pacific journal of mathematics

Vol. 184, No. 2, 1998

C

FUNCTIONS ON THE CANTOR SET, AND

A SMOOTH m-CONVEX FR

ECHET SUBALGEBRA OF O
2
Larry B. Schweitzer
We construct a nuclear, spectral invariant, dense Frechet
subalgebra C

(K) of the commutative algebra C(K) of con-

tinuous complex valued functions on the Cantor set K. The
construction uses the group structure of the 2-adic integers
on K.
We then use a smooth crossed product construction to get
a dense, nuclear Frechet subalgebra O
2
of the Cuntz algebra
O
2
. We prove the general result that a tempered action of a
locally compact group on a strongly spectral invariant dense
Frechet subalgebra of a Banach algebra is automatically m-
tempered, and obtain the m-convexity of O
2
as a special case.
0. Introduction.
Dense subalgebras of C*-algebras are well-known to be useful in the study
of C*-algebras. In the current literature, a dense subalgebra of smooth
functions is often viewed as a replacement for C

-functions on a manifold,
where instead of a manifold we have an underlying noncommutative space.
In this paper, we seek not just a dense subalgebra of smooth functions of
the C*-algebra O
2
, but an algebra with a Frechet topology, for which the
seminorms are submultiplicative. The Frechet topology is very helpful when
it comes to working with a dense subalgebra. It allows the use of functional
calculus, and more importantly allows the dense subalgebra to be written as
a countable projective limit of Banach algebras. By taking projective limits,
interesting results about Frechet algebras can be deduced from corresponding
results on Banach algebras (for example, see [Da], [PhSc]).
In 1, we dene a set of smooth functions on the Cantor set C

(K)
for each increasing sequence = c
p

p=1
of positive real numbers c
p
. This
is done by identifying K with the topological group of 2-adic integers [Ko].
Using this group structure on K, we dene C

(K) to be the inverse Fourier

transform of the set of -rapidly vanishing Schwartz functions o

K) on the
dual group

K. (The dual group

K = Z[
1
2
]/Z is the discrete group of dyadic
rationals lying between 0 and 1.) If satises the summability condition

p=1
2
p
/c
q
p
< for some q > 0, then C

(K) is nuclear as a Frechet space,

349
350 LARRY SCHWEITZER
and is a strongly spectral invariant dense Frechet subalgebra of C(K) (see
Denition 1.4). The existence of a subalgebra of C(K) with these properties
was a surprise to me, since the Cantor set is totally disconnected, and seems
to have no smooth structure. Note that for dierent choices of , we get
apparently dierent sets of functions C

(K).
In 2, we restrict to the case c
p
= 2
p
or = 2
p

p=1
(which satises
the above summability condition for q > 1), and show that the action of the
free-product of cyclic groups Z
2
Z
3
on C(K) in [Sp] leaves C

(K) invariant
and is exponentially tempered with respect to the word length function on
Z
2
Z
3
(Theorem 2.3). Thus the set of exponentially rapidly vanishing
functions (see Denition 2.2) o
e
(Z
2
Z
3
, C

(K)) is a nuclear Frechet algebra

(Corollary 2.4). By [Sp] (Theorem 2.1 below), the corresponding reduced
C

crossed product is isomorphic to the Cuntz algebra O

2
, so our Frechet
algebra is a dense subalgebra of O
2
, which we will call O
2
for short.
In 3 we show that a tempered action of a locally compact group H,
with weight or gauge (see 1, 2), on a strongly spectral invariant dense
Frechet subalgebra A of a Banach algebra (Denition 1.4) is automatically
m-tempered. Thus the smooth crossed product o

(H, A) is m-convex, and

as a special case so is the smooth crossed product O
2
we constructed above.
In 4, we show that the action of Z
2
Z
3
on C

(K) is not tempered with

respect to the word length function on Z
2
Z
3
. Thus it is necessary to use
functions which vanish exponentially rapidly with respect to to form the
smooth crossed product, as we did above. In 5, we show that the algebra
O
2
has an element with unbounded spectrum, from which it follows that
O
2
is not spectral invariant in the C

-algebra O
2
. (Hence it follows that
O
2
cannot be a set of C

-vectors for any group action on O

2
(otherwise it
would be spectral invariant by [Sc 2], Theorem 2.2), which is why we prefer
to use the notation O
2
instead of O

2
.)
1. Smooth Functions on the Cantor Set.
We view the Cantor set K as innite sequences of 0s and 1s (e.g. .00101 . . . ),
topologized as the product space

N
0, 1. One may view this compact
topological space as a disconnection of the unit circle T = R/Z, where
elements of T are written in their binary expansions. Note, for example,
that .0111 . . . is not equal to .1000 . . . in K.
The group operation on K is given by carrying down instead of up.
For example, .0100 + .0100 = .0010 . . . , and .01100 + .10100 =
.11010 . . . . (Note that if .k
0
k
1
k
2
. . . is a binary sequence in K, then k
0
+
2k
1
+2
2
k
2
+. . . denes an innite power series, with 2 as the indeterminant
variable. Then addition is equivalent to carrying down, since 2
n
+ 2
n
=
C

