or III; type
I can be handled by the previously discussed hyperniteness results
while types II
i=1
(1)
i
(x
1
, . . . , x
i1
, x
i
x
i+1
, x
i+2
, . . . , x
n
)
+ (1)
n
(x
1
, . . . , x
n
)x
n+1
, (2.2)
for x
i
/, 1 i n + 1. An algebraic computation establishes that
= 0. Thus the image of : L
n1
(/, 1) L
n
(/, 1) is contained in
the kernel of : L
n
(/, 1) L
n+1
(/, 1), and the quotient of these two
vector spaces is the n
th
Hochschild cohomology group H
n
(/, 1). The
elements of the kernel and image above are called respectively cocycles
and coboundaries. For n = 1, the cocycles are derivations, since the
dening equation (2.2) can be rearranged to
(2.3) (x
1
x
2
) = x
1
(x
2
) + (x
1
)x
2
, x
1
, x
2
/.
The coboundaries are the inner derivations implemented by elements
of 1, and H
1
(/, 1) becomes the space of bounded derivations modulo
the space of inner derivations.
One may impose further restrictions on the elements of L
n
(/, 1)
beyond boundedness. Provided that the extra conditions are compat
ible with the coboundary map, various cohomology theories arise by
using the denition given above. We will discuss here the additional
requirement of complete boundedness which leads to the completely
bounded cohomology groups H
n
cb
(/, 1) for suitable modules 1.
The matrix algebras M
n
(/) over a C
algebra of operators on
the nfold sum H
n
. A bounded map : / B between C
algebras
induces a sequence of bounded maps
n
: M
n
(/) M
n
(B), n 1, by
tensoring with the identity on M
n
(C), and is completely bounded
6 ALLAN M. SINCLAIR AND ROGER R. SMITH
if
(2.4) 
cb
:= sup
n
: n 1 < .
This was generalized to multilinear maps in [9] by using matrix multi
plication to replace the identity map on M
n
(C). We illustrate this in the
bilinear case which then has an obvious general extension. If : /
B B(H) is a bounded bilinear map then
n
: M
n
(/) M
n
(B)
M
n
(B(H)) is dened for matrices (a
ij
) M
n
(/), (b
ij
) M
n
(B) by
specifying the (i, j) entry of
n
((a
ij
), (b
ij
)) to be
n
k=1
(a
ik
, b
kj
). Then
is said to be completely bounded if the analogous supremum to (2.4) is
nite. The importance of completely bounded maps is twofold: most
naturally occurring maps on C
1
(x)V
2
2
(y)V
3
, x /
1
, y /
2
,
where
i
: /
i
B(K
i
), i = 1, 2, are representations and V
3
: H
K
2
, V
2
: K
2
K
1
, V
1
: K
1
H are bounded operators satisfying
(2.7) V
1
 = V
2
 = V
3
 = 
1/3
cb
.
This theorem makes it an easy exercise to verify that the cobound
ary operator preserves complete boundedness, and so the completely
bounded cohomology groups H
n
cb
(/, 1) can be dened whenever 1 has
A SURVEY OF HOCHSCHILD COHOMOLOGY 7
an operator space structure with respect to which it is also an operator
/module. In particular, when / B(H) is a von Neumann algebra,
we will discuss H
n
cb
(/, /) and H
n
cb
(/, B(H)) subsequently.
If / B B(H) is an inclusion of C
algebras and : B B
B(H) is a bounded bilinear map then we say that is /modular if
the following relations are satised:
a(b
1
, b
2
) = (ab
1
, b
2
), (2.8)
(b
1
a, b
2
) = (b
1
, ab
2
), (2.9)
(b
1
, b
2
a) = (b
1
, b
2
)a, (2.10)
for a / and b
1
, b
2
B. We omit the obvious extension to multilinear
maps.
In anticipation of subsequent developments, we mention that we
will be able to work with cocycles which are /modular with respect
to a suitable subalgebra /. Usually this means one with an amenable
generating subgroup of unitaries or, for von Neumann algebras, a hy
pernite subfactor. When / is a factor of type I
, II
or III, there is
a factor ^ and a separable Hilbert space H such that / is isomorphic
to ^ B(H). Any multilinear map on / which is separately nor
mal in each variable and also modular with respect to the subalgebra
1B(H) can be expressed as I where is multilinear on ^. Since
B(H) contains arbitrarily large matrix subalgebras, it is easy to estab
lish the complete boundedness of and hence of in this situation.
