REFERENCE INTERNATIONAL CENTRE FOR THEORETICAL PHYSICS INTERNATIONAL ATOMIC ENERGY AGENCY UNITED NATIONS EDUCATIONAL, SCIENTIFIC AND CULTURAL ORGANIZATION CONDITIONAL EXPECTATIONS IK QUANTUM PROBABILITY THEORY S. Twareque Ali and Gerard G. Emch 1974 MIRAMARE-TRIESTE
30
Embed
INTERNATIONAL CENTRE FOR THEORETICAL PHYSICSstreaming.ictp.it/preprints/P/74/072.pdf · Over the last twenty years various extensions of the concept of conditional expectation have
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
REFERENCE
INTERNATIONAL CENTRE FORTHEORETICAL PHYSICS
INTERNATIONALATOMIC ENERGY
AGENCY
UNITED NATIONSEDUCATIONAL,
SCIENTIFICAND CULTURALORGANIZATION
CONDITIONAL EXPECTATIONS
IK QUANTUM PROBABILITY THEORY
S. Twareque Ali
and
Gerard G. Emch
1974 MIRAMARE-TRIESTE
International Atomic Energy Agency
and
United Nations Educational Scientific and Cultural Organization
INTERNATIONAL CENTRE FOR THEORETICAL'PHYSICS
COVARIANT CONDITIONAL EXPECTATIONS IN QUANTUM PROBABILITY THEORY *
S. Twareque Ali
• International Centre for Theoretical Physics, Trieste, Italy,
and
Gerard G. Emch
Departments of Mathematics and of Physics and Astronomy,University of Rochester, Rochester, NY, USA.
MIRAMARE - TRIESTEJuly
* • To be submitted for publication.
wwm
This paper investigates the
extension of the concept of conditional
expectation to a non-commutative
probability theory, subject to some
group covariance requirements.
Over the last twenty years various extensions of the concept of
conditional expectation have been explored [3,7»H,12,13,l6]. The line of
investigation, to which the present paper belongs, deals with the possible
generalizations of the idea of conditional expectation from classical probability
theory [8] to non-commutative probability theories, in which the classical
algebra J^_ (X,y) of stochastic variables is replaced by a non-abelian von
Neumann algebra. Such a generalization, apart from its intrinsic mathematical
interest, is useful in discussions of the measurement process in quantum
mechanics [l,3,*0> an(l in the theory of coarse-graining in statistical mechanics.
[1?]. In this paper, we adopt Davies and Levis1 [3] definition of a conditional
expectation, as the dual to the concept of a measurement. We next subject a
measurement to a condition of covariance under a locally compact symmetry
group. The problem which then concerns us is that of expressing covariant
measurements in terms of certain operator-functions, these latter providing a
better basis for making contact with the theory of observables in quantum
mechanics [l]. We succeed in getting such a description for two classes of
measurements, which actually appear to be sufficient for all cases of
physical interest. We wish to point out that, in contrast to Davies'work [If],
we found it useful to stress here the measure theoretic aspects of a conditional
expectation and a measurement, A description of our results can be found at the
end of Sec. I.
I. PRELIMINARY DEFINITIONS AND FORMULATION OF THE PROBLEM
Throughout this paper X shall denote a separable, locally compact,
Hausdorff space, K(x) the space of all continuous complex-valued functions on
X which have compact supports (equipped with the standard inductive topology), and
W a separable Hilbert space, J£$¥), &($) and '&($) shall denote, respectively,
the Banach spaces of all bounded operators (on $f) under the operator norm, of
compact operators under the operator norm, and of trace-class operators under the
trace norm. It is well known [15] that the Banach space duality relations
hold between these spaces. We shall denote the positive cones of these spaces
(which in fact generate the spaces) by K(x) itfW) , etc.
-2-
Definition 1.1
A measurement on X is a bilinear map,
which is continuous and satisfies:
2) if f is a sequence of functions in K(X) such that f / I ,n -1
pointwise, then tr[£(fn,f)]y*tr f, V* f € cT'&O*. where 'tr
denotes the trace operation on «y W) •
Definition 1.2
A conditional expectation is a "bilinear map,
£*-.which is dual to a measurement through the relation
?)i(1.1)
As shown by Davies [k], Definition 1.2 generalizes the concept of a
conditional expectation of Umegaki [l6], which in turn generalizes, to operator
algebras, the concept of conditional expectation in classical probability theory
(in the sense of S.-T.C. Moy [ll]. Further, the measurement g is also a
generalization of von Neumann's expression for the "collapse of a wave packet"
in the course of a measurement [3].
