Professional Documents
Culture Documents
71 Von Neumann Algebras - Mazhe
71 Von Neumann Algebras - Mazhe
71.2 Commutant
Let M Ä BpH q be a collection of bounded operators on H . The commutant of M in BpH q
is
M 1 “ tS P BpH q tel que T S “ ST @T P M u. (71.2)
From now let M be a self-adjoint algebra of operators on an Hilbert space H which contains .
We have three lemmas.
Lemme 71.1.
If v P H and if P is the projection onto M v Ñ H , then P commutes with all operators in M .
Lemme 71.2.
If S P M 2 , v P H and ‘ ° 0, there exists a T P M such that }Sv ´ T v} § ‘. Moreover, in the
finite dimensional case, there exists a T P M such that Sv “ T v.
Proof. Consider P as in lemma 71.1, then P P M 1 , so that XP “ P X. Now, the fact that M is
unital gives v P M v from which we deduce P v “ v. Thus we have
Xv “ XP v “ P Xv P M v
because Xv “ P pXvq. The lemma now result from the fact that an element of the closure of M v
is as close as we want from M v.
Lemme 71.3.
If T P M 2 , v1 , . . . vn P H , and ‘ ° 0, there exists a T P M such that }Svi ´ T vi } § ‘ for all i.
Moreover if the finite dimensional case, there exists a T such that Svi “ T vi .
2763
2764 CHAPTER 71. VON NEUMANN ALGEBRAS
Corollaire 71.4.
In the finite dimensional case, we have M 2 “ M .
Proof. A first evidence is that M Ñ M 2 . Secondly, every commutant is strongly closed, so that
M 2 in particular is strongly closed. So it remains to be proved that for each S P M 2 is the limit
of some net in M in the strong topology. For that, consider the directed set of finite subsets of H
and consider the directed set
A “ tfinite subsets of H uˆs0, 8r
on which we say pF, ‘q • pF 1 , ‘1 q if and only if F 1 Ñ F and ‘ † ‘1 .
For each a “ pF, ‘q P A, we define TpF,‘q as in lemma 71.3, so we have, for all v P F ,
}TpF,‘q v ´ Sv} § ‘,
which proves that TpF,‘q converges to S in the strong topology.
Définition 71.6.
A von Neumann algebra is an unital ˚-subalgebra of BpH q which is equal to its double com-
mutant.
Proposition 71.7.
Let fi be a unitary representation of G on H and H1 , an invariant subspace of fipGq. We have
— the orthogonal projection on H1 belongs to M ,
— if P P M is a projection, then P H is a subrepresentation of fi, i.e. the closed subspace
fipGqP H .
Proof. First, suppose that H1 is S -invariant. Thus for every v P H1 and w P H1K , we have
0 “ xSv, w, y “ xv, S ˚ wy
where S ˚ P S by assumption. We conclude that S ˚ w K H1 for every S P S . Since S ˚ “ S , we
have that H1K is S -invariant. Now if P is the orthogonal projector on H1 , we decompose x P H
as
x “ P x ‘ pId ´P qx P H1 ‘ H1K . (71.3)
since H1K is S -invariant, or every S P S , we have SpId ´P qx P H1K , so that P SpId ´P qx “ 0.
Thus, `using ˘the decomposition (71.3), P Sx “ P SP x ‘ 0 “ P SP x. But P x P H1 , then SP x P H1
and P SP x “ SP x. We have proven that P S “ SP , it is P P S 1 .
For the second part, assume that P P S 1 , then for every v P H1 , we have v “ P v and
Sv “ SP v “ P Sv P H1 ,
which proves that H1 is S -invariant.
71.3. EXAMPLES OF VON NEUMANN ALGEBRAS 2765
Proposition 71.9.
If H1 and H2 are equivalent subrepresentations of H , then the intertwining operator W : H1 Ñ
H2 determines on H a partial isometry such that P1 “ W ˚ W and P2 “ W W ˚ .
Proposition 71.10. ` ˘
There exists a partial isometry V : Image |T | Ñ ImagepT ˚ q such that T “ V |T |.
In other words, any element of a von Neumann algebra is the product of a positive operator by a
projection. A version of this decomposition for operators in Hilbert spaces is given in lemme 57.63.
` ˘
Proof. The operator V must satisfy V |T |v “ T v for every v P H . That operator is a partial
isometry because
› ›
›|T |v ›2 “ x|T |v, |T |vy “ x|T |2 v, vy “ xT ˚ T v, vy “ }T v}2 .
Lemme 71.11.
Every von Neumann algebra is the commutant of an unitary representation.
Proof. Let M be a von Neumann algebra and consider that group G “ U pM 1 q, the group of unitary
operators on M 1 . Let T P M 1 , it reads as the sum of self-adjoint operators by
1 i
T “ pT ` T ˚ q ´ piT ´ iT ˚ q.
2 2
Thus we can restrict ourself to self-adjoint operators. Let S P M 1 , it can be written as a combination
of unitary operators:
1´ a ¯ 1´ a ¯
S“ S ` i 1 ´ S2 ` S ´ i 1 ´ S2 .
2 2
Notice that the square` root makes
˘ sense because 1 ´ S 2 is positive. So M 1 is spanned by U pM 1 q
1
and then M “ M 2 “ U pM 1 q .
T g “ T pgf0 q “ gT f0 “ gh
That proves that }h}8 § }T } because if }h}8 “ }T } ` ”, then we have a set (of non vanishing
measure) on which h is bigger than }T }. We conclude that h P L8 pXq.
2766 CHAPTER 71. VON NEUMANN ALGEBRAS
In turn, one can see that the commutant of the right hand side is M itself.
Let us now go back with the case of H8 . The map T fiÑ T8 provides an isomorphism M »
BpH q as ˚-algebras. But it is not sufficient to conclude that M is a von Neumann algebra because
the topologies do not correspond. A net T– strongly converges in M when T– f converges for every
f P M . But f is an infinite list of vectors in H , so that the strong topology in M is something like
an infinite collection of strong topology on BpH q. Every strongly open set in BpH q is strongly
open in M , while the reciprocal is not true.
‡n : An Ñ An`1
(71.5)
T1 b . . . b Tn ބ T1 b . . . b Tn b .
We pose A “ limÑ An .
Lemme 71.12.
If fi1 and fi2 are ˚-homomorphisms from A to BpH q, then
for every a P A. In other words, there is an unique way to close A in the norm topology.
If one does not ask fii to be faithful, I can take fi1 pAq “ 0 as counter-example. Thus I think that I have to add the faithful assumption.
Proof. One has }fii paq}2 “ }fii pa˚ aq}, so that we only have to prove (71.6) in the case of self-
adjoint elements of A. Using` what ˘ is said in the proof of proposition 69.63, we have (with obvious
notations) }fii pbq} “ rBpH q fii pbq . Since b belongs to one of the Ak “ nk p q and dim Ak † 8.
Now we ask the question of an actual way to represent A on an Hilbert space. First, as matrix
algebra, Mk “ BpHk q for a certain finite dimensional Hilbert space Hk . We form the Hilbert
space
H1 b . . . b Hk
with the inner product
Notice the dependence in the vectors vk . Now we define the map fi : A Ñ BpHq by
fipT1 b . . . b Tk qpw1 b w2 b . . .q “ T1 w1 b . . . b Tk wk b wk b wk`1 b . . . (71.10)
 ` ˘
More intrinsically, fi : Mk Ñ B pHk , vk q ,
` ˘
fi b Tk pbwk q “ bTk wk . (71.11)
One can show that the strong closure of fipAq is BpHq. The strong topology on BpH q is generated
by the open sets
UpS, v, ‘q “ tT P BpHq tel que }T v ´ Sv} § ‘u
with S P BpH q, v P H and ‘ ° 0. The direct limit H “ limÑ pH1 b . . . b Hk q, is given by a
vector space H and maps Ïk such that
‡
H1 b . . . b Hk / H1 b . . . b Hk`1
Ï Ïk`1
& w
H
commutes where ‡pw1 b . . . b wk q “ w1 b . . . b wk b vk`1 . The space H is the free vector space
generated by symbols of the form w1 b . . . b wk b vk`1 b . . . and we pose Ïi pw1 b . . . b wi q “
w1 b . . . b wi b vi`1 b . . . and the Hilbert space H is the completion of H .
The definition of A proceeds in the same way: Mi are matrix algebras and we pose An “
M1 b . . . b Mn with the map ‡pT1 b . . . b Tn q “ T1 b . . . b Tn b , and
‡
An / Ak`1
Ïn Ïk`1
|
A
with Ïi pT1 b . . . b Ti q “ T1 b . . . b Ti b b . . ., the space A being the free vector space generated
by the symbols T1 b . . . b Tj b b . . .
Now an element of H reads
8
ÿ
w1i b . . . b wki i b vki i `1 b . . .
i“1
where some of the Tj are subject to actually be . We are now going to prove that the strong closure
of fipAq is BpHq. Let E be the set of finites sets of elements of the form w1 b . . . b wk b vk`1 b . . .
and the directed set I “ Eˆs0, 8r.
Problèmes et choses à faire
Proposition 71.13.
Let Â
Vk “ nk endowed
`Â with ˘the standard representation of nk p q. Then the representation
fi: k nk Ñ B k pVk , vk q fulfils
` ˘2 `â ˘
fipM q “ B pVk , vk q
k
Â
where M “ k nk .
Proof. No proof.
 Â8
Notice that 8 1 nk can differ from 1 n1k when the nk and n1k do not agree. In fact, we
have
8
â 8
â
nk “ n1k
1 1
± ±
if and only if k nk “ k n1k in the sense of generalised products: each prime factor arise the same
number of time in both side. For example, 2 b 2 b . . . “ 4 b 4 b . . . ‰ 3 b 3 b . . ..
Take for example Vk “ 2 for every k and then vk “ v, vk1 “ v 1 . In this case, the representations
fi and fi 1 are equivalent if and only if v “ ⁄v 1 . This provides an Hilbert sphere of inequivalent
irreducible representations.
Proof. No proof.
Let us see an example of that result. When ⁄ “ 0, we get the standard representation of
matrices on n , so that M0 “ BpH q. When ⁄ “ 1, we can show that
TrpT q “ xV, T V y
where V “ v b v b . . . b v b . . . is a trace on M1 while there does not exist any trace on BpH q.
We conclude that M1 ‰ M0 .
71.5. CANTOR 2769
71.5 Cantor
À
Let M “ 8 k“1 Mk with Mk “ nk p q that can be seen as SpantT1 b . . .u with Tk “ for
sufficiently large k. Now take the case nk “ 2 for all k. The algebra M naturally acts on the space
of locally constant functions on the Cantor set.
More generally, for arbitrary nk , one
± can think of M as an algebra of endomorphisms of the
space of locally constant functions on nk . Problèmes et choses à faire
Still to be developed.
Ï: M Ñ
(71.15)
T1 b T2 b . . . fiÑ Ï1 pT1 qÏ2 pT2 q . . .
where the product is finite because Tk “ for sufficiently large k. One can prove that Ï fulfills
(1) Ïp q “ 1,
(2) ÏpT ˚ T q • 0,
(3) ÏpT ˚ S ˚ ST q “ }S}2 ÏpT ˚ T q,
so
ÂNthat Ï is in particular a state on ÂNM . The normn1}S} is the nfollowing. We know that S P
1 M k Ñ M but, by construction, 1 Mk “ Endp b ... b N q. The norm of S is taken as
Définition 71.15.
A factor of type II1 is an infinite dimensional factor M which accepts a non vanishing linear
functional Tr : M Ñ such that
— TrpST q “ TrpT Sq,
— TrpT ˚ T q • 0,
— the function Tr is continuous for the ultraweak topology.
So we have ª ª
pgf qpxqdpgµqpxq “ f pxqdµpxq. (71.18)
X X
Notice that if f P L8 pX, µq, we have gf P L8 pX, µq, but when f P L2 , there are no guarantee
that gf P L2 .
Let µ1 and µ2 be two measures on the set X. One says that µ2 is absolutely continuous
with respect to µ1 if every µ1 null set is µ2 null.
Proposition 71.17.
There exists an unique function
df µ
: X Ñs0, 8r
dµ
such that ª ˆ ˙ ª ª
dgµ
pgf qpxq pxqdµpxq “ pgf qpxqdpgµqpxq “ f pxqdµpxq. (71.19)
dµ
´ ¯
Applying g ´1 to the function pgf qpxq dgµ
dµ pxq and applying the theorem, we see that for every
function f , we have ª ˆ ˙ ˆ ´1 ˙ ª
dgµ dg µ
f pxq pgxq pxq “ f pf qdµpxq,
dµ dµ
so that ˆ ˙ ˆ ˙
dgµ dg ´1 µ
pgxq pxq “ 1. (71.20)
dµ dµ
Using the function provided by that theorem, we define
ˆ ˙1
dgµ 2
Ug f “ pgf q · , (71.21)
dµ
so that }Ug f }L2 pX,µq “ }f }L2 pX,µq . Moreover we have the following two important relations
(1) Ug Uh “ Ugh ,
(2) Ug Mf Ug´1 “ Mgf
where Mf stands for the operator of pointwise product with f . The second relation implies
Ug Mf “ Mgf Ug , thus we have
n
ÿ
M :“ tMf , Ug u2 “ t Mfi Ugi u (71.22)
i“1
(because the double commutant is the strong closure) which is a ˚-algebra of operators. Notice
that Ug Ug˚ “ 1 because Ug is an isometry.
We say that an action G ˆ M Ñ M is ergodic when gf “ f for every g P G implies that f is
constant almost everywhere.
Lemme 71.18. ` ˘
If the action G ˆ X Ñ X is ergodic, then M “ B L2 pX, µq .
Proof. Let us first study the commutant tMf , Ug u1 which is of course contained in tMf u1 . But we
know that the commutant of L8 pX, µq is L8 pX, µq, so that
Is there an element in L8 pX, µq which commutes with all the elements of the form Ug , or in other
words, is there a f P L8 pX, µq such that Mf “ Ug Mf Ug˚ “ Mgf ? The only element f such that
Mf “ Mgf for every g P G is f “ 1, since the action is ergodic.
That lemma shows that we didn’t construct any interesting von Neumann algebras in the
ergodic case.
Ug : H Ñ H
ˆ ˙1 (71.24)
dgµ 2
´1 ´1
pUg Ïqpx, hq “ Ïpg x, g hq pxq
dµ
The introduction of the Radon-Nikod˝m function serves to preserve the norm. The so defined
operators have the following properties:
(1) Ug Mf Ug˚ “ Mgf ,
(2) Ug Uh “ Ugh ,
(3) Ug´1 “ Ug´1 “ Ug˚ .
Now we define n
ÿ
M pG, Xq “ tUg , Mf u2 “ t Mfi Ugi u, (71.25)
i“1
and we will show (later) that
Théorème 71.19.
If the action is ergodic, then M pG, Xq is a factor.
We emphasize the progress: lemma 71.18 says that the commutant is trivial while now the
center only is trivial.
As an example, take X “ S 1 Ä and G “ , the action being g · z “ e2fii◊g z for some
irrational ◊. One can prove, using Fourier transform, that this equation is ergodic.
Let us now define
Ï : M pG, Xq Ñ
(71.26)
T ބ xf0 , T f0 y
where f0 P L2 pX ˆ G, µq is defined as follows
#
µpXq´2 if g “ e
f0 pg, xq “
0 otherwise.
