You are on page 1of 192

Chapter I.

General Concepts and Chebyehevre Inequallty.

1.1.

BasLc ldeaa.

One of the most lmportant ln partlcular

facta

Ln matheuatlcg that

and ln the theory

of lnequalltles nunbet X ca Thus, for

le the obsenratlon 2 X- > 0, wlth

the equare of a real equallty only lf

never be negative: tno posltlve

X - 0.

nrnbers

a and b, we have

(6 - /6,12 o, ,
a- 26- +b> 0.

TtrLs nay be written

ln the forn a*b

( 1 .1)
,t

2-

{ab

By replaclng

a by a- and b by b-,

we obtaln

an alternatlve

forn,

( 1 . 2)

}ab,

rhlch

elearly

holde

for

poselbly

negatlve

a md b ae rlell.

A releted

lnequallty

ls glven

in Exerclse 1.1:

( 1.3)

R lT'7.
Glven two poslttve nrnbera

-a*b
a and b, the number *+i 16 ls le called called the

the aritlmetlc geoEtrlc GFn Eaa

ren

of a and b, wtrl.le the ntrmber I{trat (1.1) tel1s

of a and b. be greater wlth

ue le that meaq.

the g.oretrr"

caa never In deallng

thaa the arlthretlc it

inequall.tlee,

la alwaye lqortant

to know when, In

or under ntrat condltlone,

an lnequallty

can reduce to an equallty.

-2 -

case of the lnequallty (G

(1.1),

we have the an$rer

from the LnequaLlty (G - /t-)2 = o

- {E)2 > 0 r,hlch we used ro deduce (1.r): a - b. Hence the arlthretlc

we have that

tf and only lf poeltlve

and gemetrlc lf a - b.

means of two

numbers a and b are equal if and only the relatlonehlp

Let us proeeed to explolt and gemetrlc meane.

between the arlthmetic

Example l.

Let e,b,c be real

numbers.

rt

follose

fron

(1.2)

that

,2*b2>2ab ( 1 .4) b2 + c2 > 2bc .2*"222ca, where ve obaenre that equallty i r an a ooly'lfa=b, holdt ln the three Btatementa tn (1.4) i. e . , a -b ' c . A d d ln g t h e e e

b-cand c o E r and dlvldtng

three lnequalltlea ( 1 .5)

by 2 ytelda

"2*b2+c2 2 a b * b e + c a ,
wlth equaltty lf and o1y lf e b o c. (For other proofe, see

Exerclse L.2.)

Example 2.

Let ua rewrlte

(1.4)

in the forn

a2 - .b + b2 >.ab b2-bc + c 2 > b c "2-""+a 2 > e " .


Now obeerve that the three (a + l)(a2 - ab * b2) - .3 * b3. Ttrus, if, rn ntrlttply

lnequalitLes

above by a + b, b + c, and c * e, rre have

'.

"3*b3 > a b (a + b ) b3+c3> b c (b + c ) "3*"3> c a (c f a )r

-3 -

w?rere we mrst now requLre positlve, or else at leaet

that

the terms a * b, b f c, and c * a be mrst be reversed. by 2, we obta{n

one of the inequalltl-es Lnequalltles and dtvlde

rf we now add the lagt

three

( 1.6)

"3
equaHty

* b3 + c3 ) .o g+i
and only lf

* u" L{i*

g{i, ".

rlth

lf

a - b - c.

EXERCI SES Alwaye lndLcate when equallty holds.

1.1.

Prorre (1.3) .

L.2,

(a) Glve another proof forn(2a-b-c)2>0.

of

(1.5)

by considerlng

inequalltlea

of the

(b) Glve yet another proof of generallty) quantltlee that

of

(1.5)

by flrst

aesuming (wlthout (1.1)

loee

e > b > c and then applytng

to the

(a - b) and (b - c).

1.3.

Apply (1.1) to (1.6) + @A)312 for a,brc

to obtain .3 * b3 * 2 0. Can you derlve

"3

= $d3lz

+ (ca)3/2 +

thlg

in another !ray?

L.4.

Prove thet ltultlply

the rtght-hand

slde of

(1.5)

ls at leaet by crarb orr

3abc. in order

(Elnr. and

the trLple

of lnequall.tlee a3 + b3 + , wirh

(1.4) "3

then add.) tr,"t 9-i|-t3

Deduce that = 3/tE;

= 3.b"r

equlvalently, if a - b - c. a,b,c.

equatlty

1f and only

1.5.

Show that

(a + b)(b + c)(e + a) 2 Sabc for poslrlve I *b c * c c +8 + -T-to' -' a

r.6.

Prove that

*b

4-

L. 7.

Pro v e th e

th a t

2.2 2'-) .22 a-b- + b-c- * c-a- > abc(a + b + c). of lnequal-lties ab bc ca+-2a*b*c. cab .3 * b3 r (a t u)3, orr more elegantly, (1.4) by appropriate

(Hint. quantities

Multiply and add.)

tri p l e

1.8.

Show that

1.9.

show that

/Z / " ' + a ' _ ra * b { 2

2'

1. 10.

Let

a rb rc rd

b e positlve

nunbers. c*d-

(a)

Show that

a*b*

tT
a *b *c *d r

ffi

(b)

Use (a) to show that

4V

L.2.

Further Basic Ideas.

We have seen in the prevl.ous sectLon how which cannot be negatlve let bl into a useful in-

we can rnanipulate an expression equallty. To devel,op this

idea in another dlrection, ^L = ^2,

{ar,ar}, = b2.

{Ur,Ur} Then

be two increasing

sequences* of real- numbers, (at ar) (b, - b2) > 0,

with equality we have

if

and only lf

a, = a2 or bl = b2.

Multiplying

this

out,

"1bl Then if obtain

^ZbZ.

^Lbz

^ZbL

we add .1b1 * arb, to both sides, an instance

factor,

and divide

by 4, we

of Chebyshev's inequal-lty, tl-bL * t2b2

(1.7)

2-2 if and only if

,^r*

^,

.bt*bz
2 We sha1l see Ln the ex-

r^rith equality exercises tensively.

a, = a, or b, = bZ.

below and in the next section For the moment, we consider

how (1.7) nay be generalized some of its consequences.

Note is

*We are using the t,erm ttsequencett for a pair {ar, ar}, of length 2, with a view towards later developments.

even though it

-5 -

that

(1.7) holds for decreasing sequences {a'ar},

{br,Ur}

a16o.

Exanple 3. and {anrbt} "t"

suppose that either

arbrmrn are positive

numbers.

Ttren {atrbm}

both increasing

or both decreasing sequences, ac-

cording as a < b or a Z b.

Ilence (1.7) applies:

(1. 8)
Note that equallty

.ffi+btfur"t+b*."t+bt 2-22 holds if and onJ-y lf a = b.

Example 4. function "o"4e

For 0 in the interval + sin40.

0 s e < T/2,

let

us consider the + sin4g as

suppose that we rewrlt"

"o"40

' l- "ot2*20 )

+ srn2+20l

in order to use (1.8) with m = rt = 2.

Then

2{ "o"2*2e

"irr242e

]=,t'-+d]t
sin 0 = cos 0, i.e.,

+ sln2o =- I "o"20 2 2' |

Therefore

(1 .e )
with equality if

4^ 4^ 1 cos'O * s in ' O = . t
and on1-y if 0 = tt/4. in the and we

We wish to mention here an important conditions under which equallty

aspect of our interest inequa].ity,

holds in a specific

can use (1.9) as our first greater

exampl-e. The function

than or equal co L/2 for all 0 = n/4, it assumes its

+ sin40 is always "o"4e 0 in 0 < 0 s n/2 (indeed, for all 0). least value, Lf2; in other words,

At the point the function

assumes lts (1.9).

mLnimun value at any point we shalL see that

0 at which the equal-ity

sign prevaLls in
rii

in a large number of cases,

-6 -

technLquee lnvolv{ng nore convenlent tool

Lnequalltles for

w111 prove to be a more powerful

or

deternlning

maxLna and mLnlma than the caIculus.

EXERCISES

1.11.

By repeated appllcatLon lncreaelng

of (1.7),

prove that

for

three posltlve tcrrcr},

(or decreaslng)

eequences tarraZI,

{Urrtr},

ttbt"t

I 2

tzbz"z

* t, , ", 2

bt * bz *._it . "t - - T . 2
condltion. be renoved? Cen

Pay partlcular reetrictlon

attentlon that

to the equallty

the aequence be posltlve

1.12.

Ttrough we ehall equall,ty

not prove lt

here explicltly,

note that

the ln-

ln the prevloue llke n, (1.8).

problen

nay be generallzed for posltLve

agaln and also arb and a poal.-

put ln a forn tlve lnteger

shon that

I [.o + b n );

t--T-J
1.13. For t ln the lntervel
, I t 2a ' l t' . ( ,t

-a*b -2

0 < t < 1, flnd


)" /ur-r

the nlnLnnn value


,rrr^..r-]a ]La .,-i ..^

of

t ; ; 7 1 + t ; ; ? l ( l t l " s ' csr cu la te th e va r u e o r

[t- .1 ]t * tr r . =l'. , [ [r
*.2J *.2J
1.14. For any pair shotr that of poel.tLve numbere a and b such th"t + 2n-, , where n is a poaltLve e3 + b2 - l, integer.

.2o * b2o >

-7

1 .15 . 1.16 .
1.17.

S honthat 1*tan"0

RlA

>f

s e c " 0 f o r 0 < e < t l|,

e x c e p t a t 0 -1t 1 4 .

shor thar for posttlv e a a n d b , (a + b )(" 2 + u 2 )(" 3 + b 3 ) < 4 (. 5 + u 6 ) .


Uelng the methoda of thls fror the orlgla to the llne section, find. the shortest disiance

x + y - 3 - 0. aectfon, flnd the shorteet - G . if oae of the but that distance

1.18.

Uelng the athods of this frm the orlgtn

to the parabolic

arc G, + 5 in (1.7)

1.19.

shorr that

the lnequaltty

is reversed whlle reoalns

seguences Lg increaeLng the equellty condit{on

the other

La decreaeLng,

unchanged.

1.3.

Chebyshevrs laequel!,ty.

Problen 1.11 lodlceted

oae poeaLble

gmerallzatLon

of (1.7).

lfe shall non glve another.

conalder two lnand {b'b2,b3}.

creaalng (or, alternetlvely,

decreaslng) gequences {arrarrar}

Ttren ae ln the prevloua sectlon, (at (aZ (al ar) (b, - b2) > 0 ar)(b , - b 3 ) > 0 ar) (b, - br) > 0 ,

"lb1 "2b2

ar b, + ar b, ^ZbZ, + .3b3= .2b3+"3b2

.3b3*.1b1 . . 3 b 1 * " 1 b 3 .
If ne now sln theec lnequalltles at and then add .1b1 * * arb,'to

^zb.

both

eldee, we arrlve

-8 -

( 1.10)
wlth equaltty lf

- "l
and only lf

'zt
JJ'

^3,

bl * bz * b3

cl.ses, {a1,

the reader "r}r{br,

a, a3 or bl - b2 - b3. In the exer"2 is asked to generallze thle to two sequencea of n numbers "', bo} :

'. .,

arb, * arb, t ( 1 .11)


nnn

. . . * a o b r,

a, * a, * ... * ao

b l+ b 2 +

and then to more than tlro sequencea. Chebyshev r e lnequsllty.

Ttre lnequallty

(1.11)

le knorrn as

Exanple 5. order.

Glven three

real

numbere, labe1 them arbrc

ln lncreaaing

Then by (1.10),

( 1.12) or 3(e2 +t2 * " 2 ), "2 * b2 * "2 + 2 (b c * c a r a b ).

.2 D

>bc*cafab,

glvlng lf

us another proof of (1.5).

(Note that

equallty

holds lf

and only

a - b = c.) Exanple 5.

Uelng (1.10) end (1.5) glves us 13*b3*o3 - - 3 - .* b 2 z "2


bctca*ab

( 1. 13)

+ c2

a*b*c

3 afb*c

3,("3 + b3 +

"3 )

> (b c + c a i a b )(a + b + c ) ,

-9 -

with

equaHty

if

and only if

a = b = c. notation.
To make another notation

Note on the summation and product more compact, we write is n tlf"t for

the sun a, f a, + . . .

* a_. n

T h u s (1 . U )

nlr"non
k=l

=i*.i,
"k] for the product ... ^L^2 8n. Thus n II k=n! k=1 and

Sinilarly,

we write

n IIa

1. n(1 + E/= I k=1 k=1

- 2L 2 3 .!.t_

,"*!,

= nf1.

EXERCISESFOR CHAPTERI (a) Use the methods of this from the origin (b) Sirnilarly, find

1.20.

chapter to find

the shortest

distance

to the plane x + y + z - 5 = 0. the shortest distance frour the origin to the

su rfa ce G + 6 + E - 6 = 0.
L.zL, Find the mLnl-mum val-ue of the function = "irr4 (. 2t4' x + l-l-- Y--l lL - y,)

f (x,y)

( ^ r4 + I 'Y =l "o"4* \L + yt) " o " u *


of x and y yield the

on 0 < x < r/2, minimum?

0 < y < 1.

Wtrat values

L.22.

Let f(x)

= 1+cos4x+sin4xon0< the inequality f(x)

x<n/2. > 4/3,'and

Showthat explain

theuseof apparenr

(1.12) ylelds discrepancy

rhis

with

the minimum of 3/2,

wt.leh follows

fron

(1.9).

L.23.

Prove (L.11). is increasing

Show that

the inequality

is reversed if

one sequence

while the other is decreasing.

-1 0 -

L.24.

General.lze (1.11) to m aequences of n terma.

(cf . Exerclge 1.11.)

L.25.

shon that then

lf

n le a posttlve

lnteger

and arrdzr..rsn

are poeltl.ve,

F mrl
)and only

1rlr"t l; ,
[" with equaHty lf

-tt% ' n
lf n a, ^2 - 8o. Ttrle generallzes

Exercl.se 1.1?.

L.26.

Shon that lntegers

(*t

+ yt)n

< (*o + yo)t

tf

m ) n, where nrn are poeltlve

aad xry are posltlve

numbers.

L.27 ,

Let x1r... rxn be poeltlve that (1 + xr)

nuoberg auch that x,

xo - 1.

Show

(1 + xo) > 2n. numbera such that x1 .. Shotr that

1.28.

Let xlr...rxn *nr,...r\obe

be posltLve

1, and let

a permutatLon of *1,...rxn.

(*t+.h l )...(xr
L.29. (a)

* *po) > 2n.


posltl.ve xry whose sum x * y is conatant. :ry ls a maximum when x - y. :ry ls congtant. show

coneLder all that

the product

(b)

ConeLder all Show that

posLtLve xry rrtrose product

the sum x + y ls a mLninrn when x - I. posltlve ntrmbere ls constant, Bhorr their

(c)

If

the suil' of three

product

ls a naxtmrn wtren the nunberc are equal .

1.30.

Flnd the rnaxluun value o, *R

for

0 < x s 4.

1.31.

Flnd themlnimumvalue

of

ryz lf

x * 2y + 3z - 6, xryrz

) 0.

-11

L.32.

Arnongall edges is

rectangular

parallelepipeds

such that

the sum of the

the same, show that

the cube has maximum volume.

1.33.

Arnongall area,

rectangul-ar

parallelepipeds

having the same surface

show that

the cube has the maxlmumvolume.

1.34.

Find the maximum val-ue of G. + '6 + G it For n positive (at numbers r1r...

*2 + yz + z2 = L.

1.35.

r8n, show that

+ . . . + at- )f'"1 * ... + - $ fln = ,,2 .


numbers such that a * b = 1. Show that L 2'

1.36.

Let a and be be positive (a+

=)- + (b +

' l?

1,

)\

f,)',

t,

wirh e q u a lit y o c c u rrln g whena=b=


Show that

L.37.

Let ^L,^2,"',a' n 1;2 j=1

x( 6,
J

.1

be p o s it L v e n u mb e rswit h . 1 . 8 . , = 1 . j= l J = 1(r,2 + L )2
n'

"j.

1.38.

For positive

a and b

wl-th a * b = 1, show that

the series

i n=l
1.39.

tf

1
(a

+ (b + 1b)" *1"

converges and that

the sum does not

tl 3.

For a > 1 and b > 1, show that

the series

f n=1 (logrb + log'a)n

converges and that

the sum does not exceed 1.

1.40.

Let aL,a2, -n r I a-r

...

,3

be positive

and letp>0,9<

0.

Show that

nl_

"f*q

=In i.rl [r ] "tj , wlth E tJ |.t


rlt ilr
a. n
.. .,g

equality

if

and onl-y if

'L= ^z =
1.41. Let arta2t ,,

be posLtive

and prq negatlve.

Prove that

n .L -" ! l-=r

t -p+q
-[*

[t il ,i,"r],i,'

-L 2 -

L.42.

Show that for posltLve bc b*c .

arbrc ab ,a+b*c af b2

ca c*a

1.43.

Prove that

for arbrc ) 0

a{t*c

"ab""
wlth equaLity lf

> (abc) 3
lf a = b = c.

and only

I.44.

Show that

for

a,b,c

) 0, we have

-t _ 3 -

Chapter II.

The Arithmetic,

Geonetric,

and Harmonic Means

2.L

Means.

A nean of two or more numbers is a function

of those of

nurnbers whose value always lies the nunbers, that is,

between the lowest and the highest In this book, we shall

between the extrenes. geonetric,

define and consider the arithnetic, power, and two symnetric The arithmetic tistics, sinply neans.

harmonic, quadratic,

mean is commonly called It

the average, or,

in sta-

the mean.

is the sun of the given numbers divided mean of the n numbers

by the number of them.


&,t Id n ,3 - , t"' , J' d n 1s

Thus, the arithmetic

^L*'2+"'+an

,n = lI
n

.-. l-= I

a.

l-

In an arithnetic terms innediately

progression,

each term is the arithnetic following,

mean of the

preceding and inmediately

since three

consecutive terms can be written of the first and last being (a--d)

in the form a - d, d, a + d, the nean

+la+d)

2
The geonetric &1, u2, "', 4r, it mean is s in ila r. is

-4.

Of n positive

numbers,

= 'r',' ('1,
of positive terms, each term is the geonetric

In a geometric progression

nean of the two adjacent terms, since

-L 4

ff i = ^ .
The harnonic mean is defined as the reciprocal of the arithnetic nea! of neah of the reciprocals of the nunbers. Thus, the harnonig

" r' ^z'

"'. -na

is

-l-* r
"l ^z

+ ...

I a n

-+ -

11 ^2

"l

* ... * -la
n

i-l

l1 ai

A .hamqrLc" progression is a Drogression such that the reciprocals of its terns form an arithrnetic progression. Thus, each tern in such

a progression ig the harnronic nean of the adjacent two. we should justify the use of the tenn rneanr', for a mean lies l{e shall seldon consider negative nunbers,

between the extreme values. so let a1, a.,


L z

, a- be n positive nunbers such that a, - - the snallis ---- -l n Then

est (it

is not necessarily unique) and a' the largest.


a,(3,<a
I_

I_

a,(a^(a IZ-

<a a, <a rnn

Adding and dividing by.n, we obtain


*
a1
I-

+ ... ^2

^l

+ an :8n,

-lt

which is what is desired. root, we have


L L t -.

If we nrultiply,

instead, and take the n-th

^t1/^ ta2 " ' a n Furthermore, if we take reciprocals,


I -an

< an

we have

!t!t al -41 l>1>l al-^2-rn

!t! r !

e. -a rn n

-a

,l.dding, we get

n ,-!*l-*... "t-"1 ^2 or

+ 1 > tt % -" o

"rj

ll -+-+

4r

.oo

t
n

(4

^2

Thus, these are indeed means. All three neans appear in elenentary Euclidean geonetry. exanpLe, if AABC has a right foot of the perpendicular For

angle at c (see Figure z.L), if H is the M is the nrid-point of

frorn c to AF', and if

AT-, then cH is the geometric nean of AH and HB, and cM is the arithnetic mean of All and HB.

- t6-

cH= ffiffi,

ttl= ry

Also, given a trapezoid ABCD(see Figute 2.2), with IIB-parallel to 6-, if t{ and N are the mid-points of lT'and ET, and if such that FF is parallel X and Y are

points on Ab-and F

to FB'through the internean of

section of the diagonals, Id and ffi-, then MN is the arithmetic AB and CD, while XY is their harnonic mean.

1 9 + CD MN= _-z

2 -= XY

- +-

11
CD

AB

Figure 2.2.

2.2.

Conparison of the Arithrnetic

and GeOmetric Means. between then.

Given

these means, we may ask if FigUre 2.lsuggeststhat, mean is at least tive

there is any relation for

at least

two nunbers, the arithnetic words, for any two posi-

the geometric nean; in other

numbels, a and b,

( 2. 1)

+>6 ,

-L 7

with equality as inequality posi tive

if

and only if

a = b.

Indeed, this

was shown in Chapter for any three

(1.1) .

In Problem I.24,

we saw'also that

nunbers, a, b, and c,

:
with equality true if and onlyi fa=b=c. for n positive
+ o ..

3,/ffi

,
ln fact, this relation is

in general

numbers:

uI

* ,2

* " r,

( 2 .2)

, v6-T

,
= rrr. This is known as the This inequality may

with equality

if

and only if

a, = d2 =

theorem of the arithmetic

and geonetric

means.

be proved in nany ways, and we present first To prove (2.2), assertion: If the product of n positive we proceed by induction

one of the simpLest proofs. to prove the following

nunbers is equal

to 1, then the sun cannot be less than n. Fron what we have done in chapter true for n = 2 since, if ara, = l, 1, we know that or ar= 1/a' the assertion we have ar+I/a, is true fot L, 2, is

2. ,k;

Suppose next that

we assurne that if "k*1

the assertion = l,

we wish to prove that, * ^r = ^I ^2 ^2 + "' * "k *

arar...akak+l

then only when

t k + 1, with equality

We may assune that at least two of the numbers-"k*1. --we call them here a, .rd the property that a, < I and "k*l--have t 1, for if all the a. are less than 1, for example, their "k*1 product could not be 1:

= ' ' ' =

- .18

(a,an*1) az"'

"k

= l.

If

we set z = 8,?,_.,,
I K+ I
1^

w have that

the product of the k nunbers hypothesis, their

zr a.1r ,..,

&v is 1, so that,

by the induction

cannot be less than k:


, * + ... ^2 * ak > k.

However, we observe that

^ L*

rZ + ... * ak *

"k*I

z>

>k+1*"k*l*"1 . = k + 1 * (*k*t - t )(f and this

z -L

=k+1

* "k*l

* tl

"l"k*l

- I

- a r), - 1 ) (l - a r ) > 0 .

last expression is greater than k + I s in c e (. k * t

This shows that

and the assertion is completely proved. is actually equal to k + 1 o n ly if

We renark that the ,.m

nit i= 1
a.
1

= ... = ^L= ^Z "k*1. As an innrediate corollary of the assertion above, we have that
,L*^2+...*"n

.l l r--

>

(a-a^-..a
LZ

n with equality only if we have the identity = ,Z = ... =

^L

" rr.

F o r, if we s e t g = n E G ,

-r9 -

^r.
g

a.a^ a 4* 4 g I

+ ...

t t g -" ' n .

+a ^I*12+... .n with equality onlyifa,=o^= - Itrl only if a', al j = -- = gg =4.

18,

a = n , or, what is the same thing, ;

Let us give another proof of (2.2), which we feel gives a different insight into (2.2). By (2.1), if the sun of two numbersis given, their This nay suggest that, nore

product is greatest when they are equal. generally, Lema. P roof. their

product is greater the closer they are.

That is,

Let x < y and 0 < a < y - x. W ehavex+a< y a n d a > 0 .

Then xy < (x * a)(y - a). T h u s a (x + a ) <ay, o r, if

we add xY - a(x + a) to both sides, xy < xy * ay - a (x + a ) = (x + a )(y - a ). l{e notice that, by the definition suchthat in (2.2), if not all the numbersare equal, then soy

of nean, there exist two of them, a, and d1

whe re A = * i= , T t ru s , if we t e t a l = A , "1.A . ^ 2, " i. = ,L * a2 - A, we have * ,2 = * and, by the Lernnaabove, ^) "I ^l ^) , ala), ot ^L^2

-2 0

"ary^*,
Repeating this nunbers all

, ",Eg;.::"
we arrive at a set of n

process at most n - 2 tines,

equal to A, whose geonetric nean of the original

nean, nanely A, is greater nunbers, nanely \/al"r*"".

than the geonetric That is to say,

"'arn^,
unless = ^l ^2 - "'-an,

< A

^l*u2+"'+a
nt

in which case we have equality.

Thus, (2.2)

is established. Note that given, their (2.2) implies that. when the producL of n numbers is equal., or, their what is the is a

sun is a minimum when they are all when the sum of n nunbers is given, equal. t)"

sane thing,

product

maximun when they are all

Exanple I.

Prove that n! < (" i

for n > 1. and

This is a sinple geometric neans, for

consequence of the theorem of arithmetic

Note that n > 1.

equality

cannot hold because the n nunbers are not equal for

Exanple 2.

A sinple

consequence of

(2.1)

is that

(2.3)

u **|>2,/ffi, x-

_2I

with equality to ninimize

if

and

O nI Yl. aX =- / - . .\l t

ID

t-f-

Thus, suppose we are required

(a + x)(b + x) f(x) = c+x

'

where ?, b, c are constants.

To use (2.3),

let

us substitute

Y=c+x.

Then
(a - c + y)(b - c + y) f(x) =

c)(b : c ) +y+(a+b-2c) - (a :

(x) with equallty if and only Lf y = lrG-llTG:-T. That is, f (x) has - ffi)z.

its minirnun 4t X = yC)-6

- cI - c, where f = Gffi

If this value of x is complex, then, since equality in (x) can never occur, f has no rnininum in the real donain.
Example 5. Heronrs fornula Given a triangle with sides of lengtls K, as ar b and c,

gives the area of the triangle,

K = f fi,
wtreres, the serni-perineter, equals *F Let us use this and

(2.2) to find the triangle of given perimeter which has the mst area.
l{e have

- 22'

K 2 = s(s - a )(s - b )(s - c )


(g

.s-4+s-b+s
rT

: c,) 3 = s (----5--J s -3 -5 s - 2

,4 =T' with equality if and only if s - a = s - b - s - c, or a = b = c. has the nost area, namely

Thus, the equilat'eral triangle IT

* ={ b=f ^' .
The inequality just derived also showsthat the equilateral triangles triangle has the least perimeter anong all E xanple 4. (a+b*")2= of a given area.

Let us ex p a n d (a + b * c )Z a n d u s e (2 . 1 ):

(^ 2 *b 2*.21

* (2ab+26s+Zca) + [ ("2* b2) , (b2 * .2) * ("2 * ,2)l

(r2 rb 2*"21

3 (a 2 * bz * .21 ,
t-

( 2 .4)

T_.^/
and only if

a+b+c-

/^'*b2*.2

,
a = b = c.

with equality if E xanple 5.

Maximizy = * 2 7 t - Z * 1 .

If we consider y as a product of two factors., x2 arrd (1 - 2x), their sun is not constant, lo that (2.2) cannot be used. sider y as the product of three factors, sum, x + x + I - 2x = l,
'I

If,

however, we con-

x,x and (l - 2x), then their be a maxinnrn,

is constant.

Hence, y will

na m ely (l/3)",

whenx=x=

| - 2 x , o rx = L / 3 .

T h is a p p lie s o n ly t o

-. 2 3 -

x ) 0, since (2.2) is applicable only to positive n u mb e rs . A t x = l/ 3 , therefore, ily we have a relative, or local, rnaximum, taking on arbitrary

large values on x <0. Exanple 6.

Maxinize )r = x {G.

The maxirnrn of y occurs at the naxirmrn of yZ = *2( r - *21, r hat i s , w he n*2=L-*2


1

,orx=*rsince {z

'=t. ,':ft'-'-,.']
-Thus, the naxinun is at (*,

b
Exercises

r,.)

2.I.

Given n positive numbers ,Z ^1,


^'*^' a t, n. aI -

, tn.

prove that

+ ...

^z ^s
2.2.

P rove the generaliza t io n o f (2 . 4 ), n a n e ly

(2.s )
with equality right

*r!, ",tffir4
= = The terrn on the ^Z "rr. is known as the quadratic nean, or root mean square, of the if and only if a, =

n numbers. 2.3. If a, b, c, d are the sides of a quadrilateral,

an&e is the sun

of two opposite angles, Bretschneiderts fornrula gives the area, K, as K2 = (s - a) (s - b) (s - c) (s - d) - abcd "or2 f, where s is the

- 24-

6.\
2 .4.

olmlperlneter. teral

Aseunlng thts

foruula,

prove that

the quadrlla-

of glven perlneter

wlth maxlq'tn area ls the aquare.

Flnd the dlnenslons volume lngerlbed and radiug

of the right

clreular cme

cyllnder havlng

of nsxlnun altltude II

ln a rlght

elrcular

of baae R;

2.5.

(a)

Find the right inscribed

circular

cone of largest

volume *hich cEn be

in a sphere of radius a , circuler cylinder of largest volune rhlch

(b)

Find'the right

can be inscribed 2 .6.

in a sphere of radius a.
< a, whefe

lfhere does the rnaxinnrnoccur of (a + x )3 (a - x )4 , a is a constant?

2 . 7.

Prove that

5
(Hint:

r_sry
Cube, expand , a n d n s e (2 . 2 ). )

s/Tm

..,j}

:i:-;

-2 5 -

2 .8.

Prove that

. 2"2 . 4%]ro :-3"s I ^f1^1^1 1-r


(o

2 .9.

(a) (b)

l,trxinize l,bxinize

(1 - x) s( r + x) ( t + (x.

z*) 2 .

s121sx- 7) ( r 1 -x )g (x + 1 ), x )0 .

2.L0.

Prove tfuat
a+b+c --..3-. \ a + b)(b + c)(c + a

2 nr.unbers , *2 , *1 ., on bI n

.sffi
the n

2 . 11.

Given n positive

, Xn, and define

quantities 61, 62;

ol=*l**2

**n

L
1 =I

xi,

62=

*r*2 *r*s *

xn- r n = x.x.. .x r. li<j

O=

x-x^ ..'x
LZ

+ '

[ i l < i 2:....i n

* ., x! 'L '2

...x. l n-

o = x-x^ ...x n Lz n

n lx .
i=l

(Tte quantiti es Orr ' .', functions on

0n are call.ed the elenentary Show that

slmetrie

* r'

*n')

1+

n n . I o1 =.'I I (1 ilr i= l

+ * i)

1r + ?[)n

by showing that

- 26'

o, I tll o|/" ,

m=l,2,"',D.

(Notice that this gives an alternative .n 2.12. when o_ = l.) (a) Prove that, their (b) 6ive if

solution of Probl.emL.22

/ the product of two positive nunbers is given,

sun is less, the closer they are. another proof of (2.2), sinilar text, to the first proof

given in.the

but keeping the product of the n nunbers

constant while valYing the sun.


2.'I3. Prove (2.2) by the third by backward induction. method of solution of Problem 2.2, i.e.,

2.L4.

(a)

When the perlmeter

of a rectangle

is glven,

show that

the

aquare has the maxlmum area.