Let G be a locally compact group, and suppose that for each g<£G there
exists a positive, norm-preserving automorphism of Zf&O onto itself. We shall
assume that this automorphism is unitarily implementable, so that we have a
weakly (hence strongly) continuous mapping g n- U of G into the set of unitary
operators of£f. The space X on which the measurement § is defined will be
assumed to be a transitive space in the sense of Mackey [9]» with respect to the
action of G (assumed to be acting on the right). Hence X will be
identifiable with the quotient G/H, of G by some closed subgroup H. In
particular, we shall take G to be a semi-direct product of the form:
G = H® X
where X itself is a commutative symmetry group of the system. Let g[f]
(= U *f>U ), g[f] and [x]g denote the transforms of f£j°(^0, f£K(x) and
respectively, under g
-3-
Definition 1.3
The measurement £ is said to be 'covariant under the action of the group
G if </-g£G, f£K(x) and
= r'l?(f,?Le])l.The conditional expectation g* is said to be covariant if it is dual to a
covariant measurement S
For feK(X), le t a(f) € Jf(#) be defined by
the resulting map a: f £ K(x) «• a.{t)e£&f} is positive and thus defines a
positive-operator-valued measure on X, which we denote by
a: EeB(x) w- a(E)eo£$0+, where B(X) is the set of all Borel sets of X. We
have, in particular, a(x) = I, the identity element in £&f).
Definition l.k
The normalized, positive-operator-valued (POV) measure f a(f), or
equivalently E » a(E), defined through Eq.. (1.3) is said to be the observable
determined by the measurement g .
The observable E <•*• a(E) determined by a covariant measurement satisfies
the relation
d 1 $ > d^)
[E]g~ being the translate of the set EeB(X) through g" .
Definition 1.5
The pair E •+ a(E), g *+ U satisfying Eq. {l.k) is said to form a
generalized system of imprimitivity. (in case a(E) is projection valued, this
reduces to a transitive system of imprimitivity in the sense of Mackey [<?])•
In the next section we obtain (Theorem 2.1) a general integral representation
for covariant measurements in terms of a Cf&P) space valued density function
x •+ I"}* (x). Next we find (Theorem 3.1) the conditions under which this density
may be written in terms of a certain bounded operator-valued function x *• T ,
namely Tl Cx) » T (f). This case covers all situations of physical interest
(including those treated in [l]), except the.case of a covariant observable with
a purely non-atomic projection-valued measure. This last case, however, is taken
-k-
care of by the results of the final section of the paper, where we analyse
a situation (Theorem U.l) in which' T. may "be unbounded. This allows us tox . . . .
answer the question as to which covariant measurements determine o"bservables
such as are projection-valued measures(Theorem k.2).
II. A CHARACTERIZATION THEOREM FOR COVARIANT MEASUREMENTS
Theorem 2.1
Let g be a measurement on a locally compact Hausdorff space X
S, : K(X)
$f being a separable Hilbert space, and let £ satisfy the group covariance
condition
where G is a locally compact group of the form
G = H®X,
with H a closed subgroup of G and X itself a commutative group. Further,
let there exist a strongly continuous representation g *+ U of G (implementing
the positive norm continuous automorphisms of &/{#) induced by G) by unitary
operators on $f.
Then, for each P, ^ is completely defined by a function.
which is positive if f €JW) , and which is measurable with respect to the Haar
measure y on X. £ is further given by the integral relation:
S(f,f) = f(x)Pr(x.)/ACc(.?t), (2.1)Jx /
vfeK(X) and f*etfkff), the convergence of the integral being in the weak topology
of tftyf). For all xeX and fcfffi), I£(x) satisfies the group covariance
condition
? u ? f u ; . (2.2)
Proof
The proof of this theorem will be given in four stages:
(a) The measure <?(f, f), for fixed f € 3"&) , will be written in terms of its
absolute value \B^\ as
-5-
f "being some \8J - jaeaaurable function with, values in
(b) For any geG and f£K(X) i t wi l l he shown tha t
and also that the map g »•»- ]<f rp-||(f), for fixed feK(x) and fG?°C#0, is
continuous.