A similar argument applies to those type II
1
factors which are stable
under tensoring with the hypernite type II
1
factor, but it should be
noted that not all type II
1
factors have this property.
3. Reduction of cocycles
In this section we will discuss a basic technique, originating in [23],
which seeks to modify a given cocycle by a coboundary so that the
resulting cocycle has some desirable properties. To motivate the ap
proach, we rst consider a von Neumann algebra / with a module
1 and we let / be a nite dimensional subalgebra of /. Then /
has a norm compact unitary group , whose normalized Haar measure
8 ALLAN M. SINCLAIR AND ROGER R. SMITH
we denote by . Let : / / 1 be a 2cocycle, and dene
L
1
(/, 1) by
(3.1) (x) =
U
u(u
, x) d(u), x /,
where this integral converges in the norm topology of 1. Then
(3.2) (v, x) =
U
vu(u
, x) u(u
, vx) + u(u
, v)x d(u),
for v  and x /, and invariance of the Haar measure allows us to
replace u by v
U
u[(u
v, x) (u
, vx) + (u
, v)x] d(u).
Since (u
(v, x) (u
v, x) + (u
, vx) (u
, v)x = 0
to rewrite (3.3) as
(3.5) (v, x) =
U
u[u
U
(xu, u
) d(u), x /.
Then
(3.9) (x, y) =
U
x(yu, u
) (xyu, u
) + (xu, u
)y d(u).
If we replace x by a / in (3.9) then
(3.10) (a, y) =
U
a(yu, u
) (ayu, u
) d(u) = 0,
A SURVEY OF HOCHSCHILD COHOMOLOGY 9
using (3.7) to obtain vanishing of the integrand. Now replace y by
v  in (3.9). This gives
(3.11) (x, v) =
U
x(vu, u
) (xvu, u
) + (xu, u
)v d(u).
The rst term on the right vanishes and the second may be replaced by
(xu, u
, v) (x, v) + (xu, u
v) (xu, u
)v = 0,
in which the rst term is 0, allows us to rewrite (3.11) as
(3.13) (x, v) =
U
(x, v) d(u) = (x, v).
If we dene to be + , then these calculations show that is a
cocycle equivalent to but with the additional property that (x, y) =
0 whenever at least one of the arguments is in /. As a consequence,
may be replaced by an equivalent cocycle which is /modular. To
obtain this extra information, consider the identity
(3.14) x(a, y) (xa, y) + (x, ay) (x, a)y = 0
for x, y / and a /. The rst and last terms vanish leaving
(3.15) (xa, y) = (x, ay).
The equation (x, ya) = (x, y)a follows similarly from (x, y, a) = 0.
For simplicity, we have only discussed the 2cocycle case, but these
techniques are generally valid and give the following.
Theorem 3.1. Let be an ncocycle on / and let / be a nite
dimensional C
, x)), 1
,
denes an element of (1
subalgebra of / then
(3.18) H
n
(/, 1) = H
n
w
(/, 1) = H
n
(/, 1) = H
n
w
(/, 1), n 1.
If / is hypernite then this theorem shows that any ncocycle
into a dual normal Mmodule 1 is equivalent to one which vanishes
whenever one of the arguments is in /. This says, of course, that any
cocycle is a coboundary and H
n
(/, 1) = 0 in this case, [23]. In [13],
a converse to this result was given.
Theorem 3.4. If / is a von Neumann algebra and H
1
(/, 1) = 0
for all dual normal /modules 1, then / is hypernite.
4. Completely bounded cohomology
In this section we discuss the completely bounded cohomology groups
H
n
cb
(/, B(H)) and H
n
cb
(/, /), where / is a von Neumann algebra
on a Hilbert space H. Both groups will be 0. The techniques for show
ing this are dierent in the two cases, but both depend crucially on
the representation theorem for completely bounded multilinear maps
(Theorem 2.1). The methods from Section 3, which show that any co
cycle is equivalent to one which is separately normal, preserve complete
boundedness. Thus, when considering a xed completely bounded co
cycle, it may be assumed to be separately normal and, if necessary, to
be 1modular for a xed hypernite subalgebra 1of /. In particular,
a cocycle may always be assumed to vanish when one of its arguments
is a scalar.
