ArticlePDF Available

On a mixed arithmetic-geometric mean inequality

Authors:

Abstract

We extend a result of Holland on a mixed arithmetic-geometric mean inequality.
arXiv:1305.3403v1 [math.CA] 15 May 2013
ON A MIXED ARITHMETIC-GEOMETRIC MEAN INEQUALITY
PENG GAO
Abstract. We extend a result of Holland on a mixed arithmetic-geometric mean inequality.
1. Introduction
Let Mn,r(q,x) be the generalized weighted power means: Mn,r(q,x) = (Pn
i=1 qixr
i)1
r, where
q= (q1, q2,··· , qn), x= (x1, x2,··· , xn), qi>0,1inwith Pn
i=1 qi= 1. Here Mn,0(q,x)
denotes the limit of Mn,r (q,x) as r0+. Unless specified, we always assume xi>0,1in.
When there is no risk of confusion, we shall write Mn,r for Mn,r(q,x) and we also denote An, Gn
for the arithmetic mean Mn,1, geometric mean Mn,0, respectively.
For fixed x= (x1,··· , xn),w= (w1,··· , wn) with w1>0, wi0, we define xi= (x1,··· , xi),
wi= (w1,··· , wi), Wi=Pi
j=1 wj,Mi,r =Mi,r(wi/Wi,xi),Mi,r = (M1,r,··· , Mi,r). The following
result on mixed mean inequalities is due to Nanjundiah [5] (see also [1]):
Theorem 1.1. Let r > s and n2. If for 2kn1,WnwkWkwn>0. Then
Mn,s(Mn,r )Mn,r (Mn,s),
with equality holding if and only if x1=···=xn.
It is easy to see that the case r= 1, s = 0 of Theorem 1.1 follows from the following Popoviciu-
type inequalities established in [4] (see also [1, Theorem 9]):
Theorem 1.2. Let n2. If for 2kn1,WnwkWkwn>0, then
Wn1ln Mn1,0(Mn1,1)ln Mn1,1(Mn1,0)Wnln Mn,0(Mn,1)ln Mn,1(Mn,0)
with equality holding if and only if xn=Mn1,0=Mn1,1(Mn1,0).
In [6], the following Rado-type inequalities were established:
Theorem 1.3. Let s < 1and n2. If for 2kn1,WnwkWkwn>0, then
Wn1Mn1,s(Mn1,1)Mn1,1(Mn1,s)WnMn,s(Mn,1)Mn,1(Mn,s )
with equality holding if and only if x1=···=xnand the above inequality reverses when s > 1.
The above theorem is readily seen to imply Theorem 1.1. In [3], Holland further improved the
condition in Theorem 1.3 for the case s= 0 by proving the following:
Theorem 1.4. Let n2. If W2
n1wnPn2
i=1 Wiwith the empty sum being 0, then
(1.1) Wn1Mn1,0(Mn1,1)Mn1,1(Mn1,0)WnMn,0(Mn,1)Mn,1(Mn,0)
with equality holding if and only if x1=···=xn.
It is our goal in this paper to extend the above result of Holland by considering the validity of
inequality (1.1) for the case W2
n1< wnPn2
i=1 Wi. Note that this only happens when n3. In the
next section, we apply the approach in [2] to prove the following
2000 Mathematics Subject Classification. Primary 26D15.
Key words and phrases. Arithmetic-geometric mean, mixed-mean inequality.
1
2 PENG GAO
Theorem 1.5. Let n3. Inequality (1.1) holds when the following conditions are satisfied:
0<wnPn2
i=1 Wi
W2
n1
1w1
wn
,Wn1
Wn
n2
Y
i=1 Wi+1
Wi
Wiwn
W2
n11,(1.2)
Wn1wn
Wnw1 n2
X
i=1
Wiwn
W2
n1
1!+wn
Wn!n2
Y
i=1 Wi+1
wi+1
wi+1
Wn11.
