IC/92/275

INTERNAL REPORT
1. INTRODUCTION
(Limited Distribution)

international Atomic Energy Agency
and
Nations Educational Scientific and Cultural Organization
INTERNATIONAL CENTRE FOR THEORETICAL PHYSICS
Conditionally exponential convex functions have been introduced and studied in [2, 8].
Berg named it "negative-definite" [21. The set of these functions, denoted Eo(G), is a convex
cone. This arises the following questions;
How to construct a compact base for Eo (G) ?
What are the extreme parts of Eo (G) ?

In fact the aim of this paper is to answer the above questions. To do so, we first study
a weight, called the Levy weight, in its abstract form. Also, we clarify how to obtain a unitary
representation from elements of Eo (G) •
CONDITIONALLY EXPONENTIAL CONVEX FUNCTIONS
ON LOCALLY COMPACT GROUPS Now, let C'(G) be the enveloping C-algebra of Ll(G) equipped with # involution
denned by: f{%) = A ( i " ' ) / + ( x ) , where A is a modular function and / + ( i ) = f(x~l). The
dual Banach space of C*(G) isB(G) [4], The set of positive linear functional in S(f?) is P(G)
and it is identified with the set of exponentially convex functions on G; i.e., the set of functions
satisfying
A.S. Okb El-Bab * n
International Centre for Theoretical Physics, Trieste, Italy. ] P V»(jfi9;) Ci Cj > 0 ,
ij-l

where gi gn E G and c\,..., cn 6 R, The set of elements from P( G) with norm equals 1 is
ABSTRACT Pi(G); this is a convex set whose sel of extreme points is denoted extPi(G) |7). Also, we write
Eo (G) for the set of all conditionally exponential convex functions defined on G and vanishing at
The main results of the thesis are: the group identity; i.e., functions satisfying

1) The construction of a compact base for the convex cone of all conditionally exponential
convex functions. c, > 0

2) The determination of the extreme parts of this cone.

where p i , . . . , gn € G and c i , . . . , cn € R [2, 8J.
Some supplementary lemmas are proved for this purpose.
Now, if Q G C(G) and p £ P(G), define the translation of a by p to be the unique
element Tpa of C*(G) such that for all b e B(G),(6p,o) = (b,Tpa). The translation operator Tp
MR: 43 is a completely positive linear map on C'(G) and is norm decreasing if p £ Fi(G). lfp,g € P(G)
and \ > 0, then Tp, = Tp T, and r^+x, = Tp + \ T,, so that T(.) is a homomorphism from P(G)
MIRAMARE-TRIESTE into the set of completely positive linear maps on C*( G) [3].

September 1992 Given the notion of translation, we can define differential as a limit of difference quo-
tients. In this way, we obtain a tangent space at each point of P\ (G).

A semitangeni vector to Pj(G) at the identity is any continuous real valued function ^
Permanent address: Mathematics Department, Faculty of Science, Al-Azhar University, Nasr City, on G satisfying
Cairo, Egypt. ii = lim \k( 1 - pk) ,

T I IT
where {Ai}t>i is a diverging increasing sequence of nonnegative numbers and {pt}*>i is a un- Now let 2, be the central support of the weak closure of ker 1 i n i y ^ G ) andlet(ker 1)]
formly convergent sequency belong to P] (G). It can be seen that the collection of such semitangent be the unit ball of ker 1. The proof of the following theorem, which is similar to that of Proposition
vectors is identified with the set EQ(G). If both iji and -f are in Eo (G), then ij> is called a tangent 1.11 of [3] is omitted.
vector to Pi (G) at the identity [ 1 ].
Theorem 2.2 The necessary and sufficient condition for the function i/> g £ 0 (G) to be lower
semi-continuous is that there exist positive linear functionals {fk}k>\ on ker 1 such that ij>(a) =
T, ( / * , Q ) fora e (ker 1) + .
2. LEVY WEIGHTS

