Eur. Phys. J.

H (2021)46:7
https://doi.org/10.1140/epjh/s13129-021-00001-0
THE EUROPEAN
PHYSICAL JOURNAL H
Regular Article

Everett’s theory of the universal wave function

Biao Wu1,2,3,a
1
International Center for Quantum Materials, School of Physics, Peking University, 100871 Beijing, China
2
Wilczek Quantum Center, School of Physics and Astronomy, Shanghai Jiao Tong University, Shanghai 200240, China
3
Collaborative Innovation Center of Quantum Matter, Beijing 100871, China

Received 21 May 2020 / Accepted 20 November 2020

© The Author(s), under exclusive licence to EDP Sciences, Società Italiana di Fisica and Springer-Verlag
GmbH Germany, part of Springer Nature 2021, corrected publication 2021

Abstract This is a tutorial for the many-worlds theory by Everett, which includes some of my personal
views. It has two main parts. The first main part shows the emergence of many worlds in a universe
consisting of only a Mach–Zehnder interferometer. The second main part is an abridgment of Everett’s long
thesis, where his theory was originally elaborated in detail with clarity and rigor. Some minor comments
are added in the abridgment in light of recent developments. Even if you do not agree to Everett’s view, you
will still learn a great deal from his generalization of the uncertainty relation, his unique way of defining
entanglement (or canonical correlation), his formulation of quantum measurement using Hamiltonian, and
his relative state.

Part I: Prologue word entanglement; instead, he called it canonical cor-

relation or simply correlation. I will use entanglement
Although Everett’s many-worlds theory is now well in this abridgment. In addition, I’ll use Dirac brackets
known, it is still a minority view among physicists. wherever possible.
There are many reasons, which I have no intention to Before the abridgment, I use the Mach–Zehnder
discuss extensively here. One of the reasons may be interferometer (MZI) to illustrate the many-worlds the-
that many physicists have not read his work seriously. ory. It appears to me that the MZI is a simple example
Everett’s theory was presented in his PhD thesis, which to illustrate all the essential points in Everett’s long
has two versions. The long version has over 130 pages thesis. In particular, the MZI is ideal to demonstrate
and was finished in 1956. It was published only 17 years interference between different worlds and the essence
later for the first time with the title The Theory of The of approximate measurement, which was discussed in
Universal Wave Function in the book edited by DeWitt detail by Everett in his long thesis and has not been
and Graham [1] and was re-published with commen- discussed much since. Near the end of this part, I also
tary in 2012 [2]. Due to Bohr’s objection, Everett had discuss the issue of preferred basis and I think that it
to shorten it. The short version became his official PhD is related to the perceptive abilities of observers. Hope-
thesis at Princeton University [3] and was published fully, this example of MZI will aid your reading of the
with the title “Relative State” Formulation of Quantum abridgment.
Mechanics in Review of Modern Physics [4] accompa- The two words, universe and world, are often used
nied by an article by his advisor Wheeler [5]. differently by different people when they discuss
On the one hand, Everett’s short thesis lacks many Everett’s theory. It was first dubbed “many-worlds”
important results in his long thesis, e.g., entanglement theory by DeWitt [1]. In this way, we say that there
(or canonical correlation) and formulation of quantum is one universe that consists of many different worlds.
measurement. On the other hand, the long version may However, Everett’s theory has recently often been called
be too long for many people’s patience, which is further the theory of multiverse. In this way, we say that there
exasperated by Everett’s mathematical notations that is one world that consists of many different universes [6].
are not familiar to modern readers. It is my hope that It rubs salt to the wound that multiverse has different
this abridgment makes a good compromise between the meanings for different people in literature [7,8]. So, to
long and short thesis. In this abridged version, I will avoid the confusion, we stick to DeWitt’s term and say
keep its structure and stick to Everett’s original state- that there is one universe that consists of many different
ments as much as possible at key points while omitting worlds.
detailed discussion and derivations. Entanglement is all It is my sincere hope that you will eventually find
over the long thesis. However, Everett never used the time to read Everett’s long thesis in its entirety, which
is richer in content than the short version and juicier
a
than this abridgment. Finally, even if you do not agree
e-mail: [email protected] (corresponding author)

0123456789().: V,-vol 123

7 Page 2 of 20 Eur. Phys. J. H (2021)46:7

S materials of different refractive indices [14]. With this

M1 M2
Photon choice of phase differences, the beam splitter functions
exactly as a Hadamard gate.
The sample S causes a phase shift θ to the wave func-
tion that passes through; the reflections by the mirrors
M2 and M4 interchange |φH  and |φV  and cause a π
phase shift. As a result, right before encountering the
M4 M3 mirror M3 , the photon’s wave function becomes
DH
1
|ψ1  = √ (eiθ |φH  − |φV ). (0.2)
2
DV
At the mirror M3 , |φH  further√splits into two compo-
Fig. 1 Mach–Zehnder interferometer. M1 and M3 are half-
silvered mirrors; M2 and M4 are reflective mirrors. S is some nents becomes (|φ √H  + |φV )/ 2 and |φV  splits into
sample material that causes a phase shift to the photon’s (− |φH  + |φV )/ 2. Consequently, we have
wave function. DH and DV are two photon detectors. Note
that M1 and M3 are not symmetric and their two sides are 1  iθ 
made of materials of different refractive indices |ψ2  = e (|φH  + |φV ) + |φH  − |φV 
2
 θ θ 
= eiθ/2 cos |φH  + i sin |φV  . (0.3)
to his view, you will certainly get entertained and 2 2
inspired by how Everett generalized uncertainty rela-
tion, defined entanglement (or canonical correlation), This means that the probability that the photon be
formulated quantum measurement, and introduced rel- detected by the detector DH is cos2 (θ/2) and the prob-
ative state. ability detected by the detector DV is sin2 (θ/2).
The mysterious part of the MZI is the following. On
the one hand, the photon wave function has two parts,
Part II: The universe of MZI |φH  and |φV , right before the detection. On the other
hand, in a single run of the experiment, there is only
one detection either at DH or DV . Suppose that DH
The Mach–Zehnder interferometer (MZI) was proposed detects a photon in one experiment; this detection is
by Zehnder in 1891 [9] and was refined by Mach in clearly triggered by the |φH  term in Eq. (0.3). So, why
1892 [10]. Due to its simplicity and flexibility, the MZI does not the other term |φV  trigger a detection at DV ?
has not only enjoyed wide applications [11] but also What has happened to |φV ? According to the conven-
often been used for illustration of fundamental sub- tional view , both |φH  and |φV  can trigger detection,
tleties in quantum mechanics [6,12]. We follow the but it is purely random which one triggers. Further-
crowd and use it to illustrate Everett’s many-worlds more, when one of them triggers detection, the other
theory. We first briefly review the Mach–Zehnder inter- part magically disappears. This is called the collapse of
ferometer (MZI) in a conventional way. wave function. Everett showed in details the collapse of
As shown in Fig. 1, MZI consists of four mirrors. The wave function would lead to two difficulties in his long
mirrors M1 and M3 are half-silvered and serve as beam thesis [1]. The first difficulty is that it would lead to log-
splitter. The mirrors M2 and M4 are reflective. Initially, ical inconsistency when there are two or more observers;
the photon’s wave function has only the horizontal com- the second difficulty is that it is inadequate to deal with
ponent, i.e., |ψ0  = |φH . After the photon encounters approximate measurement.
M1 , its wave function splits and has two components In both his long thesis and short thesis [1,4],
Everett had used branches or just elements of superpo-
1 sition instead of worlds referring to the different super-
|ψ1  = √ (|φH  − |φV ). (0.1)
2 position components in a wave function. Here in this
work, we will often use worlds as his theory is now
Note that there is an ambiguity for the phase differ- widely known as the many-worlds theory [1].
ence between the transmitted component |φH  and the We now analyze MZI with the many-worlds theory.
reflection component |φV . In general, the phase differ- We assume that the universe consists only of MZI and
ence depends on the incident side of the beam splitter. nothing else. There is no gravity. The two detectors
If δ1 is the phase difference when the photon is incident can absorb the photon with 100% efficiency. The mir-
on the left side of the beam splitter and δ2 is the phase rors M1 and M3 are at rest initially and arranged as
difference when the photon is incident on the right side, in Fig. 1 with no support or attached wires while both
then δ1 + δ2 = π [13]. The exact values of δ1 and δ2 the mirrors M2 and M4 are fixed in space. With mir-
depend on how the beam splitter is manufactured. In rors M1 and M3 movable, we can discuss the conditions
the above we have used δ1 = π and δ2 = 0. This occurs for interference to occur. If one consider a more com-
when the two sides of the beam splitter are made of plicated situation where neither of the mirrors M2 and

123
Eur. Phys. J. H (2021)46:7 Page 3 of 20 7

M4 are fixed, the analysis would become much more After the photon interacts with the mirror M1 , we have
complicated without gaining essentially new physics.
There are two different kinds of interactions in this 1  
universe of MZI: photon with half-silvered mirror and |Ψ1  = Ue |Ψ0  = √ |φH , ψ0M1  − |φV , ψpM1 1 
2
photon with the reflective mirror. We use Ue denote the
former and U0 the latter. The interaction at the mirror ⊗ |ψ0 , ψ0 , ψ0M4 , ψ0DH , ψ0DV  .
M2 M3
(0.11)
M1 can be mathematically expressed as
According to the many-worlds theory, the two compo-
nents in |Ψ1 , which are orthogonal to each other, rep-
Ue |φH  ⊗ |ψ0M1  resent two different worlds: in one world the photon
1   travels horizontally and in the other world the photon
= √ |φH  ⊗ |ψ0M1  − |φV  ⊗ |ψpM1 1  , (0.4) travels vertically. After the sample S, we still have two
2
worlds but one world has acquired a phase shift
where |ψ0M1  and |ψpM1 1  are the states of the mir- 1  
ror before and after the interaction, respectively. After |Ψ1  = √ eiθ |φH , ψ0M1  − |φV , ψpM1 1 
the interaction, if the photon continues to move hor- 2
izontally, nothing changes; if the photon moves verti- ⊗ |ψ0M2 , ψ0M3 , ψ0M4 , ψ0DH , ψ0DV  . (0.12)
cally, the mirror acquires a momentum p1 and its state
becomes |ψpM1 1 . Overall, it results an entangled state The photon is then reflected by the two mirrors M2 and
between the photon and the mirror. Similarly, at the M4 and the state of the universe becomes
mirror M3 , we have
1  
|Ψ2  = U0 |Ψ1  = √ |φH , ψpM1 1  − eiθ |φV , ψ0M1 
Ue |φH  ⊗ |ψ0M3  2
1   ⊗ |ψ0 , ψ0 , ψ0M4 , ψ0DH , ψ0DV  .
M2 M3
(0.13)
= √ |φH  ⊗ |ψ0M3  + |φV  ⊗ |ψpM1 3  , (0.5)
2
The universe still has only two worlds. Now the photon
interacts with the mirror M3 , resulting in the following
and state of the universe