In the above theorem, we do not give the condition for the equality in (1.1) to hold. In Section
3, we show that there do exist sequences {wi}n
i=1 that satisfy the conditions of Theorem 1.5.
2. Proof of Theorem 1.5
We may assume that xi>0, wi>0,1inand the case xi= 0 or wi= 0 for some ifollows
by continuity. We recast (1.1) as
(2.1) Gn(An)Wn1
Wn
Gn1(An1)wn
Wn
Gn0.
Note that
Gn(An) = Gn1(An1)Wn1/Wn
Awn/Wn
n, Gn1(An1) = An
n1
Y
i=1 Ai
Ai+1 Wi/Wn1.(2.2)
Dividing Gn(An) on both sides of (2.1) and using (2.2), we can recast (2.1) as:
(2.3) Wn1
Wn
n1
Y
i=1 Ai
Ai+1 Wiwn/(Wn1Wn)+wn
Wn
n
Y
i=1 xi
Aiwi/Wn
1.
We express xi= (WiAiWi1Ai1)/wi,1inwith W0=A0= 0 to recast (2.3) as
Wn1
Wn
n1
Y
i=1 Ai
Ai+1 Wiwn/(Wn1Wn)+wn
Wn
n
Y
i=1 WiAiWi1Ai1
wiAiwi/Wn
1.
We set yi=Ai/Ai+1, 1 i2 to further recast the above inequality as
(2.4) Wn1
Wn
n1
Y
i=1
yWiwn/(Wn1Wn)
i+wn
Wn
n1
Y
i=1 Wi+1
wi+1
Wi
wi+1
yiwi+1/Wn
1.
We now regard the right-hand side expression above as a function of yn1only and define
f(yn1) = Wn1
Wn
c·ywn/Wn
n1+wn
Wn
c·Wn
wn
Wn1
wn
yn1wn/Wn
,
where
c=
n2
Y
i=1
yWiwn/(Wn1Wn)
i, c=
n2
Y
i=1 Wi+1
wi+1
Wi
wi+1
yiwi+1/Wn
.
On setting f(yn1) = 0, we find that
yn1= Wn1
Wn
+wn
Wnc
cWn/Wn1!1
.
It is easy to see that f(yn1) is maximized at the above value with its maximal value being
Wn1
Wn
cWn/Wn1+wn
Wn
(c)Wn/Wn1Wn1/Wn
.
ON A MIXED ARITHMETIC-GEOMETRIC MEAN INEQUALITY 3
Thus, in order for inequality (2.4) to hold, it suffices to have
Wn1
Wn
cWn/Wn1+wn
Wn
(c)Wn/Wn11.
Explicitly, the above inequality is
g(y1, y2,...,yn2) := Wn1
Wn
n2
Y
i=1
yWiwn/W 2
n1
i+wn
Wn
n2
Y
i=1 Wi+1
wi+1
Wi
wi+1
yiwi+1/Wn11.
Let (a1, a2,...,an2)[0, W2/W1]×[0, W3/W2]×...×[0, Wn1/Wn2] be the point in which
the absolute maximum of gis reached. If one of the aiequals 0 or Wi+1 /Wi, then it is easy to see
that we have
g(a1, a2,...,an2)max Wn1
Wn
n2
Y
i=1 Wi+1
WiWiwn/W 2
n1
,wn
Wn
n2
Y
i=1 Wi+1
wi+1 wi+1/Wn1!.(2.5)
If the point (a1, a2,...,an2) is an interior point, then we have
g(a1, a2,...,an2) = 0.
It follows that for every 1 in2, we have
Qn2
i=1 aWiwn/W 2
n1
i
Qn2
i=1 Wi+1
wi+1 Wi
wi+1 aiwi+1/Wn1=ai
Wi+1
wi+1 Wi
wi+1 ai
:= 1
d,(2.6)
where d > 0 is a constant (depending on the wi). In terms of d, we have
ai=Wi+1
dwi+1 +Wi
.
