 2 3AA G A2G A
3 3AG 4.1
P> A > > > ,
2 5A G 2A G 2G A

3 A G 2 3GA G G2A G 3AG 4.2
L> G > > > .
2 5G A 2G A 2A G
Proof. The first inequalities of 4.1, respectively, 4.2 are the first ones in relations 3.12,
respectively, 3.13.
Now, apply the geometric mean-harmonic mean inequality:
3 3
3
xy2 3
xyy > ,
1 1 1 1 2 4.3

x y y x y
for x A, y A G/2 in order to deduce the second inequality of 4.1. The last two
inequalities become, after certain transformation,
A G2 > 0. 4.4
The proof of 4.2 follows on the same lines, and we omit the details.

Theorem 4.2. For all x 0, , one has
2
x
sin x 4 tan > 3x. 4.5
2
For all x > 0, one has
x
sinh x 4 tanh > 3x. 4.6
2
Proof. Apply 1.6 for P > 3AA G/5A G of 4.1.

As cos x 1 2cos2 x/2 and sin x 2 sinx/2 cosx/2, we get inequality 4.5. A
similar argument applied to 4.6, by an application of 4.2 and the formulae cosh x 1
2cosh2 x/2 and sinh x 2 sinhx/2 coshx/2.
Remarks 4.3. By 4.1, inequality 4.5 is a refinement of the classical Huygens inequality 1.1:
x

2 sin x tan x > sin x 4 tan > 3x. 4.3
2
Similarly, 4.6 is a refinement of the hyperbolic Huygens inequality 1.2:
x

2 sinh x tanh x > sinh x 4 tanh > 3x. 4.4
2
We will call 4.5 as the second Huygens inequality, while 4.6 as the second hyper-
bolic Huygens inequality.
In fact, by 4.1 and 4.2 refinements of these inequalities may be stated, too.
The inequality P > A2G A/2A G gives
sin x 2 cos x 1
> , 4.7
x cos x 2
or written equivalently:
sin x 3
> 2, x 0, . 4.8
x cos x 2 2
Acknowledgments
The author is indebted to Professor Edward Neuman for his support and discussions on this
topic. He also thanks the Referee for a careful reading of the paper and a new proof of
Theorem 2.4.
References
1 C. Huygens, Oeuvres Completes 18881940, Societe Hollondaise des Science, Haga, Gothenburg.
2 J. S. Sandor and M. Bencze, On Huygens trigonometric inequality, RGMIA Research Report
Collection, vol. 8, no. 3, article 14, 2005.
3 E. Neuman and J. Sandor, On some inequalities involving trigonometric and hyperbolic functions
with emphasis on the Cusa-Huygens, Wilker, and Huygens inequalities, Mathematical Inequalities &
4 J. Sandor, On the identric and logarithmic means, Aequationes Mathematicae, vol. 40, no. 2-3, pp.
261270, 1990.
5 J. Sandor, On refinements of certain inequalities for means, Archivum Mathematicum, vol. 31, no. 4,
pp. 279282, 1995.
6 J. Sandor, On certain inequalities for means. III, Archiv der Mathematik, vol. 76, no. 1, pp. 3440, 2001.
7 E. Neuman and J. Sandor, On the Schwab-Borchardt mean, Mathematica Pannonica, vol. 14, no. 2,
pp. 253266, 2003.
8 E. B. Leach and M. C. Sholander, Extended mean values. II, Journal of Mathematical Analysis and
9 J. Sandor, On certain inequalities for means. II, Journal of Mathematical Analysis and Applications, vol.
199, no. 2, pp. 629635, 1996.
10 E. Neuman and J. Sandor, On the Schwab-Borchardt mean. II, Mathematica Pannonica, vol. 17, no. 1,
pp. 4959, 2006.
11 B. C. Carlson, The logarithmic mean, The American Mathematical Monthly, vol. 79, pp. 615618, 1972.
12 J. Sandor, A note on some inequalities for means, Archiv der Mathematik, vol. 56, no. 5, pp. 471473,
1991.
13 H. Alzer, Two inequalities for means, La Societe Royale du Canada, vol. 9, no. 1, pp. 1116, 1987.
14 H. J. Seiert, Ungleichungen fur einen bestimmten Mittelwert, Nieuw Arch, Wisk, vol. 13, no. 42,
pp. 195198, 1995.
15 J. Sandor, New refinements of two inequalities for means, submitted.
doi:10.1155/2012/430692
Research Article
A Nice Separation of Some Seiffert-Type Means by
Power Means
Iulia Costin1 and Gheorghe Toader2