Ue |φV  ⊗ |ψ0M3  |Ψ3  = Ue |Ψ2 

1  
1
= √ |φV  ⊗ |ψ0M3  − |φH  ⊗ |ψpM3 3  . (0.6) = |φH , ψpM1 1 , ψ0M3  + |φV , ψpM1 1 , ψpM1 3 
2 2
−eiθ (|φV , ψ0M1 , ψ0M3  − |φH , ψ0M1 , ψpM3 3 )
Note that p1 = −p3 . The reflective interaction at the
mirror M2 has the following mathematical form ⊗ |ψ0M2 , ψ0M4 , ψ0DH , ψ0DV  . (0.14)

U0 |φH  ⊗ |ψ0M2  = − |φV  ⊗ |ψ0M2  . (0.7) It appears that there are now four worlds in the uni-
verse. But in general the four terms above are not
orthogonal to each other. There are in fact seven worlds.
And similarly at the mirror M4 , we have To see it, let us expand |ψpM1 1  as

U0 |φV  ⊗ |ψ0M4  = − |φH  ⊗ |ψ0M4  . (0.8) |ψpM1 1  = α1 |ψ0M1  + β1 |ψ⊥

M1
, (0.15)

M1
No entanglement is generated in this interaction, and where ψ⊥ |ψ0M1  = 0. Similarly, we have
the mirrors do not gain momentum as they are fixed in
space.
Initially, the universe of MZI is described by the fol- |ψpM1 3  = α3 |ψ0M3  + β3 |ψ⊥
M3
, (0.16)
lowing wave function
and
|Ψ0  = |φH  ⊗ |ψ0M1  ⊗ |ψ0M2  |ψpM3 3  = α3∗ |ψ0M3  + β3∗ |ψ⊥
M3 ∗
, (0.17)
⊗ |ψ0M3  ⊗ |ψ0M4  ⊗ |ψ0DH  ⊗ |ψ0DV  . (0.9)
M3
where ψ⊥ |ψ0M3  = ψ⊥
M3 ∗ M3
|ψ0  = 0. In the above we
Whenever there is no confusion arising, we omit ⊗ and have used that p1 = −p3 implies |ψpM1 3  = |ψpM3 3 ∗ . We
simplify the above expression as will discuss these coefficients α1,3 and β1,3 later. With
these expansions, we have
|Ψ0  = |φH , ψ0M1 , ψ0M2 , ψ0M3 , ψ0M4 , ψ0DH , ψ0DV  .
1
(0.10) |Ψ3  = (α1 + eiθ α3∗ ) |φH , ψ0M1 , ψ0M3 
2

123
7 Page 4 of 20 Eur. Phys. J. H (2021)46:7

+(α1 α3 − eiθ ) |φV , ψ0M1 , ψ0M3 

M1
+β1 |φH , ψ⊥ , ψ0M3  + β3∗ eiθ |φH , ψ0M1 , ψ⊥
M3 ∗

M1
+α3 β1 |φV , ψ⊥ , ψ0M3  + α1 β3 |φV , ψ0M1 , ψ⊥
M3

M1 M3
+β1 β3 |φV , ψ⊥ , ψ⊥ 
⊗ |ψ0M2 , ψ0M4 , ψ0DH , ψ0DV  . (0.18)

These seven terms are orthogonal to each other. In the

end, the photon is detected by the detectors and we
have a universe that consists of seven different worlds
that exist simultaneously. And the wave function of the
universe is
1
|Ψf  = (α1 + eiθ α3∗ ) |φH , ψ0M1 , ψ0M3 , ψ1DH , ψ0DV 
2
+(α1 α3 − eiθ ) |φV , ψ0M1 , ψ0M3 , ψ0DH , ψ1DV 
M1
+β1 |φH , ψ⊥ , ψ0M3 , ψ1DH , ψ0DV  Fig. 2 Worlds in the universe of Mach–Zehnder interfer-
ometer in two special cases. a The pure interference (PI)
+β3∗ eiθ |φH , ψ0M1 , ψ⊥
M3 ∗
, ψ1DH , ψ0DV  case; b the pure split (PS) case. |ΨH  and |ΨV  represent
M1
+α3 β1 |φV , ψ⊥ , ψ0M3 , ψ0DH , ψ1DV  worlds where the photon moves horizontally and vertically,
respectively. The superscript d in |ΨdH  and |ΨdV  represents
+α1 β3 |φV , ψ0M1 , ψ⊥
M3
, ψ0DH , ψ1DV  the states of mirrors have changed relative to the initial
M1 M3 state of the MZI universe
+β1 β3 |φV , ψ⊥ , ψ⊥ , ψ0DH , ψ1DV 
⊗ |ψ0M2 , ψ0M4  . (0.19) only one world and it splits into two worlds with equal
weight at the mirror M1 . These two worlds evolve in
In this final state, everything in the universe except the parallel without changing their weights before interfer-
mirrors M2 and M4 are entangled together. The photon ing at the mirror M3 . As a result of the interference,
detection should be presented as we still have two worlds but with different weights. In
the special case θ = 0, there is only one world after the
|φH  ⊗ |ψ0DH  → |ψ1DH  (0.20) interference.
Consider the second special case, α1 = α3 = 0 and
to reflect the fact that the photon is absorbed by the β1 = β3 = 1. We call it pure split (PS) case. In the PS
detector. However, in the above, to explicitly represent case, we have
the photon state before the detection, we have kept φH
and φV . This should not cause confusion. 1
We consider two special cases. In the first case, which |Ψf 2  = |φH , ψpM1 1 , ψ0M3 , ψ1DH , ψ0DV 
2
we call pure interference (PI) case, α1 = α3 = 1 and +eiθ |φH , ψ0M1 , ψpM3 3 , ψ1DH , ψ0DV 
β1 = β3 = 0. In the PI case, we have
−eiθ |φV , ψ0M1 , ψ0M3 , ψ0DH , ψ1DV 
1
|Ψf 1  = (1 + eiθ ) |φH , ψ0M1 , ψ0M3 , ψ1DH , ψ0DV  + |φV , ψpM1 1 , ψpM1 3 , ψ0DH , ψ1DV 
2
+(1 − eiθ ) |φV , ψ0M1 , ψ0M3 , ψ0DH , ψ1DV  ⊗ |ψ0M2 , ψ0M4  . (0.22)

⊗ |ψpM2 , ψpM4  The evolution of the worlds in this case is illustrated in

 θ Fig. 2b. The evolution is similar to the first case before
= eiθ/2 cos |φH , ψ1DH , ψ0DV  the mirror M3 . The crucial difference is that there is
2
θ no interference at M3 in this case. Consequently, the
+ sin |φV , ψ0DH , ψ1DV  worlds keep splitting and we obtain four different worlds
2 with equal weights. And the phase shift θ has no effect
⊗ |ψ0M1 , ψ0M3 , ψ0M2 , ψ0M4  . (0.21) on the weights of the different worlds.
We now examine the expansions in Eqs. (0.15),
This is exactly the state in Eq. (0.3). The only dif- (0.16), (0.17) in detail. The mirrors, made of atoms,
ference is that the states of mirrors and detectors are have enormous amount of degrees of freedom. However,
not expressed explicitly in Eq. (0.3). This case happens in this MZI universe, only their centers of mass are rel-
when the two mirrors M1 and M3 are very massive or evant. Moreover, their centers of mass move only along
mounted in space and unmovable. Figure 2a shows how p1 = −p3 . With these considerations, we are allowed
the worlds split and evolve in this case. Initially, there is to describe the states of the mirror M1 before and after

123
Eur. Phys. J. H (2021)46:7 Page 5 of 20 7

the interaction as the following Gaussian wave packets

1
1/4  x2 
x|ψ0M1  = exp − 2 , (0.23)
πa2 2a

and

1
1/4  x2 
x|ψpM1 1  = exp − 2 + ikx ,
πa2 2a
(0.24)

where a is the width of the wave packet and k = |p1 |.

We obtain
 1 
α1 = ψ0M1 |ψpM1 1  = exp − a2 k 2 . (0.25)
4
Fig. 3 Worlds in the universe of Mach–Zehnder interfer-
Similarly, we can compute α3 and find that α3 = α3∗ = ometer in the general case. For clarity, the weight of each
α1 . In real experiments, the wave length of the photon world, which is represented by the length of the vertical line,
is much larger than the width a, so we have α1 = α3 = is not plotted to scale. The superscript d in |ΨdH  and |ΨdV 
α3∗ ∼ 1. This is exactly the PI case in Fig. 2a. One may represents the states of mirrors has changed relative to the
want to use a photon with much shorter wave length initial state of the MZI universe
so that α1 = α3 = α3∗  1 and β1 = β3 = β3∗ ∼ 1,
i.e., the PS case. However, the interaction of mirrors
with shorter-wave-length photon is very different and we have
the MZI can consequently cease to work.
1 iθ
There is one possible way to realize the PS case
θ
as illustrated in Fig. 2b. This is to add a very sensi- |Ψf  = αe 2 cos2 |φH , ψ0M1 , ψ0M3 , ψ1DH , ψ0DV 
tive detector DP that is capable of measuring the tiny 2 2
momentum that a mirror gains after interacting with +(α2 − eiθ ) |φV , ψ0M1 , ψ0M3 , ψ0DH , ψ1DV 
the photon. If the momentum is zero, the detector is M1
+β |φH , ψ⊥ , ψ0M3 , ψ1DH , ψ0DV 
described by |DP0 ; if the momentum is p1 or p3 , the
detector has the state |DP1 . These two states should +βeiθ |φH , ψ0M1 , ψ⊥
M3
, ψ1DH , ψ0DV 
be orthogonal to each other DP0 |DP1  = 0 to reflect M1
+αβ |φV , ψ⊥ , ψ0M3 , ψ0DH , ψ1DV 
the effectiveness of the detection. With the addition of
the new detector, the interaction Ue in Eq. (0.4) can be +αβ |φV , ψ0M1 , ψ⊥
M3
, ψ0DH , ψ1DV 
rewritten as M1 M3
+β 2 |φV , ψ⊥ , ψ⊥ , ψ0DH , ψ1DV 
Ue |φH  ⊗ |ψ0M1  ⊗ |DP0  ⊗ |ψ0M2 , ψ0M4  . (0.27)
1  
= √ |φH  ⊗ |ψ0M1 , DP0  − |φV  ⊗ |ψpM1 1 , DP1  .
2 This wave function is a superposition of seven mutually
(0.26) orthogonal elements, each of which describe a world. In
the order in the above equation, we call them worlds I,
Let |ψ0M1  = |ψ0M1 , DP0  and |ψpM1  = |ψpM1 1 , DP1 . II, III, IV, V, VI, and VII. The worlds I and II are the
results of interference with the world I existing only in
We clearly have ψ0M1 |ψpM1  = 0. As a result, when the PI case while the world II existing in both the PI
we expand as in Eq. (0.15) for |ψpM1 , we should have and PS cases. The worlds III, IV, and VII exist in the
α1 = 0 and β1 = 1. Similarly, we should have α3 = 0 PS case. The worlds V and VI are new and do not exist
and β3 = 1. In this way, we have effectively realized the in either of the two special cases. To understand these
PS case in Fig. 2b, where the worlds have split twice two new worlds, we expand the interaction in Eq. (0.4)
with no interference. Note that the discussion with the with Eq. (0.15)
detector DP is a matter of principles, not for realistic
1 
realization. In real experiments, other methods may be
used to distinguish the two states |ψ0M1  and |ψpM1 1  or Ue |φH  ⊗ |ψ0M1  = √ |φH  ⊗ |ψ0M1 
tell which direction the photon is going after encoun- 2
tering the mirror M1 . −α |φV  ⊗ |ψ0M1  − β |φV  ⊗ |ψ⊥
M1
 .
We now discuss the general case. As the above anal-
ysis shows that α1 = α3 = α and β1 = β3 = β3∗ = β, (0.28)