(b)

Glven the maxlmum of people table, one slde of which ls

to slt against

around a rectangular a wall, area. determlne

the shape. which has the most table

2.L5.

(a)

Given the volume of a box. ls a minimun when it is

Prove that

lts

surface

area

a cube.

(b)

Given the surf ace area of a box wlthout Lts volune a maxLmum?

a top.

I'lhen ls

. :7. '.:i.

- 27-

. \--l

2.L6.

Construct

wlth

96 sguare inches of slding open at the top, lt wlth

a reetangular

box of

xlmum volume, ' Lnslde

trvo vertlcal snaller

partltlong boxes, each

the box whlch dlvlde

lnto

three

open at the top.

2.L7.

(a)

Show that n integers

the arithmetlc Ls greater

mean of the aguares of the first n ) 1.

, than (n + L)-l4r

(b)

Show that lntegers

the arithmetlc ls greater

mean of the cubes of the flrst


?

than (n * L )' 1 8 , n > 1 .

2.18.

Show that the k-th


l".

for

any positive

integer

k,

the arithnetlc ls greater

meaa of than

porrers of the fLrst


t

n lntegers

(n+1)*/2*,n>1.

z.Lg.

Let the perlmeter maxlmum area.

of a sector

of a circle

be gl-ven.

Find lts

2.2O. P rove that


t

(x+ y)- + (y

- 28

2 .2L_. Let a triangle hav e s id e s a , b , c .

L e t s b e t h e s e mip e rime t e r , through its vertices, Given that K = rs and and

K the area, R the radius of the circle r the radius of the inscribed circle,
^ R= abC

ff,

prove ^L_! 2r , h that 4_^ < R.

)-__ Whendoes equality hold? . rrtu^_

2.22.

P rove that, for an y real nunbers

^L'

.. , gn,

. "3, (1
Give conditions for equality.

,-"+.
the

2.23.

Let a andb

be positlve

constants. 1b ;;

> 0, find .For any e

mlnimum value of eax *

2.3.

Mean. Figure 2.2 sugges$that the hannonic mean The ffrarrnonic nean. This is true, in fact,

is less than or equal to the arithmetic and, noreover, it

is not greater than the geonetric rnean. This is quite follows directly l{e have
nr fr .L .

easy to Prove, for it and geonetric means.

from the theoren of arithnetic

1 = I1 \

;:

[,], , +J
1-ln
n

or ( 2 . 6)
n $1 < (n )1 / n III a.l [ i= 1 t )

I a.1 1

t\

{
c

-2 9

wh erea.>0for
1

all

i and equality holds if

and only if

a, = 8^ = ...
LZn

= a .

Exanple 7.

(2.6) inp L ie s t h a t

ii ["']#.
Let us expand the left side, getting

2 n.

firis can al.so be proved without (2.6) in a manner sinilar

to Exemple 4.

n I
1=l

1 " .1 a .1

i<j i, j=lr "'ril

(4. t1 l-+ la. 'l

where the index on the second sumation over all i,j frorn I to n such that i < j.

sign neans that we are to sun (For erample, if


il=3

i, j=|,2 13

[+.+] ,1, = e.;le.;l) e.}). .


ai a'

, then

o f c o u rse ,a .',|=r
= ,j.

and

.
J1

> 2, with equaLity if after

and only if terns

^i

Theqecondsurmration has Cll =

n(n-- r)

(there are n choices for i_and n - I for j, because of the requirenent that i < j, ( i ,j) ) . there

i has been picked, but

- l) choices for "r" |rrtn

Thus, the expansi o n is a t le a s t n(n-- 1 ) n +

2=n2,

and can only be n2 rhen all

the a. rs are equal.

- 30

Example8.

Mininize y = i . 11fu;, rm= *-=*.

ll-1

0 ( * t

lte have a n d o n ly

111-32 y = ** * " I -ffi if zx=l-4x,or

9 ' wit h e q u a lit y if

Exercises If x+Y +z =l,P rov e t h a t


r 1 l-

2.24 .

( 1 * * ) ( 1 *7 1 t+ f,) > 6 4 . z.zs. Mininizey=l=.*-# o n lxl ' t.


rnmbcrs, given their

2.26,

Prov6 that the hsrnonic neen of two positive sum, is greater th closer theY are.

5121 . 2.27. Minirnizey = * fr

1ft;

on -i t * t t'

2.28.

Find th rlght abqrt a rlght baec r.

cLrcular clrcular

cqre of nlnlnun volrnc cyllnder

clrctcrcrlbcd of

of helght h aad radft,r

t.

-3 1

2,4. tion

Undeternined Coefficients.

When we are to maximize a funcstands, whence of

such as y - x(l

- 2x), we cannot apply (2.2) as the function observation that y = +t(2x) (1 - 2x)], The point

but we must make the sinple we can apply (2.2) introducing

to the quantity

inside

ttte Uraltets. x is that

the coefficient

of 2 to the factor

the factors can-

now sum to a constant independent of x. not be found at sight, however.

Sonetines the coefficient

Exanple 9.

A Nornan window is a window consisting Find

of a rectangle

surmounted bY a senicircle. the relative

dimensions of the Perimeter amount of

Nornan window of fixed adrnitting I ight . If the greatest

h is the altitude

of the

rectangle senicircle, window is

and r the radius of the then the area of the Figure 2.3.

(x)

A = 2rh * t

TT

T.

1lrrc

+ 4hJ,

which is proportional to the anount of light (xx)


where C is a constant. by some positive (x) is constant. have
ly + ' l t' t + 2h = C ,

adnitted.

The perimeter

is

We desire

to multiply so that

the first

factor

in

(x) in We

constant coefficient Assurnethat

the sun of the factors exists and call


it o.

the coefficient

- 32

A =*

[ (ar) (nr + 4h) ]

and, fron

(xx), 4t+2nr+4h=2C, that is,

ct nust satisfY the condition

we desire,

s1' + tTr + 4h = 4r + 2nl + 4h(= 291'

Thisgives0,=4+nand

where A attains or r =h.

this naxinun value if

and only if

(4 + n)r = Tr + 4hr

This is the sirnpl.est application coefficients,

of the nethod of undeterd$ed

a nethod which is useful in nany areas of nathernatics' more factors, however, there afe nore consider-

when there arethreeor a ti ons.

y = *(a2 - *2) , Example 1.0. One way to naximize the function (2'2): where a is constant, is to square and use

y2 = *z (r2 - *2)2 = le*21le2 - *2) (^2 - *2) I haz'f


)? ? .a = with equality at Zxo = a' - x- ' or x n'

il

However, we need not square y. termined coefficients. We have

We can use the nethod of unde.

.)/ .

_33-

(r)

+ x], / = x(a - x)(a + x) = i= [ a x ] [ B (a - x )] [ a l{e desire that the strmof the facof x in

where o and B are to be deterrnined.

tors in (r) be constant, or, looking at just the coefficient the sun, (xx) o - I + I = 0.

Wemrusthave, also, that we can Solve for x, since we have three fsctors which nust be equal for somevalue of x, if these si^m.rltaneous equations have a solution and B, namely, that the solution we are to apply (2.2). That on a

inrposes anothef coadition

for x found by equating any pair always solutions to the equations

be ttre same. Thus, we mrst have identical , ox = o t x r B (a-x) = a + x the solution of the first being
a * =m

(xxr) and of the second

x = f f8-1= 6 - la z i?
where we have substituted

t
That is to say'

for B from (xx).

#= o f u a
o2 - z o - z = o
OT

o= 1 + 6 ,

-3 4

since cr must be positive. Using (xxx), this gives

The coefficient

B = o+1

is also positive.

a x= -l -, r/3

in agleement with the first Note that cient for in (x), it

solution. would have been useless to put in one coeffithe sum of the factors

each factor.

Our two requirements--that there be a corunon solution Thus, if then all

be constant and that the ratio efficients,

for x--depend only on

of the coefficients. we could divide

we had three undeterrnined coone, s&L leaving two

by the last

undeteflnined and one equal to unity. The reason why we must solve two equations in two variables, reduces to solving in y. a quadratic, is that there are three different of the function which factors

In the calculus,

we set the derivative

to be

naxinized cubic for

to zero and solve the resulting y, and hence the calculus

equation.

Here, we have a to sol-ve.

glves us a quadratLc

Exercises

2.2g.

Maximize y = xD(^2 - *2)n, tive integers.

where a is constant and m,n are posi-

2.30.

Maxiniz" *1^2 - *2)(2a

- x),

where a is constant

2.31.

(a)

In a given segnent of a circle, of greatest and its area.

inscribe

the rectangle between a chord

(A segment is the portion

corresponding minor arc.)

',,

t'i

- 35-

(b)

In a given segnent of an elliPse' perpendicular to a PrinciPal of greatest area.

whose bounding chord is

axis, inscribe the rectangle

2.32.

(a)

In a given segnent of a qphere, inscribe the box of greatest volune.

(b)

In a given segrient of an ellipsoid, is perpendicular to a principal greatest volune.

whose bounding plane

axis, inscribe the box of

=C) 2 .33 . 'Given the paraboloid , * * 2 * y z

0, lnscribe,

above the

x-Y Plane, (a) the cYlinder and (b) the box of naxinun surface area. 2.34. Inscribe the rectangle of greatest area under each of the curves Y =I/*,P =1,2,3, to the right of x = a > 0. (That is'

i ts ve rti ce s w i l l b e at ( a,0) , ( x,0) , ( *,#) ,

( ",#) ' )

-3 6 -

Exercises for ChaPter II

2 .35 .

Minimize the surface area of a right volume aird the shaPe of its

Prism, given its

(polYgonal') base.

2.36.

Given the area and one angle of a triangle, (a) (b) (c) the sum of the two including the opposite side; the perineter.

minimize

sides;

2 .57 .

+ 1) on x ) 0, Maxinize z = (x + 2y ) (f f - 3x - y )(2 x - Y o < y < 5.

2 .38 .
\--l

Find all positive rel a t iv e naxima of z = (x + 2y - 10)(2x + y - 26) (x + 3y -2 3 )o n x , y

0.

2.39.

Prove that the arithmetic

mean of the arithrnetic

and harmonic

tneans of two nunbers is not less than the geonetric mean of those nunbers. 2.40. Prove that for any positive nunbers, a, 'b, c, such that a + b + c = l,

( ; - 1 ) ( i - tl( i- t) > s.
2.4I. Prove that a3 * b3 * 93 + lSabc < 2(a + b + c ) (^ 2 * b 2 , c2a)(aa2 * bc2 *

111

" 2 ).

2.42.

Prove that

(a2b * b2c ,

t g^282.2. ""21

- 37

2.43.

Find the distance 2x-3Y+62-13=0.

from (4,-2,I)

to the Plane

2 .44 .

P rove that if then n

l*,=tt,' t '

) A , )r, q B , i
- l .-

= lr

"'r

llr

x/r-3
1
inside it.

(B - A ).

2 .45.

Given an angle and a point point to cut off

Pass a line

through the

the triangle

of mininun area.

-3 8

Chapter III.

The Quadratic Function and the Cauchy-Schwarz Inequal ity.

3.1.

Introduction. function Ourfirst

In this

chapter we shall

investigate

the use equation

of the quadratic ') ax-+bx+c=0. vertex

Y =^*2+bx+c observation

and the quadratic is that finding the

of the parabola y=ax2+bx+c

( 3 . 1)
is equivalent to finding

the naxinum or the mininum of the function have a nininun or a maxinun according as

(5.1); a is

as we sha11 see, we shall positive or negative.

Exanple 1.

Exanine the function

y=3x 2
for a possible

-1 2 v + 1 7
As in the case of analytic the coordinates of the vertex geometry, of the

naximun or minimun.

where we complete squares to find parabola, we have

v
.or The vertex coefficient upward. that point

=S(x2

-4x+4)

+17-L2,

=g(x-2)2*5.

of the parabola is thus at the point of the *2-t"tr

(2,5),

aD{, because the

i, positive,

namely 3, the parabola opens on the parabola, ninimun value 5 it follows

Since the vertex is the lowest point

the function x = 2.

s*2 - L2x + 17 assumes its

at the

-3 9

Let us look at the solution different point

of the problem in Exanple 1 from a

of view, which can then be extended to a wider range By completing squares, we have transformed the which is the sum
a

of applicability. function

) ^-) 3x' - l2x + 17 to the form 5 + 3(x - 2)-, 5, and a non-negative variable 5 + 3(x - 42 will

of a constant' the function

term 3(x - 2)';

hence,

be a minirnum whenever we add to 5 the case' is 0, when x = 2. 5 - 3 (x - 2)2 will be a naxi-

least possible

amount, which, in this

By the sane reasoning, rnumif x = 2. This point we subtract

the function

from 5 the least

possible

anount, again 0, when

of view al1ows us to formulate

the following

principle,

which no longer depends on the properties

of a parabola:

Let A be a constant and let function. Then the function

F(x) be a non-negative A + F(x) will be a ninimum A F (x)

whenever F(x) is a minimun, and the function will be a maximum whenever F (x) 15 a ml nl num.

Exanple 2.

Determine whether the function If

(x + 3)(x + 5) x+6

has the

any maxima or minina. function

we set x + 6 = z, then x=z-6rand o,

- 3')-(z - l), becones Q

z2-4"*3

+:--4 a

= z- 2/7

*!- q
z

+ z {5

b E - ?) ' . 28
This last

- 4.

expression is of the forn A + F(z), where A=28-4and The expression for F(z) will assume its

F(z) is non-negative.

t"l

-4 0

rnininun when /l the original x = B - 6. E xalple 3. x> 0.

- /7//Z

= 0 or when z - ,8, and the minimun value of be 2/3 - 4, which is achieved when

expression will

In E xercis e L . 3 2 , we t re a t e d t h e f u n c t io n x + l/ x f o r = (/ it / / -x )2 * 2 , l/ G )2 t h e n , b y t h e p rin c ip le , =0, o rwh e n x = 1. the

If wewrite*.f,

m i nimunvalue A =Z

occurswh e n (t -

we pose next a geonetrical problem which can be solved by ninirnizing a quadratic function, and which has many applications. ExanPle 4.
vertices that A, B, C.

Given a triangle with


Find the Point P such

the sum of the squares of the disis a nininum'

tances frorn P to A, B, C Let the coordinates

of A, B, C be

( x r , r r),
tively,

(*2 ,Y 2 ), (xa ,rr)' r esPecand let the coordinates of x and y so that the expresslon

P be (x,y).
--J ----3

Then our problem is to find


---3

S=PA'+PB-+PC-=

(3.2)

(x - xr)2 * (v - vr)2 * 1x - xr)2 + ( y - y2) 2 * ( x - xr ) 2 * (v - v ) 2


Now S may be written as the sum of two quadratic expres-

is a ninirmrn.

sions in x and Y:

( 3.3)

= A rx2 * 81* * C1

A ry 2 * B ry + C, ,

where At Bt A 2= 3 _2(xL **2 2 22 *1 ** 2*x s

xr)

'

Bz

_z(y *Y2* t
22 2 11 * Y2*Y3

y3)

'

ct

cz

-4 L -

Since A1 = A2 = 3, and since x and y are independent, S will its mininun when the two quadratic

achieve a

expressions are simultaneously only the use

mininun. of A'

Thus, by cornpleting squares, which requires Br, we have

B, and A'

$=

T I
lx i-.
T I 22 lv

2. 3(*r **z*
.
5

-s
-2t

l* t *

,2 x2 + x 3 )

+3

t_

( v r* v 2 +v 3 ),.!# l *sl2

..,

I
J

(t '

* Yz * Ys)2

t.
Hence S will whenever

* r* * 2*

_l

r[ ".':.'7
(* 1 * * 2 * * 3 )2
1 (Y t * Y 2 + Y 3 ) 3
?

+C r+C r-

I
J

achieve its

rnininun value,

cr*c2-|,*,

**z

*r ) ' - *,r ,

*y2 *vr ) z> 0 ,

*1 **2 **s
(3 .4)

Y = --T-is called

yl+y2+y3

The point

(x,y)

given in

(3.4)

the centroid the solution

or the center of (3.4) is obvious

nass of the triangle from a physical point

ABC. We remark that

of view, for the expression of a system of unit

(3.2)

is nothing more A, B, C

than the monent of inertia about a line

masses at the points

L which is perpendicular is nininized

to the plane of A, B, C at the Since it is with

Point P, and this possible

when P is the centroid. feeling

to have an intuitive

for nany of the inequalities

-4 2 -

which we deal,

it

is worthwhile

to define precisely

the physical

con-

cepts associated with sone of the inequalities. Thr Moments oL a Point-Mass .System. Let Pl, of n points in the (x,y)-plane, there having coordinates P2, "', (x'Yr), (i P' be a set (x2,Y2),
= I r 2 r " 'r i l )

(xn,yrr), and suppose that located at the point P..

is a non-negative nass n' in the

I,et L be a line from P. to L.

(x,y)-plane,

and let

d. be the (signed) distance

The sum

( 3 .s)

*1d1 *

^Zd2

* trd'

is defined to be the total Tespect to L.

first

moment of the point-nass for a positive integer

systen with

More generally,

k, the sum

(3. 6)

rrul * nra\*

*rdk nn

he to alk k- th nonen I .o f the systen with resPect to L. otal is defined to be th ek= For the ca se on (3 6) is called 2, he expr S S io eex re ,t the noment of inertia

pect t en wi of the sys ten witth re sp ctt to L ar I di ke wo erpe nd Let u st take tw pe pend cu1a li ines in the plane, al first c nsi ert he the y-axis ,aand con de the t o ta fi wo ines respect to these tw 1i ES : the x-axis and

nonents of the systen with

( 3.7)

{ Mt 1I M^ lt.z

tr*1 + t1r1 +

^2*2 ^2Y2

...

nx nn
n' n

+. . .

+ ny

We now ask the question:

( 3 .8)

Is there a point P(x,t)

such that,

if

the entire first

nass of

the systen were located at P, the total and M. are the sane?
L-

nonents Mt

.:

_ 43_

There clearly system is m, ,

exists ^Z *

such a point, + m and n-

for,

since the total

mass of the

(nt *^ 2 (mt *^ 2*
by (3. 7) ,

=Mr
=Mzt

so that,

*1*1 *
t1

^ 2*2
* t2

+mx nn
* tn

+ ...

i=1 n
1=I

m.x.
11

I m .1

( 5 .s)
n

tlrl

*
tl * t2 *

^zYz

+ ny +n n

n' n

i= 1
n

m. y.

l _- 1

It.
1=I

The point gravity

P with coordinates

given by (3.9) If

is called

the center of

or center of mass of the systen. P(x,y), it

L ls anY f-ine passing to show (see Problem 5.8)

through the point that line the total L is zero.

is not difficult

first

moment (3.5) of the systern with respect to the

Exanple 5.

Given a point-mass system consisting Prr i = I,2, Anong all "',

of n nasses rn. but

located at the points fixed line line

n, and given an arbitrary lines para1le1 to L find that

L in the plane.

with respect to which the total a mininurn. way:

second moment--that is,

the moment

of inertia--is

We proceed in the following center of gravity

We letl.be

the line

through the of

of the system and express the mornentof inertia line f para11e1 to the given line about Lo. Let L

the system about any other

in terms of the moment of inertia,

- 44-

o be the (signed) distance between t of the systen about Lo is ( 3 . 10)

and l-;

if

the nonent of inertia

I, = nrdf ,
o

+ ... * t,di , ^raz,


{ it

then the nnment of inertia

of the system about

(3.rr)

L=

tr(d t * s) ' 2 * nr ( d, * o) 2 + "'

+ nr ( dr , * o) z

If we expand the squares 1n (3 . 1 1 ), we h a v e ^n Ig = fo+ 2c L Drd, , gt I rn, , ]. 11 1 =I i= 1 where the first sumnation is the total first tnonent with respect to Since n

L,'and the second surnation is the total o Lo passes through the center of gravity, relat ion ( 3.12) Thus the line parallel of the systen will inertia. I/ = I l * o ' 2M . o

mass M of the system.

I n.d. = 0, and we have the

to L and passing through the center of gravity second nonent or nonent of

mininize the total

Exercises 3 .1. 3 .2. Find the ninimm vaLue of Find the naximrn value of
(5x + 2)(x + l) 2x+1 7 x-+4x+5

fo r xr - lz

3 .3.

Find the naxinun value of

x+5 xo+4x+5 -.

-4 5

3 .4 .

Find the rnaxinun value of

,3 5*.os4**sirr4*

3.5.

For positive

x and y, find if

the maxinun value of x and y are constrained by x + y = 1.

+y

+ -11 r +-2

*- Z

xy
3 .6.
For positive x and y, find 18 the maximun value of

Y/ L
3 .7 .
If

,/x+yl
flnd the snallest using the

ArBrC are the angles of an acute triangle, tarr2A + tan2B + tan2c. prove first

value of the expression fact that A + B + C ='r,

(Hint:

the identity

tan A + tan B * tan C = tan A tan B tan C.)

3. 8.

Prove the assertion P(x,t) total

following

Equations (3.9),

namely, that

if

is the center of gravity first

of a point-mass systen, then the d passing through

noment of the systen about any line

P is zero.

3 . 9.

Given a tetrahedron such that

ABCDin S-dimensional space.

Find the point A, B, C,

the sun of the squares fron P to the vertices

D is a ninimum.

3.10.

Given a 3-dimensional point-mass system consisting located at points Pr(x, ,yi,zi), i = 1, 2, ..., the total n. first

of n nasses m. Given a plane noment of the

II: ax + by + cz + d = 0, we define

system with respect to II to be the sum d1*1 *d2^2* *drr*rr,

-4 6

where d. is the signed distance will be taken positive The question if

fron p. to II (for

example, d. p. lies

P. lies (3.8)

above II and negative if fornulated

below II).

is easily

in three planes, we

dimensions, and, with respect to the three coordinate are 1ed to the coordinates n of the center of gravity n n
y.m.
-11

of the system

I
x-ilr

*.t. 11
M

t-

i=1

i=1

z . m.
11

M is the total first mass of the system. moment of the systen with respect is zero.

where

n M =.I- n. i=l 1

show that

the total

to any plane passing through the center of gravity 3.11. For the point-nass

system described in Problem (3.10),

the quantity

**raj* ' n raf


where d. is the distan cetoa
l-

*r'di,
given line f
of

in space, is cal1ed
rL. Given A, B,

the monent of inertia a regular tetrahedron

16

o\

the system about the line

ABCD and unit nasses at the vertices

C,D(thatis,m.=1,i=1, line

2 , 3 , 4 ); f in d

if L is t h e
BCD.

passing through A and the centroid of the triangle

3.2.

The Ineq-uali-ty of Cauchy and function u2, ..., that

-Schwarz. (3.1).

We wish to show another we are given .., b, ; in

nethod of using the quadratic two sets of real numbers this section we lift ^1,

Suppose that bI, b2,

3n

and

the condition

the numbers .k,

bk be positive.

For any value of the real nurnber x, the expression

- 47-

y = (.1*
cannot be negative, i.e.

br)2 * {arx - br)z + "'

+ (anx - Orr)t

(3. 13)

Y=

Ir("k* ft:

bk)- > o

.2

l{e observe first

that equalitY in (3.13) can hold only if each of the )') (.r,* b rr)- is z e ro ' that is, only if terns (atx - bl)-, "',
b^ b, -l ._ = - 2 = ...
a4 aa LZn

( 5. 1 4 )

b_ = -n .
a_

Wem ay write (3.13) as

y=

(i 4 ) r ' ( j,no*) .' \ k=r


{
a=

(i,'i):o

which is'of

the foqn (3. 1) with

I 1 tk2 k=
n
-2

( 3 .1 s )

I
I

b=

t
n c= \

k=1

tkbk

k=1

I al

Notice that we have inposed on (3.1) the additional condition (5'13)' nanely, that y cannot be negative. The value of a in (3.15) indicates

y cannot that the parabola (3.1) nust open upwards, and the condition that be negative neans that the parabola cannot closs the x-axis at two distinct points. Consequently, the discriminant b2 ' 4ac of the quadratic

-4 8 -

(3.1) cannot be posit ive,

which neans that

\ o ( u i ,tn'ni

ln

OT

( 3. 16)

(-I, : (-i, "-,-)' 'il(-!,


,\ b.- |
K/

This is the Cauchy-Schwarz inequality, equality useful holds is given by (S.I4). by taking

and the condition

under which

Another forn of

(3.16) which is (3.16) :

may be obtained

the square roots of the terms in

( 3.17)

I n L I _\ k=1 "nonl k = f
F

(.t-, "i)'''
suggestive, is 2 or S. since each The fo11owmeaning.

The right-hand

side of

(3. r 7 )
asa

is geonetrically distance when n

term nay be interpreted ing example illustrates

the geometrical

Exanple 6.

Given the points

A(a, ,a2,a3) and B(bl,b2,bS)

in a

S-dimensional space. the origin distance (0,0,0). formula:

Find the angle between OA and 6T-, where O denotes The three sides of triangle 0AB nay be found by the

I (* (3. 18)

2\ =ls "n) 7 / 2 lni'

loB
iAB-_

2 1

.b? = l3
-or)' *
(a2

L/2

\ui,'i)
^r-br)2

irt'". t

= -on)" (-i, I
(at , \ 7 /2

-4 9

and we nay find cosines ,

the angle o, between 0A and 0B by means of the law of

AB2 = oA2 * oB2 - 2oA.oB cos 0,, by s ubstituting the expres s io n s in (3 . 1 g ):

(at

br)2 * (az

br)z * (as - br)2 = al * 2 2 ^z*'3


- 2OA.OBcos cl,.

.o ' r.o 3*b:


J

If

we expand the terns on the left-hand we are left with

side and nake the obvious can-

cellations,

0A'0B cos 0, = rlbl

* ,Zb2 * .SbS ,

( 3. 1e )

cos0=

"1b1 "1 *

*^2bZ*^Sbs

*^3

The formula

(3.19) for

the angle o, shows, at least

in dimensions 2 and the cosine of

3, the neaning of the cauchy-schwarz inequality an angle lies zero. between -l

(s .L7):

and +1, and can be I only when the angle is arithnetical proof of (S.17) allows spaces;

The fact

that we have a purely the notion

us to introduce indeed, this

of an angle into

higher-dinensional

is precisely . spaces .

how one may set up an analytr.c

geometry ln

higher-dimensional

Exanple 7.

Use the Cauchy-Schwarz inequality if A > 0, B > 0, 0 ( x (

to find
T

the minimun

value of A csc2x + B r""2*

We have

-5 0

A crc2x t B sec2x = 1A csc2x + B ,u"2*) (sin2x *

"or2*)

> (fi- csc x sin x + r,/6-sec x cos x)z , or A csc2x * B sec'x > 6A + JE)' , with equalitY if and only lf ,r/[csc x sin x tan2x = 6secx cos x
''--)

,o T

4 lA . x = a rc f , a n q / 9 .

value of the firnction 8. Find the mininnrm Exampl,e 2 4' ' L ( 16 + x +y )(T 6 -*i .L . 4l . This expressionis essentially the rightyI b2= hand si d e o f (3 .1 6 ) w i th a , = 4, d2= x, 83 = y2 and b, = l, Ir
b3=

+ v

Hence (16+ *' * ,o)(# .

r, tr.
xy

> (1 + I + 1)2 = 9,

with equafity

if

4x -=-= L /4

rl x

x = 4,

y = 2.

t/yz
solution involving the

In exarnplesof this t)rpe, there is an alternative inequality we have


?-

between arithrnetic

and geometric neans.

For the first

factor

L6*x2+

y4 > sJr c*zya ,

-5 1

while, for the second, we have


lll

re*a*v xy
If we nultiply

,tm
we obtain
I

these two inequalities,

(16

t *2 , r4lr$

.*.*, xy

zs:m

9,

with equality only if solution. The.inequality

16 = *2 = y4; this result coincides with our first

of cauchy and schratz nay be generatized in nany

'uays, end several of these crtcnsions sre interchangeable intheirapplication to eloentary o a classical geonetry. fle wish to introduce here a simple forrn be treated at length in Chapter 4; Let {^r,^r,rr} and

inequality

rhich will

this is a special case of Minkorski's in_equality.


r'l

tbt,bZ,bSl

be two triples

of real numbers; then

( 3.20 )

,l'' ,i)"' ,r)"' (.! . br) , (,.!, . (,.!,


,(k
^s q' E =E = ^z tl

with equality occurring only if

(3 .2 r)

The proofs of (3.20) and (3.21) are obvious on the basis of our discussions in Exanple 6 abovez rf re fom the parallelogran of which two sides are C(a, + b'

OA and OB, then the fourth vertex will a,+ b* a,

be the point be

* bS), and the length of OC will

-5 2

,r,r1.J

,/

(^., * b.,)'

* ("2 * br)'

* ("3 * br)' .

In the triangle

oAC, we shal I since

have oA + AC > oc with equality only if

the point A lies on 0c.

AC = OB , (3.20) and (3.21) f o llo w a t o n c e . ' I n o t h e r wo rd s , (3 . 2 0 ) asserts that the sun of two sidles of a triangle si de . Exanpl.e 9. From the point C(O , c ) a n a n ru n s t o a p o in t P (x , 0 ), O{* ( l, a n d t h e n t o t h e p o in t F in d t he " nust exceed the third

D(f , d ) (s e e t h e f ig u re ). point P so that CP + pp is minimun. I ,0 )

The distance to be covered is

CP+PD=

and by (3.20), the sum of the two radicals


|.,,, 4Grd)o*

is greater than or equal to

(x*l-x)'=

Thus CP + PD wil.l assune its ( 3 .2L) holds, or when cx

least value

when the equality

a-= 1There is a particularly

or

x =

;J!-

l-x

c+d

elegant geometrical solution to this problern, in the form (s ,za),

which brings out the fact that Minkowskirs inequality, sa)'s essentially

that the shortest distance from one point to another segrnentbetween the two points. Reflect the point D

is the straight-line

a;

-5 3

in the x-axis

into

the point

Df (f,-d). p(x,0) also To ninion the C to then to

Then the choice of the point ninimize nininize the sun CP'+ pD will the sun CP + pDt.

(1 , 0 )
nize the latter straight-line Dr.

sun, p must lie segnent drawn fron geonetry, it

By elenentary that

follows

the minimrn is

(c * d)2 *

and that this ninimum

occurs when = x "fu

and1 - x = #=
we shall

The additive property of the sunmation sign in (s.13) suggests that we may exploit even further the quadratic function (i.l).

assumesomeknowledgeof the elenentary integral calculus, even though, with the exception of the Lenna below, the following developmentsare co n ceptually identical to t h e d e riv a t io n o f (s . rg ), above.
Suppose that we have two functions f(x) and g(x) which are both the

(3 . rs ), a n d (3 . 1 6)

continuous on the closed interval expression [f(x)t a < x < b only if of f(x). - g(x)]z

a < x < b.