FUNCTIONS ON THE CANTOR SET 351

2
n+1
. In fact, this is precisely the group structure of the 2-adic integers on
K [Ko].) It is easy to check that this addition is associative, .000 . . . acts as
the identity, and every element has an inverse. (K, +) is clearly an Abelian,
compact topological group, which I will refer to as the Cantor group.
Remark. If you dene addition by carrying up as you would for usual
binary expansions in T, then .111... has no inverse. (The inverse would have
to be .000 . . . 1, with the 1 at innity.)
If s is a nite sequence of 0s and 1s, we let .s denote the subset of K
of all innite sequences beginning with s. We normalize the Haar measure
on K so that (K) = 1. Then (.0) = (.1) so they both equal 1/2.
Similarly, (.00) = (.01) = (.10) = (.11) = 1/4, etc. Haar measure
is the natural product measure on K.
We denote the dual group of the Cantor group by G =

K. Then G is the
discrete group of dyadic rationals from 0 to 1 (i.e. G = Z[
1
2
]/Z). G can be
represented pictorially as a pyramid - see Figure 1.1.
0 1 lev = 0
1/2 lev = 1
1/4 3/4 lev = 2
1/8 3/8 5/8 7/8 lev = 3
.
.
.
.
.
.
.
.
.
.
.
.
Figure 1.1. G =

K, dual of Cantor group.
If g G can be written g = l/2
p
, l odd, then we dene lev(g) = p, the
level of g. (This function is the same as the base 2 logarithm of the non-
archimedian 2-norm on Q [Ko].) Note that there are 2
p1
gs in G with
lev(g) = p. When two elements of G are added, the sum lies at the same
level of the pyramid or above. Thus lev(g + h) max(lev(g), lev(h)).
The pairing between G and K is given by
(g,

k) e
2ig,

k
,
where
g,

k) = g(k
0
+ 2k
1
+ 2
2
k
2
+ . . . ) = (l/2
p
)(k
0
+ 2k
1
+ 2
2
k
2
+ . . . ) (mod Z).
This makes sense since g2
n
is an integer if n p. (To see that G is really
the dual group, let K
n
be the subgroup .0 . . . 0 of K, where there are n
zeros. Then lev(g) n if and only if g denes a trivial character on K
n
,
i.e. g factors to a character of K/K
n

= Z
2
n. Since there are 2
n
gs with
352 LARRY SCHWEITZER
lev(g) n, this gives all of

Z
2
n. Exercise: show that every character of K
factors through some K
n
.)
A gauge or length function on a locally compact group H is a Borel
measurable function : H [0, ) which satises (id
H
) = 0, (g
1
) = (g),
and (gh) (g)+(h) for g, h H. If = c
p

p=1
is an increasing sequence
of positive numbers, then we may dene a gauge (also denoted by ) on G
by
(g) =
_
c
lev(g)
if lev(g) 1,
0 if lev(g) = 0.
(Note lev(g) = 0 i g = 0.)
Lemma 1.2. The group G has polynomial growth, and = c
p

p=1
satises
the summability condition

p=1
2
p
c
q
p
< ,
for some q > 0, if and only if

gG{0}
1
(g)
q
< .
Proof. Note that if U G is nite, then

n=0
U
n
is contained in a nite
sub-pyramid, so [U
n
[ M < for all n, and G has polynomial growth.
For the second part,

g=0
1
(g)
q
=

p=1
_
_

lev(g)=p
1
(g)
q
_
_
=
1
2

p=1
2
p
c
q
p
,
where we used [g G [ lev(g) = p[ = 2
p1
.
Dene o

(G), the -rapidly vanishing functions on G, by

o

(G) = : G C

||
d
< , d = 0, 1, 2, . . . ,
where
(1.3) ||
d
= |
d
|
1
=

gG
(g)
d
[(g)[.
Then o

(G) is a Frechet -algebra under convolution [Sc 1], Theorem 1.3.2.

(In fact, it is a strongly spectral invariant (see below) dense Frechet subal-
gebra of the Banach algebra L
1
(G) by [Sc 2], Theorem 6.7.)
Denition 1.4. We say that a Frechet algebra A is m-convex if there
exists a family | |
n

n=0
of topologizing seminorms for A which are sub-
multiplicative:
|ab|
n
|a|
n
|b|
n
, a, b A.
C

FUNCTIONS ON THE CANTOR SET 353

A is a Frechet -algebra if A has a continuous involution dened on it. Let
A be a dense Frechet subalgebra of a Banach algebra B [Sc 2], Denition
1.1. If A is nonunital, let

A and

B denote the respective unitizations. Then
A is spectral invariant (SI) in B if every a