1
Department of Computer Sciences, Technical University of Cluj-Napoca, 400114 Cluj-Napoca, Romania
2
Department of Mathematics, Technical University of Cluj-Napoca, 400114 Cluj-Napoca, Romania
Correspondence should be addressed to Gheorghe Toader, gheorghe.toader@math.utcluj.ro
Received 21 March 2012; Accepted 30 April 2012
Academic Editor: Edward Neuman
Copyright q 2012 I. Costin and G. Toader. This is an open access article distributed under the
Creative Commons Attribution License, which permits unrestricted use, distribution, and
reproduction in any medium, provided the original work is properly cited.
Seiert has defined two well-known trigonometric means denoted by P and T. In a similar
way it was defined by Carlson the logarithmic mean L as a hyperbolic mean. Neuman and
Sandor completed the list of such means by another hyperbolic mean M. There are more known
inequalities between the means P, T, and L and some power means Ap . We add to these
inequalities two new results obtaining the following nice chain of inequalities A0 < L < A1/2 0. 1.1
Each mean is reflexive; that is,
Ma, a a, a > 0. 1.2
This is also used as the definition of Ma, a.

We will refer here to the following means:
i the power means Ap , defined by
1/p
ap bp
Ap a, b , 0;
p/ 1.3
2

ii the geometric mean G, defined as Ga, b ab, but verifying also the property
lim Ap a, b A0 a, b Ga, b; 1.4

p0
iii the first Seiert mean P, defined in 1 by
ab
Pa, b 1
, b;
a/ 1.5
2 sin a b/a b
iv the second Seiert mean T, defined in 2 by
ab
Ta, b , a
/ b; 1.6
2 tan1 a b/a b
v the Neuman-Sandor mean M, defined in 3 by
ab
Ma, b 1
, b;
a/ 1.7
2 sinh a b/a b
vi the Stolarsky means Sp,q defined in 4 as follows:

qap bp 1/pq

, pq p q /0

paq bq

p 1/ap bp

1 aa
, p q
/0
Sp,q a, b ep bb 1.8
p

1/p

ap bp

, 0, q 0
p/

pln a ln b

ab, p q 0.
The mean A1 A is the arithmetic mean and the mean S1,0 L is the logarithmic
mean. As Carlson remarked in 5, the logarithmic mean can be represented also by
ab
La, b 1
; 1.9
2 tanh a b/a b
thus the means P, T, M, and L are very similar. In 3 it is also proven that these means can
be defined using the nonsymmetric Schwab-Borchardt mean SB given by

b2 a2

, if a b
cosh1 a/b
see 6, 7. It has been established in 3 that
L SBA, G, P SBG, A, T SBA, A2 , M SBA2 , A. 1.11
2. Interlacing Property of Power Means

Given two means M and N, we will write M < N if
Ma, b < Na, b, b.

for a / 2.1
It is known that the family of power means is an increasing family of means, thus
A p < Aq , if p < q. 2.2
Of course, it is more dicult to compare two Stolarsky means, each depending on two
parameters. To present the comparison theorem given in 8, 9, we have to give the definitions
of the following two auxiliary functions:

|x| y
, x/y
k x, y xy

signx, x y,
2.3

L x, y , x > 0, y > 0
l x, y
0, x 0, y 0, xy 0.
Theorem 2.1. Let p, q, r, s R. Then the comparison inequality
Sp,q Sr,s 2.4
holds true if and only if pq r s, and (1) lp, q lr, s if 0 minp, q, r, s, (2) kp, q kr, s
if minp, q, r, s < 0 < maxp, q, r, s, or (3) lp, q lr, s if maxp, q, r, s 0.
We need also in what follows an important double-sided inequality proved in 3 for
the Schwab-Borchardt mean:
a 2b
2.5
3
ab2 < SBa, b < , b.
a/
3
Being rather complicated, the Seiert-type means were evaluated by simpler means,
first of all by power means. The evaluation of a given mean M by power means assumes the
determination of some real indices p and q such that Ap < M < Aq . 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 Ar if p < r < q.
For the logarithmic mean in 10, it was determined the optimal evaluation
A0 < L < A1/3 . 2.6
For the Seiert means, there are known the evaluations
A1/3 < P < A2/3 , 2.7
proved in 11 and
A 1 < T < A2 , 2.8
given in 2. It is also known that
A1 < M < T, 2.9
as it was shown in 3. Moreover in 12 it was determined the optimal evaluation
Aln 2/ ln < P < A2/3 . 2.10
Using these results we deduce the following chain of inequalities:
A0 < L < A1/2 < P < A1 < M < T < A2 . 2.11
To prove the full interlacing property of power means, our aim is to show that A3/2 can be put
between M and T. We thus obtain a nice separation of these Seiert-type means by power
means which are evenly spaced with respect to their order.
3. Main Results
We add to the inequalities 2.11 the next results.
Theorem 3.1. The following inequalities
M < A3/2 < T 3.1
are satisfied.
Proof. First of all, let us remark that A3/2 S3,3/2 . So, for the first inequality in 3.1, it is
sucient to prove that the following chain of inequalities
A2 2A
M< < S3,1 < S3,3/2 3.2
3
is valid. The first inequality in 3.2 is a simple consequence of the property of the mean M
given in 1.11 and the second inequality from 2.5. The second inequality can be proved by
direct computation or by taking a 1 t, b 1 t, 0 < t < 1 which gives

1 t2 2 3 t2 3.3
< ,
3 3
which is easy to prove. The last inequality in 3.2 is given by the comparison theorem of the
Stolarsky means. In a similar way, the second inequality in 3.1 is given by the relations

S3,3/2 < S4,1 3
AA22 < T. 3.4
The first inequality is again given by the comparison theorem of the Stolarsky means. The
equality in 3.4 is shown by elementary computations, and the last inequality is a simple
consequence of the property of the mean T given in 1.11 and the first inequality from 2.5.
Corollary 3.2. The following two-sided inequality
x x
1
< A3/2 1 x, 1 x < , 3.5
sinh x tan1 x
is valid for all 0 < x < 1.
Acknowledgment
The authors wish to thank the anonymous referee for oering them a simpler proof for their
results.
References
1 H.-J. Seiert, Problem 887, Nieuw Archief voor Wiskunde, vol. 11, no. 2, p. 176, 1993.
2 H.-J. Seiert, Aufgabe 16, Die Wurzel, vol. 29, pp. 221222, 1995.
3 E. Neuman and J. Sandor, On the Schwab-Borchardt mean, Mathematica Pannonica, vol. 14, no. 2,
pp. 253266, 2003.
4 K. B. Stolarsky, Generalizations of the logarithmic mean, Mathematics Magazine, vol. 48, no. 2, pp.
8792, 1975.
5 B. C. Carlson, The logarithmic mean, The American Mathematical Monthly, vol. 79, pp. 615618, 1972.
6 J. M. Borwein and P. B. Borwein, Pi and the AGM: A Study in Analytic Number Theory and Computational
Complexity, John Wiley & Sons, New York, NY, USA, 1987.
7 B. C. Carlson, Algorithms involving arithmetic and geometric means, The American Mathematical
Monthly, vol. 78, pp. 496505, 1971.
8 Zs. Pales, Inequalities for dierences of powers, Journal of Mathematical Analysis and Applications,
vol. 131, no. 1, pp. 271281, 1988.
9 E. B. Leach and M. C. Sholander, Multi-variable extended mean values, Journal of Mathematical
Analysis and Applications, vol. 104, no. 2, pp. 390407, 1984.
10 T. P. Lin, The power mean and the logarithmic mean, The American Mathematical Monthly, vol. 81,
pp. 879883, 1974.
11 A. A. Jagers, Solution of problem 887, Nieuw Archief voor Wiskunde, vol. 12, pp. 230231, 1994.
12 P. A. Hasto, Optimal inequalities between Seierts means and power means, Mathematical
Inequalities & Applications, vol. 7, no. 1, pp. 4753, 2004.

