Professional Documents
Culture Documents
The Logarithmic Mean: Rajendra Bhatia
The Logarithmic Mean: Rajendra Bhatia
The Logarithmic Mean: Rajendra Bhatia
(2 )
a + bp 1 = p
B p (a ;b) =
; 1 < p < 1 ;
(4 )
2
RESONANCE June 2008
Rajendra Bhatia is a
Professor at the Indian
Statistical Institute, New
Delhi. Some of his recent
work is on means of
matrices. But he is not a
mean theorist.
Keywords
Inequalities, logarithmic mean,
power mean, Holder mean, arithmetic-geometric mean (AGM),
arithmetic-mean geometricmean inequality (AM-GM inequality), Heinz mean.
583
GENERAL ARTICLE
A substantial part of
the mathematics
classic Inequalities
by Hardy, Littlewood
th en
applications.
B 0 (a ;b) = G (a ;b):
(5 )
a + bp 1 = p
B 1 (a ;b) := lim
= m a x (a ;b);
p! 1
2
p
a + bp 1 = p
B 1 (a ;b) :=
lim
= m in (a ;b):
p ! 1
2
A little ca lcu la tio n sh ow s th a t
B p (a ;b) B q (a ;b)
if p q :
(6 )
584
a b
lo g a lo g b
fo r a 6= b;
(7 )
GENERAL ARTICLE
w ith th e u n d ersta n d in g th a t
The logarithmic
mean always falls
L (a ;a ) = lim L (a ;b) = a :
b! a
betwen the
T h ere a re o th er in terestin g rep resen ta tio n s fo r th is o b ject, a n d th e rea d er sh o u ld ch eck th e va lid ity o f th ese
fo rm u la s:
Z1
L (a ;b) =
a tb 1 t d t;
(8 )
0
Z1
1
dt
;
=
(9 )
L (a ;b)
ta + (1 t)b
0
Z1
1
dt
:
=
(1 0 )
L (a ;b)
(t + a )(t + b)
0
geometric mean
and the arithmetic
mean.
(1 1 )
w 1
lo g w
fo r w > 1 :
R ep la cin g w b y u 2 ; th is is th e sa m e a s say in g
2 lo g u
u2 1
u
fo r u > 1 :
(1 2 )
T h e tw o fu n ctio n s f (u ) = 2 lo g u ; a n d g (u ) = (u 2 1 )= u
a re eq u a l to 0 a t u = 1 ; a n d a sm a ll ca lcu la tio n sh o w s
th a t f 0(u ) < g 0(u ) fo r u > 1 : T h is p rov es th e d esired
in eq u a lity (1 2 ), a n d w ith it th e rst in eq u a lity in (1 1 ).
In th e sa m e w ay, th e seco n d o f th e in eq u a lities (1 1 ) ca n
b e red u ced to
u 1
lo g u
fo r u 1 :
u + 1
2
585
GENERAL ARTICLE
In this form we
recognize it as one
of the fundamental
inequalities of
analysis:
t sinh t for all
t 0.
a+ b 2
2
t + t(a + b) +
2
fo r a ll t 0 : U sin g th is o n e
Z1
Z1
dt
(t +
(t + a +2 b )2
0
0
n d s th a t
dt
a )(t + b)
Z1
0
(t +
dt
p
a b)2
E va lu a tio n o f th e in teg ra ls sh ow s th a t th is is th e sa m e
a s th e a ssertio n in (1 1 ).
S in ce a a n d b a re p o sitiv e, w e ca n n d rea l n u m b ers
x a n d y su ch th a t a = e x a n d b = e y : T h en th e rst
in eq u a lity in (1 1 ) is eq u iva len t to th e sta tem en t
e (x +
or
1
y )= 2
ex ey
;
x y
e (x y )= 2 e (y x )= 2
:
x y
T h is ca n b e ex p ressed a lso a s
1
sin h (x y )= 2
:
(x y )= 2
In th is fo rm w e reco g n ise it a s o n e o f th e fu n d a m en ta l
in eq u a lities o f a n a ly sis: t sin h t fo r a ll t 0 : V ery
sim ila r ca lcu la tio n s sh ow th a t th e seco n d in eq u a lity in
(1 1 ) ca n b e red u ced to th e fa m ilia r fa ct ta n h t t fo r
a ll t 0 :
E a ch o f o u r th ree p ro o fs sh ow s th a t if a 6= b; th en
G (a ;b) < L (a ;b) < A (a ;b): O n e o f th e rea so n s fo r th e
586
GENERAL ARTICLE
in terest in (1 1 ) is th a t it p rov id es a re n em en t o f th e
fu n d a m en ta l in eq u a lity b etw een th e g eo m etric a n d th e
a rith m etic m ea n s.
