what is the logical form of probability attribution in qm

Upload: jhahab

Post on 02-Jun-2018

223 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    1/26

  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    2/26

    WHAT

    IS

    THE LOGICAL FORM OF

    PROBABILITY

    ASSIGNMENT IN QUANTUM

    MECHANICS?*

    JOHN F. HALPINt

    Department of Philosophy

    Oakland University

    The

    nature

    of quantum

    mechanical

    probability

    has often

    seemed

    mysterious.

    To shed some light on this

    topic,

    the

    present paper analyzes

    the

    logical

    form

    of

    probability

    assignment in quantum mechanics. To begin the

    paper,

    I

    set out and

    criticize several attempts to analyze

    the

    form.

    I

    go

    on

    to

    propose a new forn

    which utilizes a

    novel, probabilistic

    conditional and

    argue that this

    proposal is,

    overall, the best rendering of the quantummechanical probability assignments.

    Finally, quantum mechanics aside, the discussion here has

    consequences for

    counterfactual

    logic,

    conditional

    probability,

    and

    epistemic probability.

    Most

    of the

    interesting

    and

    difficult

    interpretive

    issues

    in

    quantum me-

    chanics

    (QM)

    are

    closely

    tied to

    that

    theory's probabilistic

    nature: QM,

    rather than making

    unequivocal/deterministic predictions, assigns prob-

    abilities to

    the

    possible

    results

    of

    observation

    or

    measurement. For

    this

    reason QM

    is

    said

    to be an indeterministic

    theory.

    However, the prob-

    ability assignments of QM, and in fact, probabilities in general, arephilo-

    sophically vexing. Indeed,

    the

    question

    of

    the

    nature of

    quantum

    me-

    chanical

    probability

    assignments

    constitutesa

    significantpart

    of the

    problem

    of

    interpretingQM.

    That

    question

    is to be

    pursued

    in

    this

    paper.

    As

    I

    see

    it,

    this

    question

    of

    quantum

    mechanical

    probability

    assignment

    has two

    parts:

    What

    sort of

    interpretation

    should

    we

    give

    to

    probability

    as

    it occurs

    in QM?

    What is the logical form of the probability assignments of QM?

    The

    first

    of these

    questions

    is

    the fundamental one. It

    is the

    traditional

    question asking,

    for

    example,

    are

    quantum

    mechanical

    probabilities

    de-

    grees

    of belief? relative

    frequencies? propensities?

    or

    perhaps something

    nonclassical?

    An

    answer

    will

    explain

    the

    meaning

    of

    "probability"

    as

    used within

    QM.

    But the second

    question

    is

    also of

    importance;

    as

    we

    *Received February

    1988;

    revised

    January

    1989.

    tI

    want to

    thank

    Arthur

    Fine,

    Alan

    Nelson,

    two

    anonymous

    referees for

    Philosophy of

    Science, and, especially, Paul Teller for comments on an earlier draftof this paper. These

    were

    extremely

    valuable. Research

    for this

    paper

    was

    supported

    by

    a

    Faculty

    Research

    Grant from Oakland

    University.

    Philosophy f Science,

    58

    (1991) pp.

    36-60.

    Copyright X

    1991

    by

    the

    Philosophy

    of

    Science Association.

    36

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    3/26

    LOGICAL FORM OF PROBABILITY ASSIGNMENT 37

    will

    see,

    it

    is in some

    ways

    a

    preliminary

    to

    the first

    of

    these

    questions.

    And, as well, this question

    of

    logical

    form is

    surprisingly difficult.

    The main emphasis of this paper is the second question. The goal here

    is

    to see how best to think about the probability assignments of QM to

    make the

    best sense of these

    assignments.

    So here

    I

    attempt to analyze,

    or,

    in

    case

    of

    vagueness, explicate,

    the

    logical

    form of

    quantum me-

    chanical probability

    attributions.

    To

    begin,

    I

    set out and

    criticize several

    attempts

    to

    analyze the

    form.

    I

    go

    on to

    propose a

    new form

    which utilizes

    a

    novel, probabilistic

    conditional and

    argue

    that this

    proposal is, overall,

    the best

    rendering

    of

    the

    quantum

    mechanical

    probability assignments.

    Finally, quantum mechanics aside,

    the discussion here

    has consequences

    for counterfactual logic, conditional probability, and epistemic proba-

    bility.

    1. Introduction.

    I

    have

    claimed that

    the

    question

    of

    the

    logical

    form

    of

    quantummechanical probability assignments

    is

    difficult. To

    begin to see

    the

    difficulty,

    consider

    a

    quantum

    mechanical

    example.

    A

    particle, say

    an

    electron,

    is

    assigned

    a state

    description

    or

    "wave

    function", If,

    a

    com-

    plex valued

    function

    defined on

    3-space. According

    to

    the

    quantum the-

    ory,

    a certain

    function

    of

    f,

    I112-

    W*f

    =

    the

    complex conjugate

    of

    If

    times If, gives

    the

    probability density

    for

    that

    particle's position.

    Roughly, this

    means that

    the

    electron

    is

    most

    likely

    to be found in

    regions

    where

    1I12

    is

    large.

    The

    straightforward

    or

    naive

    reading

    of

    such a

    claim

    is

    that,

    for

    all

    regions

    V

    of

    space,

    the

    integral

    over

    V of

    11,12

    ives

    the

    probability

    that the

    electron is

    in

    V.

    But,

    in

    general,

    there

    is a

    difficulty

    with

    this

    straightforward endering:

    on the received

    view

    of

    QM,

    that

    is,

    the

    Copenhagen interpretation,

    the

    electron is

    typically

    not in

    any region V,

    at

    least not

    in

    any

    small

    region

    V. According to this view, it is, with a few exceptions, only when a

    position

    measurement

    is

    made that

    a

    particle

    can be said to

    occupy

    a

    specific region

    of

    space. Moreover,

    the

    same

    is

    said about

    all

    interesting

    physical quantities:

    such

    quantities typically

    have

    no

    values until

    mea-

    sured.'

    Now,

    it

    would be

    wrong

    to

    assign

    nonzero

    probability

    to

    some-

    thing

    which is

    certainly

    false.

    So, assuming

    the

    received view of

    QM,

    because

    a

    typical

    electron

    certainly

    is

    not

    in

    V

    before measurement,

    it

    cannot

    have a

    nonzero

    probability

    of

    being

    in

    V

    before

    measurement.

    Hence,

    the

    straightforward reading

    of

    quantum

    mechanical

    probability

    ascription-as assigning probabilities that a particle possesses a position

    (or other)

    value-cannot

    on

    the

    received view be true in

    general,

    for

    example,

    cannot

    be true

    before measurement.

    The

    received

    view has

    gar-

    1QM assigns probabilities

    not

    only

    for

    the

    physical quantity position,

    but also

    assigns

    probabilities

    for

    the values of other

    physical quantities,

    for

    example, angular

    momentum.

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    4/26

    38

    JOHN F.

    HALPIN

    nered impressive support

    in

    the last twenty years through the work

    of

    John Bell and others.2

    This

    leads one to ask: If, for example, quantum

    mechanical position probabilities are not probabilities (in the-straightfor-

    ward, naive sense) that a particle has

    a

    position, what are they?

    In a standardQM textbook, Eugen Merzbacher notes that

    I

    12

    "is pro-

    portional to the probability

    that

    upon

    a measurement

    of

    its

    position the

    particle

    will be found

    in

    [a] given

    volume element"

    (1970, 36).

    But this

    remark eaves unclear the

    notion

    of

    probability-upon-measurement.

    In an-

    other standardtext Albert Messiah writes:

    . .

    .since

    the wave

    function .

    .

    . has a certain

    spatial extension

    one

    cannot attribute o a quantum particle a precise position; one can only

    define the

    probability

    of

    finding

    the

    particle

    in

    a

    given region

    of

    space

    when one carries

    out a

    measurement

    of position. (1976, 117,

    Messiah's

    emphasis)

    Later he makes

    a more

    general

    statement:

    One

    . . . abandons the fundamental

    postulate

    of

    Classical

    Physics,

    according

    to which all

    the various

    quantities belonging

    to a

    system

    take

    on well-defined values at

    each

    instant

    of

    time. One can

    only

    determine for each of these variables a statistical distribution of val-

    ues,

    which

    is

    the

    probability

    law

    of the

    results

    of

    measurement

    in

    the

    eventuality

    that

    such

    a

    measurement

    is

    performed. (1976,

    294,

    Messiah's

    emphasis)

    I

    take it that Messiah's eventualities may

    be

    hypothetical

    or

    counterfactual

    eventualities,

    and that

    the probability assignments "upon"

    measurement

    or "in

    the eventuality

    that

    a measurement

    is

    performed"

    are

    probabilities

    given

    the

    (possibly counterfactual) assumption

    that a measurement is

    per-

    formed. (The "probability aw" assigns probabilities even in the absence

    of

    actual measurement.)

