Lecture Notes on
C Algebras and Quantum Mechanics
Draft: 8 April 1998
N.P. Landsman
Kortewegde Vries Institute for Mathematics, University of Amsterdam, Plantage Muidergracht 24, 1018 TV AMSTERDAM, THE NETHERLANDS
email: npl@wins.uva.nl homepage: http://turing.wins.uva.nl/ npl/ telephone: 0205256282 o ce: Euclides 218a
2
CONTENTS
3
Contents
1.1 1.2 1.3 1.4 1.5
1 Historical notes
Origins in functional analysis and quantum mechanics Rings of operators (von Neumann algebras) . . . . . . Reduction of unitary group representations . . . . . . The classi cation of factors . . . . . . . . . . . . . . . C algebras . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5 5 6 8 11
5
2 Elementary theory of C algebras
2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 2.10 2.11 2.12 2.13 2.14
Basic de nitions . . . . . . . . . . . . . . . Banach algebra basics . . . . . . . . . . . . Commutative Banach algebras . . . . . . . Commutative C algebras . . . . . . . . . . Spectrum and functional calculus . . . . . . Positivity in C algebras . . . . . . . . . . . Ideals in C algebras . . . . . . . . . . . . . States . . . . . . . . . . . . . . . . . . . . . Representations and the GNSconstruction The Gel'fandNeumark theorem . . . . . . . Complete positivity . . . . . . . . . . . . . . Pure states and irreducible representations . The C algebra of compact operators . . . . The double commutant theorem . . . . . .
12
12 15 20 25 28 30 32 35 37 39 40 43 47 54 56 58 61 63 65 67 68 73 75 78 79 81 83 84 85 86 88
3 Hilbert C modules and induced representations
3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 Vector bundles . . . . . . . . . . . . . . . . . Hilbert C modules . . . . . . . . . . . . . . . The C algebra of a Hilbert C module . . . Morita equivalence . . . . . . . . . . . . . . . Rie el induction . . . . . . . . . . . . . . . . The imprimitivity theorem . . . . . . . . . . Group C algebras . . . . . . . . . . . . . . . C dynamical systems and crossed products . Transformation group C algebras . . . . . . The abstract transitive imprimitivity theorem Induced group representations . . . . . . . . . Mackey's transitive imprimitivity theorem . .
56
4 Applications to quantum mechanics
4.1 4.2 4.3 4.4 4.5
The mathematical structure of classical and quantum mechanics Quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Stinespring's theorem and coherent states . . . . . . . . . . . . . Covariant localization in con guration space . . . . . . . . . . . . Covariant quantization on phase space . . . . . . . . . . . . . . .
83
Literature
89
4
CONTENTS
Heisenberg's paper initiating matrix mechanics was followed by the `Dreimannerarbeit' of Born. In 1908 his student E. Riesz studied the space of all continuous linear maps on `2 (1912). Hilbert's memoirs on integral equations appeared between 1904 and 1906. which contains a heuristic and mathematically unsatisfactory account of quantum mechanics in terms of linear spaces and operators. saw his way through all structures and mathematical di culties.1 Origins in functional analysis and quantum mechanics
The emergence of the theory of operator algebras may be traced back to (at least) three developments.e. the abstract concept of a Hilbert space was still missing. this book remains the de nitive account of the mathematical structure of elementary quantum mechanics. However. Hilbert had been interested in the mathematical structure of physical theories for a long time his Sixth Problem (1900) called for the mathematical axiomatization of physics.
1. and proved the mathematical equivalence between matrix mechanics and wave mechanics. which he called `wave mechanics'. and introducing notions such as density matrices and quantum entropy. and various examples of L2spaces emerged around the same time. The latter may be de ned by its notion of convergence: a sequence fAn g of bounded operators weakly converges to A when ( An ) ! ( A ) for all 2 H.e. was much discussed at the time. spectral theory. Initiating and largely completing the theory of selfadjoint operators on a Hilbert space. F. which at rst sight looked completely di erent.2 Rings of operators (von Neumann algebras)
. which at the time was called `matrix mechanics'. It was von Neumann alone who. This type of convergence is partly motivated by quantum mechanics. (von Neumann's book was preceded by Dirac's The Principles of Quantum Mechanics (1930).. and Jordan (1926) all three were in Gottingen at that time. and the three wrote a joint paper on the subject (now obsolete). Born turned to his former teacher Hilbert for mathematical advice.5
1 Historical notes
1. culminating in his book Mathematische Grundlagen der Quantenmechanik (1932). Hilbert thus ran a seminar on the mathematical structure of quantum mechanics. he formulated the abstract concept of a Hilbert space.) In one of his papers on Hilbert space theory (1929). Heisenberg. von Neumann de nes a ring of operators M (nowadays called a von Neumann algebra) as a subalgebra of the algebra B(H) of all bounded operators on a Hilbert space H (i. sequentially complete) in the weak operator topology. provided that A is selfadjoint and has unit norm. Schmidt de ned the space `2 in the modern sense. The relationship and possible equivalence between these alternative formulations of quantum mechanics. Aided by his assistants Nordheim and von Neumann. a subalgebra which is closed under the involution A ! A ) that is closed (i. at the age of 23. developed the spectral theory of bounded as well as unbounded normal operators on a Hilbert space. The work of Hilbert and his pupils in Gottingen on integral equations. in which ( A ) is the expectation value of the observable A in the state . It was clear from either approach that the body of work mentioned in the previous paragraph was relevant to quantum mechanics. and in nitedimensional quadratic forms (1904) The discovery of quantum mechanics by Heisenberg (1925) in Gottingen and (independently) by Schrodinger in Zurich (1926) The arrival of John von Neumann in Gottingen (1926) to become Hilbert's assistant. Heisenberg discovered a form of quantum mechanics. In a series of papers written between 19271932. Schrodinger was led to a di erent formulation of the theory. Born was one of the few physicists of his time to be familiar with the concept of a matrix in previous research he had even used in nite matrices (Heisenberg's fundamental equations could only be satis ed by in nitedimensional matrices).
and the bicommutant M00 is simply (M0 )0 . Among von Neumann's interest in quantum mechanics was the notion of entropy he wished to de ne states of minimal information. one takes U to be the set of all unitaries in M0 . Von Neumann's bicommutant theorem implies a useful alternative characterization of von Neumann algebras from now on we add to the de nition of a von Neumann algebra the condition that M contains I. and. so that. In the opposite direction. When G is compact. given M. indeed. to some extent been realized in algebraic quantum eld theory (Haag and followers).) In the same paper. containing the unit operator I. This programme has. where U is the identity map. it does not. but for in nitedimensional Hilbert spaces this state may no longer be de ned. Von Neumann's motivation in studying rings of operators was plurifold. Density matrices may be regarded as states on the von Neumann algebraB(H) (in the sense of positive linear functionals which map I to 1). (As we shall see. viz. a von Neumann algebra is automatically normclosed as well. although in algebraic quantum eld theory the basic algebras of local observables are. a theorem of Wedderburn says that a von Neumann algebra on a nitedimensional Hilbert space is (isomorphic to) a direct sum of matrix algebras.) Finally. that is. This hope has not materialized. Instead. As we shall see. this implies that U is a multiple of a xed irreducible
. Motivation from a di erent direction came from the structure theory of algebras. so that one cannot characterize a physical system by saying that `its Hilbert space of (pure) states is L2 (R3 )'. (Note that a given group U of unitaries on H may be regarded as a representation U of U itself. in terminology to be introduced later on.3 Reduction of unitary group representations
The (possible) reduction of U (G) is determined by the von Neumann algebras U (G)00 and U (G)0 . or a similar result in which direct sums are replaced by direct integrals (see below). but are all isomorphic to the unique hyper nite factor of type III1 (see below). he knew that all Hilbert spaces of a given dimension are isomorphic. Furthermore. (Since the weak topology is weaker than the uniform (or norm) topology on B(H). still holds when the dimension of H is in nite. The representation U is called primary when U (G)00 has a trivial center. In the present context. not of the form B(H). one trivially veri es that the commutant of any set of bounded operators is weakly closed. von Neumann proves what is still the basic theorem of the subject: a subalgebra M of B(H). In one direction. where K? is the orthogonal complement of K in H. conversely. the natural topology on a von Neumann algebra is neither the weak nor the uniform one. However. there are von Neumann algebras on in nitedimensional Hilbert spaces which do admit states of minimal information that generalize I=n. When H = C n for n < 1. This alternative characterization indicates why von Neumann algebras are important in physics: the set of bounded operators on H which are invariant under a given group representation U (G) on H is automatically a von Neumann algebra. whereas the commutant of a set of bounded operators which is closed under the involution is a algebra. B(H) is itself a von Neumann algebra. when U (G)00 \ U (G)0 = C I. Here the commutant M0 of a collection M of bounded operators consists of all bounded operators which commute with all elements of M. For example. Von Neumann wondered if this. every von Neumann algebra arises in this way. The commutant of a group U of unitary operators on a Hilbert space is a von Neumann algebra. U is irreducible i U (G)0 = C I (Schur's lemma). such a state is given by the density matrix = I=n. a von Neumann algebra becomes a C algebra when one changes the topology from the weak to the uniform one. von Neumann's motivation came from group representations.6
1 HISTORICAL NOTES
For example.)
1. His primary motivation probably came from quantum mechanics unlike many physicists then and even now. This theorem is remarkable. the factors of type II1 (see below). in relating a topological condition to an algebraic one one is reminded of the much simpler fact that a linear subspace K of H is closed i K?? . von Neumann hoped to characterize quantummechanical systems by algebraic conditions on the observables. von Neumann hoped that the divergences in quantum eld theory might be removed by considering algebras of observables di erent from B(H). is weakly closed i M00 = M. so that U is discretely reducible.
Instead of decomposing H as a direct sum.0 it then makes sense to de ne ( The direct integral ) :=
Z
d ( ) ( ( ) ( )) : d ( )H
is discrete. one may take = R with Lebesgue measure . and V U (x)V is diagonal in the above sense. of course. and U ' U IK. The solution to this problem was given by von Neumann in a paper published in 1949. We then write A=
Z
d ( )A :
Thus a unitary group representation U (G) on H is diagonal when
U (x) ( ) = U (x)
for all x 2 G. To de ne the direct integral of the H with respect to the measure . When H is separable. The irreducible wouldbe subspaces of H would be spanned by the vectors p (y) := exp(ipy). write
U=
Z
d ( )U :
Reducing a given representation U on some Hilbert space then amounts to nding a unitary map V between H and some direct integral Hilbert space. This occurs already in the simplest possible cases. with A = U (x). one needs a sequence of sections f n g satisfying the two conditions that rstly the function ! ( n ( ) m ( )) be measurable for all n m. one should decompose it as a direct integral. of all sections which contains all n . in which case we. When G is not compact. such as the regular representation of G = R on H = L2 (R) that is. unitary representations may be reducible without containing any irreducible subrepresentation. Complete reduction therefore calls for a further direct integral decomposition of primary representations this will be discussed below. ^ of all To completely reduce U . There then exists a unique maximal linear subspace . but written in the thirties (the ideas in it must have guided von Neumann from at least 1936 on). so that is the space G equivalence classes of irreducible unitary representations of G. For example. and for which all sections ! ( ) are measurable. x).3 Reduction of unitary group representations
7
representation U on a Hilbert space H .
Z
is then by de nition the subset of . A section of this family is a function : ! fH g 2 for which ( ) 2 H . one may always reduce a unitary representation in such a way that the U occurring in the decomposition are primary. and secondly that for each xed the n span H . and take the sequence f ng to consist of a single strictly positive measurable function. (To do so. For 2 . one puts U (x) (y) = (y .) This means that rstly one has a measure space ( ) and a family of Hilbert spaces fH g 2 . one needs the U to be irreducible. An operator A on this direct integral Hilbert space is said to be diagonal when
A ( )=A ( ) for some (suitably measurable) family of operators A on H . so that H ' H K. When the direct integral reduces to a direct sum. but still assumed to be locally compact.0 of the space . This leads to the direct integral decomposition Z L2 (R) = dp Hp
R
.1.0 of functions for which ( ) < 1. but these functions do not lie in L2 (R). such that each H carries a representation U . and this central decomposition of U is essentially unique. one needs to assume that H is separable.
and profound papers (3 of which were in collaboration with Murray) in which the study of his `rings of operators' was initiated. a group is type I when all its primary representations are of type I. for example. A locally compact group is said to be type I or tame when every primary representation is a multiple of a xed irreducible representation in other words.4 The classi cation of factors
. all nitedimensional Lie groups and all discrete groups).e. these papers form `perhaps the most original major work in mathematics in this century'. If not. curious phenomena may occur. t 2 R. where the measures 1 and 2 are disjoint (that is. (According to I. They noticed that one may de ne an equivalence relation on the set of all projections in M. so that U has been diagonalized: the U (x) above are now the onedimensional operators Up (x) = exp(ipx) on Hp = C . An equivalent characterization of arises when we write M = U (G)0 for some unitary representation U of a group G (as we have seen. Another example. is the free group on two generators. well known to von Neumann.E. We have therefore completely reduced U . A primary representation is said to be of type I when it may be decomposed as the direct sum of irreducible subrepresentations these subrepresentations are necessarily equivalent. M is generated by its projections. that is. As far as the reduction of unitary representations is concerned. By de nition. and annihilates pH? . in which p q i there exists a partial isometry V in M such that V V = p and V V = q. all (proper) subrepresentations of a primary type III representation are equivalent. di cult. Moreover. In case that U is not equivalent to Uh . representations of the latter type may be decomposed in two alternative ways
U=
Z
^ G
d 1( ) U =
Z
^ G
d 2( ) U
^ ).
V (p) =
Z
R
dy e. respectively). is the group of matrices of the form 0 eit 0 z 1 @ 0 ei t w A 0 0 1 where is an irrational real number. one simply performs a Fourier transform V : L2 (R) ! L2 (R). discovered at a later stage.ipy (y):
This leads to V U (x)V (p) = exp(ipx) (p). In that case U is equivalent to Uh indeed. When M B(H). for in that case one may take any V with the above properties.8
1 HISTORICAL NOTES
in which each Hp is C . there exist two radically di erent classes of locally compact groups (the class of all locally compact groups includes.. i. More bizarrely. it is said to be a representation of type II. Segal. When G is wild. when pH qH. Hence when M = B(H) one has p q i pH and qH have the same dimension. Murray and von Neumann de ne a partial orderering on the collection of all projections in M by declaring that p q when pq = p. and U is not of type I. it is of type III. This induces a partial orderering on the set of equivalence classes of projections by putting p] q] when the
1. When U is neither of type I nor of type II. this always applies) then p q i the subrepresentations pU and qU (on pH and qH. Between 1936 and 1953 von Neumann wrote 5 lengthy. An example of a wild group. and z w 2 C . are unitarily equivalent.) The analysis of Murray and von Neumann is based on the study of the projections in a von Neumann algebra M (a projection is an operator p for which p2 = p = p) indeed. supported by disjoint subsets of G A reducible primary representation U may always be decomposed as U = Uh Uh . the group is called nontype I or wild. a wild group has primary unitary representations which contain no irreducible subrepresentations. the operator V is unitary from pH to qH. To reduce the regular representation of R on L2 (R).
. where n < 1: M has minimal projections.
type II1: M has no minimal projections. all projections are nite. the range of d is the interval 0 1]. q p and q 6= p). A projection p is called minimal or atomic when there exists no q < p (i.4 The classi cation of factors
9
equivalence classes p] and q] contain representatives p ~ and q ~ such that p ~ q ~. d(p) = d(q) i p]
q] 3. One then has the following possibilities for a factor M. in order to distinguish in nitedimensional but inequivalent projections. distinguishing the equivalence classes p]. For M = B(H) this generalized notion of niteness coincides with the usual one. More generally. satisfying 1. and I is nite. For this to be the possible. type I1 : M has minimal projections. Normalizing d such that d(I) = 1.
Here a projection in B(H) is called nite when pH is nitedimensional. For M = B(H) this actually de nes a total ordering on the equivalence classes. Such a factor is isomorphic to B(H) for separable in nitedimensional H. As we have seen. this is independent of the choice of p 2 p] and q 2 q]. A factor of type In is isomorphic to the algebra of n n matrices. d is unique up to nite rescaling. all nonzero projections are in nitedimensional in the usual sense. and d takes the values f0 1 : : : 1g. The study of von Neumann algebras acting on separable Hilbert spaces H reduces to the study of factors. when pH and qH are orthogonal
4. in which p] q] when pH has the same dimension as qH as we just saw..
. The range of d is 0 1]. Murray and von Neumann showed that the set of equivalence classes of projections in M is totally ordered by whenever M is a factor. all projections are in nitedimensional in the usual
sense. the dimension function d is a `renormalized' version of the usual one. for the factor M = B(H) the dimension d of a projection is a complete invariant. but M has nitedimensional projections in the sense of Murray and von Neumann. and d(0) = 0 2.1. and d takes the values f0 1 : : : ng. Murray and von Neumann de ne a projection to be nite when it is not equivalent to any of its (proper) subprojections an in nite projection is then a projection which has proper subprojections to which it is equivalent. for von Neumann proved that every von Neumann algebra M B(H) may be uniquely decomposed.e. d(p + q) = d(p) + d(q) when pq = 0 (i.e. d(p) < 1 i p is nite. Murray and von Neumann now proved that on any factor M (acting on a separable Hilbert space) there exists a function d from the set of all projections in M to R+ 1. One may say that. A von Neumann algebra M is a factor when M \ M0 = C I when M = U (G)0 this means that M is a factor i the representation U is primary. type II1 : M has no minimal projections. and I is in nite. The dimension is a function from the set of all projections in B(H) to R+ 1. For M = U (G)0 the decomposition of H amounts to the central decomposition of U (G).
type In. A rst classi cation of factors (on a separable Hilbert space) is now performed by considering the possible niteness of its projections and the range of d. yet some are nite in the sense of Murray and von Neumann. as in
H =
M =
Z Z
d ( )H d ( )M
where (almost) each M is a factor. but in other factors all projections may be in nite in the usual sense. Moreover. d(p) > 0 when p 6= 0. satisfying the above properties.
It is hard to construct an example of a II1 factor. Haag. this theory leads to a natural timeevolution on certain factors. In the sixties. a group of physicists. where. Factors of type I1 and II1 are then called semi nite.g. In view of the range of d. This idea initiated algebraic quantum eld theory. may be extended to a positive linear functional tr on M. However.. such that M(O1 ) M(O2 ) when O1 O2 . and von Neumann did the latter by himself. Godement. For he showed that on a II1 factor M the dimension function d. Kaplansky. One sometimes says that a factor is nite when I is nite (so that d(I) < 1) hence type In and type II1 factors are nite. On the mathematics side. de ned on the projections in M. the existence of a II1 factor solved one of the problems that worried him in quantum mechanics. the TomitaTakesaki theory and the examples of type III factors
type III: M has no minimal projections. in 1957 Haag observed a formal similarity between the collection of all von Neumann algebras on a Hilbert space and the set of all causally closed subsets of Minkowksi spacetime. which has helped both subjects. and Schwinger. and until the sixties only a handful of mathematicians worked on operator algebras (e. which was to play an essential role in the classi cation of factors of type III. which nowadays is called modular theory or TomitaTakesaki theory (apart from clarifying Tomita's work. Segal. and Adel'sonVel'skii. and others). and gave von Neumann the state of minimal information he had sought. This has led to an extremely fruitful intercation between physics and mathematics. From then till the present day. This `trace' satis es tr(I) = d(I) = 1. the representation U is primary i M is a factor. and even harder to write down a type III factor. and type III factors are purely in nite. an extraordinary coincidence took place. but only 5 years after he and Murray had recognized that the existence of type III factors was a logical possibility. and d assumes the values f0 1g. On the physics side. and is therefore called the KMS condition. Tomita developed a technique in the study of von Neumann algebras. with the property that tr(UAU ) = tr(A) for all A 2 M and all unitaries U in M. Dixmier. where O? consists of all points that are spacelike separated from O. U is of a given type i M is of the same type. Mautner. and M(O)0 = M(O? ).10
1 HISTORICAL NOTES
With M = U (G)0 . a subject that really got o the ground with papers by Haag's pupil Araki in 1963 and by Haag and Kastler in 1964. he believed these de ned some form of continuous geometry. Moreover. Here a region O in spacetime is said to be causally closed when O?? = O. Kadison. Von Neumann was fascinated by II1 factors. as we have seen. One of the result has been that in realistic quantum eld theories the local algebras M(O) must all be isomorphic to the unique hyper nite factor of type III1 discussed below. Thus Haag proposed that a quantum eld theory should be de ned by a net of local observables this is a map O ! M(O) from the set of all causally closed regions in spacetime to the set of all von Neumann algebras on some Hilbert space. (Hence von Neumann's belief that physics should use II1 factors has not been vindicated. The operation O ! O? on causally closed regions in spacetime is somewhat analogous to the operation M ! M0 on von Neumann algebras. and they admitted having no tools to study type III factors. This condition leads to type III factors equipped with a timeevolution which coincided with the one of the TomitaTakesaki theory. In particular. The precise connection between von Neumann algebras and the decomposition of unitary group representations envisaged by von Neumann was worked out by Mackey. led by Haag.) A few years later (1967). all nonzero projections are in nitedimensional and equivalent in the usual sense as well as in the sense of Murray and von Neumann. At the time not many people were familiar with the di cult papers of Murray and von Neumann. realized that operator algebras could be a useful tool in quantum eld theory and in the quantum statistical mechanics of in nite systems. In the hands of Connes. they were unable to provide a further classi cation of all factors. Takesaki made essential contributions to this theory). and Winnink characterized states of thermal equilibrium of in nite quantum systems by an algebraic condition that had previously been introduced in a heuristic setting by Kubo. Murray and von Neumann managed to do the former.
. Among other things. Martin. Sakai. Partly for this reason he believed that physics should be described by II1 factors. Hugenholtz. algebraic quantum eld theory has attracted a small but dedicated group of mathematical physicists.
for which he was awarded the Fields Medal in 1982. connecting states to representations. showing that representations of C (G) bijectively correspond to unitary representations of G. (Experience shows that all factors playing a role in physics are hyper nite. In its present form. the sixties brought a fruitful interaction between C algebras and quantum physics. as well as because of its connections with a number of other elds of mathematics. in which the theory of inclusions of II1 factors was related to knot theory.5 C algebras
. There is a family of type III0 factors.5 C algebras
11
provided by physicists (Araki. As we have mentioned already. The same idea lies at the
1. In the midst of the Murrayvon Neumann series of papers. in which multiplication is (separately) continuous in the normtopology. In 1941 he de ned the concept of a Banach algebra. The rst instance where this led to a deep result which has not been proved in any other way is the theorem of Gel'fand and Raikov (1943). and nally showing that the states of an arbitrary C algebra A separate the elements of A.) The work of Connes. the second one to be given to the once obscure eld of operator algebras. This was proved by constructing a C algebra C (G) of the group G. a great admirer of von Neumann. As with von Neumann algebras. These are factors containing a sequence of nitedimensional subalgebras M1 M2 : : : M. Their paper also contained the rudiments of what is now called the GNS construction. for an in nitedimensional separable Hilbert space K. this construction is due to Segal (1947). and proved most basic results in the theory of commutative Banach algebras. try and nd a C algebra which encodes this structure in some way. In 1943 Gel'fand and Neumark de ned what is now called a C algebra (some of their axioms were later shown to be super uous). the theory of C algebras turned out to be interesting both for intrinsic reasons (structure and representation theory of C algebras). The unique hyper nite II1 factor was crucial in a spectacular development. Other examples of mathematical structures that may be analyzed through an appropriate C algebra are group actions. led to the following classi cation of hyper nite factors of type II and III (up to isomorphism): There is a unique hyper nite factor of type II1 . Woods. which in turn is has been classi ed in terms of concepts from ergodic theory. the unique hyper nite III1 factor plays a central role in algebraic quantum eld theory. For 6= 0 the factor of type III is unique. for each pair x 6= y there exists a unitary representation U for which U (x) 6= U (y). namely the tensor product of the hyper nite II1 factor with B(K). (In physics this factor occurs when one considers KMSstates at in nite temperature. Moreover. and others) eventually led to the classi cation of all hyper nite factors of type II and III (the complete classi cation of all factors of type I is already given by the list presented earlier). Segal returned to von Neumann's motivation of relating operator algebras to quantum mechanics. Here the strategy is to take a given mathematical structure. and many natural examples of factors constructed by purely mathematical techniques are hyper nite as well. stating that the unitary representations of a locally compact group separate the points of the group (that is. combining operator algebras with the theory of Banach spaces. and complex domains. Gel'fand initiated a separate development. and then obtain information about the structure through proving theorems about the C algebra of the structure. who generalized von Neumann's idea of a state as a positive normalized linear functional from B(H) to arbitrary C algebras. groupoids. labeled by 2 0 1]. and others. In 1990 Jones was awarded a Fields medal for this work. There is a family of type III factors. Moreover. Powers. such that M is the weak closure of n Mn . and proved the basic theorem that each C algebra is isomorphic to the normclosed algebra of operators on a Hilbert space. foliations.1. He proceeded to de ne an intrinsic spectral theory.) There is a unique hyper nite factor of type II1 . and even led to a new knot invariant.
k v k= j j k v k for all 2 C and v 2 V 4. The abbreviation `i ' means `if and only if'. We will denote a generic Banach space by the symbol B.1. it is enough to assume that (v v) 2 R for all v (use this reality with v ! v + iw).. Either way. An equation of the type a := b means that a is by de nition equal to b. K theory. cohomology.
2 Elementary theory of C algebras
All vector spaces will be de ned over C . An equivalent set of conditions is 1. should be translated into tools pertinent to the C algebra C (X ). unless we explicitly state otherwise. (v w) = (w v) for all v w 2 V 2. (v v) 0 for all v 2 V . and. De nition 2. which we will study in detail in these lectures. Finally. which is the same as the symbol . w k.2 A preinner product on a vector space V is a map ( ) : V V ! C such that 1. The strategy is now that the basic tools in the topology of X . when appropriate. ( 1 v1 + 2 v2 1 w1 + 2 w2 ) = 1 1 (v1 w1 ) + 1 2 (v1 w2 ) + 2 1 (v2 w1 ) + 2 2 (v2 w2 ) for all 1 2 1 2 2 C and v1 v2 w1 w2 2 V 2.1 Basic de nitions
CauchySchwarz inequality
j(v w)j2 (v v)(w w)
(2. and in (de Rham) cohomology theory. A vector space with a norm which is complete in the associated metric (in the sense that every Cauchy sequence converges) is called a Banach space. whose noncommutative version is even simpler than its usual incarnation. and that subsequently these tools should be generalized to noncommutative C algebras. k v + w k k v k + k w k (triangle inequality). k v k= 0 i v = 0 3. and index theory haven been uni ed and made noncommutative in the KK theory of Kasparov. A preinner product for which (v v) = 0 i v = 0 is called an inner product. Here the starting point is another theorem of Gel'fand.12
2 ELEMENTARY THEORY OF C ALGEBRAS
basis of noncommutative geometry and noncommutative topology. where X is a compact Hausdor space. De nition 2. in its di erential geometry. The equivalence between the two de nitions of a preinner product is elementary in fact. whose noncommutative version is called cyclic cohomology. (v 1 w1 + 2 w2 ) = 1 (v w1 ) + 2 (v w2 ) for all 1 2 2 C and v w1 w2 2 V 3.1. The basic tool in KK theory is the concept of a Hilbert C module.1 A norm on a vector space V is a map k k : V ! R such that 1. homology. (v v) 0 for all v 2 V . This strategy has been successful in K theory. one derives the
2. A norm on V de nes a metric d on V by d(v w) :=k v . stating that any commutative C algebra (with unit) is isomorphic to C (X ). k v k 0 for all v 2 V 2.1)
. and all functions will be C valued. The two main examples of Banach spaces we will encounter are Hilbert spaces and certain collections of operators on Hilbert spaces. to derive the rst axiom of the second characterization from the rst set of conditions.
by the cardinality of an arbitrary orthogonal basis). Now de ne a map A on B by Av := w = limn An v. it follows that (2.7)
k Av k k A k k v k
. In view of the comments following (2.2) that A .3 A Hilbert space is a vector space with inner product which is complete in the associated norm.3)
When A is bounded. The smallest such C is the norm k k:= sup fj (v)j v 2 B k v k= 1g: (2. as it follows almost immediately from its de nition (and from the properties of the the norm on B) that the operator norm is indeed a norm. De nition 2. Conversely. Since we have just seen that A 2 B(B).) When B is a Hilbert space H the expression (2. for any > 0 there is a natural number N ( ) such that k An .1. Proposition 2. Since B is complete by assumption. Am ) + Am .3).1. It follows from these properties that an inner product on V de nes a norm on V by k v k:= p(v v) the triangle inequality is automatic.2) The number k A k is the operator norm. Let fAn g be a Cauchy sequence in B(B). Am k k v k k v k (2.5 The space B(B) of all bounded operators on a Banach space B is itself a Banach space in the operator norm. Indeed.5) for n m > N ( ). in arguments involving continuous operators on a Banach space one almost always uses boundedness rather than continuity. (2. this proves that B(B) is complete.2. A Hilbert space is completely characterized by its dimension (i. it is wise to restrict oneself to bounded operators.. This map is obviously linear. (It is easily shown that a linear map on a Banach space is continuous i it is a bounded operator. Am is bounded. because k An v . Am k for all m > N ( ). the sequence fAn vg converges to some w 2 B. De nition 2. In other words.2) imply that k A . the sequence fAn vg is a Cauchy sequence in B. To obtain a satisfactory mathematical theory. then A is bounded. we obtain k Av . For arbitrary v 2 B.4) for all v 2 B. Since A = (A . Moreover. one therefore studies operators on a Hilbert space.2) becomes
1 k A k:= sup f(A A ) 2 j 2H (
) = 1g:
(2.1. We recall this concept in the more general context of arbitrary Banach spaces.6) for all m > N ( ) and all v 2 B. To obtain an interesting theory. Note that this inequality is valid even when ( ) is not an inner product.6) and (2. but we will never use this result.e.1 Basic de nitions
13
for all v w 2 V . Am v k k v k (2. so that fAn g converges to A. it only remains to be shown that B(B) is complete in the operator norm. but merely a preinner product. with operator norm k A k equal to the smallest possible C for which the above inequality holds. we infer that A is bounded. and B(B) is a linear space. Taking n ! 1 in (2. It now follows from (2. We will usually denote Hilbert spaces by the symbol H.5). of A. Am k < when n m > N ( ). and all v 2 B. Am v k k An . or simply the norm. We de ne a functional on a Banach space B as a linear map : B ! C which is continuous in that (v)j C k v k for some C .4 A bounded operator on a Banach space B is a linear map A : B ! B for which k A k:= sup fk Av k j v 2 B k v k= 1g < 1: (2. when for A 6= 0 there is a C > 0 such that k Av k C k v k for all v. This terminology is justi ed. rather than the Hilbert space itself.
for each v 2 B one has
It follows that multiplication in a Banach algebra is separately continuous in each variable. Hence B(H) is a algebra.6 For a functional 0 on a linear subspace B0 of a Banach space B there exists a functional on B such that = 0 on B0 and k k=k 0 k. in which for all A B 2 A one has k AB k k A k k B k : (2. For v 6= 0 we may de ne a functional 0 on C v by 0 ( v) = . As in this case. we infer that k A k2 k A A k k A k k A k : (2. De nition 2. The operator adjoint A ! A on a Hilbert space. without proof. Similarly to the proof of 2.4) to estimate k A k2 = (A A ) = ( A A ) k k k A A k k A A k k k2 : Using (2.8)
The dual B of B is the space of all functionals on B.
