link.springer.com978-3-540-47367...the projection p is defined by identifying the spaces x2 and x 2...

39
94 Appendix l: Halmos' lemma Halmos' lemma: Let the map A : Xl --> X 2 denote a linear contraction between two Hilbert spaces. Then there exist an orthogonal projection P, a unitary map U and an isometry J, such that A p·U·J Proof: We explicitly define the maps. J : Xl Xl X 2 J (xl) = xl 0 . U Xl X 2 X 2 Xl U = [_: : *] 1 * ;;; with S (I-A A) Xl Xl 1 * ;;; and T (I-AA ) }{2 X 2 The projection P is defined by identifying the spaces X 2 and X 2 0 , P : X 2 Xl : X 2 P(x 2 xl) = x 2 . It is an easy exercise to show that the maps have the required specifications. But to prove that the map U is unitary, we need to know that Sand T intertwine with A AS = TA This we show as follows. Consider the identity 2 * * 2 A'S = A'(I-A A) = (I-AA )'A = T 'A .

Upload: others

Post on 26-Jan-2021

2 views

Category:

Documents


0 download

TRANSCRIPT

  • 94

    Appendix l: Halmos' lemma

    Halmos' lemma: Let the map

    A : Xl --> X2

    denote a linear contraction between two Hilbert spaces. Then there

    exist an orthogonal projection P, a unitary map U and an isometry

    J, such that

    A p·U·J

    Proof: We explicitly define the maps.

    J : Xl Xl X2

    J (xl) = xl 0 .

    U Xl X2 X2 Xl

    U = [_: : * ]1

    *;;;

    with S (I-A A) Xl Xl1

    *;;;

    and T (I-AA ) }{2 X2

    The projection P is defined by identifying the spaces X2 and

    X2 0 ,

    P : X2 Xl : X2

    P(x2 xl) = x 2 .

    It is an easy exercise to show that the maps have the required

    specifications. But to prove that the map U is unitary, we need to

    know that Sand T intertwine with A

    AS = TA

    This we show as follows. Consider the identity

    2 * * 2A'S = A'(I-A A) = (I-AA )'A = T 'A .

  • 95

    By a simple induction in nErn this extends to

    A.S 2n = A.S 2(n-' ).s2 = T2(n-' ).A.S 2 = T2(n-' ).T2.A

    We have shown that

    where pol denotes a polynomial.

    We now choose polynomials Pn(·) on [0,'] in such a way that

    converges to v't uniformly on [0, , ] By the functional

    calculus we get

    Pn(S2) ------> S in ?1i (X, )n

    Pn(T2)

    ------> T in ?1i(X 2)n

    Hence

    AS TA .

    For x, EX, 'Ne get

    p·U·J(X, ) = p·U(x,l& 0) P(Ax,l& (-Sx, )) A(X, )

  • 96

    Appendix 2: Gaussian measures

    The results in this appendix come from [1 8]. We provide the

    construction of the gaussian measure on infinite dimensional spaces,

    uniqueness of the measure and some properties that are both useful and

    have illustrative purposes.

    The numbering is independent of the numbering in other parts of

    these notes.

    A: The gaussian measure ona>

    IR

    We define the space moo as the set of

    \Roo = ... ,an"") I aiE\R

    We define a topology in moo by the metric

    all real sequences,

    for all iEIN } .

    Le.

    00

    = = l 2-nn=l

    !Xn-Yn!

    l+lx -y In nwhere

    The well known real Hilbert space is a subset of \Ra>

    n=l

    Writing \Roo-n for the subset of \Roo consisting of zeros in the

    first n positions, and identifying the space \Rn as a subset of \Roo

    by adding zeros after the first n positions, we get

    \Roo \Rn ffi 1R 00 - n .

    00

    In \R we define the cylinder sets

    B ffi 1R00

    -n

    where n are natural numbers and B runs over Borel sets in IRn

    .

    These cylinder sets form a base for the Borel sets of \Roo, I- I

    We define the gaussian content on the cylinder sets in !Roo

    by setting, for nEIN and a Borel set B C !Rn ,

    ffi \Roo-n) = ! .. .. dxnwhere dx

    1dx2... dx

    ndenotes the Lebesgue integration over \Rn

  • 97

    We have that

    'Y(moo) = 1

    and for every fixed nEm 'Y becomes a measure on mn

    We shall use the Kolmogorov extension theorem saying that if 'Y

    is a content on the cylinder sets in

    'Y(moo) =

    00

    m such that

    and if 'Y is a measure on mn for every nEm , then 'Y extends

    uniquely to a Borel measure on moo, I- IWe denote the measure by the same symbol -r .

    For a > 0 we define

    'Ya(B e moo - n ) = (2va)-n/2 f .. .. dxnB

    It is known that for different a's the corresponding measures are

    pairwise singular. For simplicity we develop the theory for a =

    In particular we shall need the measures for the numbers

    a = and

    These measures will be denoted by 'Y

    a =

    and

    Theorem lA: Consider a sequence

    numbers. We define the set

    of positive real

    00

    If the series 2an is summable then -r(E)=ln=l

    otherwise 'Y(E)=O .