    So,

    at

    least

    on

    the

    standard

    view, quantum

    mechanical

    probabilities

    are

    2Bell's (1965) argument

    shows that

    the

    hypothesis

    that

    all

    physical quantities

    have

    values

    leads to statistical predictions different

    from

    those

    of QM. (Bell assumes a statistical

    lo-

    cality condition,

    that measurement

    results are

    statistically independent

    of what

    measure-

    ments are made

    at distant locations, plus

    he in

    effect makes an idealization

    about

    mea-

    surements,

    that measurement

    devices are

    perfectly

    efficient and

    accurate,

    in order to

    derive

    the conflict with QM.) Now,

    the

    quantum

    mechanical statistics are

    confirmed

    in

    the lab-

    oratory, hence

    the

    hypothesis

    that all

    physical quantities

    have values

    is

    often

    taken to be

    proven false. However, ArthurFine and others have shown thatreasonable possessed value

    models exist (models

    which respect

    the

    observed quantum

    mechanical

    statistics) as

    long

    as detector inefficiencies

    can be

    assumed. But even

    if

    Fine's models are

    correct,

    we

    cannot

    read

    the

    quantum

    mechanical probability

    attributions

    as

    being

    about

    possessed

    values. On

    Fine's models,

    the

    possessed

    values are not

    distributed in

    accordance with the

    quantum

    mechanical

    statistics, (Bell's

    theorem assures

    this).

    Only

    the observed

    values are

    so

    dis-

    tributed. See Fine (1982)

    for some

    possessed

    value models.

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    5/26

    LOGICAL FORM OF PROBABILITY ASSIGNMENT 39

    in some way conditional on a measurement. But how so? There have

    been a number of

    attempts to say exactly what

    this means.

    In what fol-

    lows, I try to show that the extant proposals are all problematic.

    2.

    A

    Survey of Proposals.

    As

    we

    have

    seen, the probability assign-

    ments

    of

    QM

    are conditional

    upon

    the occurrence of a

    hypothetical

    or

    counterfactual measurement. Now,

    the

    resources

    for

    clearly formulating

    conditional assertions

    are at hand.

    First, mathematicians have developed

    the notion

    of

    a

    conditional

    probability, Pr(A/B),

    read "the

    probability of

    A

    given that B", and defined as P(A&B)/P(B) where P, a probability

    measure,3

    gives the

    unconditional probabilities. Also, logicians have de-

    veloped the semantics for counterfactual conditionals. Take an example:

    If the Earth's mass were larger, its gravitational attraction would be

    greater.

    This is

    to be analyzed as follows: Under certain (counterfactual) circum-

    stances

    in

    which the antecedent

    holds,

    the

    consequent

    is

    also true.

    Such

    counterfactual

    ituations

    are

    usually spelled

    out

    in

    terms

    of

    possible worlds.

    Possible

    world

    semantics

    are

    still

    controversial,

    but as a

    provisional

    def-

    inition

    we

    may take the following. (An

    A-world

    is

    defined to

    be a possible

    world at which A is true.)

    (0) A

    >

    B

    is true

    at a world w

    if

    and

    only

    if

    B

    is

    true at the

    A-worlds

    most

    similar4 to w.

    On

    this

    analysis,

    the above counterfactual about

    the Earth is

    presumably

    true because the

    worlds most similar to

    ours which contain

    a more mas-

    sive Earth will

    obey

    the law

    of

    gravitation

    and hence

    will

    involve

    greater

    gravitational

    attraction.

    Three

    facts

    about conditionals

    will

    be of

    interest

    here.

    First,

    on the

    theory of conditionals just sketched, A > (B&C) is equivalent to (A >

    B)&(A

    >

    C):

    A

    >

    (B&C)

    is

    true

    at

    world w iff

    (B&C)

    is

    true at the

    A-

    worlds

    most

    similar to

    w iff

    both

    B

    and C

    are true at

    the

    A-worlds most

    similar to w

    iff both

    A

    >

    B

    and

    A

    >

    C

    are true at w.

    Secondly,

    Modus

    3A probability measure,

    P,

    is defined so

    that 0 ' P(A)

    '

    1, P(A

    V

    -A)

    =

    1, and P(A

    V

    B)

    =

    P(A)

    +

    P(B)

    if

    P(A&B)

    =

    0.

    Furthermore,

    the domain

    of

    definition of

    P

    is

    to

    be

    a Boolean algebra. Moreover,

    P

    is sometimes taken to

    be defined on a

    sigma field of

    propositions, a set closed under countable disjunction and conjunction. If so, countable

    additivity is assumed. However, these additional

    properties

    of a

    probability measure are

    not pertinent to the present project.

    'The notion of similarity

    here

    must

    be construed in the appropriate way. As I see

    it,

    "most similar" comes to "similar enough in relevant respects". For one nice attempt to

    spell

    out this notion

    see Lewis

    (1979).

    Those

    skeptical

    of

    the

    similarity theory

    of coun-

    terfactualsshould substitute

    the

    word

    "relevant"

    for

    "similar"

    in the above

    definition. That

    change of wording helps to emphasize

    that much more needs to be said in order to clarify

    the

    operative

    notion of

    this

    definition.

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    6/26

    40 JOHN F. HALPIN

    Ponens

    is

    a valid inference

    for

    the conditionals described:

    if A

    and

    A

    >

    B

    are true

    at

    w, then

    B is true

    at the

    A-worlds

    most similar to w; but by

    assumption w is an A-world, so it must be an A-world most similar to

    itself (nothing else could

    be

    any

    more similar to

    itself ), hence

    B is

    true

    at w. So B follows from A and A

    >

    B.

    Finally,

    still

    assuming the theory

    of

    counterfactuals ust sketched, con-

    ditional excluded middle

    (CEM),

    (A

    >

    B) V(A

    >

    -B),

    can fail

    to

    be

    true.

    For

    example,

    it is

    plausibly

    false for the

    well-known

    Bizet-Verdi case:

    Either

    if

    Bizet

    and Verdi were

    compatriots,

    they

    would

    have been

    French; or

    if Bizet

    and

    Verdi were

    compatriots, they

    would have

    been

    Italian.

    Intuitively,

    this claim

    is

    false because

    neither

    disjunct

    is

    true;

    had

    Bizet

    and Verdi been

    compatriots, they might

    have

    been

    either

    French or

    Ital-

    ian. The analysis just given seems to

    bear out this

    intuition.

    This is so

    because there presumably

    are antecedent

    satisfying

    worlds

    most

    similar

    to the actual world at which both men are French and some just as similar

    at which both

    are

    Italian.

    So, assuming

    the

    usual

    truth

    functional defi-

    nition of

    disjunction, CEM

    fails

    to be

    valid

    for

    the

    semantics

    of

    >

    just

    sketched.

    Now,

    as

    noted,

    our

    definition

    of the

    counterfactual

    (0)

    is

    controversial.

    Still,

    the

    consequences just

    derived are

    generally accepted.

    It is

    these

    noncontroversial consequences which are

    most

    important

    n

    what follows.

    However, there

    is

    one

    exception

    to

    my

    claim

    about

    the

    noncontroversial

    nature

    of

    these consequences: Stalnaker's

    theory

    of

    the

    conditional does

    validate CEM. I want to digress briefly to describe this exception and the

    reason to stand by (0).

    On

    the

    original similarity theory

    of

    counterfactuals,

    Stalnaker's

    (1968),

    A

    >

    B

    is

    (nontrivially)

    true

    iff

    B

    is

    true

    at

    the

    A-world, s(A),

    most

    similar

    to the actual.

    (Here

    s is called a selection

    function; intuitively s(A)

    is

    the

    world

    or

    situation that

    would hold

    if

    A

    were

    true.)