.1. we quote.11) A algebra is an algebra with an involution. and extend it to a functional on B with norm 1. Recall that an algebra is a vector space with an associative bilinear operation (`multiplication') : A A ! A we usually write AB for A B . de ned by the property ( A ) := (A ).9) yields k A k k A k.1.1. This algebra has additional structure. Theorem 2. de nes an involution on B(H). In other words. for any Banach space B the space B(B) of all bounded operators on B is a Banach algebra. De nition 2. Replacing A by A and using (2. we pick 2 H.8 A Banach algebra is a Banach space A which is at the same time an algebra. To see how the norm in B(H) is related to the involution.14
2 ELEMENTARY THEORY OF C ALGEBRAS
k ABv k k A k k Bv k k A k k B k k v k : Hence from (2.1.5. In what follows.9 An involution on an algebra A is a reallinear map A ! A such that for all A B 2 A and 2 C one has A = A (2. an element A of a C algebra A is called selfadjoint when A = A we sometimes denote the collection of all selfadjoint elements by AR := fA 2 Aj A = Ag: (2. so that k A k=k A k.8).7 When (v) = 0 for all 2 B then v = 0. each functional de ned on a linear subspace of B has an extension to B with the same norm. Corollary 2.4) twice.13) 2 + i 2i every element of A is a linear combination of two selfadjoint elements.9) (2. we derive the crucial property k A A k=k A k2. the fundamental HahnBanach theorem. using (2. Moreover.14) This leads to k A k k A k. It is clear that B(B) is an algebra under operator multiplication.A A = A0 + iA00 := A + (2. Substituting this in (2. one shows that B is a Banach space.2) we obtain k AB k k A k k B k. This motivates the following de nition. As we have just seen.1. we will restrict ourselves to the case that B is a Hilbert space H this leads to the Banach algebra B(H).12) Since one may write A A. For later use.10) (AB ) = B A ( A) = A : (2.3) and (2.14). and use the CauchySchwarz inequality and (2.
10 A C algebra is a complex Banach space A which is at the same time a algebra.16)
In other words. this implies I = I.18) (2.20) follows from (2. such that for all A B 2 A one has k AB k kAk kBk (2.12 A morphism between C algebras A B is a (complex) linear map ' : A ! B such that
'(AB ) = '(A)'(B ) '(A ) = '(A)
(2. For this reason the condition that an isomorphism be isometric is not included in the de nition.
The material in this section is not included for its own interest.2 Banach algebra basics
15
De nition 2.20) A Banach algebra with unit is called unital. so that I I = I taking the adjoint. states the converse of this: each C algebra is isomorphic to a normclosed algebra in B(H). We have just shown that B(H) is a C algebra. Moreover. a Banach algebra with involution. Form the vector space AI := A C (2. so that (2.8.22)
.17) The same argument proves the following. that an injective morphism (and hence an isomorphism) between C algebras is automatically isometric. Note that in a C algebra the property IA = AI = A already implies. we infer that for all elements A of a C algebra one has the equality k A k=k A k : (2. as follows.21) and make this into an algebra by means of (A + I)(B + I) := AB + B + A +
I
2. Even in that special context. A unit in a Banach algebra A is an element I satisfying IA = AI = A for all A 2 A. however. We often write z for z I. and k I k= 1: (2.15) k A A k = k A k2 : (2. De nition 2. one derives k A k k A k as in the preceding paragraph.1.11 A Banach algebra in which k A k2 k A A k is a C algebra.19)
for all A B 2 A. Combining (2. a C algebra is a Banach algebra in which (2. it is enlightening to see concepts such as the spectrum in their general and appropriate setting.2. Hence the axioms in 2.16). of course.1.20) take A = I .2 Banach algebra basics
(2. Two C algebras are isomorphic when there exists an isomorphism between them. An isomorphism is a bijective morphism.16) holds. When a Banach algebra A does not contain a unit. although the axioms make no reference to Hilbert spaces at all. each (operator) normclosed algebra in B(H) is a C algebra by the same argument.15).10 characterize normclosed algebras on Hilbert spaces. For later use we state some selfevident de nitions. we can always add one. where z 2 C .1. for some Hilbert space H. It is remarkable. A much deeper result. Recall De nition 2. which we will formulate precisely and prove in due course.1. Here a Banach algebra is. One immediately checks that the inverse of a bijective morphism is a morphism. but because of its role in the theory of C algebras.1. Lemma 2.16) and (2. (2.
for n > m one has
k (A + I)(B + I) k k A k k B k +j j k B k +j j k A k +j j j j =k A + I k k B + I k so that AI is a Banach algebra with unit.1. The spectrum (A) of A 2 A is the complement of (A) in C in other words. since it follows from (2. The proof uses two lemmas. We rst show that the sum is a Cauchy sequence.15) in A. one sees from (2. De nition 2. When A has no unit. When A has no unit. the resolvent and the spectrum are de ned through the embedding of A in AI = A C . Since by (2. Ak+1 ) = I . k . z I has a (twosided) inverse in A.2. z I) always exists when jz j > k A k. Now compute
n X k=0
Ak (I .2. As we shall see at the end of section 2. Proposition 2.
m X
k=0
Ak k=k
n X
k=m+1
Ak k
n X
k=m+1
k Ak k
n X
k=m+1
k A kk :
For n m ! goes to 0 by the theory of the geometric series. De nition 2. Using (2.1 For every Banach algebra without unit there exists a unital Banach algebra AI and an isometric (hence injective) morphism A ! AI. the spectrum (A) of A 2 A always contains zero.1.23) that
the norm of A + 0I in AI. Since A is complete. contained in the set fz 2 C j jz j k A kg 2. For A = B(H).2. Furthermore. In other words.22) that A never has an inverse in AI. we have shown the following. An+1 :
Hence
k I. . such that AI=A ' C .2. not empty. A) =
n X k=0
n X k=0
(Ak . de ne a norm on AI by
k A + I k:=k A k +j j:
(2. A) .22) and (2. compact 3. We assume that A is unital.4 When k A k < 1 the sum Pn k=0 A converges to (I .
Ak (I .23) the norm of A 2 A in A coincides with
k
n X k=0
Ak . Theorem 2. the Cauchy P1 this k sequence n k=0 A converges for n ! 1. When A is the algebra of n n matrices. z I has no (twosided) inverse in A.23)
In particular. A) k=k An+1 k k A kn+1
. The resolvent (A) of A 2 A is the set of all z 2 C for which A .3. etc. Indeed. the unitization AI with the given properties is not unique. the spectrum of A is just the set of eigenvalues.1 Lemma 2. 1 Hence (A . the number 1 in C is identi ed with I.2.4.2 reproduces the usual notion of the spectrum of an operator on a Hilbert space. k I k= 1. as well as 2.3 The spectrum (A) of any element A of a Banach algebra is 1.16
2 ELEMENTARY THEORY OF C ALGEBRAS
where we have written A + I for (A ).2 Let A be a unital Banach algebra. (A) is the set of all z 2 C for which A .
Given A 2 G(A). (A . Lemma 2.5. A)k.3.z X 0
k (z0 . A). the limit n ! 1 of this power series is 1 z . A k=0 z0 . z0 .8) this implies
(2. z j k (z0 . we see that z 2 (A). Since k f (z + ) .1B ). A) = I:
n X k=0
By a similar argument. It follows that all C 2 A for which k A .5 The set G(A) := fA 2 Aj A. C k < lie in G(A).1 . the power series
converges for n ! 1 by Lemma 2.25)
k A.1 ! 1 z 1 = g(z ): 1 X 0 0. de ne g : (A) ! A by g(z ) := (z .5. by continuity of multiplication in a Banach algebra. because
z0 . as k A . choose z 2 C such that jz . with A ! A .2.26)
Hence A + B = A(I + A. take a B 2 A for which k B k < k A.4. For xed z0 2 (A). From the proof of Lemma 2. z ). To resume the proof of Theorem 2. z )k (z0 .1 B k k A.2.26) and Lemma 2. z0 j < k (A . A) n X
Ak = I:
so that. it follows from the topological de nition of a continuous function that f . Moreover.z = z. we see that f is continuous (take = in the de nition of continuity). To prove that (A) is compact.1. f (z ) k= . z0 .2.1B ) has an inverse. This set being open.1 .2. so that f . which exists because k A=z k < 1 when jz j > k A k. as k A k < 1 by assumption. By (2.1 k. A = z0 . z ) k= jz . A
1
n z .1 existsg of invertible elements in A is open in A.2.4.1(I .z . z0 ). it remains to be shown that it is closed. Because G(A) is open in A by Lemma 2. Finally. A. A). A). given A 2 A we now de ne a function f : C ! A by f (z ) := z .24)
The second claim of the lemma follows because (A .1 = . A A
Hence
g(z ) =
1 X
k=0
(z0 .1 k. which exists by (2. one nally has
nlim !1 k=0
Ak = (I .2. z0 . for k A. its complement (A) is
closed. A 1 .1 k= jz0 . A k=0 z0 .1 .1 k < 1: By Lemma 2.1 :
(2.
nlim !1(I .27)
. But f .4.z . namely (I + A. z0 j. Thus
nlim !1 n X k=0
17
Ak (I . A has an inverse.1 .2.1 k k B k < 1:
(2. z0 and C ! A . z )(z0 .2.1 k.1(G(A)) is open in A.1 A.2 Banach algebra basics which ! 0 for n ! 1.1. z . A).1 (G(A)) is the set of all z 2 C where z .1
(2.1 (G(A)) = (A). A=z ).
and hence g . In particular. this isomorphism is isometric. Lemma 2.28) z!1 lim g (z ) = 0: (2.1. we have seen that for any z 2 (A) one may nd a z0 2 (A) such that the power series (2. so (2.31). for jz j >k A k the function g in the proof of Lemma 2. so that lim sup k A k1=n r(A): (2.1 one immediately infers
(2. so that g = 0 by Corollary 2. (2. and lim k g(z ) k= 0: (2.3.2.1 = I.27) converges. Since (A) 6= .18
2 ELEMENTARY THEORY OF C ALGEBRAS
Let 2 A be a functional on A since is bounded.2. g is bounded. zAI = 0 by assumption. This is absurd. At this point the proof relies on the theory of analytic functions with values in a Banach space.3.33) cannot normconverge uniformly in z unless k An k =jz jn < 1 for large enough n.2.4.35)
. The function g. A=z ).
is a normconvergent power series in z .1 k (I .4.34) To derive a second inequality we use the following polynomial spectral mapping property.29)
Now suppose that (A) = . this sharpens what we know from Lemma 2.32)
r(A) k A k :
Proposition 2. uniformly in z . (2.27) implies that the function g : z ! (g(z )) is given by a convergent power series.7. since limz!1 k A=z k= 0 by 2.6 If every element (except 0) of a unital Banach algebra A is invertible.2.7 For each A in a unital Banach algebra one has n 1=n r(A) = nlim !1 k A k :
1 g(z ) = z
1 A k X z :
k=0
By Lemma 2. and (2. Then
n!1
p( (A)) = (p(A)):
(2.33) is normconvergent for jz j > r(A). is then de ned on where it is analytic and vanishes at in nity. so that by Liouville's theorem it must be constant. If jz j > r(A) then z 2 (A).28) implies that
z!1
C.33)
On the other hand.2. Hence limz!1 (I. which will be essential in the characterization of commutative C algebras. which says that. then A ' C as Banach algebras.A=z ).31) (2. so that (A) 6= C hence (A) 6= . This is true for all z for which jz j > r(A). By (2.5 has the normconvergent power series expansion (2. zA I is not invertible.1. and the map A ! zA is the desired algebra isomorphism.1 k and observe that limz!1 I. de ne p( (A)) as fp(z)j z 2 (A)g. accordingly.2.A=z = I. Corollary 2. For z 6= 0 we write k g(z ) k= jz j. The same theory says that (2.29) this constant is zero.2.27) converges for jz j > r(A). Since k A k=k z I k= jz j. so that (A) = C .30) (2.8 For a polynomial p on C .1. De ne the spectral radius r(A) of A 2 A by
r(A) := supfjz j z 2 (A)g:
From Theorem 2. Comparing with (2. The fact that the spectrum is never empty leads to the following Gel'fandMazur theorem. Hence A . for each A 6= 0 there is a zA 2 C for which A .
39) Here : A ! A=I is the canonical projection. i ( ) is not invertible.38). exists which contains I. choose z
19
2 C and compare the factorizations
= c
I
p(z ) .37)
and the multiplication
(A) (B ) := (AB ): (2. with unit (I).2 Banach algebra basics To prove this equality. p(A) .7. A maximal ideal is an ideal I 6= A for which no ideal I
In particular. Since n 2 (An ) by Lemma 2.2.2.8.2. Hence k An k1=n j j = r(A). Conversely. which implies that all A . An ideal I that contains an invertible element A must coincide with A.
n Y
= c
i=1 n Y i=1
(z .34) yields lim sup k A k1=n r(A) k An k1=n :
n!1
Hence the limit must exist. so that i ( ) 2 (A) for at least one i. so that for some i one must have i ( ) = z for this particular z . so that 2 (p(A)).36)
Here the coe cients c and i ( ) are determined by p and . so that A . ~ 6= A. To conclude the proof of Proposition 2. since A. for given A 2 A. We omit the standard proof that A=I is a Banach space in the norm (2. = 0. for each > 0 there exists a J 2 I such that k (A) k + k A + J k : (2. i ( )I): (2. by de nition of the in mum.2. Combining this with (2. As far as the Banach algebra structure is concerned. Hence 2 (p(A)) implies that at least one of the A . This shows that p( (A)) (p(A)). rst note that (2.8). and (2.e. one has j n j k An k by (2.39) is well de ned: when J1 J2 2 I one has (A + J1 ) (B + J2 ) = (AB + AJ2 + J1 B + J1 J2 ) = (AB ) = (A) (B ) since AJ2 + J1 B + J1 J2 2 I by de nition of an ideal.2. Proposition 2. When 2 (p(A)) then p(A) . A rightideal of A is a closed linear subspace I for which A 2 I implies AB 2 I for all B 2 A. implying that p(A) .1 A = I must lie in I.9 An ideal in a Banach algebra A is a closed linear subspace I A such that A 2 I implies AB 2 I and BA 2 I for all B 2 A. To prove (2. and (J ) = 0 for all J 2 I.35) follows. 2 p( (A)). but a given proper ideal I 6= A does not contain I.31). Hence i ( ) 2 (A). i. I is invertible.10 If I is an ideal in a Banach algebra A then the quotient A=I is a Banach algebra in the norm k (A) k:= Jinf kA+J k (2. I is not invertible. If A is unital then A=I is unital. an ideal is itself a Banach algebra.38) 2I
nlim !1 k A k
1=n
n 1=n = inf n k A k = r(A):
(2. This proves the inclusion (p(A)) p( (A)). i ( )I is not invertible. Hence p( i ( )) . when 2 p( (A)) then = p(z ) for some z 2 (A). i ( )I must be invertible. observe that. A leftideal of A is a closed linear subspace I for which A 2 I implies BA 2 I for all B 2 A.. we note that since (A) is closed there is an 2 (A) for which j j = r(A).40)
. and one cannot add I to I without ruining the property that it is a proper ideal. I ~ 6= I. i ( )) (A . so that all B = B I must lie in I. and
De nition 2. This shows the need for considering Banach algebras with and without unit it is usually harmless to add a unit to a Banach algebra A.
AB = BA for all A B 2 A). with norm k ! k= 1
hence for all A 2 A. (A) consists of all nonzero homomorphisms from A to C .4 that A .38). since ! is continuous by 2. since !(IA) = !(I)!(A) = !(A).
k (A) k=k (A + J ) k k A + J k : For A B 2 A choose > 0 and J1 J2 2 I such that (2.40) holds for A B .45) (2.46) follows. and !1 = !2
i I!1 = I!2 . from (2.43).44) (2. Each ! 2 (A) satis es !(I) = 1 2.2.3.38) that
When A has a unit. for any J 2 I it is clear from (2.3 Commutative Banach algebras
We now assume that the Banach algebra A is commutative (that is. Furthermore.42)
2. each ! 2 (A) is continuous.3. The kernel of every linear map ! : V ! C on a vector space V has codimension one (that is. and there is an A for which !(A) 6= 0 because ! is not identically zero.41)
(2.43)
(2. z 6= 0. so that ker(!) is a maximal ideal. Proposition 2. On the other hand. For the second. each maximal ideal is the kernel of some ! 2 (A). For !i 2 (A) this implies !1 = !2 because of (2.46)
j!(A)j k A k
The rst claim is obvious.1 The structure space (A) of a commutative Banach algebra A is the set of all nonzero linear maps ! : A ! C for which
!(AB ) = !(A)!(B ) for all A B 2 A.3. On the other hand.
(2. By (2. Hence k (I) k= 1. z is invertible when jz j > k A k.
(2. 1. the norm in A=I could not be given by (2. so that !(A . z ) = !(A) . ker(!) is an ideal since ! satis es (2.39) that (I) is a unit in A=I. it is obvious from (2.44).8) with B = I one derives k (I) k 1. we know from Lemma 2.20
2 ELEMENTARY THEORY OF C ALGEBRAS
For if such a J would not exist. and (2.2 Let A have a unit I.41) with A = I one has k (I) k k I k= 1.
The kernel of each ! 2 (A) is closed.2. in that the kernel ker(!) of each ! 2 (A) is a maximal ideal I! .
. since ! is a homomorphism.3. There is a bijective correspondence between (A) and the set of all maximal ideals in A. In other words.3 Let A be a unital commutative Banach algebra. We say that such an ! is multiplicative.
Theorem 2. Again on any vector space. and estimate k (A) (B ) k = k (A + J1 ) (B + J2 ) k=k ((A + J1 )(B + J2 )) k k (A + J1 )(B + J2 ) k k A + J1 k k B + J2 k (k (A) k + )(k (B ) k + ): Letting ! 0 yields k (A) (B ) k k (A) k k (B ) k. when ker(!1 ) = ker(!2 ) then !1 is a multiple of !2. dim(V = ker(!)) = 1). De nition 2. Hence j!(A)j 6= jz j for jz j > A k.2.
when !n ! ! in the w topology we obtain j!(AB ) . !(A)!(B )j = 0. This map is clearly linear. Taking A = 0 we see I IB . hence I = BA + J for suitable A 2 A J 2 I. so that IB 6= I. Hence ! 2 (A).47) ^ as a function on (A). Since I 6= A. the unit ball in A is w compact. Since the w topology is Hausdor (as is immediate from its de nition). I ker(!) since I = ker( ) but if B 2 = I we saw that !(B ) 6= 0.4 The structure space (A) of a unital commutative Banach algebra A is compact and Hausdor in the Gel'fand topology. so that there is a homomorphism : A=I ! C . By 2. so that actually I = ker(!).2. !(B )j: All in all. the absolute value of the righthand side is bounded by
1). IB is even an ideal. ! is multiplicative it is nonzero because !(B ) 6= 0. ^.2. consisting of all functionals with norm
^(!) := !(A): A (2. !n (AB ) + !n (A)!n (B ) .
!(A)!(B ) = ( (A)) ( (B )) = ( (A) (B )) = ( (AB )) = !(AB )
because of (2.2 we have (A) A . Proposition 2. By elementary functional analysis. !(A))!n (B ) + !(A)(!n (B ) .39) and the fact that is a homomorphism. When !n 2 (A) for all n. !(A)!(B ) = (!n (A) . one has
In the second term we write
j!(AB ) . hence (A) = (B ). I 2 IB . !(A)!(B )j:
!n (A)!n (B ) . where We embed A in A by A ! A
k B k j!n (A) . the w When ! 2 (A). Finally. Apply the canonical projection : A ! A=I to this equation. Also. By the BanachAlaoglu theorem. !(B )):
By (2. !n(AB )j + j!n (A)!n (B ) .6 this yields A=I ' C . since and are. Being a closed subset of this unit ball. !(A)!(B )j j!(AB ) . (A) is w compact. Hence IB = A. the claim follows.3. From (2. Hence (A) is w closed. A 2 A.2. this de nes A ^. !(A)!(B )j = j!(AB ) . there is a nonzero B 2 A which is not in I.39) and (J ) = 0. so that the limit ! 2 (A). also called w topology.3. or because !(I) = 1. Recall that the weak topology. Since B was arbitrary (though nonzero). giving
(I) = I = (BA) = (B ) (A) because of (2. This implies that topology on A is the weakest topology for which all A
. By Corollary 2.45) we have (A) 2 A1 (the unit ball in A .3 Commutative Banach algebras
21
We now show that every maximal ideal I of A is the kernel of some ! 2 (A). as I is maximal. Form
IB := fBA + J j A 2 A J 2 Ig: This is clearly a leftideal since A is commutative. In particular. !(A)j+ k A k j!n (B ) .46) and the triangle inequality. are continuous. this shows that every nonzero element of A=I is invertible. Now de ne a map ! : A ! C by !(A) := ( (A)). The Gel'fand topology on (A) is the relative w topology. Taking A = I and J = 0 we see that B 2 IB .1 in A=I. The convergence !n ! ! in the w topology by de nition means that !n (A) ! !(A) for all A 2 A. on the dual B of a Banach space B is de ned by the convergence !n ! ! i !n (v) ! !(v) for all v 2 B. Therefore.
Convergence in the supnorm is the same as uniform convergence. What's more. Eq. ^ on (A) in other words.1 ) = 1. then !(A)!(A. so the example illustrates what happens to the structure theory in the absence of a unit.49).1 (O) = f! 2 (A)j !(A) 2 Og A (2. let us note in general that each ! 2 (A) has a suitable extension ! ~ to AI.52) then follows from (2. The Gel'fand transform is a homomorphism from A to C ( (A)).48) where A 2 A and O is an open set in C .5. When !1 6= !2 there is an A 2 A ^(!1 ) 6= A ^(!2 ). We now look at an example. it is easily veri ed that C (X ) is even a commutative Banach algebra under pointwise addition and multiplication. we have showed that A 2 G(A) is equivalent to !(A) 6= 0 for all ! 2 (A).22
2 ELEMENTARY THEORY OF C ALGEBRAS
^ are continuous.43). secondly it concerns a commutative Banach algebra which is not a C algebra.2. The image of A under the Gel'fand transform separates points in (A). (2. and using (2.3. we regard the space C (X ) of all continuous functions on X as a Banach space in the supnorm de ned by k f k1:= sup jf (x)j: (2. In this connection. which is included for three reasons: rstly it provides a concrete illustration of the Gel'fand transform. 3. A is invertibe). the Gel'fand topology on (A) is the weakest topology for which all functions A In particular.5 Let A be a unital commutative Banach algebra. ^ de ned by (2. (2. relying on the axiom of choice in the guise of Hausdor 's maximality priciple).51)
4. Thus the resolvent is !(A) = fz 2 C j z 6= !(A) 8! 2 (A)g: (2. for which !1 (A) 6= !2 (A). 1. (2. so that !(A) 6= 0 for all ! 2 (A). This proves 2. The spectrum of A 2 A is the set of values of A ^) = fA ^(!)j ! 2 (A)g: (A) = (A (2. One checks that the spectrum of f 2 C (X ) is simply the set of values of f . namely !(A + I) := !(A) + : (2. The Gel'fand transform is a contraction.3 there is a ! 2 (A) for which !(A) = 0. z 2 G(A) i !(A) 6= z for all ! 2 (A). that is. the map A ! A transform. so that it is contained in a maximal ideal I (this conclusion is actually nontrivial.47) and (2.51). we obtain (2. All in all.3.31). and (2.30). 2. a basis for this topology is formed by all open sets of the form ^. so that A If A 2 G(A) (i.47) is called the Gel'fand Seen as a map from A to C ( (A)). and thirdly the Banach algebra in question has no unit.52)
The rst property immediately follows from (2. that is. ^ k1 k A k : kA
(2.50) Hence the function 1X which is 1 for every x is the unit I.47).3.e. Hence by Theorem 2..49)
x2 X
A basic fact of topology and analysis is that C (X ) is complete in this norm. When A 2 = G(A) the ideal IA := fAB j B 2 Ag does not contain I.47).54)
. Theorem 2. Hence A .53) Taking the complement. We regard C ( (A)) as a commutative Banach algebra in the manner explained. For any compact Hausdor space X . ( f + g)(x) := f (x) + g(x) (fg)(x) := f (x)g(x): (2.
that is. Note that the Gel'fand transform is strictly a contraction. and norm
k f k1 := dx jf (x)j:
The associative product de ning the Banach algebra structure is convolution. With this notation.2.
R
Z
(2. y)g(y)j
R
Z
Z
Z
=
Z
R
which is (2. Standard Banach space theory says that the dual of L1 (R) is L1 (R).3.58) 1 for some p 2 C .22) and the de nition (2. This is an important function space.6 Let X be a Hausdor space X which is locally compact (in that each point has a compact neighbourhood).
. since from (2. we see from (2.43). Moreover. which does not lie in L1 (R). this should rst be de ned on the dense subspace Cc (R).58) is simply called p. Hence for each ! 2 (L1 (R)) there is a function ! ^ 2 L1(R) such that
R
dy jg(y)j dx jf (x)j =k f k1 k g k1
R
Z
R
dy jg(y)j dx jf (x ..e. whose de nition may be generalized as follows.3 Commutative Banach algebras
23
The point is that ! ~ remains multiplicative on AI. we estimate
R
dy f (x . Finally. and subsequently be extended by continuity to L1(R). using Fubini's theorem on product integrals.56) one sees that the unit should be Dirac's deltafunction (i. with the usual linear structure.43) then implies that ! ^ (x + y ) = ! ^ (x)^ ! (y) for almost all
R
dx f (x)^ ! (x):
(2.54) that every multiplicative functional ! 2 (L1 (R)) is continuous. the RiemannLebesgue lemma states that f 2 L1 (R) implies f is the space of continuous functions on R that go to zero when jxj ! 1.55) (2. i.59)
Hence the Gel'fand transform is nothing but the Fourier transform (more generally. The wellknown fact that the Fourier transform maps the convolution product (2. y)j
R
Z
!(f ) = x y 2 R.3 that the spectrum (f ) of f in L1 (R) is just the set of values of its Fourier transform.56)
f g(x) :=
Z
Strictly speaking. The space C0 (X ) consists of all continuous functions on X which vanish at in nity in the sense that for each > 0 there is a compact subset K X such that jf (x)j < for all x outside K . and since ! ^ is bounded (being in L (R)) it must be that p 2 R.46). using the inequality below.56) into the pointwise product is then a restatement of Theorem 2. This implies
Z
The multiplicativity condition (2.57)
the Gel'fand transform (2.1.57) and (2. There is no unit in L1 (R).3. De nition 2. we see from 2. as can be seen from (2. Even if one does not actually extend A to AI. Indeed. the existence of ! ~ shows that ! satis es (2. It is clear that di erent p's yield di erent functionals. This extension is clearly unique.3.52). The functional ! corresponding to (2. and therefore certainly for the restriction ! of ! ~ to A. many of the integral transforms of classical analysis may be seen as special cases of the Gel'fand transform).47) reads
! ^ (x) = exp(ipx) (2. y)g(y):
k f g k1 = dx j dy f (x .58) that f^(p) =
Z
R
dx f (x)eipx :
(2.e. We know from the discussion following (2. since this property (which was proved for the unital case) holds for ! ~ . Consider A = L1 (R). so that (L1 (R)) may be identi ed with R.. the measure on R which assigns 1 to x = 0 and 0 to all other x).5.8). there is no equality in the bound ^ 2 C0 (R). which (2.5.
24
2 ELEMENTARY THEORY OF C ALGEBRAS
So when X is compact one trivially has C0 (X ) = C (X ). When X is not compact, the supnorm (2.49) can still be de ned, and just as for C (X ) one easily checks that C0 (X ) is a Banach algebra in this norm. We see that in the example A = L1 (R) the Gel'fand transform takes values in C0 ( (A)). This may be generalized to arbitrary commutative nonunital Banach algebras. The nonunital version of Theorem 2.3.5 is Theorem 2.3.7 Let A be a nonunital commutative Banach algebra. 1. The structure space (A) is locally compact and Hausdor in the Gel'fand topology. 2. The space (AI) is the onepoint compacti cation of (A). 3. The Gel'fand transform is a homomorphism from A to C0 ( (A)). ^ on (A), with zero added (if 0 is not already 4. The spectrum of A 2 A is the set of values of A contained in this set). 5. The claims 2 and 4 in Theorem 2.3.5 hold. ~ of a noncompact topological space X is the Recall that the onepoint compacti cation X set X 1, whose open sets are the open sets in X plus those subsets of X 1 whose complement ~ is a compact Hausdor space, the removal of some point is compact in X . If, on the other hand, X ~ nf1g in the relative topology (i.e., the open `1' yields a locally compact Hausdor space X = X ~ sets in X are the open sets in X minus the point 1), whose onepoint compacti cation is, in turn, ~. X To prove 2.3.7 we add a unit to A, and note that (AI) = (A) 1 (2.60) (2.61) where each ! 2 (A) is seen as a functional ! ~ on AI by (2.54), and the functional 1 is de ned by
1(A + I) := :
There can be no other elements ' of (AI), because the restriction of ' has a unique multiplicative extension (2.54) to AI, unless it identically vanishes on (A). In the latter case (2.61) is clearly the only multiplicative possibility. By Proposition 2.3.4 the space (AI) is compact and Hausdor by (2.61) one has (A) = (AI)nf1g (2.62) as a set. In view of the paragraph following 2.3.7, in order to prove 2.3.7.1 and 2, we need to show that the Gel'fand topology of (AI) restricted to (A) coincides with the Gel'fand topology of (A) itself. Firstly, it is clear from (2.48) that any open set in (A) (in its own Gel'fand topology) is the restriction of some open set in (AI), because A AI. Secondly, for any A 2 A, 2 C , and open set O C , from (2.54) we evidently have
f' 2 (AI)j '(A + I) 2 Ognf1g = f! 2 (A)j !(A) 2 O ; g: (When 1 does not lie in the set f: : :g on the lefthand side, one should here omit the \nf1g".)
With (2.48), this shows that the restriction of any open set in (AI) to (A) is always open in the Gel'fand topology of (A). This establishes 2.3.7.1 and 2. It follows from (2.3.5) and (2.61) that ^(1) = 0 (2.63) A ^ leads to 2.3.7.3. for all A 2 A, which by continuity of A The comment preceding Theorem 2.2.3 implies 2.3.7.4. The nal claim follows from the fact that it holds for AI.
2.4 Commutative C algebras
25
The Banach algebra C (X ) considered in the previous section is more than a Banach algebra. Recall De nition 2.1.9. The map f ! f , where f (x) := f (x) (2.64) evidently de nes an involution on C (X ), in which C (X ) is a commutative C algebra with unit. The main goal of this section is to prove the converse statement cf. De nition 2.1.12 Theorem 2.4.1 Let A be a commutative C algebra with unit. Then there is a compact Hausdor space X such that A is (isometrically) isomorphic to C (X ). This space is unique up to homeomorphism. The isomorphism in question is the Gel'fand transform, so that X = (A), equipped with the Gel'fand topology, and the isomorphism ' : A ! C (X ) is given by ^ '(A) := A: (2.65) We have already seen in 2.3.5.1 that this transform is a homomorphism, so that (2.18) is satis ed. To show that (2.19) holds as well, it su ces to show that a selfadjoint element of A is mapped into a realvalued function, because of (2.13), (2.64), and the fact that the Gel'fand transform is complexlinear. We pick A 2 AR and ! 2 (A), and suppose that !(A) = + i , where 2 R. By (2.44) one has !(B ) = i , where B := A ; I is selfadjoint. Hence for t 2 R one computes j!(B + itI)j2 = 2 + 2t + t2 : (2.66) On the other hand, using (2.46) and (2.16) we estimate j!(B + itI)j2 k B + itI k2 =k (B + itI) (B + itI) k=k B 2 + t2 k k B k2 +t2 : Using (2.66) then yields 2 + t k B k2 for all t 2 R. For > 0 this is impossible. For < 0 we repeat the argument with B ! ;B , nding the same absurdity. Hence = 0, so that !(A) ^ is realvalued, and (2.19) follows as is real when A = A . Consequently, by (2.47) the function A announced. We now prove that the Gel'fand transform, and therefore the morphism ' in (2.65), is isometric. When A = A , the axiom (2.16) reads k A2 k=k A k2 . This implies that k A2m k=k A k2m for all m 2 N . Taking the limit in (2.32) along the subsequence n = 2m then yields r(A) =k A k : (2.67) In view of (2.30) and (2.51), this implies ^ k1 =k A k : kA (2.68) For general A 2 A we note that A A is selfadjoint, so that we may use the previous result and (2.16) to compute ^ k1 =k A ^ k2 d ^A k A k2=k A A k=k A A k1 =k A 1: c =A ^ , which we just proved, and in the fourth we exploited the In the third equality we used A fact that C (X ) is a C algebra, so that (2.16) is satis ed in it. Hence (2.68) holds for all A 2 A. It follows that ' in (2.65) is injective, because if '(A) = 0 for some A 6= 0, then ' would fail to be an isometry. (A commutative Banach algebra for which the Gel`fand transform is injective is called semisimple. Thus commutative C algebraa are semisimple.) We nally prove that the morphism ' is surjective. We know from (2.68) that the image ^ is closed in C ( (A)), because A is closed (being a C algebra, hence a Banach space). '(A) = A In addition, we know from 2.3.5.2 that '(A) separates points on (A). Thirdly, since the Gel`fand transform was just shown to preserve the adjoint, '(A) is closed under complex conjugation by ^ = 1X by (2.44) and (2.47), the image '(A) contains 1X . The surjectivity (2.64). Finally, since I of ' now follows from the following StoneWeierstrass theorem, which we state without proof.
