for thesis submitted
TRANSCRIPT
TYPE I MULTIPLIER REPRESENTATIONS
OF LOCALLY COMPACT GROUPS
A.K. H0LZHERR (B.Sc.)
A Thesis submitted for the Degree of
Doctor of PhilosophY
in the University of Adelaide
Department of Pure Mathematici
FebruarY, l99z
by
CONTENTS
SUMMARY
SIGNED STATEMENT
ACKNOt,\lLEDGEMENTS
CHAPTIR I - PRELIMINARIES
Locally compact abel'ian grouPs
Concrete C*-algebras and von Neumann algebras
Classification of von Neumann algebras
Type I, von Neumann algebras and polynomial identities
5. Representations
6. Multip'liers and multipf ier representations
7. Moore groups and the regular representation
B. Induced representations, Mackey's construction
CIIAPTER II - ON THE STRUCTURE OF I¡-TYPE I LOCALLY COMPAüT
ABTLIAN GROUPS
1. Notation and e'lementarY facts
2. Some groups don't admit multipìiers
3. Sonle useful resul ts
4. The connected component of G as a direct sumrnand
5. Local direct products and div'isjble groups
CHAPTER III - MULTIPLIER REPRESENIATIONS OF DISCRETE GROUPS
1. A representations of elements in V(G,t¡)
2. Discrete finite class groups
3. The o-finite class group and the o-centre of G
4. The type I part of V(G,ur)
1.
2.
.)
4.
(i)(iv)
(v)
I
1
4
6
10
t2
t7
23
2B
JJ
33
40
44
55
58
6B
69
80
82
89
CHAPTER IV . GROUPS WITH FINITI DIMENSIONAL IRREDUCIBLE
MULTI PLIER REPRESENTATIONS
1. Preliminaries
2. The main theorems
4, ExamPles
APPENDI X
1. aO and nO
2. ^d^
and CIp^
3. The dual of Q
REFERENCES ,
103
104
110
L20
4.1
4.4
4.5
(i)
SUMMARY
Let G be a locall,v compact abelian group and o a
normallzed Boiel multiplier on G. llle are concerned primarì'ly
with the von Neumann algebra V(G,o) generated by the regular
rrr-Fêprêsêntation p of G defined by
p (g)f (x) = f (g-r x)o(g-1 ,x) '
g e G, almost aì'l X e G, and f e L2(G).
The pair (G,o) is called type I if every
ur-Fêprêsêntation of G is tYPe I.
In Chapter II, we investìgate the structure of (G,r,r) ln
the case where G is abel'ian and (G,r) is type I. In
particular, we can reduce t,he study of such pairs to the case
where G is residually finite. Furthermore, if G is separable
and divisible, then (G'r)'is type I if and only if there
exists a bicont'inuous isomorphism from G to a group of the
for'm H x H^ where H is a closed subgroup of G, which carries
t¡ to a multiplier on H x H^ that is similar to the multip'lier
or' gi ven by
,'((x,i,)(y,x)) = r(y) ,
(x,r)(y,x) e H x H^.
(ii)
Chapter III provides information about the maximal type I
central projection e in V(G,o) in the case where G is a
discrete group. Indeed we have e + 0 if and.only if there
exists a subgroup H of G such that the index [G:H] is finite,
the commutator H'has finite order and o restricted to H is
the trivial multiplier. l^le also show hovl e can be realized as
a convolution operator on t2(e). As a consequence of thjs
result we can prove that for G discrete, (G,r) is type I ifand only if V(G,o) is a type I von Neumann aìgebra and that
this occurs if and only if G has an abelian subgroup A of
finite'index in G such that o restricted to A is a trivial
multjplier. fnis result generaìizes easi'ly to assert that a
locally compact group G with normalized Borel multiplier ur
satisfies V(G,to) Ís type I.O for some natural number k if and
on'ly if G has an open abelian subgroup A of finite index in G
such that the restriction of o to A 'is trivial.
Finally, in Chapter IV these results are generalized to
an arbitrary iocalIy compact group G with Borel multiplier o.
Let e be the maximal type I finite central proiection in
V(G,r¡), then e * 0 if and only if
(i) [G:¡] ( æ
(Íi) a'has compact closure, and
(iii) there exists a finite dimens'ional o-representation
of Â, where a denotes the closed normal subgroup of G
consisting of all those elements whose coniugacy class has
compact closure. Again we can construct e as a convolution
operator on L2(G) and use this to prove that the folìowing
Ir]1j
are equivalent.
(i )
(ii)(iii)
All irreducible ur-representations of G are finite
dimensional.
V(G,o) is type I finite.
The fol l owing propert'ies hol d .
(a) [G:a] < @
(b) ^'
has compact closure.
(c) There exists a finite dimensional
o-representat'ion of ¡.
(d) n{ker n : n is a finite dimens'ional
' (ord i nary) representat'ion of ¡] = { I } .
All the irreduc'ible (ordinary)representations of the central
extension Go are finite d'imensional .
(iv)
Mackey's normal subgroup analysis is used in
conjunction with the above theorem to construct a group G and
multjp'liers or, t e [0,1] such that V(G,cua) 'is type I finite
and has a non-zero type in Part for arb'itrarily large natural
numbens n if t is rational, and is type II, if t is
irrationa'l .
(iv)
This thesis contains no material which has been accepted
for the award of any other degree or diploma in any University.
To the best of my knowledge and belief, the thesis contains no
material previously publ ished or written by any other person'
except where due referenie is made in the text of the thesis.
A.K. Hol zherr
(v)
ACKNOl,.JLEDGEMENTS
I wish to express my appreciation to W. Moran, my
supervisor, for hÍs encouragement and advice during the
time spent on doing research for this thesis, and also
for his support as a friend.
My appreciation 'is extended also to Helen Koennecke and
Chery'le Marshall who did the typing for this thesis.
1
CHAPTER I
PRELIMINARITS
This chapter is introductory in nature. It serves to
define notat'ion, to provide the background, and to introduce
some of the resul ts needed for I ater chapters. For more
detai'led properties of locally compact groups, C*-aìgebras
and von Neumann aìgebras, the reader is referred to [7' 18'
34, 40, 471.
1 Locally compact abelian qroups
Let G be a locally compact abelian group. By this v¡e
vriil always mean a topolog'ical group which is local'ly compact
and To (and hence normal ). We adopt the common usa9e
throushout of ustnn .n:,:oj -fX'åiL. to describe a locallv
compact group which is metr^izaålæ. Group homomorphisms
(isomorphisms etc.) are such strictìy in the algebraic sense
unless it is specifically stated otherwise, A bicontinuous
isomorphism 'is aìso referred to as a topoiogical
isonrorphism. Given two locally compact groups H and K, then
t{ x K denotes theìr direct product (with Cartesian product
topoìogy) and l-l is identifjed (up to topo'logical isoinorph'isrn)
with the subgroup H x ieÌ. Simjìar1y for K" A subgroup ll of
G is called a topologicai djrect summand of G if it is closed
and there ex'ists a closed subgroup K of G such thai G and
H x K are topo'logìca'l1y isornorphìc, and is called a direct
2
sunrnand if the same property hoìds for the discrete topoìogy
on G.
Let G be a locally compact abelian group. The character
group (or duaì) of G, denoted by G^, is a group whose
elements are contjnuous homomorphisms, G +ill'(where Il is the
group of comp'lex numbers of modulus 1 w'ith the induced
topology) and with multiplication defined pointwise. The
sets {x e G^ : x(0) g V}, where C is a conrpact subset of G
and V is a neighbourhood of the identity in'll', is a basis of
neighbourhoods for a locally compact group topology on G^.
If S is a subdet of G, then A[G^,S] = {x e G^ : x(s) = 1 all
s e S] denotes the annihilator of S in G^. The followìng
duatity theorems and structure theorem are well known.
THEOREM 1.1. Let G be a LocaLLy eornpact abeLian gzoup,
then
ø)
øi)
øi¿)
(G^)^ is topologicaLLy isomorphíc aith G.
If S is a subset of G mtd K is the stnalLesi;
cLosed subgroup of G eontaining S, then
AlG, A[G^,S]l = K.
If H and K ave cLosed subgroups of G such that
H s K, then A[K^,H] is topoLogicalLy isomorphíc
uví,th AIG^,Hl/AtG^ ,Kl cnd (K/H)^.
If H ís a eLosed subgroup of G, then 11 is open
(nespectiueLy eompaeÐ íf A[G^'l]l is compact
(respectiueLy open. )
Gtt)
3
PROOF. See [18, 24.8, 24.10 and 23.24).
THE0REM 1.2. A LocaL Ly cornpaet group G is topoLogícaLLy
isomorphie to lRfr x K, uhere K is a LocaLLy compact abeLian
group containíng a eompq,ct open subgroup. L'Lte ínteger n ís an
inuaz,i-ant of G,
For a proof of this result see [18, 24,30].
The following result relates the structure of G to that
of G^.
THEOREM 1.3.
(i)
øi)
ft,ii)
The duaL of a diuisibLe LocaLLy compact abeLían
group is toz,síon free.
The duaL of a conrpact totaLLy dísconnected gt'oz'Lp
is tov,sion.
lf G ís a eompaet group, then the foLLouing ate
equiuaLønt.
k) G is eonrpeted.
ft) G^ ¿s tonsion free.
(e) G is dLuisíbLe,
PROOF. See [18 , 24 .23, 24 .26 , 3.5 and 24.25] .
Finatly, we mentjon some a'lgebra'ic propert'ies of discrete
abel ian groups.
4
THEOREM 1.4.
(i)
øí)
(i¿i)
A diuisibLe subgtoup D of an abeLimt groLtp G ís a
direct sunrnand, Indeed the conrpLement may be
chosen to eontain any subset of G uhích interseets
t:æíoiaLLy uíth D.
If H is a subgroup of G such that the order of
each eLqnent ín H is Less than some fined integet',
anå GIH is torsion free, then H is a direct
sutnnanÅ. of G.
A fínite abeLiqn groLtp is isomorphic uíth a
finite product n Hp, uhere each H, is a cgeLie
groLe of prime poüer order,.
PROOF. See [18, 4.8 and 4.25] and lI2, 27 .51.
2. Concrete C*-alqebras and
von Neumann alqebras
Let H be a Hilbert space with inner product ( ','), and
tet B(H) Ue the algebra of all bounded linear operators on H
and B(H)* the Banach space dual of ts(H). The uniform
topo'logy on B(H) is given by the norm liall (a e B(H)), where
llall - trpiltil<1,8e, llaEll
B(H) is a Banach algebra under this nr¡rm and, with the
adjoint operation a + a* (defined by the rejation
(ô6,n) = (8,ê*n) for all Ë,n e H) as the invojution, B(H) is a
5
C*-algebra. Indeed any norm closed *-subalgebra of B(H) (tf¡at
is, a subalgebra invariant under the action of the involutiotl
*) is a C*-algebra. For a converse we have the following.
üEqßEI_¿¿, ( [40, 1" 16.6] ) . Let; A be a cx-aLsebra,
then thev'e eæists a x-ìsomorphísm from F\ onto a tmífot'mLy
cLosed x-subaLgebra of B(H) for some HiLbert space H.
Given a subset A of B(H) we denote by A' the commutant of
A viz. A' = {a e B(H) : ab = ba for all b e A}, by A" the set
(A')' and by CA the centre A n A' = {a e A : ab = ba all
b e Aì of A.
A *-subalgebra A of B(H) for which A" = A is called a
von Neumann al gebra (on l^l*-al gebra ) acti ng on ll.
The weak operator topoìogy on B(H) is the smallest
topology on B(H) such that for all 6,n e H, the function
B(H) +C : a +(âã,r) is continuous. Each von Neumann algebra
acting on H is automat'icalìy closed with respect to this
topology and conversely, a weakly closed *-subaìgebra of B(H)
containing the identity operator on H is a von l{eumann
al gebra (Sakai [40, 1 .20.3] ) .
Let A be a von Neumann algebra acting on H and let V be
the linear space of all continuous linear functionals on A
with respect to the weak operator topology and denote by A*
the norm closure of V in B(H)*. The o(A,A*) topology on A,
that is the smaliest topology on A such that each element in
6
A*'is continuousr'is called the o-weak (or ultra-weak topoìogy).
A* cons'ists of al I o-!{eak conti nuous I i near ftrnctional s (al so
called normal functionals) and is called the predual of A; its
Banach space duaì can be identifìed (up to isonletric
isomorphism) with A by means of the map A * (Aa)* ' ¿ + Tu,
where fu('t,) = q/(a), â u A and ú e A* ([:2, Theorem 2.4]),
Furthermore, A* is unìque in the sense that no other norm
closed subspace of B(H)* has th'is property ([47, III.3.9] ).
THEOREM 2 2 ¡47, III.3.10l ) . Euery *-isomorphíun from
q.1)on Newnann aLgebra to another is o-ueakLg continuous.
Suppose we have von Neumann alcebras Ar, (i = 1, 2)
acting on Hilbert spaces H' then a *-isomorphism o: A, +A,
is said to be spatiaì if there exists an isometry U of H.,
onto H, such that
(a)=UaU*, acA.
l^le point out at once that *-isomorphisms are not necessarìly
spatial ([40, page 119]).
3. Classi ficati on of von Neumann alqebras
Let A be a von Neumann a'lgebra act'ing on a Hilbert space
H. A self adioint element e of A is called a projection if
e2 = e; a project'ion e is called central 'if ae = ea for all
a e A (tfrat is e e CA = A n A'), and abelian if eAe is an
abelian algebra. Two projections pr Q âFO sajd to be
7
orthogonal if pq = 0.
Let p and q be two project'ions in A. If there exists an
element u of A such that u*u = p and uu* = q, then p'is said
to be equìvalent to q and we denote thìs by p - q. If there
exists a projection q, (< q) equivalent to Pr v,rê write this by
p q or q p. The relation "-" satisfies the conditions of
equ'ivaì ence and the relation rr rr i s refl exive and transitive.
Let p be a proiection in a von Neumann algebra A. p is
said to be finite, ìf for a projection p, in A, P, < P and
Pr - P imply Þ, = P; p'is said to be purely infinite if itdoes not contain any non-zero finite proiection; p is said to
be infinite if it is not finite.
Abel ian projections are finite ([40, 2.2.8)), A]so one
can easi'ly see that along with fìnite projection, all
smaller ones are also finjte. A von Neumann a'lgebra is said
to be finjte (respectively purely infinite etc.) if its
identity is finite (respectively purely infinite etc.).
Let tj, i e J be a fami'ly of mutual'ly orthogonal finjte
central projectiorrs of a von Neumann algebra A and let
z= L zj. Suppose z- p < z. Then pzj a 2.. Hence y=j = =j
(i . J), so that p = z. Hence z is also finite. Hence there
exists a unique maxìma'l finjte central projection zr in A.
Similarly, we can see that there exists a unique maximal
purely infinite central projection z, in A. Set
zz=I-r.,,-23.
I
A is saìd to be semifinite if zs = 0; properly infinite if
zt = 0; properly infinite and sem'ifinite if zL = ,3 = 0,
THEOREM 3.1 ([+0, 2.2.3]), A uon Neumarm aLgebna. A ís
wtíqueLg deeonrposea trtto a div'ect sun of thv'ee aLgebras uhich
æe finil;e, properLa infinite and semifiníte, and pweLy
infínít.e, respeetiueLy .
Finiteness, semi-finiteness can be characterized using
traces. Let A be a von Neumann algebra acting on a Hilbert
space H. A trace on A+ = {a e A : (a6,8) > 0 all E e H} is a
function p deíined on A+, with non-negative finite or infinite
values, possessing the following properties.
(i )
(i i ¡
.p(a + b) = ç(a) + ç(b), a,b . A+,
e(ra) = ¡.ç(a), À e c, a. A+ (with the convention
that o . * æ = 0),
If u is a unitary operator of A, we have
ç(uau-l) = ç(o), a . A*.
(iii)
I is said to be finite if .p(a) . +- for all a e A't; , is
said to be sentifinite if, for every a e At, e(a) is the least
upper bound of the nunlbers,p(b) for the b. A+ such that b < a
and ç(b) f **; ç is said to be faithful if the conditions
a e A* and r(a) = 0, imply that a = 0; and ç is said to t"¡e
normal if it is o-weakly continuous.
THEOREM 3.2 ([40, 2.5.4 and 2.5.7] ) . Let A be a
uon Neumann aLgebra. Ihen à is f'Lnite (r'espectiueLy semi'
9
finite) 1',f ffd. onLy íf fo7 any non-zevo a in Ar, theye eæísts
a norrnaL fínite (r'espectiueLy semifinite) traee ç such that
p(a) * 0- A is properLy infinite k'espectiuely puyeLy
infínite) í,f and only if thev'e is no nor,naL finite(r'espectiueLy semi-finite) trace on Ai ercept for the
ídentícaL zero tt'a.ee,
On a semí-finite uon Neumarm aLgebra, there eæists a
semífinite faithfuL trace.
A von Neumann algebra is said to be type I if every non-
zero central çiroject'ion contains an abel ian proiect'ion; type
II if it is semi-finjte and does not contain any abelian
projection; type III if it is purely infinite. A finite type
I (respectively type II) von Neunlann a'lgebra is said to be
tYpe I, (respectively type IIr), and a properly infinìte
type I (respectively type II) von Neumann aìgebra is said to
be type I- (respectively II*). A central project'ion z in a
von Neumann algebra A is said to be type X (X = I, II etc.)
if zA 'is type X.
THEOREM 3.3. ([+0, 2.2.10]). A uon Neumann aLgebra is
wtiqueLy deeorn¡tosed into a type I, tApe 17 a'nd type I I"I
díz.eet swnnand.
The fol I owi ng d iagram wi I I hel p 'i nterpret these
definitions and facts.
ÏII
æfI I
IIt II@
10.
fi ni te proper'ly i nf i ni te
type I
type II
type III
semi -fi ni te pure'ly i nf i ni te
tVpe I, type I I, tvoe I otvoe IIó type III
4. TypeJ von Neuman al ebras and
pol vnomi al identi ti es
Let A be a von Neumann a'lgebra with centre
C = {a e A : ab = ba all b e A}. A is said to be tYpe In
(n = 1,2, ...) if jt js *-isomorphic to C ø B(H)' where H is
a Hilbert space of dimension n. Note that C ø B(H) is nothing
other than the n x n matrix algebra over C.
THEOREM 4.1, ( [40, 2.3.2]). A type I, uon Neumarvt
algebra ean be decontposed as a direet swn of type In
Ðon Neumarm aLgebras (n e Z).
Using this result, we define a vorì Neumann algebra A to
be type I<k if the type In Part in A is zero whenever n > k.
semi -fi ni tepurel y
i nfi ni te
fi ni te properly infinite
11.
For any natural number k, let SO denote the standard
polynomìal in k non-commuting variab'les
Sk(u,, ..., ao) = f (-l )Y av(r ) uy(r) ... uv(t )
where the sum is taken over all permutations y of {1, ,..' k}
and (-l)Y denotes the signature of the permutation. Let A be
an algebra. l,le say that the ident'itV SO = 0 is satisfied
identically in A (or more brief'ly A satisfies SO) if
sk(A) = {o}
where Sk(A) denotes the set {SO(ar, ...' uk) : all ui e A,
i =tl, ..., k]. Po]ynomial identities are relevant in this
context because of the Amitsur and Levitski Theorem:
THEOREM 4.2 ([39, i.4.1]). For euery eonnwtatíue ring
R, the aLgebna of n x n ma:brì-ces Rn over R sal;ísfies Srn'
we have a partial converse of this theorem which is due
to Kaplanski,
THEOREM 4 3. An n x n matz'ín aLgebra oüer a, fíeLd does
not satísfa SZU fon k < n.
For the rest of th'is Section, we fo'llow Taylor t48l .
THEOREM 4.4. If the uon Neumann aLgebra A is nol; of
type I.n, then th.ev:e eæists a. copA ,f Mn+l fthe
T2,
naturaL nwnbet'.
(n+l)x(n+1) -compLeæ matyiees ) in A.
To see this, note that if A is not of type I.n' then
there exists a set of n + 1 nlutua'l1y orthogonal equivalent
projections in A. As in Smith 144, Lemma 9.31, a copy of
Mn..,.1 can be constructed in A.
THEOREM 4.5. Let A be a uon Neznnartn aLgebra and. n a
Then A satisfies trn Uf anð. onLy if A is of
tApe I=n.
PR00F. ft A satisfies Srn, then by 4.3 and 4.4, A is of
tYpe I=n. Conversely, if A is type I.n, then A is a direct
sum of the algebras A.O, (1 < k < n), where each AO is type
Ik. By 4.2, each AO satisfies Srn. Therefore A sa.tisfies
S2n:
PROP()SiTION 4.6. Suppose B ís a ueakLy dønse subalgebna
of the üon Neumann aLç¡ebra A, then A satisfíes trn Uf B does"
PR00F_. Let a
the I i neari ty of S
1 be a net in A converging weakly to a. By
and because multiplication in A is2n
lveakly continuous, SZn(a1, ar, ..., dZ¡) converges to
SrO(a , dZ, .. ", uzk). Similarly for the other variables.
This proves the result.
5. Representati ons
l.le give a very brief outline of definitjons and required
13.
results. For more details see 17,15, 2,301.
Let A be a Banach *-algebra. A representatjon of A in a
Hilbert space H is a homcmorphism
n:A*B(H)
which is non-degenerate in the sense that n(a)E = 0 all ô e A
impìjes g = 0. Since a *-isomorphism from a Banach *-algebra
to a C*-algebra is norm reduciñg, â representatjon of A is
necessari ly conti nuous .
The (tlil bert) dimension of H is cal I ed the d'imension of
n ônd is denoted by dim n. The kernel ker n of n is the set
{a e A: n(a) = 0}; and n is said to be faithful ifker n = {0}. Two representations n and nrare saici to l¡e
equivalent if there exists a unitary operator U : Hn, + H,,,
such that U-tn(a)U = rr'(a) for all â e A. lnle will not
usually distinguish between a representatio¡t and its
equiva'lence class"
Given a set of representations r , j . J, of A, de'f,inejtheir direct sutìt,o¡eJ nj by letting A act on the Hilbert
space tj.J
and each Hj
H, coordinate-wise. Conversely, if H = *j.¡ flj
invariant under the action of A (via n), then
n = tj.J nÏ where n, (i . J) is the restriction of n to llr.
Suppose we have two representations Tr and n' of the
Banach *-algebra A. Their tensor product is the
14.
representation n @ n' whose associated Hilbert space is
H o H , and js defined byTt 1Í'
(n e n')(a)(g * E') = tt(a)E t tt'(a)6' e
E . Hn, E e Hn,, a e A.
Denote by V(n) = {n(a) : a € fi1" fhe von Neumann algebra
generated by the operators n(a), â e A. V(") is iust the
weak closure in B(Hrr) of the complex linear span of the
operatorstr(a),ôeA.
n is callecl a factor (or primary) repnesentation if the
centre of V(n) consists of scalar multipìes of the identity
in V(n)(that is CV(") = 0.I).
PROPOSITION 5" ]. ( [40, 1.21.9] ) . Let n be a repnesenta-
tion of the Banach *-aLgebra A. The foLLouing are
equíuaLent.
(i)(í.i)
V(n) = B(Hr).
If H is a eLosed subspoce of Hn uhích is
inuavíant, that is n(a)6 e H aLL Ë e H, ô e A,
thenH = {0} oz'H=Hn.
If either of these equivalent conditions are satisfied,
then n is called ireducible.