We show that the so defined Ï is a trace over M pG, Xq, i.e. it satisfies ÏpT Sq “ ÏpST q for every S,
T P M pG, Xq. We know that M pG, Xq is generated by expressions of the form Lf Ug . When g ‰ e,
the functions f0 and Ug f0 have disjoint support, so that ÏpMf Ug q “ 0. If g “ e, the computation
is easy and we finally find
# ≥
1
µpXq X f pxqdµpxq if g “ e
ÏpMf Ug q “ (71.27)
0 if g ‰ e.
2772 CHAPTER 71. VON NEUMANN ALGEBRAS
∞
It is easy to check that this expression is a trace on t i Mfi Ugi u. Since BpHq has no trace, we
know that M pG, Xq is a non trivial von Neumann algebra. Stated in a different way, what we
just proved is that, provided that µ is a G-invariant probability measure (i.e. µpXq “ 1), the
formula ª
`ÿ ˘
ϵ Mfh Uh “ fe pxqdµpxq. (71.28)
hPG X
Proposition 71.20.
If µ is not invariant (but still µpXq “ 1), then ϵ is still a state, but no more a trace.
because Ug˚ “ Ug´1 “ Ug´1 commutes with the Mf . Now, according to the expression (71.28), the
function ϵ pick up the identity component and integrates. So we have
´` ÿ ª
˘˚ ` ÿ ˘¯ ÿ
ϵ Mfh Uh Mfh Uh “ h´1 pfh˚ fh qdµ
h X
ÿ ª
“ `h |fh |2 ph´1 xqdµpxq • 0.
X
Notice that the group does not act on H while in the previous constructions, it did act on X. We
still have the relation
Uh MT Uh˚ “ MhT . (71.29)
Now, we form the von Neumann algebra
unique (up to constant multiple) Borel measure m such that each compact set has finite measure
and which is in the same time the Haar measure: for every compactly supported functions on G,
ª ª
f phgqdmpgq “ f pgqdmpgq.
G G
H b L2 pG, mq “ L2 pG, H q,
and MT is defined by the same formulas as before. The von Neumann algebra that we obtain is
denoted by M pG, N q.
Proposition 71.21.
The module l2 pGq over M pGq is projective and finitely generated.
Proof. The fact that l2 pGq as module over M pGq is finitely generated comes from the fact that fe
by itself generates the basis tfh u as we just said.
In order to prove that the module is projective, we will prove the condition (2) of proposi-
tion 52.13. Let M be a M pGq-module and fl : M Ñ l2 pGq be a surjective module map. Consider
any › P M such that flp›q “ fe , define spfe q “ › and extend by linearity and action of M pGq. The
so defined map s obviously fulfils fl ˝ s “ Id |l2 pGq .
That statement and the proof are correct uhm? I use them on page 2814.
For each S P M pGq1 , we define fS “ Spfe q P l2 pGq. Now if S P M pGq X M pGq1 , we have
(1) SUh “ Uh S,
(2) SUh fe “ Uh Sfe .
The operators Wk defined by
pWk f qpgq “ f pgkq
commute with Uh . So they commute with M pGq and with S. Therefore we have
If G has no finite conjugacy class (except teu), then fS has to be a multiple of fe because its
norm would contains infinitely many constant non zero terms. But when S1 and S2 belongs to
M 1 pGq with fS1 “ fS2 , then S1 “ S2 because S1 fh “ S1 Uh fe “ U1 S1 fe “ Uh fS1 , while the same
computation with S2 gives S2 fh “ Uh fS2 . So S1 and S2 agree on a basis.
For this reason, if we assume that G has infinite conjugacy class, we have M pGq1 XM pGq “ .
The most famous example of such a group is G “ Fk , the group of formal words of 1 . . . k with
k ° 1.
Now fix an element of finite order h P G, and the natural homomorphism Ñ G given by
n fiÑ hn . The group G has a natural equivalence relation g1 „ g2 if and only if there exists a n P
such that g1 “ g2n . We denote by G{ the quotient of G by this relation. One class of this space is
g “ tg n tel que n P u,
and we have à
l2 pGq “ l2 p g q. (71.34)
gPG{
The operator Ug acts on l3 p gq by translation: Ug g n “ g n`1 . Now we fix a g P G and we make the
identifications
l2 p gq » l2 p q » L2 pS 1 q, (71.35)
the latter identification being given by the Fourier series of a function on the circle (seen as a
periodic function on ). In that framework, Ug acts by translation of 1 on l2 p q:
À
We define the operator U P gPG{ l2 p gq as acting on l2 p gq with Ug . That allows us to consider
CpS 1 q Ñ M pGq
(71.36)
z n ބ U n
A projection is determined by the closed space of its range, so one can order the projection by
ordering the closed subspace of H .
71.8. MORE ABOUT PROJECTIONS 2775
Lemme 71.22.
We have
dimpR1 _ R2 q “ dim R1 ` dim R2 ´ dimpR1 ^ R2 q (71.39)
for any two projections R1 and R2 in M .
For two projections p and q, we write that p ® q if there exists a partial isometry V such that
(1) V ˚ V “ p,
(2) V V ˚ § q.
One proves that this determines a partial ordering on the projections.
Proposition 71.23.
If p and q belong to M , then the projections p _ q and p ^ q belong to M too.
Lemme 71.24.
If tT– u is a net of selfadjoint bounded operators such that
(1) sup– }T– } † 8,
(2) if – ° —, then xv, T– vy ° xt, T— vy for every v P H .
Then tT– u strongly converges to an operator T , see condition (57.63).
Proof. We want the limit T to fulfil xv, T vy “ lim–Ñ8 xv, T– vy. From weak-compactness of the
unit ball, that formula defines the quadratic form Qpvq “ lim–Ñ8 xv, T– vy with }v} “ 1. The form
Q is then defined on the whole space H by homogeneity: Qp⁄vq “ ⁄2 Qpvq. The polarization
identity
4 Rexv, T– wy “ xpv ` wq, T– pv ` wqy ´ xpv ´ wq, T– pv ´ wqy
defines Qpv, wq and then defines the value of xv, T wy for every v and w. This is the candidate to
be the strong limit of T– . The question is now to know if this is an actual strong limit.
Since the net is increasing, we have xv, pT ´ T– qvy • 0, so that T ´ T– is positive which implies
that pT ´ T– q1{2 is well defined. We have
In short, that lemma claims that M contains the limits of all bounded increasing nets of
selfadjoint operators. In particular, for a net of projections P– P M with – P X,
™ ©
P– P M, P– P M.
– –
2776 CHAPTER 71. VON NEUMANN ALGEBRAS
Proposition 71.25.
Let tP– u and tQ– u to be two nets of projectors on the same directed set, and suppose that we have
operators T– : ImagepP– q Ñ ImagepQ– q such that
(1) sup– }T– } † 8,
(2) T– |ImagepP— q “ T— ,
`ö ˘ `ö ˘
then there exists an operator T : Image – P– Ñ Image – Q– with
T |ImagepP– q “ T– .
Proof. No proof.
Problèmes et choses à faire
I think that p1 _ p2 _ . . . is now well defined when pi are projections. Indeed,
ö the projections Pk “ p1 _ . . . pk form a net and, by the proposition,
the limit belongs to M . More generally, if A is a set of projections, for pPA p, one considers the set of parts of A (which is partially ordered)
ö
and then the net p– “ pP– p where – is a part of A.
Lemme 71.26.
Let M be a factor. If P and Q are nonzero projections in M , then there is a T P M such that
P T Q ‰ 0.
Proof. Let U pM q be the unitary group of M . We saw that if T commutes with U pM q, then it
commutes with all M . Assume that P T Q “ 0 for every T P M , then in particular it holds for
T “ U P U pM q and we have U P U ˚ Q “ 0.
ö
Consider the operator R “ U PU pM q U P U
˚ which is a projection in M . Since the set
Lemme 71.27.
Let M be a factor and P , Q two nonzero projections in M , then there exists a nonzero partial
isometry V P M such that V ˚ V § P and V V ˚ § Q.
Proof. Let T be such that P T Q ‰ 0 (by lemma 71.26), and V be the partial isometry part of
QT P . The image of V ˚ V is at most the one of QT P which is smaller (or equal) to the image of Q,
so V ˚ V § Q. For the same reason, the image of V V ˚ is contained in P ˚ T ˚ Q˚ H “ P Y ˚ QH Ñ
ImagepP q. Then V V ˚ § P .
Proposition 71.28.
Let M be a factor and P , Q two nonzero projections in M , then there exists a nonzero partial
isometry V P M such that
(1) V ˚ V § P ,
(2) V V ˚ § Q
and either V ˚ V “ P , or V V ˚ “ Q or both.
Proof. Let X be the set of partial isometries V such that V ˚ V § P and V V ˚ § Q. We write
PV “ V ˚ V and QV “ V V ˚ . We define a partial ordering on X by V1 § V2 if PV1 § PV2 , QV1 § QV2
and V2 PV1 “ V1 . In that case, V2 is some kind of extension of V1 .
If one considers an increasing sequence V1 § V2 § . . ., the Zorn’s lemma assures the existence
of a V bigger than all the elements of the sequence. We claim that this V is the solution of the
71.8. MORE ABOUT PROJECTIONS 2777
The dimension of the von Neumann algebra M is the set of equivalence class of projections
in M . This is a linearly ordered set.
Let M “ M pGq where G is a group with only infinite conjugacy classes (but the one of identity).
We saw that there is a trace · : M Ñ , and that @t P r0, 1s, there exists a projection whose trace
is t.
Proposition 71.29.
If T • 0 is an element of M such that · pT q “ 0, then T “ 0.
Proof. We know the injection M Ñ l2 pGq defined by T fiÑ fT “ T fe . Since the right translation
commutes with every T1 (because T1 pgfe q “ gpT1 fe q), we have T1 fg “ 0 for every g if T1 fe “ 0.
On the other hand,
· pT1˚ T1 q :“ xfe , T1˚ T1 fe y “ }T fe }2
which is positive. So if · pT1˚ T1 q “ 0, then T1 fe “ 0, and so T1 fg “ 0 which proves that T1 “ 0.
This concludes the proof that T • 0 and · pT q “ 0 imply T “ 0 because every positive element
reads as a product T1˚ T1 .
It is well defined because if P „ Q, then we have a u such that P “ uu˚ and Q “ u˚ u, so that
· pP q “ · puu˚ q “ · pu˚ uq “ · pQq by cyclic invariance of the trace. The map · 1 preserves the order
because P ´ Q is positive when P • Q.
Proposition 71.30.
The map · 1 is a bijection
Proposition 71.31.
If p “ ‘i pi and q “ ‘i qi are orthogonal sums of projectors with pi „ qi , then p „ qi . If moreover
pi † qi for every i, then p † q.
Proof. No proof.
Let T P BpH q. The closed spaces ImagepT q and ImagepT ˚ q are the left support and right
support of T .
Théorème 71.32.
Let T be an invertible operator, then it reads under the form
T “ |T |U
2778 CHAPTER 71. VON NEUMANN ALGEBRAS
where U is unitary. This decomposition is the unique one of the form T “ SU with S positive and
U unitary.
?
Proof. Let T “ P U , then T T ˚ “ P U U ˚ P “ P 2 , then P “ T T ˚ and then U “ P ´1 T . That
proves unicity of the decomposition. For existence, we pose U “ pT T ˚ q´1{2 T and P “ pT T ˚ q1{2 .
Then we check
U U ˚ “ pT T ˚ q´1{2 T T ˚ pT T ˚ q´1{2 “ . (71.41)
Now, the operator U is the product of two invertible operators, so that it is invertible. Thus the
fact that U U ˚ “ forces U ˚ U “ .
Proposition 71.33.
Every T P BpM q can be written under the form
T “ |T |V
Proposition 71.34.
The left and right support of any operator in BpH q are equivalent projections.
f V “ V V ˚ V “ V e “ V,
pP M 1 P q1 “ P M P, (71.43)
and
pP M P q1 “ P M 1 P (71.44)
where we see P M P as an algebra of operators on P H Ñ H .
For the second claim, suppose Q P pP N P q1 . We are looking for a R P M 1 such that Q “ P RP .
Notice that Q itself does not belong to M 1 in general. Take for example
¨ ˛ ¨ ˛ ¨ ˛
‹ 0 ‹ 1 1
M “ ˝0 ‹ 0‚, P “ ˝ 1 ‚, Q“ ˝ 0 ‚.
‹ 0 ‹ 1 0
¨ ˛
a
we have M 1 “ ˝ b ‚, so that Q does not lie in M 1 , but
a
¨ ˛
1
Q“P˝ 0 ‚P
1
anyway. We see on that example that the matrix R to be chosen is bigger than Q.
Proof of lemma 71.35. Let Q P pP M P q1 and R be the projection onto the closure of M QH . The
latter space being invariant under M , the projection R belong to M 1 by lemma 71.8. We are
now going to prove that Q “ P RP . Since P “ P 2 commutes with R, the operator P RP is
the projection onto the intersection of the target spaces of R and P , because P RP “ P R “ RP .
Therefore Q is smaller than P , so that M QH “ M P QH . That proves that P RP is the projection
on
P M QH “ P M P QH “ QP M P H “ QP H “ QH
where we used the fact that Q P pP M P q1 . That concludes the proof that Q “ P RP .
Lemme 71.36.
If M is a factor and P , a nonzero projection in M , then the map
M 1 Ñ P M 1P
(71.45)
S ބ P SP
is a ˚-algebra isomorphism.
Proof. The fact that the map is multiplicative and surjective is clear. The only point to prove is
injectivity. So we will prove that S “ 0 under the assumption SP “ 0. We have 0 “ M SP “ SM P ,
so that S vanishes on M P H .
The latter subspace is obviously invariant under M , but also under M 1 because for every
S1 P M 1 , we have S1 M P H “ M P S1 H Ñ M P H . We conclude that the projection onto M P H
lies in M 1 X M . Hence that projection must be a multiple of the identity or zero. Since that
space contains at least P H , the zero possibility is ruled out. Thus the projection onto M P H is
the identity and M P H “ H . Now, SM P “ 0 implies S “ 0 which concludes the proof of the
lemma.
Définition 71.37.
Let M be a factor. It is
— of type I if M contains a non vanishing minimal projection,
— of type II if M contains a non vanishing finite projection (and is not of type I)
— of type III if no projection in M is finite (but the vanishing one).
2780 CHAPTER 71. VON NEUMANN ALGEBRAS
A type II factor is of type II1 if it is finite and of type II8 if it is not finite.
Let us now examine what are the type I factors. Suppose that M is a factor of type I and P P M
is a minimal projection. Then P “ P M P Ñ BpP H q, and formula (71.44) in lemma 71.35 makes
pP M P q1 “ P M 1 P “ BpP H q. Now the map of lemma 71.36 is an isomorphism, then
M 1 » BpP H q
as ˚-algebra. Thus M 1 is a factor. This factor is moreover of type I because BpP H q has minimal
projections, namely projections on one dimensional spaces.
If we repeat the same argument with M 1 instead of M , we obtain that
M » BpP 1 H q (71.46)
Proposition 71.38.
If M is a factor of type I, there exists a separable Hilbert space H such that
M “ BpH q. (71.47)
Proposition 71.39.