123
7 Page 6 of 20 Eur. Phys. J. H (2021)46:7

The term with α represents that the photon changes its general case, after the interaction Ue , we have
direction by the mirror M1 but the mirror state does not
change. We call it reflection with no detection. The term
with β represents that the photon changes its direc- Ue |φH  ⊗ |ψ0M1 
tion by the mirror M1 while the mirror state becomes
1−α  α 
orthogonal to its original state. We call it reflection with
= |φ+  ⊗ α |ψ1  + |ψ2 
detection. So, in the world V, the photon is reflected by 2 α
the mirror M1 with detection and then reflected by the
1+α  α 
mirror M3 with no detection; in the world VI, the pho- + |φ−  ⊗ |ψ1  + α |ψ2  ,
ton is reflected by the mirror M1 with no detection and 2 α
then reflected by the mirror M3 with detection. (0.33)
The analysis with the general case illustrates a crucial
point that the photon interferes only when its different
components, φH and φV , do not cause difference in the √ √
where α = ( 1 + α + 1 − α)/2. We no longer have
rest of the universe (e.g., the mirrors and detectors). a perfect measurement. When the apparatus (photon)
Whenever the different components of an object’s wave reads either ‘+’ or ‘−’, the mirror is not in the eigen-
function cause difference in other objects, interference states of B̂M , the target of our measurement. When α
disappears and decoherence occurs. is only slightly smaller than one, the resulted states are
In his long thesis, Everett offered an insight into
quantum measurement. In his view, quantum measure- very close to the eigenstates of B̂M and can be regarded
ment is a generation of entanglement between two sub- as an approximation. Everett call this kind of measure-
systems by an interacting Hamiltonian. We now illus- ment approximate measurement. It is clear in the spe-
trate it with the entanglement-generation interaction cial universe of MZI the approximate measurement is
described in Eq. (0.28). The photon is the “apparatus” more common than the precise measurement. It is the
same in our universe, the general universe.
whose reading is given by the operator Âp . The eigen- Several caveats are warranted here. (1) The approx-
states of Âp are imation is not the result of noises or other random
factors in real experimental setup. (2) The two oper-
1 ators Âp and B̂M are introduced for theoretical illus-
|φ±  = √ (|φH  ± |φV ), (0.29)
2 tration; it seems unlikely that they can be realized in
real experiments. (3) To the best of my knowledge,
such that Âp = ± |φ± . The system is the mirror M1 , nobody appears to have studied approximate measure-
whose property to be measured is given by the operator ment thoroughly since Everett , many fundamental
B̂M . The eigenstates of B̂M are questions need to be answered, for example, the pre-
cise definition of approximate measurement.
1 The above discussion has led us to another intriguing
|ψ1,2  = √ (|ψ0M1  ∓ |ψ⊥
M1
) (0.30) issue in quantum mechanics. We use Eq. (0.32) as an
2
illustration. On the left hand side, we have a familiar
with eigenvalues being 1 and 2, respectively. Before the universe that has split into two worlds: in one world,
interaction between the photon and the mirror, we have the photon moves horizontally and the mirror stays the
same; in the other world, the photon moves vertically
1 and the mirror changes into a state orthogonal to its
|φH  ⊗ |ψ0M1  = (|φ+  + |φ− ) ⊗ (|ψ1  + |ψ2 ), original state. On the right-hand side, the same universe
2 is split to two very different worlds: in one world, the
(0.31)
photon is in the state φ+ , an eigenstate of Âp , and
where there is no entanglement between the photon and the mirror is in the state ψ1 , an eigenstate of B̂M ; in
the mirror at all. After the interaction Ue , we have Eq. the other world, the photon is in the state φ− and the
(0.28). We first consider the special PS case, α = 0 and mirror is in the state ψ2 . In fact, there are infinite ways
β = 1. In this case, we can rewrite the right-hand side to re-write this entangled state. So, which represents the
of Eq. (0.28) as reality? For us, the world where the photon moves either
horizontally or vertically is the reality since we have the
1   ability to measure the photon’s position and the ability
√ |φH  ⊗ |ψ0M1  − |φV  ⊗ |ψ⊥
M1
 to measure whether the mirror has momentum or not.
2 If a different kind of creature or instrument can make
1  
measurements according to Âp and B̂M , then the right-
= √ |φ+  ⊗ |ψ1  + |φ−  ⊗ |ψ2  . (0.32)
2 hand side of Eq. (0.32) is the reality.
What kind of world that we perceive depends on our
We have a maximum entanglement (a perfect correla- abilities of perception. These different abilities mathe-
tion) here: when the apparatus (photon) reads ‘+’, we matically correspond to different bases. Suppose |ΦU 
know the mirror is in the state |ψ1 ; when the appa- is the wave function for the whole universe. For one
ratus reads ‘−’, the mirror is in the state |ψ2 . In the group of observers OA with a given set of measurement

123
Eur. Phys. J. H (2021)46:7 Page 7 of 20 7

abilities, it can be decomposed in a set of basis as surement on a system S and will record the result in
a notebook. The observer A is aware that the system
|ΦU  = |Φ1  + |Φ2  + · · · + |Φj  + · · · (0.34) S is in a quantum state |ψ that is not in an eigen-
state of the measurement. Another observer B is out-
These |Φj ’s are the worlds perceived by OA . For side of the room. Beside knowing the quantum state
another group of observers OB with a different set of |ψ and A is to perform a specified measurement, B
measurement abilities, the universe wave function can has no interaction at all with the room and everything
be decomposed in a different set of basis as inside the room. The observer A performs the measure-
ment and records the result in the notebook. One week
|ΦU  = |Φ1  + |Φ2  + · · · + |Φj  + · · · (0.35) later, B enters the room and performs his measure-
ment, that is, taking a look at the notebook. A and
B soon find themselves disputing each other: A insists
The worlds Φj ’s are very different from the worlds
that Process 1 (the collapse of the wave function |ψ
|Φj ’s. It is possible that even the space-time that we occurred when he performed the measurement. B is con-
are experiencing may look very different for another fident that the whole room should evolve according to
group of observers. Process 2 for one week. Process 1 occurred only when
he enters the room and performs his observation by
looking at the notebook. There are five different ways
Part III: Abridgment of Everett’s long thesis to resolve the paradox or the dispute between A and
B.
This abridgment is done by mostly paraphrasing Alternative 1 To postulate that there is only one
Everett’s long thesis; Everett’s words in complete sen- observer in the universe.
tences are rarely used. Chapters in the thesis become
sections in this abridgment. Everett used correlation Alternative 2 To limit the applicability of quantum
or canonical correlation to mean entanglement in his mechanics: quantum theory fails when it is applied to
thesis; I use entanglement in the abridgment wherever observers, measuring devices, or more generally any sys-
correlation is meant entanglement in Everett’s thesis. tem of macroscopic size.
In some cases the mathematical notation of Everett has Alternative 3 To deny the possibility of the outside
been updated to more modern style, such as in the use observer B could ever be in possession of the state func-
of Dirac bracket notation. The author’s words are indi- tion of A and S, where A is the observer inside the lab
cated with italic font. and S is the quantum system that A measures.
Alternative 4 To abandon the position that a wave func-
tion is a complete description of a system.
1 Introduction
Alternative 5 To assume that the universal validity of
An isolated quantum system is completely described the quantum description by the complete abandonment
by a wave function |ψ. According to standard text- of Process 1, i.e., the collapse of wave function.
books on quantum mechanics the wave function |ψ can Alternatives 1 and 2 are clearly hard to defend.
change in two fundamentally different ways [15] Alternative 4 can be viewed as hidden variable theory.
Local hidden variable theory has been refuted by Bell’s
inequality [16].
Process 1 Observation with respect to operator Ô that Alternative 3 is a bit ambiguous, at least in my opin-
has eigenfunctions |φ1  , |φ2  , |φ3  , . . . will transforms ion. No matter what, the first four alternatives need
discontinuously the wave function |ψ to one of the additional assumptions. In contrast, alternative 5 has
eigenfunctions, |φj , with probability | φj |ψ |2 . many advantages:

Process 2 Continuous and deterministic change of the • It relies on two basic ingredients of quantum mechan-
state |ψ with time according to the Schrödinger equa- ics: (1) The wave function |ψ in a Hilbert space
tion offers a complete description of a quantum sys-
tem; (2) |ψ evolves unitarily according to the
∂ Schrödinger equation.
ı |ψ = Ĥ |ψ , (1.1)
∂t • The quantum theory applies to the entire universe.
• Measurement is no longer a special process and can
where Ĥ is the operator. be described as any other physical processes.
Process 1 is commonly known as the collapse of wave
function. The key for developing alternative 5 is to study com-
The above scheme can lead to a paradox when there posite quantum systems and exploit the entanglement
are more than one observer. Consider a room isolated (or correlation in Everett’s own words) between subsys-
in space where one observer A is to perform a mea- tems.