Consider now a von Neumann algebra / B(H) and let : /
/ B(H) be a 2cocycle such that (1, y) = (x, 1) = 0 for all
x, y /. If is also completely bounded then there is a representation
(4.1) (x, y) = V
1
1
(x)V
2
2
(y)V
3
, x, y /,
where
i
: / B(K
i
), i = 1, 2, are representations and V
3
: H
K
2
, V
2
: K
2
K
1
and V
1
: K
1
H are bounded operators. Let P be
12 ALLAN M. SINCLAIR AND ROGER R. SMITH
the projection in B(K
1
) onto the closed linear span of V
2
2
(y)V
3
h: y
/, h H. Then
(4.2) (x, y) = V
1
1
(x)PV
2
2
(y)V
3
,
and the condition (1, y) = 0 gives
(4.3) V
1
PV
2
2
(y)V
3
h = 0, y /, h H,
and thus V
1
P = 0. Replacing V
2
by PV
2
allows us to assume that
V
1
V
2
= 0. If we let Q be the projection in B(K
2
) onto the closed
linear span of V
2
1
(x
)V
1
h: x /, h H then we may replace V
2
by V
2
Q, and the condition (x, 1) = 0 implies that QV
3
= 0, arguing
as above. Thus we may also assume that V
2
V
3
= 0. Suppressing the
representations
1
and
2
for notational convenience, (4.2) becomes
(x, y) = V
1
xV
2
yV
3
= (V
1
x xV
1
)(V
2
yV
3
)
= (V
1
x xV
1
)(V
2
y yV
2
)V
3
= [V
1
, x][V
2
, y]V
3
, (4.4)
where [, ] denotes the commutator. A simple algebraic calculation
shows that
(4.5) 0 = (x, y, z) = [V
1
, x][V
2
, y][V
3
, z].
To show that is a coboundary, we rst let e be the smallest projection
such that e[V
3
, z] = [V
3
, z], and dene
(4.6) (x) = V
1
xV
2
(1 e)V
3
, x /.
Then
(x, y) = xV
1
yV
2
(1 e)V
3
V
1
xyV
2
(1 e)V
3
+ V
1
xV
2
(1 e)V
3
y
= [V
1
, x]yV
2
(1 e)V
3
+ [V
1
, x]V
2
(1 e)V
3
y
= [V
1
, x]([y, V
2
](1 e)V
3
+ V
2
y(1 e)V
3
V
2
(1 e)V
3
y)
= (x, y) [V
1
, x]V
2
(1 e)[y, V
3
]
= (x, y). (4.7)
A SURVEY OF HOCHSCHILD COHOMOLOGY 13
In this calculation we have used the fact that [V
1
, x][V
2
, y]e = 0 and
that the space spanned by [V
3
, y]H is
2
(/)invariant, so that e and
1 e lie in
2
(/)
.
The discussion above shows that H
2
cb
(/, B(H)) = 0, and we have
included the argument to illustrate the importance of the completely
bounded representation theorem. Similar calculations give a general
result, [5].
Theorem 4.1. Let / B(H) be a von Neumann algebra. Then
H
n
cb
(/, B(H)) = 0 for n 1.
When a cocycle as above maps into /, Theorem 4.1 shows that
= for some L
1
cb
(/, B(H)) but not necessarily lying in
L
1
cb
(/, /). Thus one cannot conclude immediately that H
n
cb
(/, /) =
0. This is true, however, but requires a preparatory averaging theorem,
[11, 12], which can also be found in [43].
The objective is to construct a contractive projection from the
space CB(/, /) of completely bounded maps of / to itself onto the
subspace CB(/, /)
M
of right /module maps. For a special class
of maps this is easy to achieve. If s, t / and (x) = sxt then the
Dixmier approximation theorem can be applied to t. Maps of the form
(4.8)
1
n
n
i=1
(xu
i
)u
i
=
1
n
n
i=1
sxu
i
tu
i
, x /,
where the u
i
s are unitaries in /, converge strongly to a map x sxz
where z is a central element of /, and this is a right /module
map. This motivates the correct approach which is to consider sets
of operators u
in / for which u
)u
. The
desired projection is obtained from a minimal invariant set argument,
and the representation theorem for completely bounded maps is crucial.