If M is of type In with 1 § n † 8, then M » np q and the unique trace is the ususal matrix
trace up to renormalization:
1ÿ
n
TrpT q “ Tii . (71.48)
n i“1
Proof. No proof.
In particular we have that the composition of the two minimal projection P P 1 “ P 1 P is a rank
one projection in BpH q. Now consider Q “ P P 1 and v be the unital vector in the target space of
Q, i.e. Qv “ v and }v} “ 1.
Proposition 71.40.
Let M be a factor of type I and P , a minimal projection. In the same way, let P 1 be a minimal
projection in M 1 . Let Q “ P P 1 and v P QH such that }v} “ 1. Then
M 1Q Ñ P H
(71.49)
SQ ބ Sv
is an isomorphism.
Proof. Since Q projects on the space spanned by v, the fact that S1 v “ S2 v implies S1 “ S2
and the injectivity is proved. For surjectivity, we know that, M being a factor of type I and P a
minimal projection, we have P “ P M P Ñ BpH q where the last equality has to be understood in
the sense of that T P P M P restricts to an operator on P H , and that this restrictions completely
defines T . But as operators on P H , we have pP M P q1 “ P M 1 P (lemma 71.35), while lemma 71.36
assures that the latter algebra is M 1 . Thus M 1 “ BpP H q and M 1 v “ P H .
Proposition 71.41.
In the same way, we have that
M Q Ñ P 1H
(71.50)
T Q ބ T v
is an isomorphism
Proof. No proof.
71.9. TYPE I FACTOR AND FACTORIZATION 2781
Proposition 71.42.
Let M be a factor S P M and T P M 1 , then if ST “ 0, then S “ 0 or T “ 0.
Proof. Let P be the projection on the right support of S, i.e. onto ImagepS ˚ q “ pker SqK . This
operator is equal to the limit 1
P “ lim pS ˚ Sq1{n .
nÑ8
Proposition 71.43.
Let M Ñ BpH q be a factor. Then the map
M b M 1 Ñ BpH q
(71.51)
S b T ބ ST
so that tST u2 “ BpH q, which proves that tST u is strongly dense in BpH q.
We have elements Si P M and Ti P M 1 (i “ 1, . . . , n) such that
n
ÿ
Si Ti “ 0, (71.52)
i“1
and we are going to prove that the set tSi u is not linearly independent, so that all the Si , or all
the Ti vanish. An element of M b M 1 can be decomposed under the form (71.52) in several ways.
Let Q be the projection on the closure of the space
(
pT T1 v, . . . T Tn vq P H ‘ . . . ‘ H tel que T P M 1 , v P H . (71.53)
On the other hand, the operators which commute with all M 1pnq are element of n pM q, then
´ ¯1
Q P M 1pnq “ n pnq,
and for the same reason, the operators which commute with all (71.55) are elements of n pM
1 q,
so that
Q P n pM X M 1 q “ n p q.
1. To be proven.
2782 CHAPTER 71. VON NEUMANN ALGEBRAS
for every j. Since Q is a non vanishing projector, at least one of these relations is nontrivial. That
nontrivial relation shos that the Si ’s are vanishing. If not, we have Q “ 0, which means that
Ti “ 0.
71.10 Dimensions
71.10.1 Finite and infinite projections
A projection in a factor is infinite if it is equivalent to a proper subprojection; it is finite is
it is not infinite. The notion of finite and infinite projections descents to the equivalence classes.
A factor is finite when the identity is finite.
Proposition 71.44.
If P and Q are projections with P § Q and Q is finite, then P is finite.
Q “ P ` pQ ´ P q „ P 1 ` pQ ´ P q † Q,
the last inequality being assured by the vector v. What we proved is that Q is equivalent to a
subprojection of itself, which contradicts the fact that Q is finite.
Corollaire 71.45.
Let –, — P dim M . If – is finite and — infinite, then – § —.
n– “ – ` ¨ ¨ ¨ ` –,
looooomooooon (71.58)
n times
71.10. DIMENSIONS 2783
Proposition 71.46.
The sum – ` — fulfils
(1) If – ` — exists, then it is independent of the representative P and Q in – and —,
(2) It is commutative and associative: – ` — “ — ` – and – ` p— ` “q “ p– ` —q ` “,
(3) zero is the neutral: – ` r0s “ –.
Proof. No proof.
Lemme 71.47.
If – and — are finite, then – ` — is finite when it exists.
Proof. No proof.
Lemme 71.48.
If – is finite and — is infinite, then
(1) 8– exists,
(2) 8– “ —.
In particular, all the infinite projections are equivalent each other.
Proof. Since — is infinite and – is finite, the projection — contains a subprojection equivalent to –.
What remains in — is still an infinite projection, and thus still contains a subprojection equivalent
to –. Let – “ rP s and — “ rQs. We have
`ÿ ˘
Q“ Pn ` R
n
where for each n, the projection Pn is equivalent to P , and R is a finite projection which does not
contain a subprojection equivalent to P . Thus P contains a subprojection equivalent to R.
Notice that — can only contain a countable number of copies of – because one only has a
countable number of basis vectors in a separable Hilbert space.
Let us consider the projection
ÿ
— 1 “ rQ ´ Rs “ r Pn s “ 8–. (71.59)
One can construct a partial isometry implementing an equivalence between — and — 1 using an
Hilbert hotel argument. Problèmes et choses à faire
This proof has to be finished.
Lemme 71.49.
If P is a projection and if V ˚ V “ P , then V vanishes on the target space of P K .
Proof. It is evident that V ˚ does not vanish on the range of V (apart on zero) because
xv, V ˚ V vy “ xV v, V vy ‰ 0.
Lemme 71.50.
Let M be a finite factor, the every partial isometry V reads
V “ UP (71.60)
Notice that the decomposition V “ U P is a special feature of the finite factor case. In the
general infinite case, we have for example the operator which shifts the basis vectors
¨ ˛
0 0 0 0 ...
˚1 0 0 0 . . .‹
˚ ‹
S “ ˚0 1 0 0 . . .‹
˚ ‹
˚0 0 1 0 . . .‹
˝ ‚
.. .. .. .. . .
. . . . .
It has S ˚ S “ while SS ˚ ‰ .
Proposition 71.51.
Every element T of a finite factor reads
T “ U |T | (71.61)
with U unitary.
Proof. One guess the form of U by the decomposition (71.61): U “ T pT ˚ T q´1{2 . One checks that
U ˚ U “ . Since we are in a finite factor, the unit cannot be equivalent to something else than
itself, so U U ˚ “ is forced, and U is thus unitary.
Proposition 71.52.
If N is a finite factor, then so is 2 pN q.
Proof. Let us suppose that we have a partial isometry V P 2 pN q such that V V ˚ “ 2 . We have
to prove that V ˚ V “ 2 ; if not, the identity would be equivalent to a subprojection. In other
words, we have to prove that V is invertible. It will be done by constructing an invertible operator
W such that V ˚ W ˆ is invertible.
˙ In that case, V ˚ is invertible and so is V .
a b
If we set V “ , with a, b, c, d P N , the relation V ˚ V “ 2 imposes among others
c d
relations
a˚ a ` c˚ c “ Id (71.62a)
b a`d c“0
˚ ˚
(71.62b)
Relation (71.62a) says a that |a|2 ` |c|2 “ 1, so that in the sense of the continuous functional
calculus, 2
‰ we have |a| “ 1 ´ |c| . We deduce that |a| is a limit of polynomials in 1 ´ |c| and that
2
“
|a|, |c| “ 0.
Since a and c belong to a finite factor, proposition 71.51 provide unitary elements u and v of
N such that a “ u|a| and c “ v|c|. Now we consider the unitary element
ˆ ˙ˆ ˙
u 0 |a| ´|c|
W “
0 v |c| |a|
71.10. DIMENSIONS 2785
but a˚ u|c| “ |a|u˚ u|c| “ |a||c| “ |c||a|, so that the upper-right element is actually zero. We are
left with ˆ ˙
˚ 1 0
V W “
0 x
in which we want to prove that x is invertible. Using the fact that W is unitary, we have
pV ˚ W qpV ˚ W q˚ “ V ˚ V “ , then
ˆ ˙ˆ ˙ˆ ˙ ˆ ˙
1 0 1 0 1 0 1 0
“ .
0 x 0 x˚ 0 xx˚ 0 1
Thus we have xx˚ “ in N , which in turn imposes x˚ u “ because N is a finite factor. We have
finished to prove that V ˚ W is invertible.
Lemme 71.53.
If P and Q are projections with P Q “ 0, then P is equivalent to a subprojection of Q, or Q is
equivalent to a subprojection of P .
Proof. No proof.
Proposition 71.54.
If R is not a minimal projection, then there exist subprojections S, T of R such that S „ T and
S K T.
Proof. Since R is not minimal, we have a non vanishing projection S † R. Let consider S1 “ R´S.
One checks that SS1 “ 0. Now S or S1 is equivalent to a subprojection of the other. Let
S „ T § S1 . Now, we have S „ T and ST “ 0 which means S K T .
and ˆ ˙ˆ ˚ ˙ ˆ ˙ ˆ ˙
P 0 P V˚ P PV ˚ P 0
“ “
V 0 0 0 V P˚ R 0 R
ˆ ˙
P 0
because ImagepV q “ ImagepQq implies P V “ V P “ V P “ 0. Now,
˚ ˚ ˚ is a subprojection
0 0
ˆ ˙
P 0
of because R is a subprojection of P . Thus we have
0 P
ˆ ˙ ˆ ˙ ˆ ˙
P `Q 0 P 0 P 0
„ § “ finite. (71.63)
0 0 0 E 0 P
Proposition 71.55.
If P ∞is any finite projection, one can find a sequence of equivalent projections tPi u such that
“ Pi .
Proof. No proof.
A consequence of that proposition is that any factor is semifinite or purely infinite, while a
finite factor is always semifinite.
Remarque 71.56.
A factor of type I is semifinite because a minimal projection is finite. Indeed, when P is infinite,
the projection Q such that Q „ P and Q § P contradicts minimality of P .
Proposition 71.57.
A factor is semifinite if and only if it reads pfiniteqb̄ptype Iq.
Proof. If a factor reads pfiniteqb̄ptype Iq, the type I part is a BpH q in which one can take as
approximation of the identity the sequence of projections Pi given by
#
ej if j § i
Pi ej “ (71.64)
0 otherwise
Thus, in order to understand the semifinite factors, it is sufficient to understand the finite
factors.
Lemme 71.59.
Let M be a finite factor. If P „ P K and Q „ QK , then P „ Q.
This lemma means that there is only one way tu cut H in two equal subspaces.
71.10. DIMENSIONS 2787
What is the exact statement which says that a projection is equivalent to a subprojection of the other, or the contrary?
1
“ rP s P dim M (71.66)
2
1 1
where P „ P K . Now we look at the algebra P M P , which posses its own 2 that is called 4 in M .
So we set the definition
1
“ rQs P dim M (71.67)
4
3
where Q § P and Q „ P ´ Q. Notice that P ´ Q “ QK in P H . In the same way, we define 4 by
3
“ rP ` Rs P dim M (71.68)
4
Proof. No proof.
71.10.4 Summary
When M is a finite factor, it has two possibilities: it contains or not a minimal projection.
where N can eventually take the value 8, as shows the example M “ Id.
If the factor is not finite, there are two possibilities: it is semifinite or purely infinite. In the
semifinite case
Lemme 71.61.
Let P be an infinite projection in a factor and P “ R ‘ S (orthogonal sum) with R „ P and S ‰ 0.
Then rP s “ 8rSs.
E „ E1
A1 „ A1
E1 „ E2
A2 „ A2
E2 „ E3
..
.
∞ ∞
On the left
∞hand side
∞ we have E ` n Qn∞while on the right hand side we have n Qn . Thus we
have E ` n Qn „ n Qn and then P „ n Qn .
Corollaire 71.62.
If M is a purely infinite factor, then dim M “ t0, 8u
Proof. In a purely infinite factor, zero is the only finite projection, thus any projection reads
P “ R ` S with P „ R and S ‰ 0. The aim is to prove that rP s “ Id, in such a way that there
exists only one infinite.
We have Id “ Q ` S for a certain Q „ Id. The S here can be chosen the same as the S of
P . Indeed Id “ P K ` P “ P K ` R ` S. Using the equivalences P K „ P K and P „ R, we see
that Id “ P K ` P „ P K ` R, so that Id “ Q ` S with Q “ P K ` R. Using lemma 71.61, we
have rP s “ 8rSs, and then using the lemma again on rP s, we get rP s “ 8rSs, which proves that
rP s “ rIds.
71.11. TRACIAL FUNCTIONAL 2789
Lemme 71.63.
A linear tracial functional on a von Neumann algebra is a trace.
Proof. Let U be unitary, using the tracial property we find · pU ˚ A˚ AU q “ · pAA˚ q, so that
· pU ˚ BU q “ · pBq (71.72)
for every positive B. Since any element of M reads as composition of four positive, the relation
(71.72) holds in fact for every B in M . Since every element C P M can be written as C “ U ˚ B,
we have
· pCU q “ · pU ˚ BU q “ · pBq “ · pU Cq,
and we have · pCU q “ · pU Cq for every C and U unitary. Now, every element is a combination of
four unitary, so that · pCDq “ · pDCq for every C and D, which means that · is a trace.
Notice that the existence of a faithful normal trace on M implies that M is finite because if
V ˚ V “ 1,
· p1 ´ V V ˚ q “ 1 ´ · pV V ˚ q “ 1 ´ · pV ˚ V q “ 1 ´ · p1q “ 0,
which, from faithfulness, implies that V V ˚ “ 1.
Let ‘ ° 0, an ‘-trace is a normal state Ï : M Ñ such that
Lemme 71.64.
A normal state on a finite factor is an ‘-state if and only if P „ Q implies ÏpP q § p1 ` ‘qÏpQq.
Proof. No proof.
Lemme 71.65.
If Ï is an ‘-trace, then we have
1 dimpP q
ÏpP q § § p1 ` ‘qÏpP q (71.74)
p1 ` ‘q dimp q
for every projection P .
The lemma is now proved with the assumption (71.75). If that assumption
∞ does not hold, we can
build projections Pi with Pi Pj “ 0, dimpPi q “ dimp q{ni and P “ i Pi . Then we can use the
fact that Ï is normal and carry on the same computation as before.
Corollaire 71.66.
If Ï1 is an ‘1 -trace and Ï2 is an ‘2 -trace, then
}Ï1 ´ Ï2 } Ñ 0
when ‘1 , ‘2 Ñ 0.
Proof. No proof.
The interest of this lemma resides in the fact that if Ïk is an ‘k -trace with ‘k Ñ 0, we have a
Cauchy sequence, so that limkÑ8 Ïk is an actual trace. We are thus not obliged to build traces,
but ‘-trace are sufficient.
Lemme 71.67.
Let Ï be a faithful normal state. If ‘ ° 0, there exists a non vanishing projection P P M such that
ÏpA˚ Aq § p1 ` ‘qÏpAA˚ q
ÏpRq • ÏpR1 q.
m “ mintk tel que kÏpRq • ÏpR1 q for every pR, R1 qu. (71.78)
We just saw that 1 belongs to the set, so m § 1. Let us suppose that m “ 0. One typical projection
orthogonal to R is ´ R, but Ïp ´ Rq ‰ 0 because Ï is faithful and ´ R ‰ . So we conclude
that m cannot be zero and
m Ps0, 1s.