Now we reformulate this theorem in a more concrete form. Choose a £ (ker 1)+ to be
Let V) e Eo(G) and let 9 : C'(G) -> C*(G) be a linear functional such that 9(a*o) >
of the form / / • h* * h * li, where h e Cc( G), and let pk e P( G) be the extension of fk with the
(3a*) a + a* (9a) ifaa* and a*a belong to D(d), the domain of d. Define a linear functional on
same norm. Finally, put fi - 6, - Sv where 6, denotes the point mass at •. Then
C'(G), also denoted by V>, by

V>(a) = (l,9v,a), a6 h)

Clearly, V1 is densely defined and ^|(ker 1)+ > 0; i.e., V> is a weight for the elements of Eo(G)
and it is called a Levy weight. and we get that -f is the monotonic limit of the exponentially convex functions £ /J *p* */i + .
Let W'(G) be the double dual of C'(G). We may consider G and C(G) to be con-
tained in W(G); when il is necessary to emphasize that a measure /i, say, belongs to W'(G), we Asasubsetof Eo(G) we define the set of local Levy weights to be the set of iji £ EQ(G)
such that
will write w(/j), where w is the universal representation [6].
sup{/? a (o;ifr)|a>0}=0 for all a<
Lemma 2.1 Let Mc° (G) be the set of compactly supported Borel measures on G of total mass
where
zero. For/i £ M°{G) and a 6 C'(G) we have
pa(a;ij>) = inf{^(«) + ta\a < s + tz, ;t > 0, s 6
= ( - f ,o), where
Lemma 2.3
Proof First we suppose that a takes the form w(/), / belongs to the set CC(G) of compactly
FoTip,^>,il) - (o 6 B0<,G),i> is local if ^ is so.
supported continuous functions on G. Then
Proof Suppose thai a 6 (ker 1 ),*, then for
= (1, =- I
M+ = JVj JV^, and /v\, = {a ker 1 |V<(a* a) < oo}
f(x)dx.
we have

Now, we prove the lemma in the general case. Let o £ C"(G) be the strong limit of the Pa(o; V1) = inf {0(s) •
sequence an = w(/„),/„ £ CC(G). Then
> inf {<p( s) + taja < s + tz\; a £ M^, t > 0 }
> inf {if>(s) + ta\a < s + tz\; s e M^, t > 0} = pa(a; ip) , jj

Applying Cauchy-Sch wartz inequality to the second term of the right side and then taken the limit Now, to associate a representation of G with each ^ € £o(G), define a pre-inner product
we get on Mj (G) in the form
* = lim
n
= lim ( - // dip*
(the integral in r.h.s. is a negative quaniity [2, 8], and suppose its kernel to be K$- Let Definition 3.1 [3] If i>, <p e £<, (G), we say that V< dominates v> if V> - ¥> G £b (G). If V1 and
ip dominate each other, then they are said to be equivalent. They are weakly equivalent if one is
equivalent to a positive multiplier of the other.

Then Now we construct a base for Ea{G) which is compact in some suitable topology. This
and can be done by the selection of an element on each weak equivalence class in Eo (G). Let G
be compactly generated, AT be a compact symmetric neighbourhood of the identity in G which
In this case U defines a continuous unitary representation of G on the Hilbert space H+, denoted
generates the group and let K = {$ € Eo(G)\^Ni^)(x)dx = 1}. If ^ € £b(G) satisfies
also by U$. H^, can be obtained by completing M° (C?)/iC# w.r.t. (-,-)v>-
fNl \i>(x) dx = 0, then ^N* = 0. Since JV is a generating set, we have 4> = 0 everywhere. So,
Let Ay, be the canonical projection of M^(G) into H^ and let c^ : G —> fl^,c^,(g) = forV>6 EQ(G) we can divide it by / N , V>( x) dx to obtain an element of K, which implies that K
A,(,(fi e -6 ( ). Clearly. c^ satisfies the cocycle equation c^(pft) = c^,(j) + t7^(s)c^,(/i),c^ is called is a base for Eo(G).
a 1-cocycle for U$ |5].
Lemma 3.2 Suppose that X^. is the indicator function of the set An, where A\ = N,An -
Lemma 2.4 For ^i e M° (G) we have N" - N"~l fOT n > 2 andc = a'1 sup{l + A(j/)[y £ JV} where a is the left Haar measure of
JV. Then for V> € K we have 4>(x) < f(x) where / = c £ n2 XAK.
n-l