We refer to [43] for the many details which we are omitting here, but
once the method works on the right it is easy to obtain a projection
onto left module maps by taking adjoints, and then onto bimodule
maps by applying these two results sequentially. We summarize:
14 ALLAN M. SINCLAIR AND ROGER R. SMITH
Theorem 4.2. For a von Neumann algebra /, there exist contrac
tive projections of CB(/, /) onto the spaces CB(/, /)
M
,
M
CB(/, /)
and CB
M
(/, /) of respectively right, left, and twosided /modular
completely bounded maps.
We can immediately apply this result to cohomology. Let be the
projection onto right module maps obtained above, and note that every
such map has the simple form x tx for some t /. The nature
of the construction also shows that if CB(/, /), a /, and
a
(x) = (ax), then (
a
)(x) = ()(ax). If we have a completely
bounded bilinear map (x, z) and we regard z as the variable and x
x, then there is a completely bounded map (x) such that
(4.9) (x, z) = (x)z, x, z /,
and
(4.10) (xy, z) = (xy)z, (x, yz) = (x)yz.
If is, additionally, a 2cocycle then we may apply to the z variable
in the cocycle identity
(4.11) x(y, z) (xy, z) + (x, yz) (x, y)z = 0
to obtain
(4.12) x(y)z (xy)z + (x)yz (x, y)z = 0,
from which (x, y) = (x, y) follows by letting z = 1. This argument
applies equally to ncocycles which are completely bounded and leads
to the following result, [11].
Theorem 4.3. If /is a von Neumann algebra, then H
n
cb
(/, /) =
0 for n 1.
It is important to note that, in order to apply , only complete
boundedness in the last variable is required. Similarly complete bound
edness in the rst variable would allow us to apply the projection onto
left module maps with the same eect.
We observed in Section 2 that if / is stable under tensoring with
B(H) (or the hypernite II
1
factor 1) then any separately normal
multilinear map which is modular with respect to these distinguished
A SURVEY OF HOCHSCHILD COHOMOLOGY 15
subalgebras is completely bounded. Since cocycles are equivalent to
ones which are separately normal and modular with respect to a chosen
hypernite subfactor, Theorem 4.3 applies in these cases, [5, 11].
Theorem 4.4. If / is a von Neumann algebra of type I, II
, III,
or is type II
1
and stable under tensoring with 1, then H
n
(/, /) = 0
and H
n
(/, B(H)) = 0 for n 1.
Not all type II
1
factors are stable under tensoring with 1 since the
fundamental group need not be R
+
, [38], and so this theorem leaves
open the type II
1
case. In the following sections we will discuss this
further.
5. Algebras with Cartan subalgebras
In view of the results of the previous section, the main open prob
lems in cohomology theory concern type II
1
factors. The questions of
whether H
n
(/, /) = 0, n 2, and H
n
(/, B(H)) = 0, n 1, re
main unresolved but there has been recent progress on both in special
cases. In this section we consider the rst of these questions for type
II
1
von Neumann algebras / with Cartan subalgebras. A maximal
abelian subalgebra / / is said to be Cartan if / is generated as
a von Neumann algebra by those unitaries u / which normalize /
in the sense that u/u
G is a type II
1
factor and / is a Cartan subal
gebra; the hypotheses on the action guarantee that / is a masa, and
^(/) contains both the unitaries implementing the action of G and
the unitary group of /, thus generating /
G.
We will require some deep results in order to discuss the Cartan sub
algebra case. The rst of these is the noncommutative Grothendieck
inequality proved by Pisier, [32], and further rened by Haagerup, [17],
whose version we give here. Let : /
2
C be a separately normal
bounded bilinear form on a von Neumann algebra /. Then there exist
16 ALLAN M. SINCLAIR AND ROGER R. SMITH
normal states f
1
, f
2
, g
1
, g
2
on / such that
(5.1) [(x, y)[
1
2
(f
1
(x
x) + f
2
(xx
) + g
1
(y
y) + g
2
(yy
)).