    Proof: We construct a net of functions converging to the

    indicator function of E, lE' Given A>O, we define the function00

    fA : m m

    by00

    exp(-A 1n=l

  • 98

    It is easy to prove that

    {: if x(Ef?\ = positive real number if xEEand that f?\ l pointwise for ?\----->O

    The functions

    N

    = exp(-?\ 2 for NEllin=l

    are obviously integrable over ffioo and for fixed ?\ form a decreasing

    sequence. By the Lebesgue theorem of monotone convergence we get that

    J = J lim f?\ = lim J f?\a" [Roo N- ' N- [Roo '

    in case the limit exists.

    Notice that in the case where f depends only on a finite

    number of variables xl ,x2' .. o,xn the integration reduces to

    N

    2 -N/2 1\ 2 d d= exp(-2L xn) x x 2o .. dxNn=l

    t )1+2?\a·x we computenN N

    J Ne xp(-?\ 2 2 .. dxN[R n=l n=l

    Using the substitution

    Nnn=l

    [ J[R

    N/2 N [ J 1 2 dtn exp(-2 t ))1+2?\an=l ffi n

    nN -1/2 J 2(1+2?\a ) since )dtn=l n

    [R

    ../2; .

    00

    200

    If an 00 then n (1 +2?\a ) = 00 , thusn=l

    n=l n

    J f?\ 0ffioo

    Otherwise we have that

    00

    l an < 00 ,n=l

    and by the inequality for non-negative x

  • 99

    o In ( 1+x) xwe get that

    oc

    TT (l+2Aan) < '" •n=lThis implies that

    { oco

    TTn=l

    (l+2Aa )-1/2n

    if

    if

    cc

    2 a conn=l

    '"2 a < '"n

    n=l

    By using the Lebesgue theorem

    = JIR'"

    co

    If 1an < "', we have thatn=l

    cc

    of dominated convergence we get

    lim JA->O IR

    lim TT (1+2Aan)A->O n=l

    which concludes the proof.

    The difficulty in developing the theory of gaussian measures in

    IR'" is due to the fact that contrary to the finite dimensional case

    where = we have, as shown in the above, that2(L ) = 0 .

    To shorten the notation we define the measures

    nEIN, by

    on

    (d) (2 1 2 2))d dw ,+xn xl'" xn

    If we identify the spaces IR'" = IR n ffi IRoo- n and IR n x IR

    oo- n , and denote

    by oo-n the gaussian measure onoo-n

    IR , constructed in the same way

    as the measure on IRoo

    the measure can be identified with

    the product measure ® '

    = ®n oo-n

  • Definition 2A: Let f

    100

    denote a function defined on 1R00 and

    integrable with respect to For nErn we define

    = J ,1R00-n

    where xElRn and sElRoo-n and IR n is treated as a subset of

    By the Fubini theorem Enf is well defined, and Enf

    integrable function over withn

    J = JIR n 1R00

    Lemma 3A: For fELl and nErn we have that

    00IR

    is an

    Proof: Assume that f is a non-negative function, i.e. assume

    that f20 Then we have that Enf 2 0 and

    II En fill = J(En f ) • n = Lf • II fillIR n IR

    An arbitrary f can be written uniquely as

    f = f+ - f- ,

    where f+ and f

    Then we have

    are non-negative integrable functions such that

    f+·f- 0

    I f I = f+ + f

    IIEn(f+ - f-1II 1 IIEn(f+1II 1 + IIEn(f-1II 1

    = If I = .

    Definition 4A: A function f defined on 1R00 is called tame if

    there exists an nErn such that

    where

    1R00 ------> IRn

  • 101

    denotes the natural projection.

    Notice that if is a tame function, depending only on

    the first n variables, then

    Lemma SA: The tame functions are dense in,

    L ('Y) •

    Proof: Since the cylinder sets form a base for the Borel sets in

    moo every Borel function can be approximated by finite linear

    combination9 of indicator functions of cylinder sets.

    functions of cylinder sets are tame.

    Theorem 6A: (Jessen) For fELl ('Y) we have that

    -----+l f in L' ('Y) •n-

    Indicator

    Proof: Consider E>O. Find a tame function gEL' ('Y) such that

    Ilf - gill < E/2 .

    As g is tame, there exists an such that g depends only on the

    first N variables. For nLN we have that

    and

    IIEnf - fill IIEn(f-g)11 1 + Ilf-Engll,

    liEn (f-g) II, + II f-gll, 2·llf-gll, < E .

    Corollary 7A: Let f denote a function defined on which is

    integrable with respect to If f is constant with respect to

    every finite number of variables (depending only on the tail), then f

    is constant a.e. in

  • 102

    coProof: Considered as a tame function on ffi Enf depends only

    on the first n variables. Hence E f is constant a.e. Since f isn

    constant with respect to every finite number of variables, we have for

    all m,nEIN that

    fffico- n

    fffico-m

    Emf (x, , x 2 ' ... , xm)

    Thus

    is obviously constant a.e.