123
7 Page 8 of 20 Eur. Phys. J. H (2021)46:7

2 Probability, information, and correlation the case where one value, say xj , has P (xj ) = 1 and
the other values have P (x = xj ) = 0. The latter is
This section (or chapter as used in Everett’s thesis) the case where every value has the same probability
offers a very general mathematical treatment of infor- P (xj ) = 1/m. This definition can be easily generalized
mation and correlation, which is used in later sections for many variables X 1 , X 2 , . . . , X n
to define correlation (or entanglement) between differ-
ent quantum subsystems. IX 1 ,X 2 ,...,X n

= P (x1i , x2j , . . . , xnk ) ln P (x1i , x2j , . . . , xnk )
2.1 Finite joint distribution X 1 ,X 2 ,...,X n

For a collection of finite sets, X 1 , X 2 , . . . , X n , we can = ln P (x1i , x2j , . . . , xnk ). (2.5)
define a joint probability distribution, P (x1i , x2j , . . . , xnk ),
where x1i ∈ X 1 , x2j ∈ X 2 , . . . , xnk ∈ X n . This is the prob- Similarly, one can also define information for the condi-
tional distribution. It is clear that if all the random vari-
ability that events x1i , x2j , . . . , xnk occur simultaneously.
ables X 1 , X 2 , . . . , X n are independent from each other,
We can also define the marginal distribution
we have
P (x1i , x2k , . . . , xj ) IX 1 ,X 2 ,...,X n = IX 1 + IX 2 + · · · + IX n . (2.6)

= P (x1i , x2k , . . . , xj , xj+1 n
p , . . . , xq ),
X j+1 ,X j+2 ,...,X n 2.3 Correlation for finite distributions
(2.1)
For two random variables X and Y, the correlation
where the summation is over all possible elements between them is defined as
in X j+1 , X j+2 , . . . , X n . This is the probability that
events x1i , x2k , . . . , xj occur with no restrictions on other C(X , Y) = IX ,Y − IX − IY . (2.7)
sets X j+1 , X j+2 , . . . , X n . The conditional distribution
is defined as It is clear that we have C(X , Y) = 0 if two ran-
dom variables X and Y are independent. This defini-
1 2 j tion can be generalized to group correlations. Suppose
Pxj+1 ,...,xn (xi , xk , . . . , x )
p q we have groups of random variables, X 1 , X 2 , . . . , X n ;
P (x1i , x2k , . . . , xj , xj+1 n Y 1 , Y 2 , . . . , Y m ; . . .; Z 1 , Z 2 , . . . , Z  , the correlation
p , . . . , xq )
= , (2.2) between these groups is
P (xj+1 n
p , . . . , xq )

C(X 1 , . . . , X n ; Y 1 , . . . , Y m ; . . . ; Z 1 , . . . , Z  )
which is the probability that events x1i , x2k , . . . , xj occur = IX 1 ,...,X n ;Y 1 ,...,Y m ;...;Z 1 ,...,Z 
while other variables are fixed at xj+1 n
p , . . . , xq . −IX 1 ,...,X n − IY 1 ,...,Y m − . . . − IZ 1 ,...,Z  .
1 2 n
For any function f (xi , xj , . . . , xk ) defined on sets
1 2 n
(2.8)
X , X , . . . , X , its expectation is defined as
 A special case of this group correlation is
f  = P (x1i , x2j , . . . , xnk )f (x1i , x2j , . . . , xnk )
X 1 ,X 2 ,...,X n C(X 1 , X 2 , . . . , X n ) = IX 1 ,X 2 ,...,X n
(2.3) −IX 1 − IX 2 − . . . − IX n . (2.9)

where the summation is over all possible values in sets 2.4 Generalization and further properties of
X 1 , X 2 , . . . , X n . Two variables X 1 and X 2 are indepen- correlation
dent if the joint distribution P (x1i , x2j ) = P (x1i )P (x2j ).
We shall now generalize the definition of correlation to
2.2 Information for finite distributions joint probability distributions over arbitrary sets of any
cardinality. To do this, we consider the refinement of a
For a single random variable X with distribution P (xi ), finite distribution. Consider a random variable X con-
its information IX is defined as sisting of finite number of events {x1 , x2 , . . . , xn }. It is
 possible that the event xi is actually the disjunction of
IX = P (xi ) ln P (xi ) = ln P (xi ) . (2.4) several exclusive events {x̃i,1 , x̃i,2 , . . . , x̃i,μ }. The distri-
i bution P  (x̃i,ν ) is called a refinement of the distribution
P (xi )
This is just the negative of Shannon’s entropy. If X has 
m different values, the maximum of IX is zero and the P (xi ) = P  (x̃i,ν ), (2.10)
minimum of IX is − ln m. The former corresponds to ν

123
Eur. Phys. J. H (2021)46:7 Page 9 of 20 7

where the summation is over all possible values of ν The previous definition of information is a special case
for a given xi . This can easily be generalized to multi- where all values of {ai }, {bj }, {ck } in the informa-
ple variables. For a distribution P (xi , yj ) of two random tion measure are unity. Interestingly, the correlation
variables X and Y and its refinement P (x̃i,μ , ỹj,ν ), there C(X , Y, Z) is independent of information measure.
exist two correlations C(X , Y) and C  (X , Y), respec- The advantage of introducing information measure is
tively. There is an interesting and important relation that we can now generalize information for continuous
between these two corrections variables. For example, for a continuous variable, X ,
with a probability distribution P (x), we can divide it
C  (X , Y) ≥ C(X , Y). (2.11) into finite sets Xi and use μi for the Lebesgue measure
of the set Xi . We then have
With this relation, we can generalize the correlation to
μ
 P (xi )
any probability measure over continuous variables. IX = P (xi ) ln . (2.17)
For simplicity, we consider two continuous random μi
i
variables X and Y and a probability measure M (X , Y)
over their Cartesian product. We can divide X into where P (xi ) is the probability over the set Xi . We can
finite subsets Xi and Y into finite subsets Yj . This further divide and refine the sets Xi and define informa-
naturally leads to a probability distribution P1 (Xi , Yj ), tion correspondingly. These information form a series
which can be obtained by integration of M (X , Y) over which has an upper limit. We define this upper limit
these subsets. With P1 (Xi , Yj ), we can compute the cor- as the information for X with probability distribution
relation C1 (X , Y) between X and Y. By further divid- P (x)
ing the subsets Xi and Yj , we can have another corre-
lation C2 (X , Y). By repeating the process, we have a μ
IX = lim IX . (2.18)
sequence of correlations μ

C1 (X , Y) ≤ C2 (X , Y) ≤ · · · ≤ Cn (X , Y). (2.12)
3 Quantum mechanics
As a result, the correlation between two continuous ran-
dom variables X and Y is defined as Quantum mechanics has two basic ingredients: (1) the
states of a quantum system are vectors in a Hilbert
C(X , Y) = lim Cn (X , Y), (2.13) space; (2) the time evolution of an isolated quantum
n→∞
system is given by a linear wave equation. One crucial
where n → ∞ means that the division becomes finer question is whether we need more to relate quantum
and finer, approaching the continuous limit. mechanics to our experimental and daily experience.
Suppose that f is a one-one map, U = f (X ), and g Many physicists represented by von Neumann think that
is a one-one map, V = g(Y). We have we need at least one more ingredient, Process 1, which
was mentioned at the beginning. Everett thinks that no
C(X , Y) = C(U, V). (2.14) more ingredient (or assumption) is needed.

This shows that the correlation is invariant under one- 3.1 Composite quantum systems
to-one transformation.
Consider a pair of quantum systems S1 and S2 . If their
2.5 Information for general distribution Hilbert spaces are H1 and H2 , respectively, the Hilbert
space of the composite system S = S1 + S2 is H1 ⊗ H2 .
For a random variable X with a finite set of values {xi }, If |ξj  is a complete orthonormal set for H1 and |ηj  for
we assign a positive number ai to each value xi . These H2 , a general state of S = S1 + S2 can be expressed as
ai are called information measure. If the probability 
distribution is P (xi ), its information relative to this |Ψ = cij |ξi , ηj  , (3.1)
information measure is defined as ij


IX = P (xi ) ln
P (xi )
= ln
P (xi )
. (2.15) where |ξi , ηj  is a shorthand for |ξi ⊗|ηj . The concepts
i
ai ai introduced in the last section can be applied here. Let Â
be a Hermitian operator on S1 with eigenfunctions |φi 
For multiple variables, say, X , Y, Z, with information and eigenvalues μi and B̂ be a Hermitian operator on
measures {ai }, {bj }, {ck }, respectively, and a joint S2 with eigenfunctions |ϕi  and eigenvalues νi . Then,
probability distribution P (xi , yj , zk ), their information
relative to these measures are Pij = | φi , ϕj |Ψ |2 (3.2)
 P (xi , yj , zk )
IXY Z = P (xi , yj , zk ) ln . (2.16) is a joint square-amplitude distribution of the quan-
ai bj ck tum state |Ψ over  and B̂. Note that Everett did not
ijk

123
7 Page 10 of 20 Eur. Phys. J. H (2021)46:7

use probability distribution here. The physical meaning By tracing out the subsystem S2 , we have the density
of square amplitude is discussed later. It has two matrix for the subsystem S1
marginal distributions
  1
 ρ1 = cij c∗kj |ξi  ξk | = |φηj  φηj | .
2
P (φi ) = | φi , ϕj |Ψ | , (3.3) Nη2j
ik j j
j
(3.12)
and
 Similarly, we can define ρ2 . The most important con-
P (ϕj ) = | φi , ϕj |Ψ |2 . (3.4) clusion of this section is that it is meaningless to ask
i the absolute state of a subsystem—one can only ask
the state relative to a given state of the remainder of
Correspondingly, there are two conditional distribu- the system. What Everett is discussing here is of course
tions entanglement: in a composite system where the subsys-
tems are entangled, the subsystems are described by den-
Pj (φi ) = Pij /P (ϕj ), (3.5) sity matrices not pure quantum states.