2.4 Commutative C algebras
26
2 ELEMENTARY THEORY OF C ALGEBRAS
algebra as explained above. A C subalgebra of C (X ) which separates points on X and contains 1X coincides with C (X ). Being injective and surjective, the morphism ' is bijective, and is therefore an isomorphism. The uniqueness of X is the a consequence of the following result. Proposition 2.4.3 Let X be a compact Hausdor space, and regard C (X ) as a commutative C algebra as explained above. Then (C (X )) (equipped with the Gel`fand topology) is homeomorphic to X . Each x 2 X de nes a linear map !x : C (X ) ! C by !x (f ) := f (x), which is clearly multiplicative and nonzero. Hence x ! !x de nes a map E (for Evaluation) from X to (C (X )), given by E (x) : f ! f (x): (2.69) Since a compact Hausdor space is normal, Urysohn's lemma says that C (X ) separates points on X (i.e., for all x 6= y there is an f 2 C (X ) for which f (x) 6= f (y)). This shows that E is injective. We now use the compactness of X and Theorem 2.3.3 to prove that E is surjective. The maximal ideal Ix := I!x in C (X ) which corresponds to !x 2 (C (X )) is obviously
Lemma 2.4.2 Let X be a compact Hausdor space, and regard C (X ) as a commutative C 
Ix = ff 2 C (X )j f (x) = 0g: (2.70) Therefore, when E is not surjective there exists a maximal ideal I C (X ) which for each x 2 X contains at a function fx for which fx (x) 6= 0 (if not, I would contain an ideal Ix which thereby would not be maximal). For each x, the set Ox where fx is nonzero is open, because f is continuous. This gives a covering fOx gx2X of X . By compactness, there exists a nite P N subcovering fOxi gi=1 ::: N . Then form the function g := i=1 jfxi j2 . This function is strictly positive by construction, so that it is invertible (note that f 2 C (X ) is invertible i f (x) 6= 0 for all x 2 X , in which case f ;1 (x) = 1=f (x)). But I is an ideal, so that, with all fxi 2 I (since all fx 2 I) also g 2 I. But an ideal containing an invertible element must coincide with A (see the comment after 2.2.9), contradicting the assumption that I is a maximal ideal. Hence E is surjective since we already found it is injective, E must be a bijection. It remains to be shown that E is a homeomorphism. Let Xo denote X with its originally given topology, and ^ E = f by (2.69) and (2.47), and the write XG for X with the topology induced by E ;1 . Since f ^ are continuous, Gel'fand topology on (C (X )) is the weakest topology for which all functions f we infer that XG is weaker than Xo (since f , lying in C (Xo ), is continuous). Here a topology T1 is called weaker than a topology T2 on the same set if any open set of T1 contains an open set of T2 . This includes the possibility T1 = T2 . Without proof we now state a result from topology. Lemma 2.4.4 Let a set X be Hausdor in some topology T1 and compact in a topology T2 . If T1 is weaker than T2 then T1 = T2 . Since Xo and XG are both compact and Hausdor (the former by assumption, and the latter by Proposition 2.3.4), we conclude from this lemma that X0 = XG in other words, E is a homeomorphism. This concludes the proof of 2.4.3. Proposition 2.4.3 shows that X as a topological space may be extracted from the Banachalgebraic structure of C (X ), up to homeomorphism. Hence if C (X ) ' C (Y ) as a C algebra, where Y is a second compact Hausdor space, then X ' Y as topological spaces. Given the isomorphism A ' C (X ) constructed above, a second isomorphism A ' C (Y ) is therefore only possible if X ' Y . This proves the nal claim of Theorem 2.4.1. The condition that a compact topological space be Hausdor is su cient, but not necessary for the completeness of C (X ) in the supnorm. However, when X is not Hausdor yet C (X ) is complete, the map E may fail to be injective since in that case C (X ) may fail to separate points on X .
2.4 Commutative C algebras
27
On the other hand, suppose X is locally compact but not compact, and consider A = Cb (X ) this is the space of all continuous bounded functions on X . Equipped with the operations (2.49), (2.50), and (2.64) this is a commutative C algebra. The map E : X ! (Cb (X )) is now injective, but fails to be surjective (this is suggested by the invalidity of the proof we gave for C (X )). Indeed, it can be shown that (Cb (X )) is homeomorphic to the CehStone compacti cation of X . Let us now consider what happens to Theorem 2.4.1 when A has no unit. Following the strategy we used in proving Theorem 2.3.7, we would like to add a unit to A. As in the case of a general Banach algebra (cf. section 2.2), we form AI by (2.21), de ne multiplication by (2.22), and use the natural involution (2.71) (A + I) := A + I: However, the straightforward norm (2.23) cannot be used, since it is not a C norm in that axiom (2.16) is not satis ed. Recall De nition 2.1.4. Lemma 2.4.5 Let A be a C algebra. 1. The map : A ! B(A) given by (A)B := AB (2.72) establishes an isomorphism between A and (A) B(A). 2. When A has no unit, de ne a norm on AI by k A + I k:=k (A) + I k (2.73) where the norm on the righthand side is the operator norm (2.2) in B(A), and I on the righthand side is the unit operator in B(A). With the operations (2.22) and (2.71), the norm (2.73) turns AI into a C algebra with unit. By (2.15) we have k (A)B k=k AB k k A kk B k for all B , so that k (A) k k A k by (2.2). On the other hand, using (2.16) and (2.17) we can write
(2.74) Being isometric, the map must be injective it is clearly a homomorphism, so that we have proved 2.4.5.1. It is clear from (2.22) and (2.71) that the map A + I ! (A) + I (where the symbol I on the lefthand side is de ned below (2.22), and the I on the righthand side is the unit in B(A)) is a morphism. Hence the norm (2.73) satis es (2.15), because (2.8) is satis ed in B(A). Moreover, in order to prove that the norm (2.73) satis es (2.16), by Lemma 2.1.11 it su ces to prove that k (A) + I k2 k ( (A) + I) ( (A) + I) k (2.75) for all A 2 A and 2 C . To do so, we use a trick similar to the one involving (2.40), but with inf replaced by sup. Namely, in view of (2.2), for given A 2 B(B) and > 0 there exists a v 2 V , with k v k= 1, such that k A k2 ; k Av k2 . Applying this with B ! A and A ! (A) + I, we infer that for every > 0 there exists a B 2 A with norm 1 such that k (A) + I k2 ; k ( (A) + I)B k2 =k AB + B k2 =k (AB + B ) (AB + B ) k : Here we used (2.16) in A. Using (2.72), the righthand side may be rearranged as k (B ) (A + I) (A + I)B k k (B ) k k ( (A) + I) ( (A) + I) k k B k : Since k (B ) k=k B k=k B k= 1 by (2.74) and (2.17), and k B k= 1 also in the last term, the inequality (2.75) follows by letting ! 0. Hence the C algebraic version of Theorem 2.2.1 is
k A k=k AA k = k A k=k (A) kA Ak k in the last step we used (2.4) and k (A = k A k) k= 1. Hence k (A) k=k A k :
(A) k
coincides with C0 (X ). (A ). and A.1 Let A = A be a selfadjoint element of a unital C algebra.
2. Under this isomorphism the Gel'fand transform A function id (A) : t ! t..1 I) generated by A. The spectrum A (A) of A in A coincides with the spectrum C (A I)(A) of A in C (A I) (so that we may unambiguously speak of the spectrum (A)).25).3. 3. k f k.23) and (2.4. hence k 1X .5. 1 pointwise. and is such that for each x 2 X there is an f 2 A such that f (x) 6= 0. 12 ff 12 ff < 1 pointwise. This now reads Lemma 2.7 Let X be a locally compact Hausdor space.2.1 . One has (A. A C subalgebra A of C0 (X ) which separates points on X . where Ai is A or A . we see that. 2. in general. unique. Following the terminology for operators on a Hilbert space.73) de ne a norm on AI satisfying the claims of Proposition 2.7 may then be repeated in the C algebraic setting the only nontrivial point compared to the situation for Banach algebras is the generalization of Lemma 2. On the other hand.4. For each element A 2 A there is a smallest C subalgebra C (A I) of A which contains A and I. when A is selfadjoint.3.47) is an isomorphism. Let A 2 G(A) be normal in A. Hence C (A A. It is su cient for our purposes to restrict ourselves to this case. in terms of I = 1X we may therefore write 1 ff k : 1= f X (2.6 For every C algebra without unit there exists a unique unital C algebra AI and an isometric (hence injective) morphism A ! AI. (A ). so that 0 1X .4. A. A. and I. an element A 2 A is called normal when A A ] = 0. Theorem 2.1 we have C (A A.8 Let A be a commutative C algebra without unit. and consider the C algebra C (A A. A(x) 6= 0x for all x 2 X ).28
2 ELEMENTARY THEORY OF C ALGEBRAS
Proposition 2.4.1 I) is commutative it is the closure of the space of all polynomials in A. a commutative nonunital C algebra has a unitization. Recall (2.4. C (A I) is simply the closure of the space of all polynomials in A. Using Lemma 2. in view of the fact that both (2. The structure space (C (A I)) is homeomorphic with (A).e.76) I. At the end of the day we then nd Theorem 2.3.4.1 I) ' C (X ) for some compact ^ Hausdor space X . In particular.1. for any f 2 C (X ) that is nonzero throughout X we have 0 <k f k. and I. such that AI=A ' C . The crucial property of a normal operator is that C (A I) is commutative. 1.5 to Theorem 2.2. This space is unique up to homeomorphism. However. and regard C0 (X ) as a commutative C algebra as explained below De nition 2.1 . The spectrum (A) is a subset of R. Since A is invertible and the Gel'fand transform (2.1 ) = (A ). namely the closure of the linear span of I and all operators of the type A1 : : : An .1 .3 below. we conclude that the unital Banach algebra AI called for in that proposition is not.2. so that C (A I) is isomorphic ^ : (A) ! R is the identity to C ( (A)).
We return to the general case in which a C algebra A is not necessarily commutative (but assumed unital). A.64).1 .5.6. but analyze properties of A by studying certain commutative subalgebras.5 Spectrum and functional calculus
f
k f k2 1 k=0
k f k2 1
. This will lead to important results.1 . ff = k f k2 1 k1 < 1: Here f is given by (2. in particular.1 and I all commute with each other. The passage from Theorem 2. A . In any case. There is a locally compact Hausdor space X such that A is (isometrically) isomorphic to C0 (X ). A . having established the existence of the unitization of an arbitrary nonunital C algebra. A ^ is invertible in C (X ) (i. The uniqueness of AI follows from Corollary 2. By Theorem 2.
80)
Since the spectrum is determined by the algebraic structure alone.4.31)). and apply (2.48).3 yields an isomorphism C ( (A)) ! C (A I). ^ is a surjective map from X to (A). we have proved ^(!1 ) = A ^(!2 ) implies !1 = !2 . where z 2 C . The last claim follows by combining 2.3 the According to Theorem 2. Theorem 2. In conclusion. Now replace A by A . encounter an example of the opposite situation). since the linear span of all polynomials is dense in C (A I).3 does not imply that a given algebra can be normed only in one way so as to be completed into a C algebra (we will. we infer that A. In 2. and C (A A.78).
k A k= r(A A):
p
(2. z 2 G(A) the argument above applies. Finally. according to 2.2 For each selfadjoint element A 2 A and each f 2 C ( (A)) there is an operator f (A) 2 A. ^ is realvalued when A = A .2. We now prove injectivity. Since also !1 (I) = !2(I) = 1 by (2.78) with f = id (A) . the supnorm of id (A) is r(A).78)
In particular.3. A ^ is a homeomorphism.
First assume A = A . given a C algebra A there is no
other norm in which A is a C algebra). leading to the conclusion that the resolvent A (A) in A coincides with the resolvent C (A I)(A) in C (A I).3.3 the completeness of A is assumed from the outset. such that (f (A)) = f ( (A)) k f (A) k = k f k1 : (2. for general A 2 A we have.77) and (2.47). Note that Corollary 2.4. Gel'fand transHence A forming this result back to C (A A. for all n 2 N .5 Spectrum and functional calculus
29
^.3 the function A When !1 !2 2 X and !1 (A) = !2 (A).78). that for z 2 (A) the functional A ^.4. then. Using (2. z is normal.1. Hence by 2.79) Since A A is selfadjoint for any A.5.2) this implies that !1 = !2 on C (A I). we have
!1 (An ) = !1 (A)n = !2 (A)n = !2 (An ) by iterating (2. By De nition 2.5. By continuity (cf. in fact.2.3 The norm in a C algebra is unique (that is.1.77) follows from (2.5. 2.
Corollary 2. Looking at (2. When A is normal A . (2. Corollary 2.).3. which is precisely the map f ! f (A) of the continuous functional calculus. Since f ( (A)) is the set of values of f on (A).5.1 is a normconvergent limit of a sequence of polynomials in A ^ and A ^ .77) (2.1 with (2.1) yields (2.1 (z ) 2 (C (A I)) maps A to z (and hence An to z n .2 we then conclude that A (A) = C (A I)(A).1 is continuous. Hence A. the function A spectrum C (A I)(A) is real. (2. By de nition (cf.5.1 I) = C (A I). A ^. given the isomorphism C (A I) ' C (X ) of Theorem 2. one checks Since A 2 C (X ) by 2. which is the obvious expression when f is a polynomial (and in general is given via the uniform approximation of f by polynomials).3.44).5.1 (where X = (C (A I))). with A ! f (A).16). using (2.43) with B = A.5.1. that A ^ ^ is continuous.5. z . so that k A k= r(A) (A = A ): (2. we conclude by linearity that !1 = !2 on all polynomials in A. So if we assume that A . An immediate consequence of this theorem is the continuous functional calculus. To prove continuity of the inverse.2.51). etc.1 lies in C (A I). The fact that this isomorphism is isometric (see 2.5. so that by the previous result (A) is real.5.1.3 is then obvious.1 is a normconvergent limit of a sequence of polynomials in A and A .1.80) shows that the norm is determined by the algebraic structure as well.1 I). (2. the norm of f (A) in C (A I) coincides with its norm in A. one then sees that A
The nal claim in 2.
.
18) implies that ('(A .1 ).3.1.18). (A) R+ .82) Hence r('(A)) r(A). which we would like to generalize to arbitrary abstract C algebras. then '(A .
Lemma 2.5.e.3 and 2. to elements of the commutative C algebra C0 (X ) (where X is a locally compact Hausdor space).6.82)). In particular. and so is f ('(A)) in view of (2.30
2 ELEMENTARY THEORY OF C ALGEBRAS
Corollary 2. since for those f has its naive meaning. 1. A bounded operator A 2 B(H) on a Hilbert space H is called positive when ( A ) 0 for all 2 H this property is equivalent to A = A and (A) R+ .
The property is true for polynomials by (2. A+ \ . In quantum mechanics this means that the expectation value of the observable A is always positive.83)
for all f 2 C ( (A)) (here f (A) is de ned by the continuous functional calculus. Hence (A) ('(A)). z )). Classically. for (2.1 = '((A . This applies. in particular.81) follows from (2. so that ('(A)) For later use we note (A): (2. when A 2 A+ and t 2 R+ then tA 2 A+ 2. transform A
Proposition 2.1 An element A of a C algebra A is called positive when A = A and its spectrum is positive i. For general f the result then follows by continuity.
.1 exists in A. z ) is certainly invertible in B.2 The set A+ of all positive elements of a C algebra A is a convex cone that
is. a function f on some space X is positive simply when f (x) 0 for all x 2 X . so that (2.84)
It is immediate from Theorems 2. z ). one is interested in nding a number of equivalent characterizations of positivity.5 When ' : A ! B is a morphism and A = A then
f ('(A)) = '(f (A))
(2. Positivity is one of the most important features in a C algebra it will. play a central role in the proof of the Gel'fand Neumark theorem.5. Hence we have a notion of positivity for certain concrete C algebras.6.80).A+ = 0. so that (A .3 that A 2 AR is positive i its Gel'fand ^ is pointwise positive in C ( (A)).5.6 Positivity in C algebras
De nition 2. z ). We write A 0 or A 2 A+ .
(2.4 A morphism ' : A ! B between two C algebras satis es k '(A) k k A k
and is therefore automatically continuous. when A B 2 A+ then A + B 2 A+ 3. for example. where
A+ := fA 2 ARj (A)
R+
g:
(2.
2..81)
When z 2 (A).5. and clearly also applies to closed subalgebras of B(H).
Then (A. Apply the lemma to A = B B (noting that A = A ). as 0 B .A. This is important. The opposite inclusion follows from (2.2. ) BA. )3 0. . Inverting this argument. and given one puts AR = fA 2 AR j 0 Ag. this implies k cI. 2 A+ and A+ A. B kI. Then k cI . Since k f k1 r(A) =k A k (where we used (2.6.77) with f (t) = t3 that (A. (A + B ) k k (k A k .A.5.kI 0 (since k 0) we obtain .5 Every selfadjoint element A has a decomposition A = A+ .85) we have A + B kI. hence k A k k B k by (2.78) with A ! A . + The equivalence between these two structures is as follows: given A+ B if R := A one de nes A + + B . which generalizes the cases A = B(H) and A = C0 (X ). Hence. A. Use this with A ! A + B and c =k A k + k B k clearly c k A + B k by 2. when A = A one checks the validity of . i. tj c for all t 2 (A) and all c r(A). Hence ^j c by 2.85) by taking the Gel'fand transform of C (A I). B BA.6 Positivity in C algebras
31
The rst property follows from (tA) = t (A).3.86) then follows. We now come to the central result in the theory of positivity in C algebras. because a convex cone in a real vector space is equivalent to a linear partial ordering.3 and 2. when A 2 A+ and A 2 .4.(BA.79) and (2.5. As we have seen. A k c for some c k A k. = .87) = fB B j B 2 Ag: (2.5.B A B and (2.1. and satis es A2 = A. hence A = 0 by (2.1.77). = ..3.1. Hence .kAk I A kAk I (2. from which the lemma follows by (2. a partial ordering in which A B implies A + C B + C for all C and A B for all 2 R+ . : Since (A. Finally.79)). k B k k B k]. Hence A+ fA2 j A 2 ARg. = 0. Moreover. 0. We use this lemma to prove that fB B j B 2 Ag A+. is positive. ) BA. A k c for all c k A k by 2. where id (A) (t). This proves (2.30). f. then (A) R+ . For example.B ) k c where in the last step we used the previous paragraph for A and for B separately. which is a special case of (2.. (A+ .3 When A B 2 A+ and k A + B k k then k A k k.(BA.87). which also implies that kI .A) k + k (k B k . Since (A) 0 r(A)]. the bound follows from (2. this inequality implies A + B 2 A+ .B A B =) k A k k B k (2. where A+ A.5. Theorem 2. For later use we also record Lemma 2.kI A kI.AA. hence 0 A kI . k B k I A k B k I.(t) = maxf.85) for A ! B yield . A 2 AR . The implication .
. f+ (t) = maxft 0g.77) and 2.A.A+ it must be that (A) = 0. and f. so that k c1 (A) . one sees that if k cI . Gel'fand transformsupt2 (A) jc1 (A) .2.4 One has A+ = fA2 j A 2 ARg (2.88) p When (A) R+ andp A = A then A 2 AR is de ned by the continuous functional calculus p for f = . k A k k A k.1.87). )A. = .2.t 0g. A. because . By (2. A ing back to C (A I). B by the linearity of the partial ordering.86).30). so that (A) . A ^ k1 c. ) R+ because A. Apply the continuous functional calculus with f = id (A) = f+ . using . )3 = .6. we have jc .79) and (2. The real vector space in question is the space AR of all selfadjoint elements of A. we see from (2. The inclusion A+ fB B j B 2 Ag is is trivial from (2.e.6. Lemma 2.
C C 2 A+
now yields (C C ) = 0.C C 2 A+ for some C 2 A then C = 0. When A is separable (in containing a countable dense subset) then may be taken to be countable. Namely.92) (2.
I
=I
(2. Applying the above principle gives for all A B 2 A. CC : (2.. This is because A1 A2 is the same as A2 .88) is the fact that inequalities of the type A1 A2 for A1 A2 2 AR are stable under conjugation by arbitrary elements B 2 A. As (A. (2.2 we see that C C 0. A1 = A3 A3 .n n] and vanishes for jxj > n + 1.87).e. = 0. hence (. a proper ideal cannot contain I in order to prove properties of ideals we need a suitable replacement of a unit. Hence B B = A+ .89).1 An approximate unit in a nonunital C algebra A is a family fI g 2 . Hence C = 0 by 2. and 2.88) there is an A3 2 A such that A2 . A k= 0 lim !1 !1
For example. for all A 2 A.
. The last claim before the lemma therefore implies BA.90) This is because for z = 6 0 the invertibility of AB . so that the assumption . and take In to be a continuous function which is 1 on .
B A AB k A k2 B B
(2. )3 = . which lies in A+ .2. But clearly (A3 B ) A3 B 0. implies (CC ) R. = 0 by the continuous functional calculus with f (t) = t1=3 .
Proposition 2. i.2. )3 = 0.2 Every nonunital C algebra A has an approximate unit.91)
An ideal I in a C algebra A is de ned by 2. For example. An important consequence of (2.1 = B (AB .85) by A A.93) (2. (C C ) R+ . ) BA. the C algebra C0 (R) has no unit (the unit would be 1R. D E 2 AR. z . and nally A. with the
following properties: 1.3.7. = 0 we see that (A. but an approximate unit may be constructed as follows: take = N .89) Now for any A B 2 A one has (AB ) f0g = (BA) f0g: (2.
2. which does not vanish at in nity because it is constant). so that A1 A2 implies B A1 B B A2 B .1I. z . z implies the invertibility of BA .13) we can write C = D + iE .e. so that
kI k 1 k I A .CC ) R+ .32
2 ELEMENTARY THEORY OF C ALGEBRAS
Lemma 2..16). yielding A A k A k2 I. .6. and (I ) 2. one computes that (BA .(BA. Applying this with A ! C and B ! C we see that the assumption (C C ) R. A k= lim k AI . and use (2. As we have seen. One checks the axioms. and notes that one certainly does not have In ! 1R in the supnorm.6.1 A .7. By (2.6. z ). where is some directed set (i.6 If . and this is nothing but B AB B A2 B .7 Ideals in C algebras
De nition 2.A1 0 by (2.94)
0 1]. a set with a partial order and a sense in which ! 1). z ).2.9. By (2. replace A in (2. so that C C = 2D2 + 2E 2 .
and the involution (A) := (A ): (2.2 one has (B ) R+ . so certainly I 2 I. from which we build the element B := i Ai Ai .96)
We now apply (2. The quotient A=I is a C algebra in the norm (2.92). JJ I ) k k (J J . we estimate
k J . A) I A .94) follows analogously. which lies in A. Since f 0 and f assumes its maximum at t = 1=n. Finally. so that. I J )(J .10. JJ I ) .95) Since B is selfadjoint and B commutes with functions of itself (such as (n.38). JJ I ) k k (JJ .7. one has I = B C + B . This proves 2. Hence lim !1 k J . As (B ) R+ .2 it has an approximate unit fI g.
2
from a
Theorem 2. Clearly B is selfadjoint. 1 . Take J 2 I. the multiplication (2. P .16) and (2.1 I + B ).4 Let fI g be an approximate unit in I. Since any A 2 A sits in some directed subset of with n ! 1. Note that J 2 I implies J J 2 I \ I : it lies in I because I is an ideal.2. Using the continuous functional calculus on B . with f (t) = t=(n + t).7.2. J I ) k=k (JJ . partially ordered byP inclusion.7.1 is computed in AI. Lemma 2. a simple computation shows that
X
i
Ci Ci = n. and according to 2. JJ I ) k + k I (JJ .1 I+ B is invertible in AI.93). I (JJ .1 I + B ).7 Ideals in C algebras
33
One takes to be the set of all nite subsets of A. and it lies in I because I is an ideal.39). A k2 = lim k (I A . JJ I ) k : As we have seen. all we need to prove to establish 2.78) with A ! B and f (t) = n. it follows that k f k1 1=4n.97)
Note that (2.96). Hence we may form I := B (n. one has supt2R+ jf (t)j = 1=4n. A k= lim k C C k= 0: !1 !1 !1 i i The other equality in (2.3 Let I be an ideal in a C algebra A.2 . both terms vanish for ! 1.2 t(n.6. If A 2 I then A 2 I in other words. Putting Ci := I Ai . Although (n.3 then shows that k Ci Ci k 1=4n for each i = 1 : : : n. This uses
Lemma 2. Ai . Since I is an ideal.7.6. and let A 2 A.2 :
(2.7.98)
. 2 hence k n B (n I + B ) k 1=4n by (2.6. so that it is of the form C + I for some C 2 A and 2 C . one sees from (2. it follows from (2.7.78). and since I is an ideal it must be that J I 2 I for all .3. J I k2 =k (J .1 ). one has I = I .16).3. hence a leftideal. Using (2.1 + t).2 B (n.3.2 is the property (2. Put I := fA j A 2 Ig. Hence J is a normlimit of elements in I since I is closed. J I k= 0.
In view of 2.1 I + B ). also using (2. it follows that J 2 I. so that n. AI k : !1
(2.4 and 2. so that is countable. Then k (A) k= lim k A .77) and the positivity of B that (I ) 0 1]. J J I ) k + k I k k (JJ . when A is separable one may draw all Ai occurring as elements of countable dense subset.1 : (2. Hence by 2. But I lies in I \ I . so that k i Ci Ci k 1=4n by (2. 2 .1 I+ B ).1.1. I \ I is a C subalgebra of A. 1. The main properties of ideals in C algebras are as follows.97) is well de ned because of 2. J J 2 I \ I . Hence 2 is of the form = fA1 : : : An g. every ideal in a C algebra is selfadjoint. hence a rightideal.16) that lim k I A .
2.2.
De ne : A= ker(') ! B by ( A]) = '(A). AI k . I ) + J (I . Then (2. add a unit I to A if necessary. is injective. Since ' and are morphisms. and (2. I k 1 (2.97). k (A) k k A . since I is the kernel of the canonical projection : A ! A=I. since J 2 I. By (2. AI ) (A .4) its kernel is closed. since k k= 1. Also. its image is a algebra in B. In particular. and is a morphism by (2. we nd k (A . (2.5 we have '(f (A)) = 0.18).6. and (2. By Lemma 2. I k k A + J k k I .38) that
2 ELEMENTARY THEORY OF C ALGEBRAS
To derive the opposite inequality. AI k k (A) k :
(2. For the next consequence of 2.5 The kernel of a morphism between two C algebras is an ideal.5. Corollary 2. is a C algebra.6. which by the preceding sentence is closed in the norm of B. pick any J 2 I.7.7. (2. For this speci c J we combine (2. But (A= ker(')) = '(A).7. and ' = . In particular. and (2. Conversely.16).5 and 2.100) by 2. We now prove (2.6 An injective morphism between C algebras is isometric. '(A) is a C subalgebra of B. Hence '(A).99)
= lim k (I .82) implies ('(A)) (A). since '(A) = 0 implies '(AB ) = '(BA) = 0 for all B by (2. (2.39). Hence (A= ker(')) has closed range in B by 2.11 then implies (2. The nal claim follows from the preceding one. is a C algebra morphism. Hence every morphism has norm 1.1 and the proof of 2. I k + k J I . and write k A . (2. Assume there is an B 2 A for which k '(B ) k6=k B k. AI k k A + J k : (2. noting that A = A. so that ' has closed range in B.16) in AI.7. By (2. so that f ('(A)) = 0. I ) k lim k A A(I .34 It is obvious from (2.16). I k k A A(I . contradicting the injectivity of '. The converse follows from Theorem 2. Put A := B B .3 we need a Lemma 2. By the theory of vector spaces. Corollary 2.7. AI k : !1 Letting ! 0 proves (2.80) and (2. where A] is the equivalence class in A= ker(') of A 2 A. AI ) k k (A) k2 = lim k A .2.1. its range is closed.100).19) this implies k '(B B ) k6=k B B k.98). By Urysohn's lemma there is a nonzero f 2 C ( (A)) which vanishes on ('(A)). is a vector space isomorphism between A= ker(') and '(A).
.2. I ) k lim k I .39).18). This seemingly technical result is very important.99). The second term on the righthand side goes to zero for ! 1.1. where A=I is a C algebra.97).30) we must have (A) 6= ('(A)).7.40) to nd lim k A . I )A A(I . I ) k !1 !1 !1 =k (A A) k=k (A) (A ) k=k (A) (A) k : Lemma 2.98). since ' is continuous (see 2.40) holds.3.7 The image of a morphism ' : A ! B between two C algebras is closed.101). inheriting all operations in B.101) !1 For each > 0 we can choose J 2 I so that (2.16) in A=I. (2. every ideal in a C algebra is the kernel of some morphism.7. The rst claim is almost trivial. (2. According to 2. Hence lim k A . and (2. J k : Note that k I. the space A= ker(') is a C algebra. AI k2 = lim !1 !1
k A . In particular. AI k=k (A + J )(I .5.3.7.7.98). Successively using (2. Since ' is a morphism.
we rst note that the argument around (2. For example. for all A 2 A one has (2. and normalized in that !(I) = 1: (2. (2. We now pass to states on C algebras without unit.4 implies that (A B )! := !(A B ) de nes a preinner product on A.104) which will often be used. De nition 2.8 States
The notion of a state in the sense de ned below comes from quantum mechanics. Since the upper bound is reached by B = I. we have (2. Hence from (2. sA k !(I) Hence j!(I) . we look at a state in a more general context. and the bound just derived to nd j!(B )j2 !(B B )!(I) !(I)2 k B B k= !(I)2 k B k2 : Hence k ! k !(I).85). so that k I .3 A linear map ! : A ! C on a unital C algebra is positive i ! is bounded and k ! k= !(I): (2.16). showing that ! is real on AR.1 A state on a unital C algebra A is a linear map ! : A ! C which is positive in that !(A) 0 for all A 2 A+. using (2.105) !(A ) = !(A) as !(A ) = !(A I) = (A I)! = (I A)! = !(A).6. s!(A)j !(I). Then (assuming ! 6= 0) !k j!(I . Firstly. 2. To prove the converse claim. and normalization is obvious from (I) = ( ) = 1. sA k 1.1) we obtain j!(A B )j2 !(A A)!(B B ) (2. the bound j!(A)j !(I) k A k. so that ! is a probability measure.104) with A = I. which is only possible when !(A) 0.8.66) may be copied.103) Positivity follows from Theorem 2. The normalization of ! implies that !(1X ) = ! (X ) = 1.
. Choose s > 0 small enough. For general A we use (2. When ! is positive and A = A we have. since (B B ) =k B k2 0. we show that A 0 implies !(A) 0.2. Moreover.2 The state space of A = C (X ) consists of all probability measures on X .106) In particular: 1.4. The positivity of ! with 2.8. sA)j : 1 k I . Partly in order to extend the de nition of a state to nonunital C algebras. with norm 1. each positive linear map ! : C (X ) ! C is given by a regular positive measure ! on X .6.102) The state space S (A) of A consists of all states on A. we have Proposition 2. but states play a central role in the general theory of abstract C algebras also. Theorem 2.8 States
35
2. when A B(H) then every 2 H with norm 1 de nes a state by (A) := ( A ): (2. sA k= k !(I) k I .8. By the Riesz theorem of measure theory.106). A state on a unital C algebra is bounded. Next. An element ! 2 A for which k ! k= !(I) = 1 is a state on A.
.6 A state on a C algebra A is a linear map ! : A ! C which is positive and has
norm 1. AI )j ! 0 for any approximate unit in A. Since a state ! on A is bounded by 2. ) k
Now assume that Q is not bounded by the previous argument it is not bounded on A+ .5 A positive map between two C algebras is bounded (continuous). and is consistent with 2. The derivation of (2.6.13) and 2.105).8. where A0+ etc.8. A00 C k B k for all B 2 A+ and some + .106).5.102) it remains to prove positivity.8. are positive. we have j!(A .105) may then be copied from the unital case in particular. iA.8.4 A positive map Q : A ! B between two C algebras is a linear map with the property that A 0 implies Q(A) 0 in B.6.8.108) that
!I((A + I) (A + I)) j!(A) + j2
Hence ! is positive by (2.8 For every A 2 A and a 2 (A) there is a state !a on A for which !(A) = a. we obtain from (2. ) k + k Q(A00 4C k A k : + ) k + k Q(A.13).