    Notice

    that CEM is

    true

    on this

    theory

    because

    either B

    or -B is true at

    the world

    s(A),

    hence

    either

    A

    >

    B or

    A

    >

    -B

    is

    true

    (at

    the

    actual

    world). However,

    van

    Fraassen's "The End

    of

    the Stalnaker

    Conditional",

    an

    unpublished post-

    script

    to van Fraassen

    (1982),

    showed

    that

    Stalnaker's

    (1968) theory

    leads

    to

    Bell-like conflicts with

    experiment. (The problem

    arises from the

    treat-

    ment of counterfactuals like

    "if

    a measurement

    of

    quantity q

    were

    made,

    the

    result would

    be r". On Stalnaker's

    (1968) theory

    a

    unique

    measure-

    ment-of-q-world

    is

    determined

    by

    the selection

    function,

    hence so is

    a

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    7/26

    LOGICAL FORM OF PROBABILITY ASSIGNMENT 41

    measurement result

    r. This

    definiteness

    of

    values, here a "counterfactual

    definiteness",

    is

    what

    is

    used by

    Bell

    to

    derive

    conflicts

    with

    QM. See

    footnote 2.)

    Now, van Fraassen takes the problem just described to be a reason to

    favor a revised theory, Stalnaker (1981).

    The

    latter account admits that

    realistic contexts determine no unique selection function because there

    is

    in general no most similar A-world. Instead, this revision presupposes the

    existence of a set

    {sj}

    of selection functions all of which need to be con-

    sidered

    in a

    counterfactual analysis.

    Stalnaker does so

    by utilizing su-

    pervaluations over this set:

    a

    sentence

    of

    counterfactual logic

    is

    true iff

    it is evaluated as true relative to each selection function. So, for example,

    A > B is true iff for all selection functions

    sj

    in {sj},A > B is true relative

    to

    si, that is,

    iff for

    all

    j,

    B

    is

    true at

    sj(B).

    Now,

    CEM is true on

    this

    account

    because, as

    we

    have

    seen

    above,

    it

    is

    true

    with

    respect

    to

    an

    arbitrary

    election function

    s.

    Unfortunately,

    much the

    same

    thing

    can be

    said about

    the result derived

    by

    van

    Fraassen;

    the

    conflict with

    experi-

    mental

    results

    he derived

    for

    the

    original

    Stalnaker

    theory

    can

    be gen-

    eralized to

    apply

    to the

    revised

    account. See

    Halpin (1986)

    for

    an ar-

    gument showing that

    both versions

    of

    the

    theory

    lead to

    conflict with both

    QM

    and

    experimental

    results.

    I take it, then, that Stalnaker's theory in either version is not appro-

    priate

    for

    the quantum

    mechanical

    context.

    However,

    the

    other well-known

    similarity

    heory

    of

    counterfactuals,

    David Lewis's

    (1973),

    has

    also

    seemed

    suspect to many observers. On

    Lewis's account

    (put roughly)

    A

    >

    B

    is

    nontrivially true

    iff

    A&B

    is true at some world

    more

    similar to the

    actual

    world

    than

    is

    any

    world

    making

    A&

    -B

    true.

    Now,

    Lewis assumes

    that

    worlds

    can

    be more and

    more

    similar to the

    actual world without the

    existence

    of

    any unique

    most

    simnilar

    world.

    For

    instance,

    take Lewis's

    example (1973, 20). Suppose

    that

    in the

    actual

    world a line

    is

    exactly

    1"

    long. Then a world in which the line is

    1

    /4"1

    long will be less similar to

    the actual

    world

    than

    is some

    world

    in

    which it is

    1

    1/5

    long.

    And some

    world

    in

    which

    it is

    1

    /6"1

    long

    is more similar still to the actual world

    than

    any

    of

    the

    worlds

    in which the

    length

    is

    1

    1/51.

    And so on.

    (As

    Lewis

    notes, these claims about similarity

    will hold

    only

    in

    some

    contexts;

    I

    assume such a

    context

    in

    what

    follows.) Then,

    on Lewis's

    theory

    we

    can

    truly say that,

    for

    any

    n

    >

    0,

    if

    the

    line were

    longer

    than

    1"

    in

    length,

    then

    it

    would be

    shorter than

    l/[n".

    Unintuitively,

    this does

    not leave

    any

    length that

    the line

    might

    be

    if it were

    longer

    than 1". This

    sort

    of ob-

    jection has been made many times; for example see Stalnaker (1981).

    Our

    brief discussion

    so

    far

    suggests

    that

    the worlds relevant

    for

    coun-

    terfactual

    analysis, given

    an antecedent

    A,

    should

    include

    more

    worlds

    than

    just s(A). But,

    to

    continue the

    example above,

    a world in which

    the

    line has

    length

    1

    /6"1, though

    not as similar

    to the actual

    world as other

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    8/26

    42 JOHN

    F. HALPIN

    worlds

    in

    which the length

    of

    the line

    is

    greater than

    1",

    is

    nonetheless

    a world

    that might

    occur.

    In

    some

    contexts,

    at

    least,

    it is

    similar

    enough

    to the actual world to be worthy of consideration when the antecedent is

    "the

    line

    is

    greater

    than

    1"

    long". So,

    I

    suggest

    a

    compromise

    between

    these theories which

    in

    effect takes the selection function

    to have a set

    of worlds

    as values, that is, the set

    of

    worlds similar

    enough to the actual;

    this is

    (0). This proposal differs

    from

    Lewis's

    in

    that we are

    asked to

    consider a set of

    most

    similar worlds

    (for a given

    antecedent)

    even

    though

    some

    of

    these

    are

    not

    as

    similar as others to the

    actual. And (0) differs

    from

    Stalnaker's (1981) because

    on

    Stalnaker's

    theory

    (even

    as

    revised),

    there

    is a

    kind

    of

    counterfactual

    definiteness;

    that

    is, relative

    to

    each se-

    lection function "if a q-measurement is made, then r would result" is true

    for

    some

    r

    (because

    each selection function

    picks

    out a

    unique q-

    measurement

    world).

    This counterfactual definiteness

    (relativized

    to a se-

    lection function)

    is

    sufficient to

    derive

    Bell's conflict

    with

    experiment.

    (Because

    counterfactual definiteness

    holds with

    respect

    to an

    arbitrary

    selection function

    s,

    one can also derive as true

    relative

    to s

    a certain

    claim C about quantum experiments. Because s

    is

    arbitrary, Stalnaker's

    theory

    takes

    C to be true. But C

    is

    refuted

    in

    the

    laboratory.) See Halpin

    (1986).

    On

    the

    other

    hand, accordingto (0),

    "if

    a

    q-measurement

    s

    made,

    r would result" will typically not be true for any r because in typical

    quantum

    mechanical

    cases,

    r

    does not result

    in

    all

    of

    the most similar

    q-

    measurement

    worlds. There is

    no

    sense

    in

    which

    counterfactual

    definite-

    ness is

    true according to (0) because

    on

    this definition there

    is

    no

    inter-

    mediate

    stage

    of

    truth

    value

    assignment

    at which

    sentences

    are

    evaluated

    with

    respect

    to a

    single q-measurement

    world.

    A

    number

    of

    authors

    have

    suggested

    versions

    of

    (0).

    Most

    important

    for our

    purposes

    is

    Wessels

    (1981)

    who

    gives

    an

    analysis

    for

    measure-

    ment counterfactuals.

    Her

    idea, basically,

    is

    that

    for

    antecedent "a mea-

    surementof quantity q on system s occurs", we take the possible worlds

    most similar to the actual to be

    just

    like

    the

    actual with

    respect

    to the

    system's

    quantum mechanical

    state,

    and

    its

    possessed

    values.

    Further-

    more,

    the influences

    acting upon

    the

    system

    in

    such

    most similar

    possible

    worlds must

    be

    just

    like those

    acting

    on the

    actual

    system

    except

    for

    the

    influences of a measurement device

    set

    to

    measure q.

    Unfortunately this

    is just a schematic

    description.

    It

    does

    not

    give

    a full

    account

    of

    the

    possible

    worlds

    and

    what counts as

    similarity.

    We

    would like

    to

    know

    what

    values

    are

    possessed,

    and what influences

    a

    measurement

    device

    has

    on a

    system

    in

    each member

    of

    the set

    of

    most similar worlds.

    These

    questions

    involve controversial unknowns.

    A

    fuller account will

    require

    a

    complete

    description

    of

    physically possible worlds, and

    that

    awaits a

    solution to

    the

    quantum measurement

    problem. Still,

    Wessels's account

    shows us

    how to

    begin

    to

    flesh

    out

    (0).

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    9/26

    LOGICAL FORM OF PROBABILITY

    ASSIGNMENT

    43

    Finally, we are

    preparedto address

    the main business

    at hand:

    quantum

    mechanical

    probabilityattribution.