We use this to recast the first equation in (2.6) as
n2
Y
i=1 Wi+1
dwi+1 +WiWiwn/W 2
n1
=1
d
n2
Y
i=1 Wi+1
wi+1
·dwi+1
dwi+1 +Wiwi+1/Wn1.
We recast the above equality as
ln n2
Y
i=1
(Wi+1)Wiwn/W 2
n1wi+1/Wn1!
=
n2
X
i=1 Wiwn
W2
n1
wi+1
Wn1ln (dwi+1 +Wi)w1
Wn1
ln d:= h(d).
Note that the above equality holds when d= 1 and we have
h(d) =
n2
X
i=1 Wiwn
W2
n1
wi+1
Wn11
d+Wi/wi+1
w1
Wn1
1
d.
As
Wiwn
W2
n1
wi+1
Wn1
0Wi
wi+1
Wn1
wn
,
4 PENG GAO
it follows that
h(d)
n2
X
i=1 Wiwn
W2
n1
wi+1
Wn11
d+Wn1/wn
w1
Wn1
1
d
Pn2
i=1 Wiwn
W2
n1
1dw1
wn
d(d+Wn1/wn).
Therefore, when
n2
X
i=1
Wiwn
W2
n1
10,
the function h(d) is a decreasing function of dso that d= 1 is the only value that satisfies (2.6) and
we have ai= 1 correspondingly with g(1,1,...,1) = 1 and this allows us to recover Theorem 1.4,
by combining the observation that the right-hand side expression of (2.5) is an increasing function
of wnfor fixed wi,1in1 with the discussion in the next section.
Suppose now
n2
X
i=1
Wiwn
W2
n1
1>0,(2.7)
then the function h(d) is a decreasing function of dfor
d
w1
wn
Pn2
i=1 Wiwn
W2
n1
1:= d0.
If follows that if d01, then d= 1 is the only value d0that satisfies (2.6) and we have ai= 1
correspondingly with g(1,1,...,1) = 1. We further note that for any dd0satisfying (2.6), the
value of gat the corresponding aisatisfies
g(a1, a2,...,an2) = Wn1
dWn
+wn
Wnn2
Y
i=1 Wi+1
wi+1
·dwi+1
dwi+1 +Wiwi+1/Wn1
Wn1
dWn
+wn
Wnn2
Y
i=1 Wi+1
wi+1 wi+1/Wn1
Wn1
d0Wn
+wn
Wnn2
Y
i=1 Wi+1
wi+1 wi+1/Wn1.
Combining this with (2.5), we see that inequality (2.4) holds when the conditions in (1.2) are
satisfied and this completes the proof of Theorem 1.5.
3. A Further Discussion
We show in this section that there does exist sequences {wi}n
i=1 satisfying the conditions of
Theorem 1.5. To see this, we note that the left-hand side expression of (2.7) vanishes when
wn=W2
n1
Pn2
i=1 Wi
.(3.1)
It follows by continuity that such sequences {wi}n
i=1 satisfying the conditions of Theorem 1.5
exist as long as the positive sequence {wi}n
i=1 with wi,1in1 being arbitrary and wndefined
ON A MIXED ARITHMETIC-GEOMETRIC MEAN INEQUALITY 5
by (3.1) satisfies the last two inequalities of (1.2) with strict inequalities there. It is readily checked
that these inequalities become
Pn2
i=1 Wi
Pn1
i=1 Wi!Pn2
i=1 Wi
WWn2
n1<
n2
Y
i=1
Wwi
i,(3.2)
Wn1
Pn1
i=1 Wi
n2
Y
i=1 Wi+1
wi+1 wi+1/Wn1<1.(3.3)
Now, it is easy to see that inequality (3.2) holds when n= 3. It follows that it holds for all n3
by induction as long as we have
Pn2
i=1 Wi
Pn1
i=1 Wi!Pn2
i=1 Wi
WWn1
n1 Pn1
i=1 Wi
Pn
i=1 Wi!Pn1
i=1 Wi
WWn1
n.(3.4)
The right-hand side expression above when regarded as a function of Wnonly is maximized at
Wn=Wn1Pn1
i=1 Wi
Pn2
i=1 Wi
.