Proposition 2. Thus Q is bounded on
Choosing B = C . The following result is very useful cf.
k Q(A) k n.2 Q(An ) 0 for each n. Since k A0 k k A k and k A00 k k A k by (2.
This de nition is possible by 2.86)
A+. When A = A there exists a state ! such that j!(A)j =k A k.5.2 k Q(An ) k n for all n 2 N .1 because of (2.108)
This obviously satis es (2. A0.8. which is impossible since k Q(A) k is some nite number.6. Combining this with (2. 2.1. Since Q is positive.7 A state ! on a C algebra without unit has a unique extension to a state !I
on the unitization AI. so that + for each n 2 N there isP an An 2 A+ n3 (here A+ 1 so that k Q(An ) k 1 consists of all A 2 A with 1 k A k 1).36
2 ELEMENTARY THEORY OF C ALGEBRAS
De nition 2.2An obviously converges to some A 2 A+.4. There are lots of states:
0:
Lemma 2.8. The series n=0 n.8.8.88).5 then provides an alternative proof of 2. Hence in the nonunital case we may replace the normalization condition in 2. by 2. Hence by (2.8. Using (2.1 as follows. one still has j!(A)j2 !(A A). Proposition 2.104) and (2. Hence if k Q(B ) k C > 0. we have k B k k A k 00 for B = A0+ A0. we can write 00 A = A0+ .107) + .
Let us rst show that boundedness on A+ implies boundedness on A. or A.8. we have Q(A) n.3.
The extension in question is de ned by
!I(A + I) := !(A) + :
(2. then
00 k Q(A) k k Q(A0+ ) k + k Q(A0.
De nition 2. we see that a state on a unital C algebra is a special case of a positive map between C algebras Proposition 2.5.5. + iA00 (2. and therefore on A by the previous paragraph.
For this a one has j!(A)j = jaj = r(A) =k A k by (2. Positive elements of C C are fo the form + _ with of the form !( + 0 _ and 0.9. Repeating this process. Since ! ~ a(I) = 1.111) This is because is a morphism cf. De ne a linear map ! ~a : C A + C I ! C by ! ~a ( A + I) := a + . Hence S (M2 (C )) is the unit ball in R3 . normalization yields c1 + c2 = 1. By 2.8.
. we parametrize 1 + x y + iz 1 =2 (2. Since (A) is closed by 2. )w belongs to C whenever vector space C is a subset of V such that the convex sum v + (1P v w 2 C and 2 0 1].6. S (A) is a closed subset of the unit ball of A by 2. The very simplest example is A = C .1. so that a positive functional must have c1 0 and c2 0.2 !a is a state. respectively. Since a 2 (A) one has a + 2 ( A + I) this easily follows from the de nition of . it follows that i pi vi belongs to C when all P pi 0 and i pi = 1. There is a natural equivalence relation in the set of all representations of A: two representations 1 2 on Hilbert spaces H1 H2 . In the unital case it is clear that S (A) is convex. k (A) k=k A k when is faithful by Lemma 2. It follows that the state space of a unital C algebra is a compact convex set.3.9 Representations and the GNSconstruction
37
where x y z 2 R.109) y . there is an a 2 (A) for which r(A) = jaj. so that S (A) A . We conclude that S (C C ) may be identi ed with the interval 0 1].9 Representations and the GNSconstruction
The material of this section explains how the usual Hilbert space framework of quantum mechanics emerges from the C algebraic setting. The dual is C 2 as well. Moreover. since I = 1+1.110) for all A B 2 A. Now consider A = M2 (C ). A convex set C in a .79). An important feature of a state space S (A) is that it is a convex set. we see that S (A) is closed in A if the latter is equipped with the w topology.8.7. We return to the unital case.3. so that S (A) is compact in the (relative) w topology by the BanachAlaoglu theorem. In the nonunital case one arrives at this conclusion most simply via 2. x
2. in which case S (A) is a point. (2. It follows that S (A) consists of all positive 2 2 matrices with Tr = 1 these are the density matrices of quantum mechanics.6. satisfying the bound k (A) k k A k : (2.3 that each element ! of S (A) is continuous. We identify M2 (C ) with its dual through the pairing !(A) = Tr !A. and all vi 2 C .8.1 A representation of A on a Hilbert space H is a (complex) linear map : A ! B(H) satisfying (A B ) = (A) (B ) (A ) = (A) (2. By the HahnBanach Theorem 2. it follows that k ! ~ k= 1.8. Finally.
If necessary we add a unit to A (this is justi ed by 2. which clearly satis es !a (A) = ! ~ a (A) = a. Let S (A) be the state space of a unital C algebra A. there exists an extension !a of ! ~ to A of norm 1.2.41) with A ! A + I implies j! ~a ( A + I)j k ( A + I) k.2. are called equivalent if there exists a unitary isomorphism U : H1 ! H2 such that U 1 (A)U = 2 (A) for all A 2 A. The next case is A = C C = C 2 .81).8.7. since both positivity and normalization are clearly preserved under convex sums.1. so that each element of (C 2 ) is _ ) = c1 1 + c2 2 . A representation is automatically continuous. In particular. To identify S (A) with a familiar compact convex set. The positivity of this matrix then corresponds to the constraint x2 + y2 + z 2 1.7). iz 1 . We saw in 2. Hence (2. Since w limits obviously preserve positivity and normalization.3.2. De nition 2.
38
2 ELEMENTARY THEORY OF C ALGEBRAS
The map (A) = 0 for all A 2 A is a representation more generally. The proof uses a lemma which appears in many other proofs as well. Hence ! (A) may be de ned on all of H! by continuous extension of (2. We restrict ourselves to the unital case the general case follows by adding a unit to A and using 2. and let p be the projection onto the closure of M . take a nonzero vector 2 H. and the conclusion implies that A ! p (A) de nes a subrepresentation of A on pH. de ne the sesquilinear form ( )0 on A by (A B )0 := !(A B ):
Since ! is a state. from any state ! 2 S (A) one can construct a cyclic representation ! on a Hilbert space H! with cyclic
vector ! in the following way. 2. referred to as a vector state relative to . 4. Construction 2.4 1.112)
N! = fA 2 A j !(A A) = 0g
(2.9. If V : A ! A=N! is the canonical projection. This process may be repeated on p? H. This subrepresentation is clearly cyclic. where it satis es (2.9.13) this is true for all A 2 M. then any unit vector 2 H de nes a state 2 S (A). Apply this lemma with M = (A) the assumption of nondegeneracy guarantees that p is nonzero. Hence p? Ap = 0 with p? = I . If A 2 M then ApH pH by de nition of p. with cyclic vector . Then p 2 M0 (that is. p A] = 0 for all A 2 M). Its null space
(2. Conversely.115)
it follows that ! is continuous. When A = A then (Ap) = pA = (pAp) = pAp = Ap so that A p] = 0.2 Any nondegenerate representation is a direct sum of cyclic representations. p this reads Ap = pAp. The cyclic vector is de ned by ! = V I. If is a nondegenerate representation of A on H. then by de nition
(V A V B )! := (A B )0 :
The Hilbert space H! is the closure of A=N! in this inner product. etc.7. By (2.103). such trivial may occur as a summand.114)
(2.113)
is a closed leftideal in A. A representation is called cyclic if its carrier space H contains a cyclic vector for this means that the closure of (A) (which in any case is a closed subspace of H) coincides with H. one says that a representation is nondegenerate if 0 is the only vector annihilated by all representatives of A.116)
. so that
( ! ! (A) ! ) = !(A) 8A 2 A:
(2. by means of (2. Lemma 2.8.3 Let M be a algebra in B(H).9. The representation ! (A) is rstly de ned on A=N! H! by
! (A)V B := V AB
(2. Proposition 2. Given ! 2 S (A). To exclude this possibility.115). hence a positive functional.110). The form ( )0 projects to an inner product ( )! on the quotient A=N! . 3. this form is positive semide nite (this means that (A A)0 0 for all A).
To prove that ! is continuous on A=N! . we compute k ! (A) k2 for = V B .72) quotients well to a map from A=N! to A=N! the latter map is ! de ned in (2. 2 2 H2 . On the other hand. It is trivial to verify that U intertwines ! and .117)
The equivalence follows from the CauchySchwarz inequality (2.5 If ( (A) H) is cyclic then the GNSrepresentation ( ! (A) H! ) de ned by any vector state (corresponding to a cyclic unit vector 2 H) is unitarily equivalent to ( (A) H). upon which
k ! (A) k k A k
follows from (2.
. For later use we mention that the GNSconstruction yields ( ! (A) ! ! (B ) ! ) = !(A B ): Putting B = A yields
(2. H2 of two cyclic representations cyclic vector 1 2 H1 . then
1 1 1
1
2
each contain a
(A) 1 ) = (
2
2
(A) 2 ) =: !2 (A)
(A) and
2
(A) are equivalent. Hence U is unitary.117) implies that N! is a leftideal.10 The Gel'fandNeumark theorem
One of the main results in the theory of C algebras is
Theorem 2.5 the representation 1 is equivalent to the GNSrepresentation !1 . This is important. But !(B B ) =k k2. The equality (2. which is closed because of the continuity of !. It follows from (2. This is very simple to prove: the operator U : H! ! H implementing the equivalence is initially de ned on the dense subspace ! (A) ! by U ! (A) ! = (A) this operator is wellde ned. satisfying (2.104). By (2. and 2 is equivalent to !2 . ! 2 S (A) hence it is de ned on the Hilbert space Hu = !2S (A) H! .
De nition 2. but since is cyclic for the image of U is H.115). one has k ! (A) k2 = !(B AA B ). for some Hilbert space H. so that k ! (A) k k A k k k. because it implies that the map (A) : A ! A de ned in (2.10.114) and step 2 above.91) and the positivity of ! one has !(B AA B ) k A k2 !(B B ).110) on the dense subspace A=N! of H! .9. First note that the null space N! of ( )0 can be de ned in two equivalent ways
N! := fA 2 A j (A A)0 = 0g = fA 2 A j (A B )0 = 0 8 B 2 Ag:
(2. so they must coincide.9.10. Proposition 2.2 The universal representation u of a C algebra A is the direct sum of all its GNSrepresentations ! . where A B 2 A.
By 2. it is easily checked that ! is a morphism as well.6 If the Hilbert spaces H1.119) (2. By (2.
2.116) and the fact that ! is a representation.
Corollary 2.1 A C algebra is isomorphic to a subalgebra of B(H).118)
(2.9.116) that U is unitary as a map from H! to U H! . and
!1 (A) := (
for all A 2 A.3).2.
The GNSconstruction leads to a simple proof this theorem. for ! (A) ! = 0 implies (A) = 0 by the GNSconstruction.10 The Gel'fandNeumark theorem
39
We now prove the various claims made here. which uses the following notion.120)
k ! (A) ! k2 = !(A A)
which may alternatively be derived from (2. Since is a morphism. !1 and !2 are induced by the same state.
16).1 A linear map Q : A ! B between C algebras is called completely positive if for all n 2 N the map Qn : Mn (A) ! Mn (B). In particular.10. one has U (Q(A))U . Another consequence of the GNSconstruction. and that Q is normalized.2 Let Q : A ! B be a completely positive map between C algebras with unit.8.3. in which the involution in A replaces the usual complex conjugation in C . One may identify Mn (A) with A Mn (C ) in the obvious way. which implies k A A k= 0 by Lemma 2. The involution in Mn (A) is. de ned by (Qn (M ))ij := Q(Mij ). The norm is unique by Corollary 2. which is of central importance for quantization theory. It turns out that for this purpose one needs to impose a further condition on Q. so that k A k= 0 by (2.e. By Theorem 2. where each vi 2 H.11.10. Hence ! (A) ! = 0.1. suppose that u (A) = 0 for some A 2 A. and U : ~ ~ .1). since when A = B B in Mn (A). the order has to be taken into account. (MN )ij := k Mik Nkj . When is a faithful representation of A (which exists by Theorem 2. a morphism ' is a completely positive map. such that Q(I) = I. This generalization will appear as the proof of the following Stinespring theorem. any representation of A on H is a completely positive map from A to B(H).1.2.5. which is positive in Mn (B). of course. Since n (Mn (A)) is a closed algebra in B(H C n ) (because n < 1). so that this procedure does not depend on the choice of .10. is Corollary 2. De nition 2. We rst introduce the C algebra Mn (A) for a given C algebra A and n 2 N . hence k ! (A) ! k2= 0 by (2.4 generalizes the notion of a state. the Hilbert space Hu is absurdly large in practical examples a better way of obtaining a faithful representation always exists.8.121) ~ := pH Equivalently. seen as map not from H to H but as a map from H to H is unitary. While the universal representation leads to a nice proof of 2. De nition 2. with p := WW (the target projection of W on H ). In particular.120) this means ! !(A A) = 0 for all states !. is positive.11. in that the C in ! : A ! C is replaced by a general C algebra B in Q : A ! B. Theorem 2. we get an interesting generalization of the GNSconstruction.3 An operator A 2 A is positive (that is. so that U H ! H de ned as W .6.1 then follows by taking H = Hu the desired isomorphism is u . For example.11 Complete positivity
. If we also assume that A and B are unital. one obtains a faithful realization n of Mn (A) on H C n . the morphism u is isometric by Lemma 2.10. H H . There exists a Hilbert space H . and nally A = 0 by the de nition of a norm. de ned by linear extension of n (M )vi := (Mij )vj we here look at elements of H C n as ntuples (v1 : : : vn ).10. we may assume that B is faithfully represented as a subalgebra B ' (B) B(H ). or rather of 2. To prove that is injective. By de nition of a direct sum. for some Hilbert space H .8. such that (Q(A)) = W (A)W 8A 2 A: (2. Being injective.40
2 ELEMENTARY THEORY OF C ALGEBRAS
Theorem 2. and a partial isometry W : H ! H (with W W = I). given by (M )ij = Mji . with the di erence that one now multiplies elements of A rather than complex numbers.122)
2. For example. We would like to see if one can generalize the GNSconstruction. it is obvious that Mn (A) is a C algebra in this norm. The norm k M k of M 2 Mn (A) is then simply de ned to be the norm of n (M ). this implies u ( A ) = 0 for all states !. The elements of Mn (A) are P n n matrices with entries in A multiplication is done in the usual way. the best faithful representation of B(H) is simply its de ning one. We have seen that a positive map Q (cf. then '(A ) = '(B ) '(B ).10.1 = p (A)p: (2. A 2 A+ (A) 0 for all cyclic R) i representations . a representation of A on H . i.7.
this form is positive semide nite denote its null space by
(V (A v) V (B w)) := (A v B w)0 :
The Hilbert space H is the closure of A H =N in this inner product. with the property that H1 contains a closed subspace K1 such that (W W )2 = ( )1 for all 2 K1 .129) is positive. we write
X
ij
(Ai vi Aj vj )0 =
X
ij
(vi
(Q(Ai Aj ))vj ) :
(2.126) one uses (2. then by de nition
N. Since Q is completely positive. Namely. Hence A
. one concludes that the righthand side of (2.121) follow.129)
Now consider the element A of Mn (A) with matrix elements A ij = Ai Aj . Hence W is unitary from K1 to W K1 .123) is positive. To show that the form de ned by (2. whose rst row is (B1 : : : Bn ). Looking in a faithful representation n as explained above.127) (2. respectively. This is a linear map W : H1 ! H2 between two Hilbert spaces.126) so that (A) may be de ned on all of H by continuous extension of (2.2. We denote elements of H by v w. is positive in Mn (B). The representation (A) is initially de ned on A H =N by linear extension of
(2.
P where Az = A z . The form ( )0 projects to an inner product ( ) on A H =N . respectively. for arbitrary A B1 : : : Bn 2 A we conjugate the inequality 0 A AIn k A k2 In with the matrix B . De ne the sesquilinear form ( )0 on A H (algebraic tensor product) by (sesqui)linear extension of
(A v B w)0 := (v
(Q(A B ))w) :
(2.125).3 1. it must be that B . de ned by i i i its matrix elements B ij := Q(Ai Aj ). 3. and W = 0 ? .11. The map W : H ! H . Its adjoint W : H ! H is given by (continuous extension of)
W V A v = (Q(A))v
from which the properties W W = I and (2. with inner product (v w) . To prove (2. It uses the notion of a partial isometry. 4.125)
satis es (A ) = (A) . Repeating the above argument with A and replaced by B and . This extension
Wv := V
I
(A)N
v
(2.
Since Q is completely positive. Construction 2.11 Complete positivity
41
The proof consists of a modi cation of the GNSconstruction. One has the bound k (A) k k A k (2.123)
2. If V : A H ! A H =N is the canonical projection. de ned by
N .128)
is a partial isometry. because
(2. one sees that (z A z ) =
X
ij
(zi (Ai Aj )zj ) =
X
ij
( (Ai )zi (Aj )zj ) =k Az k2 0
0. It follows that WW = K2 ] and W W = K1 ] are on K1 projections onto the image and the kernel of W .91) in Mn (A).124)
(A)V (B w) := V (AB w)
this is wellde ned.
and. where F (x) = i fi (x)Mi for fi 2 C (X ) and Mi 2 Mn (C ).4 Let A be a commutative unital C algebra.121). When the extension exists but does not preserve the unit. We may then identify Mn (C (X )) with C (X Mn (C )). use the de nition to compute (Wv Ww) = (V
vV
I
w) = (I v
w ) 0 = (v w )
) = (Ww ) for
where we used (2. Since Q is completely positive. but W is no longer a partial isometry one rather has k W k2=k Q(I) k.42
2 ELEMENTARY THEORY OF C ALGEBRAS
and which has zeros everywhere else the adjoint B is then the matrix whose rst column is (B1 : : : Bn )T .130) one then has
P
k (A) k2 = k A k2
X
ij
(ABi vi ABj vj )0 =
X
ij
(vi
(Q(Bi A ABj ))vj ) (Bi vi Bj vj )0
X
ij
(vi
=k A k2 (V
X
i
(Q(Bi Bj ))vj ) =k A k2
X
ij
Bi vi V
I
X
j
Bj vj ) =k A k2 k k2 :
I
To show that W is a partial isometry.127) and (2. This leads to 0 B A AB k A k2 B B .
. To check (2.122) follows from (2. W W = I for the projection onto the subspace of H on which W is isometric this subspace is H itself Hence (2. and the sum is nite. If A and perhaps B are nonunital the theorem holds if Q can be extended (as a positive map) to the unitization of A.11.4. one is in the situation of the previous paragraph. are dense in C (X Mn (C )). 2. By Theorem 2.1 = W (Q(A))W = WW
(A)WW = p (A)p:
When Q fails to preserve the unit.. Hence in any representation (B) and any vector (v1 : : : vn ) 2 H C n one has
X
ij
(vi
(Q(Bi A ABj ))vj ) k A k2
X
ij
(vi
(Q(Bi Bj ))vj ):
(2. The relevance of Stinespring's theorem for quantum mechanics stems from the following result. one has p = WW for the projection p onto the image of W .125).121).1 we may assume that A = C (X ) for some locally compact Hausdor space X . Thus it is no longer possible to regard H as a subspace of H . such that the extension preserves the unit I (perhaps relative to the unitization of B). the above construction still applies.123) and Q(I) = I. we use (2.128). and (2. 3.e. This trivially veri ed. Positive elements G of C (X Mn (C )) can be normapproximated by positive F 's. Elements of the form F . The proof then proceeds in the following steps: P 1. Such F is positive i all fi and Mi are positive. when G 0 there is a sequence Fk 0 such that limk Fk = G.128) to compute
W
(A)Wv = W
(A)V (I v) = W V (A v) = (Q(A))v:
Being a partial isometry. from (2. one merely uses the de nition of the adjoint.123). (w W all w 2 H and 2 H . To verify (2. viz.130)
With
= i V Bi vi . and all other entries zero. one has Qn (B A AB ) k A k2 Qn (B B ). Then any positive map Q : A ! is completely positive. i. Proposition 2. since
U (Q(A))U . (2. in this case.
12 Pure states and irreducible representations
We return to the discussion at the end of 2.. When G in item 1 is positive then each G(xi ) is positive. Take G 2 C (X Mn (C )) and pick > 0.5 that Q is continuous the continuity of Qn follows because n < 1.2. such that the support of 'i lies in Oxi and li=1 'i (x) = 1 for all x 2 X . Since G is continuous. 3.15). G(y) k < g
is open for each x 2 X . Such a partition of unity exists. G(x) k=k
'i (x)(G(xi ) .
43
6. G(x) k <
'i (x) = :
Here the norm is the matrix norm in Mn (C ). Qn (F ) is positive when F is positive. An element F 2 C (X Mn (C )) is positive i F (x) is positive in Mn (C ) for each x 2 X .11. the set
Ox := fy 2 X k G(x) . G(x) k < for all x 2 Oxi .1 one has Qn (F ) = i Q(fi ) Mi .e. Hence
k Fl . By 2. i.2 we infer that F de ned by P F (x) = i fi (x)Mi is positive when all fi and Mi are positive. G k= sup k Fl (x) . because by (2. as we have just seen. A = B B by continuity of multiplication. Now each operator Bi M is positive in Mn (B) when Bi and M are positive (as can be checked in a faithful representation). 5.12 Pure states and irreducible representations 4. On F as speci ed in 2.4. In particular. P Since Q is positive.2. Finally. where i = 1 : : : l.
.6. A normlimit A = limn An of positive elements in a C algebra is positive. A partition of unity subordinate to the given cover is a collection of continuous positive P functions i 2 C (X ). We now prove each of these claims. 4. and lim Bn = B exist because of (2. Qn is continuous.
P
2. Now de ne Fl 2 C (X Mn (C )) by
Fl (x) :=
l X i=1
l X i=1
'i (x)G(xi ):
l X i=1 l X i=1
(2. when F (x) = f (x)M for some f 2 C (X ) and M 2 Mn (C ) then F is positive i f is positive in C (X ) and M is positive in Mn (C ). We know from 2.8. by (2. 5. it follows that Qn maps each positive element of the form F = i fi Mi into a positive member of Mn (B). one has
k Fl (x) .88) we have An = Bn Bn . hence is positive.16). If Fk ! G 0 in C (X Mn (C )) then Q(G) = limk Q(Fk ) is a normlimit of positive elements.131)
Since k G(xi ) . 6. This gives an open cover of X .8. 1. G(x) k < :
x2X
2. G(x)) k
'i (x) k G(xi ) . The intrinsic de nition of this boundary is as follows. One sees that the compact convex sets in the examples have a natural boundary. which by the compactness of X has a nite subcover fOx1 : : : Oxl g.
3. for @e K .12. Every vector in H is cyclic for (A) (recall that this means that (A) is dense in H). By Lemma 2.3 below and the spectral theorem. or simply P . Proposition 2.12.44
of K which can only be decomposed as
2 ELEMENTARY THEORY OF C ALGEBRAS
De nition 2.13. referred to as the pure state space of A. it can be shown that the projections in the spectral resolution of A lie in (A)0 if A does.1 An extreme point in a convex set K (in some vector space) is a member !
! = !1 + (1 .9. if !1 = !2 = !. If p ! = 0 then Ap ! = pA ! = 0 for all A 2 A. More generally. It is a deep fact of C algebras that the quali er `closed' may be omitted from this de nition. Using Theorem 2. (A)00 = B(H) (Schur's lemma) 2.12.4 The GNSrepresentation ! (A) of a state ! 2 S (A) is irreducible i ! is pure.1 is due to Minkowski it was von Neumann who recognized that this de nition is applicable to quantum mechanics. Hence when (A)0 is nontrivial it contains a nontrivial projection p. This precisely reproduces the notion of a pure state in quantum mechanics. the de ning representation d of the matrix algebra MN on C N is irreducible. this projection lies in (A)0 . Hence \ irreducible ) every vector cyclic". A state that is not pure is called a mixed state. when (A)0 = C I and is reducible one nds a contradiction because the projection onto the alleged nontrivial stable subspace of H commutes with (A). When K = S (A) is a state space of a C algebra we write P (A). The converse is trivial. so that cannot be irreducible by Schur's lemma. We may now ask what happens to the GNSconstruction when the state ! one constructs the representation ! from is pure. When there exists a vector 2 H for which (A) is not dense in H.12.14. The commutant (A)0 is a algebra in B(H). This de nition should be familiar from the theory of group representations. The collection @e K of extreme points in K is called the extreme boundary of K . Hence \ irreducible ) (A)0 = C I". the de ning representations d of B(H) on H is irreducible as well. In the in nitedimensional case. In preparation: De nition 2. The commutant of (A) in B(H) is f I j 2 C g in other words.10 that the state space of Mn (C ) consists of all positive matrices with unit trace the pure state space of Mn (C ) then consists of all onedimensional projections.2. there is a nontrivial projection p 2 ! (A)0 by Schur's lemma.
Hence the pure states on A = C C are the points 0 and 1 in 0 1]. we will prove in 2.2 A representation of a C algebra A on a Hilbert space H is called irreducible if a closed subspace of H which is stable under (A) is either H or 0.12. When ! is pure yet ! (A) reducible.12. contradicting irreducibility. But then pH is stable under (A). Theorem 2.
. Clearly. whereas 1 maps it to . Let ! be the cyclic vector for ! . but we will not prove this. The pure states on A = M2 (C ) are the the functional mapping + matrices in (2. Conversely. )!2
(2.3 Each of the following conditions is equivalent to the irreducibility of (A) on H: 1. We are now in a position to answer the question posed before 2.109) for which x2 + y2 + z 2 = 1 these are the projections onto onedimensional subspaces of C 2 . we can form the projection onto the closure of (A) . An extreme point in the state space K = S (A) of a C algebra A is called a pure state. The rst part of De nition 2. so when it is nontrivial it must contain a selfadjoint element A which is not a multiple of I. with M = (A). Hence \ (A)0 = C I ) irreducible". where 0 is identi ed with _ to .132)
2 (0 1).
suppose ! is irreducible.e.133) in other words. Continuing with the proof of 2.12. with (2.105) with A ! A B and ! ! !1 . ^ ( ) is then bounded by (2. ? := p? ! = k p? ! k.103).12. This allows us to de ne a quadratic form (i. Combining this with 2. so that by the RieszFischer Hold xed.135) holds Lemma 2. Q ^ Furthermore.5 we have the Corollary 2. )!2. and =k p? ! k2 .137) ( ! (A) ! Q ! (B ) ! ) = !1 (A B ): 0 It is the immediate from (2. It follows that Q Moreover.137). which is positive hence !1 (A A) !(A A) for all A 2 A. There is a bounded operator Q on H such that ^ ( ) = ( Q ) for all Q 2 H.135). (2. ^ on a Hilbert space H be bounded.135) ^ can be extended to all of H! by continuity. This proves \pure ) irreducible".120) imply that Q is bounded in that ^ ( )j C k k k k jQ (2.132) for !1 !2 2 S (A) and 2 0 1]. with := p ! = k p ! k.119) show that !1 is proportional to !. and therefore equal to ! by normalization. By (2. a sesquilinear map) Q ^ ( ! (A) ! ! (B ) ! ) := !1 (A B ): Q (2.104) this yields j !1 (A B )j2 2 !1 (A A)!1 (B B ) !(A A)!(B B ) (2.12. by (2. The theorem there exists a unique vector such that Q selfadjointness of Q in case that (2. From 2. (2. A2 ) B )j2 0 jQ ^ ( ! (A1 ) ! ! (B ) ! ) = Q ^ ( ! (A2 ) ! ! (B ) ! ). one has ^( ) = Q ^( ) Q (2. and some constant C 0. p? ! = 0 is impossible. we see that there is a selfadjoint operator Q on H! such that (2.133) ^ on ! (A) ! by for all A B . whence k Q k C .4. Now use (2.136) is satis ed Q is selfadjoint. De ne Q by Q = . Hence ! cannot be pure.12 Pure states and irreducible representations
45
so that p = 0 as ! is cyclic.5 Let a quadratic form Q for all 2 H.134) This is well de ned: when ! (A1 ) ! = ! (A2 ) ! then !((A1 . so that ! is pure. The map ! Q ^ ( ) = ( ). so that ^ ( ! (A1 ) ! ! (B ) ! ) . and (2.136) holds is obvious.135) to estimate ^ (Q ) C k Q k k k 2 k Q k2= (Q Q ) = Q taking the supremum over all in the unit ball yields k Q k2 C k Q k2 .
. in that (2. for all 2 ! (A) ! . In the opposite direction. Q ^ ( ! (A2 ) ! ! (B ) ! )j2 = j !1 ((A1 .4 yields Corollary 2.12.7 Every irreducible representation of a C algebra comes from a pure state via the GNSconstruction.110) that Q ! (C )] = 0 for all C 2 A.134). with C = 1. Hence Q 2 ! (A) since ! is irreducible one must have Q = tI for some t 2 R hence (2.133) and (2. A2 ) (A1 . A2 )) = 0 by (2. Then !1 . Similarly.12.120).2.6 If ( (A) H) is irreducible then the GNSrepresentation ( ! (A) H! ) de ned by any vector state (corresponding to a unit vector 2 H) is unitarily equivalent to ( (A) H).136) by (2.9. ! = (1 . We may then decompose ! = + (1 . and k Q k C . where and ? are states de ned as in (2. Similarly for B . When (2.. ) ? .
and 0 1 . let ! be a pure state.3 and Theorems 2.12. Hence (! . is positive. It follows that arbitrary states on a C algebra may be approximated by nite convex sums of pure states.11 For every A 2 AR and a 2 (A) there is a pure state !a on A for which !a(A) = a. Since !(f ) . one has 0 !g !. by 2. ) ! = 1! . g 1X .8. since ! .8.4. so when 6= 0 it must be that ker(!x ) = ker( ). Since two functionals on any vector space are proportional when they have the same kernel. k= !(I) .61) clearly de nes a pure state.8 A state is pure i 0 ! for a positive functional implies = t! for some t 2 R+ .6 and 2.8 as follows Theorem 2.4. hence k ! . )w j v w 2 V 2 0 1]g: (2. There exists a pure state ! such that j!(A)j =k A k. In view of Proposition 2. whereas (I) = 0 implies = 0. (I). K = co(@e K ). it follows from 2. and suppose a functional satis es 0 !x .12.4. Hence ker(!) is an ideal. This is possible because the unique extension of a pure to AI = C (X ~ ) guaranteed by 2.8 that !x is pure. Fortunately.3. then fg 2 ker(!) for all g 2 C (X ). applied to C (X ) it shows that arbitrary probability measures on X may be approximated by nite convex sums of point (Dirac) measures. For example. so that !1 = t! by assumption normalization gives t = . The convex hull co(V ) of a subset V of a vector space is de ned by co(V ) := f v + (1 .8.46
2 ELEMENTARY THEORY OF C ALGEBRAS
A useful reformulation of the notion of a pure state is as follows.4). (I). ker(!g ) = ker(!).
.8. 2. In other words. then 0 < (I) < 1. We assume that A is unital if not. this possibility is excluded by the KreinMilman theorem in functional analysis. Conversely. )=(1 .108).4.8. It follows that when f 2 ker(!). When ! is pure and 0 !. !g (f ) = !(f (1 . we merely need to establish a bijective correspondence between the pure states and the multiplicative functionals on C0 (X ). the proof of 2. Hence (I) would imply ! = .12.6 and 2. (I)) and = (I) are states. Then ker(!x) ker( ). Let !x 2 (C (X )) (cf. Thus we put A = C (X ). we may now re ne Lemma 2. g)).9 The pure state space of the commutative C algebra C0(X ) (equipped with the relative w topology) is homeomorphic to X .12. (I) (I) with = 1 . This is a spectacular result: for example. and ! is pure. contrary to assumption. Moreover. and the functional 1 in (2.4. But ker(!x) is a maximal ideal.54) coincides with the extension !I of a state de ned in (2. which we state without proof. the proof of 2. De ne a functional !g on C (X ) by !g (f ) := !(fg).12. Hence !g = t! for some t 2 R+ by 2. Since ! is pure.7 etc. + (1 .10 A compact convex set K embedded in a locally convex vector space is the closure of the convex hull of its extreme points.3.12.3. if (2. use 2.12. (I ) = 1 .7.7 remains pure.1 we have = (I)!. and
. since any function is a linear combination of functions for which 0 g 1X . and pick a g 2 C (X ) with 0 g 1X . It would follow that such a C algebra has no irreducible representations. the extension of state on C0 (X ) to a state on C (X a multiplicative functional de ned in (2.138) Theorem 2. Conversely. and ker( ) is an ideal. In general. The case that X is not compact may be reduced to the compact case by passing from A = C0 (X ) ~ ) cf.