3.2 Information and correlation in quantum

mechanics
Pi (ϕj ) = Pij /P (φi ). (3.6)
Consider an operator Â, which has eigenstates |ξi  with
These distributions can be used to compute the marginal eigenvalues μi . The information of this operator in a
and conditional expectations of  or B̂. given state |ψ is defined as
A key concept introduced by Everett is relative state.
For a given state |η in S2 , there is a corresponding 
IA (ψ) = | ξi |ψ |2 ln | ξi |ψ |2 . (3.13)
relative state in S1 ,
i

|φη  = Nη η|Ψ = Nη cij |ξi  η|ηj  , (3.7)
The operator  has been assumed to be non-degenerate.
ij
If  is degenerate, that is, for eigenvalue μi , there are
where Nη is a normalization constant. For a given state multiple eigenstates |ξi,λ  (λ = 1, 2, . . . , mi ), its infor-
|η, its relative state |φη  is clearly unique and inde- mation is defined as
pendent of |ξi  and |ηi . The relative state |φη  can be   
  | ξi,λ |ψ |2
used to compute expectation of any operator  on S1 IA (ψ) = | ξi,λ |ψ | 2
ln λ
.
conditioned by the state |η in S2 . i
mi
λ
If |η is one of the basis states |ηj , we have
(3.14)

|φηj  = Nηj cij |ξi  . (3.8)
For convenience, we introduce projection operator
i

It is clear that Pi = |ξi,λ  ξi,λ | , (3.15)
λ
 1
|Ψ = |φηj  ⊗ |ηj  . (3.9)
j
Nηj with which we have

Two different relative states |φηj  and |φηk  are not | ξi,λ |ψ |2 = Tr(ρPi ), (3.16)
necessarily orthogonal λ


φηj |φηk  = Nηj Nηk c∗ij cik = 0. (3.10) where ρ = |ψ ψ|. The information of operator  now
i has a concise form

In a general state |Ψ of S, the subsystem S1 cannot  Tr(ρPi )

be described by a single state but by a mixture of states. IA (ψ) = Tr(ρPi ) ln . (3.17)
i
mi
We usually use the density matrix to describe this kind
of mixture. For the whole system, it is always in a pure
state |Ψ and its density matrix is We consider again the composite system S = S1 +S2 .
For the operator  that acts only on S1 and B̂ only on
ρ = |Ψ Ψ| . (3.11) S2 , we assume for simplicity that both of them have no

123
Eur. Phys. J. H (2021)46:7 Page 11 of 20 7

degeneracy. Their joint information is let us choose  = ρ1 and B̂ = ρ2 , and assume that there
is no degeneracy in eigenvalues vj . We have

IA,B (Ψ) = Tr(ρPA B A B
i Pj ) ln Tr(ρPi Pj ), (3.18) 
ij Iρ1 ,ρ2 (Ψ) = Iρ1 (Ψ) = Iρ2 (Ψ) = vj ln vj (3.25)
j
where PA B
i = |ξi  ξi |, Pj = |ηj  ηj |, and ρ = |Ψ Ψ|.
The marginal information for the operator  and the Consequently, we have
operator B̂ is 
Cρ1 ,ρ2 (Ψ) = − vj ln vj

IA (Ψ) = Tr(ρ1 PA A
i ) ln Tr(ρ1 Pi ) (3.19) j

i = −Tr(ρ1 ln ρ1 ) = −Tr(ρ2 ln ρ2 ). (3.26)

and This special correlation is called canonical correlation

 by Everett and it is, of course, exactly entanglement.
IB (Ψ) = Tr(ρ2 PB B
j ) ln Tr(ρ2 Pj ), (3.20) For convenience, we let C(Ψ) = Cρ1 ,ρ2 (Ψ). One may
j conjecture that, for any pair of operator  on S1 and
operator B̂ on S2 , the following inequality holds
where ρ1 = Tr2 (|Ψ Ψ|) and ρ2 = Tr1 (|Ψ Ψ|). We
define the correlation between  and B̂ as CA,B (Ψ) ≤ C(Ψ). (3.27)

CA,B (Ψ) = IA,B (Ψ) − IA (Ψ) − IB (Ψ) This conjecture has now been proved rigorously by Don-
 Tr(ρPA B
i Pj )
ald [17].
= Tr(ρPA P
i j
B
) ln . For operators x̂ and k̂ = p̂/, there is the Heisenberg
ij
Tr(ρ1 PA B
i )Tr(ρ2 Pj )
uncertainty relation
(3.21)
1
ΔxΔk ≥ . (3.28)
Without loss of generality, we assume that the dimen- 4
sion of the Hilbert space H1 is equal or bigger than H2 .
The reduced density matrix ρ2 for the subsystem S2 is a Everett conjectured that in terms of information this
Hermitian matrix and it can be diagonalized with non- relation can be written as
negative eigenvalues. Suppose that its eigenvectors are
|ϕj  with eigenvalues vj . If |ξi  is the complete basis of 1
Ix + Ik ≤ ln . (3.29)
the subsystem S1 , the relative state of |ϕj  for a general πe
state |Ψ is
This conjecture has been proved in Ref. [18], where
 Everett was not acknowledged.
|φj  = Nϕj |ξi  ξi , ϕj |Ψ . (3.22)
i
3.3 Measurement
One can show that |φj ’s are orthonormal to each
other, Everett regarded measurement as a natural process
in quantum mechanics and there is no fundamental
 distinction between “measuring apparatus” and other
φj |φk  = Nϕj Nϕk ξi |ξl  ξi , ϕj |Ψ Ψ|ξl , ϕk  physical systems. For Everett , a measurement is sim-
i,l ply a special interacting process between two quantum

= Nϕj Nϕk ξi , ϕj |Ψ Ψ|ξi , ϕk  subsystems, which results in the end that the property
i
of the measured subsystem is correlated to a quantity in
 the measuring subsystem. The measuring process has
= Nϕj Nϕk ϕj |ρ2 |ϕk  = δj,k . (3.23) two characteristics that distinguish it from other inter-
i acting processes.
Suppose that we have two subsystems S1 and S2 , ini-
Note the difference between here and Eq. (3.10), where tially in a product state |Ψ0  = |ψ0 , φ0  = |ψ0  ⊗ |φ0 .
|ηj ’s are not the eigenstates of ρ2 . As a result, we have The system will evolve dynamically under a Hamilto-
 nian Ĥ of the whole system. According to the analysis
|Ψ = vj |φj , ϕj  . (3.24) in the above subsections, at any moment, the overall
j state |Ψ(t) can be decomposed canonically as

This is called canonical representation of |Ψ by Everett |Ψ(t) = pj |ψj (t), φj (t) (3.30)
and it is of course just the Schmidt decomposition. Now j

123
7 Page 12 of 20 Eur. Phys. J. H (2021)46:7


where |ψj (t)’s and |φj (t)’s are eigenfunctions of two = −Iq (0) = − dq|φ(q)|2 ln |φ(q)|2 . (3.36)
operators Â(t) and B̂(t), respectively. The Hamiltonian
Ĥ is said to generate a measurement if the following
limits exist The correlation Cq,r (t) has increased to its maximum as
soon as t is not zero. The above analysis clearly shows
Â∞ = lim Â(t) , B̂∞ = lim B̂(t) (3.31) that the Hamiltonian ĤI generates a measurement of
t→∞ t→∞ q̂ for the system by r̂ of the apparatus. The general
case that the apparatus needle has no definite position
and they do not depend on initial conditions. initially is more complicated and was discussed in the
There is one requirement for a Hamiltonian Ĥ to gen- long thesis by Everett .
erate a measurement: Ĥ does not decrease the informa- In the above discussion, the apparatus initially has a
tion in the marginal distribution of Â. This means that definite position r0 . After measurement, the apparatus
if initially |Ψ0  = |ζ, φ0  where |ζ is an eigenfunction of no longer has a definite position. In fact, according to
Â, we should have at any time that |Ψ(t) = |ζ, φ(t). Eq. (3.35), the apparatus is in a superposition of states
The requirement is necessary for the repeatability of of different positions and the probability of its position
measurements: if a spin is measured to be up along the at r0 + qt is |φ(q)|2 . If this apparatus is of macroscopic
z direction, it should be still up when we measure it size, this means that its meter needle does not point
again along the z direction. to a definite position. We of course have never seen
In sum, a Hamiltonian Ĥ is said to generate a mea- this kind of measurement in any laboratory or similar
surement of  in S1 by B̂ in S2 if the following two con- phenomena in our daily life. To resolve this dilemma,
ditions are satisfied: (1) the correlation CA,B increases one possible way to assume that the mysterious collapse
of wave function (Process 1) during the measurement.
to its maximum with time; (2) Ĥ does not decrease the Everett found that one can resolve this dilemma within
marginal information of Â. the framework of quantum mechanics without additional
We now turn to a model proposed by von Neu- assumption.
mann [15] to illustrate the above definition of quan-
tum measurement. This model consists of a particle of
one coordinate q̂ and an apparatus of one coordinate r̂
(which may represents the position of a meter needle). 4 Observation
The interaction between them is very strong so that
we neglect all the kinetic energies. This means that the Observers are introduced as purely physical systems
whole Hamiltonian is given by and are treated completely within the framework of
quantum mechanics. In other words, observers are sim-
∂ ply usual quantum systems. If this treatment is suc-
ĤI = −iq . (3.32)
∂r cessful, it should build a consistent picture between the
appearance of phenomena, i.e., the subjective experi-
If the initial condition is a product state ence of observers, and the usual probabilistic interpre-
  tation of quantum mechanics.
|Ψ0  = dqφ(q) |q drη(r) |r
 4.1 Formulation of the problem
= dqdrφ(q)η(r) |q, r , (3.33) One can regard an observer as an automatically func-
tioning machine that has sensors and the capacity to
it is straightforward to find the evolution of this state record or register past sensory data and machine con-
figurations. When an observer O has observed the event

α, it means that O has changed to a new state that
|Ψ(t) = dqdrφ(q)η(r − qt) |q, r . (3.34) depends on α. Observers are assumed to have memo-
ries; the subjective experience of an observer is related
Let us consider a special case η(r) = δ(r − r0 ), that to the contents of its memory. As a result, the quantum
is, the apparatus needle initially points to a definite state of an observer O should be written as
position r0 . In this case, we have O
|ψ[A,B,...,C]  (4.1)

|Ψ(t) = dqφ(q) |q, r0 + qt . (3.35)
where A, B, . . . , C represent memories in the order of
time. Sometimes [. . . , A, B, . . . , C] is used to indicate
It is clear that ĤI has kept the marginal information of the possible previous memories that are not relevant
q̂. Let us consider the correlation Cq,r (t). Initially, we for the current observations.
have Cq,r (t) = 0. At time t, we have Consider an observer O who wants to measure (or
observe) the property  of a system S. The eigenfunc-
Cq,r (t) = Iq,r (t) − Iq (t) − Ir (t) tions of  are |φj ’s. Initially, the system S is in one

