Professional Documents
Culture Documents
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
American Mathematical Society is collaborating with JSTOR to digitize, preserve and extend access to
Transactions of the American Mathematical Society.
http://www.jstor.org
89
b. Y=G?iEX,
c. TAh=g implies TAh(gi+x)=g(g?+x)=h(x), hGY*, gCG, g-CG?,
and xCX.
Proof. Suppose TA and G exist. Then Y=G?iEX by hypothesis. (G.
= {ylg(y)=0, yC Y, gCG}.) If we defineT1 by the relationTi(y) = Ti(gl+x)
=x, T1 is a projectionof Y onto X [16, p. 139].
cient condition (which we shall call Akilov's theorem) for a normed linear
space to have propertyP8. His proofof the conditionwas based on Kantoro-
vitch's extensiontheorem [13, p. 285]. The proofwe presenthere resembles
the proofof the Hahn-Banach theoremand does not make use of the Kan-
torovitchtheorem.
A space is of type B1+ [2, p. 643] if it is simultaneouslyof types K5 and B1
[11, pp. 210-212]. Using this definitionwe state and prove Akilov's theorem
[2, p. 646] in the followingform [5, p. 241].
3.3 THEOREM. If thenormedlinear space X is oftypeBi+,if theunitsphere
C of X has a least upper boundxo, and if IIxolI=s, thenX has propertyP..
Proof. Let Z be a normedlinear space containingX, let p be a sub-addi-
tive, positive-homogeneoustransformationon Z into X, and let f be a dis-
tributive transformationon X into X such that f(x) < p(x) for all x in X.
SubstitutingX forthe real numbersin a proofof Banach's [3, p. 28], we can
show there exists a maximal functionFo which is definedon all of Z and
which has the propertiesthat Fo(x) =f(x) forall x in X and Fo(z) ?p(z) forz
in Z.
Let p be the sub-additive positive-homogeneousfunctiondefinedby the
relation p(z) =xoIIzfl.If we take the identityfunctionI of X onto X as
1(x) ?p(x) forall x, and I can be extended to a distributivefunctionT such
that T(z) < P(z) = xoIIzlIand T(x) =x. T is a projection, T =s, of Z onto X.
Hence X has propertyP8 and the proofis complete.
Let I,n, be the space of ordered n-tuples of real numbers al
- {
(a,, a2, , an) }, 1 !n< oo, with the norm Ilall= (E f1 | ail P)l'P, 1 <p
< oo. If the usual partial ordering-that is, a'> a" if and only if a' >a" for
all i-is introduced,it can be easily verifiedthat Ip,n satisfiesthe conditions
of Akilov's theoremand has propertyP8, s =nlP.
In a later section we will considerother examples of spaces which satisfy
Akilov's theorem.However, forthe present,we shall content ourselves with
a briefexaminationof two spaces which do not have propertyP8 but which
satisfyall but one of the conditionsof the theorem.
Sobczyk [19, p. 945] proved there is no projectionof (m), the space of
bounded sequences, onto its subspace (co), the space of sequences convergent
to zero. Hence (co) does not have propertyP,. (co) is a space of type B1+but
its unit sphere has no upper bound in the space.
Fichtenholtzand Kantorovitch [7, p. 92] proved there is no projection
of M(O, 1), the space of all bounded functionson the unit interval,onto its
subspace C(O, 1), the space of continuous functionson the unit interval.
Hence C(O, 1) does not have propertyP8. C(O, 1) is a space of type B1, and
its unit sphere has a least upper bound of norm one. Hence C(O, 1) satisfies
all conditions of Akilov's theorem except the one requiring that all sets
bounded above have a least upper bound in the space.
Proof. Assume thereis a point q of normone such that the line segments
pq and -pq do not lie on the surface of C. Then (p+q)/2 and (-p+q)/2
are inside C and are not the zero element. Hence there is a point r1in C be-
tweenthepoints(p+q)/2 and (p+q)/IIp+qII and a pointr2in C betweenthe
points(-p+g)/2 and (-p+q)/II-p+qII. Let e,=1-IIrlIIand e2=1-11r211-
e1> 0 and e2> 0. Let e equal the smallerof el and e2and let k equal the larger
of IIp+qII and ||-p+qH|. Then ifbi=(e/k) (p+g) and b2=(e/k)(-p+q), ibillI
<e. Let Y be a linear (as yet unnormed) space which contains
<e and IIb2|I
X and which is of dimensionone greater than the dimensionof X, and let
yo be an elementof Y not in X. Let K, the unit sphere in Y, be the smallest
convex set containing C and the points + p ? yo, ? q ? (yo+ bi+ b2). Then
KnX=C.
