Lecture Notes #22 (Jacob Barandes – Spring 2023)

Interference
For our next topic, we’re going to study a phenomenon that might appear to be distinctly different from
what one encounters in classical physics. This phenomenon, known as quantum-theoretic interference, turns
out to resemble the interference of classical waves in some ways, but also differs from the interference of
classical waves in other ways.
Quantum-theoretic interference is easiest to introduce through a famous thought experiment, known as
the double-slit experiment. The idea is to send particles—one at a time—toward a wall with two small slits
or holes in it, where the slits are assumed to be close together, and then record where the particles eventually
land on a detection screen on the other side of the wall. Broadly speaking, we can let A be a random variable

T
that denotes each particle’s initial conditions, we can let B denote which slit the particle goes through, and
we can let C denote where the particle ends up on the detection screen.
If one carries out this experiment using small stones, then one finds a distribution of landing sites C on
the detection screen that’s concentrated at the center of the detection screen, at a point aligned with the
midpoint between the two slits, and is less concentrated the farther away from the center one looks. One can
provide an intuitive explanation for this distribution of landing sites on the detection screen. A few stones
make it through the upper slit in the wall, and lead to a concentration of landing sites on the detection
AF
screen lined up with the first slit. A few stones instead go through the lower slit in the wall, leading to a
concentration of landing sites aligned with the second slit. Because the two slits are close together, the two
concentrations on the detection screen merge into one larger distribution centered between the two slits.
If one instead carries out this experiment repeatedly using electrons, sending just one electron each time,
then one finds that the distribution of landing sites C features a strange assortment of peaks and valleys.
Near the center of the detection screen, there is a high concentration of landing sites, but slightly off-center,
there are no landing sites. Slightly farther out, there is another concentration of landing sites, but then, even
farther out, none. This pattern of peaks and valleys continues, with the peaks getting smaller and smaller
the farther one looks away from the center of the detection screen.
This pattern of landing sites C, recorded over many trials, resembles the crests and troughs that would
arise from waves approaching the two slits, spreading or diffracting from each slit, and then exhibiting
constructive and destructive interference on the other side. This resemblance between the pattern of the
electron’s landing sites over many trials and the shape of interfering waves has suggested to some people that
DR
an electron might itself have a wavelike nature or even be a wave in some circumstances—a notion known
as wave-particle duality.
One crucial difference, of course, is that for actual waves, the pattern of crests and troughs is explicitly
manifest all at once in the observable shape of the wave as a whole. When you look at, say, waves on a lake,
you can see the crests and troughs quite directly. By contrast, for an electron going through a double-slit
experiment, the interference pattern shows up only point by point, over the course of sending electrons one
at a time over many trials. We don’t directly see a wave at all in the electron case—we only see indirect
evidence of a wave through the distribution of landing sites over many trials, like the formation of a painting
from a pointillist technique.
If there is, in fact, a wave involved in this story, then what is the nature of this wave? We’ve already seen
that a quantum system can have an associated wave function (17.12). Indeed, the wavelike behavior of an
electron in the double-slit experiment is usually associated with the electron’s wave function. The trouble
is that a wave function doesn’t live in physical three-dimensional space, but in the system’s configuration
space, meaning an abstract space whose points formally denote the system’s distinct possible configurations
or snapshots. Recall that for a single particle in physical three-dimensional space, the relevant configuration
space is itself three-dimensional, so it’s easy to confuse the two kinds of spaces. But physical space and
configuration space are conceptually different things, and this difference between them becomes manifest if
we were to imagine sending two electrons at a time, instead of just one. Then because it takes six numbers
to specify the locations of two particles, the configuration space would now be six-dimensional, rather
© Jacob Barandes 2025. All rights reserved. Email: jacob_barandes@harvard.edu
than three-dimensional, meaning that the corresponding wave function would reside in a six-dimensional
configuration space. (Technically, one needs to modify the configuration space slightly to account for the
fact that electrons are indistinguishable particles.) Although it’s possible to try to take configuration spaces
of arbitrary dimensions seriously as the seat of physical reality, as Schrödinger himself argued in his early
papers on wave mechanics, and as is advocated in an interpretative framework known today as wave-function
realism, that metaphysical posture might be a hard sell for many people.
There’s a slight modification of the double-slit experiment that makes its behavior even stranger. Suppose
that we add an indicator qubit just behind the two slits, where this indicator qubit changes its configuration
based on which slit the electron goes through. This qubit has just two configurations, which we’ll denote
by ↑ and ↓, and let’s assume that it always starts off in its ↑ configuration at the beginning of each trial.
Suppose that the indicator qubit ends up in its ↑ configuration if the electron goes through the upper slit,
and ends up in its ↓ configuration if the electron goes through the lower slit.
If this indicator qubit is included in each trial of the experiment, then the distribution of the electron’s
landing sites C over many trials will look more like for the case of small stones, without all those striking

T
peaks and valleys. However, if at some point during the flight of each electron from the wall to the detection
screen, the indicator qubit is programmed to erase its memory, so that it always returns to its ↑ configuration
regardless of its previous configuration during the electron’s transit to the detection screen, then the peaks
and valleys in the distribution of landing sites will reappear.
Treating the electron as a closed system, let’s see how these strange behaviors naturally emerge from
the electron’s unistochastic dynamics. To make our preliminary treatment easier to follow, let’s simplify
or coarse-grain the electron’s configuration space so that we can model it as a qubit with just two possible
AF
configurations: 1 and 2. We’ll take 1 to mean that the coarse-grained electron is above the middle line of the
apparatus, corresponding both to the upper slit and also to the upper portion of the detection screen, and
we’ll take 2 to mean that the coarse-grained electron is below the middle line of the apparatus, corresponding
both to the lower slit and also to the lower portion of the detection screen.
The coarse-grained electron has a 2 × 2 unistochastic matrix
 
Γ11 (t) Γ12 (t)
Γ(t) = .
Γ21 (t) Γ22 (t)

The entries of this stochastic matrix are expressible in terms of corresponding entries of a 2 × 2 unitary
matrix U(t),  
|U11 (t)|2 |U12 (t)|2
Γ(t) = ,
|U21 (t)|2 |U22 (t)|2
DR
where  
U11 (t) U12 (t)
U(t) = .
U21 (t) U22 (t)
Because U(t) is unitary, we know from (20.3) that it has an inverse matrix given by its adjoint:

U† (t) = U−1 (t).

Let’s consider the product of U(t) itself with the adjoint of U(t′ ) evaluated at another time t′ :

U(t)U† (t′ ).

The resulting matrix is again unitary, and is a function of both t and t′ . As in (20.10), we call it U(t ← t′ ):

U(t ← t′ ) ≡ U(t)U† (t′ ).

As we saw in deriving (20.11), we see that we can rearrange this definition to obtain the following composition
law :
U(t) = U(t ← t′ ) U(t′ ) .
| {z } | {z }
0 to t 0 to t′
In detail, from the rules of matrix multiplication, we have
2
X
Uij (t) = Uik (t ← t′ )Ukj (t′ ),
k=1

so this composition law consists of the following four equations:

U11 (t) = U11 (t ← t′ )U11 (t′ ) + U12 (t ← t′ )U21 (t′ ),

U12 (t) = U11 (t ← t′ )U12 (t′ ) + U12 (t ← t′ )U22 (t′ ),
U21 (t) = U21 (t ← t′ )U11 (t′ ) + U22 (t ← t′ )U21 (t′ ),
U22 (t) = U21 (t ← t′ )U12 (t′ ) + U22 (t ← t′ )U22 (t′ ).