Now, take ‘ ° 0 and choose a pair pR, R1 q which fulfils the two conditions and such that
which shows that Ï is an ‘-trace and that P 1 is the answer to the lemma.
Proposition 71.68.
There is an ‘-trace on M .
Proof. We just saw that normal states do exist, so we apply lemma 71.67 and we find a projection
P for which we suppose dimpP q “ 1{n, and we choose P1 , . . . Pn such that Pi Pj “ 0 (i ‰ j) and
Pi „ P . We define the partial isometries Vi by
Vi˚ Vi “ P, Vi Vi˚ “ Pi .
Now we define ÿ
ÂpAq “ ÏpVi˚ AVi q
i
where Ï is the normal faithful state on P M P . Thus we have
ÿ ÿ ÿ ÿ
ÂpAq “ ÏpVi˚ AVi q “ ÂpVi˚ AVi q “ ÂpAVi Vi˚ q “ ÂpAPi q,
and we find
ÿ
ÂpA˚ Aq “ ÏpVi˚ A˚ AVi q
i
ÿ
˚ ˚ ˚
“ ÏpV i A Vj V
looomooon j AVi q
loomoon
ij
PP M P PP M P (71.83)
ÿ
§ p1 ` ‘q ÏpVj AVi Vi˚ A˚ Vj q
˚
ij
§ p1 ` ‘qÂpAA˚ q,
which means that  is a trace.
2792 CHAPTER 71. VON NEUMANN ALGEBRAS
Proposition 71.69.
Every finite factor accepts an unique faithful, normal and tracial state. Moreover, if P and Q are
projections and · is the trace, then
P „ Q ô · pP q “ · pQq. (71.84)
Proposition 71.70.
If M is a factor of type II1 , then
t· pP q tel que P P PpM qu “ r0, 1s. (71.85)
where · is the trace given by proposition 71.69.
Proof. No proof.
and
b / Cp pX̃q b / Cp´1 pX̃q b
... / ...
with b ˝ b “ 0. The chain space Cp pX̃q is not only a group, but a module over rfis.
We are now going to look at the tensor product von Neumann algebra
Mp pX̃q “ M pfiq b rfis Cp pX̃q. (71.86)
In the right hand side, M pfiq is the` von Neumann
˘ algebra described in subsection 71.7.5, which is
generated by the operators Ug P B l2 pfiq defined by Ug pÏqpgq “ Ïph´1 gq, and the tensor product
is the one defined in equation (52.12).
We can then look at the complex of M pfiq-modules
b/ b b / Mp´1 pX̃q b
... Mp`1 pX̃q / Mp pX̃q / ...
and look at the corresponding Betty numbers which are integers numbers associated with any
topological space.
Proposition 71.71.
Let M be a type II1 -factor and E be a separable M -module. There exists a projection P P 8 pM q
such that E » H8 P . All such projections are Murray-von Neumann equivalent.
dimM E “ Tr p (71.88)
` ˘ ` 2 2
˘
where p P P 8 pM q is the projection such that L pM q b l p » E and whose existence is given
by proposition 71.71.
Let M be a factor of type II1 with the unique trace · given by proposition 71.69. We denote
by L2 pM, · q the Hilbert space of its GNS representation, and fi· the representation. Since · is
unique, we will simply denote them by L2 pM q and fi. We denote by P L2 pM q a cyclic vector. If
fipT q “ 0, we have
· pT ˚ T q “ }fipT q }2 “ 0,
so`that T ˘ “ 0 and the representation
` ˘is faithful. We can thus identify T P M with fipT q P
B L2 pM q and we have M Ä B L2 pM q . The basics properties of the GNS construction say also
that M “ L2 pM q and · pT q “ xT , y for every T in M .
If T “ 0, then 0 “ }T }2 “ · pT ˚ T q, so that T “ 0. That means that is separating for M .
Lemme 71.72.
If M Ñ BpH q is a von Neumann algebra, a vector in H is cyclic for M if and only if it is
separating for M 1 .
fil p›qpT 1 q “ T 1 › (71.89a)
fir p›qpT q “ T › (71.89b)
` ˘ ` ˘
for T 1 P M 1 and T P M . The domains are dom fil p›q “ M 1 and dom fir p›q “ M . These
operations are well defined because is cyclic and separating for M and M 1 .
If fil p›q extends to a bounded operator on L2 pM q, one says that › is left bounded., and the
(necessarily unique) extension is still denoted by fil p›q. We do the same for fir . Let now the map
J: M ÑM
(71.90)
T fiÑ T ˚ ,
J : L2 pM q Ñ L2 pM q (71.91)
That map J is the modular conjugation operator for the factor M of type II1 .
The algebra JM J acts on L2 pM q as well as on L2 pJM Jq. If › P L2 pM q, the action is
pJSJq› “ ›S ˚ . (71.92)
That one is inspired by the fact that JSJT “ JpST ˚ q “ T S ˚ . Since J 2 “ 1, the action of JM J
on L2 pJM Jq is given by
pJSJqpJT Jq “ JpST qJ. (71.93)
Let M be a type II1 factor. We denote H1 pM q “ L2 pM q, or simply by H1 when no confusion
is possible, and H8 “ L2 pM q b l2 . The algebra M acts on H1 by
and M acts on H8 by
fi8 pSq “ fi1 psq b Idl2 . (71.95)
We also note 8 pM q “ M b Bpl2 q, that has to be understood as infinite matrices with elements
in M . It is a type II8 factor, and the trace is a follows. If p P 8 pM q “ ppij q, then
8
ÿ
Trppq “ TrM ppii q. (71.96)
i“1
With that formula, when q P Ppl2 q has rank 1, then Trp1m b qq “ 1. Let q P n pM q and consider
the element of 8 pM q given by the following:
ˆ ˙
q 0
. (71.97)
0 0
∞ ∞
It’s trace, in 8 pM q, is ni“1 Trpqii q, but the trace of q in n pM q is n1 ni´1 Trpqii q. The normal-
ization is not the same in n pM q and in 8 pM q.
We denote by e11 , the element of n pM q given by
ˆ ˙
1
e11 “ . (71.98)
0
´ ` ˘¯
We have Trpe11 q “ 1{n, and fie pe11 q P P fir n pM q .
Lemme `71.73. ˘ ` ˘
The LM L2 pM q -modules LM L2 pM q and L2 pM q are isomorphic.
` ˘
Proof. The result comes from the fact that an element of LM L2 pM q is uniquely defined by its
value at 1 P M .
dim ` ˘ L2 pM q “ 1, (71.99)
LM L2 pM q
because ´ ` ˘ ¯
LM L2 pM q b l2 p “ L2 pM q (71.100)
Lemme 71.74.
We have the isomorphism
L2 pJM Jq » L2 pM q (71.101)
as JM J-module.
Proof. Let us prove that the map which extend the following is an isomorphism:
 : L2 pJM Jq Ñ L2 pM q
(71.102)
JSJ fiÑ S ˚
So we have to prove that for every a P JM J and › P L2 pJM Jq, we have Âpa›q “ aÂp›q. Using
the actions (71.92) and (71.93), if a “ JSJ and › “ JT J, we find
while
aÂp›q “ pJSJqÂpJT Jq “ pJSJqT ˚ “ T ˚ S ˚ . (71.104)
That prove that L2 pM q » L2 pJM Jq as JM J-modules.
4. it is not 1{n or anything like that because of the remark we did bellow the definition (71.96).
71.12. MODULES OVER VON NEUMANN ALGEBRAS 2795
Lemme 71.75.
We have ` ˘
LM L2 pM q “ LM,fir pe11 q pEn q (71.105)
where En “ L2 pM q ‘ ¨ ¨ ¨ ‘ L2 pM q (n terms).
Proof. An element in the right hand side of (71.105) is a right-linear map with respect to fir pe11 q,
i.e. a map f : E Ñ E, ` ˘
f p›1 , . . . , ›n qfir pe11 q “ f p›1 , . . . , ›n qfir pe11 q. (71.106)
Since the left hand side only depends on ›1 , the right hand side shows that f p›1 , ¨ ¨ ¨ , ›n q only
depends on ›1 . Now, the right hand side takes its values in L2 pM q, so that the left hand side
shows that f takes its values in L2 pM q.
Proposition 71.76.
Let H be a separable Hilbert space and M Ñ BpH q, a factor of type II1 . We have
(1) for every d P r0, 1s, there exists a M -module Ed such that dimM Ed “ d.
(2) There exists an unique d P r0, 8s such that H » Ed as M -module.
(3) The algebra M 1 is a factor of type II1 if and only if dimM H † 8.
(4) dimM L2 pM q “ 1.
(5) If tFn unP is a set of separable M -modules, then
` ˘ ÿ
dimM ‘n Fn “ dimM Fn . (71.107)
M
(8) We have
dimM 1 H “ pdimM H q´1 (71.110)
if M 1 is a factor of type II1
The answer is p “ IdM b projn where projn stands for the projection onto the first n components.
Using formula (71.96), we conclude that dimM En “ n.
2796 CHAPTER 71. VON NEUMANN ALGEBRAS
The answer is ˆ ˙
q 0
r“ (71.115)
0 0
whose trace is
8
ÿ n
ÿ
Tr 8 pM q
prq “ TrM prii q “ TrM pqii q “ n Tr n pM q
pqq, (71.116)
i“1 i“1
because of the discussion bellow the definition (71.96). This concludes the proof of (1).
We pass to the proof of (8). Since dimM H † 8, we know that H is a factor of type II1
by (3). We can thus suppose that H “ Ed for some d. One knows that M 1 “ JM J, so that
dimM 1 Ed “ dimJM J Ed . ` ˘
Let` us first work with d “` 1. In order to ˘ compute dimJM J L2 pM q , we are searching for
2 2 2
pPP 8 pJM Jqq such that L pJM Jq b l p » L pM q as JM J-modules. From lemma 71.74,
P “ e11 works.
Let us now study the case d “ n P .
Corollaire 71.77.
When M is a factor of type II1 , two M -modules are isomorphic if and only if they have same
dimension.
71.12.2 Dimension
We are going to associate any M “ M pfiq-module with a dimension in r0,
` 1s such˘ that for every
projection P P M , the dimension of the submodule M pfiqP is equal to dim M pfiqP “ TrpP q. We
will in fact give a dimension to every module over a von Neumann algebra accepting a trace. For
that, we use theory developed in section 52.4.
Théorème 71.78.
If M is a finite von Neumann algebra and if E is a finitely generated module over M , and if F is
any submodule, then the quotient map E Ñ E{ ClE pF q splits and furthermore the quotient space
E{Ē is projective and finitely generated.
Proof. Later.
Théorème 71.79.
Let M be a finite von Neumann algebra with a normal faithful tracial state Ï. There is an unique
function
dimÏ : tM u Ñ r0, 8s
such that
71.12. MODULES OVER VON NEUMANN ALGEBRAS 2797
0 / F1 /F / F2 / 0,
Proof. later.
Proposition 71.80.
If E is a finitely generated projective module over M , we have
` ˘
DimpF q “ Dim ClE pF q (71.118)
if F is any submodule of E.
One has to check that is it actually proved somewhere in the next pages.
We will not often explicitly write the trace Ï and simply write dimpEq. However, we will some-
times precise the ring over which we consider the module and denote by dimM pEq the dimension
of E as module over M .
71.12.2.1 Examples
Let M be a von Neumann algebra with a normal faithful tracial state Ï, and let us give some
examples of modules over M .
First, M itself is a free module of dimension 1 6 over M . If P is a projection, then M P is a
projective module and dimÏ pM P q “ ÏpP q. The fact that M P is a projective module comes from
the direct sum decomposition M “ M P ‘ M p ´ P q and the third characterisation if projective
module in proposition 52.13. Notice that M is a free module over M because M “ M .
Proposition 71.81.
For every element T P M , the set M T is a projective M -module.
Proof. Let P be the projection onto ImagepT q, so that P “ limnÑ8 pT T ˚ q1{n as already seen. We
are going to prove that M T » M P as M -module. For we define : M T Ñ M P by pST q “ SP .
In order to see that this is a good definition, suppose S1 T “ S2 T , then pS1 ´ S2 qT “ 0. But we
have
0 / MT / MP / M P {M T / 0,
dimÏ pM P {M T q “ 0.
As another example, take projections Pm with ÏpPm q “ 2´m , and let F be the infinite algebraic
direct sum M P1 ‘ M P2 ‘ . . .. By cofinality, it has at least the dimension of each of its submodule.
In other words, we have
71.12.2.2 Summary
We want to define, for each module, a dimension such that
— if E is a projective left module with E » M n P (see proposition 52.14) and P “ P 2 P
n pM q, ÿ
dimpEq “ ÏpPii q,
and does not depend on the choices.
— If E is not a finitely generated projective module, then
(
DimpEq “ sup dimpF q with F . (71.120)
Proof. Passing to the matrix algebra (lemmas 52.18 and 52.19), one can assume that the module
and the submodule are in fact singly generated.
A singly generated module over M has the form E “ M T Ñ M for some element T P M .
We saw during the proof of proposition 71.81 that M T » M P where P is the projection onto
ImagepT q. The fact that M P is a projective module was already argued on page 2797.
Proposition 71.83.
If E is a finite projective module over M and F is a submodule of E, then E{ ClE pF q is projective.
Proof. Once again, using the matrix trick, one can suppose that E is singly generated by an element
e. A left module map Ï : E Ñ M such that ÏpF q “ 0 is determined by the value of Ïpeq P M .
` ˘
The element Ïpeq is an operator on H and we define PÏ as the projection onto Image Ïpeq .
Now we define the module map
Ï̃ : E Ñ M
(71.121)
Ï̃pT eq fiÑ T PÏ .
This is well defined. Indeed if Se “ 0, then ÏpSeq “ SÏpeq “ 0, which implies that SPÏ “ 0. For
the same reason, kerpÏq “ kerpÏ̃q. We define
™
Q“ PÏ , (71.122)
Ï
71.13. POSITION OF SUBMODULES 2799
and
Â: E Ñ M
(71.123)
ÂpT eq fiÑ T Q
which is well defined because T Q “ 0 if and only if T P Ï “ 0 for every Ï, since the operator T is
bounded 7 . Problèmes et choses à faire
The explanation in the footnote is unclear; it has to be expressed in terms of nets. By construction, kerpÂq “ ClE pF q, while
ImagepÂq “ M Q, so that
E{ kerpÂq » M Q, (71.124)
while the latter is projective.
Corollaire 71.84.
When E is a finite projective module and F a submodule, we have an isomorphism
E “ ClE pF q ‘ E{ ClE pF q
Let us choose a representative r›s0 in each of the classes 8 . The following is the module isomorphism
we are searching for
 : E Ñ ClE pF q ‘ E{ ClE pF q
(71.126)
› fiÑ ÷ ‘ r›s0
where ÷ is the unique element of ClE pF q such that r›s0 ` ÷ “ ›.
When E is a finite projective module over M , we say that ClE p0q is the torsion submodule
of E.