    A

    number

    of

    proposals for

    the

    logical

    form of quantum mechanical probability attributions have been given,

    utilizing conditional

    probabilities

    and

    counterfactuals.

    To

    help

    describe

    the

    proposals,

    I

    will

    use several abbreviations.

    Let

    R

    stand

    for

    the

    state-

    ment that "value

    r

    results

    upon

    measurement".5

    Let

    M

    stand for

    "quantity

    q is

    measured"

    Now, for all AP,q, r there is

    a p such

    that the following

    holds:

    If

    f

    is the state of a

    system,

    then

    QM assigns a

    probability p to

    R

    given that

    M.

    It

    is

    the

    logical

    form of such

    assignments we wish

    to

    uncover.

    Suppressing

    mention

    of

    f

    and

    q,

    the

    Merzbacher and

    Messiah

    version of

    this attribution

    s

    roughly:

    (*)

    R

    has

    probability p

    given (hypothetically) that M.

    I

    have

    seen three

    construals of (*) advanced:

    (a) Pr(R/M)

    =

    p,

    so

    that the attribution is a

    conditional

    probability given that a

    measure-

    ment

    has

    occurred,

    (b)

    P(M

    >

    R)

    =

    p,

    so

    that the attribution s an

    assignment

    of a

    probability

    to

    a

    conditional,

    and

    (c)

    M

    >

    P(R)

    =

    p,

    so

    that the

    probability

    attribution

    is

    made

    only

    on

    the

    counterfactual as-

    sumption

    that a

    measurement has occurred.

    In

    the

    following sections,

    I

    will

    try

    to show

    that each

    of

    (a),

    (b), and

    (c)

    have

    problems

    as

    readings

    of

    (*).

    I

    will

    go

    on

    to

    suggest

    an

    alternative

    explication. But we should now ask, what features are desirable in an

    explication

    of

    (*)? First,

    such

    an

    explication

    should

    allow the

    quantum

    mechanical

    probability assumptions

    the best chance to

    be true

    over

    the

    widest

    range

    of

    cases.

    This is

    just

    a

    principle

    of

    charity.

    Secondly, be-

    cause

    this

    explication

    of

    (*)

    is

    meant as a

    preliminary

    to

    interpretation,

    we

    do

    not want it to

    prejudge

    interpretive

    issues.

    That

    is,

    we

    should,

    for

    the sake

    of

    generality, prefer

    not

    to

    saddle

    QM

    with

    controversial as-

    sumptions.

    Ideally, then,

    an

    explication

    of

    (*)

    would

    not

    beg fundamental

    questions relating

    to

    the

    quantum

    mechanical

    interpretation

    problem.

    Fur-

    thermore, such an explication should, in so far as possible, not presup-

    pose

    any particularmetaphysics

    of

    possible

    worlds. As

    we will

    be

    dealing

    with semantical issues

    usually

    discussed

    in

    the

    possible

    worlds

    frame-

    work,

    this constraint

    will

    become

    important.

    In

    any

    case,

    the sum

    total

    5For

    generality,

    r can be taken to be either a

    single

    real

    number or a

    range

    of

    these.

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    10/26

    44 JOHN F. HALPIN

    of this second

    suggestion

    is

    that

    for

    purposes

    of

    generality, the best ex-

    plication

    of

    (*) will,

    to

    stay away

    from

    controversy, be neutral with re-

    gard to interpretive issues. Finally, of course, a good explication will

    uphold the physicist's

    intuitions that

    (*)

    involves

    probabilities given hy-

    pothetical measurements.

    3. Against Proposal (a).

    The first

    proposal, (a),

    that

    quantum mechan-

    ical probability attributions like (*) are conditional probabilities of the

    form

    Pr(R/M)

    =

    p, is, at first blush, perhaps the

    most

    obvious rendering.

    The probabilities

    of

    QM are probabilities given the performance of a mea-

    surement. Conditional probabilities

    of

    the

    form

    P(R/M), at least as usu-

    ally understood, arejust this. Since Kolmogoroff, such conditional prob-

    abilities

    have

    standardly

    been defined

    in

    terms of

    standardunconditional

    probabilities as described above: Pr(R/M)

    =

    P(R&M)/P(M). However,

    so

    long

    as

    this

    definition

    of

    probability

    is

    maintained,

    a

    significant prob-

    lem

    for

    proposal (a)

    exists. The

    argument

    for this

    claim comes from van

    Fraassen and Hooker (1976) as

    follows.

    Proposal (a) presupposes

    that a conditional

    probability

    is

    defined.

    For

    this to be

    so,

    the

    probability

    that

    m is

    measured, P(M),

    must also be

    defined and

    nonzero

    (for arbitrarym).

    Not

    only

    is

    this condition

    implau-

    sible on the face of it, but for arbitrarym, it must fail. (There are sets

    {mj}

    of measurement types

    such that

    (i)

    {mj}

    s of

    uncountablecardinality,

    (ii)

    no

    two elements

    of

    {mj}

    can be measured

    at once (they are incom-

    patible) andyet (iii) QM

    makes

    probability assignments conditioned upon

    each

    of

    the

    mj.

    (For example

    take the set

    {mj}

    to be defined so

    that

    mj

    is

    the measurement

    of

    spin

    of

    a

    particle

    in

    direction

    j,

    where

    j ranges

    over

    the set

    of

    all

    directions between

    0

    and

    90

    degrees exclusive.)

    Let

    Mj

    formalize

    "mj

    s

    measured".

    It is a fact of

    probability theory

    that

    because

    of

    the size

    of

    the set

    {mj},

    not all the

    incompatible Mj

    can receive nonzero

    probability. It follows that the conditional probabilities Pr(R/M) cannot

    be defined

    for

    all

    quantum

    measurements

    m; indeed, they

    cannot be

    de-

    fined

    for

    more than

    a countable subset

    of

    the

    uncountable

    set

    {mj}.

    So,

    proposal (a)

    cannot

    in

    general

    allow us to make

    sense

    of

    the

    quantum

    mechanical

    probability

    attributions

    (*)

    which are

    given

    for

    all

    mj.

    Before rejecting (a), however,

    we should consider

    responses

    which

    may

    be made

    to the van Fraassen-Hooker

    argumentjust given. First,

    someone

    might object

    to the above

    argument by suggesting

    that not all

    of the

    un-

    countable number

    of

    measurements

    in

    {mj}

    are such

    that

    they

    each

    might

    be performed in practice. (In practice we have no chance of being able

    to

    make

    so

    many

    discriminations

    for

    the same

    reason that

    we are

    not,

    for

    an

    arbitrary

    real

    number

    s,

    able to measure

    whether

    or

    not

    an

    object

    is

    exactly

    s

    units

    in

    length.) So,

    the

    objection continues, probability

    as-

    signments

    conditioned

    upon any

    but the

    experiments

    which

    might

    be

    per-

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    11/26

    LOGICAL

    FORM OF PROBABILITY ASSIGNMENT

    45

    formed

    in

    practice are unnecessary. Furthermore

    there is no reason to

    believe that uncountable

    sets

    of

    disjoint measurements

    exist which are in

    practice performable. This objection assumes that the quantum mechan-

    ical attributions of probability

    given measurements

    not in practice per-

    formable may be neglected

    for

    the project

    at

    hand.

    However, I would

    argue

    that one

    should at

    least try

    to

    make sense

    of

    all probability attri-

    butions of QM, even

    if

    some

    are

    not

    practical.

    Not to do so would be

    out

    of

    line with

    our first condition

    on

    a good explication of quantum

    mechanical

    probability

    assignments:

    that

    we

    make sense of

    the widest

    range of cases.

    The second response to the van Fraassen-Hooker

    argument against (a)

    is to countenance nonclassical probability measures with infinitesimal

    weights. According

    to this

    response,

    the failure

    of

    (a)

    is

    a result of

    the

    standard

    unconditional

    probabilities

    assumed

    in

    the

    definition

    of

    condi-

    tional probability.

    The van Fraassen-Hooker

    argument

    against (a) shows

    that the

    probabilities

    Pr(R/Mj)

    =

    P(R&Mj)/P(Mj)

    cannot

    be defined

    in

    general.

    This is

    accomplished by showing

    that

    not

    all the

    P(Mj)'s

    can

    have

    nonzero values.

    But

    they

    can have nonzero values

    if

    the notion

    of

    a

    probability

    measure

    P is

    extended

    to

    allow infinitesimal

    values

    in

    its

    range.