The simplest application of this proposition is Theorem 2.4. For = 0 or = ! the claim is obvious.3). Proposition 2.132) holds then 0 !1 ! (cf. Hence ! is multiplicative by Theorem 2.1 and 2.12. which is maximal because the kernel of a functional on any vector space has codimension 1. it guarantees that a C algebra has lots of pure states. with 0 6= 6= !. In particular. hence !1 = ! = !2 . It could be that no pure states exist in S (A) think of an open convex cone.
1 Let H be a Hilbert space. De nition 2. The simplest application of this re nement is Proposition 2. We may further restrict this direct sum by de ning two states to be equivalent if the corresponding GNSrepresentations are equivalent.12 Every nitedimensional C algebra is a direct sum of matrix algebras. which excludes the possibility that (A) and !a (A) are equivalent (as the spectrum is invariant under unitary transformations). Furthermore.12. For example. On the other hand.12. as will follow from this section. But it is clear that. It follows from 2. We may now replace the use of 2.6 that any vector state on B(H) de nes an irreducible representation of B(H) which is equivalent to the de ning representation.1. If !a were not an extreme point in S (A).11 and the existence of bounded selfadjoint operators with continuous spectrum (such as any multiplication operator on L2 (X ). For one thing.8.139.13. we put ! ~a (An ) = an etc.139) It is obvious that the proof of 2.8 by 2. that is.13 The C algebra of compact operators
. !1 and !2 would coincide on C (A I). This section is devoted to an exhaustive study of this C algebra.8.1 still goes through. it follows from Theorem 2.8) is a closed convex subset Ka of S (A) hence it is a compact convex set.12. that there are many other pure states whose GNSrepresentation is not equivalent to the de ning representation .8. where X is connected).116) that !a 2 H!a is an eigenvector of !a (A) with eigenvalue a.
2. It would appear that the appropriate generalization of the C algebra Mn (C ) of n n matrices to in nitedimensional Hilbert spaces H is the C algebra B(H) of all bounded operators on H. concluding that the universal representation u may be replaced by r := !2P (A) ! . so that !a cannot be an extreme point of Ka .11 in the proof of the Gel'fandNeumark Theorem 2.. Since A= ker( ) is nitedimensional. it would be decomposable as in (2. B0 (H) is a basic building block in the theory of C algebras. unlike Mn (C ) (which. we know from 2. in that case. and taking only one pure state in each equivalence class.10. For it is easy to show from (2.10 it has at least one extreme point !a.12. most of these are realized on nonseparable Hilbert spaces.3. One easily checks that the set of all extensions of ! ~ a to A (which extensions we know to be states see the proof of 2.10.132). We then have A ' r (A) := !2 P (A)] ! (A): (2.12. a is in the continuous spectrum of A = (A) but in the discrete spectrum of !a (A).2. an operator A 2 B(H) lies in Bf (H) when AH := fA j 2 Hg is nitedimensional.12. has only one irreducible representation up to equivalence).3 below that (A)00 = (A) in every nitedimensional representation of A. Then apply the isomorphism 2. hence certainly for any representation ' = .1 implies that (A) must be a matrix algebra (as B(H) is the algebra of n n matrices for H = C n ). then !a cannot be equivalent to .13 The C algebra of compact operators
47
We extend the state in the proof of 2. Let us refer to the ensuing set of pure states as P (A)].8 to C (A I) by multiplicativity and continuity. Namely.9 that this extension is pure. it must be that (A) is isomorphic to an algebra acting on a nitedimensional vector space. This is not the case. In noncommutative geometry elements of this C algebra play the role of in nitesimals in general. B(H) has a huge number of inequivalent representations even when H is separable. when A 2 B(H) and a 2 (A). In other words. but a is not in the discrete spectrum of A as an operator on H (i. it follows from 2. By the KreinMilman theorem 2. upon which 2.14. The appropriate generalization of Mn (C ) to an in nitedimensional Hilbert space H turns out to be the C algebra B0 (H) of compact operator on H. For any morphism '. In other words.12.e. one has the isomorphism '(A) ' A= ker('). The algebra Bf (H) of niterank operators on H is the ( nite) linear span of all nitedimensional projections on H. there is no eigenvector a 2 H for which A a = a a ). Another argument against B(H) is that it is nonseparable in the nomtopology even when H is separable.
so that we can nd N such that k p( n . note that the weak topology on H (in which n ! i ( n ) ! ( ) for all 2 H) is actually the w topology under the duality of H with itself given by the RieszFischer theorem. since p = p for any projection p. it follows that AB1 is compact. This maximum is k A k2 by (2. I) k= 1.2. Clearly jaj =k A k. De ne fA : B1 ! R by fA ( ) :=k A k2 . it is the smallest C algebra of B(H) containing Bf (H). Secondly. so that k (An . ) k :
Since the weak and the norm topology on a nitedimensional Hilbert space coincide. The C algebra B0 (H) is an ideal in B(H). and put p := Af H]. with k n k= 1 for all n.3 (noting that A A 2 B0 (H)). If A 2 B0 (H) then AB1 is compact in H (with the normtopology). Now if An ! A then An B ! AB and BAn ! BA by continuity of multiplication in B(H). the nitedimensional projection onto the image of Af .3).13. Indeed. fA ( )j = j( n A A( n . Hence fA is continuous. Hence Bf (H) is an ideal in B(H). Then (An . Since B1 is weakly compact. Hence the unit ball B1 is compact in the weak topology by the BanachAlaoglu theorem. )) . The unit operator I lies in B0 (H) i H is nitedimensional. Hence supk k=1 k (An . since AB H = AH. Hence when A = A the property k A k2 =k A a k2 with k a k= 1 implies 2 A2 a = a a . Here B1 is the unit ball in H. choose Af 2 Bf (H) such that k A . Af ) + Af ( n . n A A )j k A A( n . fA assumes its maximum at some a. A k=k (A . ( . The third item in the next proposition explains the use of the word `compact' in the present context. the norm in B0 (H) is the operator norm (2. then jfA( n ) . But since BA = (A B ) . Then
k A n . where a2 =k A k2 .48
2 ELEMENTARY THEORY OF C ALGEBRAS
The C algebra B0 (H) of compact operators on H is the normclosure of Bf (H) in B(H) in other words. An operator A 2 B(H) lies in B0 (H) when it can be approximated in norm by niterank operators. ) k +j( . I) = . and Bf (H) is a algebra. hence k An . they coincide on pH. When n ! weakly with k n k= 1. n A A )j: The rst term goes to zero by the proof of 2. So if we can show that A 2 B0 (H) maps weakly convergent sequences to normconvergent sequences.13. one has A B 2 Bf (H) and hence BA 2 Bf (H). ) k
1 3
+1 3 + k Af k k p( n . I) k 1. Now the CauchySchwarz inequality with = 1 gives k A k2 = ( A A ) k A A k. Af k < =3.
2. In particular.3). 3.3 A selfadjoint operator A 2 B0 (H) has an eigenvector a with eigenvalue a such that jaj =k A k. let n ! in the weak topology. ) k < =3 for all n > N . for any sequence (or net) An 2 Bf (H) we may choose a unit vector n 2 (An H)? .2
1.. Af ) n + (A . i. when A 2 Bf (H) and B 2 B(H) then AB 2 Bf (H). Thirdly.e. Hence B0 (H) is an ideal by virtue of its de nition.
Corollary 2. I k! 0 is impossible by de nition of the norm (2. save for the fact that it is not normclosed (unless H has nite dimension). with equality i A A is proportional to . Since
k k2 = (
) = lim ( n
n)
k kk
n k=
one has k k 1.3) in B(H) (hence in B0 (H)).
It is clear that Bf (H) is a algebra. then A is continuous from H with the weak topology to H with the normtopology since compactness is preserved under continuous maps.13. and the second goes to zero by de nition of weak convergence. Hence k A n .
. A k < .
Firstly. The spectral theorem or the continuous functional calculus with p f (A2 ) = A2 = A implies A a = a a. the set of all 2 H with k k 1. Given > 0.
Proposition 2. .
where each eigenvalue ai has nite multiplicity. The HilbertSchmidt class B2 (H) consists of all A 2 B(H) for which k A k2 < 1.13. except for the fact that they are not closed.2. This ordering is possible because by 2.13. To show that (2.141) is independent Pof the basis. and can
only have 0 as a possible accumulation point. Since the operator N a ] is clearly of nite rank. This proves \A compact and selfadjoint ) A diagonalizable with limi!1 jai j = 0". and compact by 2. This proves \A compact and selfadjoint ) A diagonalizable". which must lie in p? H. This contradicts the de nition of p? H unless p? H = 0. Then limi!1 ( i ) = 0 for all 2 H.13.2. with limi!1 jai j = 0.2. and let p be the projection onto the closure of the linear span of all eigenvectors of A. hence diagonalizable.13. so k A i k= jai j ! 0 by (the proof of) 2. so that (pA) = pA. Let A be compact and selfadjoint. This involves the study of a number of subspaces of B(H) which are not C algebras.2.141)
where fei gi is an arbitrary basis of H the righthand side is independent of the choice of the basis.1 proves the claim about the multiplicity of the eigenvalues. By 2. so that
(
)=
X
i
j(
i )j2
(2. we take a second P basis fuigi. The trace norm k A k1 of A 2 B(H) is de ned by
4 k2 k A k1 :=k (A A) 1 2 1
(2. which implies that A restricted to p? H is zero. We now wish to compute the state space of B0 (H).2. In other words.13.9. Hence k p? A k= 0 2.13. De nition 2. As in Lemma 2.5 The HilbertSchmidt norm k A k2 of A 2 B(H) is de ned by
P a ). p)A is selfadjoint. For N < 1 and 2 H one then has
k (A . Let now A be selfadjoint and diagonalizable. since the i form a basis. this proves that A is compact.140). Hence i ! 0 weakly. N i=1 i
i=N +1
jai j2 j(
i )j2
jaN j2
1 X
i=N +1
j(
i )j2 :
k A k2 2 :=
X
i
k Aei k2
(2.
N X i=1
ai i ]) k2 =k
1 X
i=N +1
ai ( i ) i k2
1 X
Using (2. Aince I = i ei ] we then have
k A k2 2 :=
X
ij
(ej ui )(ui A Aej ) =
X
ij
(A Aui ej )(ej ui ) =
X
i
k Aui k2 :
. Hence p?A = (I.4 A selfadjoint operator A 2 B(H) is compact i A = Pi ai i ] (normconvergent sum).13.13 The C algebra of compact operators
49
Theorem 2. which by 2.142)
where (A A) 4 is de ned by the continuous functional calculus. Hence limi!1 jai j = 0. this is P jaN j2 ( i ] k= 0. Normalize the eigenvectors i := ai to unit length. Hence i=1 i i \A selfadjoint and diagonalizable with limi!1 jai j = 0 ) A compact".3 there is a largest eigenvalue. which implies that all vectors in p? H are eigenvectors with eigenvalue zero. with corresponding resolution of the identity I = i ui ] (weakly).3. The trace class B1 (H) consists of all A 2 B(H) for which k A k2 < 1.3 the compact selfadjoint operator p? A has an eigenvector. so that limN !1 k A. one has limi!1 jai j = 0.3. but which are ideals of B(H). Ordering the eigenvalues so that ai aj when i > j .13.3 one sees that A p] = 0. Finally. because limN !1 jaN j = 0. By assumption this eigenvector can only be zero.140)
which clearly converges. when A is compact its restriction to any closed subspace of H is compact. and must therefore be an eigenvector of A in p? H. Let A 2 B0 (H) be selfadjoint. the set of eigenvalues is discrete.
and put U = 0 on ran(jAj)? . follow from (2. .148) for all positive A 2 B(H) and all unitaries U . Tr A0.13. Conversely. the linearity property Tr (A + B ) = Tr A + Tr B (2. it may happen that Tr A depends on the basis it may even be nite in one basis and in nite in another). and that B2 (H) is complete in this norm. whereas UU is the projection onto the closure of ran(A). whether or not A 2 B1 (H). for all A 2 B(H) one has kAk k A k1 (2. In any case. iTr A. and will not be needed.143)
is nite and independent of the basis (when A 2 = B1 (H). + iTr A00 + . for this property implies that if jAj 1 = jAj 2 then U jAj 1 = U jAj 2 . Then compute (U jAj U jAj ) = (A A ) = ( A A ) = ( jAj2 ) = (jAj jAj ): Hence U is an isometry on ran(jAj).143).151).147) for all A 2 B(H). which does not depend on the basis. we have
4 k2 4 k A k=k (A A) 1 k (A A) 1
k2 +
X
i
4 ei k2 + =k A k1 + : k (A A) 1
Letting ! 0 then proves (2. One always has the equalities k A k1 = Tr jAj (2.50 If A 2 B1 (H) then
2 ELEMENTARY THEORY OF C ALGEBRAS Tr A :=
X
i
(ei Aei )
(2.
.145) where p jAj := A A: (2.153) and that U U is the projection onto the closure of ran(jAj). U is well de ned.143) by manipulations similar to those establishing the basisindependence of (2.150).146) In particular. when A 0 one simply has k A k1 = Tr A. Lemma 2. we use our old trick: although k B k k B k for all unit vectors . The following decomposition will often be used.107) by 00 Tr + A := Tr A0+ .141). In particular. The corresponding properties for the trace norm are nontrivial (but true). It is easy to see that the HilbertSchmidt norm is indeed a norm.149) for all A B 2 B1 (H) is immediate from (2.16). First de ne U on the range of jAj by U jAj := A . and Tr UAU = Tr A (2. The properties Tr A A = Tr AA (2.151) To prove this. (2.146)) and U is a partial isometry with the same kernel as A. for every > 0 1 there is a 2 H of norm 1 such that k B k2 k B k2 + . Then extend U to the closure of ran(jAj) by continuity. Also.152) p where jAj = A A (cf. Put B = (A A) 4 .144) 2 k A k2 = Tr jAj = Tr A A (2. Completing k (A A ) 1 to a basis fei gi . it can be shown that A 2 B1 (H) when Tr + A < 1.150) kAk k A k2 : (2. and note that 4 k2 =k A k by (2. The same trick with k A k k A k + establishes (2.6 Every operator A 2 B(H) has a polar decomposition A = U jAj (2. where Tr + is de ned in terms of the decomposition (2. For A 2 B1 (H) one has Tr + A = Tr A. One easily veri es that jAj = U A (2.
154) is sometimes seen as the noncommutative analogue of `c (X ) `1 (X ) `2(X ) `0 (X ) `1 (X ) where X is an in nite discrete set.159)
. this analogy is strengthened by the following result. Since A = U jAj 2 jAj 2 by (2.13.1) and (2. Here ^ 2 B0 (H) is identi ed with 2 B1 (H).141) as
k A k1 = k A k2 =
X
i
i X
ai a2 i
(2. Since the star is normcontinuous by 1 1 1 ? (2. so 1 1 1 that jAj 2 is compact by De nition 2. P We rst show that B1 (H)P B0 (H). Since `1 (X ) = `0(X ) and `1 (X ) = `1 (X ) = `0 (X ) . we have 1 pN ( ) k2 =k pN ( )jAjpN ( ) k k pN ( )jAjpN ( ) k1 < k jAj 2
so that jAj 2 p? N ( ) ! jAj 2 in the operatornorm topology. whose proof is based on the fact that B2 (H) is a Hilbert space in the inner product (A B ) := Tr A B: (2. with the same conclusion. for every > 0 we can nd N ( ) such that i>N ( ) (ei jAjei ) < . Lemma 2.2 implies that A 2 B0 (H).13. The proof that B2 (H) B0 (H) is similar: this time we have k jAjpN ( ) k2=k pN ( )jAj2 pN ( ) k k pN ( ) jAj2 pN ( ) k2 <
1 1
so that jAjp? N ( ) ! jAj.13. Using (2. Now pN ( ) jAj 2 obviously has nite rank for every > 0. Let pN ( ) be the projection onto the linear span of all ei .145). This immediately gives
k A k2 k A k1
implying B1 (H) B2 (H).13.17). and A The basic ingredient in the proof is the following lemma.155) (2.1.154) with equalities i H is nitedimensional.144) and (2. Proposition 2. i > N ( ). the claim about proper inclusions is trivially established by producing examples on the basis of 2. Finally.13.152).2.13 The C algebra of compact operators
51
Proposition 2. The chain of inclusions (2. Finally.156)
where the ai are the eigenvalues of jAj.4 to rewrite (2.16) and (2.157) ^ 2 B1 (H) is identi ed with A 2 B(H).2.155).150).9 For 2 B1 (H) and A 2 B(H) one has jTr A j k A k k k1 : (2. Since i (ei jAjei ) < 1. use (2.13. Theorem 2. this implies p? N ( ) jAj 2 ! jAj 2 .13. we use Theorem 2.8 One has B0 (H) = B1 (H) and B1 (H) = B0 (H) = B(H) under the pairing ^( ): ^(A) = Tr A = A (2.158) To show that this is well de ned. Let A 2 B1 (H).7 One has the inclusions Bf (H) B1 (H) B2 (H) B0 (H) B(H) (2.4 and (2.
we estimate 1 1 1 2 j j 2 j = j((AU j j 2 ) j j 2 )j jTr A j2 = jTr AU j j 1
1 2 1 1 2 ) k2 =k k1 Tr (j j 2 U A AU j j 2 ): k j j2 k2 k (AU j j 1 2 Now observe that if 0 A1 A2 then Tr A2 Tr A2 for all A1 A2 2 B0 (H).162) we then have
jTr pj j j = j
N X i=1
(ei j jei )j
k ^k:
It follows that the sequence sN := j N i=1 (ei j jei )j is bounded. 1 1 From (2.161) B0 (H) we
B1 (H).16) and p = p2 = p . choose a nitedimensional projection p. Now use the polar decomposition = U j j with (2.160) (2. By (2. Then use (2.161). so that 2 B1 (H). We now prove B0 (H) = B1 (H). we need to sharpen 2 B2 (H) to 2 B1 (H). hence B0 (H) B1 (H). In view of (2.147) and (2. we conclude that B0 (H) = B1 (H) and k k1=k ^ k.158). by RieszFischer there is an operator 2 B2 (H) such that ^(A) = Tr A for all A 2 B2 (H).143) to be nite in a suitable basis. It is clear from 2. since the sum in (2. Combining this with (2.164)
P
.159).143) is nite.154). Using the original assumption ^ 2 B0 (H) . and since it is positive it must have a limit. We turn to the proof of B1 (H) = B(H).13.155).1) for the inner product (2.151). whereas k p k= 1 in view of (2. pick A 2 H. we use (2.9 that
B1 (H) B0 (H)
with To prove that B0 (H) therefore have
(2. For ^ 2 B0 (H) and A 2 B2 (H)
k ^ k k k1 :
Hence ^ 2 B2 (H) since B2 (H) is a Hilbert space.13.163) ^ 2 B1 (H) and To establish the converse.13.152) for and (2. Hence we have (2.153) to write Tr pj j = Tr pU = Tr pU = ^(pU ) changing the order inside the trace is justi ed by naive arguments.142). so from the above insight we arrive at Tr (j j 2 U A AU j j 2 ) k k1 k AU k2 k A k2
1 1
since U is a partial isometry.162) and (2. since on account
of 2.162) since U is a partial isometry. we have
j ^(A)j k ^ k k A k k ^ k k A k2 :
jTr pj j j k ^ k k pU k k ^ k k p k=k ^ k (2. Now choose a basis of H. and de ne a quadratic form QA on H by ^(j >< j): QA( ) := A (2.160) and (2. with ^k kAk: kA (2. and note that pj j 2 Bf (H) B1 (H) the presence of p even causes the sum in (2. It is clear from 2. and take p to be the projection onto the subspace spanned by the rst N elements
from (2. as well as (2.91) we have j j 2 U A AU j j 2 k AU k2 .144) this means that k k1 k ^ k.52
2 ELEMENTARY THEORY OF C ALGEBRAS
Using (2. To do so.4 one has A1 A2 i all eigenvalues of A1 are all eigenvalues of A2 .143) and (2.9 that B(H) B1 (H) .
168).13 The C algebra of compact operators
53
Here the operator j >< j is de ned by j >< j := ( ) .1.2 and Corollaries 2.169)
. Hence B1 (H) continuity to B1 (H). Applied to the nitedimensional case.12. The state space of the C algebra B0 (H) of all compact operators on some Hilbert space H consists of all density matrices. up to unitary equivalence.160). 2. where a density matrix is an element 2 B1 (H) which is positive ( 0) and has unit trace (Tr = 1).. one has Since A ^(j >< j)j k A ^ k k j >< j k1 : jA Combining (2. evaluating (2.12.168)
^(j >< j) = ( A ). The normalization P condition k ^ k=k k1 = pi = 1 (see 2.166) (2. For example.
P
Proposition 2.2.154.13. i.164) there is an operator A. it shows that the C algebra Mn (C ) of n n matrices has only one irreducible representation.13.6) and (2.112) is zero. The pure state space of B0 (H) consists of all onedimensional projections. so that
j j >< j j = (j >< j) j >< j = (
p
p
)j >< j =k k k k ]:
Since. Corollary 2. In other words. and subsequently by ^ = Tr A. which is positive. Now note that ( A ) = Tr j >< jA this follows by such that A ^(j >< j) = Tr j >< jA. Finally. with (2. with ^k kAk kA
(2. ^ k.5 and (2. j >< j is the projection ].12. we obtain A B(H).
Diagonalize = i pi i ] cf.12. and in general j >< j =k k2 ].7 and 2.13.164).144) we obtain
k j >< j k1=k k k k :
^ 2 B1 (H) by assumption.10.13.6.166). namely the de ning one.167) (2. 3. we nd k A k=k A identi cation of B1 (H) with B(H) is isometric.3 follows from 2.13. The opposite extreme to a pure state on B0 (H) is a faithful state ^. (2.10.155)) yields 2.1. we have ^kk kk k: jQA ( )j k A Hence by Lemma 2. when all pi 0 the operator is positive. Hence A Extending this equation by linearity to the span Bf (H) of all j >< j. the condition ^(A) 0 yields pi 0.10. 2.13.10 1.144) over a basis containing k k. Note that (j >< j) = j >< j. for which by de nition the leftideal N de ned in (2.8.11 The GNSrepresentation
corresponding to a faithful state ^ on B0 (H) is unitarily equivalent to the representation ^ (B0 (H)) on the Hilbert space B2 (H) of HilbertSchmidt operators given by leftmultiplication. The next item 2.e. we have Tr ] = 1. when has unit length.10.
Corollary 2.13.1 j >.165). 2. Hence from (2.2 is then obvious from 2. one has Tr A A > 0 for all A 6= 0. the number Tr p is the dimension of pH (take a basis whose elements lie either in pH or in p?H).
^ (A)B := AB:
(2. Combining (2.165) (2. so that. and (2.3 is one of the most important results in the theory of C algebras.4 and 2. for any projection p.163) and (2. The C algebra B0 (H) possesses only one irreducible representation.13. so that the we obtain B1 (H) = B(H). Conversely.10. Using A = i ].
so that (M)00 = (M00 ): (2. maps B0 (H) into B2 (H). and we may consider the projection p = M ] onto this subspace. and regard 1 + (the direct sum of n copies of H). Hence A = A0 for some A0 2 M.170) so that the representation (2. Hence A 2 M00 commutes with p.6 or 2. Furthermore. use the GNS construction itself. and.9.9.174)
n=0
in (B + inA ) (B + in A ):
(2.171) This follows from (2.3 one has p 2 M0 . and A = Ap = pA 2 M . and its proof contains the main idea of the proof of the in nitedimensional case as well. and form the linear subspace M of H. It is easily seen that 1=2 is cyclic for ^ (B0 (H)) when ^ is faithful. although it is a statement about von Neumann algebras.144) and (2. By Lemma 2. where i lies in the i'th copy. it controls the (ir)reducibility of representations.169) is wellde ned (even for A 2 B(H) rather than merely A 2 B0 (H)). The socalled double commutant theorem was proved by von Neumann in 1929. Since H is nitedimensional.1 Let M be a algebra (and hence a C algebra) in Mn (C ) containing I (here n < 1). The idea of the proof is to take n arbitrary vectors 1 : : : n in C n . We will write H for C n . compute ( B (A)])ij = Bij A]. and if ^ is faithful the closure (in norm derived from the inner product (2.173) n 0 n 0 For arbitrary B 2 M (B(H)). For an alternative proof. Choose some = 1 2 H. Hence A = A0 2 M. _ :::+ _ n . embed M in B(Hn ) ' Mn (B(H)) by A ! (A) := AIn (where In is the unit in Mn (B(H))) this is the diagonal matrix in Mn (B(H)) in which all diagonal entries are A. Using (2. M ! (M). this subspace is closed. Proposition 2.147) and the identity
k AB k2 k A k k B k2
AB = 1 4
3 X
When 2 B1 (H) and 0 then 1=2 2 B2 (H) see (2. We rst give the nitedimensional version of the theorem this is already nontrivial. and remains a central result in operator algebra theory. with the substitutions H ! Hn . when A B 2 B2 (H) one has Tr AB = Tr BA: (2. A ! A := (A). Then M00 = M. Moreover.172)
2. construct a matrix A0 2 M such that A i = A0 i for all i = 1 : : : n.9.14.171) we compute ( 1=2 ^ (A) 1=2 ) = Tr 1=2 ^ (A) 1=2 = Tr A = ^(A): The equivalence between and ^ now follows from 2. with 2 B1 (H). It is easy to see that Mn (M0 )0 = Mn (M00 ). Now use the rst part of the proof. _ :::+ _ n as an element of Hn := n H ' H C n Now choose 1 : : : n 2 H.14 The double commutant theorem
. we therefore have = I 2 M .145).141) that for A 2 B(H) and B 2 B2 (H) one has
(2.5. Hence given 1 + _ :::+ _ n and (A) 2 (M) there exists A 0 2 (M)00 such and ! 1 + that _ :::+ _ n) = A 0 ( 1+ _ :::+ _ n ): (A)( 1 + (2. Hence (M) = M (M ).158)) of the image of this map is B2 (H).54
2 ELEMENTARY THEORY OF C ALGEBRAS It is obvious from (2.158) and (2. For example. The map A ! A 1=2 . given A 2 M00 . so = p . Since I 2 M. Recall that the commutant M0 of a collection M of bounded operators consists of all bounded operators which commute with all elements of M the bicommutant M00 is (M0 )0 .
It is easily veri ed from the de nition of weak convergence that the commutant N0 of a algebra N is always weakly closed: for if A ! A weakly with all A 2 N.14.1.4) shows that normconvergence implies strong convergence. To prove \3 ) 1".1 to the in nitedimensional situation. A) i )j < 8i = 1 : : : ng
A B ] ) = ( AB ) . we de ne two topologies on B(H) which are weaker than the normtopology we have used so far (and whose de nition we repeat for convenience). Theorem 2. and B 2 N. The normtopology is de ned by a single seminorm. As it stands. Since the weak topology is weaker than the strong topology. Hence A 2 M00
. These topologies should all be seen in the light of the general theory of locally convex topological vector spaces. A k! 0. M is closed in the weak operator topology 3. (B
A ) = lim( A B ) . and > 0.14 The double commutant theorem
55
Therefore. The equivalence between the de nitions of convergence stated in 2. (B
A ) = lim( A B ] ) = 0:
If M00 = M then M = N0 for N = M0 .175)
where A 2 B(H) and > 0. which is even a norm. 3. 1 : : : n 2 H.14. Hence (2.177) where A 2 B(H).2 and the topologies de ned by the open sets in question is given in theory of locally convex topological vector spaces. A basis for the weak topology is given by all sets of the form Ow (A 1 : : : n 1 : : : n ) := fB 2 B(H)j j( i (A .2. we consider its closure M . namely the operator norm. we adapt the proof of 2. and 4. M00 = M 2.176) where A 2 B(H).2 The normtopology on B(H) is de ned by the criterion for convergence A ! A i k A .14. Proposition 2.1. In other words. \2 ) 3" is trivial. Using the CauchSchwarz inequality (2. the norm topology is stronger than the strong topology.14. A 0 = (A)0 for some A0 2 M. To describe the appropriate re nement. Hence \1 ) 2". The estimate (2. respectively. A) )j ! 0 for all 2 H. then (
Os (A 1 : : : n ) := fB 2 B(H)j k (B . which in turn is stronger than the weak topology.173) reads A i = A0 i for all i = 1 : : : n. The following are equivalent: 1. These are vector spaces whose topology is de ned by a family fp g of seminorms recall that a seminorm on a vector space V is a function p : V ! R satisfying 2. A basis for the strong topology is given by all sets of the form
(2.1) one sees that strong convergence implies weak convergence.1. so that p = M ]. A) k! 0 for all 2 H.3 Let M be a algebra in B(H).1 is not valid when Mn (C ) is replaced by B(H). The weak topology on B(H) is de ned by the convergence A ! A i j( (A . The strong topology on B(H) is de ned by the convergence A ! A i k (A .14. A k < g
(2. A) i k < 8i = 1 : : : ng (2. A basis for the normtopology is given by all sets of the form
On (A) := fB 2 B(H)j k B . Instead of M . v) ! 0 for all . where dim(H) = 1. and > 0. A net fv g in V converges to v in the topology generated by a given i p (v . whereas the strong and the weak topologies are generated by nite intersections of balls de ned by seminorms of the form ps (A) :=k A k and pw (A) := j( A )j. which may not be closed. containing I. De nition 2. Its open sets are generated by balls in the operator norm. so that M is weakly closed. M is closed in the strong operator topology. 1 : : : n 1 : : : n 2 H.
such that the relative topology of each ber coincides with its topology as a vector space 2. relative to which all axioms for a vector bundle are satis ed.2 A vector bundle is a bundle in which 1. One should verify that a matrixvalued function p 2 C (X Mn (C )) is an idempotent (that is.1. then Mn (C (X )) ' C (X Mn (C )) is a C algebra. X (the base). Let us rst recall the notion of an ordinary vector bundle this is a special case of the following De nition 3. this proves that M00 = M. If X is a compact Hausdor space. p2 = p) i each p(x) is an idempotent in Mn (C ). Since all A 2 M and M is strongly closed. The simplest example of a bundle over a base X with typical ber F is the trivial bundle B = X F .11. A ) k < . There is an integer n m and an idempotent p 2 C (X Mn (C )) such that V X C n . Such an idempotent p de nes a vector bundle Vp . The maps are called local trivializations.4. Note that the dimension of p(x) is independent of x. A ) i k2 < 2 :
1
f
n X i=1
k (A .1. We will generically denote vector bundles by the letter V. Each subset . According to the de nition. De nition 3. this implies that A 2 M.1 (x) is called a ber of B. for every > 0 there is an A 2 M such that k (A . so that M00 M. This bundle leads to possibly nontrivial subbundles. Proposition 3. specialized to A = C (X ) in the proof of 2. with typical ber C m .3 Let V be a complex vector bundle over a connected compact Hausdor space X . with (x f ) := x.1 (x) = p(x)C n . it follows that A ! A for ! 0. each ber is a nitedimensional vector space. because p is continuous and X is connected. With the trivial inclusion M M00 .176)). B ) i k2 < 2 g Os (A
: : : n)
(cf. We factorize = ( F ). The converse is also true.
3 Hilbert C modules and induced representations
This chapter is concerned with the `noncommutative analogue' of a vector bundle. with typical ber F = V . (2. F (the typical ber). and a continuous surjection : P ! X with the following property: each x 2 X has a neighbourhood N such that there is a homeomorphism : .1 (x) ! F (where x 2 N ) is linear. A )( 1 + Noting the inclusion
n X i=1
k (A . The space Vp inherits a topology and a projection (onto the rst coordinate) from X C n . Let X in addition be connected. so that F restricted to . For Hn this means that _ :::+ _ n ) k2 = k (A .56
3 HILBERT C MODULES AND INDUCED REPRESENTATIONS
implies A 2 M in other words.1 A bundle B(X F ) consists of topological spaces B (the total space).1. for some m 2 N . with . One may think of B as X with a copy of F attached at each point. each local trivialization F : .1 (x) := p(x)C n .1 (N ) ! N F X F for which = X (where X : X F ! X is the projection onto the rst factor). The simplest vector bundle over X with ber V = C n is the trivial bundle V = X C n .