If 1 is a hypernite subalgebra whose relative commutant 1
/
equals the center Z of /, then this inequality can be strengthened for
bilinear maps which are inner 1modular in the sense that
(5.2) (xr, y) = (x, ry), r 1, x, y /.
Fix an amenable group of unitaries  which generates 1, and replace
x and y in (5.1) by respectively xu and u
) + G(y
y)), x, y /,
where F and G are normal states. When u
xu is averaged over , it
becomes E(x
x) = E(xx
/ = Z. Let : // / be
a separately normal inner 1modular map. Then
(5.4)
i=1
(x
i
, y
i
)
2
i=1
x
i
x
1/2
i=1
y
i
y
i
1/2
for any nite sets of elements x
i
, y
i
/.
There is a standard technique for applying bilinear results to linear
maps. If : / / is a bounded linear map then dene : /
/ / by
(5.5) (x, y) = (x)(y
, x, y /.
If is right 1modular then is inner 1modular. By letting y
i
be x
i
in (5.4), we obtain the following.
A SURVEY OF HOCHSCHILD COHOMOLOGY 17
Theorem 5.2. Let 1 / satisfy the hypotheses of Theorem 5.1,
and let : / / be a bounded normal right 1modular map. Then,
for x
1
, . . . , x
n
/,
(5.6)
i=1
(x
i
)(x
i
)
2
2
i=1
x
i
x
.
We can give an alternative formulation of this theorem in terms
of row bounded maps, which are intermediate between bounded and
completely bounded maps. Parallel to the denition of a completely
bounded map, : / / is row bounded if
(5.7) 
r
= sup((x
1
), . . . , (x
n
)): (x
1
, . . . , x
n
) 1 <
where the rows in (5.7) are allowed to be arbitrarily long.
Corollary 5.3. If : / / is bounded, separately normal,
and right 1modular for a hypernite subalgebra 1 with 1
/ = Z,
then is row bounded and
(5.8)  
r
2 .
This corollary allows us to obtain a useful estimate on the projection
of Theorem 4.2. If : / / is normal, completely bounded and
right 1modular for 1
)u
: u
1
are bounded in norm by 
r
x. Right 1modularity then gives the
estimate
(5.10)  
r
2 
from (5.8). It is curious that complete boundedness of is apparently
required to construct , but 
cb
is not required in (5.10).
We can now state the main result of this section which has evolved
through the special cases of [34, 6] to its nal formulation in [44]. We
will then sketch the main steps in proving it, referring to [44, 45] for
the full details.
Theorem 5.4. Let / be a separably acting type II
1
von Neumann
algebra with a Cartan subalgebra /. Then H
n
(/, /) = 0 for n 1.
18 ALLAN M. SINCLAIR AND ROGER R. SMITH
By work of Popa, [36], in the factor case and extended in [46] to the
general case, there is a hypernite subalgebra 1such that / 1 /
and 1
au, xb)
= (u, (a)xb), a, b /, x /, (5.11)
where is the automorphism a u
au of /. Cartan subalgebras
have the property that they norm their containing von Neumann alge
bras in the sense that, for X M
n
(/),
(5.12) X = sup RXC,
where the supremum is taken over rows and columns in M
n
(/) of unit
norm, [35]. Equation (5.11) lifts to the n n matrix level where,
modulo the norm preserving automorphism , rows and columns over
/ can be pulled through to the second entry. Using this fact, it is
then straightforward to show that (u, x) is completely bounded as
a linear function of x when u ^(/), and thus also when the rst
argument is in the algebra generated by ^(/). Since these maps are
right 1modular, by 1modularity of , the estimate of (5.10) allows
us to apply to (m, x), when m Alg(^(/)), in a continuous way,
and thus when m lies in the completion C
x))
1/2
. Later, Dixmier, [14], gave an equivalent formulation of
property by the following condition: given an integer n, elements
x
1
, . . . , x
m
/ and > 0, there exist n orthogonal projections
p
1
, . . . , p
n
/, each of trace n
1
, such that their sum is 1 and
(6.2) [p
i
, x
j
]
2
< , 1 i n, 1 j m.