T h e lo g a rith m ic m ea n p lay s a n im p o rta n t ro le in th e
stu d y o f co n d u ctio n o f h ea t in liq u id s ow in g in p ip es.
L et u s ex p la in th is b rie y. T h e ow o f h ea t b y stea d y
u n id irectio n a l co n d u ctio n is g ov ern ed b y N ew ton's law
of coolin g : if q is th e ra te o f h ea t ow a lo n g th e x -a x is
a cro ss a n a rea A n o rm a l to th is a x is, th en
q= kA
dT
;
dx
The logarithmic
mean plays an
important role in
the study of
conduction of heat
in liquids flowing in
pipes.
(1 3 )
4 T
;
4 x
(1 4 )
587
GENERAL ARTICLE
x fro m th e a x is is 2 x L : S o , th e to ta l h ea t ow q a cro ss
th e sectio n o f th e p ip e b o u n d ed b y tw o co a x ia l cy lin d ers
a t d ista n ce x 1 a n d x 2 fro m th e a x is, u sin g (1 3 ), is seen
to sa tisfy th e eq u a tio n
Z x2
dx
q
= k4 T;
(1 5 )
x 1 2 x L
o r,
q=
k 2 L 4 T
:
lo g x 2 lo g x 1
= 2 L
x2 x1
2 L x 2 2 L x 1
:
=
lo g x 2 lo g x 1
lo g 2 L x 2 lo g 2 L x 1
In o th er w o rd s,
A
A2 A1
;
lo g A 2 lo g A 1
th e lo g a rith m ic m ea n o f th e tw o a rea s b o u n d in g th e
cy lin d rica l sectio n u n d er co n sid era tio n . In th e en g in eerin g litera tu re th is is ca lled th e logarithm ic m ean area.
If in stea d o f tw o co a x ia l cy lin d ers w e co n sid er tw o co n cen tric sp h eres, th en th e cro ss-sectio n a l a rea is p ro p o rtio n a l to th e sq u a re o f th e d ista n ce fro m th e cen tre. In
th is ca se w e h av e, in stea d o f (1 5 ),
Z x2
dx
q
= k4 T :
2
x1 4 x
A sm a ll ca lcu la tio n sh ow s th at in th is ca se
p
Am =
A 1A 2;
588
GENERAL ARTICLE
a t bt
a t + bt
;
t(lo g a log b)
2
fro m w h ich w e g et
t(a b)t= 2
a b
a b
a t + bt a b
t
:
a t bt
a t bt
lo g a lo g b
2
a
at
a t + bt a
t
2 at
t(a b)t= 2
b
;
bt
b
:
bt
lim G t (a ;b);
A 0 (a ;b) =
lim A t (a ;b);
t! 0
t! 0
th en
G 0 (a ;b) = A 0 (a ;b) = L (a ;b):
T h e rea d er ca n v erify th a t
G 1 (a ;b) =
G
t (a ;b)
a b;
G t (a ;b);
a+ b
;
2
A t(a ;b) = A t (a ;b):
A 1 (a ;b) =
589
GENERAL ARTICLE
We have an infinite
family of inequalities
that includes the
AMGM inequality,
and other interesting
inequalities.
(1 6 )
ab
a 3 = 4 b1 = 4 + a 1 =4 b3 = 4
L (a ;b)
2
a 1 = 2 + b1 = 2
2
a+ b
:
2
(1 7 )
T h is is a re n em en t o f th e fu n d a m en ta l in eq u a lity (1 1 ).