Suppose we have an irreducible representation n and an
15
operator T such that n(a)l = Tn(a) all a e A, then by 5.1' T
commutes with all the proiections in B(Hn) and thus T 's
[15, IV.3.9].+-ísæ+i+ens+en++. This js called Schur's
Lemma.
co*L..-t
We will need a v,/ay of passing from Banach *-algebras to
C*-algebras. Suppose n is a representation of the Banach
*-aìgebra A and let a e A. Since n is norm reducing,
iln(a)il < llall , thus the supremum of iln(a)il as n runs through
all the irreduc'ible representations of A is a well defined
number which we denote by llall '. Let I be the set of â e A
such that llxll ' = 0, which is a closed self-adjoint two-sideci
ideal of A. The map a + llall ' defines a norm on the quot'ient
AlI. Endowed wjth this norm, A/I sat'isfies all the c*-algebra
axioms except that A/I is not complete in general. The
completion of AlI is a C*-algebra called the enveloping
C*-algebra of A and is denoted by C*(A). The canonical map of
A into B is a norm-reducing *-homomorphism whose image is
dense in B. when A is a c*-a'lgebra, we can identify A with
c* (A) .
THEOREM 5.2, (17 , 2.7.41). Let A be a Banach *-aLgebra
uith m approxùnate ídentLta artd r the carPnícaL map of A
into C*(A).
rf n is q iaepresentation of A' thev'e ís eæactLy
one repv'esentation p of C* (A) such that
(ì.)
IT = P o T.
ftÐ The møp t + 0 is a biiectíon of the set of
16.
('1 1,1, )
Gu)
representations of A onto the set of
repz,es entations of C* (A ) .
r is ívu'e&,¿cibLe ¿f qn¿ onLy if p is irueducibLe.
V(n) = V(p).
This result shows that the maiority of questions
concerning representations of Banach *'-algebras with an
approx'imate identity, it is enough to dea'l only with the
C*-algebra case.
The following result ensures an adequate supply of
representati ons .
THEOREM 5.3 $+l , I.9.231). Lel; A be a cx-aLgebz'a arú.
Let a ín à be non-zeyo. Ih.en they,e erí,sts an í,zryeducibLe
z,epnesentation r of A such that n(a) * 0.
A representation n of a C*-algebra A is said to be type
I (respectively finite etc.) if the von Neumann algebra V(n)
is type I (respective'ly finite etc.). A is called type I if
V(n) is type I for all representations n of A.
Let A be a c*-algebra. The set of equiva'lence classes of
irreducib'le represenbations of A denoted by A^ is called the
dual of A. The dual A^ is given the hull-kernel topology
which is defined as follows. A subset F of A^ has closure
¡- = {n e A^ : ker T f nr'.¡ ker n'}
17.
'Type l-ness' of a separable C*-algebra (tfrat is one that
contains a countable dense subset) can be characterized using
this topology. This is a deep theorem due to Glimm.
THEOREM 5.4 [7 , 9. ]. and 9.5.21 ) . Let A be a C*-aLgebra'
then A is type I íf anÅ onLy íf euera factor representation is
type I. Îf in oÅ&Ltion A is sepæabLe, then A is type I if
and onLy if A^ ís a I o topoLogieaL spaee,
6. Mul ti pl i ers and mul ti I ier representat'ions
Let G be'a locally compact group with identity ejement e"
A multiptier (or cocycle or factor set) on G is a Borel
measurable funct'ion o: G x G +'T'(the group of complex
numbers of modultls 1, with the induced topology) which
sati sfi es
r(g,h)r(gh,k) = o(g,hk)o(h,k)
ur(e,g) = ó(g,e) = 1
g,h,k e G,
g e G.
Two multipliers o, and t,l2 are similôr rrr, * az, if there
exists a Borel measurable function Y : G +'lf such that
*r(s,h) = Y(g)v(h)r(3h)-t 'r(s,h)g,h e G.
A muttipt'ier which is similar to f (the constant function on
G x G) is said to be trivial. Every multiplier is similar to
a normalized multiplier, that is one vrhich satisfies the
addi tional property
18.
,(g,g-I) = 1, g € G.
Indeed, if o is an arbitrary multiplier, then the multipljer
9,h + *(gh, (gh)-1) ,2,(g,g-I )
-r2 ,(h,h-1 )-4 ,(g, h)
is normalized. The square root here is taken in a Borel
measurabl e fashion.
If t¡ is normalized, tlten using the cocycle identity' we
have
r(g-1,h-1 ) =,¡(g-r,h-1) u,(g-1h-I,hg)
= ,(g-t,g) r(h-t,hg)
= o(h-t,hg),
and
o(h,g) r(g-1,h-1) = r(h,g),,,(h-I'hg)
= t¡(hh-1,9)r(h-1,h)
= 1,
that is t¡(h,9)-i = r(g-trh-1), al'l h,g e G.
Each normalized multiplier on G defines an extension G
of 'lt by G. It is the set "11'x G provided with the
mul ti pl ication
UJ
(s,g)(t,fr¡ = (stro(g,h),gh)
19.
(and then because æ is normaljzed, (s,g)-r = (s-l,g-1)).and a
topoìogy in which a basis of neighbourhoods of the identìty
'is composed of the sets AA-r, where A is a set of finite
positive measure for the product of ríght Haar measures on'll
and G. This is the topo'logy defined by Weil cn groups w'ith
an invariant measure (Weil [51, Appendix 1])" It is easy to
check that this topoìogy jnduces on 'F (indentified with the
ôentral subgroup'l[ x {e} of Go) its original topology and
makes Go into a topological group extension of ]t by G. Since
both'll and G are locally compact, so is G'. Go is unique'ly
determined, to within topologica'l isomorphism, by the
similarity class of o. In fact 'if o, - u¿, say
,r(h,9) = v(h)v(g)v(hg)-1 ,r(h,9), then the map
o : (t,g) -, (tv(g),g) is an isomorphism of Gtt onto G'2. By
Fubini's theorem, O iS measure preserving and it follows from
the definition of the topology of these groups that o is
biconti nuous.
Kl eppner (126 , Lemma 2I ) has g'i ven the fol'l owi ng
alternative definition of the topology on G'.
PROPOS ITI0N 6.1 , (1,26, Lemma 2l ) . The set;s
(U ,. f )(U * F)-1, uhev,e U runs through a basís of
neighbouthoods of I ìn T aná t through tlte sets of posi.tioe
measu,re in G, forTn a basis fon the neighbounhoods of (I,e) in
^ (¡¡t¡.
Let H be a closed subgroup of G. Then to restricted to
H x H is a multiplier on H which we also denote by ,, and Ho
24.
is algebraical]y isomorphic to a subgroup of Go. Moreover, we
have as an immediate consequence of Proposition 6"1, the
fol I owj ng resul t.
PROP0SITION 6.2 ([3, Lemma 1.1]). Let H be a cLosed
subgroup of G" Then the ineLusion map H' * G ^ í," o
bicontínuous isomoz'phism of H'uith a eLosed, subgz'oup of G^.
Let G be a locally compact group and o a Bore'l
mul ti pl i er on G. The twi sted L l-al gebra L t (G, o; i s the space
LI(C) of complex-valued integrabìe functjons with
mul tì pl i cat'i on def i ned bY
f * h(x) = f(g)h(g-1x) (g,s-rx)dgG
and with a *-operation defined bY
f* (x) = f (x-i )- ¡(x-1)
where - denr¡tes compìex coniugation and ¡ is the nlodular
function. One verÍfies at once that LI(G,to) ìs a Banach
*-algebra possessing an approx'imate identity, which is
determined up to isomorphic *-'isomorphism by the equivalence
cl ass of co. In general Lt (G, r.,r) i s not a C*-al gebr a (17 ,
13.3.61 ) . However Theorem 5 .2 app'l i es . The C* *compl eti on of
Ll(G,ur¡ is called the tr^¡isted group C*-algebra of (G,or) and
is denoted by C*(G,r). If t¡ is identically I' ihen u¡e
delete all mention of it so that C*(G,r) becomes C*(G) which
is called the group C*-algebra of G.
21
Let G be a locally compact group and o a Borel
multipl'ier on G. The construction of C*(G,o) is useful
because its representations correspond to multiplier
representations of G. A mult'iplier representation (or
6-representation) in a Hilbert space Hn'is a map of G into the
space of r¡nitary operators U(Hr) such that
g -+ n(g)E is measurable, E . Hn,
n(g)n(h) = ,(g,yh)n(gh)' 9,h < G.
The concepts oî 'd'imension', 'equivalence', 'direct sum',
'V(n)', 'irreducibility', and 'factor' that we defined for
representations of algebras apply to this situation equally as
well, they are independent of the object being represettted.
For details see Gaal [15, IV.1, page 145]. [l|e let the reader
make the obvious definitions. However if nt is an ot'
representation of G (i = 1, 2), then *r a n2_ (¿etined in a
manner similar to that used for representations of algebras)
i s an orrurr-FêpFesentat'ion, where uoro, i s the mul ti pi i er
(g,lh) * u,r(g,fh)rz(g,h),g,h c G. The set of equivalence
c'lasses of irreducible 0r-representations ìs denoted by
(G,r)^. An o-representation where the mul tipl ier o is
i denti cal 'ly 1 i s cal I ed an ordi nary representati on or s imp'ly a
representation of G"
Mackey has shown that n * no where
,,0(t,g) = tnn(x), (t,g) . Gs, n e z
DC
is a bijection between the set of equivalence classes of r[-representations of G ancl the set of classes of (ord'inary)
representations n0 of Go such that n0(t,g) = thrr0(1,g), al1
(t,g) . G' (see [26, Corolìary to Theorem 1]).
The following theorem establishes the desired connection.
THEOREM 6.3 (17, 13.3.5l ). Let G be a LocaLLy contpaet
group and u a BoreL rruLtipLiez' on G. Fon each u-representation
r of G, put
n'(f¡ = "(g)f(g)dg,G
f e l-l(G,c,r¡, then ' : îr -> îr' is a bíjeetion betueen
u=Tepvesentations of G and. nepresentations of the Banach
*-aLg,zbra LI (G,t'r ¡.
This together with Theoren 5.2 sets up a one-to-one
comespondence between ür-representat'ions of G and the
representations of C*(G,r). As the notation already
suggests, this correspondence preserves d'imension,
irreducibil ity, equìvalence, V(n), direct sum and primaryness
etc. We use it to transfer the topology on g*(G,o)^ to
(G,r)^. One can show that for G abelian and t¡ trivial, this
topology on (G,r)^ coincides with that given for abelian
groups in Section 1. Note that the above remark together
with Theorem 5.3 also shows that there exist imeducible
o.-representations of G for every multip'lien o. l,Je say that
the pair (G,r) is type I (or G is r¡-type I) if C*(G,r) is
o')LJ.
type I.
7. Moore qroups and the reqular representation
Let G be a locally compact group. ble adopt the fol'lowing
notation consistently throughout. Let H be a subgroup of G,
then [G:H] denotes the index of H in G, H- the closure of H
and H' the commutator subgroup of H - the subgroup of G
generated by the elements {ghg-t¡-t : 9,h e G}. If dim n is
finite for all r e G^, then G is called a Moore group (after
C.C. Moore who characterized these groups 'in t31l ). For any
local'ly compait group G, the von Neumann kernel is the closed
normal subgroup Go of G given by
G =O{ker n : ri < G^ and dim t < -}0
THEOREM 7.1. Let G be a LocaLLy cornpaet gnoup and H a
cLosed subgr.oup of finite indeæ in G, then Ho = G0.
PR00F. Suppose X e Ho and n is a finite dimensionai
representation of G, t.hen n restricted to H is a finite
dimensional represenbation of H, So rr(x) = I. It follows
that Ho s Go. Since H/H' is maximalìy almost period'ic and
lG/Ho : 1'llHoJ is finite, by Heyer [19, Satz 7.2.2], G/11o is
also maximal'ly almost period'ic, thus Ho = n{ker n : n is a
finite dimensional representation of G lifted fnom G/HoÌ z Go.
Let G be a loca1ly compact group. We denote bV Gtt the
topological finite class group of G, that is the closed
'A
subgroup of G consisting of the elements x in G such that the
set {g-lxg : g e G} has compact closure. If G = GFC, then G
is called a (topological ) FC group.
These concepts alone allow for a characterization of
Moore groups.
THEOREM 7.2 (Robertson [38, Theorem 1]). Let G be a
LocaLLy eontpact gï'oup. 'Ihen G is a Moore group if d'rld' onLy if
G satisfies the foLloníng properties.
(i)
(ii)('L1,1, )
íe:Gpçl<æ,(Grc ) '- is contpact, and
Go = {e}.
This theorem is due to Robertson. 0ther proofs can be
found in Kaniuth [23, page 233] and Poguntke [35' Satz 3.4].
The proof is far from trivial and we omit it.
The statement of 7.2 can be changed slightly' without
much difficulty, to give the following.
PROPoSrTrON 7.3 ( t38l ) . The LocaLLy conr¡>act gz'oup G ís
Moore if and. onLy if there eæists a eLosed sttbgt'oup K of G
sJrch t?tnt [G:K] ( -, Kl' is eornpaet md Ko = {e}.
Kaniuth (t231) has shown that an SIN group, that is a
locally compact group which has basis of neìghbourhoods at
the identity consisting of sets which are invariant under the
25.
G-action of conjugation, is a Moore group if and only'if it is
type I.
Moore groups can also be characterized using the regular
representation. Let G be a locally compact group. The left
reguìar representation p of G in L2(G) is defined by
(P (g)f )(x) = f (g-rx) '
X,9 € G, f e L2(G). It is clear the p is indeed a
representation. Similar'ly we define the right regular
representatiorí r
(r(g)f)(x) = f(xg)¡(x) h
9,x e G, f e L2(G), where in this context, ¡ denotes the
modular functjon. Denote by v(G) tne von Neumann algebra
V(p) generated by the operators p(g), g e G, and by V'(G) the
von Neumann algebra generated by the operators r(g)' g e G.
The folìowing theorem relating V(G) to V'(G) is much
deeper than its statement indicates.
THEOREM 7.4 ([46, Theorem 3]). Let G be a LoeaLLy
compaet gaoup. Adapt the abotse notation, thett Y '(G) = V(G)'
(on equiuaLentLy V'(G)' = V(G)).
PROPOSITION 7.5 (Taylor [48, Proposition 4.i]), Let G
t
be a LoeaLLy conpaei gz,oup. Tnen V (G) ¿s a finite uon Neunartn
26.
aLgebra if anã onLy íf G is a SIN group.
Thus if G is a SIN group which is type I, then V(G) is
type I, (tf¡at is both type I and finite. See Section 3.)
Indeed the following is true.
THEOREM 7.6 (Kaniuth 123, Satz 3l). A LocaLLg cotnpact
group G is a tr4oore gï'oup if anÅ. onLy if V(G) ¿s tUpe If .
This was first provecl by Kaniuth in [23]. An alternatjve
proof appears 'in Taylor [48, Coroìlary 1 to Theorem 4]. We
can combine al'l these results in one sjngle statement.
THE0REM 7.7, (t23, 3Bl ). Let G be a T.oeaLLy eornpact
group. The foLLouing statements ate equíuaLent.
ø)
(ií.)
('t 1,'t, )
G is a Moone group.
V(c) ¿s type lr.The foLLouing propev'ties are satisfíed.
lc: GFCI ( æ
(og) is cornpact, and
Go = {e}.
Let G be a locally compact group and e the maximal type
I, Projection in V(G). l^le will be interested jn conditions on
G such that e r 0. Kaniuth has given such conditions.
THEORIM 7.8 (Kaniuth [23, Satz 2] ). The ma,æimaL tgpe It
27.
centna.L projection ìn U (G) is non-zero if an'd only if the
foLLouÌng condítions lrcLd.
(í,)
øi)tG,Grcl < @
(GrC)'- is conrpact.
G is fupe f.
V(G) ¿s type I (or equiuaLentLy type If).
Ihe centre of GFC, that is ig . G : gh = hg aLL
h e Gpç \ has finite ivt¡7eæ ín G.
G is a Moore group.
An alternative proof of th'is theorem appears 'in Taylor
[a8] ; he al so proved the fol I ow'ing.
THEOREM 7 .9, (Tay'lor [48, Theorem 4 and Corol I ary 4] ) .
Suppose the manimaL type I, centi"aT' proiection e in V (G) í.s
non-zero, thevi the Don Newnøtn ketweL Go is corupaet anÅ eV(G)
is spatiaLLy isomozphic úo V(G/G').
As we pointed out earlier, for SIN groups, V(G) is type
I if and only if it is type If. In particular this is true of
discrete groups. Both Kaniuth and Thoma have given
characterizat'ions of type I discrete groups. We state both of
these in the following resuJt.
THEORTM 7.10 (Thoma [49], Kanjuth 1221, Smith t45l).
Let G be a. diserete gz'ottp, then the foLLoaing az'e equittaLent.
ø)
ftì.)
ø¿i)
ftu)
28.
The equivalence of (i) and (iii) is due to Thoma and that
between (i'i ) and (iii ) is due to Kaniuth. An alternative
proof of '(ii ) ìs equivalent to (ì j'i )' is g'iven by Smith t451 .
The main difficulty of this theorem lies in constructìng an
abelian subgroup of finite index in G. Different ways of
dojng th'is (us'ing different but equ'!valent hypotheses) can be
found in Isaacs and Passman [21] and Schl ichting 1421.
Much of this thesis cons'ists of results genera'lizing the
theorems of this section to locally compact groups with non-
trivial multip]iers, and we do this mostly using the methods
developed in Smith t45l and Taylor t481.
B. Induced repre sentati ons, Mackey's constructi on
Let G be a separable ìocally compact group, K â closed
subgroup, o â Borel multiplier on G and n âh o-Fêpresentatìon
of K, The induced representat'ion t13 ts an ür-representation
of G defined on a Hilbert space ll. Following Auslander l2l,we let a be the measurable function cr (g,X) = d(g.u )/du 'g € G, x e G/K, (for more details see a'lso Mackay Í271 and
Blattner t4l). Define H to be the space of measurable
functions f from G to H,, (that is,9 +(f(g),8> is measurable
for each E € Hn) such that f(gk) = n(k)-Ir(g) and
lrl e t2(G/ K,u). The last condjtjon needs a note of
explanation; observe that lf (gk)l = lt(g)l s'ince n is unitary,
so that lfl is realìy a scalar function on G/K. Then define
t'fflts)r) (s) = f(g-1 s)o (g,h(t))'%(s"r,5)-r t
29.
where h is the canonical map from G to G/K. One verifies
ihat î,ff is a unitary operator ancl that g *.lltg) is an ur-
representatìon of G. One also verifies that nffl, up to
unitary equivalence does not depend on the choice of u as the
notation already suggests. The notion of induced
representatjon is compatible with tak'ing central extensions,
as indicated in the following result. If n is an
o-representation of a group G, denote by tto the corresponding
(ordinary) representation of the group extension Go.
PROPOSITION 8.1 ([3, Lemma 1.2]). Let Kbe a eLosed
subgroup of a'separabLe LoeaLLy eonrpact gnoup G uith
norrnaLized BoreL muLtipLiey u. t¡ is an u-Yepresentation of
K, tt"n GIl)o od "|ii
ore equiuaLent.
PR00F. It is quickly checked that the map f + f', where. ^û)
f ,(x) = f (1,x), for all f in the Hilbert space of ,,OJu,., setslHt-
up the desired equivalence.
THEOREM B 2 (Mackey [27, Theorem 4.1] ). Let H and K be
cLosed subgnoups of a separabLe LocaLLy compczct gtoup G uith
Boz,eL rm,fltipLien u, srch that H E K. If r is an
u-repr.eserLtation of H, ttnn ,!f; is equiuaLent to t"1[11fl.
THEOREM B. 3. (Mackey [27, Theorem ].0.11). Let G be a
sepaz'abLe LocaLLy eornpact group uith BoreL rrtLltípLier u. If
H ì,s a eLosed subgrot'tp and n ¡, i . J a eoLLeel;ion of
6-r.epv,esentatíons of H, then (e1.-lnj)ff; ;" equiuaLent to
*r., ("rîff)'
3C
Let G be a separable local]y compact group with rnultipìier
o, ônd H a closed normal subgroup such that (H,r) is type I
(see Section 6). If n e (H,r)^ and 9 e G, then n9 will be the
o-representation of H defined bY
ng(h) = r,(9-1hg)r-1(h,g)r(g,g-1hg), h e H.
It is easy to check that the action 9 * (n -, n9) of G on
(H,r)^ satisfies n9h = (nh)9, ail h,g e G. The set {g e G:
n9 it equivalent to n] is a closed subgroup of G cajled the
stabilizer of n and is denoted bY Kn. An orbit is a subset
of (H,r)^ of dhe form {n9 : g € G} for some irreducible
r,l-representation n of H.
The dual (H,r)^ has a natural Borel structure as
defined in Auslander [2]. Fell [11., page 95] has observed
that this Borel structure is iust that generated by the
topo'logy of (H,r)^.
Now each printary rrr-Fêpl"êsêntatiorì n of G determines a
projectìon valued measurê u',, oh (ll,o)^ which is unique up to
equivalence (two measures being equivalent if they ha.ve the
same null sets), whose values are projectìons on the Hilbert
space of the representation n, and which'is a quas'i-orbit in
the sense that
(i ) un(g.A) = ur,(A) for alt Borel sets A in (H,ur)^
andgeGrand
if A is an 'invariant Borel set (tfiat is g.A = A(ii )
31.
for all g u G) then either A or its complement
has un-measure zero,
(tnat is, Þn is G-invariant and ergodic). For deta'ils see
Mackey [28, 30] and Auslander l?1, The measut. 'rrn is said to
be transit'ive if it is concentrated on an orbit o (that is
u,r((H,.)^ - o) = o). '
THE0REM 8.4, (Effros [8, Theorem 2.6]). Let G be a
sepa.z,abLe LoeaLLy cornpact grory and w a BoreL rm'út'LpLier on G-
Let H be a eLosed norrnaL suhgroup sueh th.at (H,r) ís type I.
If 1;lrc G-orbíís ¿n (H,r)^ are LocaLLy cLosed (a set is
LocaLLy cLosed ¿f it ís the intez'section of an open anÅ.
cLosed set), then v*is transitiue for each pnimary
u-r.epresentation x of G. ConuerseLA íf (H,t)^ has 17þ non-
transitiue ergodic meq.sLlres, then the G orbits in (l'lro)n are
LocaLLy cLosed.
THEOREM 8.5 (Mackey [28, Theorem 8.1]). Let G" u and H
be as ín Theorem 8.4. Let 0 be at¿ orbit ín (H,r)^ and r an
eLement in o uíth stabiLize, Kn. Then the mapping À * ÀffTT
sets up a one-to-one corvesponÃ.ence betueen the 'pr"Lmary
$-r.epresentati.ons of Kn uhose restriction to H is a ntuLtipLe
of r" and the pr'ìmar'y u-?ep?esentations x of G such that v*is
concentrated on a. Fu.r'therTnore, the two Don Natnnqnn aLgebras
Y(x) and. V(ÀTl ) oon *-ísornorphie"1t
THTOREM 8.6 (Mackey [28, Theorem 8.2]). I'et G, u and. H
be as in Theorenr 8.4. If n is an irreducíbLe u-?epresentation
at
of H, then thev,e eæì.sts a muLtípLierc of Kn/|1 and a "ctu'
representation r' of Kr uhere r' ís the Lifting of r to Krr,
srch tLuzb n'(h) = n(h) aLL h e H.
TIIEOREM 8 7 (Mackey [28, Theorem 8.3]). Let G, o, H, rr
Kn, r' and. r be as in Theorem 8.6. Eoz' each r-r-nepresentation
ì, of KnlH, denote bA \' the (r')-L-rnpresentation of Rn
obtained by conrposing )' utLth the emtonícaL m6P Kn -t Kn/H. Then
À -> À' e fi' sets up a one-to-one covrespoluJerrce (eqttíuaLent
represenl;qtionß being identified) betueen the set of prLrnat'y
,-L-nepz,esental;ions of Kn/H ar¡Å. the set of pr"imary
u-:repvesentations of Kn zohich v'edtrce on H 1;o a truLtípLe of n.
Ew,bherrnoz,e, the tuo Þon Neumarm aLgebyas V(À) an¿ U (x' ø n')
ave *-isontonphic.
l,'le can summarize Theorems 8.4 to 8.7 as follows. Let G,
rr, H, r, Krr, n' and t be as in Theorem 8.6. Then
f * (f' ø rr')J G
KTT
sets up a one-to-one correspondence between t.he primary
t-I-representations of K.n/H and the primaF) trr-Fêpresentations
x of G such that u* is concentrated on the orbit contain'ing r.