    In that

    case both numerator and

    denominator

    of

    the

    definition

    of

    Pr(R/Mj) may have infinitesimal values, yet the fraction itself would be

    a

    standard,

    finite real

    value.

    This

    currently popular

    counterproposal

    de-

    pends

    on

    the

    field

    of

    nonstandard

    analysis.

    For a

    review

    of

    this

    possi-

    bility,

    see

    appendix

    four to

    Skyrms (1980).

    So, according

    to

    this

    second

    response, (a)

    can

    be

    revitalized

    if

    we

    as-

    sign

    infinitesimal

    probabilities

    to the

    Mj's.

    But

    on

    the

    face

    of

    it,

    this

    would seem

    implausible.

    Are there

    really probabilities

    that

    a measure-

    ment

    will occur? One

    might

    think that there are

    not

    because

    said occur-

    rence depends typically

    upon

    the

    experimenter's

    free choice.

    Still, one

    may want to think of the experimenteras a complex quantummechanical

    system,

    so

    subject

    to

    probabilities.

    But

    the

    quantum

    mechanical

    proba-

    bilities

    are conditional

    probabilities

    and what we need

    for

    the

    definition

    are

    unconditional

    probabilities

    P(Mj).

    Finally,

    even

    if

    some

    other source

    of the infinitesimal

    unconditional

    probabilitiesexists,

    it would seem

    that

    these

    probabilities

    of

    measurements

    would

    sometimes

    be

    zero;

    actual sit-

    uations

    would

    sometimes

    absolutely

    rule

    out

    the

    performance

    of

    a

    given

    measurement

    mj;

    hence

    P(Mj)

    would be

    zero.

    (To

    take

    an

    extreme ex-

    ample,

    consider the case

    of

    a

    universe

    which

    contains

    only

    a

    few

    simple

    atomic

    systems.

    Because there is no measurement

    apparatus

    in such a

    universe,

    the

    probability

    of

    an

    m-measurement

    occurring

    would be

    zero

    for

    some

    if

    not

    all m.

    So, Kolmogoroff

    conditional

    probabilities

    are

    not

    defined

    here, yet QM

    would

    still

    seem

    to

    apply. QM assigns

    a

    state

    to

    the atomic

    systems

    and

    so

    assigns probabilities

    for

    measurement

    results

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    12/26

    46 JOHN F.

    HALPIN

    that could result if, per impossible, a measurement device had existed.)

    Hence, it would

    seem

    that

    this second response

    to

    the van Fraassen-Hooker

    argument, the

    infinitesimal probability proposal,

    will

    still not guarantee

    that all quantum mechanical probabilitiesare defined. Moreover, if, con-

    trariwise, there

    is

    some

    way

    to

    make the proposal work, this would re-

    quire taking a stand on several controversial interpretive issues (e.g., the

    assigning

    of unconditional

    probabilities

    o human

    systems)

    which

    we should

    avoid.

    For

    the

    sake of

    generality

    and

    avoiding controversy, a better ex-

    plication

    of

    the

    form of

    (*)

    is

    in

    order.

    The final response to the van Fraassen-Hooker argument rejects the

    Kolmogoroff

    definition

    of

    conditional

    probability. Realizing that the tra-

    ditionalKolmogoroffconditionalprobabilitiesare not always defined when

    we

    want

    them to

    be,

    both Karl

    Popper

    and Alfred

    Renyii have suggested

    theories

    for

    which conditional

    probabilities

    are

    fundamental: by fiat they

    exist,

    rather

    than

    by

    definition.

    (In

    a

    nutshell,

    these

    views

    stipulate that

    conditional

    probabilities

    exist

    given arbitrary

    condition

    A,

    with

    A

    from

    a

    set

    of

    conditions

    large enough

    to

    include

    all

    the

    incompatible Mj. Fur-

    thermore,

    for

    fixed

    A,

    the values

    of

    Pr(X/A)

    for

    variable

    X

    obey

    the laws

    of

    the probability

    calculus discussed

    in

    footnote

    3.)

    As

    long

    as this

    option

    is open

    for

    QM, the

    van

    Fraassen-Hookerargument

    is

    obviated because

    that argumentrelies on the Kolmogoroff definition of conditional prob-

    abilities

    in

    order to show that certain

    conditional

    probabilities

    are

    not

    defined.

    Indeed,

    van Fraassen and Hooker

    suggest

    that we

    understand he

    probabilities

    of

    QM

    in

    Popper's way. They

    show this to

    be consistent

    with the

    quantum

    mechanical

    probabilities.

    Van Fraassen and Hooker's

    suggestion

    that

    quantum mechanical con-

    ditional

    probabilities

    are

    fundamental

    (rather

    than

    defined

    in

    terms

    of

    un-

    conditional

    probabilities)

    seems

    right.

    But one should still

    want to

    give

    a more substantial

    positive

    account

    of

    what these

    probabilities

    are. Van

    Fraassen and Hooker suggest that the conditional probabilities are prob-

    abilities

    of

    conditionals:

    Pr(R/M)

    =

    P(M

    >

    R). Though

    I

    have no ob-

    jection

    to

    taking

    the

    probabilities

    of

    QM

    on

    the model

    of

    either

    Renyii

    or

    Popper,

    I

    think

    van Fraassen and

    Hooker's explication

    of

    these

    as

    probabilities

    of

    counterfactuals

    fails

    for

    two reasons.

    First, taking quan-

    tum mechanical

    probabilities

    to

    be

    probabilities

    of

    counterfactuals has

    unfortunate

    consequences

    which

    are

    developed

    in

    the

    next section.

    Also,

    van Fraassen and

    Hooker

    presuppose

    Stalnaker's

    theory

    of

    the counter-

    factual

    >.

    As described

    in

    the last

    section,

    this

    conditional would

    seem

    inappropriate or the context of QM; see Halpin (1986) for an argument

    that Stalnaker's

    theory

    leads all too

    easily

    to

    Bell-like

    conflicts

    with

    ex-

    periment. So,

    if

    we are

    to take the

    probabilities

    of

    QM

    as

    fundamental,

    non-Kolmogoroff

    conditional

    probabilities,

    hen

    we need

    to

    say

    more about

    just

    what

    they

    are.

    In

    section

    6

    below,

    I

    explicate

    the

    form of

    quantum

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    13/26

    LOGICAL

    FORM OF PROBABILITY ASSIGNMENT 47

    mechanical

    probability assignments

    in terms of

    a probabilistic condi-

    tional. The probabilities associated with that conditional might be taken

    as fundamental conditional probabilities. If so, the proposal of section 6

    can be taken to

    underwrite (a). So,

    I

    do not want to claim that (a) is

    wrong. But

    I

    will,

    in

    section

    6, try

    to

    give

    an

    explication of (*) which

    is

    more perspicuous.

    4.

    Against Proposal

    (b).

    Let

    PQM(R) symbolize

    the

    probability

    value

    assigned

    to

    QM

    to

    R

    given

    M.

    Now, proposal (b)

    states

    that

    a

    quantum

    mechanical assignment

    of

    probability, PQM(R)

    =

    p,

    has

    logical form Pr(M

    > R)

    =

    p. What we will show is that from (b) one can derive a weakened,

    but still undesirable,

    version

    of conditional excluded

    middle. Assuming

    (b), we have that Pr(M

    >

    R)

    =

    PQM(R). But,

    as

    well,

    because

    from

    QM

    we have

    PQM(-R)

    =

    1

    -

    PQM(R)

    and

    PQM(R&-R)

    =

    0,

    we also

    have

    P(M

    >

    R)

    +

    P(M

    >

    -R)

    = 1

    and P(M

    >

    (R&-R))

    =

    0,

    and

    so, because

    A

    >

    (B&C) is

    equivalent

    to

    (A

    >

    B)&(A

    >

    C), P((M

    >

    R)&(M

    >

    -R))

    =

    0. A version of conditional excluded

    middle follows:

    (CEM')

    P((M

    >

    R)

    V

    (M

    >

    -R))

    =

    1.

    I

    take it that

    (CEM')

    is

    an unfortunate

    consequence

    of

    proposal (b).

    On

    any interpretation

    of

    the

    quantum

    mechanical

    probabilities, objective

    or

    epistemic, (CEM')

    will

    not

    in

    general

    hold.

    Typically

    in

    a quantum

    mechanical

    world it will be a clear

    physical

    fact that neither

    disjunct

    is

    true.