A is invertible in

B if and only
if a is invertible in

A. A is strongly spectral invariant (SSI) in B if there is
some C > 0 such that for every m, there is some p
m
m and D
m
> 0 such
that
|a
1
. . . a
n
|
m
C
n
D
m

|a
1
|
k1
. . . |a
n
|
kn
,
for all n-tuples a
1
, . . . a
n
A and all n, where the sum is over those k
i
s such
that

n
i=1
k
i
p
m
, and | |
0
denotes the norm on B. Then SSI SI
by [Sc 2], Proposition 1.7, Theorem 1.17. If a Frechet algebra A is strongly
spectral invariant in some Banach algebra, then A is automatically m-convex
[Sc 2], Proposition 1.7.
Recall that by basic locally compact Abelian group theory [Ru], we have
an isomorphism of C

-algebras C(K)

= C

(G) given by the Fourier trans-

form
C(K) (g) =
_
K
(

k)e
2ig,

k
d

k.
Thus we may dene C

(K), the -C

-functions on the Cantor set, to be the

inverse image of o

(G) in C(K). Then C

(K) is automatically a Frechet

algebra under pointwise multiplication, since o

(G) is a Frechet algebra

under convolution.
We give an alternative way of describing C

(K), which is closer to the

usual denition of a C

-function on the circle T, for example.

Denition 1.5. Let = c
p

p=1
be an increasing sequence of positive
numbers. For C(K) and p N, dene the pth approximate derivative

p
() C(K) by

p
()(

k) =
(

k + 1/2
p
) (

k)
1/c
p
.
Here 1/2
p
= .0 . . . 01 K is the binary expansion for 1/2
p
, so the 1 is in the
pth spot. (If we were using 2-adic notation, 1/2
p
= .0 . . . 01 would be written
2
p
. In keeping with the analogy of K with the circle group T, we will stick
to the binary notation.) Note that 1/2
0
= 1 = 0 in K, so (

k+1/2
0
) = (

k)
and it is appropriate to dene
0
() = 0 for all . It is easy to check from
the denition that

p
() =
p
() +
p
() +
p
()
p
()/c
p
,
so
p
is not quite a derivation.
354 LARRY SCHWEITZER
It is well known that the -subalgebra of linear combinations of cylinder
functions
/ = span
.s...
[s a nite sequence of 0

s and 1

s
is dense in C(K). (Use the Stone-Weierstrass theorem.) It is also easy to
check that under the Fourier transform, / corresponds to the nite support
functions c
f
(G) o

(G).
Theorem 1.6. Assume that = c
p

p=1
satises the summability condition
of Lemma 1.2. Then the Frechet algebra C

(K) is the completion of / in

the seminorms
(1.7) ||
d
=

p=1
2
p
|
d
p
()|

.
Moreover, o

(G)

= C

(K) are nuclear as Frechet spaces. They are m-

convex Frechet -algebras, strongly spectral invariant in their respective C

-
algebras.
Remark 1.8. 1) It is interesting that lim
p

p
() = 0 for /, so the
derivative in the usual sense is always zero. (Note that for xed g G,

p
(e
2ig,

k
) is the zero function in C(K) if and only if lev(g) < p.) Hence it
would have been impractical, for example, to take sup norms of derivatives
in (1.7), since the seminorms would all be zero on /, and we could not have
got a topology equivalent to the one from o

(G).
2) In the terminology of [JiSc], (G, ) is a rapid decay group. In fact
o

2
(G) = o

1
(G) L
1
(G) by the summability condition of Lemma 1.2, and
[Sc 1], Theorem 6.8. It follows that o

(G) is a SSI dense Frechet subalgebra

of C

(G) = C

r
(G) [JiSc], Denition 1.5, Lemma 3.11, Proof of Theorem
2.6(b).
Question 1.9. What is the relationship between C

(K) and the subalge-

bra of C(K) that you get by identifying K with [0, 1]the middle thirds,
and then restricting the C

-functions C

[0, 1] to K? Note that in our dif-

ference quotients, ((

k + 1/2
p
) (

k))/(1/c
p
), we are using a nonstandard
addition, so one might expect them to be dierent.
Proof of Theorem 1.6. By Lemma 1.2 and [Sc 1], Theorem 6.24, o

(G) is
a nuclear Frechet space, and is SSI in C

(G) as noted in Remark 1.8 (2)

above. The strong spectral invariance property implies m-convexity [Sc 1],
Proposition 1.7, or apply [Sc 1], Theorem 1.3.2 to get both the m-convexity
and -algebra statements for o

(G). Since C

(K)

= o

(G) as Frechet
C

FUNCTIONS ON THE CANTOR SET 355

algebras (by denition of C

(K)), it remains only to prove the rst as-

sertion of the theorem. Let g G have lev(g) = p for p 1. Then
g, 1/2
p
) =
1
2
p
2
p1
(mod Z) = 1/2 (mod Z). Since e
i
= 1, we
have (e
2ig,1/2
p