Bullen - Means and Their Inequalities

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bullen - Means and Their Inequalities

Uploaded by

Copyright:

Available Formats

International Journal of Mathematics and Mathematical Sciences

Means and Their Inequalities

Means and Their Inequalities

Guest Editors: Mowaffaq Hajja, Peter S. Bullen,

Logic and Set Theory

Oscar Blasco, Spain R. Grimshaw, UK Ram N. Mohapatra, USA

Probability and Statistics

Some Elementary Aspects of Means, Mowaffaq Hajja

On Some Intermediate Mean Values, Slavko Simic

Refinements of Inequalities among Difference of Means, Huan-Nan Shi, Da-Mao Li,

On Huygens Inequalities and the Theory of Means, Jozsef Sandor

Mowaffaq Hajja,1 Peter S. Bullen,2 Janusz Matkowski,3

Correspondence should be addressed to Mowaffaq Hajja; mowhajja@yahoo.com

Received 24 June 2013; Accepted 24 June 2013

Correspondence should be addressed to Mowaffaq Hajja; mowhajja@yahoo.com

Received 24 December 2012; Accepted 31 March 2013

Academic Editor: Peter Bullen

1. Definitions and Terminology If M and N are two -dimensional means on J, then we

pp. 8490]. They are usually denoted by A, G, and H, since

Correspondence should be addressed to Slavko Simic; ssimic@turing.mi.sanu.ac.rs

Received 25 June 2012; Revised 9 December 2012; Accepted 16 December 2012

Academic Editor: Mowaffaq Hajja

1. Introduction are the harmonic, geometric, logarithmic, identric, arith-

An inverse problem is to find best possible approximation or, more tightly,

Finally, Lin showed in [3] that { 1 1

holds for each R+ . (, ) 1 (, ) (, ) 0 (, ) (, )

where 3.3. Proof of Theorem 5, Part (1). We will prove a general

((1 ) + )] (p, x) ++1 (p, x) +1 (p, x) + (p, x) , (41)

Theorem 7. For an arbitrary probability law of random (, )

Since 0 < < , we get 0 < < 1. Also, we get

(1 + ) log (1 + ) + (1 ) log (1 ) (1 + ) log (1 + ) (1 ) log (1 )

Hence by (48), we get () > 0, that is, () is monotone 2 (1 + )6 + (1 )6 2

Remark 10. Since () is monotone increasing, we also get

As for the part 8, applying the above transformation we

where 0 < < , = ( )/( + ).

Correspondence should be addressed to Zhen-Hang Yang, yzhkm@163.com

Received 22 July 2012; Accepted 26 September 2012

Academic Editor: Edward Neuman

the q, r, s, k, m R, a, b, c, d R with b/a > d/c 1, Hf p, q is the so-called two-parameter

Definition 1.1. Let f: R2 \ {x, x, x R } R be a first-order homogeneous continuous

Hf 0, 0; a, b  afx 1,1/f1,1 bfy 1,1/f1,1 , if p  q  0,

For simpleness, Hf p, q; a, b is also denoted by Hf p, q or Hf a, b.

L  Lx, y  x y/ln x ln y if x, y > 0 with x /

where Ix, y  e1 xx /yy 1/xy if x, y > 0, with x /

where Za, b  aa/ab bb/ab see 19.

2. Main Results and Proofs

In 14, it is important to another key decision function defined by

Note that the function T defined by

Moreover, it has revealed in 14, 3.5 that

Theorem 2.1. Suppose that f: R R R is a symmetric, first-order homogenous, and three-

Partial derivative leads to

T  v; a, b T  v; c, d  v3 T3 av , bv T3 cv , dv

Clearly, the monotonicity of Q1f is converse if q 0, k 0.