Let us see an example. Let
E “ M {M T (71.127)
with T • 0 and ImagepT q “ H . We claim that the torsion submodule of E is E itself. A module
map Ï : E Ñ M is a module map M Ñ M composed with a projection, or in other words a module
map Ï : M Ñ M such that ÏpT q “ 0. Since Ï is a module map, its fulfils
for a certain X P M such that T X “ 0. Thus we have ImagepXq Ä kerpT q, but since T • 0 (in
particular T is self-adjoint) and ImagepT q “ H , we have kerpT q “ 0, and we conclude that X “ 0,
so that Ï “ 0. The question is to know if E “ M {M T has a finitely generated submodule or not.
Let P be a projective submodule of M {M T ; by the lifting property (52.34) we have a map ⁄̃ such
that the following commutes
;M (71.129)
✏✏ ⁄
P / M {M T
Now we suppose that X is positive. This is done without loss of generality because from polar
decomposition, for every X P M , there exists a positive Y such that M X “ M Y .
Since T is positive, we can consider the spectral theorem 57.59 and the isomorphism
` ˘
◊ : C ˚ pT, q Ñ C SpecpT q . (71.130)
and then one defines the operator S‘ “ ◊´1 pf‘ q P C ˚ pT, q. Let us prove that P‘ “ S‘T is a
projection. We have P‘2 “ S‘ T S‘ T . Take an orthonormal basis of H of eigenvectors of T and
let’s call ⁄i the eigenvalues: T ei “ ⁄i ei . If ⁄k † ‘, then S‘ ei “ 0 and of course S‘ T S‘ ei “ S‘ ei .
Otherwise, we have
1
S‘ T S‘ ei “ S‘ T ei “ S‘ ei . (71.132)
⁄i
That proves that P‘2 “ P‘ , and so that this is a projection. We obviously also have P‘ Ñ in M T .
Similarly one can define Q, the projection onto ImagepXq and we have projections Q‘ Ñ Q
with Q‘ “ Y‘ X for some Y ‘ P M and Q‘ P M X. Now take ‘1 and ‘2 and look at the projection
onto ImagepQ‘1 q X ImagepP‘2 q. The latter intersection is in fact 0 because A‘1 _ P‘2 P M Q‘1 X
M P‘2 “ M X X M T “ 0. Using lemma 71.22, we get
Recall that the trace used to define the dimension has to be normal, so that the dimension function
is continuous in such a way that taking the limit ‘1 Ñ 0 and ‘2 Ñ 0 provides the expected result
from which one deduce that dim Q “ 0 and therefore X “ 0. This finish the proof that the module
E “ M {M T with a positive T and ImagepT q “ H has no projective submodules.
Proposition 71.85.
If E is a finitely generated module over M , then the torsion submodule does not contains non zero
projective finite submodules.
Proof. Later.
Lemme 71.86.
If E is a finitely generated submodule of a finitely generated projective module E, then F is pro-
jective and F » ClE pF q.
Proof. We assume as usual that E and F are singly generated. The singly generated projective
module E reads E “ M P while the general form of a singly generated submodule is F “ M T
for a positive T which vanishes on P K H and ImagepT q Ñ ImagepP q. We already proved that
M T » M Q where Q is the projection over ImagepT q.
We have ClM P pM T q “ M Q because of the direct sum decomposition M P “ M Q ‘ M pP ´
Qq from which we can build an homomorphism M P Ñ M which vanishes on M T , namely the
projection because M T Ñ M Q. Problèmes et choses à faire
I do not understand one single word about the latter justification : p
Corollaire 71.87.
If F is a finitely generated submodule of a finitely generated projective module E, then dimpF q §
dimpEq.
71.13. POSITION OF SUBMODULES 2801
` ˘
Proof. By lemma 71.86, we have dimpF q » dim ClE pF q while we know that E “ ClE pF q ‘
E{ ClE pF q. The latter makes that dim E is given by the trace of two projections:
Corollaire 71.88.
If E is a finitely projective module over a von Neumann algebra with a trace, then dimpEq “
DimpEq.
Proof. By very definition, DimpEq • dimpEq because E itself belongs to the set on which the
supremum is taken in the definition (71.120) of Dim. The corollary 71.87 provides the opposite
inequality.
Proposition 71.89. ` ˘
If E is projective finitely generated and if F Ñ E, then dim ClE pF q “ DimpF q.
Proof. The case where F is finitely generated is already done by lemma 71.86. For the general
case, suppose that the proposition does not hold. In this case, the lemma 71.86 yields
` ˘ ` ˘
suptdim ClE pHq tel que H is finite projective u † dim ClE pF q . (71.137)
On the one hand, by corollary 71.84, the module ClE pHq is a direct summand of ClE pF q which is
itself (by the same result) a direct summand of E. On the other hand, being a finitely generated
module, it reads E “ M n P for some projection P by proposition 52.14. Combining both, we have
a projection PH § P such that ClE pHq “ M n PH .
Now the finitely generated projective submodules of F form a directed system. For every
finitely generated projective submodules H1 and H2 of F , there exists at least H3 ö “ H1 ‘ H2
which contains H1 and H2 . Thus the set of PH is directed too and one can look at H PH , the
smalest projection whose range contains the range of all of the PH .
ö
We have H PH † P because if not, using normality of the trace,
`™ ˘
dim PH “ sup dimpPH q † P. (71.138)
H H
ö
Now let Q “ P ´ HPH . If F is some submodule of E, one has
§
F “ tH Ñ F tel que H is finitely generatedu
§ (71.139)
“ tH Ñ F tel que H is finitely generated and projectiveu
because `HöÑ F Ñ
˘ E which is finitely generated projective. Such a H has the form M PH , sp
n
F ÑM H PH . Notice that this F lies in the kernel of the non zero map
n
ClE pF q “ M n P Ñ M n Q
(71.140)
V ބ V Q.
ö `ö ˘ ö
Indeed,
ö since
` ˘ P , we have
öH PH † H PH P “ H PH , so that for every T P M we have
n
The three point in this demonstration that use the von Neumann algebra background (and not
only general module theory) are the following.
— First we used normality of the traceö to commute the dimension with the supremum,
— and second, we used continuity of with respect to the dimension.
2802 CHAPTER 71. VON NEUMANN ALGEBRAS
71.13.1 Summary
We have two dimension functions dim and Dim such that
(1) dim E “ Dim E whenever E is a finitely generated projective module,
(2) for every finitely generated module E and every submodule F , the module E{ ClE pF q is
finitely generated and projective, Problèmes et choses à faire
Check if one does not need the assumption that E is projective too.
` ˘
(3) If F Ñ E and if E is a finitely generated projective module, then DimpF q “ Dim ClE pF q .
From now we do no more use the “von Neumann algebra” assumption. Instead we suppose to have
a ring R and two dimensions functions satisfying these three properties.
Lemme 71.91.
If one has a projective module map fl : H1 Ñ H2 , then DimpH1 q • DimpH2 q.
Proof. Let F be any projective module for which there exists an inclusion ÿ : F Ñ H2 . That map
can be lifted because F is projective. So among all the submodules of H1 , there is the one which is
the image of F be the lifted map. That one of course contains H2 itself. Thus H2 is a submodule
of H1 and the supremum defining the dimension in H1 is automatically bigger or equal to the one
defining the dimension of H2 .
Proposition 71.92.
If
p
0 / E0 / E1 / E2 /0 (71.141)
is a short exact sequence of modules, then dimpE1 q “ dimpE0 q ` dimpE2 q.
Proof. Using last proposition, we can assume that all of E0 , E1 and E2 are finitely generated.
Indeed when a module is not finitely generated, it is still the union of the directed system of all
its finitely generated submodules.
Let F be a finitely generated projective submodule of E2 , we have the exact sequence
0 / E0 / p´1 pF q /F / 0. (71.142)
The module F being projective, we have p´1 pF q » F ‘ E0 . We do not know if the dimension
function is additive with respect to direct
` ´1 sum,
˘ but by definition of a supremum, we have the
inequality DimpE0 q ` DimpF q § Dim p pF q , and the chain
` ˘
DimpE0 q ` DimpF q § Dim p´1 pF q § DimpE1 q (71.143)
For the reverse inequality, let F be a finitely generated projective submodule of E1 , and consider
the following exact sequence of finitely generated projective module
The fact that F { ClF pF X E0 q is projective is proposition 71.83. Since F { ClF pF X E0 q is at most
a subset of F which is finitely generated, it has to be finitely generated too. We also know by
corollary 71.84 that F splits into
which is a direct sum of finitely generated projective module, so that we can use the definition of
dimension with traces (which sums up over direct sum) instead of the one with supremum. Thus
we have ` ˘ ` ˘
dimpEq “ dim ClF pF X E0 q ` dim F { ClF pF X E0 q . (71.147)
The `module F being
˘ projective and finitely generated, proposition 71.89 allows us to replace
dim ClF pF X E0 q by DimpF X E0 q and write
` ˘
dimpEq “ DimpF X E0 q ` Dim F { ClF pF X E0 q § DimpE0 q ` DimpF {pF X E0 qq (71.148)
where we also used lemma 71.91. Since F {pF XE0 q is a quotient of E2 , we have DimpF {pF XE0 qq §
DimpE2 q, and taking the supremum over all suitable F , we find the result
Proposition 71.93. ` ˘
If E is a finitely generated module and F Ñ E, then DimpF q “ Dim ClE pF q .
Proof. Hint: by the proposition, one can assume that E is actually projective.
with pUg f qpf q “ gpg ´1 hq. Each Ug is an unitary operator in H. In this case, the commutant M 1
turns out to be the von Neumann algebra generated by the regular right representation.
Problèmes et choses à faire
The operator R given by
` ˘ f pxq
Rpf q pxq “ (71.151)
2
commutes with all the regular left representation, but is not part of the regular right representation. Do we impose unitarity conditions?
∞
(2) in the strong topology, ‡ P‡ “ ,
(3) the algebra M X M 1 is generated by tP‡ u and each of the P‡ is minimal in the center,
(4) P‡ M P‡ is a factor on H‡ and we have pP‡ M P‡ q1 “ P‡ M 1 P‡ ,
(5) P‡ M P‡ is finite dimensional.
Notice that, because of point (1) and (2), if H‡ denotes the range of P‡ , then
à
H“ H‡ . (71.154)
‡
Proof. No proof.
(2) if u is any section of the field such that x ބ xux , wx y is integrable for all w P H, then u is
almost everywhere equal to the section associated with a vector of H,
(3) the set of all decompositions of vectors in H is maximal.
Problèmes et choses à faire
Notice that the set of sections that are equal almost everywhere to sections obtained by de-
composition of vectors in H form a module over L8 pX, µq.
An technical fact that will be used in much of proofs is the following.
Lemme 71.95.
If tv1 , v2 , . . .u Ñ H has a dense span in H, then for almost every x P X, the set Spantv1x , v2x , . . .u
is dense in Hx .
71.14.3 Examples
71.14.3.1 First example
À Let Hx with x P X be a field of Hilbert space over a countable set X. Then define H “
xPX Hx . That Hilbert space decomposes by pX, µq with µ being the counting measure, and to
v P H corresponds a component vx in each Hx that are taken as the section associated to the
vector v.
71.14. DECOMPOSITION OF OPERATORS AND REPRESENTATIONS 2805
Second, suppose that u is a section such that the map x ބ ux f pxq is integrable for every f
associated with a vector in H. The measured space pX, µq being countable, X is a union of finite
measure sets tXi u on each of them one can consider the characteristic function 1i . Then for every
i, the function x ބ ux 1i is measurable on Xi . That proves that u is measurable. We conclude that
it is square integrable and thus defines a function which is almost everywhere equal to the section
associated with a vector v P L2 pX, µq, namely the element of L2 pX, µq which is the class of u.
Depending on the x, the set txu X tf1 ° 0u \ txu X tf2 ° 0u has zero, one or two elements. Now
if v “ pv1 , v2 q P H with v! “ rg1 s and v2 “ rg2 s, we define the section associated with v as
` ˘
vx “ f1 pxq1{2 g1 pxq, f2 pxq1{2 g2 pxq . (71.158)
As notation, if f1 pxq “ 0, we identify this with the element f2 pxq1{2 g2 pxq instead of with p0, f2 pxq1{2 g2 pxqq.
Remarque 71.96.
The conditions 71.159 are in fact uncountably many conditions (one for each v P H) each of
them having the possibility to be wrong on a null set. One has thus to be prudent because an
uncountable union of null set might be a set of non zero measure.
Lemme 71.97.
The set of decomposable operators is a ˚-algebra.
Lemme 71.98.
Let T be a decomposable operator. We have T • 0 if and only if Tx • 0 almost everywhere.
2806 CHAPTER 71. VON NEUMANN ALGEBRAS
Proof. In order to prove that T is positive, we have to evaluate xT v, vy from the decomposition of
T and v. We suppose that Tx • 0 for almost every X P X and we compute
ª ª
xT v, vy “ xpT vqx , vx ydµpxq “ xTx vx , vx ydµpxq • 0. (71.160)
X X
where we were allowed to change pT vqx by Tx vx because the equality holds almost everywhere.
For proving the contrary, let pick a dense, countable and rational vector subspace H0 Ñ H,
and assume that T • 0. If v P H and if E Ñ X is measurable, then by maximality of the Hilbert
space decomposition there is some vector vE P H such that
#
vx if x P E
pvE qx “ (71.161)
0 otherwise.
#
vx if x P E
sx “ (71.162)
0 otherwise.
Existence of integral (71.164) assures the existence of a vector vE P H such that pvE qx “ sx almost everywhere. So we have the vE without
maximality axiom. Do we really want the condition (71.161) to hold everywhere and not only almost everywhere? Since we only use it in integrals,
I think that one does not care about a violation of condition (71.161) on a null set.
We have that ª ª
xTx vx , vx ydµpxq “ xTx vEx , vEx y “ xT vE , vE y • 0 (71.165)
E X
almost everywhere. Therefore there exists a null set N Ñ X such that xTx vx , vx y • 0 for every
v P H0 whenever x P XzN . If x P XzN , then Spantvx tel que v P H0 u is dense in Hx because
of lemma 71.95. Notice that H0 is countable and there is one null set for each element of H0 on
which we are unsure of the sign of xTx vx , vx y.
By enlarging the set N we can assume that
for every x P XzN and ai P ris. Indeed the equalities are an equation for each ai P ris and
vi P H0 . That is thus a countable set of equations; thus the set where the equations are not
satisfied is a countable union of null sets.
What we have now is that every vector in Hx (with xinXzN ) can be written as a limit if
vectors in H0 , and then xTx vx , vy • 0, so that Tx • 0.
Lemme 71.99.
We have
}T } “ ess- sup }Tx } “ mintc tel que }Tx } § c almost everywhereu. (71.168)
71.14. DECOMPOSITION OF OPERATORS AND REPRESENTATIONS 2807
Proof. The inequality }T } § ess- sup }Tx } comes from the fact that
ª
}T v} “ xTx vx , Tx vx ydµpxq (71.169)
X
which gives the bound. For the inverse inequality, remark that }T }2 “ }T ˚ T } and pT ˚ T qx “
pTx q˚ Tx almost everywhere, so that one can replace T by T ˚ T , or more simply we can assume
that T • 0. Now consider the positive and self-adjoint operator }T } ´ T . The “component”
operator is positive almost everywhere: }T } “ Tx • 0 for almost every x P X. We know that the
maximum of the spectrum of Tx is }Tx }, so that the number }T Y } ´ }Tx } is almost everywhere
positive because it is an element of the spectrum of }T } ´ Tx which is positive.