    Remark: Since the arctan of a measurable function is integrable,

    corollary 7A holds for every measurable f .

    The following corollary is often called the Kolmogorov zero-one

    law.

    Corollary SA: Let B denote a Borel set inco

    ffi If the

    indicator function of B is constant with respect to every

    finite number of variables, then B is either of measure zero or one.

    Proof: From corollary 7A the function 'B is constant a.e., by

    which it is obvious that 1B = 0 a.e. or 'B = ,a.e.

    B: The linear measurable functionals onco

    IR

    Definition 1B:

    ffico if A fulfills

    is called a linear measurable functional on

    is linear on E and measurable with respect to .

    1) A

    2) A

    is defined on a linear setco

    E C lR of full measure

  • 103

    In what follows we shall provide a representation for these

    linear measurable functionals defined on rn°O

    Proposition 2B: (The Kolmogorov inequality) For00

    we define

    n

    = l akxk·k=l

    For arbitrary E>O we have that

    nEIN and

    sup

    n

    I > E }) l / E 2 .k=l

    Proof: We start by noticing that

    1 J 1 2 d 0rn t·exp(-2 t ) tand

    1 f 2 1 2 drn t exp(-2 t ) t = 1For arbitrary, pairwise disjoint, measurable sets

    meIN we have

    Q crn°Om

    n

    lk=l

    n

    l 1 frn 2 (' 2 dxk'exp - 2xk) xkk=ln n

    ld f 2 (' 2 d l L 2 2xk'exp - 2Xk) xk a k xkk=l rn k=l rn

    n

    foo ( lrn k=l

    and since

    estimate

    the relation

    Z + 2sm(sn-Sm)

    s2 + 2s (s -s ) + (Sn-Sm)2m m n m

    we can continue

    implies the

    n

    Z l f + .m=l Qm

    We construct the sets Qm' mEIN, as the inverse images by the

    function

  • 104

    the smallest kErn such that I > E

    inf { kErn I > E }Hence

    Notice

    -1 {Qm=f (m)=

    that the function

    00xEIR f = m } •

    depends only on the first m

    coordinates

    We return to the integration. Since the functions and

    depends only on the variablesfunction

    andvariablesmonly on the firstdepend

    we get by the Fubini theorem

    o ,

    J (lQ1R00 m

    J 1Q JIR 00 m IR 00

    sn-sm being linear in the variables xm+ 1 ,xm+2'· .,xn .

    J s (s -sm n m -Qm

    We estimate the second expression using the fact that if

    then and hence

    JQm

    > E •

    L JQm

    Then we get

    2= E .

    We return to the first estimate

    n n n

    2 L 2 J L 2 2 n Qm)= Ek=l m=1 Qm m=1

    • Ii 00 therexEIR exist such thatxEIR

    00there such that= E exist

    ( xEIR00

    I I > E })sup ,

    })

    })

    which concludes the proof.

    Theorem 3B: (The Kolmogorov large number theorem) Consider a

    sequence of posi tive numbers If

    00

    2 < 00n=l

    then the

  • 105

    series

    '"= 2anXn

    n=l

    converges a.e. in rn'" with respect to .

    Proof: We shall prove that the measure of

    '"{ cc l diverges }xErn a xn n

    n=l

    is zero.

    n

    We define = 2akxkk=l

    lim supn

    lim infn

    For mEIN we have

    = +

    S S sup Is - s (x) Iq p q -

    + sup -

    lim sup - lim infn n

    S sup s (x) - infp -

    sup {s (x) -p-

    S sup Is (x) -pLm p-

    2 sup Is - Ip

    For arbitrary E>O we define

    M - I > 2E } .Hence for all mE IN

    := {co

    xErn sup - > E }PLm

    and thus

    S for all mEIN .

    Using the Kolmogorov inequality, we compute

  • 106

    'Y(M) s lim sup 'Y(Mm)mlim sup 'Y({

    00

    I - s (x) I > })x IR supm m -00

    s lim sup 22 1 2 = 0a km k=m

    We have seen that series of the form

    converges a.e. in

    i\ = 2n=l

    00

    with

    n=l

    and thus defines a linear measurable

    functional. The converse is true as well.

    We introduce some notation. Let us define

    r 0 r 1 r 0 r

    where the number takes the n-th position rand

    00on IR

    Theorem 4B: Consider a linear measurable functional i\ defined

    Then we have00

    1) 2!i\(en ) 12

  • 107

    subspace of lRoo

    = {00

    xEIR there exists an NEIN such that x = 0n for } .

    Theorem 5B: Consider a measurable set

    following statements are equivalent,

    B C lRoo

    Then the

    (' )

    (2 )

    ,(B) > 0

    ,(y + B) > 0 for all

    Proof: It is easy to see that the theorem holds in the case

    where , is the gaussian measure ,n

    in the finite dimensional space

    Let us first prove the restricted form of theorem 5B, where B

    are cylinder sets. Consider the set

    C = B Ell lRoo-n

    for an nEIN and a Borel set B C IR n

    arbi trary there exists an meIN

    consider two cases.