123
Eur. Phys. J. H (2021)46:7 Page 13 of 20 7

of the eigenfunctions |φj ’s of  and the observer is in forms the total state according to
O
state |ψ[...] . A good observation is defined as the one
that results in transforming |ψ S1 , ψ S2 , . . . , ψ Sn ; ψ[...]
O


S1
→ aj |φj , ψ S2 , . . . , ψ Sn ; ψ[...,α
O
j]
, (4.5)
|ψ S+O  = |φj  ⊗ |ψ[...]
O O
 = |φj ; ψ[...]  (4.2) j

to where |ψ S1  , |ψ S2  , . . . , |ψ Sn  are the initial quantum

states for systems S1 , S2 , . . . , Sn , respectively, and aj =
|ψ̃ S+O  = |φj  ⊗ |ψ[...,α
O
j]
O
 = |φj ; ψ[...,α j]
 (4.3) φSj 1 |ψ S1 .
Rule 2 Rule 1 may be applied separately to each
The semicolon is used here and will be used to delimit element of a superposition of total system states, the
the system state and the observer state. The require- results are superposed to obtain the final total state.
ment that the system state |φj  is unchanged is neces- S
Thus, a determination of B̂, with eigenfunctions |φk  2 ,
sary if you want the observation is repeatable. It is clear on S2 by the observer O transforms the total state
that observation is just quantum measurement (intro-
duced in the last section) with memories. 
aj |φSj 1 , ψ S2 , . . . , ψ Sn ; ψ[...,α
O
j]
 (4.6)
j
4.2 Deductions
to
If the system 
initially is in a general quantum state 
j aj |φj , the final total state after a aj bk |φSj 1 , φSk 2 , ψ S3 , . . . , ψ Sn ; ψ[...,α
described by O
j ,βk ]
 , (4.7)
good observation is jk


|ψ̃ S+O  = O
aj |φj ; ψ[...,α j]
. (4.4) where bk = φSk 2 |ψ S2 . These two rules follow directly
j from the superposition principle and are consistent with
the general framework of quantum mechanics.
This follows directly from the superposition principle Consider again the simple case where there is one
and is consistent with the general framework of quan- system and one observer. The observation results in Eq.
tum mechanics. Two features stand out in the above (4.4). If one repeats this observation, according to Rule
equations. (1) There is entanglement between the sys- 2, the total state becomes
tem and the observer and, as a result, neither of them 
O
has its independent state. (2) The result seems to con- aj |φj ; ψ[...,α j ,αj ]
. (4.8)
tradict our daily experience. On the one hand, the final j
states are superposition of many different states, each
of which corresponds to a definite observation outcome; Each superposition element in the above now describes
on the other hand, there is only one outcome in our that the observer has obtained the same result for both
daily experience. observations. That is, in each world, the observation is
Here comes Everett’s genius. Everett thinks that each repeatable. This is consistent with our experience.
superposition element in Eq. (4.4) represents a “world” Let us go one step further by considering many dif-
and the observer observes different outcomes in differ- ferent systems which are initially in the same state
ent “worlds”. Since the quantum dynamical evolution

is linear, which respects the superposition, each world |ψ S1  = |ψ S2  = · · · = |ψ Sn  = aj |φj  . (4.9)
evolves on its own and in each world the observer expe- j
riences only one definite outcome. This is in accor-
dance with our daily experience; at the same time, no
Therefore, the initial state of the total system is
additional assumption, such as the collapse of wave
function, is needed. This is the so-called many-worlds
interpretation. However, Everett himself never called |ψ0S1 +S2 +···+Sn +O  = |ψ S1 , ψ S2 , . . . , ψ Sn ; ψ[...]
O
.
each superposition element “world”; “many-worlds” (4.10)
was coined by de Witt in 1970s [1]. Everett called it
branch. The measurement is performed on the systems in the
The above observation should be the same even in order S1 , S2 , . . . , Sn . After the first measurement on S1 ,
the presence of other systems which do not interact we have
with the observer O. We thus have the general rules of

observation. |ψ1S1 +S2 +···+Sn +O  = aj |φ1j , ψ S2 , . . . , ψ Sn ; ψ[...,α
O
1] .
j
Rule 1 The observation of a quantity Â, with eigen- j
S
functions |φj 1 , in a system S1 by the observer O, trans- (4.11)

123
7 Page 14 of 20 Eur. Phys. J. H (2021)46:7

The total state after the second measurement is φ̃|φ̃ = 1 implies that
q

|ψ2S1 +S2 +···+Sn +O  |ã|2 = |aj |2 (4.17)

= aj ak |φ1j , φ2k , ψ S3 , . . . , ψ Sn ; ψ[...,α
O
1 ,α2 ]  . j=p
j k
j
(4.12) and


 q q

After r ≤ n measurements have taken place, we have M( |aj |2 ) = M(|aj |). (4.18)
j=p j=p

|ψrS1 +S2 +···+Sn +O  = aj ak . . . a
The only choice is the square amplitude measure,
j 2
M(|a
 j |) = c|aj | , where c can be fixed by requiring
|φ1j , φ2k , . . . , φr ; ψ Sr+1 , . . . , ψ Sn ; ψ[...,α
O
1 ,α2 ,...,αr ]  .
j M(|aj |) = 1.
j k 

(4.13) This square amplitude measure M is of probability

nature. Everett discussed this for general cases. I’ll use
a simple case to illustrate. Let us consider a simple sys-
Each of the superposition elements, which is one tem of spin-1/2. Suppose that there are n copies of them
of the many worlds, describes an observer which and their states are the same
has observed an apparently random sequence of defi-
nite results represented by [. . . , αj1 , αk2 , . . . , αr ]. If one 1
|ψ S1  = |ψ S2  = · · · = |ψ Sn  = √ (|u + |d),
repeats the measurement on the system S m (m < 2
r), the observer would get a memory sequence of (4.19)
[. . . , αj1 , . . . , αkm , . . . , αr , αkm ]. In each world, the observer
feels the “collapse” of wave function. where |u is for spin up and |d is for spin down. We
To make sense of the coefficients aj ak . . . a before make observations of the spins of σ̂ z . If the spin up |u
each superposition element, we need to assign a measure is registered as 0 and the spin down |d is registered as
to them. We first consider a superposition state 1, we have after measuring all the spins
 1 
|ψ = aj |φj  . (4.14) |ψnfinal  = √
j 2n
|u , d , . . . , dr ur+1 , . . . , un ; ψ[...,0
1 2 O
1 ,12 ,...,1r 0r+1 ,...,0n ]  ,

To assign a measure, we first need that each element (4.20)

is normalized φj |φj  = 1. For each element |φj , the
assigned measure is M(aj ), which is a non-negative where the summation is over all possible sequences of
function. We could change |φj  to eiθ |φj  and aj to 0’s and 1’s of length n. Most of times, we only care
e−iθ aj , then the measure M(aj ) assigned to this ele- about how many of these spins are up and how many of
ment becomes M(e−iθ aj ). However, physically noth- them down. So, we group these states according to how
ing has been changed. Therefore, we need M(aj ) = many spins are up
M(e−iθ aj ). For this to be true, it is clear that M(aj ) 
should depend only on the amplitude of aj , that is, n! 1 
M(|aj |). n
|φ̃m  = √ |u . . . d .
m!(n − m)!2 2n m ups
In reality, due to the accuracy of the measurement
or other reasons, we often regard a group of states (4.21)
|φp  , |φp+1  , . . . , |φq  as the same, i.e.,
where |φ̃m  represents a state where m of the n spins
q
 are up. The measure M for the group of m up spins is
ã |φ̃ = aj |φj  , (4.15)
j=p
n!
Mm = , (4.22)
m!(n − m)!2n

where φ̃|φ̃ = 1 is normalized. We require the additiv- which is exactly the probability that one observes the
ity for the measure, that is, state of spin up m times when making n repeated same
measurements. What happens here is that, after n mea-
q
 surements, it splits into 2n branches of worlds, each of
M(|ã|) = M(|aj |). (4.16) which is equally probable and has a different sequence
j=p of 0’s and 1’s registered. The chance being in a world

123
Eur. Phys. J. H (2021)46:7 Page 15 of 20 7

where there are m up spins (or m 0s) is Mm . In gen- Everett even considered the situation where the two
eral, if the spin is in a state |φ = a |u + b |d, we still observers are allowed to communicate their observation
have 2n branches of worlds after n measurements but results. It does not lead to any self-contradiction and
the chance being in a world where there are m up spins contradiction to our daily experience.
(or m 0’s) is Let us now return to the paradox in Section I. The
observer A made the measurement; the observer B did
n! not make the measurement directly and he obtained his
Mm = |a|2m |b|2(n−m) . (4.23) result by reading A’s notebook. In this case, we have
m!(n − m)!  A B
j aj |φ j ; ψ[...,αj ] , ψ[...,αj,A ] , where the subscript indi-
The number of branches has nothing to do with the prob- cates that the result comes from A. In each world, the
ability measure |a|2 or |b|2 ; it depends on the observation two observers A and B always agree with each other and
outcomes. have nothing to argue about. The paradox is resolved.
The above results can be straightforwardly general- Case 2 Two observers measure separately two differ-
ized to the cases where different measurements are per- ent quantities, which are non-commuting, in the same
formed on different systems and different measurements system.
are performed on the same system. The same initial state |Ψ0  and the same observation
by O1 . Then the observer O2 measures the quantity B̂,
4.3 Several observers which does not commute with Â. We apply Rule 2 to
|Ψ1  and obtain
It was pointed out at the beginning that the assumption
of Process 1 (or the collapse of wave function) would 
O1 O2
lead to self-contradiction when there are more than one |Ψ2  = aj ϕk |φj  |ϕj ; ψ[...,α j]
, ψ[...,β k]
 , (4.27)
j,k
observers. There is no such contradiction in the many-
worlds theory. Let us consider the situation where there
are multiple observers. Three different cases are to be where |ϕk  is the eigenstate of B̂. In this case, the sec-
considered. ond observation leads to further splitting. If  has NA
Case 1 Two observers observe the same quantity in the eigenstates and B̂ has NB eigenstates, then there are
same system. NA NB different worlds in total, which are represented
Observers O1 and O2 are to observe the quantity  by the terms on the left hand side of the above equa-
for the system S that is in the following state tion. The measure M of the coefficients aj ϕk |φj  gives
the probability of getting into one of the worlds if the
 observations are repeated on the same system in the
|ψ S  = aj |φj  , (4.24) same state.
j
Case 3 Two observers O1 and O2 measure two entan-
gled systems S1 and S2 : O1 measures  in S1 and O2
where |φj  is an eigenstate of Â. The observer O1 makes measures B̂ in S2 .
the first observation; we apply Rule 1 to the initial For simplicity, we assume that the initial state of the
O1 O2
state |Ψ0  = |ψ S ; ψ[...] , ψ[...]  and obtain composite system of S1 and S2 is entangled