The matrix U(t ← t′ ) is unitary, so squaring its entries |Uij (t ← t′ )|2 defines a 2 × 2 unistochastic matrix
Γ(t ← t′ ):

T
   
Γ11 (t ← t′ ) Γ12 (t ← t′ ) |U11 (t ← t′ )|2 |U12 (t ← t′ )|2
Γ(t ← t′ ) = ≡ . (22.1)
Γ21 (t ← t′ ) Γ22 (t ← t′ ) |U21 (t ← t′ )|2 |U22 (t ← t′ )|2

However, this new unistochastic matrix Γ(t ← t′ ) does not generally lead to a corresponding composition
law for the system’s original unistochastic matrix Γ(t). That is, generically speaking, we’ll find that

Γ(t) ̸= Γ(t ← t′ )Γ(t′ ), (22.2)

AF
so despite the notation for Γ(t ← t′ ), we cannot say (at this point, at least) that it has a clear physical
interpretation of describing stochastic dynamics from t′ to t. Fundamentally, this discrepancy occurs because
taking absolute-value-squares of individual entries of matrices is not compatible with matrix multiplication.
We’ll now show that this discrepancy between the true, indivisible dynamics expressed by Γ(t) on the left-
hand side of (22.2) and the incorrect, divisible dynamics expressed by Γ(t ← t′ )Γ(t′ ) on the right-hand side
is precisely why interference occurs.
Let’s compute the discrepancy between the left-hand side and right-hand side of (22.2) explicitly for our
coarse-grained, two-configuration electron, treating t = 0 as the launch time, t′ as the time the electron
reaches the wall with the slits in it, and t as the time when the electron lands on the detection screen.
For definiteness, let’s focus on Γ11 (t) ≡ p(1, t|1, 0), the conditional probability for the electron to be above
the middle line of the double-slit set-up at the detection time t, given that the electron was already above
the middle line at the initial time t = 0, again within the context of our coarse-graining of the electron’s
DR
configuration space down to just two possible configurations.
The corresponding entry of the right-hand side of (22.2) is, from the standard rules of matrix multipli-
cation, given by
[Γ(t ← t′ )Γ(t′ )]11 = Γ11 (t ← t′ )Γ11 (t′ ) + Γ12 (t ← t′ )Γ21 (t′ ).
Inserting the basic relationships Γik (t ← t′ ) = |Uik (t ← t′ )|2 and Γjk (t′ ) = |Ujk (t′ )|2 , we have

[Γ(t ← t′ )Γ(t′ )]11 = |U11 (t ← t′ )|2 |U11 (t′ )|2 + |U12 (t ← t′ )|2 |U21 (t′ )|2 . (22.3)

In a sense, this is the predicted conditional probability based on the assumption that we can divide up the
electron’s path into two simple stochastic processes: a unistochastic process from 0 to t′ and then, separately,
another unistochastic process from t′ to t.
Computing the left-hand side of (22.2), meaning the electron’s actual conditional probability, we find

Γ11 (t) = |U11 (t)|2 .

Invoking the composition law U(t) = U(t ← t′ )U(t′ ) from (20.11), we have

Γ11 (t) = |U11 (t ← t′ )U11 (t′ ) + U12 (t ← t′ )U21 (t′ )|2 . (22.4)

Comparing (22.3) and (22.4), we see that we have, schematically speaking,

|u|2 + |w|2 versus |u + w|2 .

These expressions are indeed not in agreement. Specifically,
|u + w|2 = (u + w)(u + w)
= (u + w)(u + w)
= uu + uw + wu + ww
= |u|2 + uw + wu + |w|2 .
The discrepancy therefore lies in the two cross terms uw and wu, which, despite being complex-valued
individually, add up to a real-valued quantity, because for any complex number z ≡ a + ib, one has z + z =
(a + ib) + (a − ib) = 2a.
Carrying out the multiplication in (22.4) explicitly, we have
 

Γ11 (t) = U11 (t ← t′ )U11 (t′ ) + U12 (t ← t′ )U21 (t′ )

| {z } | {z }

T
u w
 

× U11 (t ← t′ )U11 (t′ ) + U12 (t ← t′ )U21 (t′ )

| {z } | {z }
u w
= |U11 (t ← t′ )|2 |U11 (t′ )|2
| {z }
|u|2
AF + U11 (t ← t′ )U11 (t′ ) U12 (t ← t′ )U21 (t′ )
|

+ U12 (t ←
|
{z

{z
w
u

t′ )U

′
}|

21

2
+ |U11 (t ← t )| |U11 (t )| .
| {z }
{z

(t′ ) U
}|
′
}

11 (t

2
{z
w

← t′ )U11 (t′ )
u
}

|w|2

The difference between this expression for Γ11 (t) and (22.3) for [Γ(t ← t′ )Γ(t′ )]11 consists precisely of the
two cross terms:
)
Γ11 (t) − [Γ(t ← t′ )Γ(t′ )]11 = U11 (t ← t′ )U11 (t′ )U12 (t ← t′ )U21 (t′ )
(22.5)
+ U12 (t ← t′ )U21 (t′ )U11 (t ← t′ )U11 (t′ ).
DR
Using a state vector Ψ(t′ ) as in (17.16) to denote the first column of the time-evolution operator U(t′ ), we
can write (22.3) somewhat more succinctly as
[Γ(t ← t′ )Γ(t′ )]11 = |U11 (t ← t′ )|2 |Ψ1 (t′ )|2 + |U12 (t ← t′ )|2 |Ψ2 (t′ )|2 , (22.6)
and we can write the cross terms (22.5) as
Γ11 (t) − [Γ(t ← t′ )Γ(t′ )]11 = U11 (t ← t′ )Ψ1 (t′ )U12 (t ← t′ )Ψ2 (t′ )
+ U12 (t ← t′ )Ψ2 (t′ )U11 (t ← t′ )Ψ1 (t′ ),
where the right-hand side is the sum of a complex-valued quantity with its complex conjugate, and is therefore
real-valued. Recalling the simplest version of the Born rule, (17.19), we have that pk (t′ ) = |Ψk (t′ )|2 is the
standalone probability for the electron to be in its kth configuration at the time t′ , so we can recast (22.6)
as
[Γ(t ← t′ )Γ(t′ )]11 = Γ11 (t ← t′ )p1 (t′ ) + Γ12 (t ← t′ )p2 (t′ ),
meaning that we can write Γ11 (t) as a whole as

Γ11 (t) = Γ11 (t ← t′ )p1 (t′ ) + Γ12 (t ← t′ )p2 (t′ )

+ U11 (t ← t′ )Ψ1 (t′ )U12 (t ← t′ )Ψ2 (t′ ) (22.7)
′ ′
+ U12 (t ← t′ )Ψ 2 (t′ )U 11 (t ← t )Ψ1 (t ).
The first line of this equation looks like a simple marginalization rule at t′ , but the other two lines make
clear that this marginalization rule at t′ is not the whole story, due to the indivisibility of the unistochastic
dynamics.
For example, if
1 1
Ψ1 (t′ ) ≡ U11 (t′ ) = √ eiα , Ψ2 (t′ ) ≡ U21 (t′ ) = √ eiβ ,
2 2
so that the electron has equal probability of going through either slit,
1 1
p1 (t′ ) = |Ψ1 (t′ )|2 = , p2 (t′ ) = |Ψ2 (t′ )|2 = ,
2 2
and if
U11 (t ← t′ ) = aeiθ , U12 (t ← t′ ) = beiϕ ,
where a and b are real-valued and non-negative, so that

T
Γ11 (t ← t′ ) = |U11 (t ← t′ )|2 = a2 , Γ12 (t ← t′ ) = |U12 (t ← t′ )|2 = b2 ,
then
1 2 1 2
Γ11 (t ← t′ )p1 (t′ ) + Γ12 (t ← t′ )p2 (t′ ) = a + b ,
2 2
and the interference cross-terms are
U11 (t ← t′ )Ψ1 (t′ )U12 (t ← t′ )Ψ2 (t′ )
AF + U12 (t ← t′ )Ψ2 (t′ )U11 (t ← t′ )Ψ1 (t′ )
1
2
1
2
1

1 −iθ −iα iϕ iβ 1 −iϕ −iβ iθ iα

= ae e be e + be e ae e
2 2
2
1
= aeiθ √ eiα beiϕ √ eiβ + beiϕ √ eiβ aeiθ √ eiα
2