(c) The results of (t>) will "be used to prove that the Haar measure u on X
dominates the measure | £p\ in (a).
(d) Using (c), the desired form for £f(f,p) will be obtained by writing first
Ifpl * Yf • V ,
where yp is some y-measurable numerical function, and then setting
V? (x) = f.f(x) YfU).
We proceed now with the detailed proof.
(a) For a fixed f£&($), ^(f,f) (with f£K(X)) . defines a vector-valued measure
on X, with values in v $f)* i.e.,
S(f, f) - |f(f) = \ fM &*&*). .(2.3)
We shall prove that.this measure is majorizable for the norm of [f&)t in the
sense of Bourbaki [2], and that \£_f\ exists. Indeed, from the definition of
the absolute value of a measure.
1£P| - Sup lA
where Ao^Cf) = tr[A^(f) ], for all feK(X). Since g is a continuous
bilinear map on the normed spaces K(x) and tf&f), for every compact subset
V ^ x it follows [5] that g restricted to K(V) is bounded by some number
. Hence, for all f£K(V)+ ,
f\\KCV)
-6-
which proves our contention.
Thus, £« is a measure on X with, values in 3"&0 , the strong dual of
the separable normed space &(#*;. Further, £=is majorizable for the norm of
V&f), and has as its absolute value the bounded positive measure |£p|> Hence
there exists [2] a function
/f :X—> ?(&),.measurable with respect to \£f\
a n^ satisfying:
(2) for all f£a7'<^0, | jf. » (x) j |y^\ = 1, for almost all x (with respect to
\if\) 'We may write then
so that Eq. (2.3) may be rewritten as
(b) Consider next the measure |£ rp-ij for any geG. We shall prove that
for all
Indeed, we have, for any
5^pIIAI;<J
p
u ? |trfA9-J[5(/',/[p])J]|, in virtue of Eq. (1.2),
Henc e,
Also, since by Eq. (1.2), g[S(g[f],f )] = (f, g[f])» therefore, we can
similarly show that
, +for all feKlXj , whence Eq. (2.5) follows.
-7-
To prove t h e c o n t i n u i t y of t h e map g **• \g- , , | ( f ) , for f ixed fcK(x)
and fe TS ), we prove first tne continuity in the norm topology of Hffi), of
the map g<+ g[A] = U AU *, for a l l k£&&). (Since G is a group, i t i sDO
enough to prove continuity in a neighbourhood of the identity element e of G)
We have
AU/ - Ai
Hit 3
Now, <gW being an ideal in the C* - algebra
compact operator. Hence, there exists some vector
, U AU * -A i s also a6 S
£ in c£f~ for which
< U\\im II (u;* - 1 )\im I (u? - r)A ?„0, as
because of the strong continuity of g •+ U . In fact, the continuity ofo
g H- u AU * is also seen from this proof to be uniform in the unit ball ofS g. This result now implies the continuity, in the trace norm topology of,
, of the map g H- g[f] = U*fU . Indeed,o o
\\u4A'ii A 11 <
J
-+•0, as g -+• e .
The continuity of g H- \g . J(f) is now immediate, for
SIf o ItS J
hr User,p)]i'£ KCX)
U\I<J UfUf
— i| o l\ yj- Hi II K(yj (I Ug Ug j3 JJo'/-'j , where V is some compact
region containing the support of f
- • 0 , as g -+• e .
(c) We shall prove next that the right invariant Haar measure ]i on X
dominates the positive measure \So\-
This will be achieved by first showing that u dominates the measures
] for y-almost all xeZ [lO]. To do this, let EeB(x). Then the set
E1 of all (x,y)€X x X such that xy' £ E is also a Borel set, and for all
x and y, Y-PI (x»y) - X-^i^y" )• (X denoting, as usual, the characteristic
function). Applying the Fubini theorem to x^ v e
fX
x
The left-hand side, because of the invariance of ]i is equal to u(E) |g|aj (X),
" ) = x l ( y x ~ )and since Xgtxy" ) = x
E-l(yx~ ) - X[E-l]x^y)» the right-hand side is equal to
Thus,
Hence, y(E) = 0 implies |£j([E~ ]x) = 0, for all x, except perhaps on—1
a set of y-measure zero. But 1J(E) = 0 iff u(E ) = 0, so that (in virtue ofEq. (2.5) above) u(E) = 0 implies \g r -. | (E) = 0, for y-almost all x,
whence our assertion follows.