1) = 2. Let C(K). Then

(g)
d
(g) = (g)
d
_
K
(

k)e
2ig,

k
d

k
= c
d
p
_
K
(

k)e
2ig,

k
d

k
=
c
d1
p
2
_
K
(

k)
_
e
2ig,

k1/2
p

e
2ig,

k
1/c
p
_
d

k
=
c
d1
p
2
_
K
(

k + 1/2
p
) (

k)
1/c
p
e
2ig,

k
d

k
=
c
d1
p
2

p
(g) = =

d
p
(g)
(2)
d
.
The same equality clearly holds if lev(g) = p = 0, since both sides are
zero. So multiplication by (g)
d
in o

(G) is the same as applying

d
p
/(2)
d
in C

(K), and then evaluating the Fourier transform at g. (Note that p

depends on g, though.) Using this, we have
|
d
|
1
=

p=0
_
_

lev(g)=p
[(g)
d
(g)[
_
_
=

p=0
_
_

lev(g)=p

(
d
p
)(g)
(2)
d

_
_

p=0
2
p
2
d+1
_
K
[(
d
p
)(

k)[d

k
1
2
d+1
_

p=0
2
p
|
d
p
|

_
=
1
2
d+1
||
d
.
Thus the topology on /, given by (1.7), is at least as strong as the topology
induced from o

(G). We show that the topologies are equivalent. By the

Fourier inversion formula [Ru],
(

k) =

gG
e
2ig,

k
(g) =

lev(g)<p
e
2ig,

k
(g) +

lev(g)p
e
2ig,

k
(g).
Since
p
(e
2ig,

k
) = 0 if (and only if) lev(g) < p, we have
[(
d
p
)(

k)[ =

d
p
_
_

lev(g)p
e
2ig,

k
(g)
_
_

lev(g)p

d
p
_
e
2ig,

k
_
(g)

,
356 LARRY SCHWEITZER
since the series converges absolutely. Since
d
p
(e
2ig,

k
) =
p
(g)
d
e
2ig,

k
,
where
p
(g) = (e
2ig,1/2
p

1)c
p
, it follows that
[(
d
p
)(

k)[

lev(g)p
(2c
p
)
d
[ (g)[.
Thus
||
d
=

p=1
2
p
|
d
p
()|

p=1
2
p
_
_

lev(g)p
(2c
p
)
d
[ (g)[
_
_
= 2
d

g=0

d
(g)[ (g)[ 2
d
|
d
|
1
,
where
d
(g) =

lev(g)
p=1
2
p
c
d
p
. But

d
(g)
_
_
lev(g)

p=1
2
p
c
q
p
_
_
c
d+q
lev(g)
since the c
p
s are increasing

p=1
2
p
c
q
p
_
(g)
d+q
= C(g)
d+q
, by denition of
where C < by our summability assumption on . Thus ||
d

2
d
C|
d+q
|
1
, completing the proof of Theorem 1.6.
Remark 1.10. Theorem 1.6 is analogous to the well-known isomorphism
C

(T)

= o(Z). Note that the formula (2(g))
d
(g) =

d
p
(g) for
C

(K) obtained in the proof is similar to the formula (2in)

d
(n) =

(d)
(n),
n Z, for C

(T), obtained using integration by parts.

2. A tempered action of Z
2
Z
3
.
Recall that the free product Z
2
Z
3
is a
1
b
j1
ab
j2
. . . b
jn
a
2
[j
i
= 1,
i
=
0 or 1, with the obvious group multiplication. Here a, b are the gener-
ators of the cyclic groups Z
2
, Z
3
respectively. The word length function
corresponding to the generating set U = a, 0, b, b
1
is (a
1
b
j1
. . . b
jn
a
2
) =
2n 1 +
1
+
2
. Then is a gauge on Z
2
Z
3
(1).
A weight on a locally compact group H is a Borel measurable function
: H [1, ) which satises (id
H
) = 1, (g
1
) = (g), and (gh)
(g)(h) for all g, h H. We let e

denote the exponentiated word weight

on Z
2
Z
3
[Sc 1], Example 1.1.17. Then e

is easily seen to be a weight on

Z
2
Z
3
.
C

FUNCTIONS ON THE CANTOR SET 357

We may think of the Cantor set K as innite words in a and b as follows:
K = a
k0
b

k1
ab

k2
a . . . [

k K,
where

k
i
= 1 if k
i
= 0, and

k
i
= +1 if k
i
= 1. Then Z
2
Z
3
acts on K on the
left, and we may form the reduced C

-crossed product C

r
(Z
2
Z
3
, C(K)).
Theorem 2.1 [Sp], [Ch]. C

r
(Z
2
Z
3
, C(K))

= M
2
(O
2
)

= O
2
, where O
2
denotes the n = 2 Cuntz C

-algebra.
Denition 2.2. If a locally compact group H with weight or gauge acts
on a Frechet algebra A, we say that the action is -tempered if for every
m N, there exists C > 0, p, d N such that
|
h
(a)|
m
C(1 + (h))
d
|a|
p
, a A, h H.
By [Sc 1], Theorem 2.2.6, o