1 h i(θ+α−ϕ−β) −i(θ+α−ϕ−β)
i
= ab e +e .
2
Using Euler’s formula eiθ = cos θ + i sin θ from (12.35), as well as e−iθ = cos θ − i sin θ, this last expression
becomes
1h (ϕ(−(β)
(−
(
(−(ϕ(−(β) (i
ab cos(θ + α − ϕ − β) + ( (+
i sin(θ
( (α + cos(θ + α − ϕ − β) − ( (+
i sin(θ
( (α
2
DR
ab
= 2 cos(θ + α − ϕ − β)
2
= ab cos(θ + α − ϕ − β).
Putting everything together, (22.7) becomes
1 2 1 2
Γ11 (t) = a + b + cos(θ + α − ϕ − β),
2 2
where the sinusoidal term at the end represents the interference. In general, the phase angles θ and ϕ
will depend on the final time t, so different values of the final time will lead to slightly different amounts
of interference. In particular, for some values of θ + α − ϕ − β, the interference term will be negative,
partially canceling the preceding two terms and reducing the probability for the electron to end up in its 1
configuration at t.
There was nothing special about the specific entry Γ11 (t), or even the 2 × 2 case. In general, for the case
of a system with N possible configurations, one finds that
N
X
Γij (t) = Γik (t ← t′ )pk (t′ )
k=1
X (22.8)
+ Uik (t ← t′ )Ψk (t′ )Uil (t ← t′ )Ψl (t′ ),
k̸=l
where the first line looks like simple marginalization at t′ , as though the unistochastic dynamics were divisible
at t′ , and where the cross terms in the second line describe interference, and make clear that the actual
unistochastic dynamics is, in fact, indivisible.
In this more general case, suppose again that
1 1
Ψ1 (t′ ) = √ eiα , Ψ2 (t′ ) = √ eiβ ,
2 2
so that the system is equally likely to be in its configurations 1 or 2 at t′ (with zero probability of being in
any other configuration at t′ ), but now let’s let the final configuration i = 1, . . . , N at t be more general,
with
Ui1 (t ← t′ ) = ai eiθi , Ui2 (t ← t′ ) = bi eiϕi .
Then
1 2 1 2
Γi1 (t) = a + bi + ai bi cos(θi + α − ϕi − β).

T
2 i 2
So some final configurations i = 1, . . . , N will exhibit positive interference terms, meaning constructive
interference, whereas other final configurations will exhibit negative interference terms, meaning destructive
interference.

Decoherence and Division Events

AF
Let’s return to the double-slit experiment again, and formulate the interference in a slightly different way
that will eventually smooth the way toward introducing an indicator qubit. Once more, let’s suppose for
simplicity that we’ve coarse-grained the electron’s configuration space so that it consists of just two possible
configurations: 1, for when the electron is in the upper portion of the apparatus, and 2, for when the electron
is in the lower portion of the apparatus. Again for simplicity, suppose that the electron is launched from a
definite configuration j at t = 0 (say, j = 1), that it arrives at the wall with the two slits at t′ > 0, and that
it lands on the detection screen at t > t′ .
Let p1 (t′ ) be the standalone probability that the electron passes through the upper slit at t′ , and let
p2 (t′ ) be the standalone probability that the electron passes through the lower slit at t′ :

p1 (t′ ) ≡ p(electron passes through the upper slit at t′ ),

(22.9)
p2 (t′ ) ≡ p(electron passes through the lower slit at t′ ).
DR
Because the electron has some specific configuration j at t = 0, the electron has a corresponding 2 × 1 state
vector  
Ψ1 (t)
Ψ(t) = , (22.10)
Ψ2 (t)

whose two entries at t′ are

Ψ1 (t′ ),
(22.11)
Ψ2 (t′ ).

From the simplest form of the Born rule in (17.19), these entries of the electron’s state vector at t′ are related
to the corresponding standalone probabilities (22.9) according to

p1 (t′ ) = |Ψ1 (t′ )|2 ,

(22.12)
p2 (t′ ) = |Ψ2 (t′ )|2 .

At the detection time t > t′ , the electron interacts with the detection screen, so the electron is no longer
a closed system starting at the detection time. However, between the initial time t = 0 and the detection
time t > t′ , we can treat the electron as a closed system. The electron then has unistochastic dynamics
from the initial time t = 0 all the way to the detection time t > t′ , from which it follows that there exists
a unitary operator U(t ← t′ ), as defined in (20.10), that transforms the electron’s state vector Ψ(t′ ) at the
wall time t′ into its state vector Ψ(t) at the detection time t > t′ :

Ψ(t) = U(t ← t′ )Ψ(t′ ). (22.13)

That is,
    
Ψ1 (t) U11 (t ← t′ ) U12 (t ← t′ ) Ψ1 (t′ )
= . (22.14)
Ψ2 (t) U21 (t ← t′ ) U22 (t ← t′ ) Ψ2 (t′ )

Carrying out the matrix multiplication explicitly, we see that the two entries Ψ1 (t) and Ψ2 (t) of the electron’s
state vector at the detection time t > t′ are given by

Ψ1 (t) = U11 (t ← t′ )Ψ1 (t′ ) + U12 (t ← t′ )Ψ2 (t′ ),

Ψ2 (t) = U21 (t ← t′ )Ψ1 (t′ ) + U22 (t ← t′ )Ψ2 (t′ ).

T
Let’s give names to the four distinct terms appearing on the right-hand sides of these two equations:

Φ1 (t) ≡ U11 (t ← t′ )Ψ1 (t′ ),

Ω1 (t) ≡ U12 (t ← t′ )Ψ2 (t′ ),
(22.15)
AF Φ2 (t) ≡ U21 (t ← t′ )Ψ1 (t′ ),
Ω2 (t) ≡ U22 (t ← t′ )Ψ2 (t′ ).

Then the two equations for Ψ1 (t) and Ψ2 (t) above take the simpler form

Ψ1 (t) = Φ1 (t) + Ω1 (t),

(22.16)
Ψ2 (t) = Φ2 (t) + Ω2 (t).

Applying the Born rule (17.19) again, we see that the standalone probabilities for the electron at the
detection time t > t′ are given by

p1 (t) = |Ψ1 (t)|2 = |Φ1 (t) + Ω1 (t)|2 ,

(22.17)
p2 (t) = |Ψ2 (t)|2 = |Φ2 (t) + Ω2 (t)|2 .
DR
Notice here that in each expression on the right-hand side, we sum Φi (t) + Ωi (t) first, and then take the
absolute-value-square of the result. Using |z|2 = zz, these expressions yield
 
p1 (t) = Φ1 (t) + Ω1 (t) (Φ1 (t) + Ω1 (t))
= |Φ1 (t)|2 + |Ω1 (t)|2 + Φ1 (t)Ω1 (t) + Ω1 (t)Φ1 (t),
 
p2 (t) = Φ2 (t) + Ω2 (t) (Φ2 (t) + Ω2 (t))
= |Φ2 (t)|2 + |Ω2 (t)|2 + Φ2 (t)Ω2 (t) + Ω2 (t)Φ2 (t).

The cross-terms Φi (t)Ωi (t) + Ωi (t)Φi (t) appearing in these formulas, just like the cross terms appearing in
(22.7), represent interference.
We’re now ready to introduce the indicator qubit. Again, in each run or trial of the experiment, this
indicator qubit will sit between the two slits and record which slit the electron passes through. The indicator
qubit will have two possible configurations: ↑, which is the indicator qubit’s initial configuration and the
one it will maintain if the electron goes through the upper slit, and ↓, which will be the indicator qubit’s
new configuration if the electron goes through the lower slit. For now, we’ll assume for simplicity that the
indicator qubit doesn’t undergo any other time evolution.
The electron alone is no longer a closed system during the time interval from the initial time interval
t = 0 to the detection time t > t′ , but will be assumed to interact with the indicator qubit slightly before the
wall time t′ > 0. However, we can treat the composite system consisting of the electron together with the
indicator qubit as a closed system from the initial time t = 0 to the detection time t > t′ . This composite
system has a configuration space consisting of four configurations, which we’ll denote by the following ordered
pairs:
(1, ↑),
(2, ↑),
(22.18)
(1, ↓),
(2, ↓).
In each of these ordered pairs, the first label denotes the configuration of our coarse-grained electron, and
the second label denotes the configuration of the indicator qubit.
By assumption, at the wall time t′ , the composite system has zero probability of being in the configurations
(1, ↓) or (2, ↑). Let’s suppose that the joint probability p1,↑ (t′ ) for the composite system to be in the

T
configuration (1, ↑) at t′ has the same value p1 (t′ ) as in (22.9), and, similarly, that the joint probability
p2,↓ (t′ ) for the composite system to be in the configuration (2, ↓) at t′ has the same value p2 (t′ ) as in (22.9).
Altogether, we then have

p1,↑ (t′ ) = p1 (t′ ),

p2,↑ (t′ ) = 0,
(22.19)
p1,↓ (t′ ) = 0,
AF
The composite system has a 4 × 1 state vector
p2,↓ (t′ ) = p2 (t′ ).