The continuity of i£xrp]i i n ^ (as proved in (t>) above) now shows
that if'{x } is a sequence converging to e, then |g rplKE^ converges ton —X L1J
n
|gp|(E). But since U has support on the whole of X (being the invariant
Haar measure), therefore, ]g rpJ(E) = 0 for p-almost all x implies that
every neighbourhood of e has at least one point x1 for which
\8 trp-il(E) = 0. Let us choose a sequence of neighbourhoods ' fr } of e
:« lit Ji> V I-
such that V C V for all n, and further s.o that every neighbourhood Vn+i n • • • • •' ' . . e
of e contains at least one V . In each. V . let us choose a point x
such that |g r -||CE) = 0 . Then, taking the limit of n •*• », we immediately
get the result that u(E) * 0 implies |£P|(E) - 0, proving that u dominates
pi *
(d) Using the result in (c) above we have
(2.6)
for all feK(X), where Yp is a positive numerical function, measurable with
respect to u . Let us now define x "•*• n,(x) by setting
where f_p is defined as in Eq. (2.1+). Then it is straightforward to check
that rj, satisfies the stated conditions as a measurable function of x in
*fffi)i positive if f itself is positive. Also obvious is the covariance
property of FP as spelled out in Eq. (2.2). Hence finally, combining Eqs.
(2.k)t (2.6) and (2.7) we get the stated relation
The function x ^ ( x ) obtained above is measurable in the sense that
for each AzXQfr), the numerical function x '-> tr[Al^(x)] is measurable.
Definition 2.1
A measurable function x -»-Pp(x), of X into 3"W) t will be called the
density of the measurement g , for the state p and with respect to the base y
(assumed positive), if and only if it satisfies Eq. (2,l). It will further be
called,covariant if it also satisfies Eq. (2.2).
The density x*l"f (x), as defined through Eq. (2.7) is determined
completely, except at most on a set of u-measure zero, which could perhaps
depend on f . The next question thus is to find conditions under which this
null set is independent of f , for then we could use the linearity of Z (f »f)
in p. to define a linear operator T on tfffl) in the manner
(2.8)
-10-
on a suitable domain in $&).. The'remainder of this paper is devoted to
answering this question in two cases of physical interest.
By its very construction, the function x **• T will have the properties;
(a) For almost all xfX (with respect to u), the operator T :,/(#) -*• r/G£Q
is positive and linear. It is "bounded whenever it is defined on the whole of
(•fa) If Cftf&f), '$&))* denotes the set of all positive linear maps on
hounded or otherwise, then the (positive-operator-valued) function
x£X «• T £$£f&), ifffl)) is measurable in the sense that for each f belonging
to the intersection of the domains of the operators T ( x£X, except perhaps
on a set of u-measure zero), and each A€jpi°), the numerical map
x «• tr[AT (f)] is measurable.
(c) For all P in the intersection of the domains of the operators T (p_
almost all xeX),
(2.9)
and, if £ is also covariant,
Tt.3f (f)
(d) For all P which are in the intersection of the domains of the operators
•p (vi-almost all
Definition 2.2
A measurable function x •+ T of x into (7(tf$0, ZfW)) will be
called the operator function inducing the measurement g , with respect to the .