Sketch of proof. First, a convex subset of BpH q is strongly closed if and only if it is weakly
closed. So, using what is said on page 2434 about the fact that weak topology is compatible with
the involution, we know that if T lies in the strong closure of A, and if T “ T ˚ , then T is the
limit of a net of self-adjoint elements in A. That allows us to focus the proof of the theorem on
self-adjoint elements of A.
Problèmes et choses à faire
For me, the fact that this reasoning actually allows to restrict ourself to self-adjoint elements. But I suppose that it would become more clear
when one understands the link between the statement here and the statement of Kaplansky’s theorem in [511], page 32.
Consider the function f pxq “ 2x{p1 ` x2 q that provides an homeomorphism from r´1, 1s to
r´1, 1s. Thus by continuous functional calculus, if T “ T ˚ and }T } § 1, we have T “ f pSq where
S “ S ˚ and S belongs to the norm closure of the algebra generated by T .
Now if T is the strong limit of the net T– of self-adjoint operators, then S is the strong limit
of S– with S– P A and S–˚ “ S– . Indeed one can prove that S– Ñ S implies 10 f pS– q Ñ f pSq.
Since }f pS– q} § 1, thus }T } § 1.
Théorème 71.101.
The algebra of decomposable and diagonal operators are mutually commutant.
10. In the weak topology, that implication is the continuous functional calculus, but in the strong topology (the
one we are considering here), the validity of this assertion relies on the particular form of f .
2808 CHAPTER 71. VON NEUMANN ALGEBRAS
Proof. The fact that decomposable operators are in the commutant of diagonal operators and vice
versa. The strategy to prove the theorem will be to first prove that decomposable and diagonal
operators are von Neumann algebras, and then show that every operator in the commutant of
diagonal operators is decomposable. From there, the conclusion yields from the double commutant
theorem.
Suppose that T is a strong limit of decomposable operators, i.e
T “ s limT– (71.172)
where T– is decomposable and supt|T– |u † 8. We want to show that T is decomposable. For, let
H0 be a dense countable rational subspace of H. If T is limit of a uniformly bounded net, the fact
that H0 is countable makes that there is a sequence of uniformly bounded decomposable operators
Tn such that Tn v Ñ T v for every v P H0 .
We have ª
lim }Tnx vx ´ pT vqx }2 dµpxq “ 0 (71.173)
nÑ8 X
for almost every v P H0 . A general fact about L2 spaces is that when a sequence goes to zero, then
the functions (representative of the elements of L2 ) themselves goes to zero on a subsequence. So
for every given v P H0 there is a subsequence tnj u such that
for almost every x P X. The subsequence might depends on the v, bu a diagonal argument provides
a subsequence tnk u of tnj u such that for every v P H0 ,
for almost every x P X. These relations are a countable number of convergence almost everywhere.
Then there exists a null set N Ñ X such that
for x P XzN . This is a dense subspace of Hx by lemma 71.95. Since the essential supremum of
}Tnx } is the norm of Tn , we have a “double uniformly bounded” relation
pT wqx ´ Tx wx “ 0. (71.180)
It extends by continuity from H0x to Hx . The resulting operator is bounded because the
sequence Tnk x is uniformly bounded.
Now we enlarge once again the set N by the null set appearing in the definition of the essential
supremum we find that Tx is bounded for every x P XzN . All the work make tTx uxPX a candidate
decomposition of T .
71.14. DECOMPOSITION OF OPERATORS AND REPRESENTATIONS 2809
In order to check that this actually is a decomposition of T , we have to prove that pT vqx “ Tx vv
for every v P H and almost every x P X. We will prove that property by proving that the integral
ª
}pT vqx ´ Tx vx } dµpxq (71.181)
X
vanishes. For that, we choose w P H0 such that }w ´ v} § ‘ and, by virtue of (71.180), we add
Tx wx ´ pT wqw in the integrand:
ª ª
2
}pT vqx ´ Tx vx } dµpxq “ }pT vqx ´ pT wqx ` Tx wx ´ Tx vx }2 dµpxq
X Xª ª
2
§2 }pT vqx ´ pT wqx } dµpxq ` }Tx wx ´ Tx vx }2 dµpxq
X X (71.182)
“ 2}T pv ´ wq}2 ` 2}T̃ pv ´ wq}2
“ 2}T } }v ´ w}2 ` 2}T̃ } }v ´ w}2
§ constant · ‘
for every v1 and v2 in H0 and ai P ris. The set H0x “ tvx tel que v P H0 u is dense in Hx . For
f P L8 pX, µq, we define the multiplication operator
for every x P XzN . Of course, P H0 ‘ P K H0 “ H0 , so that the direct sum of the two spaces of
(71.186) is H0x . Taking the strong closure,
where the double bar denotes the strong closure. Now, for x P CzN , define Px : Hx Ñ Hx , the
projection onto tvx tel que v P P H0 u. By the same reasoning as for T before, we prove that tPx u
decomposes P .
11. My spelling corrector does not know that word. Bad word or bad corrector?
2810 CHAPTER 71. VON NEUMANN ALGEBRAS
Lemme 71.102.
If M is an abelian von Neumann algebra on a separable Hilbert space H, then M has a strongly
dense C ˚ -algebra A
Proof. No proof.
Théorème 71.103.
If M is an abelian von Neumann algebra on a (as usual separable) Hilbert space H, then there is
a decompositon of H such that M is the set of diagonal operators.
Proof. Pick a A Ñ M as in the lemma 71.102, and add an unit is needed. Then we have A » CpXq
for some metrisable compact space X by the Gelfand theorem (see [504]). We can write
à8
H“ Hk (71.188)
k“1
where Hk is invariant under the action of A and has a cyclic vector. Indeed, pick a vector v1 in
H and define H1 “ Av1 , then pick v2 in the complement and continue. We have an isomorphism
Hk » L2 pX, µk q where µk is some probability measure by vk fiÑ 1 and T vk fiÑ T̂ where the hat
denotes the Gelfand isomorphism.
Problèmes et choses à faire
That gives an element of L2 pX, µk q for each vector inside Hk . But I do not see the isomorphism. For example a simple step function belongs to
L2 pX, µk q and corresponds to which element of Hk ?
The map
2
 : Av1 Ñ L pX, µq
(71.189)
T v1 fiÑ T̂
(whose image is included in L2 pX, µq because X is compact, so that every continuous function is square integrable with respect to a probability
measure) is injective because two different continuous functions cannot belong to the same class in L2 pX, µq.
For me, surjectivity is not clear (and even wrong) because there exists many elements in L2 pX, µq who have no continuous representative.
Do we have to game with limits and exploit the fact that A is strongly closed?
À
So we can assume that A “ CpXq and H “ k L2 pX, µq. Let
8
ÿ
µ“ 2´k µk (71.190)
k“1
where, in the first line we permuted the sum and the integral using the monotone convergence
theorem. The fact that integral (71.195) is finite proves that the function (71.194) has finite values
almost everywhere, or
}Ïx }2l2 p ,µx q † 8 (71.196)
almost everywhere. Let us redefine Ïx as zero for the x where the former definition gives }Ïx }2l2 p ,µx q .
This is a redefinition over a null set.
One can check that this construction provides a decomposition of H for which M is the algebra
of diagonal operators.
Problèmes et choses à faire
To be done. . .
Let M be an abelian von Neumann algebra of operators on H and suppose that there are two
decompositions tHx u and tHy u of H using M by theorem 71.103. Let now
"ˆ ˙*
T
M2 “ tel que T P M (71.197)
T
That von Neumann algebra decomposes H ‘ H into tHx u ‘ tHy u. The matrix
ˆ ˙
0 1
S“ (71.198)
1 0
Proposition 71.104.
If N lies in the commutant of M , there are von Neumann algebras Nx Ñ BpHx q for almost every
x such that
(1) Nx “ tTx tel que T P N u2 ,
(2) pNx q1 “ pN 1 qx almost everywhere.
Proof. No proof.
Lemme 71.105.
If H is provided with a decomposition tHx u and if fipAq consists of decomposable operators, then
there are representations fix : A Ñ BpHx q such that fipaqx “ fix paq.
Proof. No proof.
Théorème 71.106.
If fi : A Ñ BpHq is any representation of a separable C ˚ -algebra, then there is a decomposition of
H such that fipAq consists of decomposable operators. There also exists a decomposition tfix u of fi
such that each fix is irreducible.
Proof. We do not prove the first part, and suppose then that tfix u is a decomposition given by
lemma 71.105, so that fix paq “ fipaqx . ` ˘1
Let M be a maximal (Zorn’s lemma) abelian von Neumann subalgebra of fipAq , and decom-
pose H so that M is the algebra of diagonal operators. Now we consider N , the von Neumann
algebra generated by M and fipAq. An element in N 1 belongs to fipAq1 and commutes with M
which is maximal. Thus N 1 “ M .
2812 CHAPTER 71. VON NEUMANN ALGEBRAS
By the property of the decomposition fix , an element of fix pAq1 has to commute with all M , so
that
fix pAq1 “ pNx q1 “ pN 1 qx “ Mx “ , (71.199)
and by Schur’s lemma, the representation fix is irreducible almost everywhere.
As an example, take A “ C ˚ pF2 q where F2 is the free group with two generators and A is
formed by taking all the linear combination and then the closure as C ˚ -algebra. One can show
that
(1) There are two decompositions of the regular representation fi of A on H “ l2 pF2 q that we
denote by tHx u and tHy u. These decompositions are a priori built on different measured
spaces.
(2) All the representations in tfix u \ tfiy1 u are irreducible and mutually inequivalent.
So the decomposition in irreducible representations is not unique at all, in contrast to the group
representation case. The point is that the regular representation of C ˚ pF2 q is a factor, while the
following theorem only assures unicity of decomposition into factors.
Théorème 71.107.
Let fi : A Ñ BpHq be a representation of the C ˚ -algebra A. There is an essentially unique decom-
position of H such that
(1) the representation fipAq consists of decomposable operators,
(2) the representations fix are factors representations for almost every x.
By “essentially unique”, one means that if
` ˘ ` 1 1 ˘
X, µ, tHx u, H Ñ tHx u and X , µ , tHx1 u, H Ñ tHx1 u (71.200)
are two decompositions, there exists a map X Ñ X 1 which is almost everywhere an equiv-
alence of measurable spaces. That is it maps µ to a measure which is mutually absolutely
continuous with µ1 , and an unitary isomorphism Hx Ñ Hx1 defined for almost every X
which maps vx to vx1 where vx1 is the decomposition of v with respect to the decomposition
tHx1 u.
Proof. Let us prove the existence part of the theorem For we proceed as in the proof of theo-
rem 71.106, but we take M “ fipAq2 X fipAq1 , this is the center of the von Neumann algebra
generated by fipAq. Using proposition 71.104, we find
` ˘
fix pAq2 X fix pAq1 “ fipAq2 X fipAq1 x “ . (71.201)
Théorème 71.108.
If G is a semisimple Lie group, then C ˚ pGq is liminal.
Proof. No proof.
Théorème 71.109.
If A is a liminal C ˚ -algebra, then every factor representation is of type I.
Théorème 71.110.
If fi is any representation of a liminal C ˚ -algebra on a separable Hilbert space, then there is a
decomposition of H such that
(1) the space Hx carries a representation which is a factor of type I,
71.15. INDEX THEORY 2813
Proof. No proof.
Lemme 71.111.
Every irreducible representation of A1 b A2 with both Ai being liminal is a tensor product of
irreducible representations of A1 and A2 .
Now take an unimodular group G such that C ˚ pGq is liminal. One knows that the von Neumann
algebras generated by the left and right regular representations are mutually commutant. We
consider the bi-regular representation G ˆ G on L2 pGq. The commutant of that representation is
equal to the center of the von Neumann algebra generated by the two regular representation and
is in particular abelian.
Using the lemma, the space L2 pGq decomposes into representations Hfi· “ Hfi b H· of G ˆ G
where fi and · are irreducible representations. From the unimodular assumption, the symmetry
f pgq Ñ f pg ´1 q is an isometry that intertwines left and right regular representations. Notice that a
function can be seen as a vector on L2 pGq as well as as an operator over L2 pGq by the convolution.
Thus we turn Hfi· into a ˚-algebra by the multiplication Hfi· b Hfi· · Hfi· .
We denote by Ĝ the set of irreducible representations of G
Théorème 71.112.
Let G be an unimodular locally compact liminal group. There is an unique measure on Ĝ, the
Plancherel measure, denoted by µ such that for every f P L1 pGq X L2 pGq (as vector in L2 pGq
or as operator by involution),
ª
}f }2L2 pGq “ }fipf q}2HS dµpfiq (71.202)
Ĝ
That theorem has to be compared to the Fourier theory for abelian groups like or S 1 , in
particular the Parseval equality. This has also to be compared with the following theorem.
where the ration dimpfiq{ volpGq is the Plancherel measure made explicit in the compact case.
71.15.1 Introduction
Let M be a type II1 factor, which, thus, possesses an unique normal, tracial, normalized faithful
state. We can assign a dimension to every modules over M by theorem 71.79. In particular, M
being an algebra of operators on the Hilbert space H , the space H is a M -module and we can
consider the number
dimM pH q. (71.205)
2814 CHAPTER 71. VON NEUMANN ALGEBRAS
Following the cases, that can be any number in s0, 8r. Indeed, taking any v P H , we have the
M -module map
fl: M Ñ H
(71.206)
T ބ T v,
so that by lemma 71.91, we have
But we saw on page 2797 that dimM M “ Trp q, see definition (52.40) and bellow. We deduce
that dimM H • Trp q, which can be any non vanishing positive real number.
Let N Ñ M Ñ BpH q be a subfactor, i.e. that N is a factor in BpH q and N Ñ M . For the
same reason as before, dimN pH q Ps0, 8r. We consider the ratio
dimN pH q
rM : N s “ (71.208)
dimM pH q
that is called the index.
The theorem that we are intend to prove is the following.
Proof. No proof.
n 4 cos2 pfi{nq
2 0
3 1
4 2
5 2.618 the golden ratio
6 3
.. ..
. .
In the dots, we have an increasing sequence of numbers bigger than 3 that converges to 4.
71.15.2 Example
Let G be a discrete group with infinite conjugacy classes and
` H, a subgroup which has` infinite
˘
conjugacy classes for its own right. Consider M “ M`pGq Ñ˘B l2 pGqq and N “ M pHq Ñ B l2 pHq .
By simple inclusion of H in G, we have M pHq Ñ B l2 pGq too.
As defined in subsection 71.7.5, the von Neumann algebra M pHq is generated by the opera-
tors Uh with h P H. We already proved in proposition 71.21 that l2 pGq is a finitely
` ˘ generated
projective module over M pGq, so that one can address the question of dimM pGq l2 pGq using the
decomposition given by corollary 71.84:
` ˘
l2 pGq “ Cll2 pGq M pGq ‘ l2 pGq{M pGq. (71.209)
´ ` ˘¯ ` ˘
On the one hand, by proposition 71.80, we have dim Cll2 pGq M pGq “ dim M pGq , and on the
other hand, one can prove that l2 pGq{M pGq is a torsion, so that
` ˘ ` ˘
dimM pGq l2 pGq “ dimM pGq M pGq “ 1 (71.210)
Is it correct that the latter dimension is in fact the trace of on M pGq which has to be normalised?