This second formulation will prove to be the more useful one here. By
xing a  
2
dense countable set of elements in / and letting be
k
1
for k 1, we may nd a countable set of projections in / so that
(6.2) can always be satised by projections from this set. We will refer
to such a collection as a set of Dixmier projections. An important step
is to be able to nd such a set for / within a hypernite subfactor.
This can be achieved by using a result of Popa, [36], stating that any
separably acting type II
1
factor / contains a hypernite subfactor 1
of trivial relative commutant 1
X))
1/2
, X M
n
(/).
We also x a hypernite subfactor 1 of trivial relative commutant.
There is no direct generalization of the Grothendieck inequality to
three or more variables, and the essential dierence appears to be that
a bilinear map (x
1
, x
2
) has only outer variables while a trilinear map
(x
1
, x
2
, x
3
) has an inner variable x
2
which is hard to reach. How
ever, useful inequalities can be obtained by xing some variables and
applying a vector functional to connect with (5.3). We illustrate this
for a trilinear map : /
3
B(H) which is separately normal and
1multimodular. To handle the outer variables we consider bilinear
maps
(6.4)
1
(x, y) = (x
, x
2
, x
3
)
(y, x
2
, x
3
), x, y /
and
(6.5)
3
(x, y) = (x
1
, x
2
, x)(x
1
, x
2
, y
, x, y /,
whereas the appropriate map for the inner variable is
(6.6)
2
(x, y) = (x, y, x
3
), x, y /.
For arbitrary vectors , H, the bilinear forms
(6.7)
i
(x, y) =
i
(x, y), ), 1 i 3, x, y /,
are inner 1modular and separately normal, so that the inequality
(5.3) is valid. This approach applies equally to the amplications
n
A SURVEY OF HOCHSCHILD COHOMOLOGY 21
to M
n
(/), and repeated averaging over 1M
n
leads to the following
result.
Theorem 6.2. Let M be a type II
1
factor with a hypernite sub
factor of trivial relative commutant and let : /
k
B(H) be a
separately normal 1multimodular map. Then, for each n 1 and
X
1
, . . . , X
k
M
n
(/),
(6.8) 
n
(X
1
, . . . , X
k
) 2
k/2

n
(X
1
) . . .
n
(X
k
).
The term n tr
n
(X
X) in the denition of
n
(X) precludes (6.8)
from implying complete boundedness of in Theorem 6.2, although
this is our objective when / has property . A step in this direction
is to observe that if P M
n
(/) is a projection of trace n
1
, then
(6.9)
n
(XP)
2
= XP
2
+ n tr
n
(PX
XP) 2X
2
.
This gives
Corollary 6.3. In addition to the hypotheses of Theorem 6.2, let
P M
n
(/) be a projection of trace n
1
. Then, for X
1
, . . . , X
k
M
n
(/),
(6.10) 
n
(X
1
P, . . . , X
k
P) 2
k
X
1
 . . . X
k
.
To illustrate the relevance of this, and to motivate what comes next,
consider a normal linear map : / B(H) on a property factor
which is modular with respect to a hypernite subfactor 1 containing
a set of Dixmier projections for /. Let p 1 be any projection. Then
p(x) = p(px) = p([p, x]) + p(xp)
= (p[p, x]) + p(xp)p, x /. (6.11)
This algebraic identity is valid at the n
th
matrix level with p replaced
by P = p I
n
. Moreover, we may nd a countable set of projections
in M
n
(1) of this form such that
(6.12) tr
n
(P
ij
) = n
1
,
n
j=1
P
ij
= 1, lim
i
[P
ij
, X]
2
= 0
22 ALLAN M. SINCLAIR AND ROGER R. SMITH
for each X M
n
(/). For an arbitrary pair of unit vectors , H
n
,
(6.11) becomes
n
(X), ) =
n
j=1
P
ij
n
(X), )
=
n
j=1
n
(P
ij
[P
ij
, X]), ) +
n
j=1
P
ij
(XP
ij
)P
ij
, ), (6.13)
for X M
n
(/) and i 1. The last sum in (6.13) can be estimated
by
(6.14) max
i,j

n
(XP
ij
) 2X,
by Corollary 6.3. The arguments in the penultimate sum in (6.13)
are P
ij
[P
ij
, X] which are uniformly bounded and tend to 0 in  
2

norm as i . Thus this sum vanishes in the limit provided that
lim
i
n
(P
ij
[P
ij
, X]), ) = 0, which certainly would result from  
2

norm continuity of
n
on bounded balls. However, this follows from
normality and the Grothendieck inequality (see [43, 5.4.3] and [7, The
orem 4.4]). Letting i in (6.13) results in the estimate
(6.15) 
n
(X) 2X, X M
n
(/), n 1,
which establishes complete boundedness of with 
cb
2.
