2ments of
both
subsystems in particular states.The problem arises because there is no projection operator that gives the probability of outcome “1” whenthe subsystem to be measured (“
A
” or “
B
”) is undetermined. That is an illposed question in the quantum theory. As Page emphasizes, this is the kind of question oneneeds to address in order to make predictions in the multiverse, where our lack of knowledge about which pocketuniverse we occupy corresponds to “
A
” vs. “
B
” not being determined in the toy model. Such illposed quantumquestions exist in laboratory situations but there are sufﬁciently many wellposed problems that we tend not tobe concerned. Also, in the laboratory it is often possibleto modify the setup so there is a measurable “label” thatdoes identify “
A
” vs. “
B
”, thus resolving the problem.But such a resolution is believed not to be possible inmany cosmological cases. (There is also an interestingrelationship between this issue and the (anti)symmetricwavefunctions required for identical particles [4].)A natural response to this issue is to appeal to classicalideas about probabilities to “ﬁll in the gap”. In particular, if one could assign classical probabilities
p
A
and
p
B
for the measurement to be made on the respective subsystems, then one could answer the question posed above(the probability of the outcome “1” with the subsystemto be measured undetermined) by giving:
p
1
=
p
A
ψ

ˆ
P
A
1

ψ
+
p
B
ψ

ˆ
P
B
1

ψ
.
(1)Note that the values of
p
A
and
p
B
are
not
determinedfrom

ψ
, and instead provide additional informationspecially introduced so one can write Eqn.1. Although
p
1
can be written as the expectation value of ˆ
P
1
=
p
A
ˆ
P
A
1
+
p
B
ˆ
P
B
1
, the operatorˆ
P
1
is not a projection operator (ˆ
P
1
ˆ
P
1
=ˆ
P
1
), thus conﬁrming that
p
1
isnot a probability with a fully quantum origin.Authors who apply expressions like Eqn.1to cosmology[5] do not claim this gives a quantum probability. Instead they appeal to classical notions of probability alongthe lines we have discussed at the start of this paper.Surely one successfully introduces classical probabilitiessuch as
p
A
and
p
B
all the time in everyday situations toquantify our ignorance, so why should the same approachnot be used in the cosmological case?Our view is that the two cases are completely diﬀerent.We believe that in every situation where we use “classical” probabilities successfully these probabilities could inprinciple be derived from a wavefunction describing thefull physical situation. In this context classical probabilities are just ways to estimate quantum probabilities whencalculating them directly is inconvenient. Our extensiveexperience using classical probabilities in this way (really quantifying our
quantum
ignorance) cannot be usedto justify the use of classical probabilities in situationswhere quantum probabilities have been clearly shown tobe illdeﬁned and uncomputable. Translating the formalframework from one situation to the other is not an extrapolation but the creation of a brand new conceptualframework that needs to be justiﬁed on its own.We are only challengingthe ad hoc introduction of classical probabilities such as
p
A
and
p
B
. We are not criticizing the use of standard ideas from probability theory tomanipulate and interpret probabilities that have a physical origin[6] . Of course we never know the wavefunctioncompletely (and thus often write states as density matrices). Our claim is that probabilities are only proven andreliable tools if they have clear values determined fromthe quantum state, despite our uncertainties about it.We next use some simple calculations to argue thatit is realistic to expect all probabilities we normally useto have a quantum origin. Consider a gas of idealizedbilliards with radius
r
, mean free path
l
,average speed ¯
v
and mass
m
. If two of these billiards approach each otherwith impact parameter
b
, the uncertainties in the transverse momentum (
δp
⊥
) and position (
δx
⊥
) contribute toan uncertainty in the impact parameter given by:∆
b
=
δx
⊥
+
δp
⊥
m
∆
t
=
√
2
a
+¯
h
2
alm
¯
v
(2)where the second equality is achieved using ∆
t
=
l/
¯
v
andassuming a minimum uncertainty wavepacket of width
a
in each of the two transverse directions. The value of ∆
b
is minimized by
a
=
¯
hl/
(2
m
¯
v
)
≡
l