The two von Neumann algebras V(¡,) and V((r'* '')1f") are
*-isomorphic. Furthermore, if the G-orbits in (H,r)^ are
locally closed, then as n vôries through (H,o)^, the above
construction yields all the primary o-representations of G.
33.
CHAPTER I I
ONT HE STRUCTURE OF O-TYPE I
LOCALLY COMPACT ABELIAN GROUPS
Let o be a normaljzed mult'iplier on a localiy compact
abelian group G. Baggett and Kleppner I3l have given a
useful criterion for deciding when (G,r) is type I. Holvever
not much is known about the structure of type I pairs (G,r).
In this chapter we investigate the structure of ìoca'l'ly
compact abel ian groups that admi t o-type I mul ti p],r'ers. In
particular, a comp'lete structure theory for a certain class
of such groups is given. (See Theorem 4.5, 5.11 and
Corollary 5.12.)
1 Notation and elementary facts
Throughout this chapter, all groups are local1,rr
compact and abelian (tnis includes discrete groups), and all
multipliers are normalized and Borel measurable, unless
otherwise sta+,ed. [,le adopt the notation of Chapter I
Section 1.
Let o be a normalized multipìier on the locally compact
abelian group G. Denote by õ the ant'isymmetrized form of o,
that is
õ(k,g) = r(t,g)/r(g,k)
?¿,
k,g e G.
PROPOSITION 1.1. ([25, Propositjon 1.1, Lemma 7"1 and
Lemma 7 .21). Let G be a LocaLLy eompaet abeLian group and u a
norrna1,ized rm'LLtipLier on G, I,/e haue
G)
Gi)
õ : G x G '>Tl'is a eontinuotæ bicLnraeter.
g * õ('rg) is a continuotæ homomorphism from G to
G
(ì.ii) õ(k,g) = 1 aLL k,9 e G if qnd onLy íf u is
tríuíaL,
The map g * ñ(.,g) is also denoted by the s¡nrbol õ. No
confusion should arise from this ambigu'ity.
PROOF. (i) The bilinearity of õ follows from the
equati on
õ(gh,k) ='(gh,k)r(k,gh)"I
= r(gh, k)r(g,h)r(g,ir)-tr1k,gh)-t
= r(g, hk)o(h,k)r(kg, h)-1r(k,g )-1
= r(g,hk)o(h,k)o(g,k)r(g, k)-1,(gk, h)-1r(k,g )-r
= o(g, hk)o (h, k)o(g, k)r(g, kh) -rr(k, h) -1'(k, g )-r
= õ(g,k)õ(h,k)'
g,h,k e G. Clearly õ(k,g) is a measurable character in k for
fixed g e G. It follows that õ(k,g) is continuous in k for
f ixed g (t18 , 22.191 ). Sirnilarly o(k,g) is continuous in g
for fixed k. By [26, Corollary to Lemma 1]' õ is
35.
continuous as a function of two variables at the identity in
G x G. Using this, we show that õ is cont,inuous. Let (g'h)
be an arbitrary point of G x G and let V be a neighbourhood of
1 in'll'. Let V, be a neighbourhood of 1 ìn T such that Vl s V.
Because õ is continuous at the identity in G x G, t,here exists
a neighbourhood U of the identity in G such that õ(U,U) s Vr.
The neighbourhood can be chosen so that õ(U'g) s V, and
õ(t<,u) s vr. Now õ(ku,su) = õ(k,g)õ(k,u)õ(u,g)õ(u,u)
õ(k,g)Vl s õ(k,g)V and õ is continuous at (k,g).
(ii) Since õ(",g) is a measurable function of G, as in
(i), it is continuous. That õ : G + G^ is a homomorphism
follows from the bil inearity of õ. Fclr the continuìty, ute
observe that the sets p = {X e G^ : X(C) s V}, where C is a
compact subset of G and V a neighbourhood of the identity in
11, is a basis for the neighbourhoods at the identity in G^.
Since õ : G x G +'ìl is continuous, if go . W = {g e G :
õ(g) e P] is fixecl and k e C, there ex'ists an open set U x U'
containing (k,go) such that õ(U,U') s V. But C being compact
can be covered by a f Ínite fam'ily Ur' ..., Un of such sets U,
and if U'= ni=l,...,n U.l is the jntersect'ion of the
corresponding U' sets, we have õ(C'U') s V. Thus if go e W'
there exists an open set U'such that g0 € U's l'1, proving
that W is open and õ continuous.
(iii) Let n be an irreducible ur-representation of G
(such representat.ions exist - see remark follo¡jng Theorem
I.6.3). If õ = 1, then n(g)n(h) = n(h)"(g) for" aì1 g,h e G,
thus by the remark following I.5.1, n is one dimensional and
to is trivial .
36
A normalized multiplier o on a locally compact abelian
group G is called non-degenerate (or totally skew) if õ is an
inject'ion. Given any subset S of G, v'/e denote by Sur the
subgroup So = {g e G: o(s,g) = 1 all s e S}. Since So is
the intersection nr.r{S}o of closed sets, it must be closed.
S is called isotropic if Sur = S and max'imal ìsotropic ifSo = S. An application of Zorn's lemma (see Hannabus II7,1.61) ensures that each ìsotropic set is contained in a
maximal isotrop'ic subgroup. The operation on sets S + Sr'r is
inclusion reversing and (uSo) = n(Sour),
So= {g e G : õ(s,g) = 1 all s e S}
= {g e G : õ(g) e A¡G^,SI}
= (õ)-r (A tG^,Sl ),
thus ñ(So) = AIG^,sl n õ(e ) and
AtG,õ(S)l = {g e G : x(g) = 1 all x e õ(S)}
= {g e G: r(s,g) = 1all s e S}
=So.
The closure of õ(G) is a cJosed subgroup in G^, thus by
r.i.1(ii),
õ(G)- = AIG^,4[G,õ(G)]l = AIG^,Grl .
S . G,nyFor a
Since Gur = ker õ, if r¡ is non-degenerate (tfrat is Gt¡ = ieÌ),
37.
then the range of õ is dense in G^.
PROPOSITION 1 2 . I'et u be a non-degenerate mul'típLiez' on
G. Then it is an open map if, and onLy i,f ß is a bicontinuous
ísomorphísm, Moreoùer,, íf G is separable anå. l;he range of ã
ís cLosed, then ã ís a bí¿ontínuous isomozphism.
PR00F. If õ is an open map, then õ(G) is open and
closed, hence õ'is onto. If G is separable and õ(G) = G^, then
by [18, 5.291, õ is open.
The condition that õ is a bicontinuous-isomorph'ism
eradicates a great dea'l of pathology and forces some order
onto the structure of the group G (for instance'it must be
self dua'l ). It is not surprising then that it is equivalent
to o being non-degenerate and type I.
THEOREM 1.3, (Baggett and Kleppner [3, Theorem 3.2]).
Let u be a non-degenerate rm'LLtipL'Ler on the LocaLLy cornpact
abeLian gnoup G, then (G,r) ís type I if, anå. onLy if
õ : G + G^ : g + õ(",g) i,s a topo\ogicaL isomorphism.
PR00F. lnle give an outl ine of the proof of thìs Theorem
for G separable. For more debails and for a proof in the
case where G is not separable, see llannabuss t17l and Baggett
and Kl eppner t3l .
Let G be separabìe. By 1.3 it is enough to prove that
(G,o) is type I if and only if õ (C) is cìosed. Let H be a
38.
maximal isotropic subgroup of G. Since o restricted to H is
trivial (1.1 (iii)), (H,r)^ is isomorphic to the abeljan group
dual H^. Suppose õ is a topologicaì 'isomorphism. Then the
G-orbit of an element in H^ is of the form {õ(.,g) : g e G}
which is closed by assumption. Thus by Theorem I.8.4, Mackey's
construction with H as the closed normal subgroup gives all
the factor representations of G. Because u is non-degenerate,
the stabiljzer of the orbit {õ(.,g) : g € G} is H, thus the
o-representations obtained using Mackey's construction are all
type I.
Conversely suppose (G,r) is type I. Again by the non-
degeneracy of o, the stabilizer of each element in H^ is iust
H. In particular the action of G on (H,.)^ is essentially
free, and under these condÍtions, the projective version of
Auslander 1,2, II, Proposit'ion 3.11 asserts that there are no
ergodic measures on (H,r)^ which are not transitive. Thus
by Theorem I.8.4, the G-orb'its in (l-t,ur)^ are local'ly c]osed.
We decluce that the subgroup {õ(.,g) : g e G} of G^ is loca]ly
closed. But we know that its closure is G^, so {õ(',9) :
9 u GÌ is an open subgroup of G^. Open subgroups are also
closed, so it must be all of G^.
PROPOSITION 1.4. Suppose Gto g S for some subset S of G
and i:(G) ís cLosed (in partí,cuLan' tLtis is trwe df u is rnn-
degenenate arú- type I)" then
õ(Sr) = A[G^,S], anã
(St¡)o = K ,
39
uhere K is the smaLLest cLosed subgz'oup containing S.
PRC0F. From our earJier remarks, õ(G) = A[G^,Gur] =
A[G^,S], so
õ(So) = A[G^,S] n A[G^,Gur] = A[G^,S], anC
(So)o = A[G,õ(So)] = A[G,AlG^,S]l = (.
We collect this result together with some other
elementary facts as a proposition.
PR0POSITION 1.5. Let u be non-degenenate and l;gpe I on
the LocaLLy conrpact and. abeLian group G. Suppose K and L are
cLosed subgroups of G such that K c L" then
(í. )
fti)
('t'1'1, )
Øu)
(Ko)o = K, õ(Kur) = A[G^,K] an¿ Kt¡ = AIG,õ(K)1.
K is eonrpact k,espeetíueLy open) if, and onLy if
Ku ís opøn k'espeetiueLy cornpact).
(L/K)^ is topoLogieaLLy isomorphie uíth Kr¡/Lr,r.
rf So is a coLLeetion of subsets in G" then
(nSo)ur = (u(Sour))-.
PR00F. (i) follows from 1.4 and earìier remarks, and
(ii) follows from (i),1.3 and I.1.1.(iv). To see (iij)' use
I.1.1. (i i i ) and observe that
Kur/Lr,r: õ(Kr¡)/õ(Lur) = A[G^,K]/AtG^,Ll : (L/K)^ ,
where the symbol rr:rr denotes topol ogica'l i somorphi sm.
40
2. Some groups don't admit multipliers
Denote by lPthe circle group - the group (under
mul ti pl i cati on ) of al I comp'l ex numbers of rnodul us one (wi th
the induced topology), byR the group of real numbers, by Z
the integers, by Q the rational numbers and by Z(n) the
cyclic group of order n. For a fixed prime p, let Z(p-) be
the subgroup of 'll consisting of elenlents whose order is a
power of p, ÂO the group of p-adic integers and nO the group
of p-adic numbers (see Append'ix) .
LEMMA 2J. Suppose the abeLían group G has the
properl;y that for aLL x e G, 1;here eæis1;s an h e G such tha't
h2 = x. rf w is a nnútipLien on G srch thn'b å(x,y) = õ(y,x)
aLL x,Y e G, then u is triuiaL.
PR00F. õ(x,y) = o(x,v)/u(y,x) = ,(y,x)/r(x,y) = 12(y,x)
implies õ(x,y)2 = l all X,Y e G. Given any x,V. G, choose
h e G such that h2 = x, then õ(x,y) = õ(h2o.y) = õz(h,x) = 1.
That iS o is symmetric, thus trivial (1"1 (iii)).
LEMMA ¿.2, (Kleppner [25, Lemma 7.5]). Let G be a{..r^
disey,ete group uith muLtipLíer u. -rf õ(x,y) = õ(y,x\aLL
X,j e G, then u ís símiLar to a muLtípLien Lífted from a
quotient G/H uhich is of erponent 2 (that is eaeh e1'ement of
G/H has ordev, at tnost 2.).
PROOF. The mul t'ipl ier õ i s synunetrjc, so by 1.1 (i i i )
41.
there exists a function Y : G +'ll such that
õ(x,y) = Y(x)v(v)v(xy)-1. simi'larly, there exists a function
r : G +'lf such that r(x,y)o(y,x) = r(x).(y).(xy)-1 for all
X,! e G. Thus ur(x,y)z = vt(x)vt(y)vt(xy)-l for al1 x,y e G.
This'imp'lies t¡ is similar to a mu'lt'ipl'iêì trr, such that
11 = l. In fact ur,(x,y) = (yr(xy)vr(x)-Iyr(y)-1)"r(^,y),
where the square root is chosen in any fashion. Let H = Grr.
Since õr(x2,y) = rr(x,v)2 = l all X,y e G, lve have x2-e H all
x < G, that is G/H has exponent 2. Since ürr[HrH is trivial
and urf = 1, there exists t, : H +'ll such that
rr(x,v) = rr(x)tr(v)rr(xv)-I and rf is a character of H. Let
r' be an exterision of rl to G (t18, 24.12)), and defjne t, bV
tr(x) =t'(x)t'"h..e the square root is chosen in any fashion
with the restriction 11(x) = rr(x) aìl x e H. Let
urr(x,y) = ,r (x,y)"2(xy).2(*)-1.r(v)-1, then ,7 = l,rr(x,Y) = rr(Y,x) ali (x,y) e l{ x G and rr(x,Y) = l ali
xry e H.
Let, n be an irreducible or-representation of G and {9o}
a set of coset representatives rnodulo H containing the
identity of G. Define tto as follows
"o(x) = n(x), atrd
tro(xgo) = tt(x)n(go), for X e H.
Clearly n0 is a multipìier representation whose associated
multipl ier ur, is simil ar to ur, (and consequently o). Further-
more, for x,y e H,
no(xy9*) = "(xy)"(go)
42.
= TT(x)n(y)"(go)
= no (x)no (vso) ,
thus urr(x,y) = l all (x,y) e H x G, and by the symmetry of r,r.,
,r(V,x) = 1, (x,y) e H x G. From this fol'lows that ur, is
constant on the H x H cosets in G x G. Indeed, if g,h e H,
then o, (gx,hy) = ,, (g,x)u,¡ (gx,hy) = ,, (9,xhy),, (x,hy) =
or(x,hy) = tor(x,vh)or(v,h) = ra(x,y)r¡(xy,h) = rg(x,y). Thus
there exists a multìpìiêr t¡ron G/H whose lifting to G is
simi I ar to t¡. Thi s proves the I emma
Part of ífr. tollow'ing result can also be obtained from
[6, Lemma 2].
THE0REM 2.3. Let G be a discnete gz'oup uhich ís eíther
eycLic ot, of the form 7(p-) or Q,. rf u is a rruLtipLiez' on G,
then u is triuíaL.
PR00F. Case 1. Suppose G is cyclic. If it is finite,
say G = L(n), then õ(.,1), being a character of Z(n), is of
the form õ(p,1) = expl2nikp/n]; it now follows from the
bilinearity of õ that
õ(p,q) = exp[2nikpq/n]
p,Q e 7(n), k e 7. If G is infinite, then G is isomorph'ic
with V., and by the same reasoning as above,
õ(p,q) = exp[2nipqo]
43
p,g € V-for some G e [0,1[. At any raie õ is symmetric. The
only quotient of a cyc'lic group which is of exponent 2 is
Z(2) or the trivial group. Since a multip]ier r^r' on V.(2) can
be assumed to be normalized, we have 1. = o'(-1,1) =
o'(1,-1) = t,'(1,1), that is o' is trivial . Hence by
Lemma 2.2, u is trivial .
Case 2. The dual of Z(p-) is the group of p-adic
integers nO (see Appendix) which is torsion free. Thus the
homomorphism õ : Z(p*) - aO must be trivial. Hence õ = 1 and
õ is trivial.
Case 3. Q^ is torsion free and divisible (see Appendix).
If x e Q^, then for each n e 7, x + x(x/n), X e Q js a
character of Q wh'ich we denote by xlln and is the only
character of Q satisfying (vI/n)n = x (use the fact that Q^
is torsion free). Thus if the map õ: Q * Q^ maps 1 to x,
then õ( .,!/n) = xI/n, that is
õ(s/m ,t/n) = -[*^] t
this is symmetric, Q is divìsible, so by Lemma 2.1, r¡ is
trivial.
LEMMA 2.4. Let us be a Boz,eL rruLtipLier on a LocaLLy
contpact abeLian group G. If H is a dense subgz'oup of G such
that the restrietion of u to H is triuíaL, them u is triuiaL.
PR00F. õ : G x G -+'ll is continuous (1.1 (i)). The
44.
restriction õ : H x H +'[' is the trivial ffiðP, thus õ = 1 and ur
is rrivial (1.1 (iii)).
COROLLARY 2.5. A BoreL muLtipLíer on 67LA one of the
foLLouing groups.' 1ll lR, aO and ar; is tniuiaL.
PR00F. 7(p-) is dense 'in'lf, Q is dense in fR and nO, and
Z is dense 'in lO (see Appendix). Now use Lemma 2.4.
we know from I.1.3,(jii) that the dual of a torsion free
discrete group (for example Q^) is connected. The following
result shows that these groups clon't admit non-triviaì
mul ti p1 ì ers .
PROPOSITION 2 6 . Let G be a cornpaet {rou7. If G is
conneeted. (or equiuaLentLy diuisibLe (I.1.3.(iii))l then a
BoreL rm,LLtipLier u on G is triuiaL.
PR00F. The homomorphistn õ : G * G^ must be trivial
because G is connected and G^ discrete.
3. Some useful resul ts
THEOREM 3.1. Let G be a LoeaLLy eontpact abeLian
subgrotp uhieh has a cornpaet open subgt'oup K" St'ppose u is a
non-d.eganerate rm,tLtipLier' then u is type I if ' anÅ. onLy if G
eontains a eompact open rnaæí.maL ísctopíc subgzoup. MoreoÐev':
anA cornpaet maæírnaL isotropic H satísfi¿s õ(H) = A[G^,H]
aná G/H is topoLogicaLLy isomorphic to H^.
45
PR09F. Suppose ur is type I. By Proposition 1.5.(ii)'
lfu is compact and open, so K n lQo is a compact open isotrop'ic
subgroup. Without loss of generality, we assume Ko : K. Let<4|+? 4<'l--
H be a(mäximal isotropic subgroup, then õ(H) (being the
continuous image of a compact set) is compact and c'losed;
õ(G) is dense in G^, thus õ(H¡ = õ(Hur) = AIG^,Hl n õ(G)
= (AlG^,Hl n õ(G))- = AIG^,Hl n õ(e )- = AIG^,H1 , the fourth
equaìity being valid because A¡G^,H] is open anC closed. It
follows that õ restricted to H is a continuous isomorphism
onto A[G^,H] so it must be open [18, 5.29]. Let A be open in
G and A9 = ff n g-lH, g e G. The set A9 is open in H, thus the
set õ(A) - u{õ(g)õ(ng) : g e G}, being the union of open sets
in G^, is open in G^. The proposition now follows f,rom 1.3.
Let G be a locally compact abelian group for each
positìve integer n, denote by Gn the subgroup {gn t g e G}
and by Gn the closed subgroup {g e G : gñ = e}.
THEOREM 3 2" Let u be a non-degenerai;e tgpe I rruLtipLíer
on the separabLe abeLíon group G, then ufr is type | (n a
fiæed ù'tteger') if anå onLy if Gn is cLosed, in G"
PR00F. First observe that the proof of 1.3 extends
readily to assert that a multiplier ur (which is not
necessarily non-degenerate) on a (separabìe) group G is type
I if and only if the range of õ : G * G^ is closed. Let or be
as in the hypothesjs of the Theorem. By 1.3, G and G^ are
topoìogically isomorphic. The range of õn is (õ(c))n = (G^)n,
thus by the above remark, rr is type I if ancl only if Gn is
46.
closed in G"
THE0REM 3.3. Let G be a separabLe abeLian LoeaLLy eorrcpaet
gToLlp anÅ u a rm,LLtipLier on G. Thqn G6 ís l;ype I if and onLy
íf ^n is type I for aLL n e V-
PR00F. Observe that 'll i s a type I normal subgroup of G.
Si nce 'lI^ i s di screte, Go-orbi ts are cl osed, so by I .B .4, al I
factor representat'ions of Go are obtained using Mackey's
construction (I.8.5, I.8,6, I.8.7). Indeed, because 'll is
central , a factor representation n o'f= G0 reduces on Il to m.X,
where m is a iardinal anci x the character t * tn (t e n') of 'll
for some n e Z (see [15, IV.7.20])" Now m.x', where
X,(t,x) = x(t), ((t,x) n e') is a multip'lier extension of m.x
to Go, and the multiplìer associated with X' is precisely tr-n.
Thus by I.8.7, each n which restricts on 'll to a multiple of x
is type I if and onìy if r-h is type I, and Go is t¡rpe I if
and only if urn is type I all n e 7-.
Suppose G, ur and H are as in Theorem 3.1 and G is a group
of exponent p for some prime p (tfrat is each element in G has
order at most p). Since G is a vector space over the field of
p e'lements and H is a subspace it admìts a comp'letnent, that is
a subgroup K (isomorph'ic to G/H s H^) such that G = H x K.
Since H is open, we actuaìly have a topological isonorphism
between G and H x H^.
Such decornpos'itìons of G wher^e we 'insist that H is
compact and open cannot a'lways be found even if we ajlow the
47.
additionaì hypothesis that rh js type I for all n e î. (or
equivaìently that Go is type I), as the following exampìe
shows.
EXAMPLE 3.4. (i ). Let H be a locally compact abelian
Then o defi ned bygroupandG=H*H^.
(x,r),(y,x) e G is a non-degenerate multiplier which according
to 1.3 is type I. Moreover, if G is divisible and torsion
free, then on is non-degenerate and type I for each n e Z
(Theorem 3.2) ,
(ii). Suppose we let H in (i) be the group of p-adic
numbers r¿-, then H^ is aìso the group of p-adic numbers (seep-
Appenc!ìx), that is G = Qp * np, in particular G is divisible
and torsion free, thus ,n i, non-degenerate and type I for
each n < Z-. By Theorem 3.1, G has a tnaximal isotropic
compact open subgroup H. [{e show that. no such H can be a
direct summand of G. For otherwise, G must be isomorphic
with H * H^; but the dual of H, H being compact and totally
disconnected, must be' torsion (I.1.3.(ii)), contradicting the
torsion freeness of G.
(iii). For each t . Z, let K., be the group
7(4) = aç: C4 = 1) and Kj the subgroup Z(2) = (C2) of Kr.
Define H to be the subgroup of nK., consisting of elements
(ui) such that a., . Ki for all but a finite number of
indjces i; and topologize H so that nK.i (with the compact
cartesian product topoìogy) becomes a compact open subgroup.
48
Let o be as in (i). ClearlY, G2 is not closed in G and by
Theorem 3,2, w2 is not type I. Furthermore, Theorem 3.3 shows
that although o is type I, Gt is not a type I group.
Given a localìy compact abelian group G and a non-
degenerate type I mult'ipìier, we can however hope for the
existence of a cìosed isotropic subgroup H of G such that G
decomposes as a product H x H^. Indeed the rest of thjs
chapter is devoted to showing that such a subgroup exists under
certain addit'ional hypothes'is on G, for example if G is
divisible and separable, or if the connected component of G is
open. )
THEOREM 3.5 (Mackey [28' 9.6] ) . Let H and K be tuo
LocaLLy cornpaet abeLiarl groups artri u a rru.LtipLien H x K. If
ûr' is the ftmction on (H t K) * (H * K) defíned bA
r,r'(hk,h'k') = *(h,h')r(k,k')r(k,h'),
(h,k)(h',k') . H x K, then u' ds amuLtipLien uhich is siniLar
to u.