Proof. By 2.13 in 13 we have

Direct calculation leads to

Clearly, the monotonicity of Q2f is converse if k 0 and k  m 0.

Proof. From 2.13 in 13, it is derived that

Simple calculation yields

  

It is evident that the monotonicity of Q3f is converse if m < 0, 2m k 0.

Proof. By 2.13 in 13, ln Hf pr, ps can be expressed in integral form

This proof is accomplished.

ii Q2L is defined by

for fixed m > <0, k 0, 2 m 2m, 0.

the q, r, s, k, m R, a, b, c, d R with b/a > d/c 1, Hf p, q is the so-called two-parameter

Definition 1.1. Let f: R2 \ {x, x, x R } R be a first-order homogeneous continuous

Hf 0, 0; a, b afx 1,1/f1,1 bfy 1,1/f1,1 , if p q 0,

L Lx, y x y/ln x ln y if x, y > 0 with x /

where Ix, y e1 xx /yy 1/xy if x, y > 0, with x /

where Za, b aa/ab bb/ab see 19.

In 14, it is important to another key decision function defined by

Moreover, it has revealed in 14, 3.5 that

Theorem 2.1. Suppose that f: R R R is a symmetric, first-order homogenous, and three-

T v; a, b T v; c, d v3 T3 av , bv T3 cv , dv

Proof. By 2.13 in 13 we have

Clearly, the monotonicity of Q2f is converse if k 0 and k m 0.

Proof. From 2.13 in 13, it is derived that

Proof. By 2.13 in 13, ln Hf pr, ps can be expressed in integral form

for fixed m > <0, k 0, 2 m 2m, 0.

for fixed k, r, s R with kr s > <0.

Inequalities 3.6 in the case of d c are due to Alzer 22.

When d c, inequalities 3.7 are changed as Alzers ones given in 24.

Putting m 1/2, k 1/3 leads to another new inequality

Theorem 4.1. Suppose that f: R R R is a symmetric, first-order homogenous, and three-time

16 Z. H. Yang, On the log-convexity of two-parameter homogeneous functions, Mathematical Inequa-

MSA a, b Sa, b Aa, b,

MSN2 a, b Sa, b N2 a, b,

MSN3 a, b Sa, b N3 a, b,

MSN1 a, b Sa, b N1 a, b,

MSG a, b Sa, b Ga, b,

MSH a, b Sa, b Ha, b, 1.4

MAN2 a, b Aa, b N2 a, b,

MAG a, b Aa, b Ga, b,

MAH a, b Aa, b Ha, b,

MN2 N1 a, b N2 a, b N1 a, b,

MN2 G a, b N2 a, b Ga, b

Definition 2.1 see 4, 20. Let x x1 , . . . , xn Rn and y y1 , . . . , yn Rn .

Definition 2.2 see 15. Let x x1 , . . . , xn Rn and y y1 , . . . , yn Rn .

holds for any x x1 , . . . , xn 0 , then is a Schur-geometrically convex Schur-geometrically

Lemma 2.5. For a, b R2 0, 0, one has

21 2t2 2 51 2t 1 2t

To prove that the function DAHSH is Schur-geometrically convex in R2 it is enough to

so 0. By Lemma 2.4, we get that DAGSG is Schur-geometrically convex in R2 .

so 0; it follows that DN2 N1 AH is Schur-geometrically convex in R2 .

DANA N2 SN2 a, b 4DSN2 SA a, b, 3.28

and, consequently, the function DAN2 SN2 is Schur-geometrically convex in R2 .

is Schur-geometrically convex in R2 it is enough to notice that

DSN1 SH a, b DSGAH a, b. 3.30

which means that the function DSN3 SA is Schur-geometrically convex in R2 .

Chow 4 established the following refinement which is a complete moment

m sup |P B | A P B| ; A k1 , B

where kj {Xi ; j i k}, 1 j k .

E|Xn |p I|Xn | > x CE|X|p I|X| > x. 1.6

Proof. By Lemma 5.4.4 in 11 and Holders