Let us now consider g1 , . . . , gn be representatives of the classes of G{H and look at the decom-
position
l2 pGq “ l2 pHg1 q ‘ . . . ‘ l2 pHgn q. (71.211)
71.15. INDEX THEORY 2815
` ˘
Since l2 pHgi q is a submodule of l2 pHq, each of dimn l2 pGq lies in the same case as equation
(71.210), so that ` ˘
dimN l2 pGq “ n. (71.212)
Thus rM : N s “ n.
One can explicitly construct examples of index for every real bigger than 4.
Let M Ñ BpH q be a von Neumann algebra and form M p8q Ñ BpH ‘ . . . ‘ H ‘ . . .q,
(
M p8q “ pT, T, . . .q tel que T P M . (71.213)
Since M p8q does not contains more information than M itself, we want to be able to say that
they are the same von Neumann algebra. The interest of ultraweak topology described in subsec-
tion 57.6.1 is that if we consider both M and M p8q with the ultraweak topology, then they are
homeomorphic in a natural way.
for every increasing net of projections. A net of projections being said increasing when – ° —
implies P– ° P— . An example of a normal map is the trace of a type II1 factor.
Théorème 71.115.
A positive linear functional on a von Neumann algebra is normal if and only if it us ultraweakly
continuous.
Théorème 71.116.
Let : M Ñ BpH q be a ˚-homomorphism from a von Neumann algebra into BpH q. The the
following are equivalent:
(1) is ultraweakly continuous,
(2) is normal,
(3) pM q is a von Neumann algebra.
Notice that we do not suppose M to be a von Neumann algebra acting on the Hilbert space
H.
Proof. The implication (3) ñ (2) is proved by using the theorem 71.115. The implication (2)
ñ (1) is only the fact that ultraweak continuity is defined in terms of linear functionals.
The difficult part is to prove that (1) implies (3). Since pM q is a ˚-algebra, proving that
it is strongly closed proves that it is a von Neumann algebra. So, let us suppose T to be in the
strong closure of pM q. By Kaplansky density theorem 71.100, there exists a net T– P pM q with
sup }T– } † 8 and
T “ s limT– “ s lim pS– q (71.215)
for some S– . Since is isometric, the set tS– u is uniformly bounded: }S– } † 8. Thus, using the
Banach-Alaogu theorem 57.70, and taking a subnet, we can assume that there is a S P M such
that
S “ uw limS– . (71.216)
2816 CHAPTER 71. VON NEUMANN ALGEBRAS
Since is ultraweakly continuous by assumption, we have pSq “ uw lim pS– q. Since the limit
T “ s limT– exists, the same exists in the weak topology (and is equal), thus T “ pSq, which
proves that pM q is strongly closed.
This is the GNS construction applied to the tracial state · . The standard form of the finite factor
M is the realisation of M as left multiplication operators on L2 pM q. The image is a von Neumann
algebra because the representation is normal and faithful.
Lemme 71.117.
An element v P L2 pM q belongs to M if and only if it is bounded.
Proof. No proof.
Lemme 71.118.
We have
M 1 “ JM J (71.221)
Notice that operators in JM J are operators of right multiplication on M because pJT JqS “
JT pS ˚ q “ JpT S ˚ q “ ST ˚ while the elements of M are left multiplications, so that it is clear that
JM J Ñ M 1 .
Proof. No proof.
Théorème 71.119.
One has
rM1 : M0 s “ dimM0 pM1 q. (71.223)
and moreover
dimM0 pH q
rM1 : M0 s “ (71.224)
dimM1 pH q
if M1 is represented on the Hilbert space H .
Proof. No proof.
Suppose given M1 and a finite subfactor `M0 and ˘suppose rM1 : M0 s † 8. Since M0 is a
subfactor of M1 , we can see M01 as subset of B L2 pM1 q . If M11 is not finite, then there is a part of
L2 pM1 q which is identified with L2 pM1 q as M0 -module. That part has obviously the same dimen-
sion as the whole L2 pM1 q (because the module structure is the same). Thus in this ` case, L˘2 pM1 q
2
contains infinitely many submodules with all the same dimension, so that dimM0 L pM1 q “ 8,
which contradicts the assumption. We deduce that
` ˘
M01 Ñ B L2 pM1 q (71.225)
Lemme 71.120.
If T P M1 , then T P1 “ P1 T if and only if T P M0 .
Proof. It T P M0 , we have
T P1 “ P1 T P1 . (71.227)
Taking the adjoint of that equation, P1 T˚ “ P1 1 , and using the relation (71.227) for T , we
T ˚P ˚
T P1 “ P1 T (71.228)
for every T P M0 .
For the reverse sense, equality T P1 “ P1 T is an equality of operators on M1 . Let us apply it
to P M1 . Since belongs to M0 too (every von Neumann algebra contains the unit), P1 p q “
and we stay with T “ P1 T , so that T P M0 .
M2 “ tM1 , P1 u2 . (71.229)
Théorème 71.121.
If rM1 : M0 s † 8, then
(1) M2 is a finite factor, in particular it has a unique normal faithful trace · ,
(2) rM2 : M1 s “ rM1 : M0 s,
(3) · pP1 q “ rM1 : M0 s´1 .
Proof. No proof.
The von Neumann algebra M2 was created from M0 and M1 , while the theorem shows that
M1 has the same properties in M2 than M0 in M1 , so that one can redo the construction starting
from M2 and M1 instead of M0 and M1 to build M3 and P2 . The index does not change and the
projection P2 still has the same trace.
2818 CHAPTER 71. VON NEUMANN ALGEBRAS
71.15.5 Subfactors
Problèmes et choses à faire
Je pense qu’à partir d’ici, j’ai inversé M0 et M1 partout jusqu’à la subsection 71.15.7.
Let M1 Ñ M0 be a subfactor of the finite factor M0 , and P : L2 pM0 q Ñ L2 pM0 q be the projection
onto L2 pM1 q. Notice that L2 pM1 q Ñ L2 pM0 q because the trace on M1 is the restriction to the one
of M0 from unicity. Now we consider xM0 , P y, the von Neumann algebra generated by M0 and P .
Lemme 71.122.
We have P pM0 q Ñ M1 .
Proof. No proof.
Proposition 71.123.
The map
p : M0 Ñ M1
(71.230)
T ބ P pT q.
satisfies
(1) ppT ˚ q “ ppT q˚ ,
(2) ppT ˚ T q • 0,
(3) ppT ˚ T q “ 0 if and only if T “ 0,
(4) ppS1 T S2 q “ S1 ppT qS2 for every S1 and S2 in M1 .
for every T P M0 .
Proof. The point (1) is the fact that M1 is closed under the involution. The point `(2) comes
˘ form
lemma 71.122 and the positivity property in M1 . For the point (3), remark that Tr ppT q “ TrpT q
because pp q “ , while by unicity up to normalisation, the trace is determined by its value
on .
Lemme 71.124.
If T P M0 , then
P T P “ ppT qP (71.232)
as operators on L2 pM0 q.
Proof. No proof.
Lemme 71.125.
If M1 Ñ M0 is a subfactor of the finite factor M0 and if P is the projection onto L2 pM1 q (as
operator in L2 pM0 q), then
M1 “ tM01 Y P u1 , (71.233)
and
M11 “ tM01 Y P u2 (71.234)
as consequence.
Proof. First, every element of tM01 YP u1 has to commute with M01 and then to belongs to M02 “ M0 .
But we proved in lemma 71.120 that when T P M0 and T P “ P T , then T P M1 . That proves that
tM01 Y P u1 “ M1 .
71.15. INDEX THEORY 2819
P pT Sq “ T P pSq (71.235)
Lemme 71.126.
We have
xM0 , P y “ JM11 J, (71.236)
and as consequence, xM0 , P y is a finite factor.
Proof. First note that the commutant of a factor is a factor, so that M11 is a factor and JM11 J is
a factor because it is algebraically isomorphic to the factor M11 . Thus the lemma proves that the
left hand side of (71.236) is in particular a finite factor.
Let us now prove the equality (71.236). One needs to show that JxM0 , P yJ “ M11 . The
commutant of the left multiplication action is the right one, while the left action of JT J is the
right action of T , so one has JM0 J “ M01 . Since JP J “ P , it is sufficient to show that
The von Neumann algebra generated by M01 and P is tM01 , P u2 that is equal to M11 by lemma 71.125.
So xM0 , P y is a factor and it is finite when M11 is finite, which happens when rM0 : M1 s † 8.
Assuming that finiteness, we thus have a trace Tr : xM0 , P y Ñ which extends the trace on M0
(by uniqueness).
Notice that, as particular case of equation (71.236) when M1 “ M0 , we have P “ Id and
M0 “ JM01 J. (71.238)
S · T “ pJSJqT (71.239)
Proposition 71.127.
If M0 is any finite factor acting on H and if Q P M01 , then QH is invariant by M0 , and
Proof. First, notice that a finite factor always has an unique trace. If Q “ 0 or Q “ , the claim
is obvious. Suppose now that TrpQq “ r{s, that is a rational number.
In that case, we have
QH ‘ . . . ‘ QH “ looooooomooooooon
loooooooooomoooooooooon H ‘ ... ‘ H (71.241)
s times r times
I can understad that in the setting of the fractional dimensions as described in subsection 71.10.3, but there, we have the assumption of no minimal
First, notice that the trace of the identity over QH is the trace of Q on H . Indeed, let
projection.
Thus, by additivity,
Trp QH ‘...‘QH q “ s TrpQq “ r. (71.243)
2820 CHAPTER 71. VON NEUMANN ALGEBRAS
On the other hand, Trp H ‘...‘H q “ r Trp H q “ r. So the identity over these two M0 -modules
are the same. Problèmes et choses à faire
Taking the dimension of both sides of (71.241), we
Is that enough to deduce that these are isomorphic as M0 -modules?
find s dimM0 pQH q “ r dimM0 pH q, which proves the statement in the case of TrpQq P .
Now, the functions dimM0 pQH q and TrpQq dimM0 pH q are increasing functions of TrpQq that
coincide on . Thus they are equal on .
Proposition 71.128.
We have the formula
1
TrpP T q “ TrpT q (71.244)
rM0 : M1 s
for every T P M0 .
` ˘
Proof. We have TrpP T q “ TrpP P T q “ TrpP T P q “ Tr ppT qP because of lemma 71.124. It is
sufficient
` to
˘ prove the claim (71.244) for T P M1 because when S P M0 , we would have TrpP Sq “
Tr ppSqP “ TrpSq{rM0 : M1 s because ppSq P M1 . Let thus T P M1 and consider the map
T ބ TrpP T q (71.245)
with T0 , Tj1 and Tj2 in M0 is a ˚-algebra. Indeed, in the multiplication of two such expressions,
we get monomials of the form
T1 lo
PoT on T3 “ T
2P
mo 1 ppT2 q P T3 .
looomooon (71.249)
“ppT2 qP PM0
By the double commutant theorem, the von Neumann algebra xM0 , P y is the strong closure of
such expressions, and
1
TrpT1 P T2 q “ TrpP T2 T1 q “ TrpT1 T2 q (71.250)
rM0 : M1 s
by proposition 71.128. Thus we have an explicit formula for the trace on xM0 , P y from the trace
over M0 .
Lemme 71.129.
If P P M1 , then we have
1
dimP M1 P pP H q “ dimM1 pH q, (71.251)
TrpP q
but P H is no more a M1 -module.
71.15. INDEX THEORY 2821
Proof. Let E be any M1 -module. Since P E is a P M1 P -module, we can define dimpP q pEq by
dimpP q pEq “ dimP M1 P pP Eq. (71.252)
That defines a dimension function on the M1 -modules, which is thus a multiple of the standard
one. We know that, as M1 -module, M1 P as dimension TrpP q, so that the proportionality factor
can be fixed on M1 P .
Problèmes et choses à faire
Lemme 71.130.
We have
dimM1 pH q dimM11 pH q “ 1 (71.253)
if M1 and M11 are finite factors.
Pourquoi ? Cela est fait dans la proposition 71.76. Et je crois que du côté de cette proposition, on trouve pas mal de réponses à pas mal de
questions ici.
Corollaire 71.131.
We have
rM0 : M1 s “ rM11 : M01 s (71.254)
when M0 P BpH q is a factor of type II1 and M1 is a subfactor of M0 .
Proof. No proof.
Théorème 71.132.
We have
rxM0 , P y : M0 s “ rM0 : M1 s (71.255)
under the same assumptions.
It turns out that, for that factor, we have R » P RP for every projection P P R. Choose isomor-
phism – : R Ñ P RP and — : R Ñ P K RP K , and form the algebra
where TrpP q can take any value between 0 and 1, and TrpP K q “ 1 ´ TrpP q. The possible values
of the index are then given by the range of the function
1 1
f pxq “ ` (71.262)
x 1´x
when x runs over r0, 1s. One easily checks that that range is r4, 8s.
M2 “ xM1 , P1 y. (71.263)
` ˘
For
` that definition,
˘ we see M1 as subalgebra of B L2 pM1 q . Now, M2 is also a subalgebra of
B L2 pM2 q , so that, defining P2 : L2 pM2 q Ñ L2 pM2 q as the projection onto L2 pM1 q, allows to
define ` ˘
M3 “ xM2 , P2 y Ñ B L2 pM2 q . (71.264)
Using that construction again and again, we get a sequence
M0 Ñ M1 Ñ M 2 Ñ . . . (71.265)
of factors defined by Mn`1 “ xMn , Pn y where Pn : L2 pMn q Ñ L2 pMn q is the orthogonal projection
onto L2 pMn´1 q. One can prove that
Proposition 71.133.
We have
(1) rMn : Mn`1 s “ ⁄´1 does not depend on n,
12. Recall that one needs a functional in order to define the infinite tensor product.
71.15. INDEX THEORY 2823
(2) TrpPn q “ ⁄,
(3) TrpPn Tn q “ ⁄ TrpTn q for every Tn P Mn .
Proof. No proof.
Lemme 71.134.
We have
pn pPn´1 q “ ⁄ (71.269)
where ⁄ “ TrpPn q does not depends on n by the proposition 71.133.
Proof. Since the map pS,`T q fiÑ TrpST ˘ q is a nondegenerate bilinear functional on Mn´1 , it is
sufficient to prove that Tr Spn pPn´1 q “ ⁄ TrpSq for every S P Mn´1 . Using point (4) of proposi-
tion 71.123, we have Spn pPn´1 q “ pn pSPn´1 q, so that
` ˘
Tr Spn pPn´1 q “ TrpSPn´1 q “ ⁄ TrpSq, (71.270)
Proposition 71.135.
The projections Pi fulfil the algebra
when |n ´ m| • 2.
Proof. Let us prove the second one. Using lemma 71.134, we find
Corollaire 71.136.
The algebra generated by
t , P1 , . . . , Pn u (71.273)
is a finite dimensional C ˚ -algebra in which there exists an unique tracial state such that for every k,
Sketch of the proof. Let a reduced word in P1 , . . . , Pk be a word which is as small as possible using
the three rules (71.271). We prove by induction that such a reduced word contains Pk at most
once. For beginning, the only reduced word in P1 is P1 itself.