PROOF. For (h,k),(h',k') e H * K, we have
(,J(hk,h'k')r(h,k) = r(h,kh'k' )r(k,h'k'), and
,(h, h' )r(hh' , kk' ) = ,(h, h'kk' ) r(h' , kk' ),
thus
/1C
ûr(hk,h'k') =üJ h'k' û) h' TJ hh' kk'k h
û) ûJ
k' UJ h' û) k h'k',h' u) k,k ü)
Ih kû)
û) k
where Y is the function on H x K defined by Y(hk) = o(h,k)
Now
ûr(h',k' )o(h',k)"(k,h'k' ) _t,r(k;h' )'(k,k' )r(h', kk' )
h
kû)(¡
OJ
kk üJ
û)
kh'üJ
k h
,k=l
Hence the result.
COROLLARY 3.6. Let H and K be Loe.aLLy eompact and
abeLian. rf u is a rnn-degenez'ate rm'fltipliez' on H x K such
that Ku = H, then u is simiLar to the rruLtipLier w' defined by
t,t' ((h,k)(h',k' )) = r(h,h' )co(k,k' ).
PR00F_. Sìnce K,¡ = H, õ(k,h) = l all (h,k) e H x K.
Hence the result follows by Theorem 3.5.
If the condjtions of Corolìary 3.6 are satisfied, then
we say that o splits relative to H and K.
THEOREM 3.7. Let ube a non-degenerate BoreL muL'bipLier
on a LoeaLLy comapct gz'oup G. Suppose G has a marimaL
isotz,opic subgroup H uhich is a topologieaL direct sutnnanã,
then u ís type I if and onLg íf it is sinr|Lar to a
rruLtipLier u, of the forrn
t
,,(ri,(x,r),{,(y,x)) = À(y),
50
(x,r),(y,x) e H x H^, for some topoLogicaL isomonphisrn
{, : H x H^ + G.
ìl
PR00F. Suppose such an isomorph'ism ü exists, Denote by
i ts dual 'i somorph'i sm, then
v
where uro is the multiplier ro((x,l),(y,X)) = r(y),
(x,r),(y,X) . H * H^. Clearly õo is a topological isomprphjsm,
thus so is õ. .3, 0t and o are tYPe I.
Converse'ly, suppose ur is type I, then G 'is topologically
isomorphic with H x H^ (see proof of 3.1), thus we assume
without loss of generaìity that G = H ,. H^. Define
p : G + G^ by [p (x,r)] (y,x) = x(x)r(y)-, then p is a
bicont'inuous isomorphism such that p (H) = A[G^,H] ancl
p(H^) =A[G^,H^]. Itfollows thatthemapp-l o õ: G+G is
a bicontinuous automorphism such that p-l . õ(H) = H and
hence p-1 . õ(H^) = H^; but õ(H^o) = A[G^,1-l^] = p(H^) = õ(H^),
thus ll^ur = H^. By [1, Theorem 1], there exists a bicontinuous
automorphism p, (respectívely rpr) of H (respectjvely H^) such
that p-I . õ(x,À) = (Vr(x),{,r(x)) for aìl (x,r) e G. Define
{l : G + G by ,1,(x,r) = (,i;I (x),r), then rp is a bicontinuous
autornorphi sm.
Since H (respectively H^) is maximaì 'isotropic, the
restriction of o to H (respectively H^) is trivial. Thus by
Theorem 3.5, we can (and do) assume that o is similar to a
(¡) v o û)0 I
51
mul ti pì i er tr defi ned byI
t,, ( (x,l), (y,x) ) = õ( (t,r), (y,1) ),
(x,r),(y,X) . G. Clearly
,r({,(x,r),,t,(y,x) ) = õ( (1,À)), (q,;I (v),1))
= [õ(,t,;I (v),,l/;t (1 ) )] (1,r)
= [p(y,1)] (1,¡,)
= r(y).
Thus when'ever G decomposes as a product ll * H^ with H
maximal isotropic, we know precisely the form which ur takes.
A multipìieF r¡ satisfying the hypothesis of the above
theorem i s cal I ed a cross mul ti pì ier.
EIUA_Lp_. Let H, K and L be LocaLLy eonrpact anå.
abeTiqn øtd u a non-degenenate type I rm'LLtipLíer on H x K x L
urch that the map i: z H x K x L + H^ * K^ * L^ satí'sfies
õ(H) = ¡'^, then
õ(f¡ = H^, õ(L) = L^ arú. Lu: = K x H.
PROI|. By 1,5, K^ = õ(tl) = õ((Hr¡)t¡) = A[G^,Hur] , so
Ho = A[G,A[G^,Hr]l = A[G,K^] = H x L (use I.1.1.(ii)), thus
õ(H * L) = õ(Hr) = A[G^,H] = K^ * L^, consequently õ(L) = ¡^
and õ(K) - H^" Fina]]y o(Lo) = A[G^,t-] - H^ x K^ = õ(H >< K),
that is Lo = H x K.
52.
Yo &0.-9
The folìowing theorem allows ur{{uronot other things) #ffi with finite products of groups.
THEOREM 3 9. Let G be a LocaLLy conrpact abeLian gv'oup
uíth a non-degenerate and tgpe I muLtipLier u, and suppose H
is a cLosed isotropic subgroup uhich ís a topoLogieaL dírect
swmnanã. of G. Then G is topologicaLly isomorphic tLth
H x H^ x Ho/H and. u is simiLar to a rruLtipLier u' of the foyrn
r' ((h,À,X) (h' ,l' ,X' ) ) = . ((h,r)(h' ,r' ) )o(*,*'),
uhexe r ís a eross muLtípLier on H x H^ and" o is a non-
degenerate anÅ. type I nntLtipLier on Ho/H.
PR00l. All isomorphÍsms stated in this proof are
topo'logical isomorphisms. There exists a closed subgroup K of
G such that G = H x K; now õ(Ht¡) = A[G^,H] = A[H^ x K^,|-l] = K^
is a topological direct summand of G^. The nrap õ is an
i somorphi sm, thus þlo i s al so a topol og'ical di rect summand,
thus there exists closed subgroups K, L of G such that
H x L Ì Ho and G: Ho x K. Collecting all this information
together, we have
G:HxKxL
Hi,¡ËHxL
L = Ho/H
KÈG/HoãH^ (use 1.6 (iii)).
This proves the first part of the theorenl about the structure
53.
of G. Now o(H) = A[G^,Hur] = A[H^ t K^ x L^,H t Li = K^, thus
by Lemma 3.8,
õ(r¡ = H^, õ(L) = L and Lo = K x H.
It now follows from Coroìlary 3.6 that o is similar to a
multiplier to where t is a multip'lier on K x H and o is a
multiplier on L. Since
lr**
is a bicontinu'ous isomorph'ism from H x K to H^ x K^ such that
i(H¡ = K^ and i(t<) = u, by Theorem 3.7, r is a cross
multip'lier on H * K È H * H^. F'inaì'ly,
all ffill = õlr
is a bicontinuous isomorphism fr'onl L to L", thus o is type I
and non-degenerate on L Ì Hr/H.
COROLLARY 3 10. Let u be a non-degenenate ru.LtipLí'er on
the fìníte abeLi,an group G, then G is isomozphie to H x H^
for some subgroup H of G arld. u is simiLav.' to a eross muLtipLier"
!R001-, Ciearly o is type I. Using the structure theory
of abelian groups (see I.1.4.(iji)) G must be a finite product
e = II!=1 H.' where each H., is a cycl ic group of prime power
order. For each i, (Hi)o = {g e G : õ(n,9) = 1 all h e Hi}.
By Theoren 2.3, ,lH.i*Hi is trivial , so (Ht), = Hl . Now
HxKHxT1Õt)
K
54,
Theorem 3.9 and induct'ion yields the requ'ired result.
C0ROLLARY 3.11. Let u be a non-degenerate nuLtipLier on
Rn, then n is euen (n = 2k), u is type I anå. is a cnoss
muLtipLie, onRk * tRk.
PR00F. õ is a continuous homomorphism. Clearly
õ(r/s.x) = r/sõ(x) all F,s e v, x e Rr, ând by continuity
õ(t.x) = t,õ(x) , t efR, X .lRr. Thus õ is a vector space
homomorphìsm; since õ is jnjective, the range õ(fRn) is
n-dimensional and so is all of tRn. By 1.2 and 1.3, u is
type I . The r'est fol I ows from 3 .9 .
Recall that every localìy compact group G can be
written as a productfRn x H, where H is a locally compact
abelian group containing a compact open subgroup. The integer
n is an 'invariant of G (see I .I,2) ,
PROP0SITI0N 3.12. Let u be a ipn -degenerate and type I
mtLtípLíeo on lRfr x H uhev,e H is a LocaLT'U conr¡tact abeLían
g?ole, then u is simiLar to 'cç uhev'e r is a cross rm,fttípLien
on [Rfr anÅ. o is a non-d.egenenate and. type I nwLtipLíer on H.
PR00F. By Corollary 2.5, R is an isotropic subgroup. It
is a topological direct summand, thus by Theorem 3.9 and
induction, o is of the desired form.
55.
4. The connec'led com Donent of G as a dìrect summa nd
l^le saw in Section 3 that for our purposes, we can
disregard direct factors offR in G, so we can and do assume
that G has a compact open subgroup H. For the whole of this
section, we suppose that G adm'its a non-degenerate type I
multjplier o, Let C be the connected component of G, then C
must be a closed subgroup of H and is thus compact. We point
out at once that C is divìsible (I.1.3.(iii)); divisible sub-
groups of discrete abel ian groups are always direct summands
,
(I.1.4.(i)); furthermore, if C is a direct product of circle
groups then C is also a topoìogical direct summand
([18, 25.31a] ). Converse'ly, Fuìp and Griffith [13, Coroì lary
3.21 have shown that a connected group C which is a
topo'logical direct sum in every locally compact abelian group
(with compact open subgroup) in which jt occurs as the
connected component of the identjty must necessarily be a
product of circle groups. In particular, there exists a
localìy compact group G v¡hose identjty component is not a
topolog'ical direct summand of G. It is not clear if the same
is true jf G is self-dual , Below vre prov'ide some conditions
on G which ensures L,hat the connected component of the
identity is a topolog'ical direct summand"
LEMMA 4.1 . Suppose ue haue abeLian groLLps H g K s G
hn topoLogy) such thnt H is díuisibLe anÅ. K/H is a direct
sutftnanÃ. of G/H, thsn K ís q, diTect sutnnand. of G.
56
PR00F. From the hypothesi s and I .1 .4. (i ) , there exist
subgroups F and M in G such that K = H x F and 6 = fl x M.
Furthernrore, M can be chosen to contain F. Since F is a
direct summand of M, there exists a subgroup L such that
M = L x F. It follows that G = H x F x L and K is a direct
summand of G.
Let G be a locally compact group with compact open
subgroup H, connected component C and non-degenerate type I
multip'lier ur. since Cur is open in/(r.5 (ii)), Cu¡ is a
direct sunrnand of G in the discrete sense if and only'if it,
i s so i n the topoì og'i cal sense .
LËMMA 4 2 , Let G" u and C be as aboue, Then C is a
topologicaL dínect swnnand. of G if anrl onLy if Cu is a
dit ect swnnøtd of G.
PR00L. Suppose G is topoìogically isomorphìc to C x K,
then G'^ = c^ ' K^ and (1.5 (i)), õ(cr) = A[G^,c] = K^ is a
topoì og'ical direct summand. The "onìy i f " part now fol I ows
from the fact that õ is a topoìogical isomorphism. The "if"part is similar.
PROPOSITI0N 4.3. Let G, u and C be as aboue. If any of
the foLLouíng properties ave satisfied,
ft)(ü)
ft,ii)
G/C is tov,síon free,
CulC is díuisíbLe,
Cu/C has botmded order',
57.
ftu)
(u)
Cu/C is cornpact oz, C is open,
là is totsion free,
Curr/C is of erponent p lp - a fiæed prine), or
Curr/C ís eonrpact oz, C is open,
then C is a topoLogicaL di:rect swmnand. of G.
PR00F. (i). See Fulp [14, Coro'l1ary 9].
(ii ). Because Cur/C is divisibl e, it must be a d'irect
summand of GrC (I.1.4.(i)). Now C is divisible, so by 4.1,
Co is a direct summand of G and so by 4.2, C is a topological
direct summand of G. (An alternative proof can be constructed
usjns (i).)(iii ¡. Since (G/C)/(Co/C) = G/Cur which 'is topolog'icallv
isomorphic to C^, js a torsion free group (I.1.3. (i ij ) ), by
I.1.4.(ii), Co/C is a direct summand of G/C, Now proceed as
. ,..1rn (1rJ.
(iv). Since Co/C is self dual (1.5 (ìii)), Cr,r/C compact
or discrete 'impì'ies Co/C is f inite and thus of bounded order.
Now use (jii).(v). Since C is divìsjble it is a direct summand, thus
G torsion free implies G/C is torsion free. Now use (i).
THEOREM 4.4. Let G be a LoeaL Ly eonpaet abeLian group
and. u a non-degenerate and. tgpe I rmútipLier', then G is of the
fonm lRn x K, uhexe K eontains a cornpact open subgroup H
(I.1.2). Let C dernte the aormeeted component of K and u,
the xestriction of u to K. tf eíther
(+)
øi,)
5B
then G has a tnanímaL ísobz'opic subgnoup uhich is a topoLogicaL
dLv,ect summand. ConseqtentLy the stv'uchune of G and u ís
cornpLeteLy determined by Theoretn 3.7.
PR00F. By Proposition 3.I2, we may assume that n = 0.
Since C is compact and connected, Propositiorr 2.6 shows that.
C is isotropìc, thus by 3.9 and 4.3, we may assume G = Curr/C.
(i). As in 4.3 (ii'i), G rnust be fjn'ite. The result now
follows from Corolìary 3.i0. (ij). By Theorem 3.1, G has a
maximal isotrop'ic compact open subgroup L. Since G is a
vector space over the fìeld of p-e'lements and L is a subspace,
it admits a complement, but L is open, thus it is a topcl'logical
direct summand.
5. Local direct ducts and di v'i si bl e rou S
Given a locally compact abelian group G with non-
degenerate and type I rnultipìier o, observe that the group
Cu/C, where C is the connected componettt of the identity, is
a total]y disconnected group ([18, i.3]). l^le can deal v¡ith
some total'ly disconnected groups by decompos'ing them as local
direct products of topological p-groups. A definition of
local products is as follows. Let Gr''i e I be a collectjon
of locally compact abel'ian groups each with a compact open
subgroup Hr. The local direct product of the Gt with
respect to the compact open subgroups H.,, denoted by
LPi.I(Gi,Hi) is the subgroup of the full (d'iscrete) direct
product ni.IGi defined bY
59
{(Sr) e IIG., t 9i . Hi for all but finitely many i e I}
and is topolog'ized so that the subgrouP tri.IHj (with compact
cartesian product topo'logy) becomes a compa.ct open subgroup
of LP.,.t(Gi,Hj). (see also [18, 6.16])"
Abel ian topolog'ica'l p-groups (or simply p-groups, P ô
fired prime) and local direct product decompositjons of
topo'log.icaì groups into p-groups are dealt with jn deta'i'l by
Braconnier tsl and to some extent by Vilenkin [50]. The
following is a brjef exposition of tlre facts we need later.
An element x of an abeljan topoìogical group js called
p-prìmary if
xpn
lim _1
n-þ
(Brar,onnier's alternative equivalent definition states that x
is p-prj¡¡ary 'if the homomorphjsm Z + G : n -'xn extends to a
conti nuous homomorph'ism ao * G ' ) An abel ì an topo'logi cal group
consisting entjrely of p-p¡imary elements is called a p-group.
PROPOSITION 5.1, (tsl ) . A disere'be abeLían gt'oup G is a
p-group if and onLy if eaeh eLement of G has ordeT a pouer of
p
PR00F. The result follows immediately from the
defi ni ti on.
60
LEMMA 5.2, (t5l). A eonrpact abeLian group G is a p-grloup
if anå onLy íf G^ is a, P-Woup.
PR00F. Suppose G 'is compact. Let x e G^. For each
x e G, by the continujty of x, xPr * f imp1i., xpnl*¡ =
x(xP ) * 1, hence x(G) is a p-group; it is also a compact
subgroup of'lf and hence fìnite. In particular xpn = 1 some
n e 2.. Conversely suppose G is a discrete p-group, let U be
an open set in G^. By the definition of the topoiogy on G^,
there exists a finite set, - {xr, ..., *k} g G and e > 0
such that
{xeG^: lx(x)-1<eallxe FlsU.
n
Let n.1
be the smallest integer such that x
], then xPn
n = max{n
that xPn
=lallxeF,thusX= 1 and let
e U. It fol I ows
p
i1
np
t+LaSn-àæ.
PROPOSITION 5.3 ( t5l ) . Let H be a cLosed subgz'oup of
the p-groLtp G, then H an"d G/H are aLso p-gro'upri. ConuerseLy'
if H ís an open p-group in G and. G/H is a, p-group" then G ís
a, p-gToup.
PR09I. The first assertion about closed subgroups 'is
obvious. Let p : G + G/H be the canonical homomorph'ism. a
is continuous, so for all X e G, we infer from the'
djscreteness of G/H and Proposition 5.1 that ^Pn.
H for some
rì, so (*Pn)o* = *o**n * 1 ôS Ír + ø.
61.
THEOREM 5 4 ( I5l ) . Suppose the LocaLLy eompact abeLùcn
groLtp G has a contpact open subgz,oup H" then G ís a p-group if
aná. onLy ¿f G^ is a p-gtoup.
PR00F. Since G^/AIG^,H] is topo'logical'ly'isomorph'ic to
H^ which is a p-group (Lemma 5.2) and A[G",H] is isomorphic
to the dual of G/H which is also a p-group, the result
follows from Proposit,ion 5.3.
LEMMA 5.5. Let G be a LoeaL Ly eompaet abeLian gvoLLp. If
x e G is p-pr,ìmaty anÅ q-pnimarg for Luo distinct prinLes p and.
g, thenx=!.
PB00.F_. Let x . G^, then by the continuity of x,
x(*)Pn = ¡1¡pn) * 1, thus x(x) is a p-adic rational, that is
x(x¡Pn = l some n. similarìy, x(^)9* = l some m, thus
x(x) = 1. This is true for all x e G^, so x = 1.
PROPOSITION 5.6, (Braconnier [5, page 48]). Let G be a
conrpact totaLLy discormected group. Dernte by P the set of
rpimes {2, 3, ...}, then there eæist eLosed subgz'oups lr1,
(p . P) of G sueh that H, is a p-gnoup and
p€
Euz,therrnore, an eLqnent x e G beLongs to H^ for a prime p ifP
and. onLy if x ís p-prinarg.
PHpG=I
PROOF. By I.1.3. (ii ) G^ is a tors'ion group, thus by
62.
{[18, A3], ø is a weak direct product of p-Eroups. Now apply
duality and Lemma 5.2. For the second part, let vO denote the
canonicaì projection map n * tp and let x be q-primary (for a
fixed prime q), then ro(x) is both p-primary and q-primary, so
by Lemma 5.5, e,(x) = l whenever p I g. It follows that
x e HO.
THEOREM 5.7 (Braconnier [5, page 49] and Vilenkjn [50,
page 86] ). Suppose G is a LocaLLy contpact abelian gz'oup uith
cornpact open subgz'oLe H st'æh that both G anã. G^ are totaLly
discorm,eeted. Denote by P the set of pr'ímes {2r 3, ...},
then thez,e eælst subgz'oups Gr, Hp of G such that G, ís a
P-gTouP: H ís a eornpact open subgroup of G, anÅp
G = LPo.p(Gp'Hp) l{= np.P Hp
ELements of
that xln *GO are uniqueLy eharacterized as those x e G such
1.
PR00l. For x e G\H, define Hx to be the grorip generated
by {x} u H. The quotient G/H i's torsìon because-it is
topo'logically 'isomorphic wìth A[G^,H] ^ - the dual of a
compact totally disconnected group which must be torsion
(I.1.3.(ii)); thus Hx, the union of a finite nunrber of compact
cosets, is compact. Suppcse we have the primary
decomposi ti ons
H=I ^Hp€r pand Hx = Iro.p Hf
63.
which we may obtain by appealing to Propos'ition 5.6' then by
the latter part of that proposit'ion the group Gp = rr,.G Hf ls
a p-group and GO n H = HO. Since GpH/H is discrete, bY
[18, 5.32] nO/tO i s d'i screte and HO 'is open i n GO. Suppose
plqandx eGO nGO, thenLemma 5.5 shows thatx= 1. Next'
suppose x e G\H. Since Hx/H = np.p Hfluo is a finite product
(recaìl the index of H in Hx is finite), x is a finite sum
i X h+p€P p
with xO e Go and h e l-1. This says n =_loo.o(Gp'Hp). Finally,
the proof thad x e GO'if and only if xP * 1 is identical to
the proof of the corresponding fact in Proposition 5.6.
Having set up the necessary machinery, bJe proceed via a
few preì im'inary results to the majn theorem.
PROPOSIIION 5.8. Suppose G is totaLLy dísconnected anÅ
admíts a non-degenez'ate and type I rm,fltipLiev' u. Choose a
conrpact open manimaL isotfopic subgroup of G (3.I) ard Let
G = LPO.p(nO,tO) be a pninaxg decornpcsítion as in fheorem 5.7.
Ihen each G
rruLtipLierp , P € P adnits a rnn-degenerate and. type I
ÛJp
such that
fo? aLL (S.i ), (hi ) . G uhere 'the p?odtrct aboue is aLuays finí|;e.
a( (si ), (hi ) ) = rp.e õo (so, ho)
PROOF. Fix a prime q. Observe that G^ = LPP.P
64.
(a;, o tG;,Hpl ) i s precise'ly the primary decomposition of G "
and the image o(Go) consists of those ñ(x), x e G such that
õ1x¡en -> 1 as Íì à ær thus o'(Go) = G;. An.appeal to
Corol ì ary 3 .6 yi el ds the des i red mul ti P1 ì er ur' . It remai ns
to verify the above product formula. Let (St),(ht) . G and
let Q be a fin'ite subset of P such that 9O e ll', hO e HO for
all p ìn the complement of Q. Then G = L x K where
L = np.Q GO and K = LPp.pfQ(Gp,HO), and we can argue as above
to obta'in non-degenerate and type I multjpliers o and t on L
and K respective'ly such that t
Ip
and
However ;( (gi ), (hi ) ) = 1 s'ince gO
õ = õ and the result follows.
HO a1ì p c P\Q, thus
LEMMA 5 9. Let be a rpn'degenerate rm,LLtipLíez' on
nf.
ü).Q 'po 61
h €p
t1 - then
(i)
(ií,)
n = 2k, G is topoLogicaLLy isomorphie Lo K x K^
u'here K is a cLosed subgt'oup of G of tiæ form n\p
ctnð. u is cv'oss rm'tLtipLíer on K * K^. (rn
pæticuLaz', u is type I. )
If H is a conrpact open maæímaL isot:r'opíe
subgz'otp of G (anã sueh a gnoup aLuays eæists bg
Propos'ition 3.1.), then H is of the forin L x L'
for some eompact subgnoups L and l-' of K anã K^
respeetíueLy.
65.
!B!9I_. (i). Since G is a finite dimensional vector space
over the p-adic field n,,, the proof of Corollary 3.11 app'lies
to (i ).(ii). Let p (respectively rf) be the (continuous)
projection map G -' K (respect'ive'ly G * K^). First we show
that ç(H) s H and V(H) s H. Since ç(H) n V(H) = {1}' the
direct sum L = p(H) * {,(H) is a compact totally disconnected
group, so'its dual must be torsion (I.1.3.(ii)). Now õ is
injective and õ(t-1 = A[G^,Lo] ìs topologically isor¡orphic to
G/Lo and L^ (1.5), thus õ restricted to K must be trivial and
L isotropic. In particul ar õ(k,ø (h)) = õ(ç (k),,l, (h) ) ,
õ(,p(f<),ç(h)) j f for all h,k e H, thus ç(h) e Ht¡ = H whenever
h e H" Similarly with rl. It is now easy to check that
H=ç(H) "ü(H).