Ψ1,↑ (t)
Ψ2,↑ (t)
Ψ(t) = 



Ψ1,↓ (t). (22.20)
Ψ2,↓ (t)

Its individual entries at the wall time t′ are given by

Ψ1,↑ (t′ ) = Ψ1 (t′ ),

Ψ2,↑ (t′ ) = 0,
DR
(22.21)
Ψ1,↓ (t′ ) = 0,
Ψ2,↓ (t′ ) = Ψ2 (t′ ),

where, from the Born rule (17.19), these entries are related to the composite system’s joint probabilities
(22.19) at the wall time t′ according to

p1,↑ (t′ ) = |Ψ1,↑ (t′ )|2 = p1 (t′ ),

p2,↑ (t′ ) = |Ψ2,↑ (t′ )|2 = 0,
(22.22)
p1,↓ (t′ ) = |Ψ1,↓ (t′ )|2 = 0,
p2,↓ (t′ ) = |Ψ2,↓ (t′ )|2 = p2 (t′ ).

By assumption, the electron and the indicator qubit interact with each other just before the wall time
t′ > 0, but do not interact with anything else from the wall time t′ > 0 to the detection time t > t′ .
During this later time interval, the two subsystems therefore have their own individual unitary time-evolution
operators. Thinking of our electron as the subject system S under study, and thinking of the indicator qubit
as the simplest possible model of an environment E, the unitary time-evolution operator USE (t ← t′ ) for the
composite system SE factorizes as

USE (t ← t′ ) = US (t ← t′ ) ⊗ UE (t ← t′ ), (22.23)
where US (t ← t′ ) describes the time evolution of the electron S from t′ to t, and where UE (t ← t′ ) describes
the time evolution of the indicator qubit from t′ to t. Here the tensor-product symbol ⊗ is merely notational
shorthand for
SE ′ S ′ E ′
U(i,e),(i ′ ,e′ ) (t ← t ) = Uii′ (t ← t )Uee′ (t ← t ), (22.24)

with i = 1, 2 labeling the two possible configurations of the electron S and e = ↑, ↓ labeling the two possible
configurations of the indicator qubit E. Hence, USE (t ← t′ ) is a 4 × 4 matrix that looks like
 S E S E S E S E 
U11 U↑↑ U12 U↑↑ U11 U↑↓ U12 U↑↓
U21
S E
U↑↑ S E
U22 U↑↑ S E
U21 U↑↓ S E 
U22 U↑↓ 
U (t ← t ) = 
SE ′
U11
S E S E S E S E , (22.25)
U↓↑ U12 U↓↑ U11 U↓↓ U12 U↓↓
S E S E S E S E
U21 U↓↑ U22 U↓↑ U21 U↓↓ U22 U↓↓

with the functional dependence (t ← t′ ) suppressed for notational compactness. Equivalently, we can write

T
this matrix as  S   S  
S S
U11 U12 E U11 U12 E
 U21S S U↑↑ S S U↑↓ 
USE (t ← t′ ) =   S U22  U21 U22  , (22.26)
 U11 U12 S
E
S S
U11 U12 E

S S U↓↑ S S U↓↓
U21 U22 U21 U22

with the 2 × 2 time-evolution operator US (t ← t′ ) for the electron multiplying each of the four entries of
AF
the 2 × 2 time-evolution operator UE (t ← t′ ) for the indicator qubit. (It’s remarkable just how complicated
things get even with just a pair of two-configuration systems.)
For simplicity, let’s suppose that the indicator qubit doesn’t evolve after its interaction with the electron.
Then from the wall time t′ to the detection time t, the indicator qubit’s time-evolution operator can be taken
to be the 2 × 2 identity matrix:
 E
  
U↑↑ (t ← t′ ) E
U↑↓ (t ← t′ ) 1 0
UE (t ← t′ ) = =1= . (22.27)
U↓↑ (t ← t′ )
E
U↓↓ (t ← t′ )
E
0 1

The state vector Ψ(t) for the composite system at the detection time is given by

Ψ(t) = USE (t ← t′ )Ψ(t′ ). (22.28)

DR
That is,
   S E S E S E S E  
Ψ1,↑ (t) U11 U↑↑ U12 U↑↑ U11 U↑↓ U12 U↑↓ Ψ1,↑ (t′ )
Ψ2,↑ (t) U21S E S E S E S E 
U↑↓ Ψ2,↑ (t′ )
   S U↑↑ U22 U↑↑ U21 U↑↓ U22 . (22.29)
Ψ1,↓ (t) = U11 E
U↓↑ S E
U12 U↓↑ S E
U11 U↓↓ S E 
U12 U↓↓ Ψ1,↓ (t′ )
Ψ2,↓ (t) S E
U21 U↓↑ S E
U22 U↓↑ S E
U21 U↓↓ S E
U22 U↓↓ Ψ2,↓ (t′ )

Recalling our expressions (22.21) for the entries of the composite system’s state vector at the wall time t′ ,

Ψ1,↑ (t′ ) = Ψ1 (t′ ),

Ψ2,↑ (t′ ) = 0,
Ψ1,↓ (t′ ) = 0,
Ψ2,↓ (t′ ) = Ψ2 (t′ ),

our matrix equation (22.29) simplifies to

   S E S E S E S E  
Ψ1,↑ (t) U11 U↑↑ U12 U↑↑ U11 U↑↓ U12 U↑↓ Ψ1 (t′ )
Ψ2,↑ (t) U21S E S E S E S E 
U↑↓  0 
   S U↑↑ U22 U↑↑ U21 U↑↓ U22 . (22.30)
Ψ1,↓ (t) = U11 E
U↓↑ S E
U12 U↓↑ S E
U11 U↓↓ S E 
U12 U↓↓ 0 
Ψ2,↓ (t) S E
U21 U↓↑ S E
U22 U↓↑ S E
U21 U↓↓ S E
U22 U↓↓ Ψ2 (t′ )
Carrying out the matrix multiplication explicitly, we see that the four entries Ψ1,↑ (t), Ψ2,↑ (t), Ψ1,↓ (t) and
Ψ2,↓ (t) of the composite system’s state vector at the detection time t > t′ are given by
S
Ψ1,↑ (t) = U11 (t ← t′ )U↑↑
E
(t ← t′ )Ψ1 (t′ ) + U12
S
(t ← t′ )U↑↓
E
(t ← t′ )Ψ2 (t′ ),
S
Ψ2,↑ (t) = U21 (t ← t′ )U↑↑
E
(t ← t′ )Ψ1 (t′ ) + U22
S
(t ← t′ )U↑↓
E
(t ← t′ )Ψ2 (t′ ),
S
Ψ1,↓ (t) = U11 (t ← t′ )U↓↑
E
(t ← t′ )Ψ1 (t′ ) + U12
S
(t ← t′ )U↓↓
E
(t ← t′ )Ψ2 (t′ ),
S
Ψ2,↓ (t) = U21 (t ← t′ )U↓↑
E
(t ← t′ )Ψ1 (t′ ) + U22
S
(t ← t′ )U↓↓
E
(t ← t′ )Ψ2 (t′ ),

where, in particular, the entry Ψ2,↑ (t) corresponds to the possibility that the electron eventually floats into the
lower portion of the apparatus after it passes through the upper slit, and the entry Ψ1,↓ (t) corresponds to the
possibility that the electron eventually floats into the upper portion of the apparatus after it passes through
the lower slit. Using (22.27) for the entries of the indicator qubit’s time-evolution operator UE (t ← t′ ), the
four expressions above become

T
S
Ψ1,↑ (t) = U11 (t ← t′ )(1)Ψ1 (t′ ) + U12
S
(t ← t′ )(0)Ψ2 (t′ ),
S
Ψ2,↑ (t) = U21 (t ← t′ )(1)Ψ1 (t′ ) + U22
S
(t ← t′ )(0)Ψ2 (t′ ),
S
Ψ1,↓ (t) = U11 (t ← t′ )(0)Ψ1 (t′ ) + U12
S
(t ← t′ )(1)Ψ2 (t′ ),
S
AF Ψ2,↓ (t) = U21 (t ← t′ )(0)Ψ1 (t′ ) + U22
S
(t ← t′ )(1)Ψ2 (t′ ),

which simplify to
S
Ψ1,↑ (t) = U11 (t ← t′ )Ψ1 (t′ ),
S
Ψ2,↑ (t) = U21 (t ← t′ )Ψ1 (t′ ),
S
Ψ1,↓ (t) = U12 (t ← t′ )Ψ2 (t′ ),
S
Ψ2,↓ (t) = U22 (t ← t′ )Ψ2 (t′ ).