(H, A) is a Frechet algebra under convolution

if the action is -tempered. o

(H, A) is also dense in L

1
(H, B) and in the
reduced C

-crossed product C

r
(H, B), if A is a dense subalgebra of a C

-
algebra B. Throughout this paper, the smooth crossed product o

(H, A)
will denote L
1
-rapidly vanishing functions from H to A [Sc 1], 2.1. We
will abbreviate o
e

by o
e
. The following theorem is the main result of this
paper.
Theorem 2.3. Let = c
p

p=1
, where c
p
= 2
p
. Then the action of Z
2
Z
3
on C(K) dened above leaves C

(K) invariant and is e

-tempered.
Corollary 2.4. o
e
(Z
2
Z
3
, C

(K)) is a dense Frechet -subalgebra of

C

r
(Z
2
Z
3
, C(K))

= O
2
, and is also nuclear as a Frechet space.
Proof of Corollary 2.4 from Theorem 2.3. Apply remarks preceding The-
orem 2.3 to see that it is a Frechet algebra. By [Sc 1], Corollary 4.9 it is
also a Frechet -algebra. Since = 2
p
satises the summability condition
of Lemma 1.2, C

(K) is nuclear by Theorem 1.6. Then by [Sc 1], Propo-

sition 6.34, Theorem 6.24, Proposition 6.13 (1), so is the smooth crossed
product.
Proof of Theorem 2.3. This is equivalent to showing that the induced action
of Z
2
Z
3
on o

(G) is e

-tempered:

()(g) =
_
K
(

)(

k)e
2ig,

k
d

k,
where C

(K), = . See (1.3) for the seminorms we will be using for

o

(G).
358 LARRY SCHWEITZER
Lemma 2.5 (Truncation of lower levels). For any specied level p N,
and for all d N and o

(G),
|
d
|
1
2
(d+1)p
||

+

lev(g)p
[
d
(g)[,
where C

(K) is the inverse Fourier transform of .

Proof.

lev(g)<p
[(g)
d
(g)[
=
p1

r=1

lev(g)=r
2
rd
[(g)[
p1

r=1

lev(g)=r
2
rd
||

=
p1

r=1
2
r1
2
rd
||

= (2
p(d+1)1
1)||

2
p(d+1)
||

.
For Z
2
Z
3
, I will let denote the set of words in K which begin with
. Written in terms of 0s and 1s, we have for example a = .1. I will let
a denote the complement of this set. Here are a few more examples: ba =
b = .01. Note that since b indicates a reduced word beginning with b,
there must be an a after the b. Similarly abab
1
a = abab
1
= .110.
The action of a on K is measure preserving, and in particular (a) =
() for all . However, the action of b does not leave the Haar measure on
K invariant: If begins in b or b
1
, or = 0, we have (b
j
a) = 1/2(a),
(b
j
b
j
a) = (b
j
a), (b
j
b
j
a) = 2(b
j
a). So we have the
following integral formulas (where is any function in L

(K)):
_
a
(b
j

k)(

k)d

k = 2
_
b
j
a
(

k)(b
j

k)d

k (2.6)
_
b
j
a
(b
j

k)(

k)d

k =
1
2
_
a
(

k)(b
j

k)d

k (2.7)
_
b
j
a
(b
j

k)(

k)d

k =
_
b
j
a
(

k)(b
j

k)d

k. (2.8)
If S is any measurable set, then nally:
(2.9)
_
S
(a

k)(

k)d

k =
_
aS
(

k)(a

k)d

k.
Lemma 2.10. If = b
j1
ab
j2
a . . . b
jn
a Z
2
Z
3
and (

k) = e
2ig,

k
, then

() =
_
K
(
1

k)(

k)d

k = 2
n
_

1
a
(

k)(

k)d

k
C

FUNCTIONS ON THE CANTOR SET 359

+
1
2
n
_
a
(

k)(

k)d

k (2.11)
+
n

i=1
2
(ni)
2
(i1)
_
ab
jna...b
j
i+1
ab
j
i
(

k)(

k)d

k.
Proof. Use induction on n, and the formulas (2.6)-(2.9).
From now on, we let = b
j1
ab
j2
a . . . b
jn
a. Note that n = ()/2. We will
estimate |
d

()|
1
by breaking

() up as in Lemma 2.10. We proceed

by applying

gG
(g)
d
to the absolute value of each of the n + 2 terms in
(2.11).
For the rst term, note that if

k = .
1
ak
n+1
k
n+2
. . . , then g,

k) = g(1+
2k
n+1
+ 2
2
k
n+2
+. . . ) = g/2
n
(2
n
+ 2
n+1
k
n+1
+ 2
n+2
k
n+2
+. . . ) = g/2
n
,