LTMMA 5 10 . Suppose ue. haÞe a separabLe LoeaLLy cornpact
abeLian p-grory u-hich is diuisibLe, If w is a non-
degenerate anÅ. type I rmt-LbípLier on G, then G ís of the form
nl far some integer n.p"
PR00F. Since G is self-dual, it must be torsion free
I.1.3.(i), thus by Rajagopa'lan and Soundararajan [36, Lemma 12],
G is a local product
G = LP.,.l(CIp,op)i
For G to be d'ivisibìe however, the index set I must be finite,
thus G i s of the form si!.p
66.
THEOREM 5.11. Let G be a Local Ly cornpaet abeLian group
uíth eornpact open subgroup anã. rÌþn-degenerate and type I
nwLtipLier u. Let C denote the eornponent of the i.dentity in
G. Sutppose Cu/C is sepaz,abLe ard. díuisi,bLe" then G is of the
form H x H^ aná u is simíLa:r to a, cross rruLtípLier.
PR00F. According to 2.6, 3.9 and 4.3 (ii), we can assume
without loss of generality that G = Cur/C. Let H be a compact
open maximal isotropic subgroup of G. Use 5.7 to write
ll= II ^HP€l' P
Now by Proposition 5.8, for each p € P, there exists a non-
degenerate and type I multiPlier ur' Gp
on such that
'G = LPPTP (no'to ) '
OJ np.P 'p
Observe that the hypotheses of Lemma 5.10 apply to each GO,
thus by Lemma 5.9, we can write
G = K * K'. l-lp p p- pL'
p
and K' = LPp-p(Kå,1.å),
=l_ xp )
where KO, Kå (respective'lV LO and Li) are cjosed subgroups of
nO (respect'ively HO) and KO 'is a max'imal isotropic subgroup
i n GO (rel ati ve to the mul ti p'l 'ier ,O ) . Let
K = LPpep(Kp'Lp)
thenG=KxK'and
67.
Kr¡ = {(s.¡). G: r((ht),(sr)) = l all (ht) e K}
= {(9i). G,np.p'p(hp,9o) = 1 all (hi) E K},
but KO is maximalty isotrop'ic in GO
Now apply Theorem 3.7.
for all p . P, thus l('¡ = K.
COROLLARY 5 T2 . Let G be a sepatabLe LocaLLy cornpact
abeLim groL{p. If G is ditsisibLe and adrnits a non-degenerate
anå. type I rruLtipLier u, then there eæists a cLosed subgroup
H of G and. a topohogieaL isomorphism front G to H * H^, such
that the image, of u wtder thís isomorphism ís simiLav' to a
rm,útipLier of the forTn
,r((x,r)(y,x)) = r(y) ,
(x,r),(y,x) e H x H^.
PR00F. By I.1.2 and 3.I2, we can assume without loss of
general'ity that G has a compact open subgroup. Since G is
self dual by I.1.3.(j) G is torsion free, thus by 4.3 (v)'
2.6 and 3.9, it must be of the form C x C^ x Co/C. Hence
C,¡/C is separable and divisible and the result follows from
Theorem 5.11.
Finally, we remark that Theorem 3.9, Proposition 2.6
together with [1, Theorem 2] show that the problem of the
structure of ur-type I locally compact abelian groups has been
reduced to that of 'res'idual' groups (see [1, page 597]).
68.
CHAPTER IIIl
MULTIPLIER REPRESENTATIONS OF DISCRETE GROUPS
Throughout this chapter, we fix a cj'iscrete group G and a
normalized multiplier ûr on G. Recall that. L2(G) denotes the
Hilbert space of square summable complex valued functions on G
w'ith scalar product ( , ) , B(12(G)) the space of bounded
l'inear operators on L2(G) and U(12(G)) the subspace of unitary
operators in B(12(c)).
We denote by r (respectìvely p) tfre ríght (respect'ive1y
I ef t) t, (respect j vely ur- I ) - representat'ion of G gi ven by
p,À : G + U(12(G)), where
(r(x)t) (g) = ,(s,x)f(gx)
(p(x)t) (g) = ,(x'1,g)t(x-19),
f e L (G),X,9 e G. To make sense of this definition, we observe
that
'¡(x) r'¡(v)!(s)
= J::ïï;;ïl:îl-,,' r( g,xy),¡(x,y) f( gxy)
= ,(x,y)[r(xy)f] (g),
IThe results contained in this chapter have appeared in l2}l.
ÂCt
all f e L (G), 9 € G, hence r(x)r(y) = ,(x,y)r(xy), and sirnilarìy
,(x,y)p(x)p(y) = p(xy).
Let V(G,o) (respectively V'(G,r)) denote the Von Neumann
algebra generated by p (respectively r), that js the weak closure
in B(12(c)) of the complex l'inear span of {p 9) : g e G}
(respectìvely {r(g) : g e G}). (See Chapter I, Section 2 and
Section 7. )
The a'im of th'is chapter is to investigate how various
statements about the maximal central type I proiectìon 'in
V(G,6) are reflected in the structure of the group G and the
multip'l'ier ur. Th'is leads to a characterization of ur-type I
discrete groups. The corresponding probleln for ordinary
representat'ions, that is if we assume o to be triv'ia'l , has
bee¡i successf ul ly deaì t with in Thoma [49] , Kani ut'h i22l ,
Smi th [45] , Formanek t10l and Schl icht'ing t4l I .
Our methods resemble more closely those found in Smith;
these are of a more elementary and algebra'ic nature than Thoma's
and Kan'iuth's "E(G)-Methoden" .
t. A representation of elements 'in V(G,o)
To begin, we construct a way to represent elements of t
V(G,r,r) as sequences in l-2(e ). For each x e G, denote by r* the
characteristic function of ixlc G. The set {eX
: x € G] is an
70
orthonornral basis for L2(G). We have
p(x)ç I = ).(x) -'r..e x
For a e B(L (G)), let a*,y = ( a(rr),r*) and a a.Xrê-X
x,y e G. The numbers ax,X € G are calTed the coefficients of
a.
LEMMA 1.1. (Kleppner 124, Lemma 1l). Let G be a disa'ete
gvol&, o a nonnaLized rm,¿LtipLier on G, ), the z'ight regular
u-vepresentatíon of G anã p the Left reguLar u-r-tepresentation
of G. Suppose â e B(12(G)),
(i)
(¿i)
ar(x) = r(x) a aLL X e G if and onLy íf
u*,y = o(x,y-i)axy- y, aLL x,y q G.
ap(x) = p(x) a aLL x e G if and onLy if
ur,y = r(y-t,x)ay- t* aLL X,! e G.
PROOF. Suppose a¡.(x) = r(x)a all x e G, then
ur,y = (a(vr) ,la*)
= ( a(r(y-l )r.) ,v*l
= ( a(r.) ,r(y)r(x- i )vul
= r¡(x,y-t)u*y-r,
all x,y e G. Converse'ly ìf a e B(12(G)) satisf ies
7i.
a = o(x,y- I )a 1al'l X,V € G, thenxyX,Y
< r(y)a(ç.),px) - 1 a(ve),r(y-r)r(x-1)pe)
= o(x,y)a a IX'Y xy-
- , u(rv-r),v*)
= ( aÀ(y)(r.) ,r*>
for all x,y e G, hence aL(y)(e.) = r(y)a(pe) aìl y e G. Itfollovrs that ar(y)(v¡) = r(y,h-r)ar(yh-t)(ee) = r(y)r(¡-t)a(ru) =
r(y)a(^f,), and hence that ¡.(y)a = aÀ(y) a1l y e G. The second
statement follciws from a sinlilar calculation.
PROPOSITION 1.2. Let G,u,). and p be as in Lemma 1.1,
Ifae B(12(c)), then
(i) ff r(x)a = ar(x) aLL x e G, then
u = In.G ano(s) ' and
(i¿) tf p(x)a = ap(x) aLL x e G, then
¿ = Ig.G agr(g-l),
uhey,e the suntnation is to be interpreted in ihe sense of ueak
operatoz, conuergence in B(12(G) ) .
PROOF. Suppose ¡,(x)a = aÀ(x) al I x e G, then
72
a(ey) = Ir.G u*,yt,
I*.e ,(X ,Y-l ) axy- t
Ig.6 '(9v,v-t)untn,
çX
Ig.6 r(v-I,9-r)ano(gY) ( ee)
= Is.G ano (s) ( tv) '
Thus the operators a and lg.e urO(S) agree on an orthonormal
basis in L2(G). It follows that tkiey are equal. This proves
part (i). Part (ii) ìs proved s'imilar'ly.
PROPOSITION 1 3. (K'leppner [24, Theorem 1] ) . Let
G,û,,p ond. x be as zln Lemma 1.1. Let a è B(L2(G)), then
(ì.) r(x)a = al(x) aLL X e G if and onLy íf
a e V(G,ur)
(ii) p(x)a = ap(x) aLL x e G if arui onLy if
â e V'(G,o)
(iii) Íhe eowmttart't Y(G,o) ' of Y(G,ur) is equaL to
V'(G,o) , Hence V and V' haue a cotftnon eentre
V N V"
PROOF. (i) Since
73
p(x)l(v)f(g) = o(x-r,g)r(g)f(x-19)
= r(x-I,g)o(¡-lg,y)f(x-lgy)
= r(x-r,gy)r( g,y) f(x-1gy)
= ,(g,y)p(x)f(gv)
= r(y)p(x)f(g)
alì X,Y,g e G,f u Lz(G) , we see that p and À commute' It
follows that each I(x) (x e G) commutes wìth all finite
conrplex linear combinations of the operators tp(g) : g e G].
But these linear combinations form a dense subspace o'f V(9,o)i
furthermore, multiplication is weakly cont'inuous in v(G,o),
consequently any a in V(G,ur) commutes w'ith each r(x) (x u G) .
Conversely, 'if ¡.(x)a = ar(x) , then by 1.2(ì)' a be'longs
to the von Neumann algebra generated by the operators p(g)'
g e G.
(ii) The proof of th'is part is similar to the a'bove'
(iii) If a e V(G,o)', then in particular ap(g) = p(g)a
aìl g. G and thus by (ij), a e V'(G,t). The remaining assert'ion
folIows from the definjtion of the centre (see defin'ition the
folìowing I.2.1).
We arra.nge some of thjs informat'ion'into a singìe statement.
THE9REM 1.4, Let G be a díscrete {roup, û a normaLised
m.tLtipliey. on G and. X the r,íght reguLar u-repÍ'esetztal;Ì'on of G'
74.
(i)
(ii)
sttppose ô e B(12(G)), then a e V(G ,u) í¡ ard.
only íf ar(g) = r(9)a aLL g < G. This oecuns if
and onLy if ax,y = ,(X,V-l)axy- ts aLL X,y e G.
rf a e V(G,ur) , then u =Lue ano(s) in the sense
of ueak operator conüergence. Ihis deconrpositíon
is wtique" that ís if a = IgeG a'no(9), then
u'g a all g u Gg
PR00F. Except for the utriqueness, all these assertions
follow immediately from 1.1, 1.2 and 1.3. Suppose we have
¿ = Ig.G a'no(ô), then a(,r.) = Ig.G u'grg and a* = ( a(r.),r*)
In.
,(x,y)uy-r*y = urr(Y,Y-1xY)
a all x e G.X
Since we are interested in central projectìons in V(G,o),
we need a method of determining when an element a e V(G,r)
be]ongs to the centre of V(G,o) 'in terms of the coefficients
ug (g . G) of a.
PROPOSITION 1.5. (KIep pner [24, page 557]). Let
Grtr tr and. p be as in Lemma 1.1. -If a e V(G'u) , then a ís in f;he
centv,e CV(Grr¡) = {b e V(G,o) : cb = bc aLL c e V(G,r)} íf and
onLy if
G u'g(rg,r*)
aLL x,y e G.
t
75.
PR0O.F. Consider the set of equivalent statements i
a e CV: ap(x) = p(x)a all x e G;
Ig.G uso(g)p(x) = IguG ano(x)p(s) ' all x e G;
I*.G uyr-rp(y)ur(¡-1,XY-r¡ = I*.G ax-rye(y)t¡(y-rx,X-1)'
all y e G; ar*-rr(V,x-1) = ô*-rn(x-I,v) all
r(x,y)uy-r*y = a*trr(y,y-Ixy) all X,! e G.
was obtained by a change o'F variable.
xrY e G;
The last equìvalence
g
PROPOSITI()N 1 (Kleppner [24, Theorem 2 and Lemma 6]).6
Ipt G be a discrete groutp and u a norrnaLized rm,¿L-bípLíer.
(i) Fon a e V(G,r), the mq I +â is ín L2(G).
(ii)
(io)
PROOF.
(ab)x = Ig.G uyby-r*r(y-I,x)
= Ir.G a*r-tbro(x,z-l) .
(äi) ( an,b/ = (ab*)", and (o*)g = ug-r.
Ihe map þ z V(G,o) * C : a + auis a finite
faithiuL nov,mal tv'aee on \,!(G,t)+ ' l;he set
of positi.ue elements in V(G,t,r) .
(ii), we have
(i) follows from (ììì) by letting a = b. For
76
(ab)x = (ab)*,. = I..G u*,=br,"
lrr; axr-rbzr(xrz-l).
The other formula follows by a change of va¡iable. The fìrst
part of (.iij) follows from (ij) and the second part from the
equation (ao)g = (a*(r.),r/ = (9.,a9/ = (avn,t.) = u.,g =
an-r:ur(e,g-1) = ug-, . The proof of (iv)is as follows:
Clearly O js finite and normal. If a > 0, then a = bb*
for some b e V(G,r); nor,r o = O(a) = â. = (bb"). = (llbnll¡z-(C))'
if and only if,a = 0. Thus 6 is faithful. For the invariance
we have O(ab) = (ab). =(ag,(b*)g) = (bg,(u*)g) = (bq)u = 0(ba).
Thus by I.3,2, V(G,o) js a finite votl Neunlann algebra (tfris
is not the case when G is non-discrete, see IV.4.1), thus
V(G,u:) has no type III, type II- attci type I- d'irect summands.
By I.3.3 V(G,r) is the d'irect sum of a type It von Neumann
algebra and a type II1 Von Neumann algebra. Indeed by I.4.1
there exi st central orthogonal proiect'i ons e,ê 1 rê2, . . . i n
V(G,o) wi th (li=, en) + e = I (ttre identity operator) and
such that enV(G,ur) ìs type In and eV(g,o) 'is type II1.
Let G be a djscrete groupr ür â normalìzed multìp1ier on
G and H a subgroup of G. For each t e V(G,r), let supp(a)
denote the support of a defined by supp(a) = {g € G: an I 0}.
Denote by V(H,o) the von Neumann aìgebra V(H,olHrH)'where
'lHrH is the restriction of o to H.
THEOREM 1.7. If ll is a subgz,oup of G, the set
77.
( = ia. V(G,r) : supp(a) c H\ is aueak operatov'closed
*-subaLgebz,a of v(G,r¡) (artd hence ís a uon Newnann aLgebr'ù, dnã'
there eæists a (normaL) *-isomorphism ' : V(H'o) * K'
Moleot)er', the coefficients of a e V(H,r.¡r) are presev'oed uttdert '
PR00F. Usjng the representation a = Ig.H ato(S) of
elements in K, we easily see that K'is a *-subalgebra of
V(G,o). To see that K is weakly cìosed' let {ai : i e I} c K
be a net converging weakly to an element a e V(G,t,l)' In
iparticular (ar(ç.),erl *(a(e.),er); but for x { H,
< ui (r.),rx) = 0 all i, hence u* = ( a(.r.),r*) 'is also equa'l t'o
0. ,
Let S be a set of coset reptesentatives modulo H'
A *-'isomorphism ' : V(H,6) * K can be defineci as follows
a'f(x) =
f e Lz(G); where g € Sn h e H are two elements such that X = hg'
and fn denotes the function h ' f(hg)r(h,g), h e H'
Note that
o(g,h)fgx(h) = f(hgx),¡(h'gx),¡(g,x)
= f(ghx)u,(hg,x),¡(h'g)
= f*(rrg)u,(h,g)
= r(g) fx(h) ,
that is r,r(g,h)f gx = f(g)fx. Hence ìf hg=h'g', h,h'e H'
7B
9,g' e G, then
(À(h- rn' ) (afû) r9 )ur
, )l (h)
This shows that the definjtion of ' does not depend on the
choice of the transversal S. Moreover, we have
lla'fll L2(c) = IgeS,h.Hl (arn) (h) l2
Isus(llafnll ¡z 1¡¡)2
ll all z [nur,h.Hl fg(h) l2
L2(G) )2
ô e V(H,r), f e L2(G). Hence a'e B(12(G)) for a'll â e V(H,r).
The next step'is to show that the map ' indeed maps into K. We
do this by caìculating coefficients. Let h e H,9 € S and
X,y€Gbesuchthatx=hg,
= il all2 (ll f ll
79.
t(çv g
Hence,
Suppose a' =
so jn partjcular,
Yg-tr¡(g,h-1),0
y-lg e l{,otherl',¡i se.
) l(h){
a'xy = < a'(çn) 'çl = a'(çr) (x)
In particular u'* = ô,at
v g
0 otherwise
xy-rür(x,y-I) if xy-t e H,
otherw'i se.
x e H. Also a'
I (h)/r(h 'g)
V(H,al) , then a'*
H which implies
a[(ç
la(rnn 1)l (rr),(g'v-r)/"1rr,g¡ if vg-i ' H'
0 otherwise
uhgy- rr(h, gy- I ),¡( g,Y- I )/r( h, g ) ifv Hg HX,€
a -rt,r(x,y-l),Xrê X'J xy
0 otherwi se
hence a' e K for all a e V(Ìl,ur). S'ince multipì'ication and
involution in V(G,o) and K are defined wholly in terms of the coefficients
a.., respectively ô'x (which are preserved undeF'), we see that I
X
is indeed a *-homorphism from V(H,r) into K.
0 for SOme a €
all X e
0
a=0allXeG,
, thus theâ*=o
80
map' is injective. To show that' is onto, we construct a right
inverse " for', that is amap ": K+V(H,to) such thatro " - idK.
If f e L2(H), we denote by f the function
f(g) e e H,f (g) =
0 otherwi se.
Defjne " by a"f = a(f )lH, f e L2(H), a e K. Since
lla¡rfllL2(H) = llaf l¡ll¡z(ti).
< llafl llL2(G)
= ll all ll fllL2_ (t-t )
as well as a"*,y = < a[(çr)" l,r*) = âx,y = o(x,y-t)u*y-r =
r¡(x,y-l) â"xy-1 for X,y € H, we see that " maps into V(H'r).
That I o " = id* foì]ow by comparing coefficients. The
asserted nornral 'ity foì I ows f rom | .2.2.
In view of this Theorem and because we are irrterested
only in the type structure of V(G,r¡), we rvill identify V(H,o)
with the weakly operator closed *-subalgebra {a e V(G,r) : supp(a) c H}
whenever H is a subgroup of G.
2. Di screte f i n'ite cl ass qroups
Recall from Chapter I Section 7 that a (discrete) group G
81.
is called a FC group (finìte class) if the set {g-1x9 : g e G}
is finjte for each x e G; and for an arbitrary (d'iscrete group)
G, GFC denotesthe subgroup of ail elements x in G such that
{g-lxg: g € G}'is a finite set. Clearly GfC ís a FC group.
For subgroups H and K, let [H,K] be the subgroup of G generated
by {n-tk-thk : h e l-1, k e K}. The group [H,H] 'is denoted by
H'. Itis a non-trivial theorem of Neumann ( ¡aZ, Theorem 5.11)
that for finite class groups G, the commutator G' is locally
finite (tnat is, each fin'itely generated subgroup of G' is finite).
hle need this result in the form of the following'lemma.
LEMMA 2.1 (Neumann [33, Lemma 4.1]). Let G be an FC g?ottp,-----.-.--------
H a subgroup of finite índeæ in G such that lH'I a -, then
lc'l < æ.
PB00r_. Let S be a set of coset representatives modulo ].l.
If s,t e S and g,h e H, then by using the equalities [a'bc]
¡a,cl ¡a,blc, [ab,c] = [â,c1b¡b,c1 (where [a,b] denotes a-lb-1ab
and ab denotes U-Iab), we obtain
[gs,ht] = [9,t]slg,hlttIs,t] ¡s,hlt ,
so we see that G' ìs generated by elements of the forrn [s'h],[s,t],
[g,h] and the'ir coniugates. The commutator H'is finÌte by
hypothesis, so there are only finitely many eìenlents of the form
[g,h]. Since S js a finite set, there are on'ly finitely many
elements of the form [s,t]. To see that there are on'ly fin'itely
many eì ernents of the f orm [s , h] , note that [s ,h.l = s- lsh , but
G is a FC group, so we have the desired property. h'e have shown
8?.
that G' is finitely generated. The result now follows because
G' is locally finite.
3. The o-finite class qroup and the ar-centre of G
Let G be a dìscrete group and t¡ a normalized multiplìer on
G. The concepts of finite class group, centre and centralizer
of G have thejr natural analogues for the pair (G,o). We
define these and invest'igate some of their properties.
Let l-t be a subgroup of G. For each x e H, the ur-centralizer
of x in H 'is defjned by
Cr,¡(x) = {g e cr(x) : ur(x,g) = o(g,x)},
and the o-centre of H by
Z,(H) = ^*.H Cr,r(x)
= {g € Z(H) : o(x,g) = o(g,x) al I x. e H},
where Cr(x) denotes the (ordinary) centralizer of x in H and Z(H)
the centre of H.
The o-finjte class group of G is defined by
A= an = {x e G : [G : Cu,,6(x)l < æ]
Note that if o is a trivial multìplier, then  = GFc. Because
Cr,nh)c- C(x) each x, we have ¡ 9 GrC. In particular ¡ is a FC
group. The fol I owi ng propos'i ti on wi I I i ust'i fy the abrove def j ni t'ion .
83
PROPOSITION 3.1. Let G be a discrete gt'oups 6 4 norTnalized
ru,LtipLiev'on G and p the Left TeguLaz' ,-r-t'ept'esentation of G.
LetxeG,then
Cr(x) = Cr,dx) = {g e G : p(x)p(g) = p(g)p(x)1.(1'.)
(ií) Cr(x) and A are subgroups of G.
PR00F. (i). Suppose g . Cr(x), that is g e C(x) and
o(x,g) = ,(g,x), then p(g)p(x).(g,x) = p(gx) = p(xg) = p(x)p(g)
o(x,g), so p(x)p(g) = p(g)p(x). Conversely, suppose g e G
satisfies p(xÍp(g) = p(g)p(x), then p(gx)/'(g,x) = p(xg)/,(x,s),
so in particular,
lp(gx)r.J (v) Ip (xg )v*l (v)
û) 9'X r¡(X,g )
for all y. G, where.r" denotes the characteristic function
of {e}. Thus
to(x-tn-1,V),/"(x-tg-tr1 o(g-1x-1,y)r.(g-Ix-Iy)T b-G'gl--
Now if we let y = gx, this expression simplifies to
,(g,x)-l = o(9-1x-1,9X)r.(S-t*-ty)/o(x,g), hence vu(S-tx-ly)
cannot be zero, but the only way th'is call occur is if gx = xg
This shows that g e C(x) and from the above expression,
r(x,g) = r(g,x). This completes the proof of part (i).
(ii). Suppose h,9 € cr(x), then
B4
p(h-1g)p(x) = p(h)-rp(g)p(x)r(rr-t,g¡-t
= p(h)-rp(x)p(g)r(h-t,91-t
= p(x)p(h)-Ip(g)r(h-r,g)-1
= p(x)p(h_ig),
hence h-rg e cr(x) and Cr(x) is a subgroup of G
cr(x-ly) > cr(x) n cr(v), we have
Si nce
tG : cr(x-ry)l < tc : cr(x) n cr(y)l
(-,
whenever X,y € A, hence ¡ is a subgroup of G and thìs proves
(i i ).