Under the identification U(t ← t′ ) ≡ US (t ← t′ ), recall from (22.15) that we already gave names to the four
distinct terms appearing on the right-hand sides of these equations:

S
Φ1 (t) ≡ U11 (t ← t′ )Ψ1 (t′ ),
DR
S
Φ2 (t) ≡ U21 (t ← t′ )Ψ1 (t′ ),
S
(22.31)
Ω1 (t) ≡ U12 (t ← t′ )Ψ2 (t′ ),
S
Ω2 (t) ≡ U22 (t ← t′ )Ψ2 (t′ ).

Hence,
Ψ1,↑ (t) = Φ1 (t),
Ψ2,↑ (t) = Φ2 (t),
(22.32)
Ψ1,↓ (t) = Ω1 (t),
Ψ2,↓ (t) = Ω2 (t).

On comparing these four expressions with (22.16) from the case where the indicator qubit was absent,
)
Ψ1 (t) = Φ1 (t) + Ω1 (t),
← [no indicator qubit],
Ψ2 (t) = Φ2 (t) + Ω2 (t)

we see that with the indicator qubit present, the sum Φ1 (t) + Ω1 (t) and the sum Φ2 (t) + Ω2 (t) are now
separated out into their individual terms.
 Applying the Born rule (17.19) again, we see that the joint probabilities for the electron and the indicator
qubit at the detection time t > t′ are given by

p1,↑ (t) = |Ψ1,↑ (t)|2 = |Φ1 (t)|2 ,

p2,↑ (t) = |Ψ2,↑ (t)|2 = |Φ2 (t)|2 ,
(22.33)
p1,↓ (t) = |Ψ1,↓ (t)|2 = |Ω1 (t)|2 ,
p2,↓ (t) = |Ψ2,↓ (t)|2 = |Ω2 (t)|2 .

We can obtain the standalone probabilities p1 (t) and p2 (t) for the electron alone at the detection time t > t′
by marginalizing over the configurations e = ↑, ↓ of the indicator qubit:
X
p1 (t) = p1,e (t) ≡ p1,↑ (t) + p1,↓ (t) = |Φ1 (t)|2 + |Ω1 (t)|2 ,

T
e=↑,↓
X (22.34)
p2 (t) = p2,e (t) ≡ p2,↑ (t) + p2,↓ (t) = |Φ2 (t)|2 + |Ω2 (t)|2 .
e=↑,↓

Notice here that these results feature absolute-value-squares and then sums, meaning that there are no
cross-terms and therefore no interference. By comparison, the final standalone probabilities (22.17) for the
AF
electron without the indicator qubit feature sums and then absolute-value-squares,
)
p1 (t) = |Φ1 (t) + Ω1 (t)|2 ,
← [no indicator qubit],
p2 (t) = |Φ2 (t) + Ω2 (t)|2

which leads to interference.

Sums inside of absolute-value-squares are known as coherent sums, whereas sums outside of absolute-
value-squares are called incoherent sums. We therefore see that with the indicator qubit present, we’ve gone
from coherent sums to incoherent sums, an important quantum-theoretic phenomenon called decoherence.
Let’s now suppose that after the indicator qubit interacts with the electron, and without sharing its
reading with any other systems, the indicator qubit ‘forgets’ its reading, and returns back to its initial
configuration ↑. We can implement this time evolution by replacing the unitary time-evolution operator
DR
UE (t ← t′ ) defined in (22.27) with a non-unitary (indeed, non-invertible) time-evolution operator ΘE (t ← t′ )
defined by
 E   
E ′ Θ↑↑ (t ← t′ ) ΘE↑↓ (t ← t′ ) 1 1
Θ (t ← t ) = = . (22.35)
ΘE↓↑ (t ← t′ ) ΘE↓↓ (t ← t′ ) 0 0

Then our matrix equation (22.30) for the composite system’s state vector Ψ(t) at the detection time t > t′
becomes
   S E S E S E S E  
Ψ1,↑ (t) U11 Θ↑↑ U12 Θ↑↑ U11 Θ↑↓ U12 Θ↑↓ Ψ1 (t′ )
Ψ2,↑ (t) U21 S E S E S E S E 
Θ↑↓  0 
   S Θ↑↑ U22 Θ↑↑ U21 Θ↑↓ U22 ,
Ψ1,↓ (t) = U11 E S E S E S E 
Θ↓↑ U12 Θ↓↑ U11 Θ↓↓ U12 Θ↓↓ 0 
Ψ2,↓ (t) S E
U21 S E
Θ↓↑ U22 Θ↓↑ U21S E
Θ↓↓ U22 S E
Θ↓↓ Ψ2 (t′ )

which gives the four equations

S
Ψ1,↑ (t) = U11 (t ← t′ )ΘE↑↑ (t ← t′ )Ψ1 (t′ ) + U12
S
(t ← t′ )ΘE↑↓ (t ← t′ )Ψ2 (t′ ),
S
Ψ2,↑ (t) = U21 (t ← t′ )ΘE↑↑ (t ← t′ )Ψ1 (t′ ) + U22
S
(t ← t′ )ΘE↑↓ (t ← t′ )Ψ2 (t′ ),
S
Ψ1,↓ (t) = U11 (t ← t′ )ΘE↓↑ (t ← t′ )Ψ1 (t′ ) + U12
S
(t ← t′ )ΘE↓↓ (t ← t′ )Ψ2 (t′ ),
S
Ψ2,↓ (t) = U21 (t ← t′ )ΘE↓↑ (t ← t′ )Ψ1 (t′ ) + U22
S
(t ← t′ )ΘE↓↓ (t ← t′ )Ψ2 (t′ ).
Using the explicit form (22.35) of the indicator qubit’s new time-evolution operator ΘE (t), these equations
become
S
Ψ1,↑ (t) = U11 (t ← t′ )(1)Ψ1 (t′ ) + U12
S
(t ← t′ )(1)Ψ2 (t′ ),
S
Ψ2,↑ (t) = U21 (t ← t′ )(1)Ψ1 (t′ ) + U22
S
(t ← t′ )(1)Ψ2 (t′ ),
S
Ψ1,↓ (t) = U11 (t ← t′ )(0)Ψ1 (t′ ) + U12
S
(t ← t′ )(0)Ψ2 (t′ ),
S
Ψ2,↓ (t) = U21 (t ← t′ )(0)Ψ1 (t′ ) + U22
S
(t ← t′ )(0)Ψ2 (t′ ),

which simplify to
S
Ψ1,↑ (t) = U11 (t ← t′ )Ψ1 (t′ ) + U12
S
(t ← t′ )Ψ2 (t′ ),
S
Ψ2,↑ (t) = U21 (t ← t′ )Ψ1 (t′ ) + U22
S
(t ← t′ )Ψ2 (t′ ),
Ψ1,↓ (t) = 0,

T
Ψ2,↓ (t) = 0.

Recalling the definitions of Φ1 (t), Ω1 (t), Φ2 (t), and Ω2 (t) once again from (22.15), we end up with

Ψ1,↑ (t) = Φ1 (t) + Ω1 (t),

Ψ2,↑ (t) = Φ2 (t) + Ω2 (t),
Ψ1,↓ (t) = 0,
AF
detection time t > t′ are now
Ψ2,↓ (t) = 0.