To a certain point, this argument extends smoothly to the multilin
ear case. There are more complicated but similar analogs of (6.11)
and (6.13) but the penultimate sum in (6.13) has, in the bilinear
case, terms like
n
([P
ij
, X], P
ij
[P
ij
, Y ]), ). Successfully taking the
limit as i now requires joint  
2
norm continuity, rather than
the separate  
2
norm continuity guaranteed by separate normality.
Baire, [1], applied his category theorem to a separately continuous map
f : [0, 1]
2
[0, 1] to conclude that there is a point (x
0
, y
0
) of joint con
tinuity for each y
0
[0, 1]. This extends to an arbitrary separately
continuous function f : A Z on complete metric spaces (see [2]
for a more general result). This is the starting point for the proof of
the next result.
Theorem 6.4. Let : /
k
B(H) be a bounded separately nor
mal klinear map.
A SURVEY OF HOCHSCHILD COHOMOLOGY 23
(a) If maps into /, then is jointly  
2
norm continuous
when restricted to bounded balls.
(b) If , H then the map
(6.16) (x
1
, . . . , x
k
) = (x
1
, . . . , x
k
), ), x
i
/,
is jointly  
2
norm continuous when restricted to bounded
balls.
The result mentioned before the theorem is used to obtain joint
continuity in the  
2
norm for a bilinear map at a point (a, 0) in
the product of two unit balls. Then joint continuity is deduced at (0,0)
and subsequently for all points (x, y) in the unit balls. The multilinear
case is proved by induction, regarding the nfold product B
n
of balls
as B B
n1
and starting again with a point of joint continuity at
(a, 0, . . . , 0).
By combining Corollary 6.3, Theorem 6.4, and by establishing al
gebraic identities analogous to (6.11) and (6.13), the following result is
achieved. Note that while we require / to have a separable predual,
equivalent to admitting a representation on a separable Hilbert space,
we do not make any assumptions on H.
Theorem 6.5. Let / B(H) be a type II
1
factor with a separable
predual and property . Let 1 be a hypernite subfactor with trivial
relative commutant and containing a set of Dixmier projections for /.
Then any separately normal 1modular klinear map : /
k
B(H),
k 1, is completely bounded and satises the estimate 
cb
2
k
.
Since cocycles can always be assumed to be 1modular and sepa
rately normal, this result gives the equalities
(6.17)
H
n
(/, /) = H
n
cb
(/, /), H
n
(/, B(H)) = H
n
cb
(/, B(H)),
for any n 1, at least when / has a separable predual (and, of
course, property ). From Section 4, we conclude that these cohomol
ogy groups are 0. To remove the requirement of a separable predual, a
technical result is needed. This states that any property factor is the
ultraweak closure of an increasing net of such factors with separable
preduals. We may now summarize the discussion of this section.
24 ALLAN M. SINCLAIR AND ROGER R. SMITH
Theorem 6.6. Let / B(H) be a type II
1
factor with property
. Then, for n 1,
(6.18) H
n
(/, /) = 0 and H
n
(/, B(H)) = 0.
This theorem was proved in [7, 8] but is only new for n 3. For
n = 1, the two cases are in [24, 42, 3] while for n = 2 we refer to
[11, 4]. The approach taken in [7] has its origins in [4].
Sections 5 and 6 describe the most recent advances in this eld but
do not cover all type II
1
factors. The factor V N(F
2
) arising from the
free group on two generators has neither property , [29], nor a Cartan
subalgebra, [48], and the groups H
n
(V N(F
2
), V N(F
2
)), n 2, are yet
to be determined. It would be very interesting to resolve this problem
for at least the case n = 2.
References
[1] R. Baire, Sur les fonctions de variables reelles, Ann. Mat. Pura Appl. 3 (1899),
1122.