The fol'lowìng theorem po'ints out the significance of ¡.
THEOREM 3.2. Let G be a &Lscrete gï,oup, w a. norrnq.Lized
rruLtípLiet,on G anå. n the u-finite eLass gvoup of G. ff V(G,ur)
'h.a.s a non-zero manímaL type I Pæt' thert
(i) [G : A] ( -, and
(ii) l¡'l < @.
The idea of the proof of (i) comes frorn Smjth [44, Theorem 9.4],
and the proof of (ii) 'is s'imilar to that of Snljth [45, Theorem 1].
hle need the folìowing lemmas.
LEMMA 3.3. Let V be a tApe In uon Neumann aLge.bra, then
fon each non-zero írreducibLe z'epresentation n of V, we have
n(\t¡ = Mn(CI) (the space of n x n matvLces oler, E). Hence the
dimension of r equaLs n.
PR00F. Let Srn denote the standard po'lynomial of 2n
variables (see Chapter I Section 4). By i.4.5, S2n(V) = tOÌ
and Srn("(v)) = n(S2n(V) = {0}, Let Hn denote the Hilbert
space of r. It follows from I.5.1 and the jrreducjbility of
n that the weak operator closure of {n(a) " a € V} in B(Hn)
equaìs B(Hn), thus by I.4.6, Srn(B(Hn)) = 0. We conclude
(us'ing i.4.3) tfiat the cljmension of n is not greater than n.
V is an n x n matrix algebr'a over the centre CV of V
(see Chapter I Section 4), that is each a e V can be written
a = Ir<i,j=n oijcij '
where o.ij u CV and c' denotes the n x n matrix whose onìy
non-zero entry is the i,jth entry which consists o'f the
identity operator in CV. Fix j,k e {1, , r}. Sínce
.ij.jkcks = crn, all L < i,i < n, n(c¡¡) = 0 ìmp'lies n(cr.s,) = 0
all 1 < i,j < fl and hence "(V) = {0} which is a contradiction.
Thus n(cr¡) I o att i,k. Now suppose we have complex numbers
ßij, 1 < i,i < h such that Ii¡ß¡jn(cij) = 0, then for any
fixed k,,c. (]. < k,.[ < o),
Ii , jßi j"(CLt Ci.jCuu) = ßkl,n(Cou) = 0,
B6
this implies ßOu = 0; in other words, the tt(c¡u), 1 < k,t < h
form a basís for n(V), thus n(V) = Mn(CI) = B(Hn).
LEMMA 3.4. Let G be a discr.ete groLtp and u anorrnaLízed
rm,útipLier on G. Supposa CV(G,u) denotes the centz,e of V(G,ur) ,
then CV (G,r) c V(^,ûr).
PROOF. Let a e CV(G,o) and X e G such that a* I 0,
then by I,7, it suffices to show that X e a. From 1.5,
u¡(x,y) uy-r"y = a^.(v,y-1xy), thus C(x) = Cr(x), but by 1.6('i),
[G: C(x)] . -, hence X e a.
LEI4MA 3.5. Suppose G is a discrete FC groups a e
namnalízerl rruLtipLier on G, p the Left reguLar u-repv,esentation
of G and H a subgz,óup of G. Suppose there eæists a pz,oiection
e in CV such that both eV and e\f are type In (n a fiæed ínteger) "
uhere V arLd \f denpte 1;Lte uon Neumøn aLçTebras V(G,r,:) and. V(H,ur)
z,espeetiueLy. rf K= {xe G: p(h)p(x) = p(x)p(h)aLL he H}=
^¡,.H Cf,,De(h) ¿s the u-centz,alizer of H ín G, then e(p(k)p(g) - p(g)p(k)j = o
for euery k e K and g e G, Consequent_Ly [K,G] is fi.nite.
PR00t. Let n be a non-zero irreducibìe representation of eV.
Since n(ea) is non-zero for some a e V, n(e)"(ea) = n(ea) I 0, so
n(el) = r(e) I O, but eI e eV", thus r restricted to eV' is non-zero.
By Lemma 3.3, n(eV" ) = Mn(C) = n(eV).
If k e K, then ep (k) central'izes eW , so Tr(ep (k)) central izes
n(ef ) = n(eV), 'in particular n(ep(k)) commutes with n(ep(S)) for
87.
every g e G, that is "[e(p(k)p(g) - p(s)p(f ))l = 0. By I.5.3,
ep(k)p(g) = ep(g)p(k). If we write e = Ig.Geno(o) (1.4), then
f*.G.*o(x)p(g)p(k) = Ix.Ge"o(x)o(k)p(g)
that is
Iyuceyk-rn- r,(s, k) - Ir¡(gk,y- I )p (y) =
IyuGevg-Ik tr(k,g)r(kg,Y- I )-lP (Y).
Equating the coefficients (using the uniqueness of decompositÍon)
and letting x = kg gíves lutnL-ro-rl = leri. By Proposition 1.6(i),
iluglz < -, so the set ikgk-tg-t ' k e K,g € G] must be finite.
Since this set generates [K,G] and is contained jn the ìoca'l]y
finite group G' , 'it fol lows that IK,G] is finite.
PR00F 0F TllE0REl,1 3.2. (i) Let en be a non-zero central
projection in V(G,ur) such that enV(G,o) 'is type In. Since
enV(G,ur) is a matrix algebra over its centre (I.4), it is of
dirnension at nrost n2 over CenV(G,r) = enCV(G,r) g CV(G,t',) ç V(¡)
(Lemma 3.4). Hence if gl, ... , gn2+1 are n2 + 1 elements of G
beì ongi ng to di st'inct cosets of ¡, then there ex'ist el ements
cr, , cn2+t e V(n,tr) such that
n2+ t n2+ tIi=r .i(e,.,0(9i)) = i1=1 cieno(9i) = 0.
with not all (cien)o(Si) = 0, but ciên. V(l,ur), so this cannot
88.
happen since the sum
V(¿,ur)p(gr) * ... o V(a,o)o (9nznr).
is a direct sum. This shows [G : A]'n2 < -.
(ii) Again we use the standard poìynomial in 2n
variables Srn (I.4). Suppose V(G,ur) has a non-zero type I
part, then there exists a central proiection en I 0 such
that enV(G,r) Ís type In. l.le have Srn(.nV(G,r)) = {0},
hence Srn(enV(¡,r)) = {0}. It follows that enV(a,r,r) is
type I.n (I.4.5)
Now let en be a non-zero central proiection in V(¡,o) such
that enV(¡,r) is type In. Using I.4.5'
Srn(.nv(¡,r)) = o, Srn_ (env(l,o)) f o.
Since the polynomial is multilinear and V(¡,r) is generated
by the elements p(g), g e a, there exist 91' , 92n-2- e a
with srn-r(.no(gr), , €np(9rn-r)) * o. Let H be the
normal subgroup of I generated by the eìements 91, , 9zn-2
and the'ir coniugates. Since A is a fin'ite class group, H is
finitely generated. Moreover, Srn(enV(H,r)) = 0 and Srn- (enV(H,,) I 0
so by I.4.5, .nV(H,r) has a nonzero type In direct summand (.,t.o<ø' '*^ù,UeV(H,o) for some e e CenV(H,r). We wish to show that e e CV(¡,r).
Since H is normal in ^,
for any k e a, the automorphism
V(H,o) + V(H,ur) : a + p(k-i)ap(k) leaves the type In summand
ev(H,ur) fixed, that is p(k)-1ep(k)v(tt,r)o 5o by the unìqueness of e'
p(k-l)ep(k) = e. It follows that e e CV(A,o). Sjnce e r ênr
B9
eV(A,o) is clearly also type In, and thus by Lemma 3.5, K'< @'
whereK='{xeA: p(h)p(x) =p(x)p(h) all he H}. SÍnceH js
finitely generated [A : K] < - and thus the result l¡'l '-follows from Lemma 2.1.
Theorem 3.2 provides us with a subgroup of G of finite
index'in G whose comrnutator subgroup is finite provided V(G'r)
has a non-zero type I part. In the next chapter' We construct
such a subgroup for arbitrary local]y compact groups, G using
the resul ts of Tayl or t48l . However, si nce Tayl or's work
depends on the results known for rliscrete groups (and s'ince our
proofs are of an elementary nature), we feel iustified in
i ncl ud'ing them here.
4. The type I part of V(G,t¡)
Let G be a discrete group and o a norrnalized multjplier on
G. Recall that the group extension Go of T by G is the group
whose under'lying set is T x G a.nd with multip'lication
(s,x)(t,y) = (sto(x,y),Xy),(s,x)(t,y) e G'. usually Go is
endowed wi th the wei I topo'logy ( i .6 ) . Holvever, for the re-
mainder of this chapter, we give Go the d'iscrete topology.
l^lhenever H is a subgroup of G, we identify H'in the obvious way
with a subgroup of G0.
PROPOSITION 4.1. Let H be a subgrottp of the discrete
group G and. $ q, TlormcrLízed rmfltip'Lier on G, Adopt the vLotation
of Section 3, then
(i) c6u,(t'x) = (cr(x))'fov'aLL (t,x) € G',
90
(í'¿) [G' : H'] = [G : H] ,
(¿¿i.) (G')rC = 40, and
(iu) (2u,(G))' = z(G').
PR00F. (j ) Suppose (s,y) e C6o(t,x), then (stur(x,y),xy) =
(sto(y,x),yx), thus y . Cr(x) and (t,y) . (Cr(x) )'. Conversely if(t,y) . (cr(x)), then y e cr(x), so (s,x)(t,y) = (t,y)(s,x).
(ii ¡ The map xl-l * (t,x)H' sets up a one-to-one
correspondence between the H cosets in G and the Ho cosets
in G0.
(jii) By definition, each X e G belongs to ¡ if and
only if [G : Cr(x)] . -. Bv (i ) th'is occurs if and only if
[Gt : CGo(t,x)] < * âll (t,x)u nu. But thjs last statement is
equivalent to (t,x) . (G')fc.
(iv) Fix x e G, then (t,x) € (zû,(c))'u if and only
if x e Z(G) and ,(x,y) = ,(y,x) all y e G, that. is if and only
(t,x)(s,y) = (s,y)(t,x) all (s,y) e G0.
Recall that for any discrete group G, Go denotes the
von Neumann kernel of G. (For a. definition see Chapter I
Secti on 7. )
LEMMA 4.2. Let G be a disenete group and u a norrnaLized
rruLtípLíer on G. Suppose that H is a subgnoup of firúte
91.
indeæ in G such 1;Vnt lH' | < - then there eæists a subgnoup K
of G such tVnt tG : K:l < - artÅ. K' = Qo = {L' : [G : L] < æ].
PR00F-. If tG: Ll < - then, since the characters of'
L/L' separate points ([18, 22.17]), we have G" = Lo I L', hence
& cñ {L' : tG : Ll < æ}. Since lH'l < -, ¿nd H/l-L is
maximally a'lmost periodic, by I.7.3 and I.7.10, H/H" is type i
and H has a subgroup. K containing H" of finite index'in H such
that K/H. is abelian. Since tG : Kl < - ênd K' c Ho = è,ihe result follows.
The next iemma is a key lemma.
LEMMA 4.3. If thez,e eæísts a subgroup H of G sucVt that
tG : H] < @ lH'l < - and. tnlHxH is tt'iuíaL fo' oorne n, then
thez,e eæists a subgroup K of G sueh tVøt tG : K] ( -, 6 = K'
and (Gu). = (K')'.
PROOF "
that H' = G.
all X¡y e H.
By Lemma 4 .2, we can as s ume wi thout I oss of general ì ty
For some map y : H + T, u,h(x,y) = y(x)v(y)/v(xv)
An easy calculation shows that
(H')' = {[o(x,y)r(x-I,y-t)r(xy,x-ly-l),xyx-Iy-i] : X,y e H].
Si nce
lt¡(x,y)t¡(x-I,.v-1 )r(xy,x-Iy- t )ln = y(xyx-Iy-t )-1,
we have (H')' < æ. By Lemma 4.2, Go has a subgroup Ntf.rt ,rlua
[G' : M] < @ and M' = (G').. Let L be the image of the proiection
9?.
M + G : (t,x) -n x. L is a subgroup of G with the property
M c Lo, hence [G' : L'] < @; furthermore, (L')' = M' and thus
K = L n H has the desired properties.
The fol I ow'i ng theorem characteri zes expì i ci tly the
maximal type I central projection 'in V(G,o). It is the main
result of this chapter.
THEOREM 4.4. Let G be a discrete group uith normaLized
rruLtipLieruandLete (r,espectiueLg en, tt = 1, , n l-)be the masímaL type L (respectiueLy type In) centnal proiectior"s
in V(G,u), t?rcn
I n<æ ne -e
and the foLlouíng az,e equiuaLent.
(d el0(b) thene eæis'bs a subgz.oup H of G such that
tG : Hl < * and lH'l < * and rlHrH ís tz'iuiaL,
(c) IG : Al . -, lA'l ( æ and G adnits a finitedimensionaL u-z,epresentation. (l d.eno'Les the.
u-fini'te eLass group of G.)
Suppose, e I 0, t\ten there eæists a l-Cimensioy¿aL
ur-vepresentation y of G, sueh that
nlç = (d'im n). ^,(, for aLL finite &imensionrtT.
.,o-represenbations n of G.
(i)
93.
(¿.1) I v(g)p(g).e= ge G.
PR00F-. Suppose e I 0, then en I 0 for some integer n.
Lemma 3.3 ensuresthat an 'irreducible representation t of
enV(G,ur) will give rise to a finíte dimensional o-representation
r : g + r(eno(g-l))* of G, where A* denotes the adjoint of A.
Together w j th Theorem 3. 2 , thi s yi e'l ds (c ) .
Suppose we have (c). Let r be a fin'ite dimensional
o-representation of G. After taking determjnants,
,n(x,y)det n(xy) = det n(x) det n(y), and we see than ,fl is
trivial, to úy Theoren 3.2 and Lemma 4.3, there exjsts a group
Ksuchthat tG: Kl <-Ço)K', lG.l <-ând(G')"= (K')'. Let
n be a.n irreducible finjte dimens'ional o-representat'ion of K,
then no: (l,x) + rn(x) is a finite dimensional representation of
Kûr (I.6), so no (1,x)(1,y)(1,x)-1(t,y)-1) = 'r(x)n(y)u(x)-1r(y)-i = I
all X,y € K, but n is irreducible so it is one dimens'ional (see
remark following I.5.1), consequently trlfrf is triv'ial. This
implies (b).
Finally suppose (b) 'is true. Since urlHrH is trivjal, H has
a one-dimens jonal o-representat.ion, hence by 'induc'ing (I.8), we
see that G admits a finite dimensional ur-representat'ion. Let n
be such a representation. Lemma 4.3 is applicable, so as'in the
preceding paragraph, n(x)n(y) = tr(y)n(x) al I X,Y € K, and â = K' ,
where K is the subgroup we obtain from Lemma 4.3, consequently
In(xyx-ly-1) = r(y,x)r(y-1,¡-1)r(yx,y-1x-1).
94.
Since the lefthand side of this expression is independent of
the way we express xyx-ly-l as a commutator, and since the
righthand s'ide depends only on the dimens'ion of n, the functìon
y : xlx-In-t + o(.y,x)r(y-1,¡-1)tr(yxy-lx-l), X,Y € G, extends
to a well defined o-representation of G that satisfies (i).
To complete the proof of thjs theorem, we must show that our
cument assumptions lead to the statement (a) and (ii ). Let
-LFr = lËT Inuav(g)P(g)'
By Propos'iti0n 1.5, f is central in V(G,o). Since g + Y(g)p(g)
is an ord'inary representation of K, f2 = f. We cla'im that fV(G,tu)
is abelian. Suppose a,b e V(G,') and n is a fin'ite d'imensional
o-representatjon of G, then
(fba), = I*ï lru*,v(z) (5a)z-.rro(z-r,x).
= Tl*T Iru *, Ir. KY ( z ) a.ubr- t *y- r' (z- t x,y- I )' ( z- I, x )
= Él f r. *, Iy. Kaybr- t xy- ir ( x) n (v ) t ( z- r xv- r ) - r
Simi I arly,
(abf)* = ¡*\ Ir.*, Ir.ruyby-r*r-r'(x)n(y)n(y-1xz-1)-1
Since K' is normal in G and xy-rx-ly. K', {y-txz-I : z € K'}
= {z-rxy-r : z €. K,}. It follows that (fa)(fb) = abf = fba = (fb)(fa).
If we write fV(G,o) as a module direct sum
95.
fV(K,ur)p(91) @ ... o fV(f ,r)p(gk)
for some set of coset representatives 91, , 9¡ modulo K,
then by representing fV(e,r) as right mu'ltiplication on
itself, fV(G,o) is a matrjx algebra over the abelian algebra
fV(K,o) and thus by I.4.2 and I.4.5 is' type I. Thjs proves (a).
By iook'ing at the irreducible representations of enV(G,to),
whenever this is non-zero, and using I.5.3, we see that eV(G,r)
has enough finite dimensional representations to separate the
points of eV!n,r). However n is a non-zero fin'ite dimensional
representation of eV(G,r), g + "(p(g-t))* is a finite dimensjonal
ur-rêprêsêntation of G, hence by ('i), n(p(S)v(g)) = I all g e è = K',
thus
,,(f ) = Þi Is.G ,,(p (g)v(g)) = I,
from which we obtain n(e - f) = n(e) - n(f) = 0. Since n is
arb'itrary, we conclude that e = f. This completes the proof
of Theorem 4.4.
Note that for t¡ trivial, this theorem reduces to results
due to Formanek [10, Theorem 2] and Schlichting [41, Satz L].
The fol'lowing theorem is a consequence of 4,4.
THE0REM 4.5. Let G be a &Lscete grory and u a
norrnaLízed muLtipLíer on' G, then the foLLoaíng are equiuaLent,
U,) V(G,ur) is tgpe I (or equiuaLentLy type Iç),
96
(ì.i) euetaA u-representatíon of G is type I,
(Ì,ii) G has an abeLian subgz'oup A of' finíte indeæ
in G sueh that for aLL x,Y . A, o(x,y) = t¡(y,x)
ftu) g6 h¡ith &iscv'ete topoLogy) ís type I.
To prove this theorem we introduce the notion of a
twisted group a'lgebra. For a discrete group G with
normal.ized mult'ipliêF rrr, the twisted group algebra A(G,r.,r) of
G consists of finíte'ly supported measures on G with multipf ica'Lìon.
(uv)x = Iy 6uyvr-i, (v-l,X),
and *-operat'ion
(u*)* = (u*-r)-,
u,V e A(G,ur), X,Y e G, where ux (x e G) denotes the measure of
the set {x}. Each r¡-1-representatioh n of G extends natura'lly
to a representation
u * Ig.G ugn(g)
of A(G,r¡) which, when no confus'ion may arjse' we shalì denote by
the same letter. Note that the maP
A(G,ur) + V(G,o) : u + p(u) = IgeOuno(9),
97.
vlhere p denotes the regular o-t-representation of G, is a
*-monomorphism with range {a e V(G,u:) : supp(a) is finitei.
In parti cul ar, (A(G,ur) ) 'is weakìy dense i n V(G,t¡) .
PR00F 0F THEOREM 4.5. Suppose (i)'is true then by the
proof Theorem 4.4, there exists a group K such that
tG : Kl ( æ: K'= Ço and ,lf*f is trivial. Since the
maximal type I central projectjon of V(G,ur) is the'identity,
by Theorem 4.4,
167 IgeG v(g)p (g)1[=
hence by the un'iqueness (1.4.(ii)I è = {e} and thus K is
abelian. From r,rlç*ç trìvial , we can novl conclude ihat
o(x,y) = r(y,x) all x,y e K. This proves (i) implies (iii).
Suppose (iii) is true. Let H be a subgroup as describeC
in (iij), and r ôh o-I-representat'ion of G. If gr, ... , gn
is a set of coset represent.atjves modulo H, then
A(G,o) = A(H,ru)p(gr) ' ... @ A(H,t")o(Sn)
where the direct sum is a module direct sum. Again, by
representing A(G,t¡) by right multipl'ication on itself , we
can embed A(G,o) in the n x n matrices over A(H,o). By
hypothesis, A(H,o) is abelian and so A(G,o) satisfies a
polynomiaì identity (I.4.2). Sìnce the von-Neumann algebra
9B
generated by n ìs the weak closure of n(A(G,r)) by I.4.5 and
I.4.6, n ìs type I. This proves (iii) impìies (iì).
The implìcation (ii) implies (i) is trjvial.
Suppose we have [G : A] < - ând ,(x,y) = ,(y,x) al l
X,) e A, then [G': A'l < - (4.1 (ii¡) and Ao is abelian,
so by the equivalence of (i) and (ii), Go is type I.
Conversely, if Go is type I, then an abelian subgroup of
finite jndex in Go projects onto an abelian subgroup of
finite index in G on which ur ìs symmetric, so (G,ur) is
type I . Thus (i i i I and ('i v) are equi val ent.
COROL|..ARY 4.6 . The subgroup A in 4.5 ('ii i) may
be taken to be Zr(l).
Thus
tG
PR00F_. G' type I implies [G' : Z((G')fC)] . - (I.7.10).
lG : zr(n)l = [G' : (zr(^))'] = [c' : Z(n0¡1 =
I z((G )rc)l < æ usìng Proposition 4.1.
COROILARY 4.7 ([3, Lemma 3.1]), If G is a discrete
abeLian group, then V(G,r) .is type î if and on,Ly íf
Z'r(G) = Gr¡ = {g . c : o(g,x) = r(xrg) = I aLL x < G} has
finite ínã.eæ in G.
PR00F. The ''i f ' part fo'l I ows trivi aì ly from Theorem
4.5. Suppose V(G,or) ìs t¡rpe I, then for each x e G, Cr(x)
contains the group 7u,(t) which according to Theorem 4.5 has
99
fin'ite index in G, thus ¡ = G and [c : Zr(G) ] . -.
COROLLARY 4.8. If G is a discrete abeLian .Çroup, then
either V(G,oJ is þape I oz, Y (G,o) is type tt.l,
PR00F. Suppcse the maximal type I part of V(G,r) is
non-zero, then by Theoren 4.4, there exists a subgroup H
such that [G : H] < * ând ,lH*H is trivìal . S'ince H 'is
automatically abelian by Theorem 4.5, V(G,r) is type I.
Theorem 4.4 is somewhat awkward for practical use
because 'it is generally dìfficult to establish if a
group and multipl'ier satisfy cond'itions 4.4(b) or 4.+(c).
One would like to replace these conditions by [G: ^] ( -,
l¡'l < - ând rn it trivial for some n. The foilow'ing
examp'le shows this cannot be done.
EXAMPLE 4.9. Let G be the discrete group H x H',
where H = JI - 7(2) and H, = (D .- Z(2) and defineJ=j=1
,((aj,bj),(ar,,b¡')) = exp t}l I;=i(ajbj' - uj',br')1,
(aj,bj),(aj,brj) e c. By 4.7 and 4.8, V(.G,o) is type n1,
despite tlre fact that G and o sat'isfy tG : Al ( -r
l¡'l < @ and o2 = 1.
EXAMPLT 4.10, Let G = 7 x Z and
o((m,n)(mln')) = exp [2nicr(mn' - m'n)].
10(
Here V(G,r) is type I if cr is rational and type II1 if cr
is 'irratinal . (Use 4.7 and 4.8).
EXAMPLE 4.11. Let p be a prìrne and let G be the
group wi th fi ni tely many generators b,â 1 ,â2 . . . . and the
definjng relations
atb = bat ì = Ir2,
bP=arP=uf
ui*k ai = baiai+ft ì,k = 'l,2,
ti, tl un'n' ) = exp I+ ( sr så - sïs2l )Sn
The elements of the form bs'a,
It is clear that the commutator subgroup of G coincìdes
w'ith the centre cf G and is equa'l to the finite cyc'lic group
<b>. This itnp'ì'ies that there is a bound on the size of the
conjugacy classes of G.
sb a1 un bo
I S sô4 a n form a normal
n
subgroup H of G that is contained in A, hence [G: A] < æ;
also lA'l = lG'i = p, so by Theorem 4.4, V(G,r) has a non'zeno
maximal type I part. However, because Z(¡) n H g Z(H) = <b>,
the group Z(n) and consequently Zr(a) has infinite index in G,
thus by Theorem 4.5, V(G,r) is not type I. l^Je have shown that
Corol'lary 4.8 does not necessarììy hold for non-abelian groups.