    For

    instance, suppose

    an

    m-measurement

    is

    not

    performed, but that

    if it were to be performed then value

    r

    might

    result.

    But also

    suppose

    that

    some

    r' #A r

    might

    result. So

    if M

    were

    true,

    R

    might

    be

    true,

    but

    also

    might

    be false. Hence it

    would

    be

    false to

    say

    either that

    if

    M

    were

    true, then R would be true, or that if M were true, then -R would be

    true. So

    the

    disjuncts

    are

    both false.

    Hence,

    we will not

    assign probability

    one to

    the disjunction.

    Because

    proposal (b)

    leads one

    to

    do

    so,

    I

    take

    this

    proposal

    to

    be

    discredited.

    As we noticed

    in

    section

    2,

    conditional

    excluded

    middle

    has at least

    one

    defender,

    Robert Stalnaker.

    So

    a

    proponent

    of

    this

    theory

    would

    be

    comfortable

    with

    the

    consequence (CEM') just

    derived and

    would

    not

    take

    the

    argument

    to be a reductio

    of

    (b). However,

    as was

    also men-

    tioned

    in

    section

    2,

    Stalnaker's

    theory

    of

    the conditional leads all too

    easily to conflict with experimental results (Halpin, 1986). Moreover,

    even

    if

    the

    case

    against

    Stalnaker's

    theory

    and

    against

    CEM

    is

    not

    con-

    vincing,

    it should be clear that these

    positions

    are

    deservedly

    controver-

    sial. It

    would

    be

    unfortunate, then,

    to saddle

    QM

    with

    proposal (b)

    and

    the resultant

    controversy.

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    14/26

    48 JOHN

    F. HALPIN

    5. Against Proposal (c).

    Philosophers

    have

    frequently

    taken

    proposal

    (c) to be the best option

    for

    explicating

    the form

    of

    quantum

    mechanical

    probability ascriptions. (For example, see Skyrms, 1982.) But there may

    be a

    problem

    with

    (c).

    Consider the case

    in

    which

    a

    measurement of

    m

    is now performed, that

    is,

    M

    is

    now

    true,

    and for which

    the result

    of

    measurement, r,

    is

    immediately decided,

    for

    example, is immediately

    registered

    in

    the memory

    of

    a measurement device.

    According

    to

    (c),

    in

    this case

    P(R)

    =

    p. (This

    follows

    simply by

    Modus

    Ponens.)

    Do

    we want

    to

    say

    that the

    probability

    of result

    r is

    p? Perhaps

    not.

    One

    might argue

    that after such a measurement

    is

    performed

    and

    the

    result, r,

    is

    in,

    the

    truth value

    of R is no

    longer chancy.

    Rather

    it is

    decidedly

    true. Hence

    (c) is wrong here because its consequent assigns a chance

  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    15/26

    LOGICAL FORM OF

    PROBABILITY ASSIGNMENT

    49

    taken to be

    flips

    of

    a

    coin

    by

    John

    Halpin

    on 14

    May 1987-suppose

    they are HHTTHTHTT-then this

    ratio for

    heads

    is 4/9

    and

    the

    proba-

    bility of H in

    this reference class

    is

    4/9.)

    When the

    reference class is

    countably infinite (and assumed

    to be ordered

    in

    a

    sequence), the prob-

    ability

    cannot

    be defined

    as a

    ratio,

    but instead

    is

    defined to

    be

    the limit

    of these ratios as

    we

    think of the reference class as

    growing sequentially.

    We

    can

    make

    this

    precise:

    Let

    A,

    be the number

    of

    A's in

    the first n

    trials. The relative

    frequency,

    Rn,

    of A's

    in

    the first

    n

    trials

    is

    defined as

    An/n. Then for the case where the number of

    trials

    is

    finite and equal to

    m, one

    identifies the probability

    of

    A

    with Rm.

    Where the number of trials

    is

    countably infinite, the

    probability

    is

    taken to

    be

    the

    limit

    of

    Rn

    as n

    approachesinfinity.

    In

    this simplest form of the

    frequency interpretation,

    probabilities are

    defined

    in

    terms

    of

    actually

    occurring

    trials.

    But

    surely

    these cannot

    be

    what

    QM

    tells us about. It is

    possible

    that

    quantum

    mechanical

    proba-

    bilities differ

    from

    the associated ratios

    (or

    limits of

    ratios), just

    as a coin

    with probability

    1/2

    of

    coming up

    heads, may

    in

    a class of

    trials come

    up

    heads

    only

    4

    times

    in

    9.

    Indeed,

    for

    the case

    of

    a

    finite number

    of

    trials-

    and we know of

    no

    sequence

    of

    quantum

    mechanical trials which is

    not

    finite-quantum mechanical

    probabilities, which are sometimes irra-

    tional, cannot in general be given as relative frequencies; an irrational

    number

    by

    definition

    cannot

    be

    expressed

    as a

    ratio.

    Similarly, even

    if

    the reference class

    is

    infinite,

    and the

    probabilities QM

    assigns

    are

    the

    right

    ones,

    that the limit

    of

    relative

    frequency

    will exist and

    be

    equal

    to

    the

    probability assigned by QM

    is not

    guaranteed.

    The laws

    of

    large

    num-

    bers

    only

    tell us

    that

    a

    large

    difference between

    the

    probability

    and

    the

    limit of

    relative frequencies

    is

    unlikely;

    it

    is

    not

    impossible.

    A hard

    core

    empiricist may

    want, despite

    these

    objections,

    to hold

    that

    probability

    is

    just

    relative

    frequency

    in

    actual and

    (typically)

    finite

    ref-

    erence classes. Such a proponentof the frequency view will hold that to

    go beyond

    the

    extant

    frequency

    is to make an

    unwarranted

    idealization

    (e.g.,

    that

    there

    is

    a "virtual"ensemble

    to

    serve as infinite

    reference class

    or

    that there

    is such a

    thing

    as

    propensity).

    This

    may

    be

    so; however,

    our

    job

    here

    is

    to

    interpretQM,

    and

    that

    theory pretty

    clearly

    does ideal-

    ize;

    for

    example,

    it

    assigns probabilities

    even

    to

    unmeasured

    quantities

    such as those discussed earlier

    in

    section 3.

    So,

    the

    probabilities

    of

    QM

    are

    usually thought

    of not as

    relative

    fre-

    quencies

    defined

    over extant reference

    classes,

    but rather as the limits of

    relative frequency in a "virtual"ensemble; that is, the limits that would

    exist

    if

    hypothetical experimental

    trials

    (the

    reference

    class)

    were an

    in-

    finite sequence.

    (Such

    a view

    is sometimes

    taken to be a sort

    of

    propen-

    sity

    view

    rather

    than a

    frequency

    view.

    In

    any case,

    on

    this

    view the

    quantum

    mechanical

    assignments

    of

    probability

    read

    as

    a variant

    to

    (c):

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    16/26

    50

    JOHN F. HALPIN

    (An infinite sequence

    of

    measurement

    is

    performed)

    >

    (the relative

    frequency of A in the

    sequence

    is

    the quantum mechanical probability

    P).

    This reading

    is no

    better

    off

    than the frequency view for actual infinite

    reference classes: just as actual relative

    frequencies

    can

    differ from prob-

    abilities,

    so can counterfactual

    sequences.

    For

    example,

    if

    we

    were to

    flip a fair (probability1/2)

    coin an infinite number of

    times,

    it

    might land

    heads

    on

    exactly the even

    trials

    (one way

    to

    get

    relative

    frequency

    I/2)

    but also might

    land heads

    on

    every try (this

    too is a

    possibility).

    Both of

    the infinite

    sequences just

    described are

    very unlikely

    to

    occur

    (if the

    trials are independent)

    but

    are, nonetheless, possible. Generally, there is

    no

    guarantee that the limit

    of

    relative frequencies

    will

    or would be equal

    to the

    probability. Again,

    the laws

    of

    large

    numbers

    only

    tell us

    that a

    difference

    is

    unlikely,

    not

    impossible.

    So, though

    we

    may expect relative

    frequency

    in

    the long run to be close to

    the

    probability, we should not

    identify the two.

    The

    reasoning

    of

    the above

    paragraph

    is

    well-known and

    usually

    ac-

    cepted by interpreters

    of

    QM. So,

    for

    the

    rest

    of

    this section

    I

    assume

    that

    quantum

    mechanical

    probability

    is

    not

    to

    be

    given

    a

    frequency inter-

    pretation. Now, probably

    the most

    popular

    view

    among philosophers

    is

    that

    quantum

    mechanical

    probability

    is

    a kind of

    propensity

    I

    will

    call

    "chance".