For any real number t, and is zero for all g(x)

cannot be negative, that is, only if

x in

g(x) = f(x)t,

is a real nultiple

Consequently, we have

( 3 .22)

y = [^b [f (x )t - g (* )] 2 d x > o tta


real values of t; this is t h e a n a lo g u e o f (3 . 1 S ). in

for all

T h e c o n d it i o n s

under which equality in (3.13);

prevails

(3.22) are not arrived at as sinply as we have obtained the Lenmabelow.

and we reserve connent until

-5 4

The inequality

(5.22) may be written

n b .., y =.1 ([f(x)l' t"


ua

- 2t f( x) g( x) + [g( x) J2) a* , o

or ( s.23)

, =Vorr(x),'0.),2 - z(/or,,.,s(x)d,.) ,rqx;rzaxo. , .f >


quadratic function / = ar t2 * bft + cl

Here we have again a non-negative of the forrn (3.1),

where we have used t in place of x, since x is re_ in the integral; the coefficients of the

served as the "duntnytt variable quadratic are

(s.2 4)

I
I
I

,, = ( x ) a* !^[ f 12 b ,=f t r - le( x ) dx

le " ' =f ( x)I2 a * I


ug

Since the quadratic function have distinct be positive: real roots,

(3.23) can never be negative, it

cannot

so that the discrininant

b ,2 - 4rrc' cannot

o <0, (/'rr(x),rol(y',rr,.rr2aj [or,.,s(,.)dr2,


OT

( 3 .2 s)

u, (.f'r,.,sr*ra*)2 : (/'rr(x),, (f,r(*)r2.,.),

-5 5 -

which is called Schwarzts inequality lfe now take up the question of when equality can occur in (3.2S), and this is equivalent to equality inPortant in nany contexts, is essential. in (3 .?2). The following Lenna is

and we enphasize the fact that the condition

of continuity

LEIri{A. Let F(x) be continuous and non-negative on the interval r'b a <x<b. Ttl F(x)dx = 0 , t h e n F (x ) is identically zero.
-ta

Proof.

Suppose, to the contrary,

that F(x) is not identically

ze"ro.

Then there is a point xo such that F(xo) I 0.

Let us denote the value Because

F(xo) by A; we nnrst have that A > 0 since F(x) is non-negative. F(x) is continuous, there cxists an intenral l* - *ol . 6, o"

*o - d ( a < *o * d, such that F(x) > A/z for every x in this (rf xo should be one of the end-points, I or b, it will half of the stipulated interval, suing argunent will tive, we have

^interval.

be enough to take

.g., b - d < x <b if xo = b; the ensince F(x) is non-nega-

also apply to this interval.)

, o =.fr(x)dx a .J_;.or(x)dx +u,. di.o


A do*6 2 * d- d d x= + ( 2 6 ) = Ad > 0 ,
o

and fron the contradiction caLly zero.

(nanely, that 0 > 0), we have that F(x) lg idettlstep in the A6.)

(rf xo had been one of a and b, then our last

chain of inequalities

above would have yielded A6/2, rather than

-5 6

To conplete the discussion of Schwarzfs inequalitlr non-negative quadratic function arising distinct reaL roots.

H saw that the

fron (3.22) could not have rwo

However, equality nay occur in (s.z.z), and there-

fore in (3.25), if, the quadratic function (s.22) possesses one real r oo t, say to. The function Then for this re a l ro o t t o , we h a v e { o t t t * la o [f(x)to - g(x)]2 i, - e (x )lz dx = 0.

continuous and nay therefore play the - g(x)]2 i, identically zero,

role of F(x) in the Lemna.- Hence [f(x)to

so that, for all x in a < x < b , g (x ) = t o f (x ). Exanple 10. va lu e 1/S . we have Let us consider the integral .t x . x 3 , a n d if * u ' l* " *o dx, which has the

lfe nay write *4. ,

f (x ) = x a n d g (x ) = x 3 ,

= (I'

*aa*)'= (/ *.* tr'o /

. .fu, ur'(r

/ / t . uuJ =l. l= l2L -l 3 7


\Jo
= x2 and

On the other hand, we may rrite g ( x) = x2.


I In this

2 *4 = *2 . x and take f(x)

case, we have

E=

/1t *oo,)t= ( f' *2.*2a*) YoIYo I '(/'


t.s obvious that

*.t /.r=|.*= (d'


the equality occurs because S. (This is a how the after

In this second instance, it f(x)

is a constant multiple of g (x ). Exanple 11. Showthat, for x > 1, tog * t

sonewhat contrived illustration,

einee Lt ls not at ell.obvloue but tt lllustretea, I{e have

i.nequallty le related to Schwarzrs lnequality, the fact, how the lnequallty uay be applled.)

'57-

(x - t1 2 = ({'..)'

=({.*nu,)'.:.1 .a, (f (.{'


= (t o g

*) r| t*'

- , )),

rh e n c e

2(; - tl'
x t-l

< log x,

ot

2 { * , -, t ) x+l

< ro g x .

Set x = e, and we have 36"g .1r i in this inequality,


1I <

( l, or e < S.

If we set x = e L t 3

va obtain

TT

2(e" " _ t)

.Iroe"=|,

or

" .4 =

2.744.

l'e renark thet this resurt rs useful in the nodern calculus course, where log x is defined ,tf.* dt/t, ard where the inequality of Schwarz rl is easily proved at the tine that the definite integral is introduced. of course' the inequarity Taylor-series of Exanple ll nay be obtained fron the

expansion, but this concept usually appears nuch later calculus course. bound for e nray be obtained by analogotrs nethods.

in the introductory An easy lorer From the inequality

t = ( / . "i "-1 . ({. o.)' "'u.) "-..1 (,{.


= (u * - l) (-e -x * l) = e x

**-r,
e

we have

and, if for

we set x = ln we have that we have e , 2. 7 0 5 .

"

r 3 *- 6

# 2.62. r f we set x- L13,

exanple,

-5 8 -

Exercises for Chapter III -,1 S howthat [x + y * z )(i S hswthat (x,+r^* 'LZ
tl , ..., X

3.12 . 3 .13 .

rr * i *;) ,\rl L

:9

f o r p o s it iv e x , y , z .

* *r,)(+ .

) r,2fot positive

n' valucs-of A and B find the ninipnn value of (0 < x . n ).

3.L4.

For positive

AB I - cosr

I + cosx

3.15. Find the largest vrlue of sx * 4R


0 < x < 2. 5.16.

in the intenral

Find the largest and:snallest values of l2x + Sy + 4z if

the

point (x,y,z) is constrained by the relation x2 * y2 * z2 = l. 3.17. 3.18 . Find the nininun value of S howthat g sec6x * 3csc6x for 0 < x < n/2.

+,t.?: T ,+ .?
#. # - # ls lz f o r p o s it iv e x , y , z .
wit h a > 0 a n d b > 0 .

2- 22

3 .1 9 . S how that
3 '2 0 .

Given the points A (0 , 0 , a ) a n d B (l, r, b ) use the inequality (x,y)-plane

(3.20) to find that point p(x,y,0) in the

so that AP + pg is a rninimrnr. Verify your result solution. * y2 ,

with an elementary geonetrical 3.21 . 3 '2 2 ' S howthat * yz + z x l <*,

l*,

"2. a 3 ), (b r b z , bg ) ,

Given four tripres of p o s it iv e n u mb e rs , (a r ,'?2, ("1, "2, "s), (dt, dz, d s ). S h o wt h a t

jr'ti

-5 9 -

("tbt"1dt + azbzczd2 + agbScrdr)4

. 4 c2,

, * t"1 "l *'!l cul a!* ollr"l ,


3.23.

"!rtaf* a l , ulr.

) 4x" - " '' ^ For x in the range 0 < x < tt/2, show that -' -" -" -" -- -+- + s in 2 x < "anx 2x (Hint. S tart with t h e id e n t it y * = 4 * x . a re re a l, 2 ' 2 n - < n ( *i**r + - "' 2 **i) ,' sectcostdt.)

3.24.

S howthat, if xrr .. . , (x, + x- * .:. 'I

+ x )z

and conpare the regult wlth (2.5). 3.25. Let alra2r...r8n be poaltive ntders a re re a l, t h e n euch thet
E

tit"f

. 1.

Shon

thst lf *lr*2r...rx'

(.a*r.*

^Z*2+

... + aoxo)z iu.'ll.l+ arxl+ ... + a x 2). s n

(Coupare the resulr wlth problens 3.18 and 3.24.) 3.26. Find the largest and smallest values of arx, * ,Z*Z * for a, , ?2, ..., 8n p o s it iv e if x+ax nn

222 *1 *2

i*i+...*+ bi b;
3.27.

b;

- l.
on the interval < b; a q x ( b, prove that r \

Let p(x) be continuous and positive and let f(x) lnb

and g(x) be continuous on a (x [ rb --. --2 . . \/,,b

tl.l t,1 r(x)g(x)ptx)dxj \va


---12

\%

Ir(x)]zp(x)dxf tt(*)lzp(x)oxf , I l'


/\r" I
The

with equality only if positive

g(x) is a constant multiple of f(x).

function p(x) is called a weight function for this form

of Schwarzrs inequality.

- 60-

3.28.

Given two sets of real nunbers trr, Prove the Cauchy-Schwarzinequality sun nn j=l k = l is non-negative.

"r ,1

a n d { o r , ..., bJ
that the double

from the fact

II

("jo* - anal2

3.29.

Given two infinite

sequencesof real nunbers,


C) a n d t br , bz, . . ., bn, " ' J , such that

r-) . . -, 4n, . . . 1^ 1, ^2, t


ot@

the series infinite


3.30.

X a1 and X b; are convergent. Show that k=l K k=l n series f, a. b. is absolutely convergent.
k=l KK

the

suppose that

f(x)

and ft(x)

are continuoue for 0 < x < co and

rhar rhe inregrals rl


wtth f (x) ttre lntegral

trt"ll2ax and rl
of f r(x).

rr'(x)l2d* .r" ftnite


f (x) = 0.

Show rhar frn

-61

Chapter

IV.

The Inequalltlee

of Hiilder

and !,llnkwekl

4.L. be the real

A l,ecna. ntober

Lt p be a real deflned by

nunber greater

than 1, and 1et q

(4.1)

l*!Pq

I :

lt

follose

at once that

q > 1.

Our prlnclpal

tool

ln thls chapter

1111 be the lnequallty

(4.2)

4 -qt= :2 p
only

xy,
real nun-

wlth bers.

eqtrallty

tf xP . 19, wtrere x and y are aoy poeltl.ve wharp-9-2 r 22

Ide note that

(4 . 1 ) f a satlefled md (4.2) becmea

|+t , n,
an Lneqnallty ehotr that resulte (4.2) relatlng the arl.thetfc and gemetrle for ratlo,nal means. I{e ehall the

nay be deduced dtrectly 2 on arlthnetlc

p and q frm

of chapter for

and geometrlc real nunbers,

mean8. {ay ar,

Then we sh811 ..., ao} md

ghow that, {b1, b2, "',

trlo sets of poeltlve bo},

(4.3)

.1 b 1 +a rb r+ " '+.o b o

+. .. *":Jt'n[ol*oro * rlJt'0, r rc. ["i."!

nlth

eqtrallty

lf

and only

lf

(
I

-6 2 -

( 4 .4)

"l "t " o l =4 =


(4.3) is called in what fo11ows.

ap =-. n bq n

The inequality generalize

Hiilder,s

inequality,

which we shall inequality

We wish to renark that H6lderrs

is an inportant

extension of the inequality that

of cauchy and schwarz. when p and q are rational, "elementary methods", processes). can-

We wish to enphasize.here the fact we ate able to deduce (4.2) that is,

and (4.3) by so-called analysis

by methods not involving

(or infinite

of course, when p is irrational, not be defined without

the functions

xP and yQ ir, (4.2) or limiting

the use of an infinite

process (for *analytic the

example, *P ."n be defined as eP(log x)r, nethodsn are necessary. sane results renarked, it Certain analytic

so that

so-called

methods will

also yield

in the case that p and q are rational, is one of our objects directly to show that

but,

as we have of

the inequalities

Hiitder and Minkowski follow the arithmetic and geonetric

fron the inequatities

relating

neans whenever p and q are rational. x and y and for rational the posltive integers p = n/n > 1

To prove (4.2)

for positive

we consider the expression for

m and n:

mm
-+

xp p

yq q

,,*il *.

-(.r n

- ,,)y*-n

cl=t -l=[ +a), 'qpm


and geometric rneans, is

which, by the inequality not less than

between arithnetic

(*tyt)r/m

= xy , Hence

with equality

only for

xP = yq.

ii, ,:

-6 3

*P*r9, pq
with equality if and onty if

xI'

xP = y9, which is (4.2). (a.r). As we nentioned

Now (4.2) holds for real p and q satisfying above, when p, andthereforeq, is irrational,

we nust use analytic inequality

methods. Whenone takes the linit

or both sides of a strict

(<), one must realize that the symbol can degenerate into the 'fless-than- o r - e qual" case (5), e.S . 0 < t / n , b u t 0 I llg (f f " l. cases involving lirnits of inequalities, inequality He n c e , in t h e

a separate proof of strict inequality

is required in those cases where, indeed, strict

i s possible. For any irrational p, then, ret us consider the cxpression xP P vq q,

- +!- -

which was used in the proof of (a.l suppose that,

for rationar

p and q, and ret us

for sone x and y with xp I yq, xP pq vq

X/.r

and let

(4.5)

xy-

,+. t

=d>0.

We shall show that there exist rational ( 4 .1) such that ( 4.6 )

nunbers po and eo satisfying

x v -,# .*,
*,; it

-6 4 -

diffcrs d/2; this

from the exprcssion on the left-hand and the fact that the expression

sidc of (4.6) with to

(4.5) by less than rational exponenrs

is negative by (4.2) then yield may conclude that yq."r,

a contradiction arbitrary closely real

(4.5),

from which we Since xP and this neans that

(4.2) holds for

p and q.

be approximated arbitrarily

by rationals,

*P = *Po + e, and yQ = yQo + e2 for po and Qo sufficiently and q, respectively, (4.1), wherd e, and ez are arbitrarily but this relationship offers

close to p

sma1l; because of no difficulty.

e, depends on ry

We have that

xv-,*.f+ t #. *] } = ,
=n *to D

_*P .*._ p qo

* =*oo,* tt - ?. rn",{- ro) 7


on the left-hand side of this

Hence the absolute value of the difference last equality


TJ

does not exceed


q

x'olp - D I ' - ---;e_-- ' o '


"o

v'o

lq' - o I
'ot

+ _+

l. r-l

I I

ll

lezl
q

. * P o lp

* y q o lq - e o l * lr ll
Since this last expression tends to zero ,, this that

po tends to p, we can choose expression is less than d/2, Hence (4.2) holds for xP = yq. 5 will

po close enough to p to ensure that which gives us the contradiction (4.1),

last

we seek. if

any p and q satisfying I{e remark that obviate

with equality

and only if

the powerful--and

urifying--resul-ts

in chapter

the need of dLstLngulshlng contexts.

between the rational

and irratlonal

cases here and .in other

-6 5 -

4;2.

Hiilder's Inequa lit y .

Wit h t h e e s t a b lis h n e n t o f (4 . 2 ), v e

are now in a position to prove Hiilder's inequality in the form (4.3). ..., arr} and { b r, b 2 , . . . , b rr} b e t wo s e t s o f p o s it iv e n u n ^ 2, bers, and let p and q be real nunbers satisfying (4.f). Let us set L e t {^ L, A p=
... ., 4P n

"?r * "\ *

(4.7)
B q-

ol.ol.

...

.r 59 n

so that 1nP1l/n 1na;l/c = AB. we also set


a, = A cr, bl = Ml , &2 = h c 2 t bZ = B d Z , ...,
a =Ac nn

=Bd

From (4.7) we have

np= np"f *
so that ( 4.8)

lP c! + . . r +

*epr APcl= AP("1 *

tl . "! + "'

*cP=1 n

Si m i l arly,

(4 .e )

dl . d 2 . " '

*dQ=1 n

By the use of (4.2) we have

- 66-

( 4 . 10 )

I I

"tbt

= A B (c rd r). < A B

r{. ull q,) Ip


ull |,"l.qJ l.p

^ ZbZ=

A B (c rd r). < A B

ab nn

=AB(cd) -nn'

<AB

[$ .$ i
we obtaln the reault

If

we add the n inequalities

of (4.10),

a rb r'+ a rb , + .,. *

otl "r r bn :

cP n

af.

. a;tl
-)'

and, by (4.8)

and (4.9)1 we have

"lbl*^ZbZ*

+anb6 I

(i. AB +) = A B ,
(4 . 3 ). E q u a lit y h o ld s in (4 . 3)

which is precisely Hiilderrs inequality if and only if

equality holds in each inequality of (4.10), whlch ylelds

the conditions (4.4). Renark. The inequality (3.f6) of Cauchy and Schwarz given in Chap-

ter 3 is a special case of Ht ild e rt s in e q u a lit y (4 . 3 ), wh e re p = 2 , e = 2 . Also, since our proof of Hiilderrs inequality above rlas based on the inequality between the arithmetic Cauchy-Schwarz inequality and geonetric neans, it follows that the between

is also a consequenceof the inequality

the arithsretic and geonetric means.

-6 7 -

4.5.

Minkowskits Inequ-ality.

In Chapter S we encountered the

special inequaLity (3.20),

t b*)']"'i [-!,"i]r /2 . [*!,,*


which, in the case of triples {ar, ar, proved by neans ofanalyti"g"ot"tty. of this an d {by inequality b2,...,

[i q ) " '
\ k=l a, ,

and {b, , b2, br}, was easily ^j} l{e wish to prove a generalization "n} The

for two n-tuples of real numbers, {ay

bn}, and f o r a n a rb it ra ry re a l n u n b e r K > l.

r esu lt is Minkowski's,ineq u a lit y :

( 4 .11)

* o')] r /K ' I ri,'"t


tr
i- =- l=

.,-J o'*] : [,1, . [,1,


r /K = 5n . _
n a

t/K

with equelity holding if

and onty if

t4,L2)

bt

bz

^z

the proof of (4.11) is a sinple application we se tP =K andq=K f=

o f Hd ld e r' s in e q u a t it y (4 . S ) ;
I

FT
that

, so that I *

= t.

T h u s , it

f o llo ws

from Hiilder's inequal ity

* bi )^ = ,1,

vn

ar(a, +

[,i, "*[,!,t" ""]


"* [,i,'r-] [,1,t"

-6 8 -

If we now use the relationship

between K and Kr and factor,

we have

,!r("t

* bi)K'

f l ri '.*l t/*[,"i i L ) ti tilr


1

+ D.l L ')

..KI

\ - K -1 K
|

(n + l 'i orIr ( "i \ i=l L ,I,

* o r ) *]

K -1 K

={[,], ',*]'/*.

[,!,

o.* I r)

t' -)[ rl r,' ,

* b i )* ]t- i ,

or

t"' [ ,1, *
which is (4.11). directly

(n

t'* bi)* ]

. : [,],.1]"* [,],or , ]"*

The condition for equality, nanery (4.L2), follows

from (4.4) if we have simultaneously that .K "1 (* - l)1 t -(rl * or) ,K ol


= ... =

iI
I

l
I
I

ffi -n

a n

n.K D
n

I I

ry

ffi
bt

or,

since (K - 1)Kf = K,

"l ..--...--:_= "r*Dl or, a +b nn

and

"r *bl

b n ;-+ T-nn

what ls the same thlng, a =- n b n

E=
;!

"l

-6 9

Minkowskirs inequality K > l.

(4.11j has been proved for arbitrary

real

lfe renark that, wheneverK is rational,

Minkowskirs inequality between the is such a

nay be regarded as a direct arithnetic

consequenceof the inequality

and geonetric neans because Hiilderrs inequality

consequence.

4.4.

Ggneralizations of the Inequalities.

lfe begin with the forn

(4. 3) of Htilder I s inequat ity,

( 4 . 13)

r /q .,0, [ { )"o[ r ,1, ,1, ,g, ] 'l


by the equation 1l +

where p and q are related

pq
If we setl/p=cand

= I.

I/q = $, and rnakethe substitution A. =


1

.l1

and

B.= b:, so that a. = Al and bi = Bi, we may writ e (4 . 1 3 ) in the forn 1

na

( 4.14)

o?'i o,]" ,1, 1 [,g, [,!,,,]u


for equality in (4.L4) now takes the form A .A ^ A
LZn LZn

The condition

(4.ls)

B ,B ^ B

= ... = _

suppose now that we are given three n-tuples of positive {Ar , "', A rr}, {t,,-..t B n } a n d t c r, -. . ,

numbers

c rr} , t o g e t h e r wit h t h re e p o s i Let us consider the

tive nunbers 0,, B, y sUch that o + B + y = l.

i.:1

-7 0 -

expresslon

i= l

B n9 ti .1. If wewrite 1 sP cT= 11

['t'*]'-o=

(Di)r-',

w e have, by (4.14), that n

1= I

I l? nl-o 11

: [,i,o, [,i,o,' ' ]" ]


Az
A =- n D n

with equality only if

At

= = ... Dr Dz

nn Apprying (4.14) to E D.= f , t]=.1= i=l I i= l we have that n

8y

after observing

Yl -0

l,

.r]-, I 1 i =l ']-c'r-c [,i,'r,|-[,i,


so that

| n

,= B . n

r=Y

l ii=l \
with equality only if

rn

B
Bl-0 i

a, Jt- " .1 n t*-

r3r n

' \Y

rrl,tiJtr!,ttJ '

Bt

q
This inrplies, finally,
that

=, z n r=

B^B

=q'

-7 L -

(+. ro)

ol'l .l I [,],o, [,g, ]u[,i,.,]' , ]" ,, ,1,


..., o r, ), { B r, ..., Bn}, {c, , Cn } a re sets. This form of Hiilderts inequality numbers{ x lm, nay be generalized X 2 m, " ' , * n ), re a l n u n b e r s

wr th equality only if tlr, proPortional

to any number of n-tuples of positive

N, for any -s e t { o l, o , , a N} o f p o s it iv e ^=,r ,,,..., such that o1 * o2 + ... * = l, namely, t{ n ( 4 .L7) i=l

... r .li ,.13,.lil [r!,.,,J"'[rl, *rr]o'... .rn]h. [,i,


and the method of deriving (4.16) fron (4.14)

The proof is by induction, nay be used.

we nay change (4.16) to a forn sinirar to (4.s) for certain applications. "r , t h a t , i f l /s = p , L /B = q , l/\ = r , *e haveA. = with S ta rti n g w i th (4 .16) , letusset Ol = ti = b, and CT = c i' "1, B, = bg, C, = r
so

"i-'

l. i *I ='.
firen (4.16) becones n f .arb,c. i!1 rrr ( 4.18)

n 1 [,i,"lJ"n 'l ] llqr [,1, [,i, "i]


the n-tuples { af,

L/t

with equality only if t"l ,


,t]l.

...,.l],{rl, .-. uf,], ana ,


change nay be effected
:i

...,

";l w i th (4.17).

are proportio n a l s e t s .

A sinilar

- 72-

Examp_le 1. Showthat

Given nine positive

nunbers with

"lbl"r

tsbg"s ^zbz"z* =

l.

. " "i 1"1 "?t "3)3 . ul. ulltf . .l * "9., 1ui |.-3-ij |.#J [---#-J :1'
(4.18), have,since From we + " +* * = r,

. c"i . ,l * "ltrtz (bi. a1, rltt/t r"i . .f . "ltt/6

.3 with equatityif andonly ir af , uf : .l = = t ,, V3, lo g "?, "l "l = x : y : t , s o r , e h e r e r /zyL ls " i. ( = s. Ra isin g * x both "! "!l of this inequality to the sixth power and dividing the desired inequality. l{e nay also generalize Minkowskifs inequality Let us take inequality (n (4.11), even nore easily.

, t " 36 = sides

by 56, we amive at

\i= I

II
if

(*i *

urrr]l/r [ri,"f]t'*, [rl,,$"* , :


a. ,i is constant)

with equality

and only if

addt.I, "flt/*

to

both'sides and use Minkowskifs inequality:

*. f i "fl"*.ilfi of J" | ' i I L ) lilr (i=r L) t

. [ri, x)r/x
c. I
L)

II l \ i=

1n

(a . * b . *

t'*, *J "r,

rtr

- 73-

with equality is constant.

in the last Thus,

inequality

if

and only if

a. + b. b. ra. 1 1= 1l+ *'] c. c. lb.


I 1.1

(4.le)

r)l f p a.r /x + fl I ur))l l/ K * I t | [i=-t L) l1]1 ii


a.

(! lI tilr

t/*
r); "fl b. and f c.
I

, II (a. \ i= I
-1

(n

+ b.

*] t'u, "r,

with equality if i sto sayal,blicl"a2

and only"if

I.

E: ]tbz

1S

constant

is constant, which

tz=

=a

nn

:b

cn.

This is

easily extended by induction t o (n

II I (xil \i=

* *i2 * ...

* *iN) *]t' * a

,.f,]"* .lr]"* : [,!, * [,!,

+ ...

.,*]"*' ['!,
xnl : xn2 :
: xnN'

with equality if and only if xr, : x12 : ... r *lN

4.5. integral

The Int,egral Inequ-alities of Hiilder and lr{inkowski. We obrain analogues of the Hiitder and Minkowski inequalities in rnuchthe

sane nanner as he obtained the integral -Schwarz inequality (3.17).

analogue (3.2s) of the cauchy-

l{e follow rather closely the nethod of proof

o f th e inequality (4.3) in S e c t io n 4 . 1 a b o v e . Let f(x) and g(x) be continuous and non-negative on sone interval (4.1); we as-

a ( x < b, and let p and q be positive nunbers satisfying sune that neither of f(x) and g(x) is identicalty avoid proving a trivial result. Let us set

zero in order to

-7 4

J trc*llPa*= Ap
rb

rb

( 4.20)

JIer * llea*= B Q,

where neither of A and B is zero; the functions q [f(x)JP and [g (x) ] are continuous since f(x).and g(x) are. Let us set

( 4 .2L)

= F(x)+
G(x)=9;

again, F(x) and G(x) are obviously = a rb a ,b^ J At[r(x)]Pd*,

.h

continuous. rb

Since

A P = J " I f ( x ) ] P d* =
a Sinilarly,

it

follows

that

t. J te C*llPd x=

J t c C *)l q d x= 1 .

Now

rb ;b x J f(*)g (x)d x = J A B F ( x ) c ( x ) d= A B J F (x )G (x )d x ,
fb

and, by integratlng of y, we have

(4.2) wlth F(x) ln place

of x and G(x) tn place

A B J F (x)c(x)d x < A B f b / t p r * llp


a

|.

. *?.]:)u" = ot(i.rq) AB. =

Hence

1/n qnel 1/e, J f (x)e (x)d x < AB= qnPl

so that,

by the definitions

of A and B,

-7 5 -

(4,22,

: , Uo,ro,,o]t' J'r,*,r(x)dx [{orro,rno.]t'n


[f(x)]P ir a nrultiple of [g(x)]q. t{e renark that (5.25) if

with equality onl.y if

(4.22) reduces imrediately P=q --2.

to the Cauchy-Schrarz inequality

For three (or lnore) non-negative continuous functions f(x), h(x) on a < x < b, and for three positive o + B + y = l, it

g(x),

numbers c, B, y such that

is an easy rnodification of the proof bf (4.16) to in its integral forrn rnay be written as

show that Hiilder's inequality


rb

( 4.23)

Jro gB nY a*

sd ,.][{ " u *],


for the exercises (see Problern4.1).

and we leave this nodi.fication three positive

For

numbers p, g, r with

t p
we also have the alternatlve for:m

G.24)

fou,nu* fi t,n u * ]" n t

[I'

d* sQ ]"t Ij' hrdx)"" ,

again, the proof is left

to the exercises (see problen 4.2). forn of Minkowski,'s

lfe conclude this chapter with an integral inequalitri inequality. interval

as in the case of (4.11), the proof dependsupon Hiilder's Let f(x) and g(x) be continuous and non-negative on the We

a ( x < b, and let K be a real nunber greater than 1.

-7 6 -

consider

the integral

1b J If( x) + g ( x ) l* u * ,

and write it
.h

in the forrn

I f(x)[f(x)
a

+ g (x) I K -t d*

. Ior(x)[r(x) + g (x ) l t -t
a

d*

which, by (4.22), is not greater than

(K-r)*'o*]l/K' + s(x)] * Ijotr,.,]Ka*Jt'[Jo,r,*,


)1 . [j're(x),*]" +e( x) l ( X- 1) K'o*,,|/ x ' ' Ij0,r,,.,
where K' =
v

Ri-

so that

i.#=t.

If

we subsfitute

r-T

for

Kr in this last expres-sion, we have

J tr ( * l + g 1 x)l K a x :

rb

* fjtrr(x)rKa*If tr(*)]Ku. [f ,r,., e(x),*u.]'-*, ]"* . ]"*] i


so that

( 4 . 2 5)

* I jott,,., e(x)]Ka*]t'*, [ {0,r,,.,,*u. ]

I/K

' Iiot",.)]Ka* ]t",

-7 7 -

which is t he form of Minkowski ts inequality holds only if of (4. 25 ) that f(x) = ag(x). Note also that

which we want. it is a trivial

Equal ity consequence

rf b (4 .2 6) l j lf(x) + g(x) \a

+ h( x ) r * u, . j

L/x

fo r three positive function s f (x ), f( x) = ag(x) = th(x),

g (x ), h (x ), wit h e q u a t it y if

fo r c o n s t a n t s o a n d B .

Exercises for Chapter IV 4.L. 4 .2. 4.3. P rove the inequality (4 . 2 3 ).

P rove the inequality (4 . 2 4 ). Use Hiilderf s inequality to showt h a t , f o r p o s it iv e x , y , z ,

36

-'x

. r!*!*y

)t

(* 5 * y 5 *

" 5 ).

4 .4.

Showthat,

for positiv e X , L

z,

27

. : (1

- +- l

L t,2 (*2 * y2 * r2). yz'

4.5.

S uppose that, for x t -1, o is a rational Showthat


(t+x)a<l with equality
+ 0,xr

nunbet s u c h t h a t 0 < c < 1 .

only for x = Q.

-7 8 -

4.6.

For x , -1, let a be a rational nunber greater than l, Show that (1+x )a > l+ c x , with equality for x = 0. a ) 0,

a > l.

4.7. '

use the result of Problen 4.6 to show that if 'c x ) 0, the nininun of the function *o assumedwhenx= (1)l-o . -ct-

s > l, and

=l_

"*

is (r - n) (3)"-1 and is

4.8,

For x t -1, lt a be a negative rational nunber. (l+x)a > l+ c D(, with equality only for x = 0.

Showthat

4.9.

Extend Problem 4.5 (and therefore problens 4.6 and 4.g) to the case that a is irrational numbers {rrr} such that ll1 rr n# by the use of a sequence of rational tr, = r.

4.10. For real nunbers x, d, c such that x ) 0, 4 > 0, o > I find the nininum value of the expressior *o 4.11. Use the identity
, l.t

"*.

"or2* showthatrfor0sxsnl},

* sin2x = 1 and Hdlder's inequality to

"o"6* with equality

* sin5x > l/4,

for x = n/4.

-7 9

4 .L2 . llsing the ideas of Pro b le m4 . ll , s h o wt h a t r f o r 0 < x i t t lT , 3" cos -x + s in " x > L / n
with equality for x ' fi/4,

4.15. For x in the interval

0 : x < n/2 and for

any real

number u > 2

0c find the ninimun value of cosx+stnx.