Since X has propertyP1, there is a projection T, || T|| = 1, of Y onto X.
Let T(yo) =xo. T carries the line segmentfromp+yo to p-yo onto the line
segmentfromp+xo to p -xo. Since p is an extremepoint of C, eitherone end
of the line segment is outside of C or both ends equal p and xo= 0. For T
to be of norm one the latter condition must hold. Then T(q+yo+bl+b2)
=q+bl+b2. IIq+bl+b2II= 1+2e/kI >1. But jjq+yo+bi+b2jl =1; so, by
Lemma 4.1, X does not have propertyP1, and our assumption has led to a
contradiction.Hence the theoremis proved.
Proof. Let x1and X2be any two points of E and consider the convex hull
C' on exi+m(axl+yo) and e'x2+m'(ax2+yO) where e, e', m, and m' are
equal to ? 1 and a is positive. C'n(XXCC if w= (ex,+ax,+e'x2-ax2)/2 is in
C. If e=e', w-=((e+a)xl+(e-a)x2)/2 and a?< 1 implies I ? 1. If e = - e
w=(e+a)(xi-x2)/2 and thisis in C if |e+a| <2/1x1-x211. This is possible
fora>O only when IxI-x2l <2; a suitable choice is, in fact, 2/'| xl-X2| -1.
Then w lies in C if a=minimum (1, 2/d-1). Hence KQXX= C.
4.5 COROLLARY. If thenormedlinear space X has propertyP1, each extreme
pointof theunit sphereC is at a distance2 fromeveryotherextremepointof C.
Proof. Form K as described in Lemma 4.4 and let K be the unit sphere
of Y. Since X has propertyP1, thereis a projection T, jjT|| = 1, of Y onto X.
Hence exi+m(axi+yo) in K must project into ex1+m(axi+T(yo)) =ex1
+m(axl+u) in C. Then axi+-u is in (C-xi)fC\(C-+xi). If xi is an extreme
point, (C-xl)n(C+xi) =O, so ax1+u=O or u= -ax1. If both xi and X2 Xl
are extremepointsof C, u - ax1=-ax2, so a mtistbe zero. Then by Lemma
4.4 the set E consistingof x1 and x2 must be of diameter 2.
By a face of the unit sphere C of the normedlinearspace X we shall mean
a maximal convex subset of the surfaceof C.
4.6 LEMMA. If the normedlinear space X has propertyP1 and if E of
diameterless than or equal to one is a subsetof theunit sphereC, thenall faces
of C whichcontainpointsof E intersectin somepointof C.
Proof. Let Y be a linear (as yet unnormed) space containing X and of
dimension one greater than the dimension of X. Form K as described in
Lemma 4.4 and let K be the unit sphere of Y. There is a projection T,
|T| =1, of Y onto X since X has propertyP1. If ex,+rm(ax,+yo) is in K,
ex1+m(ax1+T(yo)) is in C. Then ax1+T(yo)=ax1+u is in (C-x1) n (C+xi).
Hence u is in (C-xi-ax1)Qn(C+x1-ax1). Since the diameter of the set E
which contains xi is less than or equal to one, a 1, u is in (C - 2xi) \C,,
and -u is in CQ (C+ 2x1). If IIxlI =fi(x1) = If,=I1 forfi in X*, then Ht
- {x fi(x) = 1 } separates C fromC + 2x1, so u lies in H1inC. Suppose x1 and
x2 are in E and x| xjjI I x21I=1; then there exist fi and f2 in X* such that
IIfilI=If2| =1 and f1(x1)=f2(x2)==1. Let Fi-={x xEC andfi(x) =1, i=1, 2J
-u is in CQn(C+2x2) as well as CC(C+2x,). Then F&r)F2D -u. Hence all
faces of C which contain points of E intersectin a common point.