    On this

    view, probabilities

    are

    probabilities

    for

    the single case,

    and are tendencies

    or

    dispositions

    that admit

    of

    degree. Typically,

    we

    estimate

    the chance

    given

    relative

    frequencies

    as evidence and

    vice versa.

    But chances are new sorts

    of

    theoretical entities that

    are

    not to be

    defined

    in

    terms

    of

    frequency.7

    We understandthese theoretical

    entities

    in

    terms

    of

    what

    QM says

    about

    probabilities

    and

    in

    terms

    of

    the

    probability

    cal-

    culus and

    its

    theorems, together

    with the evidential relation to

    relative

    frequencies. In this way propensities take their place as primitives within

    a theoretical network.

    I

    have

    argued

    that an

    epistemic interpretation

    of

    probability ascription

    is

    not

    appropriate

    within

    proposal

    (c)

    because such an

    interpretation

    would

    suggest

    that

    QM

    is

    about belief states.

    Furthermore,

    I

    argued

    that this

    leaves one

    with

    some

    version

    of

    a

    propensity interpretation:

    as

    a

    relative

    frequency account

    will not do

    here,

    one should take

    proposal (c)

    to

    be

    about

    chance.

    Chance

    applies

    to the

    future, however,

    and

    there

    is no

    chance

    or

    propensity

    for

    the

    past

    or

    present

    to be

    different

    from

    the

    settled

    way it was or is. This is an old view that goes back at least to William

    of

    Ockham.

    I think it

    is

    also a

    very plausible

    view

    of

    chance. David Lewis

    puts

    the

    point

    this

    way:

    7This is not

    to

    say

    that chance

    is

    an

    irreducible

    property

    of

    systems;

    it

    may supervene

    on

    physical properties.

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    17/26

    LOGICAL FORM

    OF

    PROBABILITY ASSIGNMENT

    51

    This

    temporal asymmetry

    of

    chance

    falls

    into place as part of

    our

    conception

    of

    the

    past

    as 'fixed'

    and

    the future as

    'open'-whatever

    that may mean. The asymmetry of fixity and of chance may be pic-

    tured by

    a

    tree. The single

    trunk

    is

    the

    one

    possible past

    that

    has any

    present chance of being actual. The

    many branches are the many

    possible

    futures that have some present chance

    of

    being actual.

    (1981,

    277)

    We can think

    of

    Lewis's tree structures

    and

    the

    fixed-open

    distinction

    in

    terms

    of nomic

    possibility given the present state

    of

    the

    world: the pos-

    sibilities that have a chance

    at

    a moment

    are

    those

    that are

    nomically

    possible at that moment. Facts "about"now-say, that a coin lands heads-

    are fixed and

    so

    not

    chancy. (It

    is

    worth

    noting

    that the sort of

    chance

    just

    describedneed not be

    deeply metaphysical.

    For

    instance,

    Brian

    Skyrns

    takes

    chance

    to

    be just

    a

    special

    case of

    subjective probability,

    viz.

    the

    subjective

    probability

    conditioned

    upon

    a

    partition

    of

    the

    possible situa-

    tions. On

    his

    view,

    if

    a set

    of

    propositions,

    {Pj},

    form

    the

    appropriate

    partition

    of the set

    of

    all

    possible

    situations,

    and if

    PJ

    is the

    true

    member

    of

    these, then chance

    A

    is

    equal

    to

    the

    subjective probability

    of

    A

    given

    Pi.

    As

    long

    as

    the

    partition

    members

    fully

    reflect the

    state

    of

    the world

    up until a time, intuitively the present time, then chance will apply non-

    trivially only

    to

    statements about

    the future.

    In

    this

    way Skyrms

    makes

    sense

    of chance as described above

    but from

    a

    subjectivist's viewpoint.

    (See Skyrms

    1984, 107-109;

    and

    Skyrms

    1988.)

    Given our

    understanding

    of

    probability

    in

    (c)

    as

    chance,

    we

    can

    better

    understand the

    argument against (c) given

    at the outset

    of this

    section.

    Again, we are

    to

    consider the case

    for

    which

    QM assigns

    a

    probability

    to

    result

    r

    (0

    B"

    because, by hypothesis,

    the person asserting (3)

    holds that

    A

    >

    B is false and so not

    probable.

    Furthermore, (3)

    cannot

    be construed

    as A

    >

    (B

    is

    probable)

    that

    if

    Williams had faced

    today's

    big league pitching,

    then it

    would have been

    probable

    that he was

    a

    300+

    hitter.

    This

    latter

    construal

    should seem

    odd,

    unlike

    the

    original,

    because

    had

    Williams faced

    the

    pitching

    in

    question,

    his

    batting average

    would

    not have been

    uncertain

    or

    chancy;

    it would have

    been

    more than

    probable

    whetheror not he was a 300+ hitter, that is, his average would have been

    known,

    a settled

    fact.)

    Finally, we can

    return to the

    question:

    How

    is

    (*),

    the statement

    that

    R has

    probabilityp given

    counterfactually

    that

    M,

    to be

    interpreted?

    I

    will

    try

    to show that

    if we think of

    (*)

    as a

    quantitative

    version of

    might

    or

    would-most-likely conditionals,

    then the problems

    for

    (a)-(c) disap-

    pear. My suggestion

    is

    that

    (*)

    be understood

    as

    M

    >P

    R were

    >P,

    read

    "would-with-probability-p",

    means

    roughly

    "would

    in a

    set

    of

    measure

    p". That is, where

    P is a probability measure on s(A),

    A

    >P

    B is true just

    in case if b is the set of elements of s(A) at which B is true, then P(b)

    -p. So,

    for

    example,

    where

    QM assigns probability

    that

    a

    particle

    be

    found

    in

    region V,

    if measured for

    position,

    the

    proposal implies

    the

    fol-

    lowing:

    a measure

    p

    of

    s(M),

    the set

    of most similar

    position measurement

    worlds,

    have resultant

    value

    in

    region

    V.

    And,

    in

    general,

    the

    hypothetical

    probabilitiesof QM are

    to

    be analyzed

    in

    terms

    of

    probability

    measures

    on the

    set of

    m-measurement

    worlds that

    might

    be.

    Let me

    try

    to

    clarify

    the

    suggestion

    of the

    last

    paragraph.

    One wants

    to

    give

    truth

    conditions

    for

    A

    >P

    B.

    I will do

    this

    in

    terms of

    probability

    measures

    PS(A),

    on each of the sets

    s(A)

    for arbitrary entences A. That is,

    I assume

    the

    additional

    structure

    of

    this set

    of

    probability

    measures,

    one

    for

    each sentence.

    Now,

    let

    [B]

    =

    {w

    E

    s(A):

    B is

    true

    at w}.

    Then,

    the

    truth

    condition

    goes

    as follows:

    A

    >P

    B is

    true

    just

    in

    case

    [B]

    is

    mea-

    surable

    and

    PS(A)[B]

    =

    p.

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    21/26

    LOGICAL FORM OF PROBABILITY ASSIGNMENT 55

    The suggestion, then, for explicating the quantum mechanical proba-

    bility attributions (*)

    is

    the following:

    (d)

    M >P

    R.

    Given

    that the

    probability measures, upon

    which

    (d)

    is

    based,

    come

    from

    QM,

    it

    would appear

    that

    (d)

    involves

    objective probabilities, though we

    will discuss an alternative

    in

    the next section. Furthermore, given the

    discussion of the last section, it would seem that

    a

    frequency interpre-

    tation

    is not

    appropriate

    here.

    So, finally,

    it would

    seem

    reasonable to

    take the probabilities

    in

    (d)

    as

    chances; after all, they are theoretical en-

    tities that

    have

    meaning

    for the

    single

    case.

    I

    take

    it, then,

    that

    the prob-

    abilities of (d) are chances, though this is not a necessary concomitant of

    the

    proposal. (If

    it were

    necessary,

    I

    would be

    begging

    an

    important

    in-

    terpretive

    issue that

    I

    promised

    not

    to do in an

    explication

    of

    (*).

    For-

    tunately, (d) places

    no

    restrictions

    on how its

    probability

    measures

    should

    be interpreted.)

