0} a py,
hence ¥(0) <0 for #11 x.
susie xn «evo f
WC0> of ts an open subset of K
a> "aa
:
a
leeceupiew asa
h ;
wv)én >
Therefore, z,cannot be approxi
functions «
QED.“
EXERCISE 3.~ Lat 0
0
for avery 2 © Te
souurr0¥ a
(ey Suppose p is onfinice, tence t=O), » Hoh) <= vR-
bet B+ X= (UX U oe UF,
cyay)e The B's are painise
disjoint and has finite neasure.
Define £ by
oye i te cen, amd te) >0
mE)
“1 if xe5, amd eB.) *
40 90 tor every es mk ADs ee te tte
over ali n such that p(E,) > 0, Lebesgue's monotone convergence
theorem gives
fi ae 0
reek,
' Gd) By Exercise [7], t€ £ © L4G), then,
[x sleet > 0}
has o-finite measure,
The -converse implication of the exercise follows immediately
From this.
QED.
j
te iim irr Sorin nnaain
CHAPTER
EXERCISE 4+- Prove that the supsenus of any cotleation of conver
fuvetions on 14,Dl te oonves on 12,8 and that pointuiae ond upper
Linde of sequence of conver funetions are conver.
Give an example of the lover Lint of a eaquence of convex
functions vhieh ia not conves.
sowuTroN
Let acxcy )f)
results,
To give an example which sutisfies all requirenents of
the exercise, note that n(x) = -[x| is not convex on Rand
hG) = inf fxs], xeR
Theresore ve consider
fg)
fph@) = Es PEL,
Then Lim inf £, (x)
h(x), and £, 1 conver for all n
QED.
EXERCISE 2. If ¥ ta conver on 1,5 and tf @ te conver and non
decreasing on the range of y ,prove that @2e ie convex on a,d-
For ¥>0, shov that the, convezity of loge impliés the convesity
of 95 but not viee ve
sowrrow
lit Ge fib stenele Gealy, weld], The
s2) < A962) "2 G90) Co cones]
HCD) & afl) © CD49) Conen decreasing)
#{sete9#C-9 409) € 24D) + C-rIACHON) C#caNvEx)
Ths goy is convex.
Now Logy is’defined for @>0, and (x) = expx is convex
and nondecreasing . By the preceding result, #= @sloge is
convex if Loge is convex .
The converse is not true: (x) = x is convex, but Logx
ss en soon Seton,
i QED.
EXERCISE 3.~ A
euch that
june that q 4 @ continuous real function on)a,D
yt 1
eB chats eho
for alt = and y € e,Dl .Prove that 4 £0 convex.
souurtoy
Consider the set of dyadic numbers
which is dense in (0,1).0
By repeated applications of the hypothesis, we prove
easily Sy induction on m that
O) Wes G- oN ere +d-Nw re
Define fyg on [0,1] by F(a) = aC ax + (1-A)y)
BC) = 4¥G).+ G2) es
Then £ is A.s.e, and g 13 continuous.
tet v= | ner: ox) <800]
U 4s closed, atace U is the éomplenent of the set
freon sex) > ear}
and since fp 15 lisse,
Bot Sy (#) 4 contains D, hence U = (0,2) -
Therefore for 04.1,
f(x + (1 -d)y) € AEE) + 1 - 2)F0).
hence £ 12 convex.
emaRes
(a) im the proot , ve suppose only that £ Se lover
(2) the conclusion of the exercise dose not £oliow if
continulty of f Le onitted-trom the hypothesis, as shown the
following example,
exaupLes (see @] ch, vtrT)
Re tet emis
Choose « (lane) basis of the Quvector ape
base De Cede gy sand define aCe, ) =, aley) = 0 SE tri,
and extend 9 toe @-tinear map of R inte Q
If y fe convex , then g ie bounded above oh Ja,bf
hence {5 bounded {ns neighborhood! of 0, hence y is continuct
The continuity of ¢ now inpiies that @(R) containg an interval ,
contradicts the fact (R) = @ (Note that 9 {2 aot constant on
3-2a0.
Therefore q Ls aot conver » but
1h she eden.
f 44 2 complex neaouradte function ont, pie
EXERCISE 4.~ supped
@ positive meaaure on , and
fe ten = arte
bet alps yipl
Hence
Pema ten) Pru,
fe) dane that Uf], <2 for some 7
0, and [FL < oo for sone OC rem
Thus for 0 [freer > ef}?
Me
since ain [pcb]? «1, 48 fortows eet
pis [+e]
Lin int Het,» @
hence ae
(2) Lim ane TEL, > HEU, 1D
Now by Lebesgue's monotone theoren
Sli bo wpa
On the other hand
+ faites af Mla oe
hence we have’
"aml oP =a nya”
sep >a
Prey s(uee map
ebatnser?
pew ise,
“as
but
ceey
xin sup Lh Lilo
Y<
hence
(om sup 11, 4 Hel,
By (*) and (0), we obtain the Ldentity
hin Hel, = Let,
and now (4) gives the desired identity
EXERCISE Sim Assume , dn addition to the hypoth
that
is of exerotee(]
eerie
Ihe,
(8) onder what conde:
and ISL, = Is1,coo? .