    Assume that ,(C) > O. For

    such that yElRm. We have to

    m n Then we regard y as being in the space IRn, and the

    result follows by the finite dimensional case.

    m > n: We rewrite the set C as

    C = B Ell lR oo - n = B Ell IRm- n Ell !Roo-m ,

    where B Ell !Rm-n is a Borel set in !Rm. Then we refer to the case m

    n .

    To prove the full version of the theorem, take a non-negative

    integrable function00

    f:IR--->IR.00

    Consider yEIRO ' then there exists

    an mE IN such that yElRm If n m we have

    TyEn(f) EnTy(f),

    where (Enf)(zl ,z2" "zn) JlRoo-n

    and = (Z1 ,z2'" ,zn)

    By the Fubini theorem the function

  • for all nEIN .

    108

    Enf : ([;

    is integrable over IR n with respect to and

    J = JIRn IR

    Notice that f i 0 implies that Enf i 0 .

    Assume that

    J > 0 .IR

    Hence

    J > 0IR n

    Using the theorem for the finite dimensional case and the relation

    T E (f) = E T (f) ,Y n n Y

    we get that

    for all nEIN .

    Hence

    Setting f

    o < J = JIRn IR

    'B' the theorem follows.

    Corollary 6B: For arbitrary linear measurable functional A we

    have that

    C ,

    where denotes the domain of A •

    Proof:

    will prove that

    contains a linear subset E of full measure. We

    C E C •

    Assume that there exists an E \ E. For positive t we

    define the sets

    + E

  • Since E

    109

    is a linear set, the sets are pairwise disjoint for

    different indices. Using theorem 5B, we get

    Hence

    -r (Et ) > °is a family of pairwise disjoint sets with positive

    measures. This contradicts the fact that00

    -r(IR)=l

    Proposition 7B: Denote by A a linear measurable functional on

    IRoo

    Then we have that00

    l IA ( en) I2 < 00 •n=l

    Proof: For x = (Xl ,x2'. "xn" .)EIRoo

    we define

    = (0,0, .. ,O,xn+ l ,xn+2'" )EIRoo

    Thus we have that

    n

    l xkek + .k=l

    For we get that

    n

    = l XkA(ek) + =k=l

    J =00IR

    where = ) is measurable for all n and has the same

    domain of definition as A Hence exp ( i- A( ) and )

    integrable00

    are over IR

    For arbitrary u > ° we getnJ ooeXP[i'u l

    IR k=l

    11 JeXP[i,u'A(ek)t dt • Jk= 1 IR .,J2; IR oo

    n 2 2 Jn IA(ek)1 ).k=l IR

    Elementary computation ascertains that

  • 110

    Using this we get that

    nII I exp(_&u 2 2 IA(ek) 1 2 ) for all n .ffi k=l

    00

    Assume that

    for

    limn_

    2IA(en) 1 2 = 00. This yieldsn=l

    J aIR

    all u > a. By the dominated convergence theorem

    J J limIRoo !Roo n-

    J00 1 • ( 1,IR

    we get

    which is a contradiction.

    Proposition BB: Consider a linear measurable functional A. If

    then

    A a a.e.

    Proof: A(en) a for all implies that

    and define

    L a }

    s a }E+ = -E and since the measure

    E {xEE

    AI 00 = aIRa

    Let E denote the domain of definition for A

    E+ xEE

    Since A is linear, we have that

    is symmetric, we get

    +(E )

    Then it is obvious that

    for every and likewise for E

  • 111

    are constant with respect to every finite numberboth 1 + and 1E E

    of variables. From the Kolmogorov zero-one law we get that

    either 0 or and likewise with E Then

    '"Y(E+) = '"Y (E- ) =

    and hence

    '"Y({ xEE I o }) '"Y(E+ n E-}

    +'"Y (E ) is

    We are now ready to prove theorem 4B.

    Proof (Theorem 4B): Since 1) amounts to proposition 7B, it

    remains to prove 2}, i.e.00

    I a.e.n=l

    Since by proposition 7B00

    I IA (en) I2 < 00 rn=l

    theorem 3B ascertains that00

    In=l

    converges a.e. in moo and defines a linear measurable functional A .