k)+
terms depending only on , but not

k. Thus we have
2
n

gG
(g)
d

1
a
(

k)e
2ig,

k
d

= 2
n

gG
(g)
d

1
a
(

k)e
2ig/2
n
,

k
d

(2.12)
2
n

lev(g)>n
(2
n
g)
d

1
a
(

k)e
2ig,

k
d

= 2
dn+n

lev(g)>n
(g)
d

1
a
(

k)e
2ig,

k
d

2
(d+1)n

gG
(g)
d

1
a
(

k)e
2ig,

k
d

,
where the second step used the fact that g g/2
n
is a bijection between G
and g G [ lev(g) > n, g <
1
2
n
0, the third step used (2
n
g) = 2
n
(g),
and the last step is to make the estimate (2.16) below work more smoothly.
Next we estimate the second term in the expression (2.11). Note that if

k = .0k
1
k
2
. . . , then

k = .0b
j1
. . . b
jn
k
1
k
2
. . . . Hence g,

k) = g(2
n+1
k
1
+
2
n+2
k
2
+ . . . ) = 2
n
g,

k)+ terms depending only on , but not

k. Thus we
have
1
2
n

lev(g)n
(g)
d

_
a
(

k)e
2ig,

k
d

=
1
2
n

p=n

lev(g)=p
(g)
d

_
a
(

k)e
2i2
n
g,

k
d

(2.13)
360 LARRY SCHWEITZER
=
1
2
n

p=n

lev(g)=pn
2
n
(g/2
n
)
d

_
a
(

k)e
2ig,

k
d

2
n(d+1)

gG
(g)
d

_
a
(

k)e
2ig,

k
d

,
where the 2
n
in the third expression takes into account the repeats in the
1st and 2nd expression, and the last step used (g/2
n
) = 2
n
(g). (By using
the truncation of lower levels Lemma 2.5 above, we will be able to get by
with only summing over lev(g) n in the left hand side of (2.13).)
Next we estimate terms in the last expression of (2.11). Fix some i be-
tween 1 and n. Any

k ab
jn
a . . . b
ji+1
ab
ji
is of the form

k = ab
jn
a . . .
b
ji+1
ab
ji
k
ni+2
k
ni+3
. . . . Since = b
j1
a . . . b
jn
a, we have

k = b
j1
a . . .
b
ji1
ab
ji
k
ni+2
k
ni+3
. . . . Thus g,

k) = g(2
i+1
k
ni+2
+2
i+2
k
ni+3
+. . . ) =
g2
2i1n
(2
ni+2
k
ni+2
+ . . . )+ terms depending only on , but not

k. We
split into two types of terms - those like (2.12) and those like (2.13). We
have:
Case 1. 2i 1 n 0. Imitating (2.12), we get
2
(ni)
2
(i1)

gG
(g)
d

_
ab
jn...b
j
2ab
j
1a
e
g,

k
d

(2.14)
= 2
n+12i

gG
(g)
d

_
ab
jna...b
j
2ab
j
1
(

k)e
2ig/2
n+12i
,

k
d

2
n+12i

lev(g)>n+12i
(2
n+12i
g)
d

_
ab
jna...b
j
2ab
j
1
(

k)e
2ig,

k
d

= 2
(1d)(n+12i)

lev(g)>n+12i
(g)
d

_
ab
jna...b
j
2ab
j
1
(

k)e
2ig,

k
d

2
(d+1)n

gG
(g)
d

_
ab
jna...b
j
2ab
j
1
(

k)e
2ig,

k
d

.
Case 2. 2i 1 n > 0. Note that since i n, 2i 1 n < n. Imitating
(2.13), we get
2
(ni)
2
(i1)

lev(g)n
(g)
d

_
ab
jn...b
j
2ab
j
1
e
g,

(2.15)

1
2
2i1n

lev(g)2in1
(g)
d

_
ab
jn...b
j
2ab
j
1
e
g2
2i1n
,

FUNCTIONS ON THE CANTOR SET 361

=
1
2
2i1n

gG
2
(2i1n)
(g/2
2i1n
)
d

_
ab
jna...b
j
2ab
j
1
(

k)e
2ig,

k
d

= 2
(2i1n)d

gG
(g)
d

_
ab
jna...b
j
2ab
j
1
(

k)e
2ig,

k
d

2
(d+1)n

gG
(g)
d

_
ab
jna...b
j
2ab
j
1
(

k)e
2ig,

k
d

.
Now we are ready to collect terms and do the nal estimate. By Lemma
2.5, Lemma 2.10 and (2.12)-(2.15), we have
|
d

()|
1
(2.16)
2
(d+1)n
||

+

lev(g)n
[
d

()(g)[
2
(d+1)n
_
||

gG
(g)
d
_

1
a
e
g,

_
a
e
g,

+
n

i=1

_
ab
jn...b
j
1a
e
g,

__
2
(d+1)n
_
||

p=0
1
2
d

lev(g)=p
_

1
a
(
d
p
)e

_
a
(
d
p
)e

+
n

i=1

_
ab
jna...b
j
1
(
d
p
)e

__
2
(d+1)n
_
||

p=1
2
p
2
d+1
|
d
p
|

_
2
(d+1)n
(||

+||
d
)
= 2
(d+1)()/2
C
d
_
||
1
+|
d+q
|
1
_
C
d
e
rd()
_
||
1
+||
d+q
_
,
where the third step used the integration by parts formula (2(g))
d
(g) =

d
p
(g) of Theorem 1.6, and the fourth step used the fact that there are 2
p1
elements of G at level p, and (K) = 1. This proves Theorem 2.3 for all
beginning in b or b
1
, and ending in a. Now show that a by itself acts
temperedly (i.e. |
a
()|
d
C

d
||
d+q
), and then get the result for general
by composing a on either side of .
3. m-convexity of the smooth crossed product.
By Corollary 2.4, we know that o
e
(Z
2
Z
3
, C