[2] J.C. Breckenridge and T. Nishiura, Partial continuity, quasicontinuity, and
Baire spaces, Bull. Inst. Math. Acad. Sinica 4 (1976), 191203.
[3] E. Christensen, Similarities of II
1
factors with property , J. Operator Theory
15 (1986), 281288.
[4] E. Christensen, Finite von Neumann algebra factors with property gamma, J.
Funct. Anal. 186 (2001), 366380.
[5] E. Christensen, E.G. Eros and A.M. Sinclair, Completely bounded multilinear
maps and C
algebras,
Adv. in Math. 56 (1985), 93116.
[18] G. Hochschild, On the cohomology groups of an associative algebra, Ann. of
Math. 46 (1945), 5867.
[19] G. Hochschild, On the cohomology theory for associative algebras, Ann. of
Math. 47 (1946), 568579.
[20] G. Hochschild, Cohomology and representations of associative algebras, Duke
Math. J. 14 (1947), 921948.
[21] B.E. Johnson, Cohomology in Banach algebras, Mem. Amer. Math. Soc. 127,
American Mathematical Society, Providence, R.I., 1972. iii+96 pp.
[22] B.E. Johnson, Perturbations of Banach algebras, Proc. London Math. Soc. (3)
34 (1977), 439458.
[23] B.E. Johnson, R.V. Kadison and J.R. Ringrose, Cohomology of operator al
gebras. III. Reduction to normal cohomology, Bull. Soc. Math. France 100
(1972), 7396.
[24] R.V. Kadison, Derivations of operator algebras, Ann. of Math. 83 (1966),
280293.
[25] R.V. Kadison and J.R. Ringrose, Cohomology of operator algebras I. Type I
von Neumann algebras, Acta Math. 126 (1971), 227243.
[26] R.V. Kadison and J.R. Ringrose, Cohomology of operator algebras II. Ex
tended cobounding and the hypernite case, Ark. Mat. 9 (1971), 5563.
[27] D. McDu, Central sequences and the hypernite factor, Proc. London Math.
Soc. (3) 21 (1970), 443461.
[28] D. McDu, On residual sequences in a II
1
factor, J. London Math. Soc. (2) 3
(1971), 273280.
[29] F.J. Murray and J. von Neumann, On rings of operators IV, Ann. of Math.
(2) 44 (1943), 716808.
[30] V.I. Paulsen, Completely bounded maps and dilations, Pitman Research Notes
in Mathematics Series, vol. 146, Longman Scientic & Technical, Harlow, 1986.
26 ALLAN M. SINCLAIR AND ROGER R. SMITH
[31] V.I. Paulsen and R.R. Smith, Multilinear maps and tensor norms on operator
systems, J. Funct. Anal. 73 (1987), 258276.
[32] G. Pisier, Grothendiecks theorem for noncommutative C
algebras, with an
appendix on Grothendiecks constants, J. Funct. Anal. 29 (1978), 397415.
[33] G. Pisier, Introduction to operator space theory, London Mathematical Society
Lecture Note Series, vol. 294, Cambridge University Press, Cambridge, 2003.
[34] F. Pop and R.R. Smith, Cohomology for certain nite factors, Bull. London
Math. Soc. 26 (1994), 303308.
[35] F. Pop, A.M. Sinclair and R.R. Smith, Norming C
algebras by C

subalgebras. J. Funct. Anal. 175 (2000), 168196.
[36] S. Popa, On a problem of R.V. Kadison on maximal abelian subalgebras in
factors, Invent. Math. 65 (1981), 269281.
[37] S. Popa, Notes on Cartan subalgebras in type II
1
factors, Math. Scand. 57
(1985), 171188.
[38] S. Popa, On a class of type II
1
factors with Betti numbers invariants, preprint,
2001.
[39] F. Radulescu, Vanishing of H
2
w
(M, K(H)) for certain nite von Neumann al
gebras, Trans. Amer. Math. Soc. 326 (1991), 569584.
[40] I. Raeburn and J. Taylor, Hochschild cohomology and perturbations of Banach
algebras, J. Funct. Anal. 25 (1977), 258266.
[41] J. Ringrose, Linear mappings between operator algebras, Symposia Mathe
matica, Vol. XX (Convegno sulle Algebre iC