Invoking the Born rule (17.19) again, the joint probabilities for the electron and the indicator qubit at the

p1,↑ (t) = |Ψ1,↑ (t)|2 = |Φ1 (t) + Ω1 (t)|2 ,

p2,↑ (t) = |Ψ2,↑ (t)|2 = |Φ2 (t) + Ω2 (t)|2 ,
(22.36)
p1,↓ (t) = |Ψ1,↓ (t)|2 = 0,
p2,↓ (t) = |Ψ2,↓ (t)|2 = 0,

so the standalone probabilities p1 (t) and p2 (t) for the electron alone at the detection time t > t′ are
X
p1 (t) = p1,e (t) ≡ p1,↑ (t) + p1,↓ (t) = |Φ1 (t) + Ω1 (t)|2 ,
DR
e=↑,↓
X (22.37)
p2 (t) = p2,e (t) ≡ p2,↑ (t) + p2,↓ (t) = |Φ2 (t) + Ω2 (t)|2 .
e=↑,↓

Notice that these results feature coherent sums again, just like before we introduced the indicator qubit, so
the electron’s final probability distribution at the detection time t > t′ now exhibits interference.
Keep in mind that the ‘forgetting’ by the indicator qubit here is a very strong form of forgetting, with no
traces or records left on any other systems in the larger environment surrounding our double-slit experiment.
Such forgetting would be highly impractical for any real-world, macroscopic systems. By contrast, when
a human or electronic computer ‘forgets’ something, there are essentially always residual traces scattered
among air molecules and radiated light. As such, when a realistic environment that’s more complicated than
our highly idealized qubit carries out a reading on a subject system, the loss of interference—that is, the
decoherence—is, to an extremely good approximation, irreversible.
Let’s generalize the previous calculation, with an eye toward developing the tools that we’ll need later
to analyze the measurement process. Once more, we’ll study the time interval from the wall time t′ > 0 to
the detection time t > t′ . By assumption, the electron, regarded as the subject system system S, and the
indicator qubit, denoted as a highly simplified ‘environment’ E, form a closed composite system SE from the
wall time t′ to the detection time t. Also by assumption, the wall time t′ is just after the indicator qubit has
interacted with the electron, so the electron S and the indicator qubit E are individually closed subsystems
during the time interval from t′ to t.
 Because the composite system SE is closed, there exists a unitary time-evolution operator USE (t ← t′ )
that transforms Hilbert-space ingredients from t′ to t. As a reminder, because each of the two subsystems
S and E are individually closed during that same time interval from t′ to t, they have their own respective
time-evolution operators US (t ← t′ ) and UE (t ← t′ ), and the composite system’s time-evolution operator
factorizes as a tensor product, as in (22.23):

USE (t ← t′ ) = US (t ← t′ ) ⊗ UE (t ← t′ ).

More explicitly, in terms of indices, we have (22.24),

SE ′ S ′ E ′
U(i,e),(i ′ ,e′ ) (t ← t ) = Uii′ (t ← t )Uee′ (t ← t ),

where i = 1, 2 labels the two possible configurations of the electron S and e = ↑, ↓ labels the two possible
configurations of the indicator qubit E. Rather than specify a particular time-evolution operator UE (t ← t′ )

T
for the indicator qubit, we’ll now allow it to be more general.
We’ll also assume that at the initial time t = 0, the electron S is in its jth configuration (say, j = 1) and
the indicator qubit E is in its e = ↑ configuration, so that the composite system SE has a 4 × 1 state vector
Ψ(t) at least from t = 0 to the detection time t > t′ . At the wall time t′ , just after the indicator qubit has
read the configuration of the electron, we recall from (22.21) that the composite system’s state vector Ψ(t′ )
has entries
Ψ1,↑ (t′ ) = Ψ1 (t′ ),

Letting
AF Ψ2,↑ (t′ ) = 0,
Ψ1,↓ (t′ ) = 0,
Ψ2,↓ (t′ ) = Ψ2 (t′ ).

e(1) ≡ ↑, e(2) ≡ ↓, (22.38)

we can write the entries of the composite system’s state vector Ψ(t) at the wall time t′ more succinctly as
(
′ Ψi′ (t′ ) for e′ = e(i′ ),
Ψi′ ,e′ (t ) =
0 for e′ ̸= e(i′ ),

or, even more succinctly, as

DR
Ψi′ ,e′ (t′ ) = Ψi′ (t′ )δe′ e(i′ ) , (22.39)

where we use primes on our labels i′ and e′ just to emphasize that they correspond to the wall time t′ , and
where we’ve introduced the Kronecker delta
(
1 for e′ = e(i′ ),
δe′ e(i′ ) ≡ (22.40)
0 for e′ ̸= e(i′ ).

As in (22.28), the composite system’s state vector Ψ(t) at the detection time t > t′ is given by acting with
the compoite system’s time-evolution operator USE (t ← t′ ) on the composite system’s state vector Ψ(t′ ) at
the wall time t′ :
Ψ(t) = USE (t ← t′ )Ψ(t′ ).
Working in terms of individual entries, and using the usual rules of matrix multiplication, this equation
becomes X X
SE ′ ′
Ψi,e (t) = U(i,e),(i ′ ,e′ ) (t ← t )Ψi′ ,e′ (t ).

i′ =1,2 e′ =↑,↓

Suppressing the argument (t ← t ) to keep the notation as simple as possible, we can write this equation as
′

XX
SE ′
Ψi,e (t) = U(i,e),(i ′ ,e′ ) Ψi′ ,e′ (t ).

i′ e′
Invoking the tensor-factorization (22.24) of the time-evolution operator USE (t ← t′ ), together with the com-
pact expression (22.39) for the composite system’s state vector Ψ(t′ ) at the wall time t′ , we have
X X  
Ψi,e (t) = UiiS′ Uee
E
′ Ψi′ (t′ )δe′ e(i′ )
i′ e′
" #
X X
= UiiS′ Ψi′ (t′ ) E
Uee ′ δe′ e(i′ ) .
i′ e′

A summation involving a Kronecker delta collapses to a single term, and, in particular, the summation in
brackets on e′ = ↑, ↓ collapses to the single term Uee(i E
′ ) . To see why, notice that the summation in brackets

is
X
E E E
Uee ′ δe′ e(i′ ) ≡ Ue↑ δ↑e(i′ ) + Ue↓ δ↓e(i′ )

e′
(

T
E
Ue↑ + 0 for e(i′ ) = ↑,
= E
0 + Ue↓ for e(i′ ) = ↓
E
= Uee(i ′),

so our expression for Ψi,e (t) simplifies to

AF X
Ψi,e (t) = UiiS′ Ψi′ (t′ )Uee(i
E
′). (22.41)
i′

From the Born rule (17.19), the probability of the composite system SE being in its configuration (i, e)
at the detection time t > t′ is given by
pi,e (t) = |Ψi,e (t)|2 = Ψi,e (t)Ψi,e (t).
Inserting our expression (22.41) for Ψi,e (t) gives
X X
pi,e (t) = UiiS′ Ψi′1 (t′ )Uee(i
E
′) UiiS′2 Ψi′2 (t′ )Uee(i
E
′ ),
2
1 1
i′1 i′2

where we’ve relabeled the index of the first summation from i′ to i′1 and relabeled the index of the second
summation from i′ to i′2 to make clear that these are two distinct summations. Rearranging, we arrive at
DR
the complicated-looking expression
XX
pi,e (t) = UiiS′ UiiS′2 Ψi′1 (t′ )Ψi′2 (t′ )Uee(i
E E
′ ) Uee(i′ ) .
2
(22.42)
1 1
i′1 i′2

To obtain the standalone probability pi (t) for the electron to be in its ith configuration at the detection
time t > t′ , we marginalize over the indicator qubit’s two configurations e = ↑, ↓, as usual:
X
pi (t) = pi,e (t) ≡ pi,↑ (t) + pi,↓ (t).
e=↑,↓

Substituting in our expression (22.42) for the composite system’s probability pi,e (t) gives a formula that, at
first glance, looks even more complicated than before:
XXX
pi (t) = UiiS′ UiiS′2 Ψi′1 (t′ )Ψi′2 (t′ )Uee(i
E E
′ ) Uee(i′ ) .
2
1 1
e i′1 i′2

Rearranging the summations, we have

" #
XX X
pi (t) = UiiS′ UiiS′2 Ψi′1 (t′ )Ψi′2 (t′ ) E
Uee(i E
′ ) Uee(i′ )
2
. (22.43)
1 1
i′1 i′2 e
 For now, let’s look at just the summation on e = ↑, ↓ in brackets:
X
E E
Uee(i ′ ) Uee(i′ ) .
2 1
e

Recalling that the adjoint operation † is defined for any matrix X by transposition and complex-conjugation,

E†
X † ij ≡ Xji , we have Uee(i
E
′ ) = Ue(i′ )e , so we can write the expression above as
1 1

X E† E
Ue(i ′ )e Uee(i′ ) .
2
1
e

In this form, we see that we’re multiplying the matrix UE† and the matrix UE , and because these are unitary
matrices, the result is just the 2 × 2 identity matrix 1:
 
1 0

T
E† E
U U =1≡ .
0 1

The entries of the identity matrix are given by the Kronecker delta, so we conclude that
(
X E† 1 for i′1 = i′2 ,
E
Ue(i′ )e Uee(i′2 ) = δi′1 i′2 ≡
AF e
1 0 for i′1 ̸= i′2 .