The proofof our firstcorollarymakes use of Mazur's theorem [14, p. 731
that if K is a convex body in a normedlinear space X and if E is a flatsub-
set of X which contains no interiorpoints of K, then there exists a hyper-
plane H such that HIDE and K lies on one side of H.
4.7 COROLLARY. If X, Y, E, yo T, and u are as in the lemma, if x4i
= X211= 1, if xlEE, and if thelinesegment
fromxl to X2 is on thesurfaceofC,
thentheline segment
from-u to X2 is on thesurfaceof C.
the average value off over Si, i = 1, 2, 3, if s belongs to Si. It may be easily
verifiedthat T is a projection,IIT|| = 1, of L' onto L. Then ifL' has property
P1, L has propertyP1 by Theorem 3.2. But L does not have propertyPI by
Theorem 4.9. Hence L' does not have propertyP1. Lemma 2.3 then tells us
that (AL) does not have propertyPi.
5. A characterizationof certain normed linear spaces which have prop-
erty P1. The purpose of this section is to characterize those normed linear
spaces which have propertyP1 and which have extremepoints on theirunit
spheres.Since an equivalence preservesextremepoints,such a characteriza-
tion includesspaces with propertyP1, any one of whichhas an extremepoint
on its unit sphere; hence it includes finite-dimensionalspaces, conjugate
spaces, and reflexivespaces, which have propertyP1. The extremepoint is
used to introducea partial orderingin X. This partial orderingenables us
to show that a space with propertyP1 is equivalent to the space of all real-
valued continuous functionsover some compact Hausdorffspace H. From
this the characterizationfollowseasily.
Let C, the unit sphere of the normedlinear space X which has property
P1, have an extremepoint u. Let
Cu = { x I x = tx', x' E C + u, t real,t > 0}.
C. is closed [4, p. 849]. We definexl>x2, x1and x2in X, if and only ifX1-X2
is in C,. This definitiongives X a partial orderingsuch that the unit sphere
C of X consists of just those elementsx with -u_x_u [4, p. 850].
5.1 LEMMA.Let thenormedlinear space Y withunit sphereK containthe
normedlinear space X in whichthe unit sphere C has an extremepoint u. A
projectionT of Y ontoX maps K ontoC if and onlyif it maps Ku-{ yIy = ty',
y'EK+u, t_O} ontoCu.
Proof. Suppose T maps K onto C and let y=ty' be in K,, where t is a
real numbergreaterthan or equal to zero and y' is in K+u. Then y = t(y"+u)
wherey" is in K.
T(y) = T(t(y" + u)) = t(T(y") + T(u)) = t(x" + u)
wherex " ? C. Hence
T(y) = t(x" + u) = t(x') C C. where x' C + u.
Hence T maps K. onto C.
Suppose T does not map K onto C; that is, suppose thereis a y in K such
that T(y) =x is not in C. Then l|x|I> 1. We assume x is in C.-u, forother-
wise we are done. Then x+uFC.; that is, x> -u. -yFK since yEK and
hence T(-y)=-xEC since xEC. Suppose -x+u is in C.. Then u>x.
Hence -u<x?u and x is in C, a contradictionsince ||x|| > 1. Then -x+u is
not in C, as we assumed above. Thus u-y is in K. but T(u-y)= u-x is
not in Cu. Hence if T(K) $ C, T(K,) S C,; that is, T(Ku) = Cu implies T(K)
=C.
small 2u -XzWKEK'.
take X small. It remainsto be shown that forX sufficiently
In this connectionit sufficesto show that fork sufficiently
large,ku-zeKU',
forby settingX= 2/k> 0, we have 2u -XzCXK ' = K '. Take ki= max (lIx|I, 1).
Then kiu-xC4C.. Take k2so that k2u+aaCaaB, that is, so that k2u+aa=ab'
forsome b' in B. Take k3so that k3u-/3bE-f3A, that is, k3u-/3b= -3a' for
some a' in A. Then ku-z=(k1+k2+k3)u-x-a(-a-yo)-13(b+yo) =kiu
-x+a(yoj+-b') +3(-yo -a') EKu'. Hence zCKu and KudK'. But we have
previouslyshown that KuCK' so K =K'.