    We must prove that A = ". For we have00

    A(ek) = I A(en)n=l

    By proposition 8B we conclude00

    I a.e.n=l

    For arbitrary functionals "1 """n' we wish to calculate the

    measure of sets of the form

  • 112

    where

    { co I k=l, .. ,n }xE:lR a k< bk , ,ak,bk are real numbers. We shall often use the

    shorter notation

    = {co

    I }[i\>c] xE:1R

    Lemma 9B: Consider a sequence {fn}n=l of measurable functions

    in !R'" with values in lR. If {fn}n=l converges almost everywhere

    to a function f, i.e.

    f n f pointwise for a.e.co

    xE:lR

    then to every cE:lR there exists a sequence {Ck}kE:m fulfilling

    kck- - - ...., c

    and

    -r[f>c]

    Proof: Assume that [f=c] o and denote by M the

    zero-measure set on which {fn}n=lget

    does not converge to f Then we

    pointwise on the set

    IR'"

    l[fniC] n 'l[f>c]

    lR'" \ (M U [f=c]) i.e. almost everywhere on

    By using the dominated convergence theorem we get

    -r[fniC] ----.... -r[f>c] .n

    The case when the set [f=c] has positive measure now follows.

    We find a sequence {Ck}kE:m of real numbers fulfilling

    ck c and -r[f=ck] = 0 for every kEmk

    This is possible, since there would otherwise exist an

    uncountable family of disjoint sets with positive measure,

    contradicting the fact that the measure -r is finite.

  • 113

    By applying the above established result to the zero-measure

    sets [f>ck] and using the dominated convergence theorem, we get,

    = limk--

    The measure of the sets

    lim lim .k--n--

    where

    f 2>c 2 , .. , fm>c m] ,

    f 1 ,f 2' .. ,fm are measurable functions, can be calculated in a

    similar way.

    We shall apply the lemma with a linear measurable functional in

    ffioo , denoted by A. It has been proved earlier that00 00

    n

    xErnOO IJ

    2anXn with 2 < 00n=l n=l

    We denote by AN the sum of the terms with indices from

    Then by lemma 9B we get

    = lim lim ,k--N--

    where the sequence {ck}kErn converges to cErn .

    We now calculate

    1: akxk i cm }k=l

    ... .. dxnffin

    where

    to N.

    n

    M = { I 2akxk i Cm }k=l

    By choosing an orthogonal transformation in

    spanned by

    rn n sending the line

    n

    lI!!n112 = l into the line spanned by the first natural basisk=l

    vector in rnn, we get by using the transformation theorem

  • 114

    J

    112}

    00

    J

    cm/ 112

    We choose A wi th By first lettingi.e. 2a;n=l

    go to infinity the above expression reducesand afterwards m,n ,

    to

    with A

    00

    ,[A>C] = J

    c

    being a linear measurable functional in00

    IR with00

    2 1A ( en) I 2 = 1n=l

    The expression can easily be extended to00 00

    cm

    = J ... J

    c,where denote linear measurable functionals in

    00

    IR all

    with00

    and the vectors

    2 1Ai (en) I 2 = 1n=l

    {ain

    for i=l, .. ,m

    orthogonal for different indices

    i . One hereby obtains an orthogonal transformation in rnN with

    A simple set theoretical argument together with the additive

    property of the measure give us the expression

    b1

    bm

    .. = ... J ..

    a 1 am

    where A1, .. ,Am are measurable functionals in00

    IR withco

    '\ IA.(e )12 =1 for i=l, .. ,m,L n

    and {a in

    n=l

    Ai(en)}nErn orthogonal for different indices i

  • 115

    c: Linear transformations in 00IR

    We shall extend unitary transformations of

    maps in 1R00 that preserve the gaussian measure.

    onto to

    Definition 1C: A weak measurable linear transformation in 1R00

    is a map

    A y 00Ax E IR

    where

    1) The domain of definition EA is a linear set of full measure.

    2) The map A is linear.

    3) Every coordinate function Ym IR is a linear measurable

    functional in00

    IR

    There are some comments in connection with this definition.

    1) According to the earlier paragraph concerning functionals in00

    IR we have that00

    2amnXnn=l

    with

    matrix

    00IR wein

    (infinite)

    L 2 .

    thebygiventhenisAtransformation

    {amn}m,nEm where the rows {amn}nEm are elements in

    2) By the maximal domain for a linear functional

    The

    understand the set

    If we let

    00

    { I 2A(en)Xn converges}n=l

    Em denote the maximal domain for the functional and00

    set E = n Em' we get thatm=1

    EA C E and = 1

    If EA # E we extend in the obvious manner the transformation A to

    the whole E, thus getting A maximally defined.

  • A

    116

    3) It follows from the earlier results that

    is uniquely determined by its values on

    [2 C EA and that

    It is indeed

    determined by its values on the set

    where

    en (0 I 0 I •• ,0,1 1 0 . . . )

    icoordinate number n

    4) The converse holds as well; if elements of [2 are taken as

    the rows of a matrix A = {amn}m/nEm I

    linear transformation in moo

    then A is a weak measurable

    Let A denote a linear bounded transformation

    A : [2 [2 .

    Then we extend A uniquely to a weak measurable linear transformation

    in by extending the domain of definition for the matrix

    given by

    a rnn = .

    Since the rows are elements of [2 I by the Parseval identity00

    '\ a 2 =L mnn=l

    00

    l 1< Aen/em >1 2n=l

    00

    ll< en/A*em >12 = IIA* emI1

    2< 00

    n=l

    the extension is well defined.