base y (assumed positive), if and only if it satisfies Eq.. (2.9) (and will
further be called covariant if it also satisfies Eq. (2.10)) on a domain which
is dense in
III. MAJORIZABLE COVARIANT MEASUREMENTS
Lemma 3.1
Let X - ^ P P ( X ) be the density, with respect to p, of the covariant
measurement <f . Then, if for any xeX, Ff>(x) satisfies
(3.1)
for all fe/89, where \\r\\ is'a positive .number, then ||r|| is
independent of x and the'relation (3.1)is satisfied for all
Proof
Since £ is covariant, Tf>(x) satisfies Eg.(2.2) and hence, for any
xfcX
afrom which the result follows
Definition 3.1
Let i^fjid) be the density (with respect to y) for the covariant
measurement 8 • Then, Pe{x) will be called mat1orizable if, and only if it
satisfies the relation (3.1) at some, hence all, points x€X and
Lemma 3.2
If x *+ Vf> (x) is majorizable, then as a function of x it is
continuous in the trace norm topology of $"(&),
Proof
Let o denote the origin of the Abelian group X. We have, in virtue
of Eq. (2.2),
r? c*) - rP (o) i t m = I v* r^w (o;ux - rP to
<
-12-
because of the linearity, in f> of £(jf,f). Thus using. (3.1)
But since x U py * is continuous (cf. part ("b) in the proof of Theorem 2.1)
we get at once that
as x •+• o ^
Lemma 3.3
Let the covariant measurement £ be induced by the positive operator
valued function x '•+• T . Then, if T is a bounded operator at any point x,
the function x ••*• T is uniformly bounded for all x<?X.
The proof is an immediate consequence of the relation (2.8) and Lemma 3.1
Theorem 3.1
A measurement & on the locally compact Hausdorff space X, vhich is
covariant with respect to the action of the group G = H(£)X (assumed unitarily
implementable as before), is induced by a unique, (up to a set of y-measure •
zero) bounded, positive-valued function x ** T , with respect to the Haar
measure u on X, if its density function is majorizable.
Conversely, if w X ^ T ^ W i J($))* is a measurable function
such that
(a) I T ^ C ) ! ^ < I! T ff ftp H ^ , for all f<7WO andsome xeX, ||T11 being a positive number;
tr[ I f (x) T (f) u(dx)] / tr P , where f is a sequence in K(x)
+such that f SI, pointwise and
(b)
^ ^ t f p } t for all gcG and
duces a covariant measu
density function x^Pp(x) .
Proof
If x1-*- Pp(x) is majorizable then it is continuous (Lemma 3-2), and
hence as a function of x (for each fcVW it is completely determined at all
then x •* T induces a covariant measurement having a majorizable•x
-13-
iE1""
points of X. Therefore, Eq,. (.2,8) jaay he used to define the operator
function x •-»• T > which "by the stated condition would he uniformly bounded.
Uniqueness in the sense stated is obvious.
Conversely, let x •+ T x be, given as stated. Then clearly it is
uniformly "bounded (cf. Lemma 3.1). Let f£K(x) and f€$$f). Then, for all5
fCx) t
'Pl«w J< °o
Thus the integral converges in the weak topology of tfffl). It is then easy to
see that
'Xis a continuous bilinear map and hence, in virtue of conditions (b) and (c)
defines a covariant measurement in the manner stated!
IV. PURE MEASUREMENTS
To proceed further, we need the concepts of a modified and a pure
measurement.
Following Definition (l.h) of the observable determined by a measurement,
we remarked that the observable determined by a covariant measurement satisfies
the generalized imprimitivity condition:
Suppose that there exists on the Borel sets of X a projection valued measure
E H- p(E) which also satisfies the relation
In that case* by the imprimitivity theorem of Mackey [9] the Hilbert space
is isomorphic to the space L^(Xfu) of all functions §: X -»• 3i which satisfy
vhere ?<5 is a Hirbert space carrying a unitary representation of the subgroup
H of G. Further, on L-^CX, y) the operator P(E) is given "by :A J^
where X I s the cha rac te r i s t i c function of the set E.
Definition k.l
A measurement K(x) X $"($) •*• Jffl) t covariant with respect to the
representation g >-*• \J£J&$) of G-, is said to "be a modified measurement if thereg
exists a projection valued measure E * P(E) on $f (E£B(X)) such thatEM- P(E) and g >•+- U form a protective system of imprimitivity on H, in the
sense of Mackey.
To define a pure measurement ve need a few more results. These will "be
stated in Lemmata k.l to J+.3
With the Hilbert space appearing in Mackey's imprimitivity theorem
as stated above, let L 1 _. (x, \i) denote the Banach space of all Borel functions
r_ : X -*-c7*(?<0, for which the norm defined by
^ x ) ' (U.3)
is finite, where \\ • • • \\ ft^A <3-enotes the trace norm in tf&O . Two functions
in L [frtA^* y) are identified if they differ on a set of y-measure zero.