(a) Prove that UT (p> ECM) if O< rk ay Onder vhat
condttions do th
(a) Prove that ff OCS og em
fone does tt happen that 0 < aca
1 tuo spaces contain the sane functions 7
(4) Aamune thet APN& om for sone x > 0, and prove that
tin Wh, = ta ( [yee rian)
Pro
4f ezp(ams) ta defined to be 0.
souuTroN
(Let ocr cscee, ander ps. p>, hence
He) = xP As conver on [0,201
Applying the Jensen inequality te g sIEI", with #(x), + x?(fuera forerran = fiterae
len, < ret,
1 ‘The inquality is trivial for s #0 .Thus (a) is completed .
Hence
(a
tence by (2) :
Hel, < o() pee
Since e@)or se as every there 18 a maber a >0 such
thea) > te xe
pot tee tg. y with m(E,) + 6,20. Thon Hh, 1 as
poe wre byCiT(e)°? fe monber” fs hosen such thad ?
Now , we proc
Wy, + ait< wb itxeee
Suppose fy yescey f. tave been constructed , choose a,| 0
3074
oe ee
3 ot
bet a, Dix, + men) %
& tr, URL
aha) <8 cat ta Meee tetas
waning, » [ers
seee at 8, one)? 8,
6) ay eercise JO). 6,0) SO +a xe
ses ie decreasing 5 and
The sequence (6,0) # + 242
converges pointwise £0 "High
Fix ak, by (2) there is ¢ number M >a such that
© BG)
fIPap #0 (The convergence is unix
fa
formly on Band B)< =
hence finally
£1Pap <2yge for every €>0
that is ua Tf, :
(44) rises note that 16 18m a, + as finite, then we have
in ine (ay) © 8 = Hm 50p bye
consider hyn YCIEQIP + 141) = Ue EIR. By @) of
hg 20 for every m, hence we nay spply Fatqu's Jems, Since
dat hy = 24,141?
ay, fieiran 0 , hence s#s'
© Thus ve can define an injection from K to S (by the axiom of choice)
hence § is uncountabie, Thus L%(T) is not separable -
a) iii QED.
%
EXERCISE 4s
4 mazinal orthonormal eyetem w
Show that # is separable tf and onty tf 8 containe
uk ta at moat countable,
souurtow
The necessary condition follovs fron Exercise I]
Conversety, suppose [uy {n= 1,2... be # maxizal ortho-
conal system inl which is finite of countable » Let
t ;
2y tte | TE 8 Gr ree a+ 10}
shen S is countable and dense in H, hence H is separsbie
QED.
5
eat
cxcecase S.n if wa fe ste sa herd & fe 6 continous tinear
Yinetional on #, prove that H! ie @ veaton apace of dimension 1
(untene Wet, foe 8).
soLurtow
Lis continuous hence M = L"4(J0}) is closed. L #9,
hence MH, Thus 3x, @M with Lice) # 0. xenyetze with yo Mt
and to € M, Hence Lye) = bles). # 0.
te y ME, 3n © © with Ly = rLyes hence Liy= aye)=0 5
hence y= Aye @ MAME SO, hence y= Aye
Thus fys{ 12 2 basis of wt
Qe.
yenai eet tn
EXERCISE be- Bet Ju, n= tetseee be an ortho
| Skew that thie gives on example of 2 closed and bounded
et which ta noz conpact . Let @ be the set of atts @ of the
form s
e- Sau, sure leiak.
Prove that @ te compact’ (@ te called the Htbert cubs)
Nore generally, tet Jd, Be a sequence of positive nunbere
and let 5 be the eet of all = @H of the form
Prove that 5 ie compact if and onty 3s va! <0
Prove that de not locally compact.6
SOLUTION : "
Ne use the following theorem:
Amotric space K is compact iff every sequence in K has @
convergent subsequence
= VE (ythanoras’ theoren) Af a fm and
Tut 1 va, the set fu, Ugseee | ie clearly closed and
f Since Tu, - ust
bounded , but contains no convergent subsequence, hence ts not
compact. Now BY(0,1) contains the sequence
u,| which has no
convergent subsequence, This shows that 51 (0,1) is not compact,
hence 0 has no compact’ neighborhood -Thus H is not locally com
Now let $ be as in the exercise,
as suppose Sia? cas, and let Gy) = (ds
sequence in S.
peta) bea
ie: We define by induction on a the integers NyWMays.s.sand
as follows
yan the sequences (xy 1). Oy, )
w Fa (Det
0, 1¢
ro
rar Fa
imensional.
menedlern,
At Is evident that
Ul + tel = I~ tance eg
Tais shows that L3(p) 1s not strfctly convex,
Next, consider 1(u), There is a measurable set E such that
PE) > 0, and 2) > 0,
Therefore, ie
Boa a ami ght bees
limplies thet £ = g.
x
ve
Lahore
eke
2
thon HRs gh» 1 andtnd=a, e438.