(K)) is a Frechet algebra

under convolution. We say that an action of a group H on a Frechet algebra
362 LARRY SCHWEITZER
A is m--tempered for some weight or gauge on H, if there exists a family
of submultiplicative seminorms | |
n
(Denition 1.4), such that for every
n, there exists C > 0, and d N such that
|
h
(a)|
n
C(1 + (h))
d
|a|
n
, a A, h H.
If the action is m--tempered, then the smooth crossed product o

(H, A) is
an m-convex Frechet algebra [Sc 1], Theorem 3.1.7.
Theorem 3.1. Let A be a dense Frechet subalgebra of a Banach algebra
B. Assume that a locally compact group H acts isometrically on B and -
temperedly on A, where is a weight or gauge on H. Then if A is strongly
spectral invariant in B, it follows that H acts m--temperedly on A.
Proof. If is a gauge, we may replace with the equivalent weight 1 +
so that 1 [Sc 1], 1. Let | |
n

n=0
be some increasing family of
seminorms for A, where | |
0
is the Banach algebra norm on B. Recall that
by [Sc 1], Theorem 3.1.18, it suces to show that for every m N, there
exists q, d N and D > 0 such that
|
h1
(a
1
) . . .
hn
(a
n
)|
m
(3.2)
D
n
_
(h
1
)(h
1
1
h
2
) . . . (h
1
n1
h
n
)
_
d
|a
1
|
q
. . . |a
n
|
q
,
for all n-tuples h
1
, . . . h
n
N, a
1
, . . . a
n
A, and all n. By the strong
spectral invariance,
(3.3)
|
h1
(a
1
) . . .
hn
(a
n
)|
m
C
n
D
m

k1+...knp
|
h1
(a
1
)|
k1
. . . |
hn
(a
n
)|
kn
.
In the sum, if k
i
= 0, then |
hi
(a
i
)|
ki
= |a
i
|
ki
since the action is assumed
isometric on B. If k
i
,= 0, then |
hi
(a
i
)|
ki
C

(h
i
)
d

|a
i
|
p
since the action
is -tempered on A. Since the k
i
s are p, there is some upper bound for
the C

, d

, and p

s - call it q. (Note we consider m xed as we let n run, so

p is also xed.) Then by (3.3), and since the seminorms are increasing, we
have
|
h1
(a
1
) . . .
hn
(a
n
)|
m
(3.4)
C
n
D
m
q
p
|a
1
|
q
. . . |a
n
|
q

k1+...knp
_
(h
i1
) . . . (h
ip
)
_
q
,
where i
1
, . . . i
p
include all indices for which k
i
,= 0.
Note that
(h
i
) (h
1
i1
h
i
)(h
1
i2
h
i1
) . . . (h
1
1
h
2
)(h
1
)
C

FUNCTIONS ON THE CANTOR SET 363

_
(h
1
)(h
1
1
h
2
) . . . (h
1
n1
h
n
)
_
since 1 and is submultiplicative. Also

k1+...knp
(1) p
n
so by (3.4),
we have
|
h1
(a
1
) . . .
hn
(a
n
)|
m
(Cp)
n
D
m
q
p
_
(h
1
)(h
1
1
h
2
) . . . (h
1
n1
h
n
)
_
q
|a
1
|
q
. . . |a
n
|
q
.
This clearly gives (3.2).
Corollary 3.5. Let A, B, H, and be as in Theorem 3.1. Then the set
of -rapidly vanishing L
1
-functions from H to A forms an m-convex dense
Frechet subalgebra o

(H, A) of L
1
(H, B). In particular, o
e
(Z
2
Z
3
, C

(K))
is an m-convex Frechet algebra.
Proof. The rst statement follows from the m--temperedness of the action,
and [Sc 1], Theorem 3.1.7. For the second statement, the action of Z
2
Z
3
is e

-tempered by Theorem 2.3, and C

(K) is SSI in C(K) as we noted

in Theorem 1.6 (see [JiSc], Denition 1.5, Lemma 3.11, proof of Theorem
2.6(b)). Hence Z
2
Z
3
acts m-e

-temperedly by Theorem 3.1 and this smooth

crossed product is m-convex.
4. Action of Z
2
Z
3
is not -tempered.
We show that the action of Z
2
Z
3
on C

(K) is not tempered with respect

to the word length on Z
2
Z
3
, so that it is necessary to use the larger
weight e

to get a tempered action in Theorem 2.3. Let C

(K) be the
characteristic function
a
of the set of innite words in K beginning in b or
b
1
. Let = b
j1
a . . . b
jn
a Z
2
Z
3
. Then by Lemma 2.10,
|
d