Let o ^ y ^ C X , y) denote the corresponding space where such functions are not
identified, and let K J « A ( X ) be the subset of functions in L^^lX, y) which
are continuous and have compact supports. K^/^tX) is dense [2] in
fr/ovx (X, y) with respect to the semi-norm defined in (U.3). Also, let X be
the isometric,positive linear map
X:
which on elements in J*C^)+ of the form £<g>£, for $€.&, is defined by
One has then
= [XCU,- 1 5 -
for a l l g € Q and ji-alaoat a l l xeX.
Lemma U.I
The mapping X; ffffi'•*• L^-^CX, u) defined above is invariant under the
unitaries of ' {P(E)}" (the von Neumann algebra in JtW) generated by the
projection operators P(E) defined in (1*.2) above), and conversely, any operator
in J£&) which leaves A invariant is a unitary operator in ' {P(E)}''' Explicitly,
for all iaft \{Ul ®%t&) - \ti®l) if and only If U is a unitary operator
in 'fr(E)}"
The proof is easy and ve omit itj|
Lemma \,2.
Let g "be a measurement whose observable is a projection valued measure
E H - P ( E ) . Then, for all feK(X) and fetf&O, Z satisfies <f(f,f=) =
? i?>Uf>lf) for all unitary UC <P.(E)>".
Proof
If g* i s the conditional expectation dual to £ , then
tr
I being the identity element in JCffl), and where we have set (by the usual
method of extending a measure)
Thus,
^ * C E , I ) = P(E) • (U.6)
If now A is a positive operator in JfC#), ve have,
S*CEf]F , lAl! - A ) > o ,
for any E, F€B(X), since l!Ally(j ) x ^ A* Thus,
o < ff*CFnF, A ) < /JA// (fCEn'F, i) = |Af p( EnLet F' denote the complement of F in X. Then,
0 < £*(E.C) F , A) P ( F ' ) < IAJJPCE O F ) P ( F ' ) = O ,
Thus,
f*(EnF, A) p e r ) ' = o, •
-16-
so that,
S*(tr\r,A) = e*CEf\F, A ) [ P ( F ) •«• P(r'j]
= <?*(EnF, A) POO.
Similarly,
^(EAF, A) = P(F) (EOF, A),
so that,
e*(Ef]F, A)P(F) = P(F) £""CEnF, A ) .
(U.8)Replacing F by F' in ( .7) we get
£*Uf\F' , A) P(F) ' = P(F) g*(E OF', A) = 0 .
(U.9)
Adding (U.8) and C^.9) we have finally,
S*tE, A) P(F) . - P(F) £* (E , A) .(U.10.)
Hence, for a l l AeJ&ffl and a l l E€B(X), £*(E,A) commutes with every element
in 'fc(E)}11. If ^ f rCE)}" i s unitary, then clearly
&* g*(E, A) U = ff*(E, A),
so that, for all f / W )
O ( , ) f] = tr
i.e. ,
tr [A
from which the desired result follows
-IT-
Lemma k. 3
Let the measurement g he induced by'an operator function x *+ T , with
respect to the base u. Then, for Zl£ j£(ffl unitary, £ satisfies
Sit, Itffr) = <?(*,(=>). for all feK(X) and pc/OP)-, if and only if Tx
satisfies T^fft*) - T^Cf) for u almost all x£X and all f in the common
dense domain the operators T .
The proof is trivial§
In general a modified measurement need not "be invariant under the
unitaries of ' {p(E)>" in the sense that g(,t,Kftt*) need not "be the same as
We shall, therefore, introduce the concept of a pure measurement.
Definition U.2
A modified measurement § will be said to be pure if and only if it
satisfies the condition <f(f,p) = £(f£/f>U*), for all unitary U£ {P(E)}".
Remark
A covariant measurement whose observable is a projection valued measure
is necessarily modified and in fact, in virtue of Lemma ^.2, is also pure.
We state next a lemma, concerning the domains JD of the operators T ,
which will prove useful in the sequel.
Lemma ktk
Let x *-+ T induce a measurement g which is covariant. Assume that
JD (dense inji?') is invariant under the action of the unitary operators
U , V*x£X, i.e.4 Uv^oto Ux " ^ ^ » ^ € X , Then £) J^isx. for all xeX,