As the inequality in (3.4) becomes an equality with this value of Wn, we see that inequality (3.2)
does hold for all n3.
Note that it follows from the arithmetic-geometric mean inequality that
Wn1
Pn1
i=1 Wi
n2
Y
i=1 Wi+1
wi+1 wi+1/Wn1Wn1
Pn1
i=1 Wi n2
X
i=0
Wi+1
wi+1
·wi+1
Wn1!= 1.
As one checks easily that the above inequality is strict in our case, we see that inequality (3.3)
also holds. We therefore conclude the existence of sequences {wi}n
i=1 satisfying the conditions of
Theorem 1.5.
References
[1] P. S. Bullen, Inequalities due to T. S. Nanjundiah, Recent progress in inequalities, Kluwer Acad. Publ., Dordrecht,
1998, 203-211.
[2] P. Gao, A note on mixed-mean inequalities, J. Inequal. Appl.,2010 (2010), Art. ID 509323, 8 pp.
[3] F. Holland, An inequality between compositions of weighted arithmetic and geometric means, JIPAM. J. Inequal.
Pure Appl. Math.,7(2006), Article 159, 8 pp. (electronic).
[4] K. Kedlaya, A weighted mixed-mean inequality, Amer. Math. Monthly,106 (1999), 355-358.
[5] T. S. Nanjundiah, Sharpening of some classical inequalities, Math Student,20 (1952), 24-25.
[6] C. Tarnavas and D. Tarnavas, An inequality for mixed power means, Math. Inequal. Appl.,2(1999), 175–181.
Department of Mathematics, School of Mathematics and System Sciences, Beijing University of
Aeronautics and Astronautics, P. R. China
E-mail address:penggao@buaa.edu.cn
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
Let ℙ denote the collection of positive sequences defined on ℕ. Fix w ∈ ℙ. Let s, t, respectively, be the sequences of partial sums of the infinite series ∑w k and ∑s k, respectively. Given x ∈ ℙ, define the sequences A(x) and G(x) of weighted arithmetic and geometric means of x by A n(x) = ∑ k=1n wk/sn x k, G n (x) = ∏ k=1n x kwk/sn ,n = 1, 2, . . . Under the assumption that log t is concave, it is proved that A(G(x)) ≤ G(A(x)) for all x ∈ ℙ, with equality if and only if x is a constant sequence.
Article
Full-text available
We give a simpler proof of a result of Holland concerning a mixed arithmetic-geometric mean inequality. We also prove a result of mixed-mean inequality involving the symmetric means.
Article
In this note we give Nanjundiah’s proofs of his mixed geometric-arithmetic mean inequalities; in particular his use of inverse means is explained.
Article
In 1992 Frank Hollad [1] stated the following inequality (1) (A 1 A 2 A n )<SUP>1/n</SUP> greater than or equal to 1/n (G 1 G 2 ...G n ) where A(k), G(k), k = 1, 2,..., n are arithmetic and geometric means, respectively, of positive numbers a(1), a(2),,...,a(k). In 1994 Kiran Kedlaya [2] gave a combinatorial proof of (1). In 1995 Takashi Matsuda [3] gave another proof of(1). In 1996 B. Mend and J. Pecaric [5] proved the following generalization of inequality (1) involving power means: (2) if S > T then, m(r,s) (a) greater than or equal to m(s,r) (a), where m(r,s) (a) is defined by the following definition (1.2). In this article a more general inequality, which concern weighted power means, is proved.