Finally, suppose X is compact, £ © €(X) is non constants
Therefore If] takes a value b < HEI. Let
acfxs leool> ders dot
nef
leonl< Hien Zo] ;
‘Then A and B are nonenpty, closed, disjoint sets in x.
By Urysohn's enna, there is # continuous~
X01] wea, eae
Hage E, thong ff, and
Tet tel = pce 8
Thus CC) ts aot strictly convex «
ato,
EXERCISE 26- Lee ¢ Be the space of abt continuous funetions onto,11
vith the oup nom, Let W conatee of all f €C for vhich
:
i :
:
Prove that M ta elo)
no alenent of minimal norms
4 conver oubset of C ubich containe
sowuTrow
It is clear that M is closed and convex. Consider the Last
assertion of the exercise - (2543)
Let g(t) = @ = 20) PMA LZeoey f 6 10,2)
p
Te f+ TAM + Ht 4s clear that £, © My since
+ 1
hays flo Coco.
+
Now since
Vegle = 2 and He,t, f1, ve have
HA Je aw ~ee
rerefore
ainftl: fen joa.os
oe :
suppose f € M with HEP, <1 (This will lead to a contradiction).
thus fof GM, Atefl, © 1), 20 Me may suppose £ is real-valued
Since Lei, SHEE, |, the inequality
fis »fie i fe "a
Hel, 1, hence Lee) = 1 fer t © (0,1). £ 48
smplies IEE,
yeal-valued, hence £(t) * +1 or cL. The continuity of £ inplies
either £41 or £41, hence £ &M, This contradiction shows that
If fem, then WE, >2
This nin [EE +1, and there is no element in M wbich
Tew f
hag nininal norm.
REMARC.~ Rece22 that avery clored conver set in 2 Atibert space has
fb enique element of minima nore, Z) and (3) stow thet ents fatte
in the case of sone ce
snach spaces
EXERCISE 3.- Get W de the aut of att f €24([0,11), relative to
el
show that M ie a closed conver
infinitely many elonente of
1 of UF (C0413) whioh eontaine
inal nom.
sowuri0N
Tt is clear thet £ € M inplits
ie reese eat ca a
tence M contains infinitely many elenents of minins1 norm,
ote that in(dJand{G], M is the set fxs tx
suitable continuous linear functional, hence M is closed and convex)
QED.
wa], vere Lis a
EXERCISE Wen tet f Be @ bounded Linear functional on @ eubspace of
a Hibert apace By Prove that f has a unique normapreserving rsten~
Mon to d bounded Linear functional on B, and that this exteuston
vantohes on HE
gue measure, suck that i
9s
SOLUTION . (We do not use the Hehn~Banach theorem).
Suppose £ : M—€ Ss bounded. Hence f 4s uniforaly conti
nuous, hence £ extends uniquely by continuity to W, Thus’ we cam
suppose that M is closed, so that M is a Hilbert space itself,
Therefore, there exists 4 unique x @ M such that
£0) = Od, y © and HED = Ie
Define F by
FOO) yew
then Flys fy Flybe and WFD = De = HED.
Now. suppose Fy is « nom-preserving extension of f.
Then FY) = x + yew
Since Fyfy* £4 We have
Gals Gm . ren
hence x- x, @M! This, by Pythagoras’ theorem,
2a a
It wad see.
‘Therefore,
Deh xt, Webs Bed and rts et
imply x = x,
10, hence Fle Fs
QED.
EXERCISE 5. Construct @ Bounded Linear functional on sone subipace
of sone t}(h) which hee two (hence infinitely many) diatint nom-pre
rotig naan actenaions t0 11 (4).
souurt0w
Let x= fa,b} and define pClaf) * HCIRI) = 1, H(@) = 0 and
OO = 2, Then L2(p) = C7, and Mey), = bel + Iyt
1 £1 Cclo] "Cts defined by £(,0) = x, then f has ow
nora-proserving extensions (Recall thar Liq) = LH) )
hots: fon axty.
QED.
IXERCISE U.- Let X be q normed Linear apace (WIS), and Let I* be ite
dual apace, that {a the set of all bounded linear funettonaty on T%
vith the nom Ut = aup frien! « tel N
Thus Hf + £120 a nm ake Ix1 <1 im (19), hence
Ifa} sonverges to £ in X%,
REMARK In che sane manner as (a), wé can prove that L¢x)1) —
Ite ap{leetitel ca |
sa maanea oped es
(®) Te is clear from the definition of IIfll, for £ € x4,
oe Wel = 1G) 16 Len,
‘aad In,Isbt
(evteontny, By ts tne
by the Hahn Banach theorem, if x X then there exists
JU >taL tence
, this Sestifies the notation HEI)
anf © such that [ZF +1, and £(2) + Ix, Thus
edb.
Now xB, is clearly s Linear asp, hence this ds an ison-
etry of X into xi
7
(c) 184s] 1s a bounded sequence, thon Lf(3,)I is bounded
in €, since ‘i
lea,)1