Let o be the mult'iplier
S" i(
4
10
The foliowing theorem is an appl'ication of 4.5 to jnduced
characters.
THEOREM 4 12. Let G be a díscz,ete group and N a normaL
stbgroup of G, If ), Ì.s a aharacten of N, that ís a I-dimensíonal
nepz.esentatíon of N, d.enote bU Kx = {g € G : gng-i¡-re ker }.
aLL n e Nj the stabilì,zen of À., then the üon Newnann algebra'
Y
^ genez'ated. bg the induced z'epz'esentatíon ),+f;, is t14pe I
íf, and onLy if K^ contaíns a subgroup A sueh that [K¡. t A] < æ
ma Ar c ker À.
PR00l. Let P c K^ be a set of coset representatì ves
modulo N (includ'ing the identity of K^). For each x e K^ denote
bV sx the unique element in P such that xN = o*N. Let rf be the
functionG+[rl ; x+cr;rX. The relat'ionsû(n) =n, r](xn) =ü(x)n
and ú(nx) = {,(x)x-lnx, for x e H, n e N are evident. It js clear
that i : x -> r(V(x)) is an extension of À to KÀ and that the
multiplìer o(x,y) = l(ú(xV))r(,p(x),t,(V))-1 assoc'iated with thÍs
extension can be factored through K./N to yield a multìplìer ur
on K^/N. We may assume without loss of generality that o'is
normalized. By [24, Theorem 7], the von Neumann algebra V^
generated by the induced representation ^^fi
ls *-isomorphic
to V(K^/N,o), thus by 4.5 we deduce that V^ ìs type I if, and
only if there exists a subgroup A in K^ such that both ('i)
[Kr : A] < - ând (ìi) A' g N, o(x,y) = Õ(y,x) aLL X,Y e A,
hold. It remains t,o show that (ii) 'is equivalent to A'c ker ^.
Suppose (ii) is true. Let x,y e A. S'ince x-ty-lxy e I'1,
u(xv) = q(yxx-ly-lxy) =,/,(yx)x-Iy-1xy, but from o(x,y) = o(y,x)
follows r(V(yx)) = r(q,(xy)) so that À(x-tt-rxy) = 1. Conversely
ìf A'c ker À c N, then fon x,y e A,
10Í
r(xy) = r(u(yxx-rv-rxv)) = r(.p(vx)) À(x-tt-1xv) = r(u(vx))
and thus o(x,y) = o(y,x). Thìs completes the proof.
103.
CHAPTER iV2
GROUPS WITH FINITE DIMENSIONAL IRRTDUCIBLE MULTIPLIER RTPRESENTATIONS
Let G be a localìy compact group and u a normalized
multiplier on G. Recall that V(G) denotes the von Neumann
aìgebra generated by the ìeft regular (ordinary) representation
of G (1.7). In Chapter III we defined and. investigated the
structure of the von Neumann algebra V(G,or) in the case when
G is discrete. In the more generaì situation when G'is ìoca'l'ly
compact (but n,ot necessarily discrete), the definjtions are
sjmìlar. Indeed, the map p : G + U(12(G)), defìned by
p(g)f(x) = o(g-I,x)f(g-1x)
a'lmost everywhere in x, all g e G and f e L2(G), is a
t,-I-representatjon of G and generates a von Neumann algebra
also denoted by V(G,t¡). p 'is called the ìeft regular o-l-
representation of G.
In th'is chapter, we determine necessary and suff ic'ient
cond'itions on G such that the maximal type I¡ centraì
projection in V(G,ur) js non-zero (respect'ively the ident'ity
operator in V(G,r)), and construct this proiection expl'icitly
as a convolution operator on L2(G). This extends the results
of Chapter III
2After finishi.ng this thesis, I received a preprint entitled'rThe type structure of multiplier repiesentations rvhich vanishat Infinity?r by E. Kaniuth and G. Schlichting, which containsresults that are similar to those presented in this chapter.
10
For the case of ordinary representations, th'is probìem
has been successfully dealt with by Kamith [23] and Tayìor [48]
We point out that it'is no longer the case as'it was
for G discrete (see III.1.6(iv) and I.3.2) that V(G,r) is
necessarily finite. For an examp'le see 4.1. It is for this
reason that our methods allow a characterìzatjcln of the type
I, part, but not the type I part in V(G,o).
Recall that for a locally compact group G, Go denotes
the von Neumann kernel
G ñ{ker n : n e G^ and dinl n < -}.
and G* denotes the topological finite class group of G
wh'ich consists of all elements of G that belong to a relatively
compact conjugacy class (I.7). If ur is a normalized Borel
multiplier on G, then Go denotes the central extens'ion of
G (I.6). For a subset ll of G, H- denotes the closure of H in
G, and if H is also a subgroup, H' denotes the commutator
subgroup of H.
1. Prel 'imi nari es
Throughout this section, G will denote a locally compact
group and o a Borel multip'lier on G.
o
1ûl
LEI'4MA 1 . 1. Let A be a subset of Gu and. Let h : G0 -,' G
denote the eanonicaL proiection, then
(í) A- is cornpaet if anÅ onLy íf h(A)- is
contpact,
(i¿) ¿f A- is conrpact, then h(A)- = h(A-).
PR00F. A- compact 'impl ies h(A-) compact, but h(A-) ) h(A)'
hence h(A)- is compact and h(A-): h(A)-. Conversely, if h(A)-
'is compact, then by [18, 5.24], h-I(h(A)-) is cornpact; but
h-1(h(A)-) r h-lh(A) ) A, hence A- is compact and
h-1(h(A)-) ) A-, that is h(A)- = hh-1(h(A)-): rr(A)-. This
completes the proof.
C0ROL|-ARY 1.2. Let G be a LocaLLy conrpact gt'oup and u
a norTnaLized Borel muLtipLíer',
(¿)
(ii)(GFC)' = (c')rc.
rf one of (G¡rç) '- arld ((G')FC) - is cornpact,
then so is the ot'her and (Gp) ' = ht((G')FC)'-1.
PR00F. For (i), let A equal the coniugacy class of some
(t,x) e Go and in ('ii), let A = ((Gur) FC)'. Now use Lemma 1.1.
Combinìng 1,2 wtth I.7.8 shows that the maximal type
I, central proiect'ion'in V(G) is non-zero jf and on'ly if the
maximal type I, centra'l pnoject'ion in V(Go) i s non-zero. Thi s
follows also from Tay'lor [48, Proposition 5.2].
106
Let En, n e v-be the maps on L2(G') g'iven by
tnf (t,x) = tnfnr-nt{s,*) ds,
for alrnost all (t,x) . Gt, f e L2(G'). (Compare this with
K]eppner 124, page 5631.) Cìærty EnL2(Gur) consists of all
those functions f. t2(G') such that f(t,x) = tnf(1,x) for
almost all (t,x) e Go. It follows that the En, n e Zare
mutua'lly orthogonal idempotents. For n e 7-, En is just
convolution by the measure obtained if you multipìy the
measure on Go supported on 'fwhjch restricts to Haar measure
on 11, w'ith tne character fn of T'g'iven Oy fn(t) = th, t e T.
It is easy to check that En commutes w'ith the right and left
regu'lar representation of Go, so by I.7.4, tn is in the centre
of V(G0).
THEOREM 1.3. With the aboue notatíon' tVÌ.e E-n, rl € 7.
are mutuaLLy orthogonaL centz'al proieetions in V(Gu) md
the tht ee pon Netnnann aLgebr'as En V ( Gu') , V (G , on ) ond(i)
ft¿)
nülV(X_n+ff. ) - *e Don Neumavtn aLgebz'a generated by
the induced. z,epresentatíon Ï-n*$I - dve spatiaLLy
ísomorphic,
In.Z En = I (the identity operator).
PR00F. (i) Let t denote the left regular representation
of Go and on, n € V-, the left reguiar on-representatir¡n of G.
Observe that the representation space or X-ntfl is .just EnL2(Go)
10
and that Enr - .$'. It follows that Entl(Go) = v(u-n+$')./\ n
Denote by i the injection map i : G + Go and by q the
map
0:E L2(G)+12(G):foin
(Inle insist that Haar measure on Go be scaled so that the
measure of'ìf Ís 1.) Since Haar measure on Go is the product
of Haar measure on G and Haar measure on'ìi', we ha.ve
I r(t,x) | 2d(r,x) = I, [I" t,t'x)l' o'] dx
GüJ
I o(r) (x) | 2 dx,
G
thus q is an 'isometry. The spatial monomorphism
En\/(co) *g(t-2(e)) :T+0o JoO-t
is weakìy contÌnuous and maps Enr(1,x) to pn(x). Since
tnEnr(t,x)= Ën.(t,x), the von Neumann algebra generated by
{Ent(r,x) : x € G} is precisely EnV(G'). Thus any element
in EnV(Go) 'is the limit of operators T with the property
O(T) e V(G,orì), thus by the weak continuity of ö, O(T) e V(G,rn).
In other words, the range of O is a subset of V(G,rn). By
Sakai [40,1 .1.6 "21, the range of q ì s weakly cl osecl; i t contai ns
the operators 4 o En.(t,x) o ó-l = on(x), X e G and therefore
must equal V(G,rr).
10
(ii) From the theory of characters, the direct sum
nnr. Zn is the regular representation of T,
hence by the non-separable version of I.8,2,
we see that the induced representation
(* n.z.xn ) +et
is the regu'lar representat'ion of Go. Since inducing commutes
with taking direct sums (non-separable versÍon if I.8"3), we
have
*¡.7(xnrfi') = (* nuzxnl ^,f;.'.
Thus v(e0) = *nuv.v(xn+ff),
which is the required result.
LEMMA 1.4. ( r3l ). Let n be a u-z,epnesenta'bion of the
Dernte by p-ker r the eLosed. nonnaLLocaLLy compacl; group G,
subgnoup,
{x e G : n(x) = y(x) ,I, foz. some y(x) e 'llJ
of G. Then u is sirníLay to a rn¿LtipLierLífted from G/p-ker n.
PB00F. Let { be the canonical homomorphism from the
unitary group U(Hr), where H,,, denotes the Hjlbert space
associated with n, to the quotíent t-r(Hr)/T. According to Feldman
and Greenleaf [9], this map has Borel transr¡ersal, that is a
Borel map o, back from U(HîT)/11 to U(Hn), taking the identity
10
element ìn U(Hn)/'lf to the identity element in U(Hrr) such that
rf o a is the identity map. From this we obtain the commutative
di agram
e Iu(Hn) gu(Hr)/1t
,þi+dIt
u(Hn)
where n' is the map ü o {., o n. C'learìy n' is a multip'lìer
representation of G associated with a multiplier ur, Sôy, and
since rp " n = ! o Trt, (,r' is similar to t¡. From the definition
of n'' it follows that o' is constant on p-ker îT cosets in G x G,
that is, Ít'is ljfted from a multjplier on G/p-ker n.
LEMMA 1.5. Let G be a LocaLLy conrpact gnoup ui'bh
norrnaLized BoreL rru.LtipLer u. Suppose the maæinnL type I,centraL pz,ojection ín Y(G) (or equiuaLentLy in y(eo)) ¿s
non-zero, and that G admits a finite dimensionaL u-r,epz,esentation
lrr then K = h((G')") is conrpacb (uhev,e h is the canonicaL pz,ojeetíon
Go * G) an-d. u is simiLar to a nuLtipLíen uhich ds Lifted fnom
G/K.
Pß00F. That K is compact follows from 1.7.9. Suppose (t,x)
then since (s,y) -t sn(y), (s,y) . G', 'is a finite dimensional
(ordinary) representation of Go, we have I = tn(x), that isX e p-ker n. The result now follows from Lenlma 1.4.
. (G')" ,
1L,
2. The main theorems
THTOREM 2. 1. Let G be a LocaLLy cornpcct group uith norrnaLized
BoreL rruLtipLier u.
equiualent.
Then the foLLouing three conditions are
(i) The maæimaL type I, centz,al pz,oiection e ín
V(G) (oz, equiuaLentLy" the manimaL type I,
centraL pz,ojection in V (G')) is non-zey,o and. thene
eæists a firn te &imensí.onal. u-z.epz.esentation t
ol v.
(ii) Tke ma.ßnimaL tgpe I, centraL pz,ojection d in
V(G,ur) is non-zero.
(i'ùi) tG : Gpçl ( @: (GfC)'- ís conrpact and G adnits a
fì,nite dímensionaL ct-r,epresentation n .
PR00F. (i) and (iii) are obvjously equivalent by I.7.8.
Suppose (ii) is true and ler n be a finite dimensional representation
of dV(G,r). If we compose representation wjth the projection
V(G,o) + dV(G,o) : a + da, we obtain a representatioh r of V(G,o).
By I.6.3, the left regul êF o- l-representat'iot1 p of G corresponds
to a representation p'of Ll(G,,u-t¡ which v¡e know to be faithful
( [2,t9.3.6]). Thus p' is a *-monomorphism of Ll(G,r-1) to v(G,o)
and n o p' is a representation of ¡t(6,r-i). Again by I.6,3, this
representation .ôrr.sponds to a ,-l-representation of G and it is
easy to check that this o-l-representation is just g * n(dp(g)),
g e G. Hence g + n*(dp(g-l)), where * denotes the Hilbert space
adjoint, is a finite dimensional ür-representation of G. 0n the
other hand, by Theore 1.3, the max'imal type I, central projection
11
'in V(Go) is non-zero. This shows (ii) implies (i). That
(i) imp]'ies (ii) will be proved together with Theorem 2.3.
LEMMA 2.2 . Suppose K is a compact norrnaL subgz'oup of G
mtd u is Lif'bed fz,om a rruLtipLíez' u' on G/K, then K is also
a eornpaet norrnaL subgz,oup of G' and Gu/K is topoLogieaLLy
isomoz,phic to (G/K)"t.
PROOF. Let rf be the mforph'ism Go + (G/K)o' , (t,x) * (t,xK).
The kernel of this map is {(1,k) : k e K}, that is the image of K
in Go, thus the induced map from GolK to (G/K)' is an iso-
morphism. This isomorphism preserves the Haar measures on these
groups; they uniqueìy define the topoiogjes on these groups,
thus Go/K and (G/K)'' uru also topologically isomorphic.
THEOREM 2. 3. Suppose G is a LocaLLy eonrpacb gnoup uith
BoreL rru.LtipLier u. Sttppose the maæimaL type It centraL
pr.ojeetion d ín V(G,c.r) is non-zero. l,/e asswne (usíng 7.5 and
2"1) that u ¿s tifted fz,om a muLtípLier u' of G/K, uhere K ís
the eonrpaet nonnaL subgroup K = h((G') ") (h being the canonieaL
pnojeetion Go + G) , Ihten d is the opez,ator L2(G) + L2(G)
defdned by
f(k-r¡)dÀ(k),K
aLmost aLL x e G, f e L2(G), uhere À. is Haar measu.re on G
normaLízed sucTt that x(K) = l. Fuztherrnone, for eaeh n e Z, d
df(x¡ =
ís the ma,æimaL type I, centz,aL pnoiection in Y(G,rn) and dv(G,r¡n)
111
is isomozphic to V(G/K, (r')n).
PR00F. First we give a proof, as promised, of the statement
'(i) impf ies (ii)' of Theoren 2.I. Let cr : L2(G) * Lz(G) be
defi ned by
af(x) = f(k-1x) dr(k),K
almost all x e G, f e L2(G). The proof that a is a central
idempotent in V(G,rn) = EnV(G') (and hence in V(G')), and that
qV(G,on) and V(G/K6')n) are spatially isomorphic is sjmilar to
the proof of t'he correspond'ing facts about En in Theorem 1.3.
Si nce
v(G,rn) - v(G/K,(r')n) * v(e/r,(r')n)t
a^(where r denotesþrthogona'l
1.3'
q.Lbtlqe'V.<¿L
€ome+èñefi,i) , lve have by Theorem
V(c') = V((c/K)'') * V((G/Kt")r
but (G/K)'' and G'/(G'). are topologically isomorphtc (2.2) and
V((G')/(G'). ) is spâtjally isomorphic lo the maximal type I,direct summand of V(Go) (I.7.9). In particular we have d = a I 0
Now assume d I 0, then by (ii1 implies (i)r of Theorem 2.I,
the maximal type I¡ central projection in V(G') is non*zero, thus
by the same argument as above, we reach the desired conclusion.
1i:
COROLLARY 2.4. Suppose the manimaL type It centraL
projeetion d in V(G,o) ís non-zero, and Let n e 7-, then the
foLLouing equations obtain,
G" = h [((Gt)"] = h [n {ker r : r is a finite dimensiortnL
representation of G^ such that
nlrr(t) = tnÌl
' {g e G : there erists y(g) e Tl such that
,r(g) = y(g)I for aLL finite dimensionaL ,î-
repnesentatíons of Gj, .
uhez,e h : et * g denotes the eanonical pz'ojection.
PR00F. Let K = h[(G').] and denote the last two sets in
the above equality by H and L respectively. That Kc- H c L
is clear from the definit.ions and the property that a on-
representation n of G extends to an ordinary representatìon îT' of Go
such that n'l'(t)= tn. Using the proof of Lemma 1.5, we assume that
t¡ i s I j fted from a mul t'ipl i er on G/1. The fi ni te dimensi onal
representations of dV(G,rn) separate the points of dV(G,rn), hence
n'(9) = 1 if and only if t'*(dp(g-1)) = r*(d) for all such
representations nT where no denotes the ürn-representation g * n*(dp(g-1))
and p is the regular urn-representation of G. Thjs happens if and
only if p(g)d = d. If g e K, then c'lear'ly p(g)d = d. Conversely,
suppose g I K and p(g)d = d. Let U be a neighbourhood of gK'in
G/K of finjte Haar measure, not containing K. Let rp' be the
characteristic function of U and ,þ Y^lifting to G, then rl e dlz(e ),
is continuous at g and satisfies V(gK) = {1}, ü(K) = {0}. Since
ll¿
p(g)d = p(g), we have
,(g-1'x),¡l(g-1x) = ú(x) ,
almost all X e G. Substitut'ing g = x int<l this formula gjves
Q = lV(t)l = lV(g)l = 1 which is a contradiction. tlle have shown
that n"(g) = I if and only if g. K, thus
K = n {ker n' : n is a finite dimensional
representation of dV(G,rn) ]
)L
If we let n = 0, we obtain the remainder of the corollary,
thatisG=1.
LEMMA 2.5. (Moore [31], Lemma 4.1). Let G be a LoeaLLy
Conrpact gï.oup uith norTnaLízed BoreL muLtipLiez' u, Each
iw.edueibLe u-r'epresentation of G is finite dime'nsionaL íf and
onLy if euevA u-Tepresentation of G ís finíte,
PR00f. The if part is clear because an irreducible
to-FêpFêsentation is finite if and oniy'if it is finite dimensÍonal.
Let A be the twisted group C*-algebra C*(G,o) (see I.6). By
TheorsnI.6.3,'it is sufficient to prove the corresponding statement
of the Lemma for representations of A. If I is a two sided primitive
ideal of A, then I is the kernel of an irreducible representation of
A. According to our hypothesis, n has dimension n for some n.
Since r¡ ìnduces a homomorphism for A/I to the n x n matrices, A/I
satisfies the polynomia'l identitY Szn (I.4). Let Fn be the set
i1i
of primitive ideals I such that A/I satisfies the polynomial
identity Szn, then by assumption F = UFn is the primitive'ideal
space of A. For each subset K of F, the kernel I(K) of K is
n I (I € K). The closure of K in the kerne'l-hull topo'logy on
F is K- = {J e F: J r I(K)}. hle show that Fn is closed in
this kernel-hull topo'logy. Sjnce S2n(A) c I for each I e Fn,
the polynomial Srn 'is satisfied in Al I(Fn). Moreover, ifJ e Fi then Srn(A) c. I(rn) c,l, hence J e Fn. Thus Fn is closed.
t^le define for each closed subset K of F and for every
representation n of A, Pn(K) to be ihe proiection onto
H-(K) = {x e H-: n(a)x = 0 all a e I(K)}. According to1I' TI
[16 : Theorem 1.9:l , f -' Pn(K) extends to a countably addjt'ive
projection valued measure on the Borel set,s of F with P,r(K) in
the centre of the von Neumann algebra V(n) generated by n. Since
F = UFn, Hn = I(H,r(Fn) - Hn(Fn-r)). We define a subrepresentation
nn of n by restrictihg n to the invariant subspace Hn(Fn) - Hn(Fn-r).
Then n = Si nce S 'is satisfied211
in A/I(Fn) as noted above, the algebra nn(A) and hence its weak
closure V(nn) satjsfjes Srn = 0. Now any von Neumann algebra
satisfying this identity is finite (I.4.5). Moreover V(r) is the
direct sum of the V(nn) and since the direcl sum of finite a'lgebra
is finite (I.3), V(n) is finite as desired.
Let G be a LocaLLy eorrpact group and u a norvnalized
[nn, and n,.'(a) = 0'if a e I(Fn)"
BoreL rm,útipLier on G" The foLLouing are equíuaLent:
THTOREM 2.6.
(¿) V(G,tr) ís t11pe If.
11r
ø¿) ALL irz,educible u-nepresentation of G are fínitedimensíonaL.
(iii) The foLLouirry conditions hoLd.
(d [G : GrC] ( *,(b) (GrC)'- is cornpact,
(e) G adniis a finite &imensionaL u-nepresentatíon,
(il G = {e}.o
(iu) Gu ís a \koore gr¿oup, that is aLL its iz'reduc.ible
(ordLnany) repnesentations are fíníte dimensionnL.
PR00l. (i) 'implies (iii). By Theorem 2.1, tG: GpçJ ( @s
(GfC)'- is conípact and G admits a finite d'imensional o-representation.
Further by 2.3 and 2.4, V(G,tr) type I, implies
{e}=K=h((G')o)=Go
(iii) impì'ies (i). By 2.I, the maximal type I, central
project'ion in V(G,ur) ìs non-zero, so using 2.3 and 2.4, we obtain
V(G,ur) is type I.
(ii) implies (iii). Let A = C*(G,o) be the twisted group
C*-algebra of G. Since each representation of A is finite
dimensional, n(A) is contained in the compact operators for each
irreducible representatioh n of A, A is CCR (or lim'inal), thus
by Dixmier [7,5,5.27, A is type I, consequently V(G,r) is type I.
To see that V(G,o) is finite, apply Lenma 2.5.
(iii ) impìies (iv). If n 'is a finite dimensional o-representatjon
of G, then the n-fold tensor product n o ... @n is a finite djmens'ional
1i
,n-r.pr.rentation of G, hence V(G,rn) Ís type I, for each n e Z-
(recall that we know (i) and (iij) to be equiva'lent) and by Theorem
I.3, V(Gt) is type Ir. It follows from 1.7.6 that Go is a Moore
group.
(iv) impf ies (ìi ). If n is an irreducible t¡-representation
of G, then (t,x) -' tn(x) being an (ordinary) representation of Go,
must be finite dimensional by assumption.
Before we proceed to examples,'we need one other result, a
result obtained using Theorem 4.5 of Chapter III' which generalizes
Moore [31, Theorem 1] and Tay'lor [48' Theorem 2]
THEOREM 2. 7. Let G be a LoeaLLy cornpact gttoup and i': a
noz,maLized Bov,eL tnuLtipliez' on G. The foLlouing are equiuaLe.nt.
(í) V(G,o) í.s type I.¡, fot'some íntegen k, that is non-zez'o
ma,æ¿maL type In centz'al pz'oiections occur in
V(G,o) only if n < k.