Now, let a word containing twice Pk . What is between two successive occurrences of Pk is a
word of Sk´1 . That word is reduced, and thus contains Pk´1 only once by induction hypothesis.
From rule (71.271c), the operator Pk commutes with Sk´2 , and then the rule (71.271b) reduces
the word (because Pk´1 appears only once).
Now, a general reduced word of Sk has only one Pk and thus has the form m1 Pk m2 where m1
and m2 are reduces words of Sk´1 . Of course, m1 them self decomposes in n1 Pk´1 n2 where ni are
reduces words of Sk´2 , and the process continues.
Every reduced word in S9 look like
Some comments
2824 CHAPTER 71. VON NEUMANN ALGEBRAS
— The sub words are made of consecutive projections. For example the sub word pP5 P3 q can
be rearranged as the two sub words pP3 qpP5 q by rule (71.271c).
— The sub words begin by P5 , P8 and P9 . Notice that 5 † 8 † 9. It will always be like that.
If not, a rearrangement is possible.
Now fix a n and consider the set of all the reduced words of Sn ; there are only finitely many of
them which will be labelled as Wi . We look at the matrix whose element ij is given by
TrpWi˚ Wj q. (71.276)
Each of Wi˚ Wj is a long string of elements of Sn and the cyclic property of the trace allows us to
do ` ˘ ` ˘
Tr p. . .qp. . .qpPn Pn´1 . . .q “ Tr pPn Pn´1 . . .qp. . .qp. . .q . (71.277)
Then, using the Markov property of proposition 71.133, we can extract Pn . That matrix is positive
semidefinite because it is formed of scalar products. If not, that would means that the assumption
M0 Ñ M1 is wrong.
f0 pxq “ 0 (71.278a)
f1 pxq “ 1 (71.278b)
fn pxq “ fn´1 pxq ´ xfn´2 pxq. (71.278c)
Since fn p0q “ fn´1 p0q, all these polynomials pass by the point p0, 1q. By induction, one can see
that ˆ ˙
1 sinpn◊q
fn “ n´1 , (71.279)
4 cos2 ◊ 2 cosn´1 p◊q sinp◊q
so that the smallest positive root of fn is
1
` ˘, (71.280)
4 cos2 fin
and fn is negative on the interval between 1{4 cos2 pfi{nq and 1{4 cos2 pfi{pn ´ 1qq.
Suppose ⁄´1 “ rM : M0 s and ⁄ ° 1{4. Now, assume that
1
⁄‰ (71.281)
4 cos2 pfi{nq
for any value of n P . In this case, we will found a negative diagonal entry of the matrix (71.276),
which is impossible because a semipositive definite matrix has no negative diagonal element.
Let
Qk “ ´ P1 _ . . . _ Pk . (71.282)
From lemma 69.11, the operator Qk is in fact a projection in a finite dimensional algebra, and is
thus a polynomial in the Pi ’s.
Lemme 71.137.
We have
fk p⁄q
pQk´1 Pk Qk´1 q2 “ Qk´1 P ` kQk´1 , (71.283)
fk`1 p⁄q
and
fk`1 p⁄q
Qk`1 “ Qk ´ Qk´1 Qk´1 , (71.284)
fk p⁄q
if fk`1 p⁄q, . . . , f1 p⁄q ‰ 0.
71.16. MODULAR THEORY 2825
Proof. No proof.
Corollaire 71.138.
The operators Qk´1 , Pk and Qk`1 are projection, and the product is a positive operator.
That product is conjugate-linear in his first argument and linear in the second. The algebra M
acts at left by multiplication on L2Ï pM q, by extending
T · S “ T S. (71.287)
The space L2Ï pM q has a distinguished vector: the one represented by P M that we call v. It is
a cyclic vector for M , i.e. the space M v is dense in L2Ï pM q. It is also separating, because T v “ 0
with T P M implies T “ 0.
Conversely, if H is an Hilbert space and v P H is cyclic and separating for a von Neumann
algebra, then the expression
ÏpT q “ xv, T vy (71.288)
defines a faithful (because separating) normal state on the von Neumann algebra. Thus, in fact
H » L2Ï pM q via an unitary isomorphism compatible with M .
A natural question arising is to know the commutant of M acting by left multiplication on
L2Ï pM q. Let us suppose that Ï is also a trace, so we drot the index and we write L2 pM q instead
of L2Ï pM q. We define
J : L2 pM q Ñ L2 pM q
(71.289)
T v fiÑ T ˚ v.
Lemme 71.139.
The map J has the following properties.
(1) J is conjugate-linear.
(2) It is an isometry in the sense of conjugate linear operators.
(3) J 2 “ .
(4) JM J Ñ M 1 .
(5) xSv, JT vy “ xJS ˚ v, T ˚ vy for every S and T in M .
2826 CHAPTER 71. VON NEUMANN ALGEBRAS
Proof. The proof of (2) is a computation using the definition of J, the definition of the inner
product and the fact that Ï is a state:
xJT1 v, JT2 vy “ xT1˚ v, T2˚ vyÏpT1 T2˚ q “ ÏpT2 T1˚ q “ ÏpT1˚ T2 q “ xT1 v, T2 vy. (71.290)
Définition 71.140.
A modular isometry for M with respect to the cyclic vector v is a map J : L2 pM q Ñ L2 pM q
which satisfies
(1) Jv “ v.,
(2) J is a conjugate linear isometry,
(3) J 2 “ 1,
(4) JM J Ñ M 1 ,
(5) xJSv, T vy “ xS ˚ v, JT ˚ vy for every S, T P M 1 .
Notice the difference between the claim (5) of lemma 71.139.
Lemme 71.141.
If v is cyclic for M , then it is separating for M 1 , and if it is separating for M , then it is cyclic for
M 1.
Proposition 71.142.
If J is a modular isometry, then JM J “ M 1 . In particular, M and M 1 are of same type.
Proof. We have to prove that T P JM J whenever T P M 1 . Let S and T be in M 1 , and let us prove
that JSJ and T commute. For that, we are going to prove that
which is sufficient because the product is nondegenerate. For all S and T in M 1 , one successively
has
xJSJT v, vy “ xJSJT v, J 2 vy “ xJv, SJT vy “ xv, SJT vy
(71.293)
“ xS ˚ v, JT vy “ xJSv, T ˚ vy “ xT JSJv, vy,
where we used the extra property Jv “ v which defines a modular isometry.
Now we repeat the same with S1 “ A˚ S and T1 “ T B where A P JM J Ñ M 1 and B P JM J.
The operators S1 and T1 belong to M 1 because S and T belong to M 1 . We find
Using the fact that v is cyclic, the vectors Bv and JAv separately span dense subspaces, so that,
as operators, T SJT “ T JSJ because they are equal on a dense subspace.
Proposition 71.143.
If J is a modular isometry for M (with respect to v), then it is a modular isometry for M 1 (with
respect to v) too.
Proof. The fact that JM 1 J “ M is obtained by application of J at left and right of both sides of
JM J “ M 1 . For the last one, compute
Proposition 71.144.
If ÏpT q “ xv, T vy is a trace, then JSv “ S ˚ v for every S P M 1 .
Proof. When T runs over M , the element T v spans a dense space, and we have, for T P M and
S P M 1:
So the inner product of JSv and S ˚ v on any T v are the same. That proves that JSv “ S ˚ v.
Théorème 71.145.
For every von Neumann algebra M and every cyclic and separating vector v, there exists a modular
isometry.
Proof. No proof.
Corollaire 71.146.
The algebras M and M 1 are always algebraically isomorphic.
71.16.2 Example
Let M pGq where G is a non unimodular locally compact group. Thus M pGq “ Cc pGq2 . A
function f P Cc pGq defines an operator on L2 pGq by
ª
pf ›qpgq “ f phq›ph´1 gq dµphq (71.297)
G
If G was unimodular, this would be a tracial weight, but in the non unimodular case, Ï is not even
a trace on Cc pGq.
Lemme 71.147.
We have
` ˘
Ïpf1 f2 q “ Ï f2 ‡pf1 q (71.303)
That lemma says that the non triviality of ‡ measures at what extend Ï is not a trace.
Proof. No proof.
If Ï is a trace, then the class of identity in rG, G0 s is cyclic and one can make a GNS
construction. Notice that ‡z is not a ˚-automorphism because
‡z pf ˚ q “ ‡´z̄ pf q˚ . (71.304)
and let H0 “ A · v Ñ H .
Lemme 71.148.
We have
}‡1{2 pf q˚ v} “ }f v}, (71.306)
We have
` ˘ ` ˘
}‡1{2 pf ˚ qv}2 “ Ï ‡1{2 pf ˚ q˚ ‡1{2 pf ˚ q “ Ï ‡´1{2 pf q‡1{2 pf ˚ q . (71.308)
Now we apply the relation (71.307) with z “ ´1{2, and we find that the latter line is equal to
` ˘ ` ˘ ´ ` ˘¯
Ï ‡´1 pf q‡0 pf ˚ q “ Ï ‡´1 pf qf ˚ “ Ï f ˚ ‡1 ‡´1 pf q “ Ïpf ˚ f q “ }f v}2 . (71.309)
Proposition 71.149.
The formula
Jpf vq “ ‡1{2 pf ˚ qv (71.310)
Proof. No proof.
14. Since A is not a C ˚ -algebra, we cannot invoke a GNS construction here.
71.16. MODULAR THEORY 2829
‡z “ T z ST ´z . (71.312)
We have
` ˘ ` ˘
ÏpS1 S2 q “ TrpS1 S2 T q “ TrpS2 T S1 q “ TrpS2 T S1 T ´1 T q “ Tr S2 ‡pS1 qT “ Ï S2 ‡pS1 q ,
(71.313)
thus ‡1 has the same property as before to measure the lack of tracial property of Ï.
Once again, we pick v, the vacuum of the GNS representation, i.e. the representative of the
identity. The map
J : Sv fiÑ ‡1{2 pS ˚ qv (71.314)
is a modular isometry for M with respect to v.
Proof. The fact for the map (71.315) to be an isometry means that xT ›, T ›y “ xT ˚ ›, T ˚ ›y. This
is equivalent to
x›, T ˚ T ›y “ x›, T T ˚ ›y, (71.316)
which means that › is a trace vector.
Théorème 71.151.
If v is a cyclic and separating vector for M Ñ BpH q, and if
then there is a strongly dense ˚-subalgebra A Ñ M and a one parameter group of automorphisms
‡z : A Ñ A (z P ) such that
(1) The map z fiÑ ‡z pSq is holomorphic as map from into BpH q in the sense that it accepts
a power expansion ‡z “ T0 ` zT1 ` z 2 T2 ` . . .
(2) If z P i , then ‡z extends to a strongly continuous one parameter group of automorphisms
on M which only depends on Ï when the other conditions are satisfied.
(3) The maps Ï and ‡ are related by
` ˘
ÏpST q “ Ï T ‡pSq (71.318)
S: H Ñ H
(71.320)
T v fiÑ T ˚ v.
For a general v, this is not an isometry (this is even unbounded), but we can try to have in idea
of what is S ˚ . Since S is conjugate linear, the equation to solve is
We have xT1˚ v, T2 vy “ xS ˚ T2 v, T1 vy. Since the Hilbert space H is build from a GNS construction,
the inner product is in fact given in terms of Ï:
` ˘ ` ˘
xT1˚ v, T2 vy “ ÏpT1 T2 q “ Ï T2 ‡pT1 q “ Ï ‡´1 pT2 qT1 “ x‡´1 pT2 q˚ v, T1 vy “ x‡pT2˚ qv, T1 cy,
(71.322)
so that
S ˚ pT vq “ ‡pT ˚ qv, (71.323)
which is an equation for S ˚ in the same time as for ‡. After computations, one shows that
Conclusion: the automorphism ‡´1 (which is the inverse of ‡1 ) can be deduced from S. Moreover,
one can prove that S ˚ S is a linear (no more conjugate linear) positive, selfadjoint operator and
Thus, in fact, the whole set t‡z uzP can be recovered from S and thus from Ï.
Notice that equation (71.325) makes no sense in the infinite dimensional case. More analytic
work is needed to be right.
This is anˆequivalence
˙ relation.
0
Since is a partial isometry, the automorphism “t has to send it to another partial
0 0
isometry, i.e. ˆ ˙ ˆ ˙
0 0 Ut
“t “ (71.327)
0 0 0
where Ut is unitary. I we apply “t to the equality
ˆ ˙ˆ ˙ˆ ˙ ˆ ˙
0 0 S 0 0 0 0
“ , (71.328)
0 0 0 0 0 0 S
we find ˆ ˙ ˆ ˙
0 0 0 0
“t “ , (71.329)
0 Ut˚ 0
so that
Ut˚ –t pSqUt “ —t pSq. (71.330)
The conclusion is that –t and —t only differ by an inner automorphism.
When M “ BpH q, all automorphisms are inner and all the one parameter families of auto-
morphisms are equivalent to the identity.
71.16. MODULAR THEORY 2831
AutpM q
Ñ OutpM q “ . (71.331)
InnpM q
We recall that if M is a von Neumann algebra of operators of the Hilbert space H , a vector ›
is cyclic if H “ M › and it is separating for M is T › “ 0 if and only if T “ 0.
Ï : 2 pM q Ñ
ˆ ˙
T11 T12 1` ˘ (71.332)
fiÑ Ï1 pT11 q ` Ï2 pT22 q ,
T21 T22 2
S› : H Ñ H (71.333)
the closure. Thus everywhere it makes sense we have S› pT ›q “ T ˚ ›. The operator S› has the
polar decomposition
1{2
S› “ J› › . (71.334)
1{2
The antilinear part J› is said the modular conjugation while the operator › is the modular
operator associated with the pair pM, ›q. We have › “ S›˚ S› .
Now, instead of referring to the cyclic separating vector, we can consider a faithful normal
state Ê and think about its associated cyclic separating vector in the GNS cyclic representation
pfiÊ , HÊ , ›Ê q. In this setting, the modular group of automorphism of the pair pM, Êq is ‡ Ê
defined by
` ˘
‡tÊ pT q “ fiÊ´1 it ´it
Ê fiÊ pT q Ê (71.335)
` ˘
for T P M and t P . Here Ê is the modular operator associated with the pair fiÊ pM q, ›Ê .
Théorème 71.153.
The modular group ‡ Ê : Ñ AutpM q associated with the pair pM, Êq satisfies the following con-
ditions (the modular condition):
(1) Ê is ‡ Ê -invariant, that means that Ê “ Ê ˝ ‡tÊ .
(2) for every x, y P M , there exists a bounded function Fxy on the strip
Let H be an Hilbert space and an acting von Neumann algebra M of compact operators on
H . Let › be a cyclic separating vector. We define
j› : M Ñ M 1
(71.338)
x fiÑ J› xJ›
Tomita’s theorem shows that this map is well defined and that it takes values in M 1 .
If › is a trace vector (definition 57.8), the map j› is the involution:
j› pxq “ x˚ (71.339)
A standard form of a von Neumann algebra M acting on H is a self dual cone H ` and an
antilinear involution J such that pM, H , H ` , Jq satisfies
(1) JM J “ M 1 ;
(2) JxJ “ x˚ for every x P M X M 1 ;
(3) J÷ “ ÷ for every ÷ P H ` ;
(4) pxJxJqH ` Ä H ` for every x P M .