Proof Let y £ G and p, = Se + Sv. Then for each V> e Jf we have

- ^"(e) = + t(yx) + £ £b(G) ,

Proof To prove the first equality, we suppose that/i,^ € MC°(G). Then
and by integration we get
- / i * V*^ + (?) = - //nHzpy

=- ip(xy) d(6g *v)(x)

If y £ iV.then
Talcing v = p, the first equality follows.

For the second one we have dx + f 1>{x) A(j/) di + f dx+ dx

J

= " / / ' ^ - 6«) * * * (&< - 6 -) + ( dn(y) by symmetry of N and sup {iKy)|y K} <C.
Now, if aj e G, then x e An for fixed n and hence it belongs to JV". So, there exist
i, • • •,!/» € JV such that £ = yi y2 . . . yn. Then

= - ff^(xgy) dfi(x) dfi(y) = -0"(s

We must notice that we used the fact that fi(G) = 0. //

and we get

3. A COMPACT BASE FOR Eo (G)

In this section we construct a base for Eo{G). First we begin with the following defini- Let S be a separable compact convex set. A subset F of 5 is called a face if each line
tion. segment in 5 whose interior intersects F is contained in F. The complementary set F1 of F is the

r r
union of all faces of S disjoint from F. If F' is a face and F is a closed face, then F is called a Definition 4,1 We say that $ £ EQ(G) generates an extreme ray in £o(G) if each of its
closed split face. In the latter case, S is the direct convex sum of F and F'; which means that every dominated elements is either a tangent vector or weakly equivalent to $.
element x £ S can be written uniquely in the form Theorem 4.2 Suppose that 1/1 £ £o(G) has a lower semi-continuous Levy weight. The
-X)z, y£F, necessary and sufficient condition for V> to generate an extreme ray is that it takes the form

Now we notice that if is compact only if G is discrete, but we can compactify K by \(l-p)+h with X > 0 , p £ ext Pi(G) - {1} and /*£hom(G,ii)
adding a point at oc. 1*1$°° (x) = |W 3 |"'. It is clear that / „ , ^°°U) cte = 1, and ^ belongs
to E o (G). Let AT be the con vex hul 1 of K" and i/>°°. Proof First we suppose that \jj is weakly equivalent to 1 — p for p g Pi(G) — {1}. If
] ip £ Ea (G) is dominated by (1 - p)i then p is bounded. Hence, there exists pi £ {1}', X > 0
Theorem3.3 Suppose that L°°( G)\ is the unit ball of L°°(G) equipped with the o(L°°,L )
and h £ Horn (G,R) such that <p = X(l - pi) + k. So, 1 -pdominates X( 1 - pi) in EQ{G).
topology and suppose also that p : K -» L°°(G)i is given by p( V>) = V / / . wh ere / is defined as
No w supposing that \ < 1 and since (1 — p) — X (1 —pi) € Eo(G) thenwecanfinda constant
in Lemma 3.2 and /C is compact in the topology induced by p. Then the set {V>°°} is a closed split
K > 0 such that
face of K.
Proof It is clear that L°°( G)i is compact in the a( L™, L1) topology, and we only need to
prove that p(K) is closed. Suppose {V»n}™i € ff and such that p(^ B ) -> i/> £ L"(G)i and
suppose also that V> = / p - For g £ L1 (G) and h £ C* (G), where p = (/i# * / i ) / , we have This implies that K + p - \p\ £ Pi(G). Since p £ ext Pi (G), pi is a convex combination of
the orthogonal exponentially convex functions 1 ,p. But pi g {1}' then pi = p and we have 4> is
I ip(x)h* *h(x) dx = / y?(i) 9(1) dx = lim / p(pn) g(x) dx extreme.