    Theorem 2C: Let u denote a unitary

    transformation. We extend u00

    to a weak measurable linear map in m

    by the matrix {u =mn Then the gaussian measure

    is invariant under the transformation U

    Proof: Consider nEm and ak/bkEm I k=l / .. ,n We define

    B { xEm00 I ak

  • and that

    -r(B)

    b,

    (21T)-ft n J ..a,

    117

    bn

    J .. .. dx nan

    U(B) = { I ak < bk for k=' , .. ,n } ,*where U denote the extension of the adjoint of U Since the rows

    *in the matrix of U are the columns in the matrix of U

    that00

    U(B) = { I a k< 2UikX i bk for k=' , .. ,n } .i='

    We define the functionals00

    we get

    k=', .. ,n ,

    U are pairwise orthogonal,

    fork=" .. ,n}

    and observe that since the rows in

    -r(U(B)) = -r{ I ak< bkb, b n

    (21T)-ft n J.. J .. .. dx na, an

    D: The gaussian measure on

    Consider the linear space00

    = { I for all iEIN }lC = ( z, ,z 2 ' ... , zn ' .. ) z.ElCl.and the Hilbert space

    00

    £2 (lC) = {co 2 I z 12 < 00 }Z E lC n

    n=1and introduce in lCn , nEIN , the gaussian measure

    where

  • 118

    and

    and dx,dy, ... dxndYn indicates the Lebesgue integration

    in 1R 2n .

    Then all former results concerning the gaussian measure are easilyIX)

    extended to this new gaussian measure on

    E: Hilbert-Schmidt enlargements

    Let }{, denote a real or complex Hilbert space. A

    Hilbert-Schmidt enlargement X, of }(, is itself a new Hilbert

    space containing as a linear dense subset and such that the

    inclusion mapping of }{ into X is a Hilbert-Schmidt operator, i.e.IX)

    for every orthonormal basis where 11,11 and

    denote the norms corresponding to the inner products

    respectively.

    and

  • 119

    Lemma 2E: Let A : HxH [ denote a map which is conjugate

    linear in the first term and linear in the second. If there exists a

    constant K fulfilling

    IA(X,y) I s K·llxll- flyII for all x,yEH,then there exists a linear bounded operator

    A : H H

    fulfilling

    A(X,y) for all x,yEH.

    We have the following fundamental result.

    Theorem 3E: There exists an orthonormal basis

    which is a complete orthogonal system in X .

    Proof: We define

    in H

    A(X,y) HxH ---..., [.

    By lemma lE there exists a constant K fulfilling

    !A(X,y)! K-llxll·fIyll for all x,yEH

    by which there exists an operator J : H H with

    for all x,yEH

    It is easily seen that J is a strictly positive (hence self-adjoint)

    operator and that if

    then00 00

    denotes an orthonormal basis in

    00

    H ,

    tr(J)

    00

    l IIen 11 2 < 00 ,n=l

    i.e. J is a trace class operator.

    Now choose an orthonormal basis

    eigenvectors for J with eigenvalues

    {bn}nEIN in H consisting of

  • 120

    We then have00

    1 ) The series {An}nEm is convergent, i.e. l An < 00 sincen=l

    J is of trace class.

    2) The orthogonality follows easily for n,mEm

    = = An = AnOOn,m

    3) It is obvious that {b } is total in H,n nEm

    Corollary 4E: There exists an orthonormal basis {bn}nEm in

    H, and a convergent series of strictly positive numbers {An}nEm

    such that

    {00

    '\ x bL n nn=l

    00

    l An0 I xn 1 2n=l

    Proof: We choose {A } as constructed in then nEm

    proof of theorem 3E and the corollary follows.

    F: The gaussian measure on Hilbert-Schmidt enlargements (cf. [23])

    Using the orthonormal basis in H and the positive

    eigenvalues {An}nEm from the above paragraph we get

    = = A for xEHn n n n

    Let us define00

    [2 { lC00 l IXnl 2 < 00 }x En=l

    00

    I 2 { c00 l Anlxnl2 < 00 }x En=l

  • 121

    we can define the gaussian

    By identifying

    !t - L2

    )( - 12

    via the orthonormal basis {bn}nEIN in !t ,measure a = on the Borel sets of )( as the measure

    a'Y

    on the corresponding Borel sets in L , and then get

    = 'Ya (Z2)

    Since unitary transformations of L 2 onto extend to orthogonal

    weakly measurable linear transformations in and the measure a'Y

    is invariant under these transformations, the measure is invariant

    under the corresponding extensions of the unitary maps between Hilbert

    spaces. In particular, we have that the measure does not depend on

    the chosen orthonormal basis in !t used for identifying !t with L2 .

    We would like to show that the gaussian measure does not depend

    on the selected Hilbert-Schmidt enlargement. This can be done by

    showing that if !t, and denote two different Hilbert-Schmidt

    enlargements then there exist a Hilbert-Schmidt enlargement Kfulfilling

    i=' ,2 .