    Finally, I should say how my suggestion,

    M >P

    R, for the probability

    attributions

    of

    QM,

    fares

    against

    the sort

    of

    argument brought against (b)

    and

    (c). First,

    the

    argument against (b)

    started with the

    quantum me-

    chanical

    probability assignments,

    and concluded

    that

    a

    certain disjunc-

    tion, (M >

    R) V (M

    > -R), has probability one, an unfortunateconse-

    quence

    in

    general.

    Let me

    run

    through

    that sort

    of

    argument with (d)

    assumed

    as the form

    of

    quantum

    mechanical

    probability

    attribution.

    want

    to show

    that

    no

    undesirable

    consequence

    is

    forthcoming.

    As

    before,

    we

    have hat

    PQM(R)

    =

    p,

    PQM(-R)

    =

    1

    -

    p,

    and

    PQM(R&-R)

    = 0.

    So,

    the

    suggestion (d) implies

    M >P

    R,

    M >('-PI

    -R,

    and M

    >0

    (R&-R). By

    the truth efinition

    have

    ust given,

    this

    amounts

    o:

    PS(M)([R])

    =

    P, PS(M)([-R])

    =

    1

    -

    p,

    and

    PS(M)([R&-R])

    =

    PS(M)([R]

    n

    [-R])

    =

    0.

    It

    follows

    from these

    facts

    that

    PS(M)([R]

    U

    [-R])

    =

    PS(M)([R

    V -R])

    =

    1

    +

    (1

    -

    p)

    =

    1. So, by the truth definition for >P just given,

    M

    >1

    (R V -R).

    This

    result

    is not

    a

    problem;

    indeed it

    is

    required by the se-

    mantics I

    have

    given, following

    without the

    assumption

    of

    (d).

    Further-

    more, PQM(RV -R)

    =

    1.

    So, the argument

    that led

    (b)

    to

    a

    problematic

    conclusion

    leads (d) only

    to

    a

    triviality.

    Also

    notice

    that

    the

    problem

    described for

    (c)

    cannot

    arise

    with

    (d):

    Modus Ponens

    is not allowed for the

    would-with-probability-p

    conditional

    >P,

    so one

    cannot,

    from

    M

    >P

    R

    and

    M,

    derive claims

    about

    post-measurement

    chances.

    (Modus

    Ponens

    does

    not hold

    because

    the

    probabilitiesdescribed by

    M

    >P

    R

    relate to a measure over a set of pos-

    sible

    worlds; they

    are not

    probabilities

    that

    something

    is

    true

    in

    the

    actual

    world.) Finally,

    notice

    that

    because

    (d)

    involves a

    counterfactual

    condi-

    tional

    which

    is

    to

    be

    analyzed

    in

    accordance with

    (0),

    one

    might worry

    that

    (d) prejudges interpretive

    issues or

    saddles

    QM

    with

    unfortunate

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    22/26

    56

    JOHN

    F.

    HALPIN

    metaphysical

    baggage

    and controversy

    over

    possible

    worlds

    semantics.

    If

    (d)

    did

    so, it would,

    like (b)

    and

    (c),

    fail one

    of

    section two's tests

    for

    a good explication of quantummechanical probability assignment. For-

    tunately, the

    counterfactual analysis

    of

    (0)

    is

    generic,

    leaving details

    of

    a theory of

    counterfactuals

    open.

    Finally,

    I should repeat

    that proposal (d)

    is

    not

    inextricably opposed

    to (a),

    so

    long

    as the

    conditional

    probabilities

    of

    (a)

    can

    be taken

    as fun-

    damental. Indeed one might

    identify

    conditional probabilitywith the

    sort

    of conditional chance

    described

    in

    (d).

    However,

    before this

    identification

    can

    reasonably

    be

    discussed,

    one

    would

    need to

    go

    more deeply into the

    analysis

    of counterfactuals.

    For

    example,

    conditional

    probabilities

    are

    usually supposed to obey the product rule: Pr(A&B/C)

    =

    Pr(A/C)

    Pr(B/A&C).

    To

    see

    if the conditional

    chances also

    obey

    the

    product

    rule-

    that

    is,

    to evaluate

    the claim

    that

    if

    C

    >P

    (A&B),

    C

    >q

    A,

    and

    (A&C)

    >r

    B,

    then

    p

    =

    q *

    r-we would

    need

    to

    sketch

    the

    details

    of

    how

    possible

    world

    sets

    s(C), s(A&C),

    and their

    measures are related.

    This,

    however,

    goes beyond

    the scope

    of this paper;

    we

    have set aside

    such

    interpretive

    issues as the details

    of

    the

    analysis

    of

    counterfactuals. So,

    I

    leave a

    study

    of the

    relationship

    of

    (d)

    to

    conditional

    probability

    for

    another

    occasion.

    7. Epistemic Probability. In section 5, we rejected the possibility that

    quantum

    mechanical assignments

    of

    probability

    are

    personal

    probabilities

    because QM

    is

    clearly not about personal

    belief states.

    But

    we set

    aside

    the alternative "Epistemic

    Probabilities"

    proposal

    that the

    quantum

    me-

    chanical

    assignments

    are

    really prescriptions

    or instructions

    or

    these

    states.

    On this

    view, QM

    indicates

    appropriate

    degrees

    of

    plausibility,

    that

    is,

    quantum

    mechanical

    assignments

    of

    probability

    give

    the

    degrees

    of

    belief

    one

    should have toward

    statements

    about measurement

    results. One

    way

    to

    read these

    prescriptions

    would be

    a variant to

    (c):

    (c')

    If

    an

    m-measurement

    s

    performed,

    then

    (one

    should)

    assign

    de-

    gree

    of belief

    p

    to

    R.

    This new

    proposal, (c'),

    is a conditional

    prescription.

    (Note

    that

    (c')

    like

    most prescriptions

    can

    be

    defeated

    by

    other

    considerations;

    for

    example,

    one should

    not

    assign personal probability

    p ($ 1)

    to

    R

    in

    the case

    in

    which one looks

    and

    sees

    that

    R

    is

    true.

    So, (c')

    is

    perhapsobjectionably

    vague.

    Also note

    that

    one

    might

    want to

    rephrase (c') by dropping

    the

    words "one

    should";

    this would

    make it

    clear that

    (c')

    is

    an

    instruction,

    that is, a command, ratherthan a normative claim. I prefer to leave this

    vague.)

    Now,

    there

    is

    little

    doubt that

    we

    do,

    from

    QM, get

    information

    per-

    tinent to our

    personal probabilities.

    The

    proponent

    of

    chance will

    typi-

    cally

    hold that

    chances

    mandate

    personal

    probabilities

    (in ways

    to be

    de-

    This content downloaded from 141.210.2.78 on Tue, 12 Nov 2013 15:57:24 PMAll use subject to JSTOR Terms and Conditions

    http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp
  • 8/11/2019 What is the Logical Form of Probability Attribution in QM

    23/26

    LOGICAL FORM OF PROBABILITY ASSIGNMENT 57

    scribed later

    in this

    section). But

    the

    proponent

    of

    (c') suggests rather

    that chance doesn't

    play

    a

    role

    in

    the

    quantum

    mechanical

    probability

    attributions *), but that these involve only instructions (Leeds 1984). As

    we have seen in section 5, (d) is perhaps most plausible if we understand

    it in terms of chance. Leeds's argument, then,

    has

    the potential to cut

    against (d). Indeed, let us suppose for the rest

    of

    this section that the

    probabilities

    of

    (d)

    are chances.

    I will

    attempt

    first to

    suggest

    a

    pre-

    sumption in favor of chance

    and of

    (d) over (c')

    in

    part by showing that

    Leeds's arguments against

    chance

    in

    QM,

    and so

    against (d), are not

    forceful. There is surely something

    of

    interest

    to

    (c'),

    a

    connection be-

    tween

    QM

    and

    personal probability,

    even

    if

    it

    is

    not

    an

    appropriate

    anal-

    ysis or explication of (*). Secondly, I will discuss the relation between

    personal probability

    and the counterfactual

    probabilities suggested

    in

    the

    last section.

    Leeds gives an argument

    for

    the "incompatibility"

    of

    QM and realism

    via "a skeptical attack

    on

    the notion

    of chance

    in

    QM" (1984, 568). He

    argues that when we

    use

    QM

    to make statistical

    predictions

    or

    explana-

    tions, we

    need

    only

    take into

    account

    the

    wave function from

    which the

    appropriate probabilities

    can be

    deduced;

    there

    is

    no

    need to

    mention

    "chance".

    I

    would

    argue

    that

    though

    this is

    true,

    it

    by

    no means

    excludes

    chances from QM; rather