(i.¿) G has an open abeLían subgroup ll of finite inÅ.eæ
in G such that the restriction of u to H ís triuiaL.
(íii) The ír'redacibLe u-r'epresentations of G are of
dimension at most k for some ittteger k.
PR00F. (i) impfies (iii). tle follow the proof of Taylor
[48, Theorem 2]. From I.4.5 we know that Sr_O = 0 is satisfied
in V(G,o). Suppose n is an jrreducib'le o-representation of G,
then r' : g + n*(g-t) is an irreducible o-t-representation of G.
Now, as jn the proof af 2.I, the left regular representation of
11i
L1(G,o-i) Ís a *-monomorphism into V(G,o). Hence L1(G,ür-1) also
satisfi.t srk. It follows that v(n') satisfies sr¡, consequently
?r' and r are of dimension at most k.
(iii) imp'lies (ii). Following the proof of Moore [31, Theorem
11, we consider the underlying group GO of G (with discrete
topology) and the corresponding twisted group algebra A = A(G¿,ur)
(see definjtion following III.4.5). To each r-1-representatiorì n of
G, there corresponds an algebra representat'ion no of
A : u * Ig.G uno(S). First we show that alì representatiohs no
obtained from irreducible o-I-representations r of G separate the
points of A.
The twjsted group algebra A acts on Lr(G,r-t1 as follows
u.f(x) = Igesupp uugf(g-1x)r(9-1,x),
almostall XeG, lreA, fe Ll(G,r-Ð. Givenanon-zerouinA,
it. is clear that u.f I O for some f e Ll(G,o-l). Let n' be an
irreducibìe representation of Ll(G,t,-1) such that n'(u.f) I 0 (I.5.3).
According to I.6.3, there exjsts an irreducible o-1-representation
rT of G such that
( n'(f)E,d ( n(x)E,.ù f(x) dxG
all Er¡ € Hn,f e Ll(g,ru-t¡. Using the invariance of Haar measure,
we obtain
ug I( rr'(u.f)t,nl =I
G
( n*t,rù ,(g-l,x)f (g-lx) dx
111
un In , ngrt,.ù r(g,x)f (x)
( nnn*E,.¡) f(x)f,
SrO(n(A)) = n(Sr¡(A)) = 0 for each suchn, that SrO = 0 is satisfied
I
I
dx
u dx
Hence 0 I "'(u.f) = no(u)"'(f). It follows that n'(u) I 0.
This shows that A has a separating family of representations
of degree at most k. I^le now infer, from the fact that
in A.
g
Let n be any irreducible t¡ ]-representation of GO and n' the
correspond'ing representation of A. By I.4.6 and I.S.1, Sr¡ is
satisfied in B(H?T) and we conclude that the dimension of n is at
most k. It now follows from III.4.5" that G has an abelian sub-
group K of finÍte index in G such that rlf*f is symmetric. The
closure H of K is an open abeì'ian group and if o = rlH*H, then the map
ã : H + H^ defined ¡V ã(g)(h) = o(h,g)/o(g,h) is continuous
(II.1.L), so it follows that o(g,h) = r(h,g) for all h,9 e H and
consequently (II.1.1), ,ittrH is trivial .
(ii ) irnp'lies (i ). Let 91,
representatives modulo H, then
, g¡ be a comPlete set of
A(G,ur) = *f=r A(H,r)p(g)
is a matrix algebra over the abelian algebra A(H,o), then as in
the proof of III.4.5, V(G,or) is tYpe I.O.
rzt,
4. EXAMPLES
Let G be a locally compact group wjth Borel multiplior t¡.
In Chapter III we saw that for G discrete, (G,o) is type I ifand only if GtD is type I and this occurs if and only if V(G,o)
is type I; and in this chapter, this has been generalized to
assert that for G locally compact, ihe following are equivalent.
(i ) G has only finite dimensional ur-Ì^êpresêntations, (ii ) Go is
a Moore group, and (iii) V(G,o) is type Ir.
Example II.3.4(iii) shows that (G,o) type I does not necessarily
imply that Go ,is type I and Mackey (129, Sect'ion 7l ) constructs a
non-type I group G such that V(c) is type I.
EXAMPLE 4.1. Let G be the group IR x IR and o the multiplier
r(x,y)(x',y')) = eioxy'
(x,y), (x',J') e lR X IR , r,¡here a is an irratjonal number. By
II.1.3, V(G,o) is type I. However, all irreducible ür-representatìons
of G are infinite dimensional [3, Theorem 3.3], thus by 2.1,
V(G,o) is not fjn'ite.
EXAMPLE 4.2. Let G and t¡ be as in III.4.9, then V(G,r) is type
Ir1, )êt G satisfies the first two conditions of 2,1(iii), thus G
admits no finite dimensional o-representations.
Now we expand on Exarnple 4.10 of Chapter III. For each
t = e2nio . 11, (o . lo,2¡r[), lve obtain a multiplier ort on L x 7-
defined by
T2
,r( (m,n) , (m' ,h' ) ) = t*n' ,
(m,n),(m',n') . Z x V. Theorems II.2..3 and II.3.5 show that
(up to similarity) all multipiier on Z- x V are of this form.
hle say that t is rational (respect'ively irrational) if q js
rational (respectively irrational). As'in Example IiI.4.10, we
see that V(V-xV,u¡) is type I if t is rational and type II1
if t is imational. Suppose o is rational , Sây a = p/q with
p and q relatively prime. hle wish to knovr more about the
trr-FêpFêSêntatjons of G and to this end, we util'ize Hannabuss
Il7,Theorem 4.1] which says that the irreducible or-representations
of G are all induced from characters of a maxìmal isotrop'ic sub-
group (tnat is a subgroup H such that Ho = H; see definition
following II.1.2). Now cìear'ly the subgroup H ='{(m,nq) : n,n e Z}
of G = ZxV- is maximal isotropic and G/H is isomorphic to Z(q)
thus the ot-representations of Z x V- are all of dimension q.
EXAMPLE 4.3" Let Ap (p a fixed prime) be the group of p-ad'ic
integers (for details see Appendix) and G the group VxZ. x ap
wi th mul t'i pì i cati on
(arbrx) (z' ,b' ,X') = (a + a',b + b',x + x' + ,r(ab')),
(a,b,x),(a',b',X') e G, where ú : L* op is the canonicai injection
of Z onto a dense subgroup of no (see Appendix). We topoligize G
so that lO becomes a compact open subgroup. With this topo'logy,
G becomes a I ocal ly compact separabl e topo'log'i ca1 group, For
each t e 'lf, we define a multiplier
12i
of = ûJt o k,
where k : G -> V- x V- is the canonical- homomorphism k(m,rì,X) = (m,n).
Given an irreducible ürt-representatiorì n of V x L, denote by
'rT' the or- representation of G obtained by composing ri with k.
Since nO'is compact, the abelian group dual op^ is discrete,
thus by I.8.4, every quasi-orbit on nO^ is transístive and all
the irreducible oa-representations of G can be constructed using
I.8.5, I.8.6 and i,8.7 .
Identífy,the abelian group dual nO^ of nO with the subgroup
^ = {s e 'lf : s = exp [Zrik/ pn],k ,t1 e Z] of 'lt using the cor-
respondence t\ x ap * T : (srx) -,' sx, where x * s*, s e  is the
continuous extension from Llo tO of the homomorphlsm V- -¡ Â :
n * ,n (see Appendix).
Let t e 'ïf and S e À, then the stabilizer of s in G is
all of G. Therefore all the or-representations of G are given
by I.8.7. Suppose we have a ot-representation y of G which
reduces to a multip'le of s on ÂO (recall that s is viewed as
a representation of nO). First we define an extension of s to
G as follows
: G+'ll': (a,b,x) *sx.s
The multiplier associated with this extensíon is preciseìy
72,
((a,b,x)(a' ,b' ,X')) * st a b xt b' xtS âr ,x S ,X'a ,
-5ab
6 ((a,b,x), (â',b',x' ) ).5
Thus by I.8.7, y must be of the form Tr'S', where r is an rts-t-
representat'ion of v- x v, (Note that orr-los = of as required. )
As s ranges through Â, we obtain all irreducjbìe oa representations
of G.
0bserve that ts-I is rational if and only if t is rational.
Hence the irreducible oa-representat'ions of G, t e T are all
infinite dimensional if t is irrational and are all fin'ite
dimensional if t is rational. It follows from 2.I and 2.6
that V(G,o¡) 'is type it if t is rational and the type If part
in V(G,oa) is zero if t is irrational.
Furthermore, if t is rational, as noted earlier, by varying
s, the ir"reducible urrr-t -representations of 7-x T. can be chosen
to be of arbitrary dimension.
The same is true of the oa-representations of G. Thus,
ai though V(G,r,r¿) is type I¡, Theorem 2.7 shows that V(G'tor) has
a non-zero maximal type In part for arbjtrary large n. This
phenomenon does not occur in the case where G is discrete
(combine III.4.5 and 2.7).
A.:
APPENDI X
Since there is no representation of the p-adic integers
ap, p-adic numbers nO and Q^ available for reference, which is
suitable for our purposes, we use this Appendix to set forth
such a representation, notation and some of the properties of
these groups.
1 no and co
The maps V(p'*n) * Z(pr) : x + x(mod pr) (p a fixed prime)
form an inverse system of (discrete) groups. Let Ap = Z(p"),
r = I,2, ; thus nO'is the closed subgroup of the compact
group tt;=r Z(pr) consisting of sequences (*n) such that
Xn = Xn+r(mod Rn).
lO is a topoìogical r'ing (under pointwise multiplication)
and each ideal prn' = {x + * x: pr-times, x e aO} is closed
(recal'l aO is compact) and is the kernel of the honromorphism
aO + Z(pr) , (xn) + x.i hence ptlp is open and the quotient
n'/prn' is isomorphic to Z(pr). In fact the groups prl' form
a neighbourhood base at 0.
The homeomorphism r/ : Z -> Lp: m + (rn(mo¿ pn)) is an injection.
Let x = (xn) . ^p.
Since ú(xr) - X e prlp, ìimn_r-ú(xn) = x;
ttrus ,/r(Z) ìs dense in lO. Simiìarìy pYZ is dense in prn'.
Furthermore, if (m,p) = 1, it is not hard to see that AO + lO :
x + mx is a continuous (and hence topoìogical) automorphism.
Note that the action of Z on lO is compatible with the ring
A
multipì'ication ìn aO, that is mx = x * ... * x = (m)x, in
parti cul ar pfx = ,lr(pr)x = (0, , 0, pFXt , pFxz ) (r zeros ) .
PROPOSITION 1.1. (Serre [43, II Proposition 2] ) . (i) For
æt eLement of A, (z,espeetíueLy Z(pr)) to be inuez,tib\e, it isneeessarA and suffícient that ít is not diuisbiLe ba p.
(¿i), If U denotes the gyoup of inuettíbLe eLements ofLrt euery eLement of Ap can be uritten tmiqueLy in the forrn
pnu uith u e U and. n > 0. (An eLement of lJ is caLLed, a unit).
PR00F_. It is sufficient to prove (i) for Z(pr); the case
of aO wil'l follorv. Now if x e Z(pY) does not belong tct pV,(pY),
its image in V(p) is not zeyo, thus invertible; hence there
exist !¡z e Z(pr) such that xy = 1 - pz, hence
xy(1 + pz + * pr-rzr'r) = 1,
whi ch proves that x 'i s i nverti bl e.
0n the other hand, if x = (xn) e aO is not zero, there
exists a largest integer r such that x. is zero; then x = pru
with u not divisible by p, hence u e U bV (i). The unjqueness
of the decomposition js clear.
PROPOS iTI ON .2 . Each eLosed stbgroup of Lp is aLso an ideaL
ín L, qnd aLL non-zerc íd.eaLs in L, are of the forrn pY AO for some
integer r > 0.
A
PL00t--. Let I be a non-zero ideal 'in nO. Since n prar = {0},
there exists a largest integer r such that prl' ¡ I; letx € I c prl
O be such that x . pr+1Á0, thus x = pru where u
is not cÍivisible by p (and is tlrerefore a unit). It follows
that pr1 - pruu-I e Iu-l g t, I f pra' and so I = ptlp.
Let l-l be a c'losed subgroup of aO. If x e H, then
nx = ú(n). x e H all o e Z-. Since lt(Z) is dense in n,, and
the ring multiplication in nO is continuous, yX € H ally . Ap, thus H is an ideal.
Next, we,define the field of p-adic numbers to be the field of
fractions of lo. Since elements x,x' 6 Ap are uniquely representab'le
in the form x = pmu,X' = pm'u', we have x/xr - orr-fi'(r(r')-I)),so one sees immediate'ly that nO = aO tp-11 and every non-zero
element of nO can be written uniquely in the form pnu with n e Z,
u e U. Addjtion and multipl ication in CIp are are follov,rs: given
two elements in oO, one can write them without loss of generalìty
in the form pmx, pt*sy, where s > 0, ffi e v, xry e tJ., their product
is p2ln+'s*, and their sum pm(x + psy). Endow no r^rith the topology
such that ÂO becomes an open subgroup. hlith this topoìogy CIp
becomes a I ocal ly conrpact topo'logi cal fi el d contai n'i ng Q as a
dense subfield. Finally, nO is metrizable, a.convenient metric
being d(pnt*,pny) = 2-tîin(m'n),rì1,n e Z,x,v € u; the restriction of
d to lO is also a metric.
pObserve that A and Qp are torsion free and nO is divisible.
A
Ap^ and CIp^
Embedd 7(p*) (respectively Z(pr)) the natural way onto a
subgroup ri, = {exp(Z"ik/Rr) : k,r e L} (respectìvely Âr) of '11".
Let  (respectiveìV Âr) have the discrete topology. For fixed
À e ft, the map Z- -> Iy : n + Àn extends to a continuous homomorphism
Âp * Â : À + rx defined as follows: if À e Â,^ ÀX = ÀXF, where
x = (xn). It is easy to check that ix is well defined.
Suppose x. ap ,xf l,thenbecause A ='{x e ÂO: lX(*) -11<6}
is an open set, there exists an integer r such that RraO gR. Itfollows trom álementary considerations that o(ptop) = {l}. Let
r be the smallest integer with this property. This gives rise to
the following commutative diagram of continous maps
2
X->op 1t
txt0
+ z(p")
where h is the canonical projection, o the isornorphism
(xn)Rrno + X" ând x' is necessarily of the form x'(a.) = Àa,
a e Z(pr) for some À € ^r
c ^.
It follows that x(x) = rx.
Converse'ley, for each À e Â,n C ¡, x * lX is a character of lO.
Summing up, we have up^ = ^,
A[Ap^,ptAp] = À" and by the
Pontryagin duality theorem [8,24.8] a pairing (À,x) = ¡X,
ÀeÂ,XeAO.
The method to find nO^ is simjlar; we claim that nO^ is
topologicalìy isomorphic to nO, the pairing being
lo/ptno
A
(pt*,pty) = exp lzni(xy)-r-s/pt*tl,
râs € T.,x,y e U. First we note that for each psx, the
map pry * (pt*,pry) is a (continuous) character of nO. To
show that each character is of this form, let x. ep be non-
zero. As before, we select the smallest jnteger r (possibly
negative) such that *(ptop) = {l}. For not'ional convenjence,
whenever pFx . ep, we deem xn = 0 for n < 0. The epimorphism
0 : np * n : psx + êXp[2nix.-rlpt-s] has kernel pt¡p. Again
we construct a commutative diagram of continuous maps
and using the previous correspondence n^ = Âp, we see that
x(psx) = [0 o h(psx)]V = (psx,p-ry) for some y e U 9 Ap.
Finally, it follows easiìy from the definition of nO^ that the
isomorphism 0p -, np : prx * (prx,.) is bicontinuous.
3. The dual of Q
+'lf
f
np
+h
n /p"ro I n
(
Let x e ú^. For positive integers n, let an =l(l/n!), then
= on_1,fì = 2,3, Conversely, given a sequencen
on)
{cn} c 'lf such that (on)n = on_1, n > 1, then x(m/n!) = (on)t
defines a character of Q ( [18,25" 5] ). Thus Q^ is the projective
limit of the groups Tn = lf, n = 1, ... ; the mappings being
tn + Tn+r : a. J. One of course has to check that the dual
topo'logy on Q^ agrees with the inverse limit topology of the Tn
A
Suppose {cln} e Q^ and onk = L, n = 1, ... Then for
n=!,2, ,1= (onoo)n=onkno=ono-r, thus or., = lalln, so Q^ ts torsion free. We divide elements in Q^ as
follows. Given tcn] . Q^, let ßn = on+m(nfit)(m fixed) where
(nf,t) (m fixed) wher. (nfrt) = (n + m) !/n!m!. The sequence
{ßn} is asain rn rnirrhar i, ßn = rÍl*t)n = on-1nn1Tm¡ = ßn-r.
Also (un)*' = on*rtñitt = on. hle have shown that if x t Q^
corresponds to {crn}, then À = {ßn} is the unique (because
is torsion free) element in Q^ such that im! = x.
L
2
I
9
REFERENCES
D.L. Armacost and W.L. Armacost, rUniqueness in structuretheorems for LCA groupsr, Can. J. Math. 50(1978), 593-599.
L. Auslander and C.C. Moore, rUnitary representations of solvable
lie groupst, Mem. Amer,. Mabh. Soe, 62(t966).
L. Baggett and A. Kleppner, tMultiplier representations of abel.ian
groups', J. Fmct. AnaL" 1.4(1973), 299-324.
3
4
5
6
R.J. Blattner, r0n induced representationst, Arner. J,
83(1961) , 79-98.
Ma.th.
J. Braconnier, 'Sur les groupes topologiques lccalenent compa.ctsr,
J. d.e Math, Puz,es et AppLic. 27(1948), 1-85.
L.G. Brown, rLocally compact abelian groups with trivial nultipliergroup' , J. Fmct. AnaL. 7(I97L) , I32-L39.
,ß
7. J. Dixmier, C -aLgebras (North tlolland 1977) .
E.G. Effros, rTransformation groups and C*-algebrast, AwøLs ofMath. B1(1965), 38-5s.
J. Feldman and F.P. Greenleaf, rExistence of Borel ttansversalsin groupst , Pe"c. J. Math. 25(1968), 455-461.
10. E. Formanek, rThe type I part of the regular representationr,Can. J. Math. 26(L974), 1086-1089.
11. J.M.G. Fe11, rA neu/ proof that nilpotent groups are CCRI, Proe
Amer. MatLt. Soc. L3(1962), 95-99.
72. L. Fuchs, fnfini'oe AbeLian Gz,oups. VoLume f . (Academic Press
1970.
R.0. Pulp and P.A. Griffith, tExtensions of locally compact.
abelían groups. IIt, Trans. At¡pt,. Mabl,¿. Soc. 754(L97L), 357-363.
L3
1,4 R.0. Fulp, 'Splitting loca11y conpact abelian groupsr,
Michigøt Math. J. 1,9(7972), 47-55.
S.A. GaaI, Linean AnaLysis and Representation Theory (Springer
7s73) .
15
16 J. Glinm,
72(t962),
rFarnilies of inducecl representationst , Pac. J. Math
885-911.
19
17. K. Hannabuss, rRepresentations of nilpotent 1ocal1y compact
groups', J. Funct. AnaL. 34(7979), 146-765.
18. E. Hewitt and K.A. Ross, Absbraet Hamnonic Analgsis, I(Springer 1963).
H. Heyer,' rDualität lokatkompakter Gmppent , Lectute Notes inMath. 1s0(1970).
20. A.K. Holzherr, rDiscrete groups whose multiplier representationsare type I', J. AustraL. Math. Soc, (Sev"Les A) 31(1981), 486-495.
2T I.M. Isaacs and D.S. Passman, rGroups with representations ofbounded degreer, Can. J. Math. 16(1964), 299-309.
22. E. Kaniuth, rDer Typ der regulären Dar.stellung diskreter Gruppenr,
Math. Ann. 182(1969) , 334-339.
23. E. Kaniuth, rDie Struktur der regulären Darstellung lokalkornpakter
Gruppen nit invariarrter Umbegungsbasis der Eins t , Math, Ann.
194(1971) , 225-248.
A. Kleppner_. rT'he structure of some induced representationsr,DLtl<e Mq.th. J. 29(7962), 555-572.
A. Kleppner, rMultipliers on abelian groupst, Math. Ann. 158(1965),
lL-34.
A. Kleppner, rContinuity and rneasurability of rnultiplier and
projective representations' , J. Etmct. AnaL, L7(L974), 2L4-226.
27. G.W. Mackey, rlnduced representations of locally conpact groups. Ir.AnnaLs of Math. 55(1952), 101-139.
24
25
26
28. G.W. Mackey, rUnitary representations of group extensions. I.lActa Math. 99(1958) , 265-377.
29 G.W. Mackey, rlnduced representations and normal subgroupsr,
Procee&Lngs of the Intez,nntíonaL Symposiwn on Linear Spaces
(Perganon Press 1961) , 379-326.
30 G.W. Mackey, The Theory of Unítaz'y Gz,oup Representations (The
University of Chicago Press , 7976).
37" C.C. Moore, rGroups rvith finite dirnensional irreducible 'representationsr, Tv,ans. Amer. Math. Soe. 166(t972), 401-410.
32. B.H.
Pt'oc
Neunann, I Groups wj-th finite classes of conjugate elenents I ,
Lond. Math. Soc. 1 (1951), 778-787.
D. Poguntke, tZwei Klassen lokalkompakter: naximal fastperiodischerGruppent, Monatsh. fw Math. 81(1976), 15-40.
M. Rajagopalan and T, Soundararajan, tstr-uctu:re of SeIf-dua1 torsion-free metric LCA groupst, Ftmd. Math. 65(1969), 309-316.
J.R. Ringrose, tlectures on the trace in a finíte von Newma:rn
algebrat , Lectuïe Notes ín Math. 247(L970/L).
L.C. Robertson, rA note on the structure of Moore groupsr,
BULL. Amer. Mafh. Soc. 75(1969), 594-599.
L.H. Rowen, PoLynonrLal Iden'bLties in Rdng Theoz,y (Academíc Press
1980) .
S. Sakai, C -aLgebras Øtd W -aLgebras (Springer 797I).
33. B.H. Newrnann, rGroups with finite classes of conjugate subgroupsr,
Math. zeit. 63(19ss) , 76-96.
34. G.K. Pedeïsen, C*-ALgebras arld. theiz, Aubomorpltism Grotrys
(Academic Press 1979).
35.
36.
37.
38.
59.
40.,t ,t
47. G. Schlichting, tEine Cirarakierisier-ung gewisser diskreter Gruppen
durch ihre reguläre Darstellungr , Møtusenipta Math. 9(1973), 589-409.
42
43
44
45
46
47.
48.
49.
G. Schlichting, t Polynornidentitäten und Permutationsdarstellungen
lokalkompakter Gruppen t, Inuentiones Math. 55(1979), 97-L06.
J.-P. Serre , A Cowse in Ar|thnetic (Springer L973).
M. Smith, rGroups algebras' , J. ALg. 1B(l-971) , 477-499 .
M. Snith, 'Regular representations of discrete groupst , J. Pmct.
AnaL. 77(1972) , 40L-406.
M. Takesaki, 1A characterization of group algebras as a converse
of Ta¡inaka-stinespring-Tatsuuma duality theoremt, Amen. J. Math.
91(1969) , s29-s64.
M. Takesaki, Theory of Openator Algebras. f. (Springer 1979)
K.F. Taylor, rThe type structure of the regular representation ofa locally compact groupt, Math. Awt. 222(1976), 27I-224.
E. Thona, tEine Characterisierung diskreter Gruppen von
Typ It , fnuenl;iones Math. 6(1968), 190.-196.
50. N. Ya. Vilenkin,Amer. Math. Soc.
'Direct deconpositions of topological groups IIz,urtsLations (SerLes f ) 8(1962), 78-185.
II"
51. A. Weil, L'Integraticn døts Les Gz'oupes TopoLogt)ques et ses
AppLicatíons (Herman 1938) .