3. Recall the de nition of Mn (A) in 2.11.1 Vector bundles
. as follows. whose ber above x is . any bundle is locally trivial in the speci ed sense.1 (x) provides a homeomorphism between the latter and the typical ber F . A complex vector bundle is a vector bundle with typical ber C m .
1. when V = X C n and Vp is the vector bundle described prior to 3.3)
If this number is nite one says that the free module is nitely generated.1)
Since C (X ) is commutative.3. Here a free (right) module E for an algebra A is a direct sum E = nA of a number of copies of
.
R (B )A1
: : : An := A1 B : : : An B:
(3.(Vp ) as a nitely generated projective C (X )module by (3.
De nition 3.(V) is a rightmodule for the commutative C algebra C (X ): one obtains a linear action R of C (X ) on .3.e.(V) the function x ! ( (x) (x))x lies in C (X ). Thus one de nes the space . Conversely. and idempotent p 2 Mn (A).(V) of all continuous sections of V. This is a vector space under pointwise addition and scalar multiplication (recall that each ber of V is a vector space).1. For example.1.3. Following the philosophy of noncommutative geometry.(Vp ) = p n C (X ) (3. When m < 1. a section maps a point in the base space into the ber above the point. In particular. There is a bijective correspondence between complex vector bundles V over X and nitely generated projective modules E (V) = .. Hence the vector space p m A is a right. a leftaction as well. In other words. the action of p on n A commutes with the action by A given by rightmultiplication on each component. it is easily shown that any complex vector bundle over a paracompact space can be equipped with such a Hermitian structure. For any algebra A.2) A fancy way of saying this is that . . of course. we see that .(X C m ) is a nitely generated free module for C (X ). a Hermitian structure is simply given by restricting the natural inner product on each ber C n of X C n to Vp . called projective.4) under the obvious (right) action of C (X ).(V) for C (X ). in the trivial case one has the obvious isomorphisms
' C (X C m ) ' C (X ) C m ' mC (X ): (3. which continuously depends on x. leading to . Let us now add further structure. and thereby a vector bundle Vp as described prior to 3. More precisely. The rst step is the notion of a section of V this is a map : X ! V for which ( (x)) = x for all x 2 X .
Theorem 3.1 (x).3: any vector bundle is of the form Vp . One may then choose the idempotent p 2 C (X C n ) so as to be a projection with respect to the usual involution on C (X C n ) (i.5 A Hermitian vector bundle is a complex vector bundle V with an inner product ( )x de ned on each ber . i. on which A acts by rightmultiplication.(V) as a certain modi cation of a nitely generated free module for C (X ). one has p 2 C (X Mn (C )).e.1.4 Let X be a connected compact Hausdor space. Thus we have achieved our goal of describing vector bundles over X purely in terms of concepts pertinent to the C algebra C (X ). one calls p m A a nitely generated projective module for A.Amodule. one
. for all 2 . This lead to the SerreSwan theorem.1. given such a module p n C (X ). When V is nontrivial.
This is an immediate consequence of Proposition 3.
Using the local triviality of V and the existence of a partition of unity. we now try to describe vector bundles in terms of C algebras.4).1 Vector bundles
57
The essence of the proof is the construction of a complex vector bundle V0 such that V V0 is trivial (where the direct sum is de ned berwise) this is the bundle X C n . this is.(V) by
R (f ) (x) := f (x) (x):
(3.(X
C m)
A itself. one obtains .. Moreover. when X is a connected compact Hausdor space. Describing the bundle as Vp .
Here x v]H is the equivalence class in G H H of (x v) 2 G H .12) for all 2 E and B 2 B. All required properties are easily checked.2 Hilbert C modules
. R maps B linearly into the space of all linear operators on E . and x]H = xH is the equivalence class in G=H of x 2 G. for which we shall write B := R (B ) . This is an example of the strategy of noncommutative geometry. still requiring that all bers have the same dimension (which may be in nite). and satis es R (AB ) = R (B ) R (A)).9) h B iB = h i B B (3. De nition 3. and take a unitary representation U of H on a Hilbert space H . = 0 (3.10) h iB 0 (3. implies that every q 2 G=H has a neighbourhood N .1.1.5) by this action is a Hilbert bundle over X = G=H .11) h iB = 0 .8).2. A Hilbert bundle is de ned by replacing ` nitedimensional vector space' in 3. Any other Hermitian structure on Vp may be shown to be equivalent to this canonical one. The following class of Hilbert bundles will play a central role in the theory of induced group representations. This map is well de ned because of (3. The proof relies on the fact that G is a bundle over G=H with projection (x) = x]H (3. and is a local trivialization of G H H .58
3 HILBERT C MODULES AND INDUCED REPRESENTATIONS
has p = p in addition to p2 = p). This fact. linear in the second and antilinear in the rst entry.7) and typical ber H ..8) This leads to a map : . whose proof we omit.6) and typical ber H . which satis es H (xh) = H (x)h: (3.1 A Hilbert C module over a C algebra B consists of A complex linear space E . and the quotient H := G H H = (G H )=H (3. Note that the projection is well de ned.1 by `Hilbert space'. with projection ( x v]H ) := x]H (3.6 Let H be a closed subgroup of a locally compact group G. so that = ( H ) : . The simplest example of a Hilbert bundle is a Hilbert space. seen as a bundle over the base space consisting of a single point. where 2 E and B 2 B. given by ( x v]H ) := ( x]H U ( H (x))v).1 (N ) ! N H is a di eomorphism. What follows generalizes the notion of a Hilbert bundle in such a way that the commutative C algebra C (X ) is replaced by an arbitrary C algebra B. satisfying h iB = h i B (3.1 U (h)v).
3.1 (N ) ! N H . A rightaction R of B on E (i.2. A Hilbert bundle with nitedimensional bers is evidently the same as a Hermitian vector bundle. A sesquilinear map h iB : E E ! B.e. There is no reason to restrict the dimension of the bers so as to be nitedimensional. Then H acts on G H by h : (x v) ! (xh. Proposition 3.
3.1 (x). so that = 0 by (3. where the supremum is taken over all states ! on B.(H) C (X ) over B = C (X ) for . (2. the B ~ action on E ~ extends One rst completes E ~ valued to a Baction on E . with hA B iA := A B .2. Any Hilbert space H is a Hilbert C module H C in its inner product.(H) is de ned by (3.
2. Then use (2. a norm
h B iB = B h
iB :
(3.86).16) one uses (3.2.. Expanding. 2. In the third example the norm in . one simply takes the quotient of E by its subspace of all null vectors and completes. It is useful to note that (3.13) coincides with the C norm by (2. When all conditions in 3. For (3.9) and (3.12).4 A preHilbert B
~ in the norm (3. where k (x) k= ( (x) (x))x so that it is easily seen that E is complete. and (2. It is easily checked that the required properties hold by continuity. Any C algebra A is a Amodule A A over itself.12).2. The inequality (3.18) is immediate from (3. One checks that (3. for in the inequality h iB 0.16) (3.e.0 (H) the function h iC (X ) is de ned by
h
where the inner product is the one in the ber .3 In a preHilbert B inequalities 1
iC (X ) : x ! ( (x) (x))x
(3. ~ module (and hence in a Hilbert Bmodule) one has the Proposition 3.18)
To prove (3.13) is indeed a norm: k k2 equals p supf!(h iB)g.11).1.16).17) (3.3. and replace by = k k. Let H be a Hilbert bundle H over a compact Hausdor space X .(H) is k k = supx k (x) k. hence h iB = 0 (when B is unital this is follows by taking B = I otherwise one uses an approximate unit in B). obtaining a Hilbert C module in that way.17).8 and (3.17) we substitute h iB . (2. The space of continuous 2 sections E = .8. de ne a preHilbert B ~ module E ~ as in De nition 3.16). Note that the norm (3.18) then allow one to extend the B ~ sesquilinear form on E to a Bvalued one on E .(H) of H is a Hilbert C module . The rightaction of C (X ) on .2
1. ~ B Corollary 3. Since each map ! !(h iB ) is a seminorm (i.2 Hilbert C modules
The space E is complete in the norm
59
except for positive de niteness) by (3.91).1).14)
Example 3. The completeness of B and (3.13).91).16). which is actually positive de nite because of Lemma 2. and write E B. Many Hilbert C modules of interest will be constructed in the following way. The Baction on E is automatically nondegenerate: the property B = 0 for all B 2 B implies that h iB B = 0 for all B .2.2.14). Using (3. ~ .15)
h
k Bk iB h iB k h iB k
k kk B k k k 2 h iB k kk k :
(3.10) imply that
1 k k:=k h iB k 2 : (3. the rst term equals h iB h iB h iB .13) is only a seminorm.
.12). ~ module E ~ can be completed to a Hilbert Bmodule. except ~ B Given a preC algebra B that the nal completeness condition is omitted.1 are met except (3. obtaining E . the supremum is a seminorm.13) We say that E is a Hilbert Bmodule. Recall that a preC algebra is a algebra satisfying all properties of a C algebra except perhaps completeness. so that k k de ned by (3.
so that. A map A : E ! E for which there exists a map A : E ! E such that h A iB = hA iB (3.88).5.19) for all 2 E is called adjointable.2. so that T is bounded. and bounded. one may wonder what C (A A) is.1. When the righthand side vanishes for all it must be that h Z iB = 0 for all . using (3. C (E B) is a Banach algebra.2). so that its satis es (2.6 In a selfdual Hilbert C module E B the C algebra C (E B) coincides with the space L(E )B of all bounded C linear and Blinear maps on E . To prove the nal claim. Under a further assumption (which is by no means always met in our examples) one can completely characterize C (E B).19) of the adjoint that k A k2 k A A k then use Lemma 2. To check (2.18) that k T k k A A k.10) it implies that h iB B = 0.19) that for xed 2 E the map A ! h A iB from C (E B) to B is positive. To prove boundedness of a given adjointable map A.85) and using (2. one has T = h A A iB . Each element A 2 C (E B) satis es the bound
hA A iB k A k2 h iB (3.2.13) on E . for by (3. which plays an important role in the induction theory in 3. (A )B = A( B) for all 2 E and B 2 B).14) this also shows that A 2 C (E B) when A 2 C (E B).2. the map Z ! h Z iB is in C (E B).12). respectively.
. de ned with respect to the norm (3. As a normclosed space of linear maps on a Banach space. for xed 2 E .
The property of C linearity is immediate.11).13) that if An ! A then An converges to some element.5 An adjointable map is automatically C linear. We write C (E B) ! E B.19) then leads to (3. we note that. Hence supfk T k j k k= 1g < 1 by the principle of uniform boundedness (here it is essential that E is complete). it follows from (3. The space C (E B) is normclosed.5 we only need to show that a given map A 2 L(E )B is adjointable. To establish Blinearity one uses (3. Proposition 3.9). the space C (E B) is a C algebra. which by de nition is A . The adjoint of an adjointable map is unique.
Equipped with this involution. x 2 E and de ne T : E ! B by T := hA A iB .2.60
3 HILBERT C MODULES AND INDUCED REPRESENTATIONS
~ (de ned over In Example 3. Blinear (that is. This map is isometric (hence injective). which is precisely A . It is clear from (3. and (3. A Hilbert C module over B is called selfdual when every bounded Blinear map ' : E ! B is of the form '( ) = h iB for some 2 E . since A is adjointable. for xed 2 E de ne 'A : E ! B by 'A (Z ) := h AZ iB . it is almost trivial to see that A and H are the closures of A ~ ) and of a dense subspace D. Uniqueness and involutivity of the adjoint are proved as for Hilbert spaces the former follows from (3.2. Theorem 3. and the map A ! A de nes an involution on the space C (E B) of all adjointable maps on E . In view of Theorem 3.16) one infers from (3. The (de ning) action of C (E B) on E is nondegenerate. A A Hilbert C module E B de nes a certain C algebra C (E B).15). so that Z = 0.72) is easily seen to map A into C (A A). It then follows from (3.11. the latter in addition requires (3.20). Indeed. as one easily veri es from (3. In the context of Example 3.2. Here we used the fact that B = 0 for all and B in the linear span of hE EiB implies B = 0.13) that k A k< 1. On the other hand.16) and (3. The map : A ! B(A) given by (2.20) for all 2 E . (2. hence for = Z .13) and the de nition (3. and with the norm (2. By selfduality this must equal h Z iB for some .19) and (3.1.18) once again.2. one has k T k k A A k k k. Finally. Replacing A by A A in (2.
and call this a dual pair. One may compute this object by taking a faithful nondegenerate representation : A ! B(H) it can be shown that C (A A) is isomorphic to the idealizer of (A) in B(H) (this is the set of all B 2 B(H) for which B (A) 2 (A) for all A 2 A). De nition 3.
In preparation for the imprimitivity theorem.2.22) is obtained by taking (B0 (H))
to be the de ning representation see the paragraph following 2. we introduce the analogue for Hilbert C modules of the C algebra B0 (H) of compact operators on a Hilbert space. Proposition 3.13.19).18) one nds the bound (T B )
3.2 the C algebra C (H C ) coincides with B(H).2.3 The C algebra of a Hilbert C module
k TB k k kk k :
(3.21) follows by taking (C0 (X )) to be the representation on L2 (X ) by multiplication operators (where L2 is de ned by a measure with support X ).2) one obtains
C0 (A A) ' A: A.3. In Example 3. Hence (A) is an ideal in C (A A). and (3. Its subalgebra B0 (H) of compact operators has an analogue in the general setting of Hilbert C modules as well. When E = B = A (see Example 3.1 The C algebra C0 (E B) of \compact"operators on a Hilbert C module E B is the C subalgebra of C (E B) generated by the adjointable maps of the type T B . This is the C algebra most canonically associated to a Hilbert C module.28)
C0 (H C ) = B0 (H)
whence the dual pair B0 (H)
(3. and T B Z := h Z iB : (3. (3. The signi cance of the notation introduced at the end of the de nition will emerge from Theorem 3.2. The word \compact" appears between quotation marks because in general elements of C0 (E B) need not be compact operators.1) one has
This leads to the dual pair A A 2. and also as a matter of independent interest. one veri es without di culty that C0 (E B) is a (closed 2sided) ideal in C (E B).2. C (A A) is the socalled multiplier algebra of A.2. the proof of 2. From (3. where 2 E .4.3 below. one infers that A (B ) = (AB ) for all A B 2 A.16) and (3.2.2. When C0 (E B) has a unit it must coincide with C (E B).29)
H
C.2.23) We write C0 (E B) E B.1.3.5.21) C (B0 (H) B0 (H)) = B(H): (3.
.26) where A 2 C (E B).5 that C0 (E B) acts nondegenerately on E.
(3. one therefore has C (A A) = (A) ' A cf. For E = H and B = C (see Example 3.3.5. When A has a unit.22) Eq. One thus obtains
C (C0 (X ) C0 (X )) = Cb (X ) (3. because every bounded operator has an adjoint. so that it is a C algebra by Theorem 3.27)
One sees from the nal part of the proof of Theorem 3.3 The C algebra of a Hilbert C module
61
Using (3. When A has no unit.2 1. Using the (trivially proved) properties = TB (3.24) B B AT = TA (3.25) B B T A = T A (3.3.
de ned by linear extension of '(T A ) = .13) with B or with A we write k kB and k kA .30)
in combination with the rightaction R (A) := A . where A 2 C0 (E B).10). the map ' from the linear span of all T A to A.26).26) that each L (B ) is adjointable with respect to h iA .33) The fullness of E B and the de nition of C0 (E c A) imply that L : B ! C0 (E c A) is an isomorphism. and (3. it is normpreserving by Lemma 2. Now observe that iB ) = T A (3. and Lemma 3.5.34).9) etc.11). A Hilbert C module E over B is called full when the collection fh iB g. hence = 0 by (3. one sees from Corollary 2.14) and (3. we use (3. and (3.
1 k kB=k T A k 2 :
C0 (E C0 (E B)) ' B:
. where run over E . Hence E c = E . It follows from (3. and that L maps B into C (E c A).30).1 and the fact that the linear span of all T C is Bf (H).35) From (3.6. (3. The space E is equipped with two norms by applying (3.13). in particular. (3.72).7 that '(C0 (A A)) = A.62
3 HILBERT C MODULES AND INDUCED REPRESENTATIONS
One has T A = ( ) see (2. (3. but has the conjugate action of complex scalars. whose norm is majorized by the norm of B in B. It is trivial from its de nition that L is a morphism. The expression h iC0 (E B) := T B (3. The properties (3.31) We call A := C0 (E B) in the references to (3.13) to show that h iA = 0 implies k h i3 iB is positive by (3. The leftaction L (B ) := B of B on E implements the isomorphism
(3. is an isometric morphism as well. the conjugate space E is equal to E as a real vector space. is dense in B.12). Extending ' to C0 (A A) by continuity.7.34) For 2 E we now use (3.16) one nds that L (B ) is a bounded operator on E with respect to k kA .7. as E is complete in k kB by assumption.30) with = . A similar de nition applies to preHilbert C modules.23) with Z = . this implies B k= 0. (3.9). Since h h iB = 0. so that k kA =k kB . (3. Moreover. (3. (3. (3. from E B one obtains E C0 (E B).30) and (3. Eq. It follows from the previous paragraph that L (B ) extends to an operator on E c (denoted by the same symbol).27).12). The completeness of E as a Hilbert Bmodule is equivalent to the completeness of E as a Hilbert Amodule.10). respectively. In particular. Since : A ! B(A) is an isometric morphism. Theorem 3. In other words. The map L is injective because E is nondegenerate as a rightBmodule. It is. To prove (3. From (3.27) with B ! A one then derives the converse inequality to (3.32) L (h for the de nitions in question imply that B = Zh T A Z = h Z iA = TZ iB : (3.3 Let E be a full Hilbert Bmodule. de nes E as a full Hilbert C module over C0 (E B).32) to nd that (3.11) follow from (3. Let E c be the completion of E in k kA we will shortly prove that E c = E .29) follows from De nition 2. injective. Given a complex linear space E . applying (3.13.24).2. and (3. When A has a unit it is obvious that ' is surjective in the nonunital case the existence of an approximate unit implies that the linear span of all is dense in A.27) one derives k kA k kB : (3. the isometric nature of L . and (3.13). below one should substitute A for B when appropriate.14).3.
37)
Moreover.23) with B ! C. one may therefore construct a Hilbert C module. Using B = L (B ) (see 3. denoted by E C.42) This leads to the inclusion C0 (E CP ) L (A).4 Morita equivalence
The re exivity property B M B follows from (3. For later reference we record the remarkable identity
h
which is a restatement of (3.4 Morita equivalence
63
We have now proved (3.13) is easily seen to contain IB . (Remarkably. We write A M B and A E B.30). the null space of (3. the righthand side of (3.23). and (3. proving that A E B implies B E A.4. and (3. We now claim that
L
(A)(
1
B
2
2
(3.
iC0 (E B)Z = h Z iB
(3. this yields h iB = T C .2 Morita equivalence is an equivalence relation in the class of all C algebras. if one looks at (3. the de nition of B .3.36)
Proposition 3.31).28).37) this satis es (3. One T double sequence f i2 i2 g such that N 0 i 2 2
.14). The proof of transitivity is more involved.3).33). and using (3.
The imprimitivity theorem establishes an isomorphism between the respective representation theories of two C algebras that stand in a certain equivalence relation to each other. one picks a Ci i is an approximate unit in B = C (E2 C).39) (3. and E ! E2 .10).31).3.3. but in fact coincides with it. (3.41)
C0 (E C) = L (A):
Using (3. To prove the opposite inclusion. it is easily shown that B B 2 = 1 B h 2 2 h 1 1 iB iB 2 : L (T 1 h 2 2 iB 1 ) 1
Now use the assumption C0 (E2 C) = B as in (3. we eventually obtain
T C1 B 2 1 B 2 = L (T B1 h 2 2 iB 1 ): (3. Finally noticing that as a Hilbert C module over A the space E is full by de nition of C0 (E B). so that in constructing E one only needs to complete E1 B E2 .
3.4. which establishes the dual pair B B B. the space E carries a leftaction L (A): the operator is bounded on E1 ) := (A 1 ) B B E2 and extends to E .38). 1 B 2 ). which carries a rightaction R (C) given by
R
(C )(
1
B
2
) :=
1
B(
2
C ):
(3.41). De nition 3. with B ! C. Symmetry is implied by (3.) Apart from the rightaction R (C).40) (3. we can de ne a sesquilinear map h iC on E1 B E2 by h 1 B 2 1 B 2 iC := h 2 h 1 1 iB 2 iC : (3. the proof of Theorem 3.3 is ready.41) becomes 1 B 2h 2 h 1 1 iB 2 iC .19) with B ! C. (3.38) as de ned on E1 E2 .9) and (3. Substituting this in the righthand side of (3. When A M B and B M C we have the chain of dual pairs A E1 B E2 C: We then form the linear space E1 B E2 (which is the quotient of E1 E2 by the ideal IB generated by all vectors of the form 1 B 2 .10) as explained prior to (3.38) With (3.1 Two C algebras A and B are Moritaequivalent when there exists a full Hilbert C module E B under which A ' C0 (E B).23) with B ! C.
36). In practice the following way to construct dual pairs. Proposition 3. where Mn (C ) and C act on C n in the usual way.4 Suppose one has ~ and B ~ two preC algebras A ~ module E ~ a full preHilbert B ~ on E ~.38) then has limN N i 2 2 C yields
lim N
N X i
T C1 B
i 2
1
= B i 2
L
(T B1 1 ):
Hence L (A) C0 (E C). In the nitedimensional case one has Mn (C ) C n C .4 we rst complete E ~ and 2 E ~ on E ~. Therefore. the matrix algebra Mn (C ) is Moritaequivalent to C . is useful. such that E ~ can be made into a full preHilbert A ~ module with respect a leftaction of A to the rightaction R (A) := A the identity h iA (3. Proposition 3.23). This is immediate from (3.3 The C algebra B0 (H) of compact operators is Moritaequivalent to C .3.4.45) (3. Here is a simple example of this concept. with dual pair B0 (H) H C . implying that A M C. where E is the completion of E ~ to a Hilbert Bmodule E .2. the action of A ~ extends to an action of A on k A k k A k k k for all A 2 A ~ E .34) and its ~.3. Then A M B.23) with (3. By (3. one has Mn (C ) M Mm (C ). In particular.4.42).44) ~ Z = h Z iB ~ ~) relating the two Hilbert C module structures (for all Z2E the bounds
h B B iA ~ hA A iB ~
~ and B 2 B ~. so that the Bcompletion E of E ~ coincides with the Acompletion E c of E ~ that converse for 2 E is.45) the leftaction L (B ) := B extends to an action of B on E c .43)
Since Mn (C ) M C and C M Mm (C ).46)
~ as a Hilbert Bmodule. We leave the details as an exercise. we complete E to a Hilbert Amodule E c by (3. This proves transitivity. and combining both inclusions one nds (3. E c = E . for all A 2 A
k B k2 h k A k2 h
iA ~ iB ~
(3.46).64
3 HILBERT C MODULES AND INDUCED REPRESENTATIONS
P i h i Z i = Z from (3. where Mn m (C ) is the space of complex matrices with n rows and m columns.
. and therefore Morita equivalences.29).
hz wiC = zi wi
(3. and a short computation using (3. one derives (3. one has the dual pair A E C. As in the proof of Theorem 3. This equivalence is implemented by the dual pair Mn (C ) Mn m (C ) Mm (C ). The double Hilbert C module structure is completed by specifying (hz wiMn (C ) )ij = z i wj from which one easily veri es (3. Similarly. which implies Using Corollary 3. with dual pair A E B.
that A ' C0 (E B).5.
3. as explained in Example 3. We will repeatedly use Proposition 3. one nds that k ~ (A) k k A k : (3. which imply that k ~ (A)V A )0 .4 in what follows see 3.1. (3.49)
(3.
g)0 on E by 1.48)
(3.4. (3. which is of interest also in a more general context.117) of course applies here. used with the de nition of k A k for A 2 C (E B).5.g. It follows Since E is a full preHilbert A from (3. the operator ~ (A) may be it follows that ~ is continuous.5 Rie el induction
65
~ module.50) and ~ k2= (A^ (3.51) On the other hand.9. The continuity of ~ follows from (3.1 Suppose one has a Hilbert C module E B.10. into which it is turned after use of this proposition. where it satis es (2. the goal of the Rie el induction procedure described in this section is to construct a representation of C (E B) from a representation of B.3. de ne the sesquilinear form (
(^)0 := ! (h
i B ):
(3.1). Construction 3.1. The representation ~ (C (E B)) is rstly de ned on E =N ~ := V ~A (A)V
Since ! and h iB are positive (cf.1. and (once again) the fullness of E . Its null space is ~ = f 2 E j (^)0 = 0g: N (3. in the proof of Theorem 3. (3. As a corollary.47)
g)0 projects to an inner product (g) on the quotient E =N ~ .50).
.4.3. Since E =N ~ de ned on all of H by continuous extension of (3. The analogue of (2.52) (3.5.4 is a special case of 3.20).2. e. The form ( is the canonical projection. which will be used.47).52) to B. and (3. In particular.3. obtained by choosing E = B = A. The GNSconstruction 2. then by de nition
~^ ~ ) := (^)0 : (V V ~ is the closure of E =N ~ in this inner product.9.50)
~ is dense in H ~ . implies that k A k= supfk ~ (A) k ! 2 S (B)g:
(3.18) in succession.49).44) is inspired by (3.9.3 shows that ~ is faithful (hence normpreserving) when ! is.3. In order to explicate that the induction procedure is a generalization of the GNSconstruction 2. Theorem 3.. this form is positive semide nite. the Aaction on E is injective. one infers a useful property. hence faithful.10. Given a Hilbert Bmodule E .53)
A similar argument combined with Corollary 2.1 one sees that
k A k2 = supfj!(A A)j j ! 2 S (A)g: Applying (3.3 and 3.44). we rst induce from a state ! on B. The Hilbert space H ~ H ~ by 3. each A 2 A automatically satis es (3. If V ~ : E ! E =N ~ 2.19).2. Clearly.5 Rie el induction
To formulate and prove the imprimitivity theorem we need a basic technique. Using (3.9. Given a state ! on B. rather than from a representation . from the proof of Theorem 2.11)).36).3.
(3.58) where I is the unit operator on H the extension in question is possible.10) and the positivity of : B ! B(H ). Hence ( ~ ~ )0 0 by (3. Paraphrasing the comment after the rst version of the construction.117).50) and (3. one then uses (3.5. the text after 2. ~ : E ! E H by Then de ne a linear map U ~ := U : (3. To interrelate the above two formulations. this is bounded by k A k2 (v (h iB )v) . this proves (3. with cyclic vector .57) the map U by (3. The Hilbert space H is the closure of E H =N in this inner product.
Take a faithful state ! on B the condition implies that ~ (A) 0. any subspace of C (E B) may be subjected to the induced representation . 1.
. Since the second factor equals k V ~ k2 . Construction 3.58) according to (3. the sesquilinear form ( )0 is de ned on E H (algebraic tensor product) by sesquilinear extension of ( v w)0 := (v (h iB )w) (3.3 Start from a Hilbert C module E B.2 Let A 2 C (E B) satisfy h A iB 0 for all 2 E .10) to nd ( ~ ~ )0 = (v (h iB )v) with := i i Bi .54). Given a representation (B) on a Hilbert space H . With slight abuse of notation we will write (A) for ( (A)). The situation is depicted in Figure 1.59)
To prove that the form de ned in (3.59).20) and the property k (A) k k A k (cf. the general construction looks as follows. This form is positive semide nite. Also. (3.2).5. With ~ = i i vi and vi = (Bi ) (where is a cyclic vector for P (B)).54). leading to the representation (A) on H . The representation (C (E B)) is then de ned on H by continuous extension of (A)V ~ := V (A I ~ ) (3.54) where v w 2 H . one assumes that is cyclic. de ned by (V ~ V ~ ) := ( ~ ~ )0 (3. the argument below is repeated for each cyclic summand see 2.61) (U ~ therefore quotients to a unitary isomorphism U : H ~ ! H .9. which By (3. 3.60) According to (3.57) and (3.14). it is not di cult to verify that is nondegenerate when is. This particularly applies when one has a given (pre) C algebra A and a homomorphism : A ! C (E B).56) N = f ~ 2 E H j ( ~ ~ )0 = 0 8 ~ 2 E H g:
2. Of course. The null space is N = f ~ 2 E H j ( ~ ~ )0 = 0g: (3. with inner product ( ) .57) where V : E H ! E H =N is the canonical projection.55) As in (2.9. we may equally well write (3.58) duly intertwines ~ and . The form ( )0 projects to an inner product ( ) on the quotient E H =N .116).66
3 HILBERT C MODULES AND INDUCED REPRESENTATIONS
Lemma 3. this map has the property ~ ~ )0 : ~ U ~ )0 = (^ (3. we assume that (B) P is cyclic (if not.49) and (3.1). one veri es that is nondegenerate when and are. Then A 0. one computes k (A)V ~ k2 = (v (hA A iB )v) from (3. and (2.54) is positive semide nite. When one starts from a representation (B) rather than from a state. and (3. because ( ) and h iB are. since k (A) k k A k : (3. is faithful when is. Similarly. Further to an earlier comment.47).
In the opposite direction. When (A) is a representation on a Hilbert space H there exists a representation (B) on a Hilbert space H such that is equivalent to the Rie elinduced representation de ned by (3.1 There is a bijective correspondence between the nondegenerate representations
of Moritaequivalent C algebras A and B.6.
See Figure 2. ~ : E E H ! H by linear extension of We rst de ne U ~ U
v := (h
iB )v:
(3. Conversely. Let the pertinent dual pair be A E B.62)
. and move on to construct (B) from Rie el induction with respect to the dual pair B E A. de ned with respect to some representation (A) and the dual pair B E A. a given representation (B) is equivalent to the Rie elinduced representation . Taking (A) = (A) as just de ned. taking (B) = (B). We then construct a unitary map U : H ! H which intertwines and .3. we pass to the imprimitivity theorem. one has (A) ' (A).6 The imprimitivity theorem
67
A
HH HH HHH H j H

E
B

H
H
?
induction
Figure 1: Rie el induction
A
E
B
@@ @ @ R @ @@ R @ 
H
B
E
A
@
B
H := H
@
H '
Figure 2: Quantum imprimitivity theorem: H ' H and
3. one has (B) ' (B). Starting with (B). relabel this representation as (A).6 The imprimitivity theorem
After this preparation. we construct (A) with Rie el induction from the dual pair A E B. preserving direct sums and irreducibility.
Theorem 3.58) and the above dual pair. This correspondence is as follows.
58).36).
f (x) := f (x. using (3.4.14). To verify that U intertwines and .4. This also proves that the bijective correspondence (B) $ (A) preserves irreducibility: when is irreducible and 2 . and thus arrives at a contradiction.1 x.1 at the group level. one immediately sees that the construction works in the opposite direction as well.67)
(3. all read from right to left. (3.68) It is easy to check that the multiplication is associative as a consequence of the associativity of the product in G. one puts = as above. we use (3. Hence we see from the construction rstly of H = H from E H .66) is seen to be (B )UV ( V v). Hence U (B ) = (B )U for all B 2 B.3. and (3.10. and also in the theory of induced representation as originally formulated for groups by Frobenius and Mackey. this theorem leads to a new proof of Corollary 2.1 = x and (xy). with ! . C (G) consist of all complexvalued functions on G. de ned by linear extension of (cf. 3. Using (3.57) and (3. the righthand side of (3.3.1 = y. whenever ~ 2 N or In particular.58). and secondly of H from E H ~ descends to an isometry U : H ! H . Now put = and H = H . decomposes = 1 2 representation (B) as = 1 .19) and (2. we see that the range of U unitary. the C algebra B featuring in the de nition of a Hilbert C module and in Rie el induction is a socalled group C algebra. This shows that ~ 1 1 v1 U ~ 2 2 v2 ) = (V ( 1 V 1 v1 ) V ( 2 V 2 v2 )) : (U (3. so that U is (B) is nondegenerate. This is made into a algebra by the convolution
3.54) and (3. Combined with Proposition 3. to compute
U (B ) V (
V
v) = (h L (B )
iB )v
(3. then decomposes the induced isn't.1 ).66)
where the leftaction of B 2 B on 2 E is as de ned in 3. A representation of C (G) on a Hilbert space H is de ned as a morphism : C (G) ! B(H).68
3 HILBERT C MODULES AND INDUCED REPRESENTATIONS .2). that U
UV (
V
~ v) := U
v = (h
iB )v:
(3. In many interesting applications.62).63)
~ is indeed C linear. the properties (2. and use (3.13. U V ~ 2 N . the operation de ned by (3.63) is then seen to be equal to (V 1 v1 (h 1 2 iA )V 2 v2 ) .57) from right to left. (3. where ~ 2 E H .5.64) ~ annihilates ~ . (2.23).54).68) is an involution because of the properties (x.65) and (3. since ' . The righthand side of (3.7 Group C algebras
f g(x) :=
and the involution
X
y z2Gjyz=x
f ( y )g (z )
(3.18) with ' ! Note that U and (3. so that C (G) = C n(G) . Thus writing L (B ) = B .1 ):
. one obtains ~ 1 1 v1 U ~ 2 2 v2 ) = ( 1 v1 T B1 2 2 v2 )0 : (U
Now use the assumption A = C0 (E B) to use (3. and subsequently (3. In similar vein.3. Using the proof that the Morita equivalence relation is symmetric (see 3.18) with ' ! . 2 2 leads to It is easy to verify that = 1 = 1 .65)
Using the assumptions that the Hilbert C module E B is full and that the representation ~ and hence of U is dense in H .3). As a vector space.65) from right to left. We start with the de nition of the group algebra C (G) of a nite group with n(G) elements one then usually writes C (G) instead of C (G). with ! .