    If f denotes a continuous function defined on !t with values

    and f 2 are equal on )(,

    in and f, and

    respectively, then

    are continuous extensions of f to !t, and

    hence equal

    almost everywhere.

    Definition , F: Let and denote

    Hilbert-Schmidt enlargements of a Hilbert space !t, •

    !t, is finer than !t2 if the identity

    I : !t ---> !t

    extends to a continuous and one-to-one map

    We say that

  • 122

    It is obvious that if *1 is finer than *2' then there exist

    a constant C > 0 such that

    for all xEl{. The smallest C fulfilling the above is the operator

    Lenuna 2F: Consider seminorms {II· Iln}nEIN and define 11'11 * in l{

    setting

    Ilxll; = 2n=l

    If a sequence fulfills

    and for every nEIN

    ------> 0p,q

    then

    a ,

    a .

    Proof: The proof amounts to an adjustment of the well known

    method of verification that the countable direct sum of Hilbert spaces

    is complete.

    Lenuna 3F: Let l{1 ' : l{ ------>

    is a continuous linear functional, then there exists a Hilbert-Schmidt

    enlargement l{2'2 •

  • 123

    Proof: Use theorem 3E to choose an orthonormal basis

    in H, such that {en}nEm are pairwise orthogonal with respect to

    Then

    with

    and

    n ' 1

    Expanding aEH we get

    a

    II e 11 21

  • 124

    * is well defined on

    We define the inner product

  • 125

    00

    2 Ilepll; < 00 •p=l

    Then the completion *2 of *' 0p

  • 126

    and thereby

    o .

    Lemma 4F: If denotes a

    Hilbert-Schmidt enlargements of a Hilbert space X,

    sequence of

    all finer than

    then there exist aa fixed Hilbert-Schmidt enlargements

    Hilbert-Schmidt enlargement

    nEIN .

    XO'

  • and conclude that

    127

    is well defined.

    Let denote the completion of and let {ep}pEm

    be an orthonormal basis in *,. Then we have00 00

    2 2 2p=l p=l n=l

    00 00 00

    2a 2 = 2a - c2n n nn=l p=l n=l

    We have to prove that H is a

    00

    22-n < 00 •n s l

    Hilbert-Schmidt enlargement of

    * . Take nEm We must show that H is finer than *n It isobvious that the inclusion mapping

    is continuous.

    It remains to prove that the extension of I over is

    one-to-one. Assume that {Xp}pEm C * is a Cauchy-sequence in the II- II

    Then for every nEm----> 0- norm and that IIxp 11 0

    We know that for every n

    ----->, 0 .p,q

    the enlargement is finer than *0 and

    hence

    Since

    ----> 0p

    implies

    o .

    we get by lemma 2F that

    00

    '\ a -llx 11 2L n p nn=l

    II xp II ----> 0pwhich proves that the inclusion mapping of H into *0 is one-to-one.

    Since the inclusion mapping of into *0 is one-to-one, the

    injection of H into *n must be one-to-one as well.

    We are ready to prove the announced principal result.

  • 128

    Theorem SF: To every pair of Hilbert-Schmidt enlargements of a

    Hilbert space there always exists a Hilbert-Schmidt enlargement that is

    finer than any enlargements from the pair.

    and H2'

  • 129

    the functionals {An}nEm are continuous in both the - norm and

    the II· 11 3 - norm and that these functionals are dense in Ut, 1)' and (K3'

  • References

    [1] V. Bargmann, On a Hilbert Space of Analytic Functions and an

    Associated Integral Transform Part I,

    Pure Appl. Math. Sci. 14 (1961), page 187-214.

    [2] O. Bratteli and D. W. Robinson, Operator Algebras and Quantum

    Statistical Mechanics II, Springer Verlag, (1981).

    [3] J. Cook, The mathematics of second guantisation, Trans. Amer.

    Math. Soc. 74 (1953), page 222-245.

    [4] W. Feller, Acknowledgment, Proc. Nat. Acad. Sci. USA, Vol 48

    (1962) page 2204.

    [5] R.J. Glauber, Coherent and Incoherent States of the Radiation

    Field, Physical Review 131 (1963), page 2766-2788.

    [6] Alain Guichardet, Symmetric Hilbert Spaces and Related

    Topics, Lecture Notes in Mathematics 261, Springer Verlag

    [7] R.L. Hudson and J.M. Lindsay, A non-commutative martingale

    representation theorem for non-Fock Quantum Brownian Motion,

    Journal of Functional Analysis 61 (1985), page 202-221.

    [8] R.L. Hudson and K.R. Parthasarathy, Quantum Ito's Formula and

    Stochastic Evolutions, Commun. Math. Phys. 93 (1984),

    page 301-323.

    [9] R.L. Hudson and R.F. Streater, Ito's formula is the chain

    rule with Wick ordering, Physics Letters, vol 86A, number 5,

    page 277-279.