4.14. Let 0(x) be continuous and positive on a < x _1b, and let f(x) and g(x)be continuous and positive on a < x < b. prove the fol-

lowing forrn of Hiilderts inequality with 0(x) as weight function:

t'o < Io.,*,g(x)g(x)dx jotrtr,]p o(x)axl [ [


where p aad q ar two positive 4.f5 . For a, b, c, positive, where n > 1.

0(x)a*]t'0, "lo,r(xrq

numbers such tLat | * | = l. pq

f in d t h e ma x irn u n a lu e o f a x + b v

4-L6. Use the result of Problem 4.10 to prove the fundamental inequality (4.2) for real p and q s a t is f y in g 4.L7. Let p, g, r be rational x, y, z be positive.
xP Pqr

(4 . 1 ).

lll nunbers such that -+-+-=

pqr

1, and let

prove that
yq ,'

- + 4 +-

xyz

with equality if

xP = yq = zT.

-8 0 -

4.18. Prove the following analogue of problem 4.14 for suns: .i {bl, (i = L, 2, ..., "', n ) b e p o s it iv e , a n d le t { a rr . . . ,

Let

a rr} a n d

br} be two s e t s o f p o s it iv e n u n b e rs . I f p a n d q , a re (4.1). then

positive real numbers satisfying n I .u -

a.b.c.
I 1 I

1=l

.1.,][,i, "r] t'0, [,i, '^ ol


ap = -. n bq n

with equality only if

_4=
uf
4 .1 9. s how that (x + 2y + 4 z ) + .
4 .2O . S howthat, for xr, .. . ,

?. h

2 4 9 , f o r x , r . z p o s it iv e .

x n p o s it iv e , 2
I
>

(xt

2x, + ..:

* 2n -1 x r, ) l,-L + +-

-xr

*2

&\ "tr+ -l

- D2

4.2L. For a, b, c, d positive

find the rninirnum value of the expression

s=r+?-. n+;. riT. r+=.


4 .22 . If a2 * b2 * c2 = l, s h o wt h a t

L+*C b
4.23 .' lf

o'
c+a

t2 ,
a+ b

>

a2 * b2 * c2 = L, s h o w t h a t l< (^ 2-P * b2 -p * (^ 2 * q * b 2 * q * q 2 * 9 1 P , (4.1).

" 2 -P ; 9

where p and q are real nunbers satisfying

- 8t-

4.24.

Let p and q be such that posLtlve It continuoua < l,

p > I and q '> l. on the flnlte exlsts

Let f(x)

and g(x) be

functLons

lntervalaS!x<b. k such that

Llp + l/q

show that

there

a eonstant

a \a /\a

b rb 1 r / n 1 , b o . ll/ q I t edx < kf / rPa*; It s -oxl

4 . 2 5 . S u p p o e th a t f(x) a n d.f' ( x) ar e c' ontlnuous 05 x < eod thrr e for the tntegrars (x) lPdx and f lf {-lt'(x) lqax are flnlre, wherep
and q are poel.tive lntegral numbers nith f'(x), t/p + l/q - 1. t*g If f(x) f(x) ie the g. (See

of Lts derlvative

show rhar

Problen 3.30, wtrlch Ls the case for p = q ...,

- Z.) Show that

4.26.

Let xrr

xrr

xn be poslttve

nu6ers.

4(x, + 2x, * 3x, + .. . + nx-) (*1 + 2 *... t' -* x^


zn

+ g) x'

: o 2 (o + 1 )2

4,27.

Ler xl,

',,t

*t

be posltive

nunbers.

Show that

0 alf *' {] \k-1 [i, *]'

rn

, nu(n r)6. +

-8 2 -

Chapter V.

Convexlty

and Jensenrs Inequallty

5.1. lnterval two points I:

Convex Functlone.

Let y - f(x) f(x)

be a functlsr

deflned

ln some for any

a < x ( b, and euppoee that x, and x, Ln I,

has the property

thar,

( s . 1)
Ttten we shall that f(x) say that

uf-{--t1 . -1. 2 J:
f(x) is

r(xr) + r(xr)
2-.
t{e shall only that asaume throughout there are ftmctLone

convex on r. and we renark

ts contLnuoug on r, pathologLcal tlhat (5.r)

of a rather satlsfy

nature

which are not contlnuous

on I and whlch tems is that,

(5.1).

saya in Lntuitlve

and geonetrlcal

between x, end xrr the graph of f(x) cannot lle above the chord Jolning the

p o ln t s P r(x r, f (x r)) a n d P Z G Z , t (x r)) ; see Flgure 5.1, where m represents the

a*1

nldpolnt

f(x,

+ xr) of x, and xr.

If

the Lneqtrality in (5.1) ls reversed, we shall call


Exanple 1. We have Co,nslder the functlon f(x)

the functLo,n f(x) concave.


I:

= *2 on the intenral

--(s(o.

,[+] _[rr:n],,
", Chapter I that and for .lt and x, in I,

f( xr ) + f( xr ) 22

22 *1t*2

Lr ls an elementary

congequence of

(1 . 3 ) ln

-8 3 -

|.*r
so that (5.1) is fulfllIed,

* *z)2

t--T-J

22 a*t**, 2',
*2 is convex on I.

and the function

Exanple 2. (5.1) is satlsfled

The function

f(x)

= x is convex on the entire However, because equality x-axls.

x-axis

because tn (5.f),

for every x.

prevalls

the function,tr It wllL

ls eleo conceve on the entire rrot be necessary to discuss observatl-on that, if f(x)

concave functlons is

aeparately is

because

of the sinple It

concave, then -f(x)

convex. of the con-

ls possibLe for

a sLrnple function lntervaLs;

to be convex on certaln before

intenrals

x-axis

and concave on other lte should llke

proceedLng to nofe general

sLderations,

to gLve an example of the behaviour = sin x has the property Indeed,

Just descrlbed. -sLn x ls

Example 3.

Ttre functLon I:

f(x)

that

convex on the lnterval

0 < x S fi. x,

from the identlty

sLn x, * sln
t

. = sr-n

*1 **2 Z

*r - - *2 cos -Zx.' * xo and 0 < sln on I, it

and from the observatton follows that

that

xI _x2 cos -]-

< I

sin x, * sl-n x,

2
*.t * *l

<sin ry,

- sin

z-

t-

/.6.ln

x, + sln

2l
-sin x satlsfies equivalently

Hence the functlon

(5.1) and is convex on I; the functl.on Ilollever,

sin x is concave on I. T < x < 2n, convex (or

on the lnterval sLn x is

the functlon Flg. 5.2 -sin

x is concave); see Flgure 5.2.

-8 4 -

It every

is

clear

th a t

if

w e h / ere to

devi se

a method of

veri fyi ng of

(5.1)

fo r

given

function,

we should

be confronted

r4rith a variety

formidable

tasks,

each tailored of that

to a specific function

function

and depending on the special (or concavity) of the

properties

to determine

the convexity

given function. cient condition

Consequently, we should like that (5.1) be-satisfied. which involves

to establish

a simple suffiand the usefurness in the calculus, somewhat different

The sirnplicity elementary notions

of our first

condition,

lead us to a plan of development for from the plans of the earlier the Lnequalities inequalitles give first independently

this

chapter which is In the earlLer although

chapters.

chaptera we developed we showed how those Here we shall and

of the cal-culus, to lntegrals that (5.r)

could be applied a sirnple crirerion

as well

as to sums.

be satisfled

(Theorern 5.1),

then proceed with a development of Jensenrs inequaLLty based on (5.1). THEOREM 5.1.
a non- neg a ti v e is convex on I.

Let f(x)

be defined on rhe intervql


f" (x) everyrrrhere

I:

a < x < b and have

s e c o n d d e ri v a ti v e

on a < x < b.

Then f(x)

P r oo f. B y T ay lor fs

For any two points


th e o re m,

x, and x, in I,

rhe poinr

](x,

+ xr)

ts in I.

we have

f ( xr ) = r [+J *
w h e r e*r . L. i (" ,

["'-

_r.:l,,[1]l 2
) - l2

.l[", +]",,,o,,,
+]2r,,{*r)

+ xr), a nd

r ( x r)=r[+]

*[*'-

* t * * r J * , [ 1t _2. }, L T * Tlz j- [ 2 2 )

-8 5 -

where f(xr)

|(x,

+ xr) .V 2.

*2.

I f ne n<n add these expresslo,ns for f (xr)

and

and dlvide by 2, we have


f(x,)+f(x^)
LL

2
Noll the negatLve second gLnce

= t[+J

* St*, - *r)2[r"(:ir) + r"Gz)].


side of this so that last equatLon cannot

term on the rlght-hand f"(x) > o by hypotheels,

'F#l u
whlch ls (5.1), so that Iile note, on I, aa a corollary ls then -f(x) 0f all convex, or, the crLteria I,

f(t)

+ f(xr)

the theoren is proved. to Ttreorem 5.1, equlvalently, that f(x) Lf f"(x) is ls never posl.tLve

concave qn I. that a functlon be

givl.ng sufffcient theoren 5.1 wl-ll

conditlona turn'out

convex on an lnterval moet ugeful

to be the al.rnpLest and conslder will poseess be in-

becauee moat of the functions at least tn certain

that we ehall intervals

a second derlvatlve terested.

in whlch we shall

Ilowever, we gr.ve some er(ampleg to show the type of exceptlonal (5.r) but not possessing even a first f(x) = derlvative.

case satLsfying

Exanple 4.

The function

lxl

tffustrared

in Figure 5.3 fails

to

have a derivative property (xylr) that

at x = 0, yet hae the

any chord JotnLng two points on the graph cannot ao that the function

and xrrY2)

ll.e below the graph, satlsfles intultive Fig. 5.3 (5.1). feature

Let ua note another of convextty whlch will

pro\re useful

ln the eequel:

For each x

-86-

in I, is,

the graph of a convex function line which intersects

on I possesses a line

of support,

that

a straight

the graph of the funct.ion in one or of the graph lies below the line x, the for x # 0 of

more points support.

in such a way that

no point

rn the event that will be the line At the point

the graph has a tangent at the point of support, as in the case of line l"l

tangent line

in Figure 5.3. origin will

x = Or any straight

y = mx through tire

be a line

of support provided that m is E4ample 5. ,/ I :


I
I

chosen so that -1 < rn < 1. be the polygonal is

Ler rhe graph of f(x)

path shown in Figure 5.4, where f(x) a piecewise linear


that (5.1) ls

l,
tl l' tl ll tl

function.

It

is clear
f(x)

I
tr

sati sfi ed

and that

t r a*l*2"3*4b

.f

possesses a line pointxinl:a(x<b.

of support at every

Fig. 5.2. quite notion

5.4 The ideas behlnd Jensenrs inequality far-reaching, f(x), and are based solely

Jensenfs Inequality.

are

simple, yet surprisingly that,

on the

for a continuous function cannot lie

the chord between any tr{ro points This last as property

on the graph of f(x) of a convex function,

below rhe graph of f(x).

which can be expressed analytically

( s.2 )
for

f(ox, + (1 - o)xr) < o f (x r) + (1 - o )f (x r)


x,,x, on I and for any real number cr with O < o ( 1r must no w

any pair

be shovrn to follow

from (5.1) whenever f (x) is continuous. if ..., f(x) x' sarisfies in I, (5.1) in an interval I,

We begin by showing rhat, then, for any n points *1 *2r

-8 7 -

(s . 3)
the lnequallty directly (5.3) Ls a useful tf forn of Jeneelts lnequallty, Zk. and folLons if

frorn (5.1)
2

n ls e power of 2, say n -

For exanple,

n = 4 = 2-, we have

tirf
we now proceed, by a forn holde for any positlve let

(xr) + r(xr) + r(xl) + r(xo)1.


prLnclple*,

of the iaductlon n.

to ahow thet

(j.3) poelof 2,

iategcr

Suppose nor that lnteger such that

n le aa arbLtrary n * u ls e porer

tLve lnteger;

n be a pouitlve firen, for

say, n + n - 2J.

any polnt 1

xo ln Ir

'I
If

xtf

:"*xrr*nn^1

n +m -J
*(*,
g o r.

sfrir ( x t )

+ . . . + f (x rr) + n f (x o )).
lles in I, whence

xo =

* x, *

...

+ xrr), then xo clearly

* xo f

* (* ,

...

+ xrr)

m*n

* Ttre foru

of the prlnclple

that

ne uae for

to as ploof

by backlrard tnduciio,n, Pr, ls true for

the proof , soretluee ref erred proceeds by ehowing flrst thet a sequence many lndlces We obeerve that n and then that we have already the es-

of proposltlons truth

lnflnltely of Po-1.

of Po lnplles the truth

the truth of (5.3)

tabltshed

for

inflnltely

nany valuee of n, namely, the

powera of 2.

-8 8 -

shlch reduces to (5.3). Now (5.2) lndeed, lf follows easily for any ratlonaL value of a Ln 0 < c < 1; integers wlth r s e, then

d = r/s,

uhere r and s are positlve

.l

_ .__1' r(v 'r +

(s - r) s

,r*") _. cf(xr) z- I.

+ (1 - c)f(xr).

l{e renark be en arlltrary polnte Ln r.

that real

we have not yet used the contlnuity nunber wl.rh 0 s a S L, and let

of f(x).

Now let

x, and x, be any two 0 < c^ < l and

Let us asaune that

there exLsts a value q,_ wlth oo (5.2) does not hold,

a pal'r of nr.urbers xlrxo

in r such that

1.e.,

( s.4 )

f(cox, + (1 - cro)xr) t oof(xr) + (t - ao) f.(xr). there must exist an lnterval J about the point

By continulty \

- oo*1 + (f - oo)*2 such rhar

(s.5 )

f(x) > aof(xr) + (1 - a o )f (x r).


between the two sLdes of (5.4),

Let d > 0 be the dtfference

1..,

(s.6 )

f(cox, + (1 - co)xr) - o o f (* f )
xl as close to xl as ne please, to cto as we please,

(1 - c ro )f (x r) = d > 0 .
xj as eLose to x2 as lre pl-ease, so that

IJe nay find and rat,lonal

ct as close

-8 9 -

!'*
whence f (X')

c'*i * (1 - o ' ) *i

oo*1 * ( 1 - oo) x2 = \
of inplies

,
thar

= f (4).

Now (5.25 wtrh rational

(s .7)
and (5.4) lnplfes

f (I') S o'f ( x i) + ( t - o ' ) f ( x i)


that

-f(Io ) I -oof( *r )
whence

tt - oo) f( xr ) ,

lr(x') - r(4)l = lo'r("i) + (r - o')r(xi) - oof(*r) (r - oo)r(xz)l s a'lrr*i) - r(xr)l * tr - a')le<xi) r(xz)l * lo'- ool(lr(xr)l* lr(xr)l) s lrtxi) - r(*r)l+ lr(xl) - t(xz) lo,- ooll[rrxr)l+ lr(xr)l). l+
Each of the three ter:ns ln the Last llne of thls *L, laet inequalLty this can be nade

leee than d/6 by approprlare (5.5) and (5,7)

cholces of xl,

and ot;

irnplLes that arbltrary real

are l-n contradLction,

so that

(5.2)

hotds for

aln0<c<1. Let us say brlefly lty does not lnclude of the function what we have shorm. Ttre definitlon (5.f) of convexthe

the assumption of continuity; at the nidpoint

Lt says merely that

value

of any two points

x, and x2 cannot x, and xr.

exceed the average of the values Ihe property (xrrf(xr)) lf (5.2), on the other lte

of the functlon hand, requlres

at the polnts that

the entl-re chord between

and (x1f(xr)) asrsumesall

above the graph of the curve between x, and x, rn thls latter case, is

a in (5.2)

real values 1n 0 < a < L.

(5.1)

is a slmple consequence of (5.2):

Lndeed, the contlnulty the continuity

of f(x)

also an elementary consequence.

However, without

of f(x),

-9 0 -

lre can prove no more than (5.2) this result. Let us proceed dlrectly

for ratlonal

c;

(5.3)

Ls a speclal

caee of

to the prlncLpal

lnequality

of thle

chapter.

THEOREM 5.2. deflaed for

(Jensenf 6 Inequallty) r:

Let f (x) be a conrl.nqogs fqnctlon f(x) sartdfy (5.1) i-rr r. on Then,

ln the lntewaL *2

a < x < b, and ret

any tl,

r xr,ln

I and for any poglgLve nnoberg o1, o2, ...r

euch thet ,8.c. r, L-I 1 (n


(5.A)

t' Irlrot*t.;
(5.g) E \ -

'\ I

1-1

at f(*r)

'

wlth equaLtty ln functlon.

*z

"'

- x_r aE if ;

f (x) reduces to a llnesr

Proof. f(x)

Let ue shon that on r,

(5.9)

reducee to an equality in the case that ...r ' x_. n

Ln the case that f(x) ts llnear be-

ts lLnear

or nore precleely,

tlteen the'smalleet g(x) -,Ax f B,

anil largeet

valuee of the xrr I-

rndeed, for

( 5 . g ) g ( X crxr) -A (E crxr) +B - A( Xor xr ) + B( Ial) - Xor ( A:r t+ B)


wtrich sholrs that functlon g(x). ln the case that f(x) ls not a llnear functlon, equality prevails i. (5.g) whenever f(x)

- f,cr r g ( x r ) ,

Ls the llnear

To pi'ove (5.8) eeed by LnductLon: and let ris conslder .'', %,

lre prok,

Let us assume that k * 1 polnts xr,

(5.8) holds for n - L,2r..., \, *k+t

...,

on r and k + I posltlve

numbers orr

crn*, such that c, + ...

* ok * ok+1 - l. "we have

-9 1

k+L
L- l^I

.l -orf(x r)=(1

-% * r )

1-1
frmr the lnductlon

Sl'nce

n fffof/(f - ot*f) - 1' Lt follows hypotheale rhat

(s.10)

,1,Ffr

o,

rk r(xr)>r[ I 1 - \*r ^ r J*l , \l-=1

ot

'

wtrence, lf rce use (5.2), k+l r k c. \

(5.11)

> (r - or+r)t[rl, r={; ,lro.t,*r)

*rj * or*r r(*t*r)

' tlrlr"t*tJ'
SLnce (5.8) holdg for, n (5.8) - 1 and n - 2, the x, are equal, then equallty (5.8), clearly then, lf holds now follows ln (5.8). for rf, all n. If all

1k+1

.r

on the other steps, rre

hand, equal.Lty holds tn have equaLitl x, ...

we retrace

our induction

= *n rrp to (5.10).

Fron (5.10) we then have

wtrence lt

follows

that

*i

= ...

- xk - *k+t.

5.3. Lnequallty lnequallties that

Conne glven by (S.8) allows

. ue another unlfLed

Ttre forn of Jensentg approach to moat of the extent,

of the preeeding

ctrapters.

I{e have shorrn, to a great and ge@etrlc

the lnequallty

between the arithnetl.c

meens nay be used

-9 2 -

to derlve

nany of the prlnclpal lt

lnequalLties.

Ae we mentl,oned Ln the

rntroductlon, matheftrtl"cal lnsight,

ls not wLse to abandon the many approaches to a glven for one then loses both lntuLtion LnequalltLes treated and in this

problem or concept,

and, in the case of the elementary among the varlous we shall

book, the interplay rn thls widely sectlon

inequalltles. in the forn of exereLses, how

show, nostly

Jensenrs inequality

nay be used. = l*|

ExanpLe 6. I: -6 ( x ( -, -

The function and its

f(x)

r"

convex on the Lnterval llnear pleces, I - x for (5.8), or ln

graph consLsts of tro

0 ( x <

and y = -x for -@< xS0. fotm (5.3), shows thar

Jensenrs lnequallty

the sinplest

t,lr'rl*il' lrlr"r*rl
ln

n 1=1

l I* ., rL=l Equallty for holds only lf all

l"i I
are the sane, or, all the xi lie and this shows the need of

the xi lf

the condLtlon

of lLnearlty,

on one or the other

the two linear

pleces

of the graph of

l"l.
=*Pfo.0<x(-r ls p )0, wemust

ExaTple 7. distlnguish

For the functionf(x)

two cases, p > 1 where f(x) is concave (lf The slnple forn

convex, and p ln the range Ls lLnear and we have lnplles

0 < p < 1 where f(x) nothing that to prove).

p = 1, f(x) (5.3)

of Jensenfe lnequaltty

- 93-

( 5.12 ) and

P>l-'

(s.1 3 )

r
only lf
t.....-t

1+

... + x \ D o l'
)... = *oi

*[.i.
thls

..

..

{],

o<P<1,

rtlth

equallty

x, -

ls a stronger

result

than that

of ExercLse 1.25. F(x) - *P ls

For negatlve vaLues of p with irnplles that

x > 0, the functlon

convex, and (5.3) ...

( 5. 1 4 )

rx- t I r

* x rD nl'

I
only if

n
x, a ...

)a xn.

*["]

"'

**:],

P<0t

with

equaltty

For p

- -1r

(5.14) glvee us the

lnequalLty Sectlon 3.

between the ar{thnetlc If we apply the firll

and harmonle means dLecuesed ln Ghapter 2, force of Jengenrs lnequality three Lnequalitles (5.8), for we have a,

the followlng n wtth Ec.-1: 1-1 1

r\redghtedil forms of the last

posltive

( s .ls )

[,i,"*J' ,lr"'*l
[,if,.J' ,lro'*l
n n

p>0,

xl. : o;

(s.16)

0<p<

1,

*i t

( s.1 7 )

[rlr'*']'
only tf x, a ...

,lrot*l
r x. n

p<0,

*i t ot

wlth equallty

\;

-9 4 -

Exercises

5.1.

Shon that (5.3),

the funetl.on

f(x)

log x Ls concave on 0 < x

< o, and u8e

aa nod{fled

for a concave functlon, end geonetrLc means:

to prove the tnequallty

between the arlthmetlc

( 5 .18 )

(xrx, . r. -*o)t/o.*1

**2*"'**o

5.2.

If a1r ...r *1, "', ylelds

I, and if. rlrcl x, are poeitlve nunbere, show ttiat (5.g), apprled to log x,

oo are poeLtlve numberssueh ttrat

( s.t 9 )
(lhts

ala2 *L *2 lnequallty

'"

% xn S o 1 * 1 * o 2 * Z + . . . o o " o . calred the raaighted forn of the and geometrlc meane.)


p and q are poeltlve tn problem 5.2 lnplies

Ls frequenrly

lnequallty 5 .3.

betweco the arl.thnetic


and lf (5.19)

If x and y are poaltLve, tlP + I/q - 1, show that

nuders

such that

(4 .2) of Chap-

ter 4, namely, ( 5 .20 )


__p __q **., < + Y' r ^'=p'o tool ln the proof of [iilderrs laequaltty.

whlch was the prLncipal

5,4.

Extend the result

of Problen 5.3 to the case of n poeltLve

nunbers
nl

* 1 , ...r so that

xn and n posltlv e n u mb e rsp r, . . . ,

pn such that

* i, i-

t,

(s.2 1 )

-P r

< -3 -+ s +

*it . *;'
Po

.n

Pn
n

-9 5 -

5.5.

Use the facc that xlogxls a>0rB ;' 0r

cdl vexforx> 0to

show that,

for

5.6.

Extend the result *1,

of the precedlng

problen pr,

to the following: pn be posittve

Let numbers

... r x,, be posltlve ;pl =

numbers, let

...,

such that

Shoi that DX -nn n.

(s.2 2)

(nt*t + . . . * orr*,r)p1"1r'''*p"1"s *P1*1 ...


a > 1 and B > 1, show that

5.7.

If

oG sGs te'-#)
5.8. Ict xrr xrr ".r xn be posltlve . numbers lees than ?r. (x- + '.. Show that

sln x, stn x,

sin xo < sLnnt-

f x r

5.9.

Extend the inequaltty f(x) "1, be posltLve "'r

of Problen 5.8 to the following I:

resuLt.

Let

and co-ncave on the Lnterval on I. Show that f (X, + "'

a < x < b, and Iet

*D be n polnts

f (xr)f (xr)
or, what Ls the sane thl.ng, f( x r), ... r f(xo)

f ( xr r ) S
that

[f

ill",
value of the arithextended? be further

"
mean of the values

the gemetric

does not exceed the functlonal ..., xn. Can thls inequality

netlc Eean'of xrr

-96-

s . 1 0 . For p > 1 and q > L wlth l/p + Llq=


lnverse funetLon of f(x) - *P-1 for

1, show rhat g(y) x > 0, y > 0. If

yq-l

ie the

A1 ie the area

between the graph of the curve and the x-axl.e between 0 < x a a, aqd Lf Az ls the area between the graph of the curve and the y-a:d.a be-

tweeo0<y<brshowthat

ab< Ar-*o, =
wlth equallty tf aad only if

rP

t- ,
Note that thls glves another

bq

aP - b9.

proof of (4.2)

tn Chapter 4.

5.11 .

Develop the Ldea of Problen 5.10 in the foLlolrlng f(0) f(xr) - 0, be a contl.nuous increaslng < f(xo) for x., ( x,,, so that function f(x)

way.

Ler y

f(x), Ls,

of x for

x _z 0, that

Lz LZ

hae an Lnveree functiqr

\-,

x - g(y), g(0) a 0, deflned on eone interval 0 < y < Ior where yo *y be :to. Define A, and A, as or f(x)dx, Az g(y)dy, and prove - {" f that

S Ar * A2 - J f(x)dx + J g(y)dy, "5 oo


rl.th equaltty only lf b - f(a). The result is often called youagfe

:lb

inequallty.

5. L2.

S ho d th a t,

fo r

0 S x < 1, ?T ;X < XS 1n z_1

**R
*5** oa xi0. Evaluate

5.13.

Let x-

g(y) be the lnveree functLon of y

rhe lnres:*.t

f,z 8(r)dy.

.:

-9 7 -

5.L4.

Use the fact for positive

that

f(x)

= log(1 + x) is concave on x > 0 to show t hat, dZ, "', dn, + ar) .ir (r + an)' .. ^2r " "rr.
s hor , r

numbers al,

(1 + A)t > (t + ar)(l \ where A is the arithrnetic

mean of the numbers arr

5. 15.

Us e th e t h a t, fo r

fa c t

th a t

f(x )

= 1og(1 + .I)

is ...,

convex 4n,

on -co ( x ( 6 to

p o s i ti v e

n u mbers al , & 2,

(1 + ar) (1 + a r)
wh e re g is th e g e o me tri c mean of the

(1 + a rr) > (1 + g )t ,
numbers a1r "', ^2,

" r,.

5.16.

Let ar, Show that

"',

ao and br,

"',

br be two sets of positive

numbers.

k=lK

* K * il .11" ftor/". il c*- b,-)u'. K


k= l K k= I

5,L7.

Let a1 , "'.r let

ao and brr

"',

br be two sets of positive

numbers, and

B" be the geometric mean of ar, "', bo. Show that


r rn (9" *, Bb)'^ .

ar, and 86 the geotnetric mean

of br,

(.t

+ br)(a,

+ b)

(ar, * brr).

5.18.

Let y = f(x), x > 0, and let f(x),

f(0)

= 0, be a continuous increasing defined

furtction

of x for of

x = g(y) be the uniquely *. Let E(x,y),

inverse

function

0 < y < yo.

p(x,y),

and q(x,y)

be non-negative Show

functions that, if

defined for F(0,0)

x ) 0, y > 0, sueh that I O, then

6p = pdx + qdy.

= 0 and.t*, ,a1b

p (a ,b ) s

oo

J p( x,f( x) ) dx + J q( e( y) ,y) dy.

-98-

5.1-9.

Ict

y = f(x),

f(0)

= 0, be a contLnuous increaslng function f(x)

function

of x for

x > 0, and let inProblen5.11. to obtain

x = g(y) be the inverse Rotate thecurvey= of rotatLon

of f(x),

as deeerrbed 0andxa

betweenx=

a solld

of volume V* and let x = g(y)

T be the x-coiirdib.an ".o

nate of Lts centrold; y

and rotate a solid

about the y-axi"

0 and y = b to obtaln fe y. Show that

of rotation

of volune V- whose

centrotd

I"'o ' . iu* * f v y *


5.4. Inequall'tles for a triangle. One of the most Lntereeting the derLvation of certaln

ele-

mentary appltcations lnequallties of a trLangle.

of Jensenrs theorem'ts xr, *2,

ln the case that I{e avold

x3 are the meagures of the angJ.es l-nequalltLegrr,

the use of the words I'trlengle context Ln natheoatl.cs.

whlch are used in another

E?<ampl.e 8. expresslon we know that

If

xr,

t2, x, f

x3 are the angles of a triangle, sln x, for

examl.ne the Now

sLn x, * sin sLn x is

posslbl_e maxlma or mlnlma.

co'ncave for

0 < x < Tr, so that

xrfxr*

"1"I
if x, = *2 = xr, 1.e.,

3l

]t"in

x, * sin x, * sin xr).


* x^t1^1

l l e n c e ,sl n xa * sl ,n x, * si n *3.3

"t"F

#)'

wlth equalLty only


trlangre. But

only in the case of an equilateral equaltty--the

*1 f *2 * *3 - r' 3 sin Tt/3 - 36

so that 12.

case of a rnaxlmun--givee

the value

-9 9 -

rf ry2. expresslon

*1,

*2,

x3 are the angles of a trlangle, posslble

examlne the since sin x

sln x, sin x, sl-n x, for

maxLma or mlnlna. extenelon,

ls concave' lte nay apply eLther Problen 5.8 or its to obtaLn

Problem 5.9,

stn x, sin x, eln x, <


with equallty, and therefore

"t "1o3

* *f * *1

16 la , "in3 [ ntrlch 1e agaln

a maxlmum, whelrever *1 - *Z - *3, trlangle.

the case of an equllateral In problens of thls

klnd,

lt

frequently rre see this

happens that occurrence

one muat restrlct ln the next exanple.

the range of the angles

Lnvolved;

Exanple 10. exptessLon

rf

*1,

*2'

x3 are the angles of a trlengle, poesible

examlne the I{e

tan x, * tan x, t tan x, for tan x ls

naxima or nlnLma. 0 < * < nl2,

poLnt out that in ordet that

convex only on the Lntenral intenal,

ao that,

to restrlct

xl , x2, x3 to thls Ls acure. By (5.3)

lre assume at the outset

the trlangle

we have

xr*xr*x" tan whence 3 tan $. -f-:

1 < j(tan

x, * tan x, * tan xr)

t"tr x, * tan x2 + ran x3, x, - x, = x3r 1.e.,

wlth equallty

onl,y for

only for

the equllateral

trlangle,

ln w?rlch case the expreeslon

tan xl + tan x2 + tan x, has a mlnlmum value

36.

-1 0 0 -

Exercl,eee

5.2O. Let x1r

n'

*3 denote the angles of a triangle.

Flnd the uaxluum or ytret further restrLctlong

mlnlmrm of the expreesLon glven, and lndlcate


osy be neceasary. (a) (b) (c) (d) coa xl + cos x, * cos xr. coe xl tan xl sec xl cos x2 cos x3. tan x2 ten x3. * sec x, f edb xr.

(e)

tan xr/2 + tatr xrlz + tar. xrl2.