O1 O2 N
|Ψ1  = aj |φj ; ψ[...,α , ψ[...] . (4.25) 
j]
j |ψ S1 +S2  = aj |φj , ϕj  , (4.28)
j=1
The observer O2 makes the second observation; we
apply Rule 2 and obtain where N ≤ NA , NB . There is no interaction between
 S1 and S2 during the following observations. The total
O1 O2 initial state is
|Ψ2  = aj φk |φj  |φj ; ψ[...,α j]
, ψ[...,α k]

j,k
  O1
|Ψ0  = |ψ S1 +S2 ; ψ[...] O2
, ψ[...] . (4.29)
O1 O2
= aj |φj ; ψ[...,α j]
, ψ[...,α j]
. (4.26)
j

After O1 observes  in S1 , the total state becomes

This shows that the first observation by O1 leads to
splitting of different branches of worlds, and the sec- N
ond observation by O2 of the same quantity causes no 

O1 O2
splitting and furthermore O2 observes the same result |Ψ1  = aj |φj , ϕj ; ψ[...,α j]
, ψ[...] . (4.30)
j=1
as O1 . It is in accordance with our daily experience:
two observers measuring the same quantity on a given
system always obtains the same result. This result can There are now N different branches of worlds. The
clearly be generalized to any number of observers. observer O2 then observes B̂ in S2 and transforms the

123
7 Page 16 of 20 Eur. Phys. J. H (2021)46:7

total state to coordinate between the proton and the electron. The
former describes the motion of the hydrogen atom as a
N
 whole in space and time; the latter is usually a bound
 O1 O2
|Ψ2  = aj |φj , ϕj ; ψ[...,α j]
, ψ[...,β j]
. (4.31) state that gives us the size and shape of the hydrogen
j=1 atom. The situation is similar for macroscopic object
that we see daily. For example, the wave function of a
No more splitting and there are still N different cannonball can be written roughly as
branches of worlds. In each branch, when O1 observes
b b
the result represented by |φj , O2 observes the result |ψc  = |g(X) ⊗ |ψX  = |g(X), ψX , (5.1)
represented by |ϕj . The observation results of O1 and
O2 are correlated. It is easy to check that if O2 observes where |g(X) is a Gaussian wave function well localized
 b
first and O1 second, the end state is still |Ψ2  . It is clear at the position X and |ψX  is the bound state giving us
that the observations of O1 and O2 do not influence the size and shape of a cannonball located at X. When
each other. Furthermore, if O1 repeats its measurement the centroid wave function |g(X) evolves over a long

of  in S1 , then the total state |Ψ2  is turned into period of time, it can spread to occupy a large region of
space. Therefore, in the general case, the cannonball is
N
 not necessarily well localized and its state is given by
 O1 O2
|Ψ2  = aj |φj , ϕj ; ψ[...,α j ,αj ]
, ψ[...,β j]
 . (4.32) 
j=1 |ψc  = b
|a(X) ⊗ |ψX  dX, (5.2)

We would have the same total state if O1 had observed

 in S1 twice in a row before O2 observed B̂ in S2 . This where |a(X) is any smooth and normalizable function.
shows that in every world O1 would not know whether In this general state, there is an entanglement between
O2 has made an observation on S2 or not if there is no the centroid position X and the rest of the coordinates
direct communication between them. In other words, of a cannonball.
the entanglement in the state (4.28) cannot be used for However, in most of systems, we can separate the
communication. centroid coordinates from the coordinates for the rela-
tive motion in the Schrödinger equation; as a result,
b
|ψX  does not depend on X explicitly. In this case, we
b
can take |ψX  outside of the above integral and have
5 Supplementary topics |ψc  = |a ⊗ |ψ b . When |a is not well localized, the


state |Ψc  represents a kind of “smeared out” cannon-

We have presented an abstract treatment of measure- ball. In contrast, we only see cannonballs with definite
ment and observation completely within the frame- positions and momenta in our daily life. This dilemma
work of quantum mechanics, which are in correspon- can be resolved by noticing that cannonballs are never
dence with our experience. Upon observation and mea- truly isolated and they are constantly be observed or
surement, there is splitting into different worlds: in measured by photons and other objects. We assume that
each world to an observer there appears a collapse of someone magically set up a cannonball in a superposi-
wave function (or Process 1); however, with all the tion state of two well separated and localized Gaussians,
worlds combined, the evolution is always unitary. This that is,
approach has at least three advantages: (1) it is logically
self-consistent; (2) it does not involve any additional 1  
assumption, for example, the collapse of wave function; |ψc  = √ |g(X1 ) + |g(X2 ) ⊗ |ψ b  . (5.3)
2
(3) it is completely quantum mechanical with no use of
classical concepts. The initial state of this cannonball and its observer is
|Ψc+O
0
O
 = |ψc ; ψ[...] O
, where |ψ[...]  is the state of an
5.1 Macroscopic objects and classical mechanics observer, which can be photons or other physical objects
that can distinguish the difference between the positions
In the many-worlds theory, there is no more divide X1 and X2 . After the interaction between the cannon-
between quantum world and classical world. Macro- ball and the observer (which happens in a very short
scopic objects are also described by wave functions. time), a splitting of worlds happens and the total state
However, we do experience in our daily life a classi- becomes
cal world where macroscopic objects have definite posi-
tions and momenta, moving around according to clas- 1 
sical mechanics. Below is a rough explanation how the |Ψc+O
1
O
 = √ |g(X1 ); ψ[...,X 1]

2
classical world emerges from quantum mechanics with O

no detailed proof. + |g(X2 ); ψ[...,X 2]
 ⊗ |ψ b  . (5.4)
Let us first consider a simple case, the hydrogen
atom. Its wave function is essentially a product of a cen- In one world, the observer finds the cannonball well
troid wave function and a wave function for the relative localized at X1 ; in the other world, the observer finds

123
Eur. Phys. J. H (2021)46:7 Page 17 of 20 7

the cannonball well localized at X2 . This is why we do macroscopic distinguishable states of a Geiger counter,
not observe “smeared out” cannonballs. The cannonball discharged or undischarged. As a result, the state of a
and other similar macroscopic objects with well localized Geiger counter can be simply written as
wave functions will move approximately according to the
classical mechanics. After a certain period of time, the |ψ G  = a1 |ψ[U ]  + a2 |ψ[D]  . (5.8)
wave packet will spread out so much to cause another
splitting of worlds. The detailed account of how long a Consider a particle which is detectable by a Geiger
well-localized wave packet will follow the classical tra- counter. The total initial state is
jectory is given by Ehrenfest time [19]. Everett did not
use the concept of Ehrenfest time in his thesis. |Ψp+G
0  = |ψ p ; ψ[U ]  , (5.9)

5.2 Amplification processes where |ψ p  is the state of the particle. If the wave func-
tion |ψ p  is not well localized so that it has a part |ψop 
In our abstract discussion of measuring process in
the previous sections, we have simplified the coupling outside of the Geiger counter and the other part |ψip 
between the system and the observer (or the appara- inside the Geiger counter, i.e., |ψ p  = a |ψop  + b |ψip .
tus). In reality, there is a chain of intervening systems After the particle encounters the Geiger counter, the
linking a microscopic system to a macroscopic appara- total state is transformed to
tus. Each system in the chain of intervening systems is
correlated to its predecessor, resulting an amplification |Ψp+G
1  = a |ψ̃op ; ψ[U ]  + b |ψ̃ip ; ψ[D]  . (5.10)
of effects from the microscopic system to a macroscopic
apparatus. We have a splitting into two worlds: in one world the
We use Geiger counter as an example to illustrate Geiger counter is discharged, and in the other one the
this amplification process. A Geiger counter contains a counter is undischarged. This is similar to the splitting
large number of gas atoms that are placed in a strong in Eq. (5.4).
electric field. The atoms are metastable against ioniza-
tion. More importantly, the product of ionizing one gas 5.3 Reversibility and irreversibility
atom can cause ionization of more atoms in a cascad-
ing process. This chain reaction correlates large number In the usual treatment of quantum mechanics, there are
of gas atoms: either very few or very many of the gas both Process 1 (the collapse of wave function) and Pro-
atoms are ionized at a given time. cess 2 (unitary evolution). It is obvious that Process 1 is
To put the above discussion in a mathematical form, irreversible and Process 2 is reversible. This difference
we write the state of a Geiger counter in terms of its can be quantified by introducing another information
individual gas atoms
 Iρ = Tr(ρ̂ ln ρ̂), (5.11)
|ψ G  = aij...k |φi , φj , . . . , φk  , (5.5)
ij...k where ρ̂ is a density matrix of a quantum system. If
the system changes according to Process 2, we have
where |φi , φj , . . . , φk  represents a state where the first ρ̂ = U ρ̂U † , which does not change Iρ since
atom is in the ith state, the second atom is in the jth
state, . . . , the nth atom is in the kth state. The super-    
Iρ = Tr ρ̂ ln ρ̂ = Tr U ρ̂U † ln(U ρ̂U † )
position terms on the right-hand side of the above equa-    
tions describe either large number of ionized atoms or = Tr U ρ̂ ln ρ̂U † = Tr ρ̂ ln ρ̂ = Iρ . (5.12)
few ionized atoms. Due to the chain ionization, there
are almost no terms for medium-sized number of ion- For Process 1, we consider a simple case where the sys-
ized atoms. By choosing a medium-sized number, we tem is in a pure quantum state, that is, ρ̂ = |ψ ψ|.
can place these superposition terms in two groups The measurement is for the quantity Â, whose eigen-
 functions are |φj . After the measurement (Process 1),

a1 |ψ[U ]  = aij...k |φi , φj , . . . , φk  (5.6) there is a probability of | φj |ψ |2 of the measured result
ij...k is |φj . This means that the density matrix becomes

and ρ̂ = | φj |ψ |2 |φj  φj | , (5.13)
 j

a2 |ψ[D]  = aij...k |φi , φj , . . . , φk  . (5.7)
ij...k
and
The primed summation is over all terms with few num-   
ber of ionized atoms, and the double primed summa- Iρ = | φj |ψ |2 ln | φj |ψ |2 ≤ Iρ = 0.
tion is over all terms with very large number of ion- j
ized atoms. |ψ[D]  and |ψ[U ]  represent, respectively, two (5.14)