There is a projection T, ||T|| = 1, of Y onto X since X has propertyP1.
Then T(K) = by b Lemma 4.1 andT(K. ) == by Lemma 5.1. But we have
shown above that T(K ') = T(Ku) - Cu. Hence our assumptionis false and A
must have a least upper bound. This completes the proofof the lemma.
We now quote a theorem of Clarkson's [4, p. 8471 with the notation
modifiedslightlyso as to be consistentwith the notation of our previous
work.
Proof. Suppose C(H) has propertyP1. Then by Lemma 5.2 every set of
elements of C(H) which is bounded above has a least upper bound. Stone
has shown [20, p. 186] that if C(H) has this property,then everyopen set in
H has a closure which is open. Then by a theoremof Hewitt's [8, p. 326],
the closures of disjoint open sets in H are disjoint. From this we conclude
that H is extremallydisconnected.
C(fI) is a space of type B1 and its unit sphere has a least upper bound e
of normone wheree is the elementidenticallyequal to one on H. If H is ex-
tremallydisconnected,every set in C(H) that is bounded above has a least
upper bound [20, p. 186]. Hence C(H) is a space of typeB1+whose unitsphere
has a least upper bound e of normone. Then by Akilov's theorem,C(H) has
propertyP1.
6.4 THEOREM. The unitsphereS of C(QH),thespace ofall boundedcontinuous
real-valuedfunctionson an extremallydisconnectedHausdorffspace H, is the
closedconvexhull of its extremepoints.
Proof. Let K be the convex hull of the extremepoints of S; let F be any
elementof S; let e be a real numbergreaterthan zero; let N be the smallest
integergreater than 2/e; let the elementsof H be denoted by x's. Our first
step is to show that thereis an elementof K withine of F, that is, thereis an
elementg in K such that ||F-gj <?e. Let
El= I{xIF(x) > 1 -e.
E1 is open [15, p. 40]. Then E' =cEl, the closure of E1, is open and closed
since H is extremallydisconnected [8, p. 326]. Hence CE1', the complement
~~Kc~L
= ki (-CEI N
gi + E CE? + (1 - k) (CE + E CE)
N
= (1 - E)CE1 + ? CE'
j=2
. . . . . . . . . . . . . . . . . . . . . . . . . .
n N
gn= E(1- E)CE+ E CE1 for n<N,
i=1 i=n+l
* . . . . . . . . . . . . . . . . . . . . . . . .
N-I
gN = E (1 - tE)CE' - CEN.
i=1
If we now choose a sequence { e(n) } of E's, such that e(n) = 1/n, we can
finda sequence of functionsge(n) in K such that lim g
g) = F. Hence F
is in the closureof K; that is, S is the closed convex hull of its extremepoints.
6.5 COROLLARY. If the normedlinear space X has propertyP1 and if the
unit sphereC of X has an extremepoint,thenC is theclosedconvexhull of its
extremepoints.
Proof. X is equivalent to C(H), the space of all real-valued continuous
functionsover some compact Hausdorffspace H, by Lemma 5.4. Then C(H)
has propertyP1 and hence H is extremallydisconnected.Theorem 6.4 tells
us the unit sphere of C(H) is the closed convex hull of its extreme points.
Then C is the closed convex hull of its extremepoints since an equivalence
preservesextreme points, line segments,and distances.
6.6 COROLLARY. If the normedlinear space X has propertyP1, if X is
infinite-dimensional,
and if theunit sphereC of X has an extremepoint,then
C has an infinitenumberof extremepoints.
Proof. If C has a finitenumber of extreme points, it is not the closed
convex hull of its extremepoints. Hence it does not have propertyP1 by the
corollaryabove.
Our next theoremgives a necessaryand sufficient conditionfora normed
linear space X to be equivalent to a space of type Bj+ which has a unit ele-
ment.A space of type Bj+ whichhas a unit elementis in turnequivalent to an
abstract (M)-space C(H).
6.7 THEOREM. A normedlinear space X whichhas propertyP1 is equivalent
to a space Y oftypeBj+ whichhas a unit elementif and onlyif theunitsphereC
of X has an extremepoint.