= lim f pn(x)h* *h(x) dx < 0 .
Let V be any measurable subset of G with finite measure. Then
f ip(x)dx=\im I p{tpj(,x) dx = lim / <pn{x)/f(x) dx > 0
Jv » Jv " Jv
and hence >p(x) > 0 a.e. and the same is true for ij>. Hence V> >s a.e. equals to a unique element
of E(G) and it remains to prove only that for \j> £ Eo(G) such t h a t / ^ VC1) dx = l,i> £ K.
In fact, since VKe) < |W 3 |-' = V°° we put X = ^(e)/^°°. If A = 0 then 1^ £ Eo(G) and
hence V» € K C ^ - If X = 1, then fsi ^>(e) dx = 1, from which we get i/i(x) = V»(e) = ^°°
and V £ if C K. Finally, we suppose that 0 < X < 1 and we put ip(x) in the form ij>(x) =
X^°° + (1 - X ) <p(x), where^ x) = C^(i) - V(0))/(l - X) £ £b(G). Then
Conversely, let 1/1 be lower semi-continuous. Using Theorem 2.2 in its reformulated
form, we can see that $ is bounded if it generates an extreme ray in Eo(G). Hence, V1 is weakly
equivalent to an element of Et> (G) in the form 1 -p:p6 P\ (G). Noting that {1} is a closed split
face in Pi(G), we have to prove that p 6 {1}'. In fact, if p = X + (1 - X)pt is the representation
of pin {1} and {1}', then 1 - p = (1 - X)( 1 - pi) and we get 1 - pi is weakly equivalent to
1-p.
Now suppose that p £ ext Pi (G) — {1}. Then p can be written in the form p = (pi +
P2)/2 withp] j- pj from P\(G). Hence 1 - p = (1 - p i ) / 2 + (1 - pi)/2 contradicting the
extremity of ^ and we have p g ext Pi (G) - {1}. //

-X)

so that \li £ /f C K. The decomposition of 0 into a convex sum of $°° and an element of K is
easily seen to be unique. //
Acknowledgments

4. EXTREME RAYS OF Eo (G) The author would like to thank Professor Abdus Salam, the International Atomic Energy
Agency and UNESCO for hospitality at the International Centre for Theoitical Physics, Trieste.
Since Eo(G) is a convex cone, one way to understand its structure is to characterize its
extreme rays.
 REFERENCES

1. Akemann, C.A., Sequential convergence in the dual of a W*-algebra, Common, Math.

Phys. 7 (1968) 222-224.

2. Berg, C, J.P.R. Christensen and P. Ressel, Harmonic Analysis on semigroups, Springer-

Verlag,(1984).
3. Combes, F, Poids sur une C'-algebra, J. Math. Pures Appl. 47 (1968) 57-100.

4. Dixmier, J.,C*-algebra, Springer-Veriag, 1977.

5. Falkowski, B.J., First order cocycles for SL( 2, c), i. Indian Math. Soc. 41 (1977)245-
254.
6. Hille, E. and R. Phillips, Functional analysis and semigroups, A.M.S. Colleg. Publ. 31,
Amer. Math. Soc. (1957).

7. Johansen, S., An application of extreme point methods to therepresentationof infinitely

divisible distributions, Z. Wahrscheinlichkeitstheorie Verw. Beb. 5 (1966) 304-316.
8. Okb El-Bab, A.S. and M.S. El-Shazly, Characterization of convolution semigroups,
Proc. Pakistan Acad. Sci. 24, No.3 (1987) 249-259.

T T