Inserting this result into our formula (22.43) for the electron’s standalone probability pi (t) at the detection
time t > t′ , we obtain XX  
pi (t) = UiiS′ UiiS′2 Ψi′1 (t′ )Ψi′2 (t′ ) δi′1 i′2 .
1
i′1 i′2

The Kronecker delta δi′1 i′2 is equal to 1 if i′2 = i′1 and is equal to 0 if i′2 ̸= i′1 , so the summation on i′2 collapses
to the single term for which i′2 = i′1 , thereby giving us
X
pi (t) = UiiS′ UiiS′1 Ψi′1 (t′ )Ψi′1 (t′ ).
1
i′1

Now that we’re down to just a single summation, we’re free to relabel the summation index i′1 back to i′ to
make the notation simpler: X
UiiS′ UiiS′ Ψi′ (t′ )Ψi′ (t′ ).
DR
pi (t) =
i′

Using zz = |z| again to write

2

UiiS′ UiiS′ = |UiiS′ |2 , Ψi′ (t′ )Ψi′ (t′ ) = |Ψi′ (t′ )|2 ,

and restoring the argument (t ← t′ ), we obtain

X
pi (t) = |UiiS′ (t ← t′ )|2 |Ψi′ (t′ )|2 . (22.44)
i′ =1,2

From the Born rule, we know that pi′ (t′ ) = |Ψi′ (t′ )|2 is just the probability that the electron is in its i′ th
configuration at the wall time t′ , and recalling (22.1), we know that |UiiS′ (t ← t′ )|2 defines the entries of a
unistochastic matrix ΓS (t ← t′ ),
ΓSii′ (t ← t′ ) ≡ |UiiS′ (t ← t′ )|2 . (22.45)

We can therefore recast our expression (22.44) for the electron’s standalone probability pi (t) at the detection
time t > t′ more succinctly as
X
pi (t) = ΓSii′ (t ← t′ )pi′ (t′ ). (22.46)
i′ =1,2
This equation has precisely the form of a linear marginalization relationship (9.1) relating arbitrary stan-
dalone probabilities pi′ (t′ ) for the electron at the time t′ to the corresponding standalone probabilities pi (t)
at the time t. This marginalization relationship looks just like our original marginalization formula (16.15),
but with t′ playing the role of a new ‘t = 0.’
Applying our original marginalization formula (16.15) to obtain the electron’s standalone probabilities
pi′ (t′ ) at the wall time t′ gives
X
pi′ (t′ ) = ΓSi′ j (t′ )pj (0). (22.47)
j=1,2

Inserting this formula into the right-hand side of (22.46) yields

 
X X
pi (t) = ΓSii′ (t ← t′ ) ΓSi′ j (t′ )pj (0).
i′ =1,2 j=1,2

T
Rearranging, we find  
X X
pi (t) =  ΓSii′ (t ← t′ )ΓSi′ j (t′ )pj (0).
j=1,2 i′ =1,2

This formula is precisely a linear marginalization relationship between the electron’s standalone probabilities
AF
pj (0) at the initial time t = 0 and its standalone probabilities pi (t) at the detection time t:
X
pi (t) = ΓSij (t)pj (0), (22.48)
j=1,2

with an overall stochastic matrix ΓS (t) given by

X
ΓSij (t) ≡ ΓSii′ (t ← t′ )ΓSi′ j (t′ ), (22.49)
i′ =1,2

or, equivalently,
ΓS (t) ≡ ΓS (t ← t′ )ΓS (t′ ). (22.50)
DR
The relation (22.50) is precisely of the form (15.1) for a stochastic matrix that’s divisible at t′ , so it
would be natural to call t′ a division event. What we learn from this calculation is that the interaction at
the wall time t′ between the electron and the indicator qubit, in which the indicator qubit reads the electron’s
configuration, corresponds to a division event for the electron’s stochastic dynamics. As a corollary, we see
that t = 0 is not special in being a division event, but that division events for systems will occur whenever
an ‘environment’ of some kind reads the system’s configuration.
We can learn even more from the foregoing calculation. Notice that if the electron’s probability distribu-
tion pi′ (t′ ) at the wall time t′ happens to be pure, so that pi′ (t′ ) = 1 for a specific value of the configuration
i′ and is equal to 0 for all other configurations, then the marginalization formula (22.46) reduces to

pi (t) = ΓSii′ (t ← t′ ) ← [if the electron is definitely in its configuration i′ at t′ ].

This formula tells us immediately that ΓSii′ (t ← t′ ), as we might have expected, is the conditional probability
for the electron to be in its ith configuration at t, given that it’s in its i′ th configuration at t′ :

ΓSii′ (t ← t′ ) = p(i, t|i′ , t′ ). (22.51)

Turning this equation around and invoking the definition ΓSii′ (t ← t′ ) ≡ |UiiS′ (t ← t′ )|2 from (22.45), we have

p(i, t|i′ , t′ ) = |UiiS′ (t ← t′ )|2 .

Re-expressing the right-hand side in dictionary form, akin to (16.9), gives

p(i, t|i′ , t′ ) = tr(US† (t ← t′ )PSi US (t ← t′ )PSi′ ).

Using the cyclic property of the trace, (11.27), we can move the matrix U† (t ← t′ ) over to the other side of
the trace:
 

p(i, t|i′ , t′ ) = tr PSi US (t ← t′ )PSi′ US† (t ← t′ ) . (22.52)

Comparing this expression with the formula (16.23) for the standalone probability pi (t) in terms of a trace
over the product of the ith configuration projector PSi and the electron’s density matrix ρS (t),

 

pi (t) = tr PSi ρS (t) = tr PSi US (t)ρS (0)US† (t) ,

we see immediately that in conditioning on the electron’s configuration i′ at the wall time t′ —or, equivalently,
conditioning on the indicator qubit’s reading at t′ —we have effectively replaced the electron’s density matrix

T
ρS (t) with
US (t ← t′ )PSi′ US† (t ← t′ ),
which reduces at the wall time t′ itself to just the single projector

PSi′ = ei′ e†i′ .

AF
That is, conditioning on i′ at t′ is equivalent to collapsing the electron’s density matrix at t′ to PSi′ . In
other words, after conditioning on i′ at t′ , we should use t′ in place of t = 0, we should use US (t ← t′ ) in
place of US (t), and we should use PSi′ in place of ρS (0). Hence, whatever the electron’s state vector or wave
function ΨS (t′ − ϵ) was at a small time ϵ just before t′ —that is, just before the interaction with the indicator
qubit—we should replace that state vector with ei′ starting at t′ :

ΨS (t′ − ϵ) 7→ ei′ . (22.53)

This formula represents the simplest version of wave-function collapse, which we see arises from a special
case of conditioning.