7. so that cannot be degenerate.
U (x) := ( x )
(3. This assumption is not necessary. Proposition 3.68) are continuous in the L1norm one has k f g k1 k f k1 k g k1 (3.68).5. The Banach space L1 (G) and the Hilbert space L2(G) are de ned with respect to the Haar measure.72) k f k1 = k f k1 : (3. by
f g(x) :=
Z
it is evident that for a nite group this expression specializes to (3. (f ) = 1 (f ) 2 (f ) for all f i U (x) = U1 (x) U2 (x) for all x. The converse follows by choosing f = x .72) is a straightforward generalization of the case G = R cf.3. We also assume that G is unimodular that is.71)
k f g k1=
Z
G
dx j
Z
=
Z
G
dy f (xy. but simpli es most of the formulae.1 There is a bijective correspondence between nondegenerate representations of the algebra C (G) and unitary representations U of G. each left Haar measure is also rightinvariant.10.67).1 that k f k1 =k f k1 .1 )g(y)j
Z
which is (3. Unimodularity implies that the Haar measure isR invariant under inversion x ! x.71) and (3.70) yields ( e ) = I. When U1 (x) = V U2 (x)V for all x 2 G then evidently 1 (f ) = V 2 (f )V for all f 2 C (G). Let now G be an arbitrary locally compact group (such as a nitedimensional Lie group). which preserves unitary equivalence and direct sums (and therefore preserves irreducibility). It is elementary to verify that is indeed a representation of C (G) when U is a unitary representation of G. The involution is given by (3. which therefore is independent of the choice of by Corollary 2. Recall the de nition of a Banach algebra below 2.7 Group C algebras
69
Proposition 3.70)
We can de ne a C norm on C (G) by taking any faithful representation .7. We denote Haar measure by dx it is unique up to normalization.1 )j
G
dx jf (x)j =k f k1 k g k1
. This correspondence is given in one direction by X (f ) := f (x)U (x) (3. The proof of (3. This time we have
G
dy f (xy.1. and putting k f k:=k (f ) k.71) and (3.56).3. one checks that these operations make C c (G) a algebra this time one needs the invariance of the Haar measure at various steps of the proof. It is obvious from invariance of the Haar measure under x ! x. we construct a faithful representation on a Hilbert space.1 .69)
and in the other by
x2G
where x (y) := (xy). so that the involution is certainly continuous. Similarly. When G is compact we choose the normalization so that G dx = 1.1 )g(y)
(3.68) from Cc (G) to L1 (G). (2. initially on Cc (G). In order to equip L1 (G) with a C norm.2 The operations (3. and vice versa.72).
G
dy jg(y)j
Z
G
dy jg(y)j
Z
G
dx jf (xy. Putting x = e in (3.73) Hence L1 (G) is a Banach algebra under the continuous extensions of (3. The convolution product is de ned. Since C (G) is a nitedimensional vector space it is complete in this norm. As in the nite case.
7.77)
Cr (R) ' C0 (R):
(3. Proposition 3.7. the leftregular representation L on L2(R) is Fouriertransformed into the action on L2 (R) by multiplication operators.2.59) maps L1 (G) into a subspace of C0 (R) which separates points on R.7. the deltafunction e is a unit in C (G). One easily veri es that L (f g) = L (f ) L (g ) and L (f ) = L (f ) ..
Introducing the leftregular representation UL of G on L2 (G) by
UL(y) (x) := (y. Lemma 3.94) for all A 2 Cc (G). Eq.7.4.e.4 The Banach algebra L1(G) has an approximate unit I in the sense that (2. Take I = N N .
.1 this basis is partially ordered by inclusion.1 .7.75) (3.78)
It follows from the discussion preceding 2. so that each N is invariant under x ! x. The Lebesgue dominated convergence theorem then leads to (2.5 The reduced group C algebra Cr (G) is the smallest C algebra in B(L2 (G))
k f kr :=k L (f ) k :
Perhaps the simplest example of a reduced group algebra is obtained by taking G = R.7.68) and the invariance of N under R inversion. The linear map L : L1(G) ! B(L2 (G)) is a faithful representation of L1 (G).78) follows from Lemma 2.1 for C algebras). for xed x). When G is nite. For f 2 Cc (G) one therefore has lim I f = f and lim f I = f pointwise (i. satisfying k L (f ) k k f k1 .
2 L2 (G).6 One has the isomorphism (3. and k k=k k1 . but this distribution is not in L1 (G).
L is injective. Pick a basis of neighbourhoods N of e.7. For general locally compact groups one would like to take the Dirac `function' as a unit. Hence k f k1 = 0 by Lemma 3. Hence (3.94) hold for all A 2 L1(G).4. which is the characteristic function of N times a normalization factor ensuring that k I k1 = 1.70
3 HILBERT C MODULES AND INDUCED REPRESENTATIONS
Proposition 3. To prove that L is faithful. so that f = 0 and
To nish the proof of 3. By construction.92) then holds by virtue of (3.74) that L (f ) = 0 implies f
= 0 for all
containing L (Cc (G)). since Cc (G) is dense in L1 (G). In other words. In order to apply Lemma 2.7. theR inequality (2. and hence certainly for = I .7. and therefore for all A 2 L1 (G). One has I f (x) = N N dy f (y. Cr (G) is the closure of the latter in the norm
De nition 3.7.1x)
it follows that
L (f ) =
(3.79) Since the Fourier transform turns convolution into pointwise multiplication.7. we now note from (3.(2. seen as a Banach algebra as in 3. we rst show that L1 (G) possesses the analogue of an approximate unit (see 2.1 x) and f I (x) = N N dy f (xy.6 that the Fourier transform (2. we need to verify that k f kr =k f^ k1 : (3. (2.3 For f 2 L1(G) the operator L(f ) on L2(G).4.3.7 below.92) . It is clear that for every p 2 R there is an f 2 L1(R) for which f^(p) 6= 0.74)
is bounded.3.76)
Z
G
dx f (x)UL (x):
The boundedness of L (f ) then follows from Lemma 3. de ned by
L (f ) := f
:
(3.93) holds as an equality.
we have j( (f ) )j (F F )L2 (G) for all 2 H. de ned as
f^( ) :=
Z
By the same arguments as for G = R.83) is immediate from (3.75) and L in (3.
. Let G irreducible unitary representations U of G such representations are necessarily onedimensional. the analogue of Proposition 3. one obtains ^ ): Cr (G) ' C0 (G (3.84)
Thus reading (3.1 only holds for a limited class of groups. Unfortunately. that is. We explain this notion in a more general context. Alternatively. The generalized Fourier transform f so that G 1 ^ f 2 L (G) is a function on G. de ned by Z (f ) := dx f (x)U (x) (3. one has
p
k
Z
X
d (x) f (x) k
Z
X
d (x) k f (x) k :
(3. The CauchySchwarz inequality then leads to j( (f ) )j k f k1 k k2 . Let us agree that here and in what follows. by its matrix elements
Z
G
dx f (x)( U (x) ):
Since U is unitary.82) as a Bochner integral. We have now found a C algebra which may play the role of C (G) for locally compact groups. (3.8 Let X be a measure space and let B be a Banach space.5 then leads to (3. ^ of ^ is nothing but the set of characters on G. Lemma 3. one may de ne (3. a unitary representation of a topological group is always meant to be continuous. it is understood as a weak integral.7 Group C algebras
71
^ be the set of all This example generalizes to arbitrary locally compact abelian groups. Lemma 2.81) We return to the general case. The following result generalizes the correspondence between UL in (3. The Bochner integral X d (x)f (x) can be manipulated as if it were an ordinary (Lebesgue) integral. A function f : X ! B is Bochnerintegrable with respect to a measure on X i f is weakly measurable (that is.7. where F (x) := k k jf (x)j.82) is most simply de ned weakly.83)
The integral (3.82)
is bounded. in a sense always R obvious from the context.7.82) as a Bochner integral. It will always be directly clear from this whether a given operator.84).76) to arbitrary representations. Hence we need a di erent construction.or vectorvalued integral may be read as a Bochner integral if not.12.7. for each functional ! 2 B the function x ! !(f (x)) is measurable) there is a null set X0 X such that ff (x)jx 2 X nX0g is separable the function de ned by x !k f (x) k is integrable. Then (f ). For example. where G is not necessarily abelian.3.7 Let U be an arbitrary unitary representation of G on a Hilbert space H.83). De nition 3.80)
k (f ) k k f k1 :
( (f ) ) :=
(3. with
G
G
dx f (x)U (x):
(3.
so that it can be extended to all of H by continuity.7.
It is obvious from (2.7.87)
where the sum is over all representations (L1 (G)) of the form (3.85) and (3. 2
De nition 3. and only one representative of each equivalence class of such representations is included. To go in the opposite direction we use the approximate unit once more it follows from (3.7. if is reducible then E (Ix)] = 0 for all x 2 G by the previous paragraph this implies E U (x)] = 0 for all x. and in the other by
(3.9. we infer that k U (x) k 1 for all x.12. This correspondence is given in one direction by (3.83).7.
k (f ) k k f k k f k1 :
.72
3 HILBERT C MODULES AND INDUCED REPRESENTATIONS
Theorem 3. It follows from (3.82) and (3. Given U .111) and (3.85). Conversely. if U is reducible there is a projection E such that E U (x)] = 0 for all x 2 G. which is the same as k U (x). and therefore irreducibility. and (3. in which U is an irreducible unitary representation of G.82) that (f ) E ] = 0 for all f . given U (G) one nds (L1 (G)) satisfying (3. Finally.11 There is a bijective correspondence between nondegenerate representations
has
of the C algebra C (G) and unitary representations U of G. proving that must be nondegenerate.7.
the Banach algebra L1 (G) which satisfy (3.1 are both contractions this is only possible when U (x) is unitary.86) that for any representation (C (G)) and f 2 L1 (G) one
(3.88) that one may extend to a representation of C (G) by continuity.
Theorem 3. Since is continuous.88) Hence the restriction (L1 (G)) satis es (3.9. Since k (Ix) k k Ix k1 = 1.85) (from which the continuity of U is obvious) that U (x) (f ) = lim (Ix) (f ) . hence is reducible. The proof of nondegeneracy makes use of Lemma 3. or from the following argument.1 ) k 1.82).3.1 y). Conversely.2 one puts
k f k:=k u (f ) k
k f k:= sup k (f ) k
(3.7.
The equivalence between the two de nitions follows from (2.139) and Theorem 3.82).2 that any nondegenerate representation of a C algebra is a direct sum of cyclic representations the proof also applies to L1 (G).85)
Recall from 2. one has lim (I ) = I strongly.9 There is a bijective correspondence between nondegenerate representations of
U (x) (f ) := (f x )
where f x(y) := f (x. Hence also k U (x. given by (continuous extension of) (3. Inspired by 2.82) indeed de nes a representation.7.83). and unitary representations U of G.9 it then follows from (3. it follows from easy calculations that (f ) in (3.83) by 3. Hence U (x) = lim (Ix) strongly on a dense domain. The unitarity of each U (x) follows by direct calculation. This correspondence preserves irreducibility. It is bounded by Lemma 3.4.86) where u is the direct sum of all nondegenerate representations of L1(G) which are bounded as in (3. The property U (x)U (y) = U (xy) then follows from (3.9.7.71). The nal claim then follows from Schur's lemma 2.7.1 k 1 We see that U (x) and U (x). This bijection preserves direct sums. Equivalently.85) stands for a cyclic vector of a certain cyclic summand of H. Thus in (3.10.83).82) de nes U on a dense subspace of this summand it will be shown that U is unitary.10 The group C algebra C (G) is the closure of the Banach algebra algebra
L1(G) in the norm
(3. C (G) is the closure of L1 (G) in the norm
This theorem suggests looking at a di erent object from Cr (G).1. and therefore corresponds to U (G) by Theorem 3.
3.89) For general locally compact groups.90) Cr (G) = L (C (G)) ' C (G)= ker( L ): A Lie group group is said to be amenable when the equality Cr (G) = C (G) holds in other words. and an automorphic action of G on A such that for each A 2 A the function from G to A. A C dynamical system (G A ) consists of a locally compact group G.1 x)) G x (f (x.94) x y (A) = xy (A) ( AB ) = ( A ) ( A ) (3.7.91) by (2. de ned by x !k x (A) k. is continuous. We have just seen that all locally compact abelian groups are amenable.8.8 C dynamical systems and crossed products
k f k1 :=
is nite.95) x x x (A ) = (A) (3.7.
Z
dy f (y) y (g(y.99)
f g(x) := f (x) :=
turn L1 (G A ) into a Banach algebra. An automorphism of a C algebra A is an isomorphism between A and A.79) coincide. It follows from the PeterWeyl theorem that all compact groups are amenable as well. and t (f ) : ! (t). one has C (Rn ) ' C0 (Rn ): (3. Proposition 3. where R acts on S by t : ! (t). a C algebra A. It follows from De nitions 2.2 and 3.1 that ( (A)) = (A) for any A 2 A any automorphism hence k (A) k=k A k (3.96) for all x y 2 G and A B 2 A.18) and the uniqueness of the unit. noncompact semisimple Lie groups are not amenable.2 Let (G A ) be a C dynamical system. looking at 3. One A has a unit. one has (3. Hence a general C dynamical system is a noncommutative analogue of a dynamical system. and de ne L1(G A ) as the space of all measurable functions f : G ! A for which
3. one may extend to an automorphism I of the unitization AI by I(A + I) := (A) + I: (3.92) by (2. the second de nition of C (G) stated in 3.1 An automorphic action of a group G on a C algebra A is a group homomorphism x ! x such that each x is an automorphism of A. In particular.5 we see that (3.8.79).8 C dynamical systems and crossed products
73
In conjunction with (3. L (C (G)) is faithful i G is amenable. so that the norms (3.11 implies that for ^ one has (f ) = abelian groups C (G) always coincides with Cr (G). The operations
Z
G
dx k f (x) k
(3.80).98) (3. When A has no unit.97) (3.87) and (3.2.8. one has (I) = I (3. The reason is that for 2 G ^ f ( ) 2 C . In other words. However. The term `dynamical system' comes from the example G = R and A = C0 (S ).1 ) )
.93) De nition 3.
so that k ! ( x (A . and de nes a unitary representation of G on H! .73) one follows the same derivation as for L1 (G). U (x) is a partial isometry.92) or (3. (3.84) and (3.7.102) and (3. B ) (A .102). Hence !( x (A .74
3 HILBERT C MODULES AND INDUCED REPRESENTATIONS
As usual.91). one sees that U (x)H! = ! (A) ! . This correspondence is given in one direction by (f ) =
Z
G
dx ~ (f (x))U (x)
(3. Consider the GNSrepresentation ! (A) on a Hilbert space H! with cyclic vector ! .102)
This operator is well de ned. The integral (3. Proposition 3. we have assumed that G is unimodular with a slight modi cation one may extend these formulae to the nonunimodular case. In order to generalize Theorem 3. where U is a unitary representation of G. for which L1(G A ) = L1 (G). Note that (3. B ) x (A .9.4 describes the way unitary representations of the Poincare group are constructed in algebraic quantum eld theory.9 is as follows. Hence ! ( x (A)) ! = ! ( x (B )).7.97). and ~ is a nondegenerate representation of A which for all x 2 G and A 2 A satis es
U (x)~ (A)U (x) = ~( x (A)):
(3. B )) = 0 by (2. If ! (A) ! = ! (B ) ! then !((A .5 Let (G A ) be a C dynamical system.100)
Here is an elegant and useful method to construct covariant representations. B )) ! k2 = 0 by (2. whereas (3. whose closure is H! because ! is cyclic.3 A covariant representation of a C dynamical system (G A ) consists of a pair (U ~ ).102) with (3.8. B )) = 0 by (3. In any case.8. that not all covariant representations of a C dynamical system arise in this way a given unitary representation U (G) may may not contain the trivial representation as a subrepresentation cf. de ne an operator U (x) on the dense subspace ! (A) ! of H! by
U (x) ! (A) ! := ! ( x (A)) ! :
(3. Theorem 3.7. The completeness of L1 (G A ) is proved as in the case A = C . There is a bijective correspondence between nondegenerate representations of the Banach algebra L1 (G A ) which satisfy (3.8. the generalization of Theorem 3.8. For x 2 G. Furthermore.104)
.101) imply that
(U (x) ! (A) ! U (x) ! (B ) ! ) = ( ! (A) ! ! (B ) ! ): This shows rstly that U (x) is bounded on ! (A) ! . and covariant representations (U (G) ~ (A)). Secondly. Taking A = x. in which ! is then taken to be the vacuum state on the algebra of local observables of the system in question. however.93) implies that
U (x) ! = ! :
(3.120).83). To verify the properties (3. so that it may be extended to H! by continuity. using (3. so that U (x) ! (A) ! = U (x) ! (B ) ! . and suppose one has a state ! on A which is Ginvariant in the sense that
!( x (A)) = !(A)
(3.1 (B ) in (3. Recall (3. we need De nition 3.103)
Proposition 3.72) and (3.94) implies that U (x)U (y) = U (xy). Hence U (x) is unitary. which is unitary from H! to the closure of U (x)H! .101)
for all x 2 G and A 2 A.103). (3.4 Let (G A ) be a C dynamical system.98) is de ned as a Bochner integral the assumptions in De nition 3.120).8 are satis ed as a consequence of the continuity assumption in the de nition of a C dynamical system.101). Note.
The SO(3)action (3. restricted to S 2 .5. one may identify G=H with the unit twosphere S 2 in R3 .6 Let (G A ) be a C dynamical system.
.
Generalizing 3. If G is a Lie group and Q is a manifold we assume that L is smooth. satisfying L(e q) = q and L(x L(y q)) = L(xy q) for all q 2 Q and x y 2 G. in which (U ~ ) is an irreducible covariant representation of (G A ).10. The equivalence between the two de nitions follows from (2.108)
where the sum is over all representations (L1 (G A )) of the form (3.7.8. and only one representative of each equivalence class of such representations is included. We write Lx(q) = xq := L(x q). De nition 3.104) and (3.105).109) is then simply the usual action on R3 . There is a bijective correspondence between nondegenerate representations of the crossed product C (G A ) and covariant representations (U (G) ~(A)).105) ~ (A) (f ) = (Af ) (3. The approximate unit in L1(G A ) is constructed by taking the tensor product of an approximate unit in L1 (G) and an approximate unit in A. we put De nition 3.8.9. at least at a heuristic level. Theorem 3. Here we simply say that a covariant representation (U ~ ) is irreducible when the only bounded operator commuting with all U (x) and ~ (A) is a multiple of the unit. This correspondence preserves direct sums.9 Transformation group C algebras
75
in the other direction one de nes Af : x ! Af (x) and ~x (f ) : y ! x (f (x.7. The crossed product C (G A ) of G and A is the closure of the Banach algebra algebra L1 (G A ) in the norm
k f k:=k u (f ) k
(3. The rest of the proof may then essentially be read o from 3. We assume the reader is familiar with this concept. and therefore irreducibility. where is a cyclic vector for a cyclic summand of (C (G A This bijection preserves direct sums.
3.83).8. where Q is a locally compact Hausdor space.1.3. and x is de ned as follows.139) and Theorem 3. Equivalently. when G = SO(3) and H = SO(2) is the subgroup of rotations around the z axis.9. in which A = C0 (Q).9 Transformation group C algebras
We now come to an important class of crossed products. This action is given by
x y]H := xy]H (3.107)
where u is the direct sum of all nondegenerate representations of L1 (G A ) which are bounded as in (3.7.7 Let (G A ) be a C dynamical system.11.106) ~ )). The main example we shall consider is the canonical action of G on the coset space G=H (where H is a closed subgroup of G). C (G A ) is the closure of L1 (G A ) in the norm
k f k:= sup k (f ) k
(3.6 etc.1 y)). The proof of this theorem is analogous to that of 3. 3.7. and therefore irreducibility. The proof is identical to that of 3.104).1 A (left) action L of a group G on a space Q is a map L : G Q ! Q. For example. If G and Q are locally compact we assume that L is continuous.106).109) where x]H := xH cf.9. and puts U (x) (f ) = (~x (f )) (3. This correspondence is given by (continuous extension of) (3. (3.
we then obtain i Q (f~) Q (~ (3.120) (3.119)
i Q (qk ) Q (ql )] = 0 ~ ~ ~ i Q (T ~ ) Q (T ~ )] = 0
~ ~
(3. which is not de ned on all of H. Firstly. Since Cc (Q) is dense in C0 (Q) in the supnorm. It can be shown that the set HU smooth vectors for U is a dense linear subspace of H. and (3.116) From the commutativity of C0 (Q). Using (3. f~ 2 Cc ~ ) := i~dU (X ) Q ~ (X (3.122)
i Q (T~ ) Q (ql )] = l : k ~ k ~ ~
k
~
l
. ^ X Y ]) ~ ~ i Q (X ~ ) Q~ (f~)] = Q~ ( X f~): ~ ~
(3. It is quite instructive to look at covariant representations (U ~ ) of (G C0 (Q) ) in the special case that G is a Lie group and Q is a manifold. so that one may form the commutative C algebra C0 (Q) cf. where the Gaction is given by L(x q) := q + x.110) Using the fact that G is locally compact.114) Xf dt where Exp : g ! G is the usual exponential map. Hence (G C0 (Q) ) is a C dynamical system.112) makes sense when is a smooth vector for a U this is an element 2 H 1 of for which the map x ! U (x) from G to H is smooth. f ~ k= 0 lim k (f (3. ~).91). Moreover. respectively. it is easy to prove that the continuity of the Gaction on Q implies that ~) . To see this. given by (3.4. given a unitary representation U of a Lie group G on a Hilbert space H one can construct a representation of the Lie algebra g by
When H is in nitedimensional this de nes an unbounded operator.121) (3.1 q): x (f
for all f~ 2 Cc (Q).115) Q~ (f~) := ~ (f~): (3. and that the operator idU (X ) is essentially ! . and may be seen as a generalization of the canonical These equations hold on the domain HU commutation relations of quantum mechanics. The relations (3.112)
1 . one therefore obtains l . Eq. given a Lie group action one de nes a linear map X ! X from g to the space of all vector elds on Q by ~(q) := d f~(Exp(tX )q)jt=0 (3.100).119) then become k ql = k
i Q (X ~ ) Q~ (Y ~ )] = Q~ (.117) .111) x!e x ~) : q ! f~(x.(3.76
3 HILBERT C MODULES AND INDUCED REPRESENTATIONS
Assume that Q is a locally compact Hausdor space. one sees that this function is continuous on all of G. A Gaction on Q leads to an automorphic action of G on C0 (Q). and ~ 2 Rnf0g we put reexpressing it in in nitesimal form. consider the case G = Q = Rn .118) (3.113) Secondly.94) and Hence the function x ! x (f~) from G to C0 (Q) is continuous at e (as e (f~) = f (3. Taking f = q l .117) ~ ~ g)] = 0
~
d U (Exp(tX )) dU (X ) := dt jt=0 :
(3. so that e has a basis of compact neighbourhoods. the l'th coordinate function on Rn . The meaning of the covariance condition (3. 2. (3. on H1 one has selfadjoint on HU U dU (X ) dU (Y )] = dU ( X Y ]): (3. the same is true for f~ 2 C0 (Q).113). (3.100) on the pair (U ~ ) may now be clari ed by 1 (Q). For X 2 g. If X = Tk is the k'th generator of Rn one has k := Tk = @=@q k .
1 x y.1 x. and the inner product h iC := ( )L2 (G) : (3. Since is evidently faithful.128) ~ on E ~ is as de ned in (3. we conclude that the closure of (Cc (G G)) in the operator norm coincides with B0 (L2 (G)). The preHilbert C module Cc (G) C is de ned by the obvious C action on and E Cc (G).99) read
G
dx sup jf (x q)j
q 2Q
(3.9 Transformation group C algebras
77
~k ) with the quantum position and momentum observables.110). It is easily veri ed that the HilbertSchmidtnorm (2.123) (3.3 Let a locally compact group G act on Q = G by L(x y) := xy.
G G
G
dy f (xy. regarded as a dense subalgebra of C (G G).124) and (3.125)
Z
As always.1 x)j2 : (3. (It should be remarked that Q~ (qk ) is an unbounded operator.1 y): It is not necessary to consider the bounds (3.129) h iCc (G G) := (y) (x.) De nition 3. We start from Cc (G G). Proposition 3.1. G is here assumed to be unimodular. the Gaction L on Q is not indicated in the notation C (G Q). we conclude from (2. Clearly.45) and (3.1 q):
G
dy f (y q)g(y. To prove this. Hence one may identify Q~ (qk ) and Q~ (T respectively. This. one shows directly that there is a bijective correspondence between the representations of Cc (G G) and of C . The transformation group C algebra C (G Q) is the crossed product C (G C0 (Q) ) de ned by the automorphic action (3.13.127) 2 Since this is clearly nite for f 2 Cc (G G).141) of (f ) is Z Z k (f ) k2 = dx dy jf (xy. (3. Here is a simple example. L1(G C0 (Q)) is then identi ed with the space of all such functions f for which
k f k1 =
f g(x q) =
Z
is nite cf.120) .1 even Bf (L2 (G)) is dense in B0 (L2 (G))). follows from the previous paragraph if we can show that the norm de ned by (3.154) that (Cc (G G)) B0 (L2 (G)). and therefore the isomorphism C (G G) ' B0 (L2 (G)).98) and (3. ~ = Cc (G G). although the construction clearly depends on it. Following the proof of Theorem 3. we proceed as in Proposition 3.4.6. in which we take A ~ = Cc (G). Since (Cc (G G)) is dense in B2 (L2 (G)) in the HilbertSchmidtnorm (which is a standard fact of Hilbert space theory).(3. is the case if all irreducible representations of the algebra Cc (G G) are unitarily equivalent to . in turn.46).126)
.126). the equality (C (G G)) = B0 (L2 (G)). the operations (3.9.108) coincides with the operator norm of ( ).3. so that is a representation of the algebra Cc (G G).125) that (f ) (g) = (f g) and (f ) = (f ) .122) are valid. Then C (G G) ' B0 (L2 (G)) as C algebras. In this realization. but one may show ~k ) always possess a from the representation theory of the Heisenberg group that Q~ (qk ) and Q~ (T common dense domain on which (3.97). whereas the Cc (G G)valued inner product The leftaction of A ~ on Cc (G) is given by (3.9.2 Let L be a continuous action of a locally compact group on a locally compact space Q. We de ne a linear map : Cc (G G) ! B(L2 (G)) by (f ) (x) :=
f (x q) = f (x. and B2 (L2 (G)) is dense in B0 (L2 (G)) in the usual operator norm (since by De nition 2. One may identify L1 (G C0 (Q)) with a subspace of the space of all (measurable) functions from G Q to C an element f of the latter de nes F 2 L1 (G C0 (Q)) by F (x) = f (x ).4.124) (3.1 x) (y):
(3. Conventionally. B ~ = C.1 q)
Z
One veri es from (3.
11).135). E We make a preHilbert Cc (H )module Cc (G) Cc (H ) by means of the rightaction = f :x! R (f )
Z
Here f 2 Cc (H ) and 2 Cc (G).10. we nd
G G=H H
! (h
iCc(H ) ) =
Z
G=H
d (q) k
Z
H
dh (s(q)h)U (h)
k2 :
(3. and any measurable map s : G=H ! G for which s = id (where : G ! G=H is the canonical projection (x) := x]H = xH ). given by UR (h) (x) := (xh): (3.4 we take ~ = Cc (G G=H ).4. In 3.4.132) Namely. satisfying Z Z Z dx f (x) = d (q) dh f (s(q)h) (3. seen as a dense subalgebra of A = C (G G=H ) as explained prior to A (3.4. This proves (3.10.131).10 The abstract transitive imprimitivity theorem
We specialize to the case where Q = G=H .109).3. This will be done on the basis of Proposition 3.135)
where U is the unitary representation of H corresponding to (C (H )) see Theorem 3.68) and (3.7.1 ) (3. with corresponding unit vector 2 H . Condition
. To prove (3.123) ~ = Cc (H ). (3. Combining (3. and the Gaction on G=H is given by (3.71). The Cc (H )valued inner product on Cc (G) is de ned by
H
dh (xh. one has Z dh f (h)U (h. For simplicity we assume that both G and H are unimodular. so that h iCc(H ) is positive in C (H ) by Corollary 2. We need to construct a full Hilbert C module E C (H ) for which C0 (E C (H )) is isomorphic to C (G G=H ).10) are easily veri ed from (3. This leads to the transformation group C algebra C (G G=H ). seen as a dense subalgebra of B = C (H ) B ~ = Cc (G).136).78
3 HILBERT C MODULES AND INDUCED REPRESENTATIONS
3.9) and (3.130)
h
iCc (H ) : h !
Z
Interestingly.134)
The properties (3.11). Theorem 3.137)
Since this is positive.1 )f (h): dx (x) (xh):
(3. where H is a closed subgroup of G. respectively. we take a vector state ! on C (H ). and (3.133) R (f ) = which should be compared with (3.82). and
H
G
(3. both formulae may be written in terms of the rightregular representation UR of H on L2 (G). Hence for f 2 Cc (H ) L1 (H ) one has
! (f ) = (
(f ) ) =
Z
H
dh f (h)(
U (h ) )
(3.11 (with G ! H ).131)
h
iCc (H ) : h ! ( U (h) )L2 (G):
(3.136) for any f 2 Cc (G). We note that the Haar measure on G and the one on H de ne a unique measure on G=H . this proves that (h iCc (H ) ) is positive for all representations of C (H ).1 The transformation group C algebra C (G G=H ) is Moritaequivalent to C (H ).
in that (U ( ) (x) ( ) (q)) = q see (3.124) and (3. Math. ~ = Cc (G G=H ) on Cc (G) and the Cc (G G=H )valued We now come to the leftaction L of A inner product h iCc (G G=H ) on Cc (G).
Using (3. In order to explain the idea of an induced group representation from a geometric point of view.4.131). There is a natural inner product on the space of sections . and that Cc (G) Cc (G G=H ) is a preHilbert C module with respect to the rightaction of Cc (G G=H ) given by := L (f ) cf. the action (3.140) Since the leftaction x : y ! xy of G on itself commutes with the rightaction h : y ! yh of H on G. but an argument that is su ciently elementary for inclusion in these notes does not seem to exist.1 yh):
the crucial condition (3.136).138) (3. This implies that the function (q h) ! (s(q)h) vanishes almost everywhere for arbitrary sections s. When the measure is Ginvariant (which is the case. The group G acts on the Hilbert bundle H de ned by (3. we return to Proposition 3.138). one needs to show that the Hilbert C modules Cc (G) Cc (H ) and Cc (G) Cc (G G=H ) are full. and (3. this implies that = 0. 3. To complete the proof. and that the bounds (3. using (3. Since one may choose s so as to be piecewise continuous.10.46) are satis ed.142)
where is the measure on G=H de ned by (3.11 Induced group representations
U ( ) (x)
( )
(q) := U (x)
( )
(x. one may check that L is indeed a leftaction. and ( ) is the inner product in the ber . Induced representations of C algebras.139)
H
dh (yh) (x. Note that di erent identi cations of the ber with H lead to the same inner product.45) and (3. it is easy to verify R (f )
The theory of induced group representations provides a mechanism for constructing a unitary representation of a locally compact group G from a unitary representation of some closed subgroup H .1 x]H ) (y)
Z
(3.4.1 (q) ' H .6). (3.12) easily follows from (3.137) as well. The Gaction U on the vector bundle H induces a natural Gaction U ( ) on the space of continuous sections .125).A.(H ). This is indeed the case.130). and 2 Cc (G).143)
. Adv. The Hilbert space L2 (H ) is the completion of the space .(H ) by
3. for example. This section is evidently continuous. given by (
( ) ( )
) :=
Z
G=H
d (q) (
( )
(q )
( )
(q))
(3.6. These are given by
L (f ) (x) =
Z
h
iCc (G G=H ) : (x y]H ) !