5.5. propertLeg

Ftrrther

DgvaloF@te. cdrvex

Wc wlch functLon8

to point ntrlch

out sme

geoetrlcel

of contlnuous

relnforce

our fntultl.on

about such frlrctlons.' use' are typtcal

ftre requlte

we obtatn,

ag well

es Lhe aethode we of one varlable are

of the way ln wtrlch real

fuuctlons

treated. we returrr three polnts to the georetry of Flgure 5.1 and let If xl, *2, *3 be any

of I euch that
x a -x l

x, . *2 . *3.
xt-xr

we uae the ldentlty

;^ - #*z x3-xl
where o1 * o2. functlon f(x)

1*

x3 -x l

x^ J

A rx , * C2 * 3 , (5.2) for a contlnuous cmvex

1, then Jensenfs inegualtty ylelds

r(x2) "ffit(1).ffir(xg)
or, after nultlplylag by x3 - *1,

-1 0 1 -

( 5.23 )

(x3 - xr)f(xr) < (x3 - *2) f (xr) + (x 2 - * t )f (x r).

If we subtract (x

I - *t)f(xr)

from both eides o f (5 . 2 3 ), we h a v e < (x Z - x a )(f (x r) - f (x r)),

(xl - *r)(f(xz) - f (x r)) or

f(xr) - f( xr ) *2 - *1
If noy we subtract

f (x r)

- f (x r) * 3 -* L

(x3

xr)f(xr)

from both sldes of (5 .23r, rrc obreln - f (x r)),

(x:

xr)(f(xr) - f( x3) ) < (x 2 - * t )(f (x f )

f(x r) - f ( x r ) s f ( x r ) - f ( x r )
*3-o 1
and 1f re conbLne the two lnequalLtles

*3- *2
involvLng

'
guotlents, we heve

(5.24)

f(xr) _ t(ltl
x24

f(xr) - f(*r)
* 3 -* 1

*1

f(xr) - f(xe)

If

we lnterpret

(5.24)

geometrlthat

cally,

we 6ee (Figure 5.5), Ls nothlng that

I
I I
I

(5.24)

Eore than the

statetEat

elope AB < alope AC < elope BC; so that, as x2 lnereases wlth x,

xt

i-r!
t'zJ

ta

I I
xl x^

held ftxed,

eay, the slope of the increaselenentary

chord AB ls uqrotonlcaLly Flg. 5.5 lng. Non we know frm

-L 0 2 -

analysis lirnit. written

that

a bounded, monotonically x for

increasing

function

must tend to

Let us write as

x, and Ax = x, - x, > 0; then (5.24) rnay be

f( xr ) - f( xr ) *3- *1
where ing s ide. the left-hand of side is, is as we have remarked bounded by the flxed

a monotonically quantity

increas-

functlon Hence

Ax which

on the right-hand

,r* Ax+o exists

f(x)

- !(x Ax

- Ax)

(Ax > 0) left-hand derl_vative f '(x) of

at the poLnt x; this In a simil-ar

is

the so-called

f (x) at x.

fashion

roe may shohl that

the linrlt

,. Ax+o exi-sts at the point f(x) at x. x; this

f(x+Ax)-f(x) ax

(Ax > 0)

is

the so-called

righr-hand result.

derivative

fl(x)

of

Thus we have proved the following

THEOREM 5.3.
I: a <x<b at

A continuous conyex function


deri vati ve f' (x)

f(x)

in an interval
deri vati ve

h a s a l e ft-h a n d

and a ri ght-hand

r l (x)
T

each point

of a < x < b.

Furthermore_, f t(x)
a ( x ( br = ff(x) .

< ff(x).

We remark that is differentiable

i f,

at

a poi nt f ' (x)

x in

ft(x)

= f](x),

then

f(x)

at x , w i th

= f ' (x)

Our observation
suggests t hen ( 5. 24) f or that

concerning

the slopes of the chords in Figure from (5.24).

5.5 *Zi

we may obtain

even more information

Let *L.

s u ffi c l e n tl v

snall

Ax > 0 we have x, * Ax < x, - Ax, so that

frorn

we have

-1 0 3 -

f(xr+A x)-f(x r) Ax
I n' t he limit a s Ax -+ 0 w e th en have

f (x r)

- f (x 2 - A x ) Ax

r l( x r ) < f ' ( x r ) ,
and, since f'(xr) shows that
The right-hand non-decreasing is tru e of th e

< fl(xr)n

we have fl(xr)

< ff(xr)

whenever *1.

*2, which

derivative
fu n cti on l e ft -hand th e facts of

f I (x) is a monotonically T
x i n -a (

x < b; the same f r(x) .


in the di scrrssi on above i n t he

deri vati ve obtai ned

We s u m m a ri z e s o m e o f form of a th e o re m.

THEOREM 5.4.
l_n I: a- line L.

the graph of a continuous funcrion

f(x)

satisfying xo in I;

(5.1) is,

a < x < b possesses a line L: y - f(x^) = m(x - x^) 'o-o

of support at. each point such that no point

tha!

of the graph lies

below

Ttre slope rn of the line of definiteness, usual derivative to be f'(xo) ftf;f".1

L of Theorem 5.4 may be taken, + f '(xo)1, so that rn will exists

for

the sake the

coi.ncide with

whenever that

derivative

(see Problem 5.23).

Exercises

.2L.

Show that function x,

a necessary and sufficient f(x) be convex.on I is that,

condition

that

a continu.ous *2, .*L,

for any three points that x, < x, ( xrr

in the intervall:a(x<bsuch

-1 0 4 -

*r. f (xr)
*2 *3

t 1 I > 0.

t(xr) f(xr)

5.22.

Prove that used in thls the rlght

the rlght-hand

and left-harrd

derl.vatlves that ff(x)

fi(x)

and f f (x) from

sectLon have the propertles ft(x) is

ls continuous

and that

contLnuous from the left. x on I where f t(x)

5.23.

Show that

the set of pointe

# fi(x)

ie denumerable.

l{e present ueeful context powetful

next a claesical I{e shall

Lnequallty show that

ritrtch hae proved to be very l.ts Lnterpretatlon ln the

ln nany contexta. of the theory generalizatlon

of convex--and

concave--functlone

can lead to a

of Jeneenfs theorem. and non-negatlve for a 1i x < b. Then

Let f(x)

be contlnuous

( s.2 5 )

b-a

Iororf(x )d x < lo s [ - * a
for f (x) constant. l-t frm

, f ,,.,u-]
I,Ie reproduce the slnple claesl-cal of. convex - n, of vLew of the theory and then rrlte

with proof

equallty

only

Ln order

to analyze

the point 4b f{*)a*,

functlons.
t

FLrst,

we set r 6-* Then

g(x) - f(x)

so that

{"

S(x) = O.

b-a

tb 1 J 1og f(x)d x = b -"


I -- b-a

1b J lo g (n + g (x ))d x
a

1b J troe n + tog(l * Ekllla*.

-1 0 5 -

If

we apply

to the 1a8t term the lnequality

( s.2 6 )
with equallty only for t

log (1 + t )< r,
0, we have

rfb . lfb tP ." < b - a l-t"t r(x)dx los' + #; lIt P ) - los n - loe{5ft- J r(x)axJ,
h

slnce

{-

e(x)dx'

O.

Equallty

holds only if

g(x)

0, that

ie,

only if

f(x)

ls the constent m. ltre analysla we observe that tloo y of the proof (5.26, turns on the alnple than the fact inequelity that (5.2d), ead

le nothing

rore

the concave func0i is slnce slmply the

1og(1 + t)

has a line

of support -

et the poLnt t of aupport

log(1 + t) tangent tlon that

hae derivatlve y - r,

1 at t

0, the line

ltne

and t\e

Lnequaltty + t) lles

1og(1 + t)

< t ls elnply

the aaeery - t.

the cunre y = log(L

below the line functlon f(x)

of eupport

suppose ndr that a contlnuous byyg(x) f(x) - f(x)

we have a continuous F'(y) deflned (b -")-t - 0.

on a s x < b and acsrged

concave functlon ona< x<b.


lr

on the range of valuee {o f(x)dxendlet is coilcevc, it

Lern=

- n, so ttrat d" g(x)dx of support all at y : n,

slnce F(y)

vtll M guch

have a llue thet, for

that '

La, there

1111 exlat

a nuder

y deecrLbed above,

( 5 :27 )

r(y) < F( n) + u( y - n) .
(5.27, and dlvlde by b - a, we have

If

we lotegrate

-1 0 6 -

1 P -,M fb b _ a l t(t(x))dx < F(n)* fr - a J (r(*l - n)dx A

F(u) +

vP fr

J 8(x)ax- F(n).
theoren, Ln the

I{e state forq

our reaultr

wtllch lc an lntegral

form of Jelrcente

of a ttreorcn.

lItEORElf 5.5. conttnuqrs ltren

Let f(x)

be eonrtnuous gg e < x ( br aad let such that F(f(x)) la deflpd

F b-e a

concave functloa,

oo a < x < b.

(s.28) r+
If F 1g convcx,

fr.r.*r)dx< r[*
a thcn

. Jor<*lu*J

(s.zs, r*
Equallty

> . f.,r,o)dx r[+= fr,.,u.J


f(x) is coostant.

holde onLv lf

Exerciges 5.24. Let f(xry) D-:


f-n

be coatlnuoue

and poeLtlve

in the dlee

(x - x^)z + (y - y^)2 , t2.

strow that

II

t" r(x'v)dxdv <lost# r(x'v)'r'dv] ' 11

- s - lSOLUTIONS Chapter I

1.1.

If

we dl.vlde both sides of (L.Z) by 2 and add ("2 + Az)lt, to both we obtaLn * b2 , * b2 * ab "2 ra.1 !)2

sldee,

^2 T.T

-T=

l- - ) ,

whtctgupon taklng holds Ln (1.2) tt

the square root, and only lf

yields

(1.3).

sLnce equaltty true in (1.3).

a = br the same is

L.2.

(a)

I,le have (
I

(2a - b - c)2 > o

(2b-c -a )2 ro
(2c-a-b)'20r holdlng a +b, everynrtrere if and only If tf 2a b * c,

t,
t

wl.th equallty

2b n e * a, 2clnequalltles,

or a - b - c.

we expand these of the forn

we obtaln

three inequallttes

4"2 * b2 + c2 2 4ab r 4ca - 2bc. Then addlng the three by 6 glves (1.5). inequalltles thus obtalned and d1vldlng

(b)

We have

^2 -2 ^.+.2

= ( a- r .l2>( T) 2 = 1( a - b) + ( b - c) f '2
-ab+bc-ac -b2,

> G - b) ( b - c )

tr t

a'+b'+c'2abfbc+ca,

- s- 2 lrlth eque[ty or a - b - c. L.3. B y (1.6) ana (1.1), w e h a v e

\-/:

if

and only if

a : e = 0 and a _ ! c b _ c,

"3

* b3 + c3 >

"b

(s!+i) * o" [Lfg


+ ca6

""(gji)

> ebfifr + bc6

312+ + @a) 312, - ( 6s7312 ( ca)


with equaUty lf and onLy if a - b - c. tng ,312, b312, .312 fo, a,b,c

Alternatlvely, ln if (1.5)

etrnply edbarltut inequall.ty,

gives the decired

wtrtch reducee to equallty

and onry if

a3l2 - v3l2 - "312, or I - b - c. frorn the hlnt. Equallty holde if and

1.4.

The proof onlylfa-b-c.

ls stralghtfomard

1.5.

If

we nultlply

the three

inequalltles

arb
whtch cme frm

> zffi,
(1.1),

b r c > z{fr,
we obtaln

c *a > 2fi:a,
Equallty holda

the deslred result.

lfandonlylfa-brc.

1.5.

Flrgt

solutl.on. .f br ab;a r--)

Fron Bxerciee

1.4,

we have

rb + + uc1-2-Jc1 r ^^ f9l_e1 + ca(-fJ by Zlabe glves

> 3abc.
reeult, wlrh equallty

ltulttplytng

thls

the deelred

lfandonlylfa-b-c.

Second eolutlon.

Notice

that

by (1.1),

for

all

x >. 0,

x+ l> 2,

-s-3-

nlth

equaUty

if

and only if

x = 1.

Therefore

+*+*gFwlth equallty if Third Solutlon.

(i.fl * L!*os)("L+oe) + 2 6,
and only Lf 3 - : - * - 1, or a - b - c. and then the inequall.ty We have We shall_ use (1.1)

deduced at the end of problem 1.4. a*b C b*c a

"l"b

:' rZ6+2lfi-rcn)6 C -tT

t,-6-

>z.4$s..F'l]-t3- 6.
L ,7. Multiply the fr.rst rnequarlty tn (r.4) by c2, the eecond by by b2. upon addltlon of thege new lnequarltlee end the third "2, and

dlvisLon by 2, the deetred reeult r.c obtalned. andonlylfa-b.e. 1.9. Flrst solutlon.

Equality holde lf

Dlvlsion by ebc of the inequaltty

in probla

1.7

gLves thls lnequality.


seco'nd solutlon. have SL * !" (see the second solutr.on of problen

r.6).

we

;-;* T

. c8

ll-ra h a., -'l + c.,- th .r\ - tL t G ; J b* t : * f , J . + + f l: J ) a * b * c ,

wlth equaHty lf 1.9.

and only lf

a - b - c.

As we sau ln gxanpre 2, a3 + b3 > ab(a t b), wrth equalr.ty tf and only lf a ' b. rf we n'rtlply thle by 3 and add a3 * b3, we obtain
.. ? ? 4(a' + b-) Renark.

> (a + b)J, "

dB deelred.

I{e have shown that

a3 +u3 = (. t u ) 3 ( + ) , -

G * b ) > a b ( a+ b ) ,

-s-4-

using (1.1).

Ttrus, we have ended up w"lth an inequallty

etronger

than the one we began wlth.


1.10. (a) By (1.1), we have

( 1)

f f i' r G + c )(b + d )
holdlng if and otrly if

,
a f c-bid.
at Ifwe the deel.red

wLth equellty

take the squere root and dlvlde by 2, ne arrlve lnequallty. (b)

rn (1) above, tre grouped a * b * c * d as (a + c) + (b + d). If we lnetead group it and use elmilar lnequalltles
?t

as (a + b) + (c * d) or (a + d) + (b + c) then additlon of theee three

lnequalltlce,

gives
> 2(ab f bc * cd f da f ac f bd), lf a * c - b + d, a + b - c * d,, Now dlvldlng ineguallty. by 12 and

f,(a ltlth

+ b t c + d)tf

equaUty

and only

r b and a + d - b f c, or a - c - d. taklng square roota glvee the deelred

1.11.

suppoee that the sequeneesare r.ncreaer.ng. then {brcrrbrcr} aleo increasing; note that positlvlty generally. Ttrus (1.7) glves

te

ls necessarT for thls to hold

Applylng

(1.7)

again and notlng Equaltty

that

+ lf

is poaltlve, and only ie eaaily tf

we get

the deelred b1"1 - b2.2, ls true lf

lnequallty. and b,

holde It

a1 r a2 or thle

- b, ot "l all

- "2.

eeen that

and only lf

but one of the sequencea {arrar},

{u rru r}, {c'cr}

i s consr ant.

I
_s_5_

1.12.

The generallzatlon

of Probleo 1.11 le that for n increaalng (or k ...,'e - 1, we have

decreaelng) posltive sequencest1Or.*r],

I I

a11e2r"'anl a*arr"'ao, I --l-T)tT-J , ['{u1 2


elth stant equaltty tf and only tf all (see Exerclse

ftt
'l

..}3]... ...t-t-J r."{5al. .

I I

I
but one of the Bequencea Lg couproof, in fact, of a more general 1.24 for a full

-l

tnequallty). have

Therefore, if

arbrxlr"'rro

are po'ltrvc

cumbere, we

I I I

a-r'-'ts n+bFl

It*"'fU-

I r* ' . .-*" y F n = i" - t " - !f I r - u-

nlth equaHty lf and o n ly L f a - b. ,/ . . taklng n.h roote glvea

..f"\*] -t--J ."'t--J

lrrl

l-

I I
: il
il

L e t t ln g I , - r - . . . rU' n - 1 an d

rl I l i

I I iI

a f&-b"lr/o,'T'* b

f-7-l

rl

1'13' Ler ue denote(i#tt' vioua probrenr'

bv a and laplt

bv b. rrrenbv the pre-

'l iI

.6* b 6={ t . * u) 6- #, J 3
2'
wlth equauty lf and only lf a - b. rn 0 s t s 1, thls 1g equlva-

I I

I
tl

rl

leattoL-t2-2t,ortrfr-t 1.14. It follorr frm probleu 1.12 that

il

| I

iI

T .t_ _ _ z _ J
with equaHty lf and mly if

-24 , .2n a--+b-- -

1-2 E lu,2 lo _ r -21'.


a - b.

- i ,.

:I I

ll
ll

l
ll
I

I I

;l ;l

rl 'I

'l I

-s-6-

1.15.

UeLug problenr L;L2, we have

1 + tan80 = 14 + (t"rr20)4 t except wlrere 1 - tan20, or 0 - nl4 1.15. rn the second ineqtiality

*<t 2'

+ tan2e)4

- $ "."80,

of the eolutlon to Exercrse 1.12. ret

n - 3 and U, - 1, !2. - 2, lr3 - 3. Ttre deal.red reEult folloss {mnediately. I{e have equallty lf and only lf a - b.
1.17. Let d //x'+ yt be the dlstance frm the orr.gfn to a polnt on

th llne.

Ttrenby (1.3),

d >#r-*,
trfth equaltty !t:( y - 3lz. That is, the nr.nlnr.n dlgtance ls 3ltz.

1 . 1 8 ' L e t d ffi arc.

b e the dletancefr m the or tgtn r o a polnt on t he

Thanby uelng (1.3) trrlce, d zx+Y , ( E+' i12 j-

AzfrzF

wlth equaltty 1.19.

at x - y - al4.

The mlnimum dlatance re a126. ar)(b, - bz) s 0, wlth equality

under the hypotheses, we have (at *f and only lf


ln the text

"L - "2or

b, - b2. proceedlng wlth the ergument


wlth the oppoelte lnequality slgn.

ehons that

we get (1.7)

L .20 ,

(a)

rf d - /*2 * f;7
on the plane,

is the dlgrance frm


we have

the orlgrn

ro a polnt

thea by (1.12)

u 2 r(x+v+ z) 2- 25
rrlth equallty at -'7, y x z - Slj. The nlnlnun dletance Ls

therefore

516.

-s-7-

(b)

We now apply (1.12)

twlce:

,2, (x + I + z)2= <G* E* G>a__ 4


The nlnimum dLetance ie thue a136, L.zL. the lnequaltty (1.12) y1elda achleved at x - y - z clg.

r(x,y)$["r"2* = . [#]'
- t' ' 1'

"o,r*.[j?J' .o"rl
'"o.r* - (-2r")'"o"r* LtY'

ft" roz"+cos2x)'-t,

wl.th equallty ri md only 1f eln2x tf=41 orel n* '* , c o' * - f tvrr- y n-\.*'{5 L.22. If we apply (1..12), roe obtaln \ f (x) >
1 7 rn

itr '

+ c o s ' x + s in a x )a

- 413, elnce equality rrould

However, 413 te not the mLnlnumvalue of f(x)

. sln x, whlch le inpoeslble. requl.re 1 - coe x Tttl's exmple should aerve to forewarn the reader that in looklng for extrema by uslng fnequallties, lty L.23. ls actually attalned. '.., bo} be two increaelag sequenceg one must be Bure that equal-

Let {ar,

"',

8n} and {br, Then, lf (ar a , )(b l

of real nunbers. (1) or.

t < J, - bJ) > 0,

" l b i r" JbJ=tr o,*"Jb, If ne sun all such inequalLtles and then add n a,b, to both sldee E
l rl rr

of the reeultlng

lnequalltyr

rre obtaln

-s-8-

I tro, = L -l .f,qrb,,f. J -l t- r1
2
clearly

nn

[,!, [,1,"J 'J


Convereely, lf equality In (1) for alL l < J. Therefore "1 - "r, one of the

Dlvislon conatant, holds ln particular,

by n- produces (1.11). then equallty (1.11),

Now Lf one of the sequencee is holds. holde ln

then equality holds for

equality

i = 1 and J - n.

or b, = b_, but slnce each sequence ls monotontc, Jngequences rnust be constant. If stlll the ttro Bequencee lrere both decreaeing, Bo that (1.11) ls valld. If

then (1) would

hoId,

one sequence le increasfn (1) le

lng whtle revereed, revereed.

the othepLe so that

decreaslng, ls valld

then the LnequalLty wtth

(1.11)

rhe sense of inequallty

L.24,

We shall show that tf

{arr,

...,

.Lo} are Lncreaalng (or elee dem, then

creaslng) sequencesof posltive numbersfor 1. 1, ..., (1)

n a .1 I irrl, r,.I,..J It -l
lf and only lf

rn m nmm.n
IF

> II r I n o
,-r t=l

rl Jlt "tJ

lrlth equauty constant.

al-l but one of the sequencea le

We ehall ptoceed by Lnductlon. ln (1.11). {.rr}f-r, Slnce rP -l

I{e have eeen (1) for m 2 -

Ttrus, assume(1) for m - p - 1, p ) 3, and let 1 - 1, ...r tb p be p poeltlve lncreaelng sequences.

trlr"*ljr_l

l" an lncreseing sequence, (1.11) glvea us (t


=

(2 )

J'l

if; 1-1

ttj

bt,1,,Ir"rrj ,t-, ht "nr]'

P -l

lf r

wlth equallty tf and only tt

fP-l

In

trlr"rrir-,

or t"o,)|-,

r" consretrt.
i" llr"n:

Applylng the Lnductlon hypotheeis and obeenrrng.that posit{ve, we have

[* ]i."J[* = *,1,.,,) ,i, ,i,r,]t:rl F,i,r,]


_ II + tr.*. , n a: 1-1
Jlr wlth equallty 1 - 1, ...r lf and only tf all are conctant. bur oae of the Bequenccs t.rr)|-r, Conblning thla wlth (2) ylelds coodltlon. p - l,
prtl

(1), together wl.th ih. ptop"r equality L.25, t{lthout lose of generalltrr (othernlse, s)tmetrlc). rye sinply

! aay aasrse that ths

relaber

= = ... s 8n "L "2 errd note that the reeurt Le of Chebyshevfs lnequaltty

Then, by the generalizatlon

glven ln Exerclee L.24,

where r*e have let all followg. Eenark. L.26. Flrst solutlon.

m sequences be the sane.

ttrus the reault

Thre reeurt w111 be generalrzed in chapter 5. Let x > y and z < (1 + ,o)t,
n

ylx.

Then we rleh to prove

that (1 + ,t)t

o"

nlrtl'

"*.Aq, "*

- s-10 -

Not oaly doee the rlght-hand left,

slde contaln

more terms than the tetm

but each term Ls aleo greater Hence the lnequallty

than the correspondl.ng ls true.

on the left.

Second eolutloa.

Let t be such that

(t)n

(IJt

t.

Then slnce

x,y ) 0, werravelfyn . (rJo . t, ttlt . <"J" . 1, and

ttoJt* t?'t".
uulttplylnB L.27.

t(t)" * ()"1" - 1- t(rJ"* (IJolt.

by tmn gives the deeLred result. wl.th equaltty for I b, we have

Slnce a * b > 26,

1t + xr)... (1 + xo) ,z JfL)..


* 1 -...rx o t1 .

. (2Fn)

- 2D, wlrh equallty for

1.28.

By the

aame method

as ln

the

precedlng

problem

we have

(*1 + *p1)... * *oo)> (2/6F;).'..12ffiJ (*o


Equallty holds for x, t ... = xo - 1.

- 2n.

L ,29 .

(a)

Let x * y - c, a constent. equallty x - y (et x - y. cll). c.

Ttrenby (1.1), xI 5 .214, tLrt

In other wordg, xy ls a oaxlmun wten

(b)

Let xy' x - y.

Ttrenby (1.1), x * y > 2G, wlth equallty for

Thue the nLnlmun value of x + y ls achleved when

x - y (- lf).
(e)

The proof ls the aame aa ln part (a), exeept Ehat ne uae the Lnequallty Oitg of (1.1). = 3tffi of Exerclge 1.4 ln place

s-11 -

Remark.

Parts

(a) and (b) are referred (c).

to as dual properties.

There is an obvl.ous dual to part

1.30. To maxlnlr" *ffi


Regardlng *2(to constant, - *2)

,t

to naximlze a product

lts

square, *2(t6

- *2).

"" 16, we see that

of two factors

whose sum ls

the maxlmum occurs,

by Exercise L.29(a),

,nhao *2 = lb - *2 or. * = ZfZ. then 8.


Alternatlvely, we could

The maxlmum value

or * R

have u s e d (1 . 1 ) d ire c t ly :

x{L6 - x' =

t-

= *2 t( tg - *2 ) - r . -2
the same solutLon, ae Ln the By

1.31.

There are tno ways of expressing prevLous exerclge.

Here, however, we present

only one.

Problen 1.4, we have

= Ktlz ft*trrylti)l = *(=Ilgit


wlth equaHty tf and only lf x = 2y - J2(= 2),

= L, ,
or * = 2, I - l,

z = 213.
1.3i. Let x,yrz tex. be the lengths of the edges emanatlng from a flxed By ver-

The volume le then xyz wLth 4(x + y + z) constant.

Problem 1.29(c), when the flgure

the volume is a maxinum when x = ! = zs l_.e., ls a cube.

1.33.

rf

xryrz

ate as Ln the prevlous

problen, zx).

then the volume y n ryz Ttrose valueg of xryrz

and the surface whlch naxlmize

area Ls 2(xy + yz *

v also maxlmize v2 (xy) (yz'l kx). tn v2 being a constant, - zxt whlch ls equlvalent y2 will

The sr:m of be naxLnLzed Hence

the three faitors wtrenever xy = yz the flgure

to x = y = z.

of maxinum volume ls a cube.

- s-12 -

1.34.

By Froblen

1.25 ylth

k - 4, we have

[*2 *" 2 * r 2) Ll4


t-, or G,+ /f l + rlE s 3314, with

tG+ G+ G
, equaHty lf and only lf x - y z - Ll6. } 8o. Problen

1.35.

I{lthout

loaa of gerteralltlr

!r may assum" rl

= ^2a..

= * s ... Henee * al d2

= ] and by Chebyshev's lnequallty, an ,

L.23, we have a.f...+a


I nln

1 *... a.

1!

1 * ... + a a- . I " ln ro

1 -1 ,

nn rrlth equaHty lf 1 .36 . Flrst by (1.3), and /only lf a, - '.. - "o.

( a+ r J ' * G + f r z =* (. * * * o * * )' - i t * * * i l ' ,


wlth equailty 1f and only lf a *!
or a - b. By the prevtoue exerclse,

- u +f

, ot (a - b)(1 - #-

O,

1 *l
a

b-a*b

r - J-

=4. ''
a - b i
1
I

rLth equalLty,

again, lf

and only lf

Therefore

', 1. l _t 25 (. *?-' + (b +=b)- i(r + 4)-' 7 . =

!. (" *
1t

Though by (r.1) + (b +
1t

we have a * * t 2 and b + + 2 2, ot aD -

?-

i)'

= 8, the value 8 cannot be achleved alnce thie Recall Exerclse L.22.


1.25 rtlth k 2, we have

wanld requlre a - b - 1.
L.37. Ualng the inequaltty

of Problen

- s-L3 -

.*J'* .+Jf ,1,[.,=[,[,[", t

:lr

n L

n .-2 + I gl .JJ

jlr

wlth

equaHty

if

and only

tr- = "'

= a

nn

=- 1 .

= ..' a, ., ! "1 Now by Problem 1.35, tf

= a + 1 . na n

t +>-n-- =rr2, Jlr"J $_ Lo1


j=l ' = ro = 1/o. Ilence the result with equallty agaln at al = ...

n-z

follows.

1-.38.

From Exerclse 1.12'

/ =

(a +

f,)^' +

lr

(b +* , "

;ht'*
lf

o* * * h " -

-n 2tt-1 '

as Ln Exerclse 1.36, wt t h equallty Therefore

andonlylfa=b=L12.

n=1

<"+]1"+1u+f)n

=i , # =+.
are positl.ve. Uslng the

1.39.

Since a,b ) 1, both log"b fact that log'a = L/log"b,

and log'a we have

logrb*log'a)2, wlth equality lf and only lf log.b = b. - 1, or a Therefore

.I I' n=l (log"u + rosp

!-1.

n-1 2rl

- s-14 -

1.40 .l { enayasauB ew1thout1os e o f g e n e ra I . 1 t y t h a t " 1 = ' 2 > . . . ttren {al}l-r


1Fr one. holds

1" a decreaslng sequence whlIe {.1}l=,


the result a, ' followa " ' = a from Exerclse

ls an lncreasing
and equality

Therefore lf

1.23,

and only'Ln lf

1.41,

As ln the Last exerclee, let a1 2 a, 2 ... talli=1


r Qr ll

= ro.

men {al}l-f

*d

are both lncteaslng seguerices' so that by (1.11),

the re-

eult follqrs.
Renark. fhe aame lnequalLty holde for p;q > 0.

L,42.

Fron (1.1) we have


bc b*c-b+c / ca 'c*a-cta ab 'a*bgivee -a*b

4 inequallty.

Addlng these three Equallty holds lf

lnequalLtles and only tf

the deslred

a - b - c. a < b < c. Then

1.43.

t{lthout

lose of generalltlr

tr may asaume thet

1og a < 1og b < log c, so that by (1.10) (1og a + log b * log c),

a log a tb

log b r c 1og

"

log(.tbb"c)
whlch gtves a-b-c. Req4!&. the desired

> tog(aue)
result.

a{btc 3

Equallty

holds

lf

and only

lf

The lnequallty

clearly

generall.zes

to n numbers.

L.44.

If

we add 2(ab * bc f ca) to Lnequallty


t

(1.5),

we obtaln

(a + b * c)-

> 3(ab * bc i

ca),

a *b *cr 3
wlth equal{ty lf and only

- { r ,,|ffijg3
lf a - b - c.

'
Now applylng Exerclge

1.4, we obtain

.{3 with

@=ffi-36 ;" ,
lf and only lf ab 'bc . b - ca, or a - c.

equallty

- s-16 -

CIIAPTERII:

SOLUTIONS

2.1.

Flrst

solutioa.

Froa (2.2),

we have

! f'l n tt'q'" 'rirJ't", -^, -.. . +!


wtti.ch glvee .ar a6 ff J-J3 us the deslred ... a--3tl t.

= (4. ", "3


lnequalLty. gay. Equallty a, j

til

a rl/ a

- 1'
lf aad only ao . -r

holde ...

Then tn-

^z
llence,

"3

"z
- 8o. \ ls clearly

. "2 -. t3

1.

"1

' "' "1 - "2 -

Secoad,golutl.oa, Aeeune that lt ganeralfty, lnductlve


(1)

Ttte lnequallty

ttue for n

l and n - 2. loss of By the

ls true for n - k.

I{e can rssrrtne, vtthout

that rk+I 1" the enallegt of the arrarr...rak+l. asarrnptlm,

a*2+. . . + % > k. t3
"z
"1

I{e also have that

(at
or

"r.*r)t1

- octl) > o,

(2,

tk+l

" k * fu + r_ a = r.
"1 "1
!.nequallgy. clearly also true: nE1 Then rlr 6i 1. Let t ",

Addlng (1) and (2), Renark. {b1,...rbo} wlth equa[ty

we get the deelred theoren ls

The followtng

be a permutatlon of {arr...r.o}. lf and only lf a, - b, for all

- s-17 -

2,2.