Measurement in the Double-Slit Experiment

DR
To clarify what’s going on here, let’s now study the time interval from the initial t = 0 to the wall time t′ , just
after the indicator qubit in our double-slit experiment has interacted with the electron. This time interval
will capture the simplest kind of measurement in quantum theory, and will serve as the basic paradigm for
measurements of quantum-theoretic observables more generally—even beyond the special case of random
variables, which are represented by diagonal matrices.
By assumption, the initial configuration of the composite system SE consisting of the electron S and the
indicator qubit E at t = 0 is (j, ↑) for some specific value of the electron’s configuration label j—say, j = 1.
The composite system therefore has an initial state vector that we can take to be
   
Ψ(1,↑) (0) 1
Ψ(2,↑) (0) 0
Ψ(0) =    
Ψ(1,↓) (0) = 0, (22.54)
Ψ(2,↓) (0) 0

or, equivalently,
(
1 for (j, e0 ) = (1, ↑),
Ψj,e0 (0) = δj1 δe0 ↑ = (22.55)
0 for (j, e0 ) ̸= (1, ↑),

where we use the special label e0 here to emphasize that this index refers to the indicator qubit’s configuration
at the initial time t = 0.
 This state vector is simple enough that it admits a special kind of factorization. To start, let’s introduce
two 2 × 1 state vectors ΨS (0) and ΨE (0) that are respectively defined to be the configuration-basis vectors
e1 and e↑ , which we can write as mathematically identical column vectors:
 S   
Ψ1 (0) 1
ΨS (0) ≡ ≡ = e1 ,
ΨS2 (0) 0
 E    (22.56)
Ψ↑ (0) 1
ΨE (0) ≡ ≡ = e↑ .
ΨE↓ (0) 0

That is,
(
1 for j = 1,
ΨSj (0) = δj1 =
0 for j = 2,
( (22.57)
1 for e0 = ↑,

T
ΨEe0 (0) = δ e0 ↑ =
0 for e0 = ↓ .

Notice then from (22.54) that

   S 
1 Ψ1 (0)ΨE↑ (0)
0 ΨS2 (0)ΨE↑ (0)
Ψ(0) =    
0 = ΨS1 (0)ΨE↓ (0), (22.58)
AF
or, equivalently, from (22.55), that
0 ΨS2 (0)ΨE↓ (0)

Ψj,e0 (0) = δj1 δe0 ↑ = ΨSj (0)ΨEe0 (0).

We can capture this factorization by writing the composite system’s state vector as the tensor product
(22.59)

Ψ(0) = e1 ⊗ e↑ . (22.60)

If we assume that the indicator qubit remains unchanging until the interaction with the electron just
before the wall time t′ , then for a small time interval ϵ, we can model the time evolution of the composite
system up until that interaction as the unitary 4 × 4 matrix

USE (t′ − ϵ) ≡ US (t′ − ϵ) ⊗ 1. (22.61)

DR
Here US (t′ − ϵ) handles the time evolution of the electron from t = 0 to t′ − ϵ, whereas the time evolution
of the indicator qubit is trivial during that time interval, as is captured by the identity matrix 1. The
interaction takes place from t′ − ϵ to t′ , and a simple way to model it is using the unitary 4 × 4 matrix defined
by
X2
USE (t′ ← t′ − ϵ) ≡ PSi′ ⊗ REe(i′ ) ≡ PS1 ⊗ RE↑ + PS2 ⊗ RE↓ , (22.62)
i′ =1

where RE↑ is a 2×2 unitary matrix that leaves e↑ unchanged and RE↓ is a 2×2 unitary matrix that transforms
e↑ to e↓ :
RE↑ e↑ = e↑ , RE↓ e↑ = e↓ . (22.63)
For example, we could take
   
1 0 0 1
RE↑ ≡ ≡ 1, RE↓ ≡ , (22.64)
0 1 1 0

where, by coincidence, RE↓ happens to have precisely the same form as the first Pauli sigma matrix σx , as
defined in (12.58). Altogether, then, the time-evolution operator USE (t′ ) for the composite system from 0
to t′ is given by
USE (t′ ) ≡ USE (t′ ← t′ − ϵ)USE (t′ − ϵ). (22.65)
Remember that all we’re doing by introducing this time-evolution operator is defining the entries of a
corresponding unistochastic matrix ΓSE (t′ ), as usual, where this unistochastic matrix captures how the
composite system SE probabilistically evolves with time.
We’re now ready to calculate the composite system’s state vector Ψ(t′ ) at the wall time t′ , just after the
measurement interaction between the electron S and the indicator qubit E. Acting with the time-evolution
operator USE (t′ ) on the composite system’s initial state vector Ψ(0), in accordance with the general rule
(17.12), we find

Ψ(t′ ) = USE (t′ )Ψ(0)

= USE (t′ ← t′ − ϵ)USE (t′ − ϵ)Ψ(0)
" 2 #
X  
= Pi′ ⊗ Re(i′ ) US (t′ − ϵ) ⊗ 1 [e1 ⊗ e↑ ].
S E

i′ =1

T
It is in the nature of tensor products that they sequester indices on matrices in precisely such a way that

(X ⊗ Y)(v ⊗ w) = (Xv) ⊗ (Yw). (22.66)

Hence,
AF Ψ(t′ ) =
2
X 

i′ =1
 h i
PSi′ US (t′ − ϵ)e1 ⊗ REe(i′ ) e↑ .

By definition, acting with US (t′ − ϵ) on ΨS (0) ≡ e1 gives the electron’s state vector ΨS (t′ − ϵ) at the time
t′ − ϵ just before the measurement interaction with the indicator qubit,

US (t′ − ϵ)ΨS (0) = ΨS (t′ − ϵ),

and acting on the result with the projector PSi′ yields

 !
DR

 ΨS1 (t′ − ϵ)


 for i′ = 1,
  
 0
ΨS1 (t′ − ϵ)
PSi′ ΨS (t′ − ϵ) = PSi′ =
ΨS2 (t′ − ϵ) 
 !

 0

 for i′ = 2

ΨS2 (t′ − ϵ)
= ΨSi′ (t′ − ϵ)ei′ .

Meanwhile, we know from (22.63) that

REe(i′ ) e↑ = ee(i′ ) .

Hence,

2
X
′
   
Ψ(t ) = ΨSi′ (t′ − ϵ)ei′ ⊗ ee(i′ )
i′ =1
2
X  
= ΨSi′ (t′ − ϵ) ei′ ⊗ ee(i′ ) .
i′ =1
Taking the limit ϵ → 0 of a very short measurement interaction, we therefore end up with
2
X  
Ψ(t′ ) = ΨSi′ (t′ ) ei′ ⊗ ee(i′ )
i′ =1
= ΨS1 (t′ )[e1⊗ e↑ ] + ΨS2 (t′ )[e2 ⊗ e↓ ]
   
1 0
 0 
S ′ 0

= ΨS1 (t′ ) 
0 + Ψ2 (t )0
0 1
 S ′ 
Ψ1 (t )
 0 
= 0 .


ΨS2 (t′ )

T
The result is therefore precisely the required form (22.39) of the state vector for the composite system at
the wall time t′ .
At the end of this interaction, at t′ , we know from (22.22) that the composite system SE has joint
probabilities
AF p1,↑ (t′ ) = |Ψ1,↑ (t′ )|2 = |ΨS1 (t′ )|2 ,
p2,↑ (t′ ) = |Ψ2,↑ (t′ )|2 = 0,
p1,↓ (t′ ) = |Ψ1,↓ (t′ )|2 = 0,
p2,↓ (t′ ) = |Ψ2,↓ (t′ )|2 = |ΨS2 (t′ )|2 .

Hence, if we marginalize over the electron’s configurations i = 1, 2, then we see that the indicator qubit’s
standalone probabilities at t′ are

p↑ (t′ ) = p1,↑ (t′ ) + p2,↑ (t′ ) = |ΨS1 (t′ )|2 ,

(22.67)
p↓ (t′ ) = p1,↓ (t′ ) + p2,↓ (t′ ) = |ΨS2 (t′ )|2 .
DR
These equations mean that |ΨSi′ (t′ )|2 not only tells us the probability of the electron being in its i′ th config-
uration at t′ , but also the probability of the indicator qubit being in its e(i′ )th configuration at t′ :

pe(i′ ) (t′ ) = |ΨSi′ (t′ )|2 . (22.68)

This result is another, more general form of the Born rule—the Born rule for measurement outcomes or
readings on measuring devices. That is, |ΨSi′ (t′ )|2 also has the meaning of the standalone probability for
the measuring device in this case—namely, the indicator qubit—to be in an outcome-configuration e(i′ )
corresponding to the appropriate value of the electron’s configuration label i′ .