T h e seco n d term o n th e rig h t is th e b in o m ia l m ea n
B 1 = 2 (a ;b): T h e seco n d term o n th e left is o n e o f a n o th er
fa m ily o f m ea n s ca lled H ein z m ean s d e n ed a s
H (a ;b) =
a b 1 + a 1 b
;
2
0 1:
(1 8 )
C lea rly
H 0 (a ;b) =
H 1 = 2 (a ;b) =
H 1 (a ;b) =
H 1 (a ;b) =
p
a b;
H (a ;b):
a+ b
;
2
590
1 = 2 (a ;b)
H (a ;b) H 0 (a ;b);
(1 9 )
GENERAL ARTICLE
fo r 0 1 : T h is is a n o th er re n em en t o f th e a rith m etic{
g eo m etric m ea n in eq u a lity.
If w e ch o o se t = 2 n ; fo r a n y n a tu ra l n u m b er n ; th en w e
g et fro m th e rst in eq u a lity in (1 6 )
2 n (a b)2
(n + 1 )
a 2
a b
L (a ;b):
n
b2 n
U sin g th e id en tity
n
n
n + 1
n
n
n+ 1
a b = a 2 b2
a 2 + b2
a2
+ b2
1
1
a 2 + b2
;
w e g et fro m th e in eq u a lity a b ov e
(a b)2
(n + 1 )
Yn a 2 m + b 2 m
L (a ;b):
2
m = 1
(2 0 )
a2
+ b2
2
Yn a 2 m + b 2 m
:
2
m = 1
(2 1 )
If w e let n ! 1 in th e tw o fo rm u la s a b ov e, w e o b ta in a
b ea u tifu l p ro d u ct fo rm u la :
m
m
1
Y
a 2 + b2
L (a ;b) =
:
2
m = 1
(2 2 )
T h is a d d s to o u r list o f fo rm u la s (7 ){ (1 0 ) fo r th e lo g a rith m ic m ea n .
C h o o sin g b = 1 in (2 2 ) w e g et a fter a little m a n ip u la tio n
th e rep resen ta tio n fo r th e lo g a rith m fu n ctio n
lo g x = (x 1 )
1
Y
m = 1
2
;
1 + x 2 m
(2 3 )
fo r a ll x > 0 :
591
GENERAL ARTICLE
W e ca n tu rn th is a rg u m en t a ro u n d . F o r a ll x > 0 w e
h av e
lo g x = lim n x 1 = n 1 :
(2 4 )
n!1
a;
a n + bn
;
2
b0 = b
bn + 1 =
a n bn :
(2 6 )
592
GENERAL ARTICLE
= A
t= 2 ;
t= 2 G t
= G
(2 7 )
t= 2 :
a 02 =
..
.
a 0n + 1
p
a+ b
A1 =
; b 01 = G 1 = a b;
2
p
a 01 + b 01 0
A 1=2 =
; b2 = G 1 = 2 =
A
2
q
a 0n + b 0n
0
; bn + 1 =
a 0n + 1 b 0n :
2
Somewhat
unexpectedly, the
logarithmic mean
can also be
realized as the
outcome of an
iteration closely
related to the
Gauss iteration.
1=2 G 1
a 02 b 01 ;
(2 8 )
k 22
= tr T T =
Xn
i;j = 1
jtij j2 :
1 = 2
A
X
X
B
+
A X B 1 = 2 A t X B 1 t d t
:
2
2
0
2
2
(2 9 )
RESONANCE June 2008
593
GENERAL ARTICLE
2 e H ;e K jjH K jj2 :
(3 0 )
A sh o rt a n d sim p le p ro o f o f th is ca n b e b a sed o n th e rst
o f th e in eq u a lities in (2 9 ). T h e in eq u a lity (3 0 ) ca p tu res
th e im p o rta n t fa ct th a t th e m a n ifo ld Pn h a s n o n p o sitiv e
cu rva tu re. F o r m o re d eta ils see th e S u g g ested R ea d in g .
Suggested Reading
[1]
G Hardy, J E Littlewood and G Plya, Inequalities, Cambridge University Press, Second edition, 1952. (This is a well-known classic. Chapters
II and III are devoted to mean values.)
[2]
[3]
[4]
[5]
594
[6]
R Bhatia and J Holbrook, Noncommutative Geometric Means, Mathematical Intelligencer, Vol.28, pp.3239, 2006. (A quick introduction to
some problems related to matrix means, and to the differential geometric
context in which they can be placed. )
[7]
Tung-Po Lin, The Power Mean and the Logarithmic Mean, American
Mathematical Monthly, Vol.81, pp.879883, 1974.
[8]