Flrst shev's

solutlon. tnequallty rn.

Ttre slnplest

method of proof

is

to apply Cheby.i = bl,

ln the form (L.11) Ttrue

of Chapter I wlth

L = 1r2r...

,.r2 a i + a l r ... r a L z

j>

( ^t+ "Zt
(.,, lf

... + ar r 12

whlch gl.ves us (2.5),

wLth equalLty

and only lf

"L - ^Z

=a

Second aolutlon.

Let us e:rpand (r!r.r)2

and uee the slnple reault

rh a t 2 a b = .2 * u 2 ("e. ( 1.2) , chapter r ) :


^ ' rlr"i*'rlr"r", [rlr"*J ( n r2 n

1"1 ,!r'i* ,1, * ^!t


non^

1a. 1."!,*(n-1)Iaj L=l -

tz

L=1

tzr.z = o-/-" i L-^L

Fz
{f and only if

i-1

llence (2.5) E ..' ^L"2

follows = 8n.

lmedlately,

with

equallty

Th:!r1L solutlon. ln texts useful,

The following

proof,

wtrlch frequently where (2.5)

appearg

on probabtltty

and statLstics

ls eupecially polnt.

Ls lndependent of the methods used up to thls

Let A - * B-a,and - y'ni'l a A!7r. ni=l1

rhen

- s-18 -

nA

os

1-1

I("r-A )' r

r t ( . ? - 2 A e, + t2 )
1- 1^r , lln^

tra

- oQ' - zll
c

1-1 ^ fi

a, + L t'
i-l I E

- oQ' - 2A"[i

,lr.r,|

+ nA4

- ,r(Q2- zlz + e2) = ,r(Q2- l2),


r,,

wtreaee A- < Qo wlth

equaldty

ff

and only-1 if

a, - A for

every i,

. or, what ls the aane thlng, 2,3, Tte naxlntn will

lf

... and only lf a, - ^Z -

- !o.

occur at the sinultaneoue maxlnum of


tc

(e - a)(s - b)(e - c)(e - a) and nlntnun of abcd cos'!, guchexl.gtg. Ihefltgt occura at e - a'"g -! "ti-t, - 6 - c-B

tf -d,

or a r b - c o dr wtrlle the second occurs only ln the case of a equare. Renark. eaally uslng The followl.ng geonetrlc nore general

Both occur

theorem can be proved of greateet

nethods: ls

the n-eLded polygon polygoo.

area rrlth

glven perlmeter

the regular

Let V be the volume of the

( o,H)
eylLnder, r the radlus altltude. From the flgure forn of of lta

( r ,h)
bace, and h lts Ttren V n.2h.

and the tno-lntercept a llne, re have that


r lr

r and h

are related by

E,

1. E -

- s-19 -

Ttrus $e may vrlte

v - 4nR2u<-r*l <!"> <**t


= 4nn2n

ffi +zt. *'l'';aa:a


t----j
of the cyllader

'
ot r -+,
4/9 the volume

w 'tththe ma:c l mum o c c u r r ln g e n f r - H- i, of V wh


h i
H

Ttrug the volune

le at nost

of the cone.

2.5.

(a)

In the flgure, ls

the altLtude of the ls easy

a * x and the rafl-ue


-tt t

baee le Va' - x', to see that for

It

x must be positive

the cone of naxlmum voltrme, ls lrrelevant to our of the cone

but thls proof. Ls

The volune

u- $ t.2 _ *2 ) ( a + x)

-T,ti=,,#,(a - x).
Since the aun of the factors
z

Ls conetatl-t,

2a, ttre na:rLnun

af . gccurs when :-# cone. (b)

x E 8 - x,

,^ or x - al3.

Ttris deter:rnlnes the

Let V be the voluue der. Then, fron

of the cylln(xry)

the flgure,

v - 2nx2y
and

22 *2 '+ y -8.

- s-20 -

Ttrus

[+J',
wlth equallty when L-yz-\.,
.J

2^2

or x r a

a 'E ' '

Reqark.

I{e could have also rr:cltten y

2r*2ffi

or v r ztry(a2 -.y2). ae before.

rn elther cese, we aquare and proceed

AB ln the precedLng problem, we aee that we can

solve thLe problem wlthout squerLng when we u8e the latter foru, V - 2ny(a - y )(a + y ). / 2,6. Ttre maxLmn of (a + x)3(a - x)4 occura at the naxlnum of
,8 (-)f x .3 ,a -x .4

-,

ntrlch 1" (f)'

.?a-7

, by (2.2),

occurrlng

whcre

alx

a-x

-'-Z:'orx--alr'. Ttrls technl.que of lntroducing Iggg. to nake e conatant It ls generalized eum or other quantlty

the proper ls wldely

coefficlents

appllcable. aE ln

Ln the wel.ghted foms

of lnequalitles,

Sectlon 5.2.

2.7.

subetitut" rl eo. .r, oi for b'


ehos that

and Bo oo. t{e are theo requlred to

i 1 I I (al + u[)tai + b;) (aj + b;) = (.1.2.3 + b1b2b3)' ,

.i

or, after expane{onand canceltratloo,

*'l"3oi .l.lrlr + cululal * +ulrl"f r|l.lr + r'i"lo3


, 3"l"l"lbrbzbg + *laful.1a2a3
It eufflces to ahow that the flret quantlty ln parentheses on

- s-21 -

the left larly (2.2).

side ls at least the fLrst

term on the right

and stnLfrom

for the second quantities. Equallty holds lf - brbrar. *i -

But these follow directly

and only if

blbZ.3 = brbra, wtltten,

ararb, = ararb, and "1.2b3 Dlvldlng these sets of equalLt{.es as i = Lr2r3, rre get *3*1

and lettlng

arlbr,

*1*2
*3

*2*3
*1

*2'

* 1 " * 2 -* 3 .
al a' a" 1g,6=*-:" is lnplied by the sets of equalltl.ee. Fur1D2o3 thermore, thla eet aleo Lnplles the orLglnal two. Ttrue, they are that equivalent, and our lnequallty becmeg an equality lf and only lf

the arg and the brs are proportlonal. Note. lnequalLty This lnequalLty (4.15). ls a special case of ltiilderts

(2.2), Lffi= 2.8. Fron


frm whlch the deslred and o,nly if In thls inequallty follors "4. in the mean. imedlately. Equallty = .3 =

holde if

^L - ^2

Remark. geometrlc

lnequalltyr

ril have the exponents

mean goLng over to coeffLcl.ents

ln the arlthnetlc

ThJ.s !1111 be generaltzed

and expanded in Chapter V.

z.g . (a) tl e hav e(1 - * )5 ( 1 + x ) ( L + 2 x ) 2< lis ( r - * ) + ( r t - * ) + e ( r + z * ) l t=


1 8 ) L ',

,,

lrith (2,2)

equaHty

lf

and only only if

lf

1 - x - 1 + x t - L all the factors

2x, or x - 0. If

ls appllcable

are posltive.

- s-22 -

they are not,

exactly

two of then muet be negative "od

for

a But

maxlmun to occur.

These muet be (1 - x)5

(1 + x).

these cannot both be negative. plete. Remerk. Problen 2.8.

Hence, the eolutlon

ie c@-

See-the remark followtng See also Problen 2.46.

the eolutLon

to

(b)

Note that

x f 5 - 5x - 7 = 11 - x hae the cmon equela 8.

solutlon A.

x - 3, ntrere each factor Thue, we have

Now x + 1l-" 4 - |. z lrc-J -

( 2x+ 2) s f(x+5 )-(5 x- 7) ( 11 - x) '

1',ro

lfgg\l3 2'13 '

_ ,38 -

The naxlmun Ls thus 238 nay be negatlve.

x - 3. on x ) 0, only ure factor "t Ilence, (2.2) ls appllcable for ftndtng the

uaxlnurnl and the solutl.on la conplete. 2 .10 . B y (2.2) fot n - 3 a n d n - 2

a+ E +c - - W .+ r!3 m 3

-tffi
In each caae, equallty holda lf

Jm-3ffi.
and only lf a o b o c.
n- 1

2.LL. :fre (l) terna of oo contaln eaeh x, exactly (;_i) , b I (2 .2 ),

tlnes.

Eence,

- s-23 1

o,
(l)

,n-1

] =[ ;_<l-it1<ltII *T/o -

ntr

[- r-t '

1-14

o1/o , n

(*)-ffi since
and only lf x, n *Z ...

=
= *r1. nn'

* ' <l : lt.

Equalltv holds 1r
of orr

Now by expanalon and deflnltlon

n II (1 + xr) l--1

- 1+ . Il--Ior > t + .
n

I - t f iloi' "
mBl

by the blnonlal
tt xrr'

nlq'n'-

I tlXtl-o-lmtn-m- (1 +t6-Jo nnEquallty holde lf and only lf

theorem.

2.L2.

(e)

Flrat x < y.

golutlon.

tet

the two nunbere be x and y. , then a:(

Aesune that

If e ls Such that I < a ..I

cloger than x and y ln the sense that lax fl we are requlred to show that .1 ax + ; y < x+y.

are "oa ]V . f - *. Ttrua,

Fron the condltlons on a, we obtain * . ot |1, ! " )r, o . t;lt < x + I r a B d e el r e d . (a - l)x. + |r - (1 "* Seiond solutLon. Let xy - rs be euch that r and s are closer
Then x ( r, and, Lf we uultlY - rr we obtain

than x and y and 1et x < y. ply thls lnequallty

by the posltlve ot *y + 12 < (x*

quantlty y)r,

xy - xr < yr - t2, f I, - r

or x+

y t t **-

es deslred.

- s-24 -

(b)

If

not all

the numbers in say al and a'

(2.2)

are equal,

then there muer be ^2, where

two of then,

such that

,1 a G ,

G ' (rlr"1)^":
a) arc closer ai. * ^).

. 1la

rf we let al - G and a) - ararlG, then al and


than a, and a, are, art or and therefore, by part (a),

a, i

Repeatlng

tlre process at most n-l -

tLnee,

we arrlve

at a get

= ... of nuubere alf - al LZn

al - G euch that

* i" '* [";io- [F,J"". ttrus, (2.2) r.B true and equallty holde if tl '' 2.13. rt w111 be ehorynthat (2.2) is true for n = 28, where r Ls any Loteger, and that tt n'k - 1. (2.2) ts true for tr - k, then it Ls true for and only lf

I'[e shall therr be able to conclude that (2.2) is true l


2L, eo that

for all n.

We have seed Ln (2.1) t h a t (2 . 2 ) ls true for n lre may assrae thet (2.2\) ls true for n 2t, or that
-

(1)

iJr"=
"L^2r '..

,2t

( ,' lr' [rlr"1 '


numbera, and where equality Conelder any set of 2r*1 By (1) and (2.1), holds - 2.2t 82t.

where .1r.. lf

. ra2t are any posltlve -

and only tf

nnmbere, 81,e''1a2tt

dpt+Lr...,a2rtL.

we have

."4

r$

- s-25 -

F'l"J:F,-i',"'] j;,,tr

or 1 1 ,r*L
rlth

,1*t

Pt+t lF

.!,"t'[r1r"J
lf and only if a,

'
- "Z ... a,t,

equal.lty

holdlng

u 2 " +l '" Z t+Z


thet le'
lf

' "'

' ar - *r ,
a,

and ar ar ..."Z"
* ... (2.2) r

and only lf

- "2r *L"Zr *2...a 2r*Lt


Ihue by the prlnall n 2t, r*rere

clple of natheuatical rls any lnteger.

- "Z lnductlon,

e2r*L. holde for

that is,

(2,2) holda for arbltrarlly

large r.

Nor aaerne that (2.2) ls true for n r k:

(2' ',

*i",' [,],',J"o'
tet

for any a tlt.1r"'rap.

k-1
1

tlrtt'

Ttren by ( 2)
i\

"n'fo(k-l)a +an) * rfr.*' [r!r.Jt'*,


or

- s-26 -

{ = ,1r", [tlt'r]'n'
r,.r" .f;-l >
k-1

.k

f "t'

ot
k-l

rlr",

[I1",]*,
(2,2) holdg fot all n.

... ntrere equallty holds lf and ont-y lf - a_t - t = "l - "2 r k-l = tr, w?rlc hle t o 8 a y , a l : t h u s (2 .2 ) 1-T - ^ Z -. . . t_f. i holds forn=k1 lf lt h o ld e f o rn -k , . T h e re f o re , b y t h e prlnclple 2,L4. of backrard lnductlon,

Ttre oaly ualn steps that need be added are those that prove the equallty eondltfon. { Along wlth aesumlng {nductlvely r ... - Gn Lnplles a,
- Gn and A - G (alao,

that {*,

re easuoe that
lmpllee thet \

- "k.
\

Then
- A).

Gk, - Gk+l {

thar

Thla tnpllea

that 41. r ...


.l . ... -

- 1

and

%+1

{*f

, or fu+t
for

Ttrerefore,
equallty. It le

8k+1 ie a neceesary condlt{on eufflclent.

also clearly

2.L5.

For.n

1, (2.2)

ls clearly glven, \,,

true.

Now aagune (2.2)

for n

k.

If

we uae the notatlon th* A+t,

a mean of arr ... r8k, ie not greater each of 81r...r "2 ... r ap. A+f . Ttrus b 2 0. Now

whl.ch ls at leaet lf

Furthermore,

b - 0, then a, *

F
_i"t*"t*r + r ' 3 - ] l a r - '- - 8
Thue

k{ + { ,*b 1-'\

*[*- ' ' .

- s- 27 -

(\*r)o*t- (\ *gfrin*t - ^l*t + (k+ r)({)k(r+) * qp&(nlk-l<ni ,-r2 *


1 + "' * (#-r)n*t

+ (k + l)(\)k(t+) ' d*t

-d*t*rd
t" r At

Lr.r

r-

"t*f

> (arar..;1)aOr1 ,
\ theorem. for n Equaltty * l. - k holde tf holds if and

where r,e have ueed the binmlal ' only lf then, b - 0. Ttlue (2.2) for all .

tg true

By inductlon, and only lf

(2.2)

ie true is,

n and equalrty

b - 0, that '\--l' ' Reuark.


6

"1 - "Z-

I{e have uged, and proved,

the lnequallty

(1 + h)- > 1 + nh, wtrere h > 0 and n le a poer.tlve integer, equaLlty holdlng tf and only lf lnequallty, n > 1. known as Bernoulllfs h o 0 or n - 1. inequalfty,la rn fact,

wlth

thls

true for any real ia reveraed.

If 0 s n S 1r the sLgn of the lnequallty

See problen 4. 2.L6. For n - 1, (2,2) ig vacuously satr.afted. rf rc > 1, then *k,
la each ceee lf

Assume(z.z) for n - k. x k ) l, wlt h e q u a lit y


we get

* k -1 , * k l * k -2 , . . . ,
Lf x - 1.
f

and only

Addlng,
oor * 1),

k*k

> (* k-1

* * k-2

or
k T (x - 1 ) > (x k -l + . . .

+ 1 )(x - 1 ) - *k - 1.

- s-28 -

Rearranged,

thls

le

(1)

(k+1)x k -1 < k * k * l, and only lf x = 1. Slnllarly, Lf O . * g 1, * 1 ), o r

wlth equaHty lf xk s xk-l, ...,

*k i 1 , wh lc h g lv e s k x k < (x k -t + . . .

t-k(*-1)> rk -1 . A gar'n, we get (1). s u b s t lt u rrn g x -F q , wh e re c 1 r. . . rc 1 of (2.2)), we

ere to be deternlned (they depend on the alr...ra1

k+1 (k + rycr..."k- 1 < ot"f*t..."f; k+l I+ 1 )1 /k ." 1 * t* ..* " f


by the inductlve hypotheals. bri... Lettlng c, -.br/bk+l, wtrere

rbO af algo to be deternl.ned, ne have

ol*t * rl*t * ... + olil = (k + 1)b1b2...bk+1,


efter addtng 1 to both eldee and nultlplying Thue, lettlng nf*t - al. 8r.ves (z.z). through by blll.

Equality holds rf and only

lf x - I and - ... "1 "k, ". #


"1 ' 2.L7 . "' - L. B y (2.2) rye have (nr1l/n (1 . 2 . . . n )L ln . -

...

-#

- 1, or

Ttrue, by inducti.on, (2,2) ls proved.

1 f 2 * ... * n n

' * ' ft" * 1 )-$ .,


whence

n! < lAJ$n.
Equallty caanot hold for any value of n > 1 because no two of

the fectorg

Ln n! are equal.

- s-29 -

2.18.

(a)

Ttrls problen le elnply the dual of Problen 1.30'


box whlch mintmlzes same es that nanely, uhich the surface naxlnlzes

Tttus, the

area ln thl-s problenr Is the

the volume in Problem 1.30'

the cube. the box ln the plane of the mlsslng reflectl.on forn a box poeare

(b)

Flret face.

solutlon.

Refleet

The orLglnal all slx

box and thls

sessing twlce

faces and whoee surface

area and volune

those of the orLglnal. ls a cube.

The volume of thle

new box le

maximlzed when lt

Ttris naxlmum oceura at the be half a

naxlmum of the origlnal cube, whose blggest Second golutlon. and c the other voltse.

box, whlch must therefore the one opposite

face ls

the nLselng face.

Let arb be the edgea of the mLeeing face edge. Let S be the aurface area and V the

ltren S - ab t

2ae * 2bc and V - abc, 8o that

= i(gtt , v2 - f,t"u)(2ac)(2bc)
wlth half equaltty at ab = Zac - 2bc, or I - b - 2c' solution. Thle la a eube, ae ln the flret

2.L9.

Let r be the radiue of the clrcle g the angle of the seetor. perlmeter If

and

P le the

and A the area of the aector' 72912 -

then P - 2t * r0 and s'

Thus, rhe mortm'maree ie p2!L6, where < - |tzr)(ro) f;enlz. 0 -2 ra d l a n sa a d r=P 14.

_ s_30_

2 .20 ,

Uslng (2.21 wlth n = 3 a n d (2 . 4 ), we h a v e

4tEW=46/@-fa*
l.7 t.t

+y+z)2

-i [(x
Equallty

+ y) + U + z) + ( z + *) Jt s ( * + y) z + ( y + z) 2 + ( z + x ) 2,
holdg tf and only if x = y - z.

2.2L.

I{e begla irtth

the lnequality

$ + z)(z + x) ( x + y) 2 Br yz,
r*rtch la ptoved eaetly holde lf y and only if by ualng z. \ (2.1) (eee problen 1. x). Equallty

X-y=

Let ue substLtute > O).

I - 8r

r I - c (note that xryrz - I - b, z - c),

We obtaln - c) - &(,2

abc a 8(e - a)(a - b)(s wtrence abc , 2K

or sabc E gs(e - a)(s - b)(s

aF=;'

o r 2 r s R,
and only lf lt a = b = c, the equilateral is sho$n that trlengle.

wlth

equality Regark.

lf

ro geonetry,

the square of the disls R(R only for 2t), and

tance between the centers hence that lateral R - Zr 2 0,

of the two clrclee colncide

The centers

the equi_

trlangle,

of course.

2.22.

Froo (1.2),

we have that

ei;q.+(t,rt

+It - "rll .$<r. + 1 - "z)


lr - ", l> , #G2 *,.L .3) lr-.rlrr#r^o*1-ar),

- s-31 -

with

equallty

ln the flrst

lnequall.tlee

of each llne

lf

and only

r r f a r l - l t - a r l , l"2 l - lr - a r l, ..., l.o [ 1 1- a r l, a ndi n the second lf "'r 8n2 and only if "1 .1 = 0, L 0=.r< "2> lfor 0, e2 I 0, 1 _ .3 2 0, Addlng, 0' 1> 0; i.e.r i-1r...ro.

we obtaln

(where an+l = al)

F-_ + .-L' - { " t +( 1 - " r rr4r ) - r n t, ^ 1 1


wlth equallty tf and only tf lf

E
condltione absrre are

both1equallty

satiefled.

That is, l "1 "2n

n is even, 1 "3 ^4-... r an_l - 1 - "o - al,

or al for - "3 3 "' - a'_l - "; If r ^2"4 n Ls odd, "' - I - "o-1'ao - r - a1, ... arr - 1a,

any a such that

0 S a s 1. t3 - r

al - 1 "2or ^L=

^2-

"'

- an

1 2'
We have, by (2.1), of

2 . 2 3 . (a ) This problen wae solved Ln problen l.


that ', A < (Pl4)-,

where A and p are the area and perlneter

the rectangle. (b)

slnce to slt of lte wal1'

each peraon has a wldth, around the teble sldee. rf

the naxlmum number of people the perlmeteg of three

deternlnes

we reflect wlth

the table trrlce

in the plane of the and twLce

we get e new table

the perLmeter

- s32-

the area of the orlglnal. four eldee included le a square. 18 half in its

This new table, perr.meter. rt

however, haa alr. has naxlnu* table area

when lt

Hence, the orr.glnal a square, the l0nger

has mextmum

area ntren lt the wall. Nolg. 2,24, Uclng (2.6),


lllr .r r

glde belng agarnst

Conpare wLrh problen

2.1g.

we have

G + : ; ( 1+ ! y ( 1 5 - 1 + ,* .i * ) +

* .+ .

# .* ,.#
' 3 ',3 .-3

> 1 + f '*\x+ ytz' - 6 + $ ;;:5 3) -: ' > I + g x+Y+2 r 3 rI ; |1 1 3 r2 '3'

3 r- ' x + y lz t

-l+ g + 2 7 t 2 7 -6 4 . Equallty holde if Renark. and only if x - y - z - L/3. 1 * * = ZA ,

!{e cannor use the lnequalitlee

1*

= ,fL

, , *I

r rA,

for equaliryholdsln all caees lf


the hypothesls that

and only if

x = y = z = l, whLch contradlcte

x + y I z - L. 2.25 . Flrat eolurton.

In other worda, equality would never hold. B y (2 . 6 ),

, - r - = 1* * =;F=--;T
,2
---{

IL- x' + lr xa)

t-J
wlth equauty if and only r.f x - 0. guaranteee that L - *2 la poaltlve.

Ttre requr.rement that

l*l

. r

- s-33 -

Second eolutlon.

Since

u ---2
'

n r-x

,+

2
1*x'

l-x'

4'

elearly

y ls nlnlnLzed

at x = 0, where ! = 4.

2.26.

By the leltna of Sectlor- 2.2,

(x+a )(y -a ))x y , where 0 < x < y and 0 < a<y - x. x+y: !tr1s glvee

x*y (x + a )(y - a ) '

*. + ' , . +;.# ;r o r
whlch la wtrat ls requlred.

; +

11

'rh,
xy
Hence, by

"'F-

2.27.

on the lnterval (2.6),

speclfied,

all

terms are poeltlve.

v =*
t,,82 .

**

**

*r+. *

. *.

T # * T+E

En-tT?I-+
tf and only lf

x) + 2(1 + 2x) - 8 ,
1 - x = 1 + x - 1 * 2x, or x - 0. altltude, and R the radl.ua

wlth

equallty

2.28.

Let V be the volume of the cone, It lts of lts base. Then, ae in Problern 2.4,

I+h=r RH
and

- s-34 -

tj

v - + nR2H$ nr2nt$r,*r,gr, ', 3 > g rrrll L. tr . .Thi ( A' E' EJ 1 ( 13 q ?

' f,nr - h,

wlth

equalr.ty

when

3, or * - *, * -

H - 3h.

The vol,ne

of

the cone Le thus at leagt 2,29. Flret

914 tttat of the cyllnder.

aolutlou.

Squarl.ng, we have

y2 - (*2)t("2- *r)ro -

i&rrr(.2

- *2r2o,
Thuc,

shere we $ant na - 2n o 2n/n. - 0, or o

,'=f,[ffi*"-I,*,t#'#1" ,
".+* 2 - ^ 2
tflthout
-XrOfxr

wlth equallty

a ,_-F

Y m.

/t+g

Second eolutloa.

squaring,

we have

, - 44 -

ctB t

to* 1 t 1 8 ( a x ) ln ( a * x ) n -

B- 1 Wewloh to have ua - nB * n = 0 and x r 4o -1 -E -T T a ' o r I @ "^-: --' -v - fr-1-2; ' so that ffi arrt
rm
, -r

/;A

Ttrla ehowg that y la naxlnlzed

at x - fq=

/t+ s vm
we wlsh to maxlnlze

2.30.

I{fth

three undeter:mlned eoefflclents,

crgry,

(a :r)tB (a- x) ltv( a + x) l( 2a - x) .

- s-35 -

We have (1) and

c- B+ Y- 1- 0

* - 6 f 1 "- F + a - f f i a ,
or (2 ,
and (3)

2( y+t;-

( c+ 11 Q- ,r )

2(B-1)-(a+r)(g-z).
o and B fron these three equatlons, re optaln t 3 -g t 2 + g y -1 -0 .

rf *e ellnlnate

Ole root of thte equatlon ie y


a

- 1, rhlch lfves c-

B - 3, eo that

* poLnt at trtrlch the product le greateat. Ttre other rootg i le a are Y ' (7 t 3rs) 12, both of uhtch are poar.tlve. rlowever, fron (2), a - ttl (Z - y), a n d, slnce c ) 0, \ < 2. B u t th e n u 8 -,2 > Thua,y - ( 7 - 3t5>l 1- . Ttr ua,r heonly

O and B - ( 1 - 8) lZ.O.

poeelble coefflclenta are c = - B 3, y - 1.


2.3L. (a) Let the radlue of the cl.rcle the

be r and the dLstance frm

chord to the center be a, as tn the flgure. on the circle the rectangle. Ttre polnt is one corner If A is the we-have (xry) of

area of the rectangle,

- s-36 ,a ^

*z + yz - r2 and A = 2x(y - a) = z/r-(y

_ a), or

t2 - 4(r2 - y z )(y - a )2 - 4 (= -y )(r+ y )(y -a )(y -a ). Introduclng two undetermlned coefflcients, o and $, we have

az and we want (1)

$to ( r - y ) lt B ( r + y ) l( y - a ) ( y - a ) ,

- o+g*2- 0
solutlon to

and a cmm

(o (r-I )-y -a 1

I o ( r - ]) - B( r + y) .
o f t h e a e i" l(o r + a )l(L + c ),

Tte eolutlon of t h e f ire t and a solutlon

of the gecond ie

(2 )

- - -dTBr - E=' , Y- o- B

wlrere we have ellnLnated B fron (l). - t/(s - 1), or

Ttrus, (ar + a)t$

+ c) _

,
slnce O must be poaltlve. r ) 8r ^2 * gr2, ^2 /*

B - c - 2 Le poaltlve, , 4ra + 4r2 (a # 2t)', ot t o.

Aleo,

slnce

a ff that y - *t. * R>.

Fron (2), we have, therefore, sjnce r > y > a, rhle determlnea 0, we get half a

the rectangle.

Ife reuark that whenra

aquare, aa we ahould.

- s-37 22 + \tb a

(b) Let the elllpee O"


(xry), then

1 and the chord y - d.

If

x ) 0, le a corner of the rectangle

on the elllpee,

A - 2x(y- d) - zau(|)Gfl,
where we have

Q t * ( f ) 2= r '
From part (a), the naxlmum of A qccure at

Yr

!*/<52 *a 'D ' y D

b4 , or

/,, d+/d- +8b


-(a)

2.32.

Lt the sphere b. *2 + y2 * ,2.> 0. lng the Begpent be z - a and (xryrz) octant, rf

,2 and let

the plane bound-

v le the volume of the box

a corner of the box on the sphere in the flrst ,, then V - (2x)(2y)(z - a) < 2(x' + y')(za) r + z)(z - a), wlth equaliry the , undeternined tf and only if c and B: x y.

2(r - z)(r

We lntroduce

coefflclerrts

vf r [ c( t - z)]tB( r + z) l( z - a ) ,
problem 2.31(a), where - c + B + 1 - 0, and, ae ln

"- ffi - fi+r-7;{1.,or


Slnce r ) 8r I - c - 1 ls poeitlve.

Ttrerefore, the oaxl.nun

occurs etz-N.*

ffi1.

- s-38 -

(b) 'Trre nethod of solutlon is analogous to that ueed Ln probleu 2.31(b). Ler rh e e lt lp s o L d b e + . * + 4 at bz cz

l and the

boundl.ng plane be the corner

z - d ) 0.

If

V is

the volune and (xryrz) wlth *,y ) 0, then

of the box on the elllpeold

v - 4xy(z- d) - aabc$)ffl *l ,
* afrt * (l)z - t, the naxinrm occure, by Part (a), at xla - ylb, and'z c 2.33. (a)

and,alnce tflt

+ J.,r -t

r o r " -| ta * /rc ).

Let the radlue of the cylinder be r, the helght z, and the gurfaceareaS . Then z+12 -c a n d s -I n t 2 * 2 rt z -

n 2rn(c + r - ,2, - Zrrr(t t z) - 2nr(r, - r)(r - rZ), where tLtz o -c, f,l * tZ o 1, and tL r t2. I,le lntroduce undeternined coeffieienta, c and B, wLth

s - *F r(or, - ar)(Br- tur).


We requlre that 1 - o * B 0 and -

0r, 1+ o
eo that

Br^ B- 1'

(g + 1) r , J-Z g+2

Br ^ B- 1'

$r

' 2* \ / ' ?* r |- , r r ,
r-r
lz
-6

- s-39 -

TtrenB andc-B +la re with A -{ti* 9t, ti rr(r , - tL.Z, + A)

positlve,

go that our eolutlon

ls'

x-B-r'-

t r(r,

+ A )(2 t , - t t - A )

fr-

c72 tr(zti_ - tL tZ a ( r 2- r r ) A - r i - r i + r r r r ) _r_ _ , _ i ,r 3r z - 3rL z ^ I

2 tz

lmrr-A
r+ffi -=____f
the last

t2, * (rt

rr)A

rr* r" f A 33
-.4

1+A

;
the fact that A - V(rl + rr)' - 3trtr' c 21.

ctep followg frm

ltrle ls valid (b)

only tf x 3 c, r*ri.ch holds lf

and only lf

Let (x"yrz)rxryrz lts surface area .

2 0, be one corner of the box and let S be T t re nS - 8 x y + 4 x z + 4 y z - 8 x Y + 4 (x +y)2.

For fixed z, thls le maxLmlzed, ae ln Problen 2.32(a), wtren y we then want to naxlnlze S .- 8x2 * gxz wlth z + x2 - c. - x; , Ttrue S - 8x(c + x - xa). By the solutlon to Part (a)t thta Le naxtunlzed at * - |tr + frT-3E-). Note thst the box la lngcrlbed ln the cyllnder of part (a).

- s-40 -

2.34.

If

A tg the area of the recthen

tangle,

A-(x-a)y-!1x-a) xy

'fr<t -u,
- ,*#,n-t1frtn-1<r _i)
t) = c.#ln-t[o - c*m
LPJ
1 - (-AP-l 'n "

(x,y)

* (t - il n

wlth equattty if and only if


2.35.

#j5;

r t . pa/(p _ 1). i , o" x


the altltude h, and one

Let the aurface area be s, the volnme v, elde of the base s. be a'and lte perlneter

To specLfy the ehape, let p wtren s - 1. rhen v -

the area of the base has2 and

B -2ae2 * hps =z ^e 2 * t r , n " * ] u t "

. te

= te , "p2h2'4)Ll3 + v2)Ll3
and where .equallty holds lf and only if

wtrich la glven,

glves h/s - 4alp.)