    [10] P. Kristensen, L. Mejlbo, E.T. Poulsen, Tempered distri-

    butions in infinitely many dimensions 1,11,111

    I Commun. Math .Phys. 1 (1965) page 175-214.

    II Math. Scand. 14(1964) page 129-150.

    III Commun. Math. Phys. 6(1967) page 29- 48.

    [11] W.H. Louisell, Quantum Statistical Properties of Radiation,

    Wiley 1973.

    [12] J.E. Moyal, Quantum Mechanics as a Statistical Theory,

    Proceedings of the Cambridge Philosophical Society 45 (1949)

    page 99-124.

    [13] Klaus M0lmer and Wojtek Slowikowski, A new Hilbert space

    approach to the multimode sgueezing of light,

    J.Phys. A:Math.Gen 21 (1988) page 2565-2571.

    [14] E. Nelson, The free Markoff field, J. Funct. Anal. 12(1973)

    page 211-227.

    [15] I.E. Segal, Tensor algebras over Hilbert space I, Trans.

    Amer. Math. Soc. 81 (1956), page 106-134.

  • [ 16]

    [ 17]

    [18 J

    [19 J

    [20]

    [21 ]

    [22]

    [23 ]

    [24J

    [25]

    [26]

    131

    I.E. Segal, The complex-wave representation of the free

    Boson field, Topics in Functional Analysis, Adv. in Math.,

    Supplementary Studies Vol 3 (1978), page 321-345.

    I.E. Segal, J.C Baez and Zhengfang Zhou, Introduction to

    algebraic and constructive guantum field theory, Princeton

    Univ. Press, to appear.

    G.E. Silov and Fan Dyk Tin, Integral, Measure and Derivative,

    Nauka, Moscow (1967), (Russian).

    B. Simon, The Euclidean (Quantum) Field Theory,

    Princeton University Press (1974).

    w. Slowikowski, Ultracoherence in Bose algebras, Advancesin Applied Mathematics 9 (1988) 377-427.

    w. Slowikowski, Bose algebras of operators and the Wigner-Weyl theory, to appear.

    w. Slowikowski, Quantization of Questionnaires, Math.Comput. Modelling. Vol 10 No.9 (1988), page 697-709.

    w. Slowikowski, Concerning pre-supports of linear probabilitymeasures, Lecture Notes in Mathematics 794, Proceedings

    Measure Theory Oberwolfach 1979, Springer.

    W. Slowikowski, Measure Theory Applications to Stochastic

    Analysis, Proceedings, Oberwolfach, Germany 1977, Lecture

    Notes in Math. vol 695, page 179-191, Springer 1978.

    A. Unterberger, Les operateurs metadifferetiels, Lecture

    Notes in Physics 126 (1980), page 205-241.

    L.V. Zyla and Rui J. P. deFigueiredo, Nonlinear System

    Identification based on a Fock space framework, SIAM J.

    Control and Optimization, Vol 21 NO.6 (November 1983),

    page 931-939.

  • real wave representation 2,3,53,72,77,90,91

    second quantization 30Stone 29,30tame 1°°,101 , 102total 16,53,66,67,92,118,120ultracoherent 57vacuum 1,4,10,20,23,65,83,85,92value of 66,90weak measurable linear trans-

    formation 115,116,121Weyl 1,45,52,91Wick 22,45,59,79,81,83,88Wiener 16,20Wigner 3,86

    Subject index

    annihilation operator 1,3,6,8,10,21,25,61,64,79,81,83,85,92

    anti-commutator 1anti-normal 92base space 1,2,4,10,23,65,83,85,

    87Bose albebra 1,2,3,4,6,10,11,20,

    22,23,65,79,83,84,85,88- extende 36,65,83- Fock space 1,2,3Campbell-Baker-Hausdorff 50,51,

    59,91,92coherent 2,33,34,43,44,63,64,66,

    83,89,92commutation 9,10,11,12,45,55,59,

    60,65,82,85,93complex wave representation 2,53,

    66,69,70,71,78,80,87,90conjugation 1,2,3,53,54,55,57,69,

    73,74,76,77,80,84,88,89,90creation operator 1,3,6,8,10,21,

    25,64,79,81,83,85,92cylinder set 96,97,101,107derivation 1,6,10,11,30Fourier transformation 25,31,32,71free commutative algebra 1,4- product 1,4,18gaussian content 96- measure 3,68,76,78,80,86,96,99,

    107,115,116,117,118,120,121,122Halmos 27,94Heisenberg 33,43Hermite 2,31,32Hilbert-Schmidt enlargement 68,71,76,

    80,86,87,90,91,118,120,121,122,125,126,127,128

    Kolmogorov extension theorem 97inequality 103,105

    - large number theorem 104- zero one law 102,111Leibniz rule 6,30,31,36,37,39,61,82linear measurable functional 102,103,

    106,108,109,110,111 ,113,114,115,117,122,123,128,129

    Nelson 27one-parameter group 29t- picture 65,84,88

    product 59,61,62,63,64,65,78,88value 72,90