( )|
1
=

gG
[
d

( )(g)[ =
1
2
n

gG
(g)
d

_
a
(

k)e
2ig,

k
d

=
1
2
n

gG
(g)
d

_
a
e
2ig,

k
d

=
1
2
n

gG
(g)
d

_
a
e
2i2
n
g,

k
d

1
2
n

lev(g)=n+1
(g)
d

_
a
e
2i2
n
g,

k
d

= 2
(n+1)d

_
a
e
2i(k0+2k1+... )
d

= 2
(n+1)d
( a) = 2
()d/2
C
d
,
364 LARRY SCHWEITZER
where the
1
2
n
went away because there are 2
n
elements of G at level n + 1,
and we used the fact that (g)
d
= (2
n+1
)
d
for g at level n + 1, and that
() = 2n. We have proved:
Lemma 4.1. Let = b
j1
a . . . b
jn
a. Then |

(
a
)|
d
is bounded below by
something directly proportional to (2
d/2
)
()
, and thus grows exponentially
fast with respect to the word length (). Hence the action of Z
2
Z
3
on
C

(K) in Theorem 2.3 is not -tempered.

5. O
2
is not spectral invariant in O
2
.
Theorem 5.1. The smooth Cuntz algebra O
2
= o
e
(Z
2
Z
3
, C

(K)) has
an element whose spectrum contains the unbounded interval (3, ). Hence
O
2
is not spectral invariant in the C

-algebra O
2
.
Proof. Let g = ab Z
2
Z
3
. Then Z

=< g > Z
2
Z
3
and C

(Z)

=
C

(g)) C

r
(Z
2
Z
3
) O
2
. Also [
Z
(g
n
) = [2n[. Thus C

(Z) O
2
=
C

(Z) o
e
(Z
2
Z
3
) = o
e
(Z). Assume that o
e
(Z) is not invertible
in o
e
(Z), but that is invertible in the bigger algebra C

r
(Z). Then if
were invertible in O
2
, it would be invertible in O
2
and so in the smaller
C

-algebra C

r
(Z)[Dix]. Thus is not invertible in O
2
. It follows that for
any o
e
(Z), is invertible in o
e
(Z) if and only if is invertible in O
2
.
So the spectrum is the same in either of these smooth algebras.
Lemma 5.2. The function e
n
2
in the Frechet algebra o
e
(Z) has the un-
bounded interval (3, ) contained in its spectrum.
Proof. Represent Z on C via n e
rn
, r R. Then
r
() =

nZ
(n)e
rn
gives a family of simple o
e
(Z)-modules. The value of
r
(e
n
2
) =

nZ
e
n
2
e
rn
ranges continuously between

0
(e
n
2
) =
+

n=
e
n
2

_
+

e
x
2
dx + 1 =

+ 1 3
and + as r ranges between 0 and +.
Since the spectrum of e
n
2
is the same in O
2
and o
e
(Z) by our preceding
remarks, we have proved Theorem 5.1.
References
[Ch] M-D. Choi, A simple C*-algebra generated by two nite-order unitaries, Can. J.
Math., XXXI(4) (1979), 867-880.
C

FUNCTIONS ON THE CANTOR SET 365

[Da] A.M. Davie, Homotopy in Frechet algebras, Proc. London Math. Soc., 23 (1971),
31-52.
[Dix] J. Dixmier, C*-algebras, North-Holland Publishing Co., Amsterdam, New York,
Oxford, 1982.
[JiSc] R. Ji and L.B. Schweitzer, Spectral invariance of smooth crossed products, and rapid
decay locally compact groups, K-Theory, 10 (1996), 283-305.
[Ko] N. Koblitz, p-adic numbers, p-adic analysis, and Zeta-functions, Springer, New
York, 1977.
[PhSc] N.C. Phillips and L.B. Schweitzer, Representable K-theory of smooth crossed prod-
ucts by R and Z, Trans. Amer. Math. Soc., 344(1) (1994), 173-201.
[Ru] W. Rudin, Fourier Analysis on Groups, New York, Interscience Publishers, 1962.
[Sc 1] L.B. Schweitzer, Dense m-convex Frechet subalgebras of operator algebra crossed
products by Lie groups, Internat. J. Math., 4(4) (1993), 601-673.
[Sc 2] , Spectral invariance of dense subalgebras of operator algebras, Internat. J.
Math., 4(2) (1993), 289-317.
[Sp] J. Spielberg, Free-product groups, Cuntz Krieger algebras, and covariant maps, In-
ternat. J. Math., 2(4) (1991), 457-476.
Received October 10, 1994 and revised October 18, 1996. This article was presented at
the West Coast Operator Algebra Symposium, September, 1994, UCLA.
Abratech Corporation, Suite 255
475 Gate Five Road
Sausalito, CA 94965
E-mail address: lsch@svpal.org