2.36.

3/pY 3h6ay 2as2 -*nn" - G*,? uz>rlio, " -3@^ {E ,"V T P 2 ffi

also ' (nris


let

Let K be the area of the tr{angle, a be the opposl.te slde, and ret

let

o be the gl.ven aagle,

b and c be the lncludlng

eLdee.

Wehave K + bc eln c.

(a) By (2.1), b + c ) z{TE, =?ffi,


lfb-c.

rrrh eqirdlltyrf andonly

- s-41 -

(b)

By (2.1) and the 1aw of coaines, + c2 - 2bc cos o "2 -b2 > 2bc(1 - coa ol $]@ wLth equallty lf and onlY if b - c. when the trtangle LB

(c)

Slnce a and b * c are both ntnlnlzed

leoecelee, 3o ls theLr errm, the perlmeter. 2.37. On the domaln glven, at mst by (2.2), + ?v + 11 - 3x-- v + 2x - v + 1]3 - - [x - 64 one factor Ln z ls negatLve. ThuSt

"=i
wlth equallty orxr
2.38.

lf

and ooly lf

x * 2y'

11 - 3x - ! - 2x - I * 1'

2r:y-L.

In the graph on the rl'ght' the elgaa Lndlcate of z in that aleo posltlve regLon. tlve tors dmaln. the sign z Le

in the shaded Poslfec-

0n the other all three

reglon,

ln z ag wrltten

are ia

posltlve

and hence there maxlurrm there, La no abeolute

no relAtlve elnce there the firet be lJ'rftten

maxLmumof z.

A11 relatlVe

naxLna ln wtrere z 'Can

octant a8

are therefore

Ln the ehaded reglon,

- s-42 -

z - (x + 2y - 10)(2 3 - x - 3 y )(2 0 - 2 x y ), wlth all factorg posltl.ve.

To uge (2.2), we Lntroduce two undeterrnlned coefficlente, a and B:

* 2y - 1 0 ) lt B ( 2 3 x - 3 y ) ] ( 2 e- 2 x - y ) , z ft[a(x
rhere we $aat the aun of the factors and y, or c - Z - gTtrug bo be lndependent of x

0 and 2ct - 1 - 38 - 0, eo that O 5, B - 3.

" =*[sr*sJ3
wlth equallty

-225,

ff aid only tf 5(x + 2y - 10) - 3(23 - x - 3y) y, or x I 3, Y - 26 - 2x - 5, wtrich |e |n the shaded reg1on. Then, by (2.1),

2.39.

Let the atrnbere be a and b.

+f+ * 1 11 +f+ +-a!-]>iGb, 21 2 -rl- 21' 2 ii-ur ! ;*bJ


wlth equaHty {f and oply lf lf the arltluetlc lf a b. and harmonlc means are equal, thatla, and only

2.4O.

FLrat aolution.

From (2,6),

we have that

i .***=

e or

bc * ca+ab > 9abc,

1 - (a + b + e) + (bc * ca * ab) - abc > 8abc.,


i1

wtrlch, when factored,

ylelde

(1 - a)(1 - b )(1 - c ) l 8 a b c , or

,*.- t,,*- t,(-

'

e,

- s-43 -

wlth equaHty lf Secoad sclutlon.

and ooly Lf a - b - c - +' tJe dealre to ghow that

(1 - a)(1 - b )(1 - c ) > S a b c . slnce 1 - a o b + c' Ec.' thla follows fron Problem 1.5. Equallty 2.4L , holde lf and only lf . a - b i - +

B y (1.5) and (2.6),

(.2 + a2+ c2r<3**.*,


or (a + b +

> (bc* car au)rS***#

=',

,9abc. A ls o b y (2 . 6 ), we h a v e + b2 + "2, ")("2 (a+b+c)(bc+ca*ab)>ge b c . A d d ln g t h 8 t rro ln e q u a llt lc e

glvee ue (a + b + c)(a2 + b2 + "2 * b" * ca t sb) > 18abc,

(a + b +

")(2.2

+ 2 b 2 * 2 " 2 ) > (a f b + c ) (a 2 + A 2 + c 2 -a b -b c -c a ) * = "3 * b3 + c3 - 3abc i 18abc,

18ebc

"3 rlth 2,42 .

* b3 + c3 + 1 5 e b c s 2 (a + b

+ c )(a 2 + b 2 + c 2 ),

equaHty lf

and only Lf a - b = c' B Y (2 . 6 )'

.Flrgt 8olutlon.

[4 ("u2
or
with

*4

bc2

=e * -Ll1r52 * b"2* ""2) caz)'--

("4 + b2c + c2a)(au2 + bc2 + ca2) , g^\2"2,


equallty if and only lf e - b - c.

- s-44 tt,

Second aolution. ab' + bc'


,t,

By (2.2),

a-b + b-c f c-a 2 3abc and equallty lnequa1lty. lf and only lf a - b = c.

+ ca' > 3abc, both with givce the deslred

Multlplleation

2 . 4 3 . The nlnlnun of the dletance d fron (4r'2rL,


on the ptane le the dl.stance we seek.' Ry (2.5), we have

to any poing

(xryrz),

d*l v4+ 9+ 36

=lW2*stt+j'*rffi
- 1,

.l?l| .rFfl| +361+2x*3v- !Zr 13*7 - ' "[ 6 l- ,3 2 -'l ., >


at wlthequality +1 ,K 11

tf
1

' +

-+,or

x I 2617, - - +, v

ls the polnt on Ehe plane cloeest to ThuB, q, -;, ? " - +. (4, -2, L); lts dlstence frm the plane le l. Reo4rk.
x - 4, y - -2t plane:

For thLs problem, lt


2 I lnto

would be slnpler
forn

to substltute
of the

the nornal

of the equatlon

2x-3v + 6 2 ' L 3

'2 2*( - l) 2+02

toobtainu-l
2 .44 . Fron (1.1), we have

l-t.

@,iFr*r-A+B-yr)
Er(r*r- Eyr) rr+@ - A), - ofi" - A)+
wLth equallty *i*It-AiB. tf and only lf x* - A' B - yl (1 ' 1,

- s-45 -

2.45.

Along the raye of the angle, 0, we set up oblique coordi-

nate a:(es, as ln the fl.gure. If (xry) is the glven point,

x ' Y)

then, by the two-l.ntercept forn of a stralght line,

i * * = t.
Ttre area of the triangle, (2.6) , K, is gtven by Fb sin 0, and thue,

K = * absLn = :E-#-g (*r e


,xys rn 0

'T
lrlrh polnt equallty (x,y)

t 5; ; l
i= +

]2

=2xYsrn e,
= 2, or a =
2xrb=

lf andonly if

2y.

The

ls then the nld-polnt

of that

slde of the trlangle.

- s-46 -

SOU'TIONS CIIAPTER FOR III

3.1.

If we eet 2x + 1 y or x - irr - 1), thc expresston (3x + 2) (x + L) | (2x * 1) becornee

<f<t - 1) + 2)(i(r - 1) + 1) - 3o2+ 4"+1 4y v

ar

'trr * i * ot- +i(6 - H, * ++ zf.:' ,


rhogo ntntuun value
orttren y-Li,lf3,

fCA

* ztil

o r* -*

rrt T $ + 2 )z (x

la achlevedntrea d5t-{12 vy
1 ).

- o,

?.2..

file rrlte

*." x-+4+5

+1

, end rcnark rhlt

rhe targecr

value of the functlon lhe denoolnator

occura wtren the denoutnator lc a nlnln'n. when (x + 212 - 0 or x Ia 7.


the orlglnal

ie a nlnlnun

-2, and the

value of the aaxlnun of the glven functloo


3.3. Let ug set y 3 x + 3 or x becooea y - 3, eo that

expreeelon

y ' - z y+ 2,
{e

-1 Z * ? - z - w _ u fi l + z E-'

1 - __,- - _

denmlnator of thls last expreeelon de a ntnlmuu rtren o ry - fr , ot x- fr # 3; themaxlmunvalue of

$ -rQ)2 o 0 ,

the glven expreselm ls then

3.4,

Agalu, the naxlnum of the functlon denoulnator, and, ln partlcular,

occurs wtren we nlninlze "o"4*

the

wtrenwe nlnlnlze

+ sln4x.

- s-47 -

By (1.9), that thls

we know that the mlnlmun of cos4* + sln4x le 1/2, and mlnlunrn Ls achleved rdren x n nlL. Bence the maxfioun

value of the glverr functlon x = rl4. 3.5. '

le 6/11, r*rlch la achleved wtren

If we 8pp1y (1.3) of Chapter I to part of the denmlnator, have

we

* ,2+ y2 +.+'ft* +y)2.+,**+r'


xy

- l*!r'A2x y ' 2 2'


wLth equallty crly lf x s y.

' x * y - 1,

On the other hand, 26s

or U(:<y) > 4, rrlth equallty x y. Hence

under the same condlt!.one, nanely thet

t2, t . i,#2 2 L7
eo that 7 xy < 7 =L4

x 2+v 2+7+4* 4 - . { . * 7 2 2

3L'

Itrus the naxLnum te L4l3L and ls achieved when x y - LlL. 3.6. For the flrst factor of the denonlnator we have

2+ (x +r)(+|l
ulth equallty have

= z + 2 + *.i

> 2 +2 + 2- 6,
For tbe gecond factor we

when y/x - 1, or x - y.

f +.ffi +
wlth equaltty when lffit

> 1 + Z - i. ,

ffi
: 1. - 1 or x * y Ttre two equallty

co'ndltlons w111 be conpattble le 18/18 1.

tf x - y - !12, and the maxlmtrm

- s-48 -

3.7.

Ttre ldentlty detalls.

given ln the hl.nt ls easy to show, and we onlt the by S the quantlty tan A + tan B * tan C,

If we denote first

we have frou (2.2) that


S > 37tan
?-

A tan B tan C = 3'6,

or S3 > 275, or S 2 36, -tanB =tanC.

with equaltty lf and only lf

tan A -

It now follorile fron (2.4) that

tan2A+ tan2B+ tan2c= l rt r fG/5 l2 ,


wlth equallty agaln lf and only lf ten A tan B - tan C, so that

the mlntnrum value

of the given

expressLon ls 9 and le achLeved

only ln the case of the equLlateral trlangle.


Remark. ln Grapter V. A general approach to problems of thle ktnd ts glven

3.8.

Suppoeethat the equatlon of & te ax * by * c = 0. then wrltten

If

& te

ln normal form, x cos td + y sln to - p = 0, where fr------5 tcoa o, - al/at * bt, sln o = 6l/at i b', then, when (xry) ls replaced by (xorlo), the expresslon xocos t * yo eln o - p ls the In par-

(slgned) dlstance fron the potnt (xorlo) to the lfue J. ticular, the dlstance d, of the polnt (xrrlr) slno lles - p. We re ma rk f irs t to [.

ls slnply o+!stno -p=O

xl cos o+yl elnce (iryl

t h a t ic o a

on {.

then the total

flrst

rfrtr momentof the systen about f, ts

If M ls the total rnass

of the sYsten'

- s-49 -

IlN F ' .F .

.l-ottt 1-1--

l_tr(x l 1-1^

c o a u J* y , s in o - p )

' [rlr"*rJ"o" '* [rlr"rtrJ"* o - Pill'l


- Cil{)coe ar + C$r)stn o - pM j - ttfi coa ut + sfu 6 - p) = 0.
3 .9. Let the coordlnates o f A rB rCrDb e (x rry 1 , z r), (x rrl2 rz 2 ), (x4r! O,zo), reepectLvely.
lhen, tf

f*

rB

(x 3r l 3 r z 3 ) ,

the coordLnateg of p are (xryrz),

we have pl,2 + pg2 + pc2 + pD2

+ (v-Y)2 + 1z- zr)21 ,lrr,*t*r)2


2

-4 *2 -

4 2x fx. +
1-1 4

1=1 I 4

fx

+ 4v2 -2y Iy r +
L-l

i=1

Iv l

* +r2 rr "r * rt ,trr?

I{e have here the strm of three lndependent quadratlc expreseLons of the form (3.1), and lf we conplete the equares, we see that

the nlnlmun occurs at the polnt

*r * *2 * *3 * *4 , -Ts l I'-ta-fr. * 3.10. The deflnltlone problen.

,, _yr+

yr+ yr+ y4

"L*

,2*

zr * zo

and notatLon are glven ln the statenent of the

Let ax + by + cz * d - 0 be the equatlon of the plane, normal form x coa o * y cos g * z cos y - p - 0

whlch !re. trensform to lts

by dtvldtnB by /^2 * b2 + c2 . lie


' ,8

Weremark flrsr
the equatlon

rnat (frl7)

nusr

on thls plane and eatlsfy


.1..

t+.

- s-50 -

T cos a +y

cos B *7

cos Y - p - 0.

A e L n P ro b le m 3 . 8 r. t h e

dlstaoce d, of (xr,yr,zr)

to II ts glven by flrst

xl cos a + yl cog B * z, cos Y - p, so that the total Doment ls nn f t.dr a - f n*(x.a cos c + yr cos B * z, cos Y - p) -1 1 a a flt l:t

'[rlr"*t;coscr* [r]r"trJ"ot u* [r]=r""


= Ui cos o + Ut cos B +fr .!t(i co s o + ! cos g +; coe 1 - Mp cos y - p) - 0.

r$

.r P

rt

,J"o"'-'[ir'J

3.11.

Ttre total lf

second mdtrent, or moaent of Lnertfa, ls the sum = re2 + PB2 + Pcz + PDz PB2 + Pc2 + PD2, Slnce / Paeses through BCD, pg2 + pc2+po?

gince PA = 0 and rn, = 1 (J : !,2,314). the center of gravlty, or centrold'

of trlangle

takee lts nlnlnun va1ue, name1Y,

cr * c2 - t,t

c, t xr* *r)' - *,t, * yz* yz)Z,where andc,


glven ln Exanple 4.

are the quantltles

. 3.L2. Flrst solutlon J = (1 + 1 + L)2 = (Ix E * # W * # Al' f+ Yz t vy


= * * +. )(x + y * z ) wt t h e q u a llt y f o r x = ! - z .
In problens of thls sortr that ls, where we

Second solutlon.

replace 1 by

. c, the relatlon

between the arlthnetl'c lltus, we have

and

geometrlc means Ls usually appllcable.

x + y *223i r - ,l ' "a n d i .i * * .tffi .

- s-51 -

If

we nultlply

thege two Lnequalltles,

we obtal.n

( x+ y + z ) a * .+ * ) . s 3 W;l JXltz

e;
for

ulth

equallty

for x y' -

z slnce thle ls the condltlon lnvolvlng arlthnetlc

equellty acen3. 3.13.

ln both lnequalltlea

and geogetrlc

Ttre ldea ig the aane aa ln the eolutlon of problem 3.L2. the flret rnethod of solutlon we have

For

n2- (1 + 1+ ro'o L)2't*ve{+ {

"'*r**{J

r'[tr
wlth equallty
Deang, we have

or..*J(*1 +... *ro) ,


for *1 ... - *n. uelng arlthmetlc and gcmetrlc

*1*. ..* * o > " ffi = f

(*1 r- { *
... wLth equallty for x. IN 3.14 . rf wewrltel-cos have xn2

+ ... . *"J=o2, -")[t


r x .

a rn 2 f

a n d l* c o g

x -2 " o "--2f

z ,,"

1- +;+ #'|"""2;.2"""2f
whlch le eseentially ln Exanple 8. Ttrus, by the Ceuchy-SchwarzLnequallty,

the forrn of the functlon whlch ie treated

- s-52 -

2x

2x

z c s c T+v gec

t=

2x. ,z B 2 x -. . 2 x (A c a c2 x ztzsec 7)(arn 7+cos T)

, [,fr""" srn.fr"." ] "'"i]' - +r^ vl)z + . f ]


Ttre nlnLunm occurs uhen equallty prevalle, namely, whenever

IExEx vz cBcz ltz Bec z or rhenever 2 arc ,"ffi x, .x-.T co8Z. sanZ
3.15.. By the Cauclry-Schrnrz lnequallty, we have

y * ,rG

s G2 + 4\Ll2 (*2 + (r - *2))112 - s.


prevall,e, shich happena

Ttre naxlnqn occurd whenever equallty whenever

i'frfr7
t-

'

= foru of the or whenever'x315. The


If rre set x - cog 0, to be oaxLnlzed

functlon suggeste e second solution: thcn etn 0 - YL - xz. , so thrt

the functlon

is 3 cog 0 + 4 eln 0 = 5 cos(0 - 0o),'where go - arc a"o * . Itre naxlmum value of 5 coa (0 -- e ) le 5 and occurs when -o cog(0 - 0 ) - 1or wh e n 0 ' 0 ^ the o - a rc t ' a n 4 1 3 , rit rlc h I ' s

*r"
3.16.

""

ti.

rr""t 8olutlon.
we have that

By the Cauchy-Schvarz lnequallty

(12x * 3y + 4z)z s 1tz2 + 32 +,r2r(*2 +y2 + 12, - L32,


Bo that -13 < LZx * 3y + 4z s 13. rften x/12 - ylLz - zl4. The extrele valueg occur

Ttre largest value, 13, occurs when

x - L2lL3, ! - 31L3, z - 4113, whlle the arnalLest value, -13, occura ntren x - -L2lL3, y - -3 l1-lt z - -4ll.l3.

- s-53 -

?.L7.

It

ls enough to c@pute the mlnlmumvalue of by 3.

then nulttply

".c6* By Problen 1.10, we have

+ csc6x and

+ cec6x , {1"""2x + cac2x)3 wlth equalLty for sec x = csc x, "..6* 2' ot 8. r 1t14. Now, by the Cauehy-Schwarzlnequall.ty, we have

$"""2*
I

+ cec2x1? + - f,t{"""2x + csc2x)(cos2x sln2x)13 "tr, *)213= f,f Crl'lt = 24 n L6,


g!9 same value as

-tt
ln the flrst
A

rs i[(aec x co a x * csc x

. rrtth equallty

= ttl[, for cos x - sln x, or x lnequalltv. -

llenee the nLnlmun value of 3 eec6x

* 3 cgc"x ls 3.16 3.18.

48r and occurs r*trenx = ttl4. we have

Uelng the Cauchy-Schrarz loequallty,

I+L +Z_ 1 x * . - L J _ * I 2' 3 6 frJ T' , 6 6 ' 1 6 1 6 ' 2


'2 '
I-+I-+ z 236

z = r!*!* '3

L \ L lz l* - * I l_ *L l - ' 6) 6' 3 [2

2t1.l2

2
,

wlth equalltY for x I y = z. 3.19. l{e nay wrlte the glven functlon as

- (x*y+z) [+

**.+]

-3 - 3.

=!rc* y)+ (y+ z)+ (z+rrrffi .#.*] +


Bi the reeult of Problem 3.12, thls laet expression"is not lesg

than i

1t

. 3- - 3, so that

- s-54 -

#" + **; . f ;r+ ' g 2 -t' 3 t2 .


Equallty holds lf and o'oly lf x r z. - I

3.20. ln have + PB- /IlIz,-Ji AP

rtrlch, by (3.20), ls not less than

. Ttre rnlnlmum of AP * PB occurs when we have equallty or when trlcel * = 1{; - ;, or when x - y = ;-|:5. to the geonetrlcal .

ln (3.20), The geoneproof ln dropped

proof Lg slnoet tdentlcal

Exanple 7.1f *e obsctve that the perpendlcular llnes froo A and B to the (xry)-plane foin a plane.

3 . 2 1. tfe h a ve th a t (:ry *yzt,x) 2

s( *2+

y2+r 2) ( y2+ ,2**21,

l*y + yz + zxl s x2 + y2 + rz .
Equallty oceurg whenever x3,22. y - z. to the expresalon

Let uB apply the Cauchy-Sehwarzlnequallty

{cOaO}12}2 {t(arbr)(crdr) + (azbz)krdr) + (albg)(ctdr)+ (a4b4)

( cr 2 * + { 1 ( a r b r ) 2 ( a zb z) + ( a r u r ) z][{ cr a r 1 z+ a z) 2 t"r a r )2l }2

* t"ioi "la|*"b|)'r.?u?*.?ul* "la?r>z


* + +u! s ("i + al+"frcuf +ulrt"l cl+"fltaf a!* a!r.

- s-55 -

3.23.

8y (3.25) we have

t < u.Jt"ff""o"2, u.Jt" " - f"." t cos dt [J*"""t.


' lEfi-T'f,* * ! "o" zt)atJLtzq '

,i,r* * elnz*r)Ltz

when." 4*2 ( tan x (2x * sln 2x), fron r,rtrlch the deelred lnequallty follona.

3;24.

. If re wrlte x. - I x., w hive 11(* t + "2 * .o ro f * o)2 (1 .x, * 1 .* Z* oeo { 1.xo)2

s ( t 2 + 1 2 + o r er r zlt*f + ... *r ll.o ( *i


wlth equaUty lf 3.25. *1 - ... - *o.

g o r o**fl

rf rru epply the cauchy-schwarz lnequallty slde of

to the rlght-hand

[rlr""J 'lrlrq<*r)J '


we have

rn

r2

(n

r2

2l = F z rF lr F ' [rlr"'*'J* [rlr"'J [r!r"'.iJ ,lr"r*',


xflfi fi xl - xn for all l and k, or xl r ... - q ,-k/q o r - 1r . In P to b l e n 3 .1 8 a , - Ll2, Ll3, ar - Ll6 so thst "Zo 1/n. t1 * 4 2 *.3 - 1 , and ln Pr oblen 3.24 ar r ... - a'
A general Chapter V. theory of lnequalltles of thls sort will be glven ln

rn

r2

aa waa deglred. Equallty holds if and only !.f

1
- s- 56-

I I

3.26.

rf rrc rrlte

... . "lrt * "2*2*... +.o*o torftJ + O"[3,

dren, by -theCauchy-Schwarz incquallty,

r.r. *... t"r', +"o*ol. r e *"]:Jt"tffr)' .[*J ["]i .r 1'''


' l1 o t r o r o * " io o j '
x_

I I

rlth cquallty wtrenever ;fi , ,, value, -(+i { r or *

.i-

xi.

x,

r .. .

&

- tfr

. Ttrentntnun

E .. "lultl/2,

r" echte'ed chengtng stgna by the

i I I

'l
3 . 2 7 . Conglder the quedrgtlc f,unctLonof't:

,l

rb --- ---, v-f ta tfiilr(x)t-ffi8(x)12>0, l

I I

rhtch ls non-negetfve and can therefore aot have tro dletlnet real roote. Ife have y - A"2 + Bt + C, nhere

I I

[f ^ - {- l tB

P2 (x)l"p(x)dx

I l

B - -z I r1x1g1x)p(x)dx tgl

pl

c[ s ( x) ]2p ( x) d x. l"
Ttre fact thar t2 - 4Ac < 0 inplles rhar

pl

:l

r2n(*)a*] r2n(*)a*J . U:rr(x) [fr<.rrtx)n(x)axl' \ra , ' (ra [/t,r,o I


;l

ii'l

'l

ll i ,l

- s-57 -

rhe quadratlc y can venlsh for one value of t, caae we have equallty. t6(il f(x)to

eay tor ln ntrlch

But then, by the Lema, alnce


tt

- /t-Glg(x)lz

le non-negarive, g(r)

- to f(x),

ao

that g(x) 1s a multiple of f(x).

t.:t.

r{errrlte o =

,i, nlra"ioi

2arbrfub. .bi). *

rr weaum riret
1a thle etro-

wlth respect to kr ' rnatlon, we obtal.n

'reuenberlng

that J ts kept fked

o=

rlrr"lcuf

* ol.* ... + bi) - r1ol(elb1 arbrr ror+'aJn)+ *

+b?(.i ul+... +"l)). * z 1'J- r


Now we aun over J, regardlng the three terns in parentheses ag conatantc wlth respect to J, and we obtal.n

o s (" i + a zr+... +.3r*i *al* ooo {oll - 2(arb, * arb, { roo + a bn)- +
t

.? , , , , , + (bi +b;+... + ui)tai + ai+... + ai) ,

whlch glves we aee thet

us the lnequallty.

To deternine

when equallty

holds,

each of the Bquares Ln the double sum glven 1,e.,

ln the

statement of the problen must vanioh,

"Jb*-a*br-0
for all J and k. Thle neans that 3 = 3 for all J and k, eo Lhat oJ ok "1 "z "o ..''q'

E"qf

- s-58 -

3.29.

Ue must ehow that the serLes deoote the partial eum

Crl.ao*l

ls ccvergenr. Let So

so - l"rbrl + larb2l { roo + lro\l


Slnce It So Ls monotone increasl.ng (that

.
n), By

lar so 3 sn+l for all


of n.

eufflcea

to ehow that

go ie bounded ladependently we have

the Cauchy-Schwarz

inequallty,

so = l"rl lurl + ... + larrl = t l.r[' { orrr l.ol2l}ttorlt ... + lbofri,, + lbol
tlow the aerlee of p o s lrlv e t e ru s , i, l* 1 2 -. t . r: - A ls c o n -

versent, rhat l"rl' + ... . ,":Tl'=i, .r,l-lrrrr.rr, eo


t r lorf g -r + lurrl2s, Rr where *!rtl.
8o ilrat
3.30.

Hence s ALlzBLlz, so
nlrlaaull
M and N, thet

]"lB

s' exlete.
inequality

rndeed,we have .n".


we have, for

By schnarzte

any poel.tlve

r2(n) - r2(u) = lNz r ( x ) f, ( x ) d x


M

rN

^ f lz
[/tt' ( x) ]

Ll2 tu"l
a Sequence

We know that

there

must exlst

{"n} w l thl l n:c k+o

s6

such that

il*tf(1|l?

= Q, and hence;fur t(1)


number d such that xo, and thle ie finlte.

= 0, for, orhencl.se
tf(x) )2 , a for the
Ife

there nuet exlst alL x greater

a poeltlve

than sone fixed {tr(x)l2ax

would corntradict

asaumprlon ttrat know that,


N

Next, gtven e ) 0, Mor

for all

M and N greater
N,

than some flxed

/ tr<*l l'd* . e ,

lf ' (x )l-d x

< e ,

S -58a -

s o th a t,

fo r

all

x > Mo and * t

t Mo, w e have

lx t r ( *) r - t r ( x n ),' l ,l i " rrr* l t' u * l t/' t' 2 .|l .


which s h o w s th a t l i m [f(x) x+o ]2 = O, w hence 1i m f(x) x+@

lf ' (x )l2 d x
K

'x.

= 0.

- s-59 -

Solutlons

to Chapter IV

4 .1.

By wrltLng F(x) - [f(x)]P a n d G(x) ' g(x). - tc(x)lB, where a - 1/p andBmay recast (4.22) Ln the forn

tg( x) l e, or f ( x) - tF(x) Js and 1/q, eo thet o + B -1 rwe

= igu* f ,r,*,101c1x) [fr,*,u*,J"[{t",.,*Ju.


Suppoee next that we have three functlong a + g frg'h and three positive
r]

constenta orBrY lrlth lt follolrs that

* y - 1.

Itren B'+ Y

- o, ftom whtch

*+1fr'r.
If we wrlte rb | -c 6.Yd* _ tgn J a

.rr 'at/

1 B

f-ralrr-a hl*l

-J--11'a

dx,

rhen

. [l',..]"[fjts ..]'* Jbr*frra* -.*

= .4"[[l', *]*]'" [f, .-)*[t"


.4"[" u*J', - 11" ..Jt["
as deal.red. 4.2. In the last lnequallty, changesI'

se tp - *,

o ' fr

t' f,

uake ana

the

fo, G - gB , and H - hY.

- s-60 -

4.3.

I{e apply H'61derts inequallty

to the identl.ty

3 - 1+ r + 1 - !+,^

t6

tri, #+r*

3ra * L_L<G

= cxyz" xyz. l r l z 1 l* * * 5 1 /3 ( xs + y5 + l* * * *
where rhe indtces a =
raise both sides

"t) ' 1 6 ,

f , B - +,

y =

heie been used. If we now


inequallty,

to the slxth

poner,

we have the stated

wlthequalltytfx=y=2. (Note that and that thle problen is analogous to problena is posslble: 3.12

an alternatlve

solutlon

1* 1 * ! t
x
so that

z -' Y x y z , "E

( * .+*)t =3sv/#.
Aleo

*5 +y5 +15r tr @,
so that the nultipLlcatLon of the problen of the last two inequalltles aleo ylelds

a eolutlon lnequallty. our proofe arlthmetLc

by means of the arlthnretlc-geometrlc-mean because we have baeed between the

Ttrls should not be surprLslng, of the ttiilaer inequalitLes

on the Lnequality

and geometrlc meane.)

4.4.

If

we wrlte

3-l +1 +1 we have
!

'r-JIT;m

1 ,=+* ' lt*

1,

- ; iBr

1, y2l3-f TF I -

3 . (l *i * Llttt(ll.

* b'tt<*' * 12+ z21L/3,

- s-61 -

where we have uied c - t,


laet Lnequallty ylelds

B - i, +,

t - *. ' 3

Cublng both sldes of the


3.L2;

the result,

whlch 1g the aame as Ptoblen

note the obvioua alternatlve the eolutlon of Problen 4.3.

eolutlon

and the remarks at the end of

4.5.

ltrlte wlth

c - m/n, so that m < n.

0 < n/n < 1 and m and n are posltive

lntegers between

Slnce I- * x > 0, we nay appl"y the lnequali.ty and geometrlc means to the expreeslon

arithmetlc

(t + x)a to obtaln

(1 + x)o -

(1 + x)m/n = t(t

+ 11n 1n-nrl/n

1 s f,ttr + x ) + ( 1 + x ) + . . . + ( 1 + x ) + 1 + 1 + " '

+ 1l ox

(n - m)rl = - * tr(r * x ) * ftr* * n) - 1**x n--n


Equallty inequallty ls attalned only tf all the factors

-1*

uaed ln the above

are equal, namely, only tf

1 t x = 1, or x = 0.

4.6.

SLnce o > 1,( l we have l. ' (1 + cx)-r* rlth equality o'nly for
1l n

t,

and Problem 4.5 ytelds


1

s 1*;

. ctx = 1*

x, to the

x = 0.

Ralse thls

l-ast lnequality

power 0,

and we have

r+cu( <( t+ x) 0.
4.7. In the lnequallty that Yo > 1 +0(Y' wLth eguallty nultlpLy this only for last y 1) = L - o f oYr yo - ay > 1 - c. :i If we now of Problem 4.6, let us substltute y for 1 f x' ao

- 1, so that

lnequallty

by Ao, we have

- s-62 -

(ly)o - oao-l(ly)

> (r - a)eo.

If we set a - otro-l and set z = Ay, we have


,Q wl.th equality only for z> (L - o)A0

",

1
A (l.e.r I = l-). 0

B u t A -(9 )F . " o a 'ct' '

the value of the minlmumls (1 - ol <filo-l.

4.8.

Suppose first enough so that

that

1 * ox > 0. < 1.