G
dy f (xy.44).139). since h iCc(H ) = 0 implies that the righthand side of (3. 13 (1974) 176257.1 q)):
(3.(H ) of H .141) satis es (U ( ) (x)
( )
U ( ) (x)
( )
)=(
( )
( )
):
(3.11 Induced group representations
79
(3. in the sense that it gives a necessary and su cient criterion for a unitary representation to be induced.141)
One should check that U ( ) (x) ( ) is again a section.1 then turns out to be equivalent to a complete characterization of induced group representations. the operator U ( ) (x) de ned by (3.5) by means of U (x) : y v]H ! xy v]H : (3.137) vanishes for all . when G and H are unimodular).c(H ) of continuous sections of H with compact support (in the norm derived from this inner product). Enthusiastic readers may nd the proof in M.3.140) is clearly well de ned.1. Rie el. Theorem 3. since the Gaction on G=H is continuous. de ned on ( ) 2 . (3. Also.
11.1 ) have the same null sets for all x 2 G. The section ( ) thus de ned is independent of the choice of x 2 .146).
Proposition 3.1 q)) (q) := d (dx (q( ) U (x )
s
( )
(x. Given 3. it can be shown that it is still quasiinvariant in the sense that ( ) and (x.2 In the context of 3.c(H ) is given by ( ) :=
Z
G=H
d ( (x)) ( (x)
(x))
(3.11.c (H ) in the inner product (3.11. When is Ginvariant. The property U ( ) (x)U ( ) (y) = U ( ) (xy) is easily checked. We now give a more convenient unitarily equivalent realization of this covariant representation.c (H ). where is given by (3. so that it is bounded.146). For this purpose we note that a section ( ) : Q ! H of the bundle H may alternatively be represented as a map : G ! H which is H equivariant in that (xh.c (H ).1. it is therefore a unitary operator.1 ).136).1. De ne the Hilbert space L2 (H ) of L2 sections
of the Hilbert bundle H as the completion of . For ( ) to lie in . In this realization the inner product on .144) with (3.140) de nes a unitary representation of G on 2 L (H ).144)
Proposition 3.147)
by
( )
( (x)) = x
(x)]H
where : G ! G=H is given by (3.146).3 The Hilbert space H is the completion in the inner product (3. the RadonNikodym derivative q ! d (x.146) (3.1 Let G be a locally compact group with closed subgroup H . The representation U ( ) (G) is said to be induced by U (H ). the expression (3.145) The pair (U ( ) (G) ~( ) (C0 (G=H ))) is a covariant representation of the C dynamical system (G C0 (G=H ) ).142).1 ( (x)) because of (3.11.148) of the set of continuous functions : G ! H which satisfy the equivariance condition (3.141). One then modi es (3.1 (q))=d (q) exists as a measurable function on G=H . the projection of the support of from G to G=H must be compact. where the measure on G=H is de ned by (3.80
3 HILBERT C MODULES AND INDUCED REPRESENTATIONS
When fails to be Ginvariant. with inverse U ( ) (x. and can be extended to L2 (H ) by continuity.145) is nothing but the rightaction (3.110).143).
projection of whose support to G=H is compact.7). and let U be a unitary representation of H on a Hilbert space H . guaranteeing the property (3.141) to
U ( ) (x)
( )
.144) simpli es to (3. Since U ( ) (x) is invertible.
One easily veri es that the squareroot precisely compensates for the lack of Ginvariance of .
De nition 3. and the
. this follows from a simple computation. Hence U ( ) (x) is isometric on . Consequently.1) of (C0 (G=H )) on L2(H ) this rightaction is at the same time a leftaction.
(3. de ne a representation ~( )(C0 (G=H )) on L2(H )
~(q) ( ) (q): ~ ( ) (f~) ( ) (q) := f (3.1 q)):
(3.148)
the integrand indeed only depends on x through (x) because of (3.11. because (C0 (G=H )) is commutative.1 ) = U (h) (x): Such a map de nes a section
( )
by
Note that the representation (3. The map x ! U ( ) (x) given by (3.
154) for all y 2 G. de ne a representation ~ (C0 (G=H )) on H (cf. This pair is unitarily equivalent to the pair (U ( ) (G) ~( ) (C0 (G=H ))) by the unitary map V : H ! H( ) given by V ( (x)) := x (x)]H (3. and (3.1 y (x.3) by (3.152).110).1 y)]H = y (x.1 x):
(3. we de ne the induced Gaction U on by y U (x) (y)]H := U ( ) (x) ( ) ( (y)): (3.151)
Similarly. it should be obvious from the argument leading from (3.
3.151).156) ~ 2 C0 (G=H here f ~x(q) := f~(x. (3. Comparing (3. any covariant representation (U ~ ) of (G C0 (G=H ) ) is unitarily equivalent to a pair of this form. in the realization H the representation (3.11. Theorem 3.1 = U ( ) (y) (3. (3. we obtain U ( ) (x) ( ) ( (y)) = U (x) ( ) (x.11.1 y)]H : Hence we infer from (3. we then have Proposition 3.155).1 x)) U (y) (x) = d d( ((y(x))
s
(y.4 In the context of 3.1 q).147).12.152).1 and 3. In the preceding section we have seen that the unitary representation U (G) induced by a unitary representation U of a closed subgroup H G can be extended to a covariant representation (U (G) ~ (C0 (G=H )).11.1 = ~ ( ) (f~) ~ for all f 2 C0 (G=H ). where is given by (3.7)). The original imprimitivity theorem of Mackey. (3.140). Conversely.149) that U (y) (x) = (y.151) that (3.1 (y))) = U (x) x.12 Mackey's transitive imprimitivity theorem
81
Given (3.3.144) in the above derivation yields
.151) and (3. for all x 2 G and f Any unitary representation U (H ) leads to a covariant representation (U (G) ~ (C0 (G=H )) of (G C0 (G=H ) ). is a covariant representation of the C dynamical system (G C0 (G=H ) ). states that all covariant pairs (U (G) ~ (C0 (G=H )) arise in this way.1 x): (3. and consider the C dynamical system (G C0 (G=H ) ).2. 3. Recall (cf. and (3.141).12 Mackey's transitive imprimitivity theorem
.1.149) Using (3. given by 3.153) in the sense that V U (y)V . This leads to a bijective correspondence between the space of equivalence classes of unitary representations of H and the space of equivalence classes of covariant representations (U ~ ) of the C dynamical system (G C0 (G=H ) ). which preserves direct sums and therefore irreducibility (here the equivalence relation is unitary equivalence).145) reads ~ (f~) (x) := f~( x]H ) (x): (3.3. An analogous but simpler calculation shows (3.147). as well as the de nition x (y) = x y]H = xy]H = (xy) of the Gaction on G=H (cf. 3.110).152) Analogous to 3.8.1 = ~ (f~x ) (3. The pair (U (G) ~ (C0 (G=H ))). satisfying the covariance condition U (x)~ (f~)U (x).11.11.141) by (3.154) holds.147).6.153) with (3.155) V ~ (f~)V .1 Let G be a locally compact group with closed subgroup H . which historically preceded Theorems 3.3) that a covariant representation of this system consists of a unitary representation U (G) and a representation ~ (C0 (G=H )).150) Replacing (3.1. where U is given by (3.149) to (3. where is given by (3.10.
where the norm on the lefthand side is in H ~ .1.159) and (3. which relate the representations of C (H ) and C (G G=H ). with Theorems 3.146) is indeed satis ed by the lefthand side. by H . and the norm on the righthand kV ~ obtained by Rie elinducing side is the one de ned in (3. one indeed has U It is obvious that N = ker(U U.138). one shows that ~ k k k.5. (3.158)
~ ).162)
G
. Let us start with a technical point concerning Rie el induction in general. Using (3. The Hilbert space H de ned below (3.131). The explicit form of the correspondence remains to be established. When Then U the image is dense this map extends to a unitary isomorphism U : H ! H . with G ! H . and (3. ~ quotients to an isometric map between E H =N and the image of U ~ in H . so that. using the invariance of the Haar measure on H .136).132). (3.10.160) (3. Otherwise. the representation (C (E B)) is equivalent to the representation (C (E B)).157)
for all ~ ~ 2 E H .54).157). one nds that the induced representation of C (G G=H ) on H is given by Z (f ) (x) = dy f (xy. comparing with (3.2 Suppose one has a Hilbert space H (with inner product denoted by ( ) ) and a ~ : E H ! H satisfying linear map U ~~ U ~ ~ ) = ( ~ ~ )0 (U (3. (3.8. one obtains (
v
~ (U
w )0 =
~ vU
Z
H
dh ( UR (h) )L2 (G) (v U (h)w) =
Z
H
dh
Z
G
dx (x) (xh)(v U (h)w)
cf. ~ : Cc (G) H ! H de ned by linear extension of Consider the map U ~ U
v(x) :=
Z
H
dh (xh)U (h)v:
(3. We will use a gerenal technique that is often useful in problems involving Rie el induction. It is clear that U desired unitary map U : H ! H . On the other hand.1 x]H ) (y) (3. Using (3.161)
w)H =
Z
H
dh
Z
G=H
d ( (x))
Z
H
dk (xk) (xh)(U (k)v U (h)w) :
Shifting h ! kh. and will therefore be denoted by this symbol.10.13).58).2 one obtains the (3.
=
We use this lemma in the following way. we continue to denote the Hilbert space H de ned in Construction 3.82). and (3.1 and 3. and using (3. applies to H .47).148) one has
(3. de ned by continuous extension of ~ ~ := U ~ (A I ~ ): (A)U (3. as follows from the invariance of the Haar measure. unitary between H and the closure of the image of U In any case.2. so by Proposition 3.7.12. It follows that the induced space H from a preHilbert C module is the same as the induced space constructed from its completion.7.57). of course. however. one veri es ~ (Cc (G) H ) is dense in H . will play the role H in 3.13).159)
Note that the equivariance condition (3. Using (3. The same comment.12.11 and 3. Lemma 3. U is ~.158) and (3.82
3 HILBERT C MODULES AND INDUCED REPRESENTATIONS
The existence of the bijective correspondence with the stated properties follows by combining Theorems 3. which allow one to pass from (C (H )) to U (H ) and from (C (G G=H )) to (U (G) ~ (C0 (G=H )).148).3. To avoid notational confusion. starting from the preHilbert C module Cc (G) Cc (H ) de ned in the proof of 3.12.1. respectively. from (3.6.
This time the pairing between states and observables is given by h Ai = Tr A: (4. mixed states are identi ed with probability measures on S .1 see Theorem 2. bounded. the state space has a convex structure. which assigns the expectation value h! f i to a state ! and an observable f . consisting of the observables a pairing h i : S AR ! R 1. Hence the space AR of observables may be taken to be C 1 (S R). interpreted as the state space a real vector space AR. We may. However. In the same sense. There is a pairing h i : S AR ! R 1 between the state space S of probability measures on S and the space AR of observables f . then the space of all probability measures on S is precisely the state space of A = C0 (S ) in the sense of De nition 2. In general.162) satis es the equivariance condition (3. and we take AR = C0 (S R). The observables of the theory are functions on S one could consider smooth.146). with the di erence that in (3.8. continuous. etc.162) corresponds to the covariant representation (U (G) ~ (C0 (G=H )) by the correspondence (3. or L1 (S R). measurable. In these examples.138).162) lies in H . or some other other class of realvaued functions. but this is not our concern here. For example. C0 (S R). In addition.2) In elementary quantum mechanics the state space consists of all density matrices on some Hilbert space H the pure states are identi ed with unit vectors .(3.2. this expectation value will be unsharp. This pairing is given by
4.8. or all compact selfadjoint operators. whereas the set of observables is a real vector space (barring problems with the addition of unbounded operators on a Hilbert space).4) A key di erence between classical and quantum mechanics is that even in pure states expectation values are generally unsharp. More generally. one should specify the dynamics of the theory. it is a simple exercise the verify that the representation (C (G G=H )) de ned by (3. whereas in (3. etc.1)
The physical interpretation of this pairing is that in a state the observable f has expectation value h f i.83 this looks just like (3.138) lies in Cc (G). and is an eigenvector of A. The only exception is when an observable A has discrete spectrum. one should check that the function (f ) de ned by (3. The observables are taken to be either all unbounded selfadjoint operators A on H.1 The mathematical structure of classical and quantum mechanics
4 Applications to quantum mechanics
h f i := (f ) = d ( ) f ( ):
S
Z
(4. Indeed. Cb (S R).106) of Theorem 3. if S is a locally compact Hausdor space. say that a physical theory consists of a convex set S . or all bounded selfadjoint operators. The situation is quite neat if S and AR stand in some duality relation. in a pure state (seen as the Dirac measure on S ) the observable f has sharp expectation value (f ) = f ( ): (4. therefore. Finally.5.3) In a pure state one has h Ai = ( A ): (4.8.
In classical mechanics one starts from a phase space S . in the classical case. in that h f i2 6= h f 2 i. whose points are interpreted as the pure states of the system. in quantum mechanics the space of all density matrices on H is the state space of the C algebra B0 (H) of all compact operators on H
.104) .
or C0 (S R).2. It is desirable (though not strictly necessary) that Q preserves positivity. and L(H) denotes some space of selfadjoint operators on H. A positiveoperatorvalued measure or POVM on X in a Hilbert space H is a map ! A( ) fromP to B(H)+ (the set of positive operators on H).6.1. On the other hand.1. in that
4. and POVM's ! A( ) on S in H. with Borel structure . The physical interpretation of quantum mechanics is a delicate matter. De nition 4.8. given by Z Q(f ) = dA(x) f (x): (4. It follows from 2.
. satisfying A( ) = 0. then the space of observables is the dual of the (linear space spanned by the) state space. In the C algebraic approach to quantum mechanics. This notion is familiar from the spectral theorem. This means that one has a classical phase space S and a linear 0 1 map Q : A0 R ! L(H).5 that a positive map automatically preserves selfadjointness. and A( i i ) = i A( i ) for any countable collection of disjoint i 2 (where the in nite sum is taken in the weak operator topology). There is a bijective correspondence between quantizations Q : C0 (X ) ! B(H). A PVM is usually written as ! E ( ) it follows that each E ( ) is a projection (take 1 = 2 in the de nition).5)
for all f 2 C0 (X ) this implies that f 2 C0 (X R) is mapped into a selfadjoint operator. etc. if we take AR to be the space B(H)R of all bounded selfadjoint operators on H. The state space of AR is then given by De nition 2. with the same choice of the state space. Given the physical meaning of a classical observable f .2 Quantization
Q(f ) = Q(f )
(4. Ideally. A quantization of X consists of a Hilbert space H and a positive map Q : C0 (X ) ! B(H).1 Let X be a locally compact Hausdor space.3 Let X be a locally compact Hausdor space. When X is compact it is required that Q(1X ) = I. This general setting allows for the existence of superselection rules. where AR stands for C (S R).13. A projectionvalued measure or PVM is a POVM which in addition satis es A( 1 \ 2 ) = A( 1 )A( 2 ) for all 1 2 2 .10. Also recall De nition 2. one needs to specify the physical meaning of any observable A 2 AR. such as B0 (H)R or B(H)R. We will not go into this generalization of elementary quantum mechanics here.6)
The map Q is a representation of C0 (X ) i
S
! A( ) is a PVM. a general quantum system is speci ed by some C algebra A. whose selfadjoint elements in AR correspond to the observables of the theory. and concentrate on the choice A = B(H).1.6. This provides the physical meaning of al least all operators in the image of Q.84
4 APPLICATIONS TO QUANTUM MECHANICS
see Corollary 2. which choice discards what happens at in nity on S . A(X ) = I.2. with the intrinsic notion of positivity given by 2.8.2. We are thus led to the following De nition 4. Here C0 (X ) and B(H) are. Note that the above conditions force 0 A( ) I. Proposition 4. It is quite convenient to assume that A0 R = C0 (S R). regarded as C algebras.2 Let X be a set with a algebra of subsets of X .8. In practice. rather then vice versa see Theorem 2. a given quantum system arises from a classical system by `quantization'.4 of a positive map. one then ascribes the same physical interpretation to the corresponding quantum observable Q(f ).13. and when X is noncompact one demands that Q can be extended to the unitization C0 (X )I by a unitpreserving positive map. of course. There is an interesting reformulation of the notion of a quantization in the above sense. as well as the (approximate) unit.
By Proposition 4.9) 2) = ( 1 2 ): for all 1 2 2 H. this functional has the same properties. a quantization Q : C0 (X ) ! B(H) is a completely positive map.2. the system is localized in Q is p ( ) := Tr E ( ): (4.3. The are called coherent states for S . for arbitrary 2 H one constructs a functional ^ on C (X ) by ^ (f ) := ( Q(f ) ) Since Q is linear and positive. and De nition 4.9) guarantees that we may de ne a POVM on S in H by
S
4.3 Stinespring's theorem and coherent states
85
The precise meaning of (4. The ensuing map ! A( ) is easily checked to have the properties required of a POVM.
Combine Theorem 2.11.2.4. Given Q. and consider an embedding ! of S into some Hilbert space H. in view of 2. There exist a Hilbert space H .2.11)
.
! A( ) is contained in the statement that the number
When X is the phase space S of a physical system.2. The positive map Q corresponding to the POVM ! A( ) by Proposition 4. Given the assumptions.3.7)
is the probability that. Let S be a locally compact Hausdor space (interpreted as a classical phase space). the system in question is localized in S. for each pair 2 H a POVM ! A( ) in H de nes a signed measure on X by means of ( ) := ( A( ) ). For 2 one then puts ( A( ) ) := ( ).11. Moreover.
S
Z
]:
(4. Conversely.6) is expressed by this equation.2 are satis ed. Hence the Riesz representation theorem yields a probability measure on X . This yields a positive map Q : C (X ) ! B(H) by R ( Q(f ) ) := X d (x) f (x) the meaning of (4. in a state .6) will emerge shortly.8) By Proposition 2.3 is given by
Q(f ) = d ( ) f ( )
In particular.
Corollary 4. one has Q(1S ) = I. the situation is therefore described by a PVM ! E ( ) on Q in H. Condition (4. in a state . The probability that.11.3 Stinespring's theorem and coherent states
A( ) =
Z
d ( )
]
(4.1 = pE ( )p for all 2 . a projection p on H .2.10)
where ] is the projection onto the onedimensional subspace spanned by (in Dirac's notation one would have ] = j >< j).7 and 2.4 Let ! A( ) be a POVM on a locally compact Hausdor space X in a Hilbert space H . one veri es that the property E ( )2 = E ( ) is equivalent to Q(fg) = Q(f )Q(g). de ning an operator A( ) by polarization. the physical interpretation of the map
p ( ) := Tr A( )
(4.1 implies that the conditions for Stinespring's Theorem 2.3. there should be a measure on S such that Z d ( )( 1 )( (4.4.6 we may as well assume that X is compact. a unitary map U : H ! pH .4. When X is a con guration space Q. Approximating f g 2 C (X ) by step functions. We will now construct a class of examples of quantization in which one can construct an illuminating explicit realization of the Hilbert space H and the partial isometry W . it is usually su cient to take the positive map Q to be a representation of C0 (Q) on H. and a PVM ! E ( ) on H such that UA( )U .2 with Proposition 4. such that each has unit norm (so that a pure classical state is mapped into a pure quantum state).
These operators satisfy the commutation relations (say.13) and (4. in p the case S = T R3 considered above one may pass to complex variables by putting z = (q . (4. We then map L2 (T R3 d3 pd3 q=(2 )3 ) into K := L2(C 3 d3 zd3 z exp(.122) one has the simple expression ~(f ) = U Q(f )U .
4. for (2.23) (4.4 Covariant localization in con guration space
In elementary quantum mechanics a particle moving on R3 with spin j 2 N is described by the Hilbert space j = L2 (R3 ) Hj HQM (4. z )= 2 q = (z + z )= 2): One may then verify from (4.9) then holds with d (p q) = d3 pd3 q=(2 )3 .13) ): (4.24)
.q) =2
(p q )
(4. One realizes the state space as a closed subspace of L2 (S ) (de ned with respect to a suitable measure).and momentum operators in the Schrodinger representation. 2 ipq+ipx e. given by (4. where k = 1 2 3.15) (4.12)
in H = L2 (R3 ).12) that V pV .20) (4. and S Jk (angular momentum).16) form the core of the realization of quantum mechanics on phase space.22) (4. The basic S physical observables are represented by unbounded operators QS k (position). so that = (p q).18) where Hj = C 2j +1 carries the irreducible representation Uj (SO(3)) (usually called Dj ). one nds that Q(qi ) and Q(pi ) are just the usual position. where Q is the con guration space of the system.i~ kl S S Jk Ql ] = i~ klm QS m PkS PlS ] = 0 S J S ] = i~ klm J S Jk l m S P S ] = i~ klm P S Jk m l
(4. For example.14) and (4.2 we now put A = C0 (S ). when S = T
R3
4 APPLICATIONS TO QUANTUM MECHANICS = R6 . sandwiched between the projection onto the subspace in question.zz)=(2 i)3 ) by the unitary operator V .16)
W ( ) := (
It follows from (4. one may take
1 2 (x) = ( ).11.14) (4. Extending the map Q from C0 (S ) to C 1 (S ) in a heuristic way. In speci c cases the projection p = WW can be explicitly given as well. In Theorem 2.9) that W is a partial isometry.(x. (A) = A for all A. on the domain S (R3 ) Hj )
S QS k Ql ] = 0 PkS QS l ] = .17) V (z z) := e 2 zz (p = (z . This should be contrasted with the usual way of doing quantum mechanics on L2 (Q). B = B(H).1 = pfp: Q
Eqs. The representation (C0 (S )) is given by ( f ) ( ) = f ( ) ( ): Finally. (4. Eq. ip)= 2.19) (4.1 is the projection onto the space of entire
1
p
p
functions in K.86 For example. We may then verify the statement of the theorem by taking
H = L2(S d ):
The map W : H ! H is then given by
(4. Pk (momentum). and de nes the quantization of a classical observable f 2 C0 (S ) as multiplication by f .21) (4.n=4 e.
In addition. The commutation relations (4. and their j j (C (R3 ))). given by unitary representation UQM QM
j (R a) (q ) = Uj (R) (R.4 Covariant localization in con guration space
87
justifying their physical interpretation. one checks that the canonical leftaction of E (3) on E (3)=SO(3) is identi ed with its de ning action on R3 . a)): UQM
(4. Eq. Pk .26) ~ is seen as a multiplication operator on L2 (R3 ).152) with ~QM This is a neat explanation of spin in quantum mechanics.29). S S Rather than using the unbounded operators QS commutation relak .1.3.4. as well as the fact that ~QM 3 Identifying Q = R with G=H = E (3)=SO(3) in the obvious way. 3 commutativity of C0 (R ). (4.21) are a consequence of the covariance relation (3.8.7. Generalizing Proposition 4. The commutation relations (4. in terms of which the position operators are given by R (see QS = dE ( x ) x cf.115) play the role of quantized momentum observables. check that the unitary map V : Hj ! j de ned by V j (q ) := j (e q ) intertwines U j and U j .2. (4.27) for all x 2 G and 2 cf. representation (3. The associated PVM ! E ( ) on R3 where f j in HQM 4. we therefore state the situation in terms of the pair (UQM (E (3)) ~QM 0 j of the transformation group pair.152). j of C (R3 ) on Hj by Moreover. a nondegenerate representation QM C algebra C (E (3) R3 ) (cf.3) is E ( ) = Ij.156).7 this is equivalent to a covariant representation (U (G) ~ (C0 (G=H ))) on H. or.25). in which case the quantum system itself is said to be irreducible. (4. It is then not hard to j (E (3)) ~ j (C (R3 ))) is a covariant representation of the C verify from (4.3 one may instead assume one has a PVM ! E ( ) on G=H in H and a unitary representation U (G). Proposition 4. we de ne a representation ~QM 0 QM j (f ~) = f~ Ij ~QM (4. one then has PkS = j (P ) and J S = i~dU j (T ) see (3. The system of imprimitivity (U (G) ~j (C0 (G=H ))) is equivalent to the one described by (3. we have Theorem 4.
. 3.20). with given by (3.1 (q .2 An irreducible quantum system which is localizable in Q = G=H and covariant under the canonical action of G is characterized by an irreducible unitary representation of H .24) follow i~dUQM k k QM k from (3. and J k . Such a tions. (3. The physical interpretation of the PVM is given by (4.19) then re ects the k k R3 j is a representation. and (4.26).4.25)
In terms of the standard generators Pk and Tk of R3 and SO(3).25) is unitarily This follows from Theorem 3. the spectral theorem for unbounded operators. The momentum and angular momentum operators are most conveniently de ned in terms of a j of the Euclidean group E (3) = SO(3) n R3 on Hj .110).113) and the commutation relations in the Lie algebra of E (3). (4.12.4.22) .11. respectively. The representation UQM j equivalent to the induced representation U . To see this.152).151). then by de nition describes a quantum system which is jlocalizable in R3 .1 = E (x ) (4. and (3.4. j (E (3)) de ned in (4.2. given by 3.18).1.8.8) the operators de ned in (3. a nondegenerate representation of C (G G=H ) on a Hilbert space H describes a quantum system which is localizable in G=H and covariant under the canonical action of G on G=H . which satisfy U (x)E ( )U (x).1 An irreducible quantum system which is localizable in R3 and covariant under E (3) is completely characterized by its spin j 2 N .26).(4.112).151) and (3. Generalizing this approach to an arbitrary homogeneous con guration space Q = G=H . equivalently.25) that the pair (UQM 0 QM 3 dynamical system (E (3) C0 (R ) ). and covariant under the de ning action of E (3). By 3. It is natural to require that QM be irreducible. and by Proposition 4. The corresponding covariant representation (U j (E (3)) ~j (C0 (R3 ))). it intertwines the HQM QM j as de ned in (4. is equivalent to the one described by (4.
de ned with respect to the canonical Gaction on G=H .88
4 APPLICATIONS TO QUANTUM MECHANICS
This is immediate from Theorem 3. and must be of the form (U ~ ).11.11.29) Every (ordinary) covariant representation is evidently a generalized one as well.2. The claim follows. (2.11. Using the covariance Since x y = xy .12. Comwhere ( )0 is de ned in (2. writing the twosphere S 2 as SO(3)=SO(2). where arises from a continuous Gaction on X by means of (3.
. and U is a representation. To avoid confusion.3. and given by Un ( ) = exp(in ).3) by U (x)A( )U (x).2 etc.3 by H induced representation (3.31) ~ (G) quotients to a representation U ~ (G) on H ~ .128).2 we therefore omit the representation occurring in 2. we denote the Hilbert space H and the repre~ and ~ . our goal is to show that (~ H in 2. one veri es that ~ (x)f ~ (x)g )0 = (f (U U g )0 (4.28) This condition may be equivalently stated in terms of the POVM ! A( ) associated to Q (cf.28) and the unitarity of U (x).2 Let (U (G) Q(C0 (G=H ))) be a generalized covariant representation of the C dynamical system (G C0 (G=H ) ).1). with corresponding covariant representation (U ~ ) of (G C0 (G=H ) ) on the Hilbert space H . By Theorem 3.5. Hence U ~ puting on C0 (G=H ) H and then passing to the quotient.12.1 A generalized covariant representation of a C dynamical system (G C0 (X ) ). consists of a pair (U Q).152).5. and suppose that U (G) is reducible.5 Covariant quantization on phase space
Let us return to quantization theory.1. We apply Theorem 2. The following notion.151).3 with B(H). which for all x 2 G and f satis es the covariance condition U (x)Q(f~)U (x) = Q( x (f~)): (4. This class actually turns out to exhaust all possibilities. We identify B above.3 and the sentation in Construction 2.11. A class of examples of truly generalized covariant representations arises as follows.11.
4. For each unitary irreducible representation U of SO(2) is labeled by such an n. respectively the space de ned in 3. Indeed. There exists a unitary representation U (H ). such that (pU (G) p ~ p) and (U (G) Q(C0 (G=H ))) are equivalent. as described by 3. U condition (4.1.1. and Q : C0 (X ) ! B(H) ~ 2 C0 (X ) is a quantization of C0 (X ) (in the sense of De nition 4. and a projection p on H in the commutant of U (G).1 to the case where the representation ~ is replaced by a quantization Q.30) U := x (f ) U (x) : ~ is clearly a Gaction. This is veri ed from Finally. and (4. De nition 4. What follows generalizes Theorem 3. ~ (x).151) will still be called H and .12. 4.2 and 2. Theorem 4.123). this system must be of the form (U ~ ) (up to unitary representation on H equivalence). the projection p de ned in 2. Let (U (G) ~ (C0 (G=H )) be a covariant representation on a Hilbert space K.110). one checks that (U ~ ) is a covariant ~ . putting H = H. Of course.11. and ask what happens in the presence of a symmetry group. ~ (x) on C0 (G=H ) H by linear extension of For x 2 G we de ne a linear map U ~ (x)f (4.127).12.11.. as in the formulation of the theorem ~ ) may be identi ed with ( H ). Pick a projection p in the commutant of U (G) then (pU (G) p ~ p) is a generalized covariant representation on H = pK. (U ~ ) is described by Theorem 3. (3. For example.2. one infers that SO(3)covariant quantum particles on S 2 are characterized by an integer n 2 Z.28).5.3. where H is speci ed in 4. since a representation is a particular example of a quantization.2.2 commutes with all U (2. which generalizes De nition 3. and (3.1 = A(x ): (4. is natural in this context.8. where U is a unitary representation of G on a Hilbert space H.11.
1977] C Algebras. Cambridge University Press. and J. Springer.G. Ringrose 1983] Fundamentals of the Theory of Operator Algebras I. R. Math. Lecture Notes in Mathematics 582.C. Proc. 2153. Oxford. Fields Institute Monographs 6. pp. R. Amer. N. Heidelberg. Springer. In: Kadison.S.V.V. Mackey.V. 1979] C Algebras and their Automorphism Groups. Pure Math. 1978] Induced Representations and Banach algebra Bundles. (ed. G. Oxford University Press. and D. New York. 1994] Notes on the GelfandNeumark theorem.K. 1989] Analysis Now. Decomposition of States. Bratteli. 167. Amsterdam. Symmetry Groups. Kadison. R. Academic Press.and W Algebras. J. and R. Locally Compact Groups and Banach Algebraic Bundles. 1978] Unitary Group Representations in Physics Probability and Number Theory.R. K. 1982] Operator algebras . 1995] Hilbert C Modules. Benjamin. Models in Statistical Mechanics. Benjamin.M. Pedersen. NorthHolland. In: Doran. Soc. 1968] Induced Representations. Academic Press. Pedersen. New York. I: C . Symp.R.. J. A. Mackey. Vol. Connes. A Toolkit for Operator Algebraists.LITERATURE
89
Literature
Bratteli. G. Cont. E. Lance. Kadison.) Operator Algebras and Applications. Berlin. 1994] Noncommutative Geometry. Doran 1988] Representations of Algebras. Kadison. Vol.) C algebras: 19431993. Springer. 2nd ed. Academic Press. Dixmier. G. Academic Press. Amer. and D. Boston. G.the rst forty years. O.V. New York.E. 1996] C Algebras by Example. New York. pp.W. II: Equilibrium States. Cambridge.K. Springer. Math. WeggeOlsen. R. J. New York. New York. G.W. R.W. Davidson. 118. Robinson 1981] Operator Algebras and Quantum Statistical Mechanics. 1993] Ktheory and C algebras.S. and J. San Diego. Takesaki. R. Providence (RI). 1979] Theory of Operator Algebras I.G.W. Soc. Springer.V. American Mathematical Society. 2.R. London. Robinson 1987] Operator Algebras and Quantum Statistical Mechanics. Berlin. Ringrose 1986] Fundamentals of the Theory of Operator Algebras II. Vol. Math. (ed. LMS Lecture Notes 210. Fell.
. Fell. Kadison. 1963] The Mathematical Foundations of Quantum Mechanics. Academic Press. Providence. Berlin. Providence (RI). Mackey. M..W.M. Benjamin. O. 38(1).