Proof. Suppose the unit sphere C of X has an extreme point u. Let
x1>x2 mean x1-x2 CCu.Then X is equivalent to a space C(H) of all continu-
ous real-valuedfunctionson some compact Hausdorffspace H, by Lemma 5.4.
The element e which is identicallyequal to one on H is the unit element in
C(H). Since C(H) has propertyP1,everyset in C(H) which is bounded above
has a least upper bound, by a previouslemma. Hence C(H) is a space of type
Bj+ and has a unit element e.
Suppose X is equivalent to a space Y of type Bj+ which has a unit element
e. e is an extreme point. Let U be the equivalence between X and Y and
considerthe point u = U-1(e) in X. lull= 1 so u is in C. Suppose u = (x] +x2)/2
where x1 and x2 are in C.
U(Xl) + U(x2) yl + Y2
=
U(u) = e-
2
-=____ 2_
2
where yi and Y2 are in S the unit sphere of Y. Then Yi = Y2 = e since e is an
BIBLIOGRAPHY
1. G. P. Akilov, Necessaryconditionsfor theextensionoflinear operations,C. R. (Doklady)
Acad. Sci. URSS. N. S. vol. 59 (1948) pp. 417-418. (Russian)
2. , On theextensionof linear operations,C. R. (Doklady) Acad. Sci. URSS. N. S.
vol. 57 (1947) pp. 643-646. (Russian)
3. S. Banach, Theoriedes operationslineaires,Warsaw, 1932.
4. J. A. Clarkson,A characterizationof C-spaces,Ann. of Math. vol. 48 (1947) pp. 845-850.
5. M. M. Day, A reviewof A kilov's"On theextensionof linear operations,"Mathematical
Reviews vol. 9 (1948) p. 241.
6. , Lecture notesfor theoryof linear operations,University of Illinois, 1947, un-
published.
7. G. Fichtenholtz, and L. Kantorovitch, Sur les operationslineaires dans l'espace des
functionsbornees,Studia Mathematica vol. 5 (1934) pp. 69-98.
8. E. Hewitt, A problemofset-theoretictopology,Duke Math. J. vol. 10 (1943) pp. 309-333.
9. S. Kakutani, Concreterepresentation of abstract(L)-spaces and themean ergodictheorem,
Ann. of Math. (2) vol. 42 (1941) pp. 523-537.
10. ofabstract(M)-spaces, Ann. of Math. (2) vol. 42 (1941)
, Concreterepresentation
pp. 994-1024.
11. L. Kantorovitch,Linear operationsin semi-ordered spaces, I, Rec. Math. (Mat. Sbornik)
N. S. vol. 7 (1940) pp. 210-279.
12. Rdume, Rec. Math. (Mat. Sbornik) N. S. vol. 2 (1937)
, Lineare halbgeordnete
pp. 121-168.
13. ---, Sur la theoriegenerale des operations dans les espaces semi-ordonnes,C. R.
(Doklady) Acad. Sci. URSS. N. S. vol. 1 (1936) pp. 283-286.
14. S. Mazur, Ober konvexeMengen in linearennormierten Rdumen,Studia Mathematica
vol. 4 (1933) pp. 70-84.
15. E. J. McShane, Integration,PrincetonUniversityPress, 1944.
16. F. J. Murray, On complementary manifoldsand projectionsin spaces Lp and lp,Trans.
Amer. Math. Soc. vol. 41 (1937) pp. 138-152.
17. S. B. Myers, Spaces ofcontinuousfunctions,Bull. Amer. Math. Soc. vol. 55 (1949) pp.
402-407.
18. R. A. Phillips, On linear transformations,
Trans. Amer. Math. Soc. vol. 48 (1940) pp.
516-541.
19. A. Sobczyk, Projectionofthespace (m) on its subspace (co), Bull. Amer. Math. Soc. vol.
47 (1941) pp. 938-947.
20. M. H. Stone, Boundednesspropertiesin functionlattices,Canadian Journal of Mathe-
matics vol. 1 (1949) pp. 176-186.
UNIVERSITY OF ILLINOIS,
URBANA, ILL.
FLORIDA STATE UNIVERSITY,
TALLAHASSEE, FLA.