Choose a posltive

integer

N l-arge

0 < -{/N

Then, by Problen 4.5,

we have

( r +x) a /N= t- $ *,

or

(1 **)o /n =*_ t-F*

1+frx
:,_-=-* TZ
NO

L+ 'r*c F * '

x- < 1. since L - ""2t a n N, we have

If

we ral.se this

last

lnequallty

to the Power

(1 r x )0 > (1 + f t r n = 1 * F x
We observe that tf 1 + cx Ls negative, trivially valid.

= L + c llc .
the lnequal-lty 1n the state-

nent of the problem ls

4.g.

For c < 1 and f or any of the rational lim r- - o, we have n

ntmbers r- euch that n

r (1 + x ) n < 1 * rrrx r so that, ln the llnlt,

- s- 63 -

(1 +x)0=

lirn
r+(l n

(1 +x)n<

lim (1 * r-x) r +rI n except for

- 1*

crx,

which proves the inequality,

the conditions

of equality. (1 + *)o num-

Suppose now that x t O and 0 < o < 1; we must show that is strictl-y less than 1 + ox. To this end, let

r be a rational

ber such that o < r <.1. 0.9. 1, lt

Since (t + x)o = [(1 + x;o/r]r of problen 4.5 that

and since

fol-lows from the result

(1 + x )o / r < 1+9x:
so that

r'

(t+x)0 <

(t *

* ")''

If

we apply the result


regard .0;

of Problem 4.5 to the e x p re s s io n (l-+ ? * )t ,


x as the quantlty x in

where l^te nolt

Problem 4.5, we have

(1+9x)r<1 +
and thls

rc lx r

1*o,x, The extension to problems 4.5 expended are in 1_/n = 0,

gLves the desired resul-t. (tet

and 4.7 is now imrnediate. here. taken, rt

us comment on the labor in analysis

is a wel-1-known fact inequality

that when l-imits

a strict

must be replaced by an inequal_ity e.g., 0 < 1_/n, but 0 = is required H

which equality

is to be considered,

so that a separate proof of strict cases when, indeed, a strict

inequality

in those

inequality

is possible.)

4.10.

Consider f irst equality *o-0*> minimize

x0 wlth

1 + y = *.

Ttren (1 +y)0>1+aywith

for y = 0 or x = 1, so that x0 > 1+o(x-1)or 1 - ct,with equality for x = 1. However, we wish to

*o - a* rather

than x0 - 0,x, and, in order to do this,

- s-64 -

ne make a change of ecale by setting A0, we have

x = Az.

But if'we

multiply

by

*oAo - crx A0 - (1 _ 0)A0,


or

(rd)a - oAa-l(* A) > (1 - c)Aa, I 1trus the nlnluum value is

where lre rrant oAcr-l = a, or A =

)o

ir
(1-a)Ao=(1

"))o-1,

aod occurs at the value x - A.

4.11.

I{e have 1 = cog2x.

L+sln 2 x . 1

5 (cos 6x + sin6 x ) L / 3 1 , 3 1 2 + L 3 l2 > 2 1 3 , 2 l3 6 o s 6 x + e rn 6x ) L 1 3 , whence I 3 or x o n/4.

+ 4(cos6* srn6x). Equalitvhords nrrenever W=


alternative solutLon

SF

may be found by means of Problen

t . t"0.
4.!2. Agaln, we have 1.= cos2* . 1 + slo2* . 2131.3 . 1

1cos3* + sln3x)

+ 13)

Ll3 * ,Ll3 6oa3x + sln3x )zl3 .

whence L < 2. Equaltty

("o"3* + sln3x)2,

whlch ls

equival.ent

to the eolutlon.

holds wtrenever (tro2*)3/ 2 -

(po"2*)312
I

,3

ot

x=1114.

13
I "o"2* + sln2x, whlch ls not greater

4.13.

I{e start than

wlth

rhe ldentity

c-2

zx)alz *
["'"

" (srn2*)a/ 1',"8h. ,#t] ,

- s-65 -

whence

a.-2
1 < (cosox + slngx)zlo(2,
0

1 . 1cosox + slnox)Z 2o-2 ,


,0 ,

wtrence 2'-l

"o"o*

+ sl#x,

wlth

equallty

for

"o"o*

sin0x =- - r ,

x = r1 4 .

4.L4.

tet.Llp

- c and Llq-

3, so that o+

B = 1, and consLder the integral not

{ttt*l

lotc(x)o(x) l8a*, whlchls lolc(x)lBo(*)a*= /.btr{*)o(x)

less rhan t{rt*lO(*)a*)o({bc(x)O(x)dx)8. rf we set f - Fo, g' cB,


/a then F = pl - fP and e "Ll$ g9, and ne harre

t'0, t'o (r(x)lpo<*la*J = )qo(x)a*] 0(x)dx [f Jor(*)r(*) lio,r,x)


with equaltty lf (f(x))p = (g(*))9.

4.15. .

Set p have

, so that L/p + Lln = l.

ny H6lderrs l"nequallty, we

axfb

+ < (a P * 6n11/n1*o
and therefore

"o

- *nr l/n- c( aP+uP) l/P,


when

lrlth

equallty,

the maxLmum, occurrlng

_C

ap n x

- f fbP i'
c -x

or whea

xt-

ca

ilf ',
n rl/ n '
)

l- il^.f ilr l D ta

- s-66 -

4 .16 .

Let p > l be real and l e t q - -+ p -r,

s o rh a rl* 1 P q the functLon xP - (py)x, whose minlmum is


p (1-p)yP-l-(1-p)yq

t. -

lJ e c o n s id e r

ln Problen 4.10.

Ttrus

*P:(py)x>(1-p)y9,

so that

"P

+ (O - l)yq

> p x yr or

1P * ( o : r ) y e > x y , pp'
frm wtrich (+.2) foll-ows.
Frm Problen 4.10 lt follorls tliat the

mlnl.mumoccurg wtren \-/ _1 _ (pl;p-l I p


p

or nrhen

*P

yP -l ,

*P = y9,

whlch ls the condLtion for equality

1n (4.2).

4 . L 7 . W ese t p -*

(n >n ),9

= f, fr t s) , and]-

r - *- i.

Then

* * + . + +{ = **oi rn (t- * - l " '


nrxP + nsyQ + (mr - nr - em)zr mr between the arlthnetlc Ls not less than na and geometrLc means, By the lnequaLlty last expresslon

f*ott ,9ne "r(mr-ot-"df/tt ()


wlth equaLtty oaly lf xP yi

- *6 ,oi
= zt.

r(1 -

* -;

z ,tyz,

- s-67 -

4.18.

We wrl.re HiiLderfs lnequallty

ln the forn

= , where At ,i;l'n'1,0[,!,rJ"o[,i,,,1"0
Then

' rl, tr - bl'

I.rrr", "11t

rl/o

",

= [(rrcr)l/ntar"r)l/e

< ([ercr)t/n{Irr"r)l/q = (I"l"r)l/pt[r!"r)l/q,


wl.th equallty tf
D

"i"t

tl"

"

..-

tP" nn
.q D-C '

nn in the statennt of the problen.

which le the same sg the form glven

4.19.

Fl.rst eolutlo,n.

Let us use Problen 4.18 wtth p = q = 2.

I{e have

L +2 +4 -

r1 Y ,*'.8

L+6

z+G #.

4,

or

7<(x.l+y

z . * z . loLt ?( * . ; * r ' t ' ,

wlth

equality

for

*l#- 6 l # =n l # ,o r x - y last lnequallty we obtaln By the arlthmetlc-gemetrfc

z.

If

we

aquare both eldes of thls result.

the deeLred

Seeond eoluti.on. have flrst

meane relation

we

xt-2 y+4 2- 7m
Tz

txyz

V.*

- s-68 -

lrlth equaLtty only tf

x = y = z.

Also,

i. 3 + !
7 wlth ities, equallty only if

z ix

7 /l z v z y
If

x - y - z.

we nultiply

these two lnequalholds for

we obtaLn the deslred

result,

and, since equallty

the saue values of xryrz, x = y = z. (We renark

we have the condltion lf

of equalLty:

that

we had used the Lnequalltles

ry.3 @ =3 @
r.
x

2.4
y 3 z L >3/ t xyz' -

we obtaln true for

that

(x t 2y * tr:')(! * ! + !-'1 ,- g 3G4 '-xyz' two lnequallties being x = 2y = 42, the conditlon =

316, whrch le condl-tions

enough, but the last eguality, the first

have dlfferLng

a n d t h e s e c o n d l= ; = t ,
x=y=2.

whlch therefore

cannot yteld

4 .20 .

{n Let c, = 2-, i o 1, ' ", wrlte

rt.

Itren .!rc,

t:1-

.B-zi - zntL- z. we
L/2

2t*1'2=l.i=

Lfr Fr "t = (Izi"r)t"[l*{

T-

so that,

on squarlng, n

4(2n - L > 2 <

1=1
and, on dlvldlng equallty only lf ,{,
1

lz'

.,] , [,i,+,J
with

both sldes by 4, we obtaLn the result,

*1 =*2-

"'

= x . n

- s-59 -

Another

soluti.on

parallels

the sec@d eolutlon

of Problen

4 .r9.
4.2L. Let us use Problem 4.18 agaln wlth p n q2. We have

a * b * c * d -< ["+. #

*;fo + #]t"[.(c+d) + b(d+a) + t/t, + c(a{:b) uto*"lJ *


Thug re obtain

= b, c3 = c, cO where we have used c, - a, c2 - d. the tnequallty

la(eH)

(a+b+c+d)2 + b(d+a) + c(a+b) + d(b+c)l

< -

c u'

wlth

equallty

o,n1y lf

c * d = d * a.= a * b - b + c, or a b c - d. therefore S .
ml.n

Ttre value

of the mlnl.mum ie

2.

4.22.

We shall

= b2, n .2, use Problem 4.18 wlth c, - a2, "Z "3

and wlth

poq-2.

Weheve

L-12+b2+cz

l2 ro I ^ 21L -["2tu*'l * b2("+") * .G-+hz " 'r"orj''' Lb+c "+' i"j


whence

.2(u*") + b2(c+a) + c2(art)


Norrequalttyholds forb+c'

(e= -LJ-D b*c

.2
' c{e,

2
.r' c a*b

cta-a+b,

ora-b-

c.

Thus r

- s-70 -

w hen - b - c , S 2 +, a

6a"

and S = {, 6b"

s 2 +, 6c-

whence 1/S < 6a

l /s s 6 b 3 ; l /s s 6 " 3 . r f we add these last


l /s < 2(a3 + b3 +
onlytfa=b-c=

lnequalitles,

we have

"3),
I

wh e n c es 2

z( a3+t3+c3)
wh e n c es > f 3 1 2 .

wlth

equallty

R,

4.23.

l'fe uge Problen 4.1-8 lrlth

any p and q satisfyLng

(4.1).

we havc

L - a 2 + b 2 + .2 - ( *, ^. ^2**.r .u2

+!.

". "2)

= |.+
(aP

* ot * * .zluet"a"2* ueb2 .'"2 * ) "e"2;1/c, bP cP


"2-P)9(^24 * b2{l * "2t)P > 1, wlth equallty

or

.,,2-p + b2-p *

lf a = b - c.
4.24. Ttre case that asatrne that l/p l/p + L/q = 1is < 1. Hdlderrs ineguallty, so that lre may

+ l/q

Ttre number r def ined by

;= t-F-a
ie greater Pr gr fr than 1, go that, by Problen 4.17 extended to arbl.trary we have

1.11

d*d*{,
Pqr

r*.

By (4.24) we have b rb

I rer ax< l/ a \a
=(b

rea*j''' [i "
_ alLtt[orn

t'n[io"u*]t"

dx

]"'[1'-'u*J"n,

- s-7t_-

wlth

equeltty
'

only if

f(x)
1t-

and g(x) are conatant.

Ttre constanr k

turrrs out to be (b - a) " ^ . 4.25. fire aolutlon lnequallty that, for paral1els lnstead that

of Problen 3.30, where 1re uae Hiilderte Indeed, we have first

of Schwarzrs inequallty.

any M and N,

,( r2 (w ) - r2 (u ) = l N zt( x) f x) dx . .M rN r Upr N _ \i./q ' lt trllrM/tm)1"y lPa"l lr ' ( * ) lqa* l lr


the detalls fron this point on are essentlarly identlcar t,o

of thc eolutloa

of Problem 3.30,

and we omlt theu.

4.26.
.t

rf we aPply the reeulte

of Problen 4.18 to the ldentlty

\-'

n(n-+l ')-1+z? +3+...* n = t r . + + : _T - r' z + r r . . . + - - . _ 1 E.


where ne aet
n6

1 ztxZfr+

...

nrrx

t#o

-.F'

"i

- i,

"1

= q.

and b, = 11fi,

we have, for p - 4 -2 ,

n'(n+L)t,, --'(*1 wlth equallty

+ 2x

+ z + ...+ n x o )(i t.
... E X

...

T" n

n\

for

*1

= - *2

4.27.

Use the ldea of the solution


tdentlty

of Problem 4.26, together wlth

the

n-(n * 1)' _ ,3 , _---T--.? 23 + ... + n3

t,

F.g - .!r* 4tt *f,'tt,


\ vr-, andtakep-3,9-3/2Ln rhe lnequallty

o f P ro b le m 4 . 1 8 .

J.-; d,!

- s- 71 -

Solutions

for

Chapter V

5.1.

If

f(x)

= 1og x,

then f '(x)

= l/x f(x)

and f"(x)

= - L/x2, wtrich l-s negative 0 ( x ( -, so that the

for x > 0. inequality tl

By Theorern 5.1, is reversed in

is concave for Ttrus n

(5.3).

locl

1t 1= los . * * t"r x, = ) ,r=, *J Z ,1, *r"gx.=1og 1og J1*t/t=loq,,ffix ,F..n ,]r"1'" ) *l (5.18). slnce 1og x is not linear x, = ... = xn. over any intervar, equallty

*r)-

whlch is hol-ds lt

(5.18) only if

5.2.

we know from problem 5.1 that with the direction

1og x is concave for x > 0 . reversed, we have

B y (5 . 8) ,

of inequality

t.*f.i,or*rlz i-c, 1og. = 1ogf, *lt , x, \i = 1 ^ -/ i=l


i=l ^ whi c h gLves (5 .1 9 ). Eq ual -i ty hol ds onl y if * 1 = ...

=x

n o, * cx,,= l.

5.3.

In (5.19), Hence, for equallty

let

us set n = 2, dL=

L/p, aZ = L/q,

so that

A1 t 0 and A, ) 0, we have Al1d2 A1 = A2. only if If

S olAt

* ,2A2, wlrh we have

only if

ne now set A, = xP and A., = !q, xP = yq.


nPi

(5.20) with

equality

5.4.

In (5.19), and follow with

we set oi = L/pL, so that the steps of the solution only if

iltoi

= 1.

If

we set A. - *r' (5.2L),

of Problem 5.3, we obtain

equality

*l-

Pr

= *2

Pz

=x

Pn n

5.5.

For any o, > 0, g > 0, we have immediately f(x) = x log x (verify that f'r(x)

frorn the convexlty for x > 0) that

of

ls positive

- s-72 -

"#
@
whence t#l equallty ' . holds only

rore+it . f ros * t losB, "


(oo3\Llz'. lf o = B slnce x log x Ls ttnear nowhere,

5.6.

Ttre fact lPl

that

x l-og x is convex (Probl-em 5.5)

aesurea ua thet,

lf

1, Pi > o,

(ot*t + "'
S Pl*l

f p xo)J-oe(nrx, "' +
lo8 x, + "'

* po*o)

* po*r, log xo

Pr*r . = rosf [1
whlch le (5.22r. Equallty

Pr,*ol *o ),
= ... = "r. x > 1. Equallty

holds only for tl

5.7.

Obsenre that holds only lf

the functlon c = B.

-6

tog x l-g convex for

5.8.

If

we play a blt

nith

the gLven lneqtralltyr

we have, slnce

e{n x, > 0r

that

1og sin *1 * ... r 1og sr.n xo I n 1-og "t"[4*]


or

(r 1*sln 1 < los sinfi F I i olrt"t nlr\j


Ttrle le the form (5.3) Ln this case, f(x) of Jeneents inequalLty After for a concave functlon, that the functlon to obtain the

- log sln x.

verifyLng

Lo9 sln x ls

lndeed concave, rre may reverse Equallty

the steps *1 .

requLred lnequaltty.

holds only for

- s- 73 -

5.9.

If

f(x)

ls twice dlfferentLable,

then F(x) = 16t f(x) cannot be positive

is concave on

I because Prr = (ff r' - f '')lt' posLtive. satisfylng then

because ffr cannot be

Ilowever, the concavity nay as foLlows. If

of F(x) nay be proved ln a more on I,

x, and x2 are any two points

'[+] =roq'[+]
2. l"s
f f(x'

) + f (xrF1

T-J
+*t"r f(xr)

(f ts eoncave)
(1og x Ls concave)

f rosr(x r )
* I r<*r) *
Thus, if xrr

r(*r)'

* z'
*1

, xn are poLnts of tr, we have

1og f

+. .. * x r nl

"J

, ! tog f(xr)

+ or. r*

tot

f(xo),

. t f ( x r )t ( x r ) . . . f( xr r ) Jr ln t[:#
as desLred. Equallty holds only for *1 = ... thls woul-d imply

bl
= *rr, for lf that f(x)

F(x) = 1og f(x) tlal is

were linear

is exponenf(x)

and therefore concave. Renark.

convex, contradicting

the assumptlon that

By using the stronger

form (5.8) of Jensenrs lnequallty we have, for posltlve

for

a concave, rather ...,

than convex, functlon, Eo, = 1,

numbers 0rr

0n with

rogr[rlr,r-rJt

rlro,

rogr(xr),

- s-74 -

or

. tr(xr)rot... [r(xo)]oo t[rlr*t],


whlch l.nvolveg the weLghted arlthmetlc - *rr. f(x) = *P-1, f(O) - 0, le increaelng drlch we Day or becauee y > 0 and and geometric meane. Equaltty

holds only wtren *L = "'

5.10.

Because p - 1 > 0, the functlon for all

x > 0 and hence has an inverse from the relation y =:(p-1,

functlon

g(y),

calculate

namel-y, x - y1/(l-1), funetlon

tlp + Llq, - 1, x - y9-1,

whlch ie an increasing

vanlehea at y * 9.

Now

fa p-r.<tx=f a P ^ r -l*^ ^1 o
and the flgure below ehows that

= o, = I)ro-tu, f ,
o

the area of the rectangle

wlth

sLdee a hords

and b cannot exceed the sum of the areaa A, and Ar. only lf b f(a) - "P-1, or, what Ls the same thlng,

Equality lf aP -

bQ.

(e (b) ,b)

(a, f (a) )

(a, f (a) )

Flg.

- s- 75 -

5.1-1.

An lntultlve

geometrlcal

proof

ls

clear

if

we nodtfy solutlon. for

the solutLon

glven

to Probl-em 5.10 and use the flgures a sol-utlon case that that ls applicable f(x)

in that

However, we glve exanple, Ln the

ln a wl.der context; and g(x)

the functlons

are not contlnuoua,

the solutlon

nay be nodlfled If a llne

ln vl.ew of the nature and x = g(y),

of the dlscontLnuitLes. then the forn of A, + A2 suggests

we set y = f(x) Lntegral

1( crB) ,
I (0,0) wlth the upper llmtt

V dx*xdy, SLnce y le poeltlve

to be deternlned.

and

g(b) ) ar we have

A r* A r=

I (0 ,0) t (0 ,0)

(s ( b ) , b )
ydx*xdy-

(a, f(a) )

ydx 1(s(u),u)

1 (e ( b ) ,u )

Ydx*xdY-

( 0,0)
b

d(:ry) 1(s(t),t)

(s( b) ) ' b = g(b) xIl; | (0,0)


> ab. that, as we traverse

Ilere,

we have agsumed we meet the polnt I above.

the curve from the orlgln, (g(b),b), as ls

(a,f(a)) The other the point

before the point posslblllty, (a,f(a)), that

the case ln Figure (e(b),b)

we meet the polnt

before meetlng

Ls not llJ-ustrated,

but in thls

caee we have

A rfA r=

(o'o )

1(a, f (a )) d (x y )J 1(a,f (a ) )

(e(b) 'b)

1(a , f (a )) *dy J

(0,0) z ab,

- s-76 -

which glves oE E 3 g(b),

the same reeult. whlch ls

Ttre case of equallty II.

holde r*tren b - f(a),

shown in Figure

5.L2.

Uel-ag the reeult

of the precedlng

problem, we have

.* . J""rn t dt + fsro-lt
oo

at

=(-cosair)+x s ln -lx + or eos a * ax ;! x sirr-l* + /;, Equarlty hords onry for

and the reeulr follows by settlng a a r/2. x=1. 5.13.

ustng the reeult of Problen 5.11, we have, slnce f(x) 8(y) are lncreaaing,

- x5 + x and

ru . J"{*t * *)a* + Jbgly)dy,


OO

wLth equeltty forb


equallty, a > 0. 1.e., Hence

= a5 + a.

To evaluarc 12 e(y)dy, wemuat have o


ie the only solution for

= 1; this a- + a - 2, or e

L.2- Itt*t *
o

x)dx *

g(y)dy,

B (r)dy - 4 / 3 .

5.14.

Using the foru of (5.3) we have

for

conceve functlone

(with

equer.lty reveraed),

f; !.:i

- s-77 -

r's[r.j, 3 ->r T 1og(1 * nj o tlr


(1 +so >
with equality only lf ar" n II (1 + ar)r & k-l
E ... E I .

"t)

5.15.

Ttrat f(x) ft'(x)

= Log(l + e*) + .*)-2 ...r

ls

convex follows for fron all

fron x.

the fact

that

= .*(1

is posltlve follows n

Thus for that

any n real

numbere xr,

xn) it

(5.3)

t1*e

*1*"' +:(

] "t - lr [ t * "* n] ,

['

...

[".t

"*]""]"

k-1 (

*.\ ft|,t

l
or ak - exk,

Slnce the an are posLtlve, we oay set 1o8 A - * k ' whence


., (1 + (ara, ... "o)r/n)n

n < ,II_11 + an), k-1 holds "o. of Problen 5.15 nay be only lf x, '"

whlch ls what we want to show. or, what Le the same thing,

Equallty

al = " ' =

5.16.

Ttre last wrltten

dieplayed as

lnequallty

ln the solutton

t. + (ara,...

"o)t'" = Jrar

* 1)1/o.

- s- 78 -

If,

in thls

inequalityr

rr replace

"k

by a,/b1,

we have

. jr4 /" ' f nlrol'". * ,|]"", Jr[,


and, if we nultiply
inequality. Equality it

both sides ur nuf/D, w obrain rhe desired


holds 7f. ar/b, = .. . = arr/brr.

We remark that

is quite

simple to show that

k=lK by applytng

il "t/" * ftorl"* .ft-"|/" r.ft-,* + bk+ "o)t/'


k=lK k=lk k=l resul-t of this
1t^

the orlginaln

problem to the product

k=l

n ([ar . + buJ + .n) ' ' ^' . rr

5.L7.

Ralse both sides of the lnequallty and recognize

in Problen 5.16 to the n-th 4.

power, the

the geometric means g, and

In accordance with

remark at the end of the sol-ution


n

of Problem 5.16, we also have

(s.+8b+g")t
K =I

wtere g" 1s the geometric mean of the posltlve

numbers crr

...,

cn.

5.18.

The method of solution as before, write A1 = {t

of Problem 5.11 uray be adapted here. and oz= f,b q(g(v),y)dy,

If

we set,

n(*,f(x))dx

rhen we nay

At * AZ ln terms of the llne

integrals

A .,*A r=

1(s(b),b ) p (x , y )d x f J (0,0)

q (x , y )d y -

1 (e (b ), b ) j p (x , y )d x (a , f (a ) )

A .,*A r=

1(a,f (a ) ) p (x , y )d x + q (x , y )d y I (0,0)

1 (a , f (a ) ) q (x , y )d y , J (e (b ), b )

- s-79 -

\.,

eccording as (a'f(a)) the path of integratlon q(xry)

or (g(b)rb) startlng

Ls encountered flrst at the orlgin.

as we traverse and

slnce p(xry)

are both posLtl.ve,

we have, Ln both caaes above, that

A r +A ? Z

) (0,0)

r(e(b) b ) ' p (x , y )d x + q (x , y )d y

1 (e (b )' b ) = F(x,I) l' - F (g (b ), b ) | (0,0)


> F (a rb ), and

A l +A 2 >

;(a ,f

(a ) )

J (0,0)
F(xry)

p (x , y )d x + q (x , y )d y
(a, f (a) )

(0,0)

- F(a,f(a))

> F(a,b) Bquallty occurs only lf b = f(a), or a = g(b)

5.19.

rf

we rotete

the curve y = f(x)

about the x-axLs,

the volume of the u* - {" ny2a*

solid

of rotation - {t

betrreen x = 0 and x = a is where 1- is the dlstance

given by fron

and F v* trold

n*y2d*,

(0,0)

to the cen-

of the solld

along the x-axis. By the result

stnllarly,

v-- - 1b n*2dy ana yo

9 V., - {b n-2ar. -yo

of problen

5.1-g, we have

F v* + t v.,= f ,*r'* * l'n*r, ayz[ x-y'"o ^2a2,


for rhe functlon F(xry) - t *'r'has the properry rhat eo that, in dF - n:ry2dx + rm2y dy. thts ease, holds when b = f (a),

Equaltty

- s-80 -

xv
(a) The function restrict triangle

t'"=i

'fl

^ 2 lt (^ ))2 .
so that rile must so that the

5.2O.

cos x is

concave on 0 ( x < t/2,

the angles x1r *2, *3 to be less than rl2, is acute. By (5.3) we have

L. x, * cos xr.* 5(cos with equality if

cos xr)

< cos = n/t.

"r

+_*z+ *:
3

= L/2,
* eos x, * trlangJ_e.

*1 = *2 = *,

Hence cos xl

+ cos x3 assumes a maximtnnvalue 3/2 for (b) As ln the preceding

the equilateral-

problem, rde must restrict

ourselves

to an

acute angle because cos x is concave only over 0 < x < rT/2. By Problem 5.9 we have

cosxr cosx2 cosx3 .


with equality, for

"o"3taa.pJ3]
aBain for

= L/8,

and hence the maximum value of the expression, x, = *2 = xy the equilateral triangle.

occurring

(c) Agaln, we must restrict

ourselves to the interval-

0 < x < Tt/2, should be

where both cot x and 1og cot x are convex (tnis verified; use Theorern 5.1). We have, by (5.3),

fact that

log cot x, * 1og cot xZ * 1-og cot x, S 3 log cot or or wLth equality log(cot cot xl for xl cot xZ cot *3) . 3 1_ogcotgr cot x2 cot x3 < (cot *L = *Z = *3. 3 =

*1**2**3

3-)

L/36

trIe observe that

the maximum of the given expression tan xl tan xz tan x3 assumes its

occurs minimum,

whenever the reciprocal wtrich Is 3/1 .

- s-81 -

(d) The function we have

sec x is convex on 0 < x < Ttlz, so that,

b y (5 . 3 ) ,

whence sec xl * sec x, * sec *3 , 3 sec n/3 wlth equallty for tie equrtateral triangle; - 6, the mlnLnun value

the gLven expresslon ls then 6.


(e) Ttre functl.on tan xlT ts convex on 0 < x < fi, so that, by (5.3)

t""[#J
or

/xr f x,, * x.1

f{."o xrlz + tan xrlz + tan xr/3)


xl

xr

xt

ta n f+
go that trLangle.

ta n f+

tao, t' :.3


ls achleved

r an Ttl6- 31 6 ,
ln the ease of the equllateral

the ml.nLmtrn 3 | 6

5.2L .

r f we set yl = f(xr),

lz=

f (x r),

y 3 = f (x 3 ), a n d re c a ll f ro n a n a lyt l c

geonetry that the fomula

for the area of a trl.angle

^-+
wlLl be posltlve to (xrrl3), lf, (xr,y2) othenslse, precleely A w111 be negatlve. the dlettnction

*1 vl tl *z v2 tl
Y3 1l
the edges fron (,xrrY1) to

*3

when ne traveree the LnterLor

of the triangle

ls to our left; 5.5 ttrat thls

I{e eee fron Figure

between convex and concave functLons.

- s-82 -

To prove the result equlval.ent to

algebralcally,

note

that

the condltLon

*rY l 1
0<

*2-*1 *3-*L

tz-tt It-It

0 0

= (xr-xr)(rr-rr) - (xr-xr) (rr-rr),

and to

*1-*2 0< *2 *3-*2

yL-yZ Yz t3-yz

0 I0 gLvee

= (xr-xr) (v,-yr) - (xr-xr) (rr-rr),

and that these two Lnequau.ttee gLve us (5.24>, whLch in turr us (5 -23). The ateps precedr.ng (5.23) are reverslble,

so that the

condLtl.on ls both neceaaary and suff{clent. 5.22. Let x and xo 1le ln a ( x ( b, x


*o. We have already \

"

eeen that

ri(x)sW
for Ax > 0. lncreaeingr slnce f(x) is conrinuous and ri(x) of this

is nonotonlcally aa x * *o,

tr may take the llnlt

lnequallty

x ) xor to obtaln

L1n fi(x) x')x


' x)xo

<

f (xo +/lx)

- f (x)

Thus, as Ax + 0, Ax > O, we have tin fl(x) ? x*,r *r*o o But we know already that <

ff (xo).

fi(xo)

S fi(x)

for xo ( x, wtrence

- s-83 -

l].n fi(x) xn(


*t*o The last o

> fi(xo).

two LnequalitLes

show that

lln fi(x)
x+:( *t*o o

- ff (xo),

whlch ls rltrat we want.

' lhe result - f'(x^)

that

ltn ff(x) x+tx *a*o o Is hmdled analogously.

5.23.

Let J* - ff(x)

- f](x); x.-

lf If

Jx - 0 for

some value x ln I, an lntenral lf xril

then f'(x) J* on the Jx, > 0

exLste at the polnt y-axla by.I*: f'(x)

J* t 0, we def{ne Nore that J", and J*,

< y < fi(x).

x, and if

and J1t > 0, the two l-ntervals for both f'(x) each lntenral and ff(x) J-- contatns x

have no polnta functlons

ln comon, slnce cannot

are non-decreaslng a rational polnt

of x.

on the y-axle, polnta, that

there le,

be more Lntervale of polnts

Jx than there

are ratLonal

the eet

x such that

Jx > 0 Ls denumerable. ueee the faet that ri(x) aad ft(x) are

I{erremark that non-decreaslng. a function every point lnterval F(x)

the proof lt

Actually,

can be proved by elenentary and left-hand Ft(x)

means that at of the

poeeeaelng rlght-hand

derivatinee

of a < x < b hag a derl.vatlve poeelbly

at al-l polnte

except

at a denumerable eet of poLnt*r

- s-84 -

5.24. t{e set r r<*,yldxdy and call g = f - rnr so rhat // g a*ay = + {/ "r rlto
r Let us use L --Z (5.26) ln the foLlowing calculatl.on:

rr rt , = 1 // roe(r + g)dxdy II log f(x,y)dxdy + fir- D. nr- D,

lo g n .# //lo e ( r + t) a xa y
r
1. = m, // + D. fi a"ay 1og nr-

(loe'*
which proves the result. lf f ts constant,

Equallty holds only lf

g = 0, that Ls, only

You might also like