123
7 Page 18 of 20 Eur. Phys. J. H (2021)46:7

So, Process 1 decreases the information Iρ but never we can predict approximately the position of a particle
increase it. One can prove rigorously that this is true with its current approximate position. If Process 1 were
for any ρ̂ not just for pure states. true, after the measurement, the particle would be in
In the many-worlds theory, even though only Pro- an eigenstate of position and its momentum would be
cess 2 is recognized, an observer can still feel similar too uncertain to make any meaningful prediction for its
irreversibility on the subjective level. When an obser- future position. This contradicts well-established exper-
vation is performed, it leads to a superposition of many imental facts. Everett has offered more detailed analysis
different worlds. From this time forward, since the uni- along this line and pointed out the inadequacy of Pro-
tary evolution is linear, these worlds are parallel, evolve cess 1 in approximate measurement.
independently, and no longer influence each other. The
observer in each world has only information in his 5.5 Discussion of a spin measurement example
world, knowing nothing about other parallel worlds. As
a result, for an observer in a given world, this process is Consider the z component of a spin-1/2 with the Stern–
also irreversible since he cannot in principle get to know Gerlach setup. In this measurement, a particle of spin-
the state before the measurement based on the informa- 1/2 passes through a magnetic field that is inhomoge-
tion available in his world. This irreversibility implies neous along the z direction. The measurement is essen-
that there is a fundamental limit on the knowledge of tially to couple the spin and the orbital of the same
the entire universe. particle. For simplicity, we keep only the coupling part
However, the irreversibility discussed here appears of the Hamiltonian and approximate only the constant
not related to the second law of thermodynamics, which and linear part of the inhomogeneous field
reflects a different kind of irreversibility. There are two
ways to see the difference. First, the former is of quan- ĤI ≈ μσ̂z (B0 + zB1 ), (5.15)
tum nature while the latter is also valid in classical
systems. When one mixes two piles of sand of differ- where μ is the magnetic moment of the particle. The
ent colors, there is clearly no quantum process involved initial state of the particle is assumed to be
but the mixing is irreversible as dictated by the second
law. Second, when the universe keeps splitting into more  
and more worlds, more and more systems get entangled |Ψ0  = φ0 (z) c1 |u + c2 |d , (5.16)
together. In this sense, the irreversibility associated with
the world splitting is for an open system, whereas the where φ0 (z) describes a wave packet along the z direc-
second law of thermodynamics is for a closed system. tion and |u (|d) is the eigenfunction of σ̂z with eigen-
These are definitely not the final words on the relation value 1 (-1). One can solve the Schrödinger equation.
between the second law of thermodynamics and the split- If Δt is the time that the particle takes to traverse the
ting of worlds, which warrants further study. In fact, field, we have
Tegmark discussed the second law of thermodynamics 
within the framework of many-worlds theory using a tri- |Ψ(Δt) = φ0 (z) c1 e−iμ(B0 +zB1 )Δt/) |u
partite partition of the universe [8]. My personal view is
that to discuss the second law within quantum mechan- +c2 eiμ(B0 +zB1 )Δt/) |d . (5.17)
ics one has to follow von Neumann, who defined quan-
tum entropy for pure states (different from the well-
known von Neumann entropy) and proved quantum H- This is an entangled state between the spin and the
theorm [20–22]. orbital. The wave function has split into two: one
with momentum μH1 Δt and the other with momen-
5.4 Approximate measurement tum −μH1 Δt. With long enough flying time, these two
parts will become well separated in space: the upper
In many situations, we have only approximate mea- wave packet for the spin up state |u and the lower
surements, where the apparatus or observer interacts wave packet for the spin down state |d. The measuring
weakly with the system and for a finite time. It is “apparatus” here is the orbital degree of freedom of the
hard to understand these cases with Process 1, which particle, which by all means is microscopic.
requires that all measurements result in a precise pro- In many situations, one can regard states for com-
jection to an eigenstate of a measured quantity. The posite systems such as Eq. (5.17) as a non-interfering
position measurement appears to be the best example mixture of states by ignoring phases in superposition
to illustrate this difficulty. elements. For example, it is correct when calculating
In any situation, we do not know the precise position marginal expectations for subsystems. For the state
of any particle. One possible way to understand this (5.17), it is alright to regard it as a mixture if one cares
with Process 1 is that the measurement indeed results either the spin or the orbital but not both. The phase
in a precise position but the observer has only impre- relations between different superposition elements are
cise information. This view is clearly wrong. In prac- important. For this Stern–Gerlach system, it is possi-
tice, for example, when tracking high-energy particles ble to recombine the two terms in Eq. (5.17) in another
with cloud chambers, we can measure the approximate inhomogeneous magnetic field and restore the original
positions of a particle successively. This means that state in Eq. (5.16) [23]. For this to happen, one cannot
disregard the phases.

123
Eur. Phys. J. H (2021)46:7 Page 19 of 20 7

6 Discussion There are a number of interpretations of quantum

mechanics, most of which are equivalent in the sense that
We have shown that our theory (the many-worlds the- they agree with all the physical experiments. To decide
among them, we must go beyond experiments and discuss
ory) can be put in a satisfactory correspondence with
the fundamental nature and purpose of physical theories.
experience, and gives us a complete conceptual model
Every theory has two separate parts, the formal part and
of the universe with more than one observers. In this the interpretive part. The formal part consists of a purely
theory, the wave function is a basic description of phys- logico-mathematical structure that consists of a collection
ical systems, including observers, and the probabilis- of symbols and rules for their manipulation. The interpre-
tic assertion of quantum mechanics can be deduced tive part is a set of association rules that relate the formal
from this theory as subjective appearances to the symbols with the experienced world. There can be many
observers. This theory constitutes an objective frame- different theories which are logical consistent and correct in
work in which puzzling subjects, such as classical phe- explaining the perceived world. In this case, further criteria
nomena, the measuring process, the inter-relationship such as usefulness, simplicity, comprehensiveness, pictora-
of several observers, reversibility and irreversibility, can bility, etc., must be used to select the theory or the theo-
be investigated in a logically consistent manner. ries. In particular, simplicity refers to conceptual simplicity
In light of his new theory, Everett discussed in length not ease in use. It is harmful to the progress of physics that
other interpretations of quantum mechanics existing at a physical theory should contain no elements which do not
his time. They are correspond directly to what we observe.

a. The “popular” interpretation. The wave function

|ψ changes continuously and deterministically with
a wave equation when the system is isolated but
Part IV: The legacy of Everett’s theory
changes probabilistically and abruptly upon obser-
vation. Although Everett published his short thesis in Review of
b. The Copenhagen interpretation. The wave function Modern Physics, a well-known and respected journal, along
with Wheeler’s supporting article [4, 5], his theory of the uni-
|ψ is regarded as just a mathematical artifice which
versal wave function had received little attention for many
one uses to make statistical predictions. All state-
years [3, 24]. In 1962, Everett was invited by Podolsky to a
ments about microscopic phenomena are meaningful small workshop at Xavier University, where he lectured on
only within a classical experiment setup. his theory for the first time in public. However, this did not
c. The “hidden variable” interpretation. The wave mean that his theory was getting wide recognition; it only
function |ψ is not a complete description of a sys- showed that his work was not completely forgotten [3].
tem. There are additional hidden parameters in the Everett’s theory began to be noticed widely in the
correct and complete theory that is to be developed physics community only after DeWitt started to promote
in the future. The probability in quantum mechan- it as the many-worlds theory around 1970s with Graham’s
ics is the result of our ignorance of these hidden help [1, 3, 24]. As a result, Everett’s theory is now widely
variables. known as the many-worlds theory. Only a limited number of
d. The stochastic process interpretation. In this the- specialists know it as the theory of the universal wave func-
ory, physical systems are undergoing probabilistic tion or the “relative state” formulation of quantum mechan-
changes at all times. The discontinuous and proba- ics. Now Everett’s theory has to be mentioned in all seri-
bilistic “quantum jump” is not the result of obser- ous discussion of the subject known as the interpretation of
vation and measurement but is fundamental to the quantum mechanics [25].
systems themselves. To celebrate the 50th anniversary of its publication in
2007, the influential magazine, Nature, put the many-worlds
theory on its July cover; and BBC produced a special pro-
Acknowledgements This work is supported by The gram called “Parallel worlds, Parallel lives”. This theory is
National Key R&D Program of China (Grants Nos. now extremely popular among non-specialists.
2017YFA0303302, 2018YFA0305602), National Natural Sci- Among physicists, the many-worlds theory is still a
ence Foundation of China (Grant No. 11921005), and Shang- minority view but it is gaining momentum. Hawking [26]
hai Municipal Science and Technology Major Project (Grant and Gell-Mann [27] were famous names among its early
No. 2019SHZDZX01). supporters. Current influential advocates include David
Deutsch [6], Max Tegmark [8], and Sean Carroll [28].

7 Appendices References
In Everett’s long thesis, there are two appendices. In the first 1. B.S. DeWitt, N. Graham (eds.), The Many-Worlds
one, Everett offered detailed proofs for many mathematical Interpretation of Quantum Mechanics (Princeton Uni-
relations in the main text. In the second one, he offered his versity Press, Princeton, 1973)
view on theoretical physics in general. Here is the summary 2. J.A. Barrett, P. Byrne (eds.), The Everett Interpre-
of the second appendix. tation of Quantum Mechanics: Collected Works 1955–

123
7 Page 20 of 20 Eur. Phys. J. H (2021)46:7

1980 with Commentary (Princeton University Press, 18. I. Bialynicki-Birula, J. Mycielski, Commun. Math. Phys.
Princeton, 2012) 44, 129 (1975)
3. P. Byrne, The Many Worlds of Hugh Everett III (Oxford 19. Y. Zhao, B. Wu, Sci. China Phys. Mech. Astron. 62,
University Press, New York, 2010) 997011 (2019)
4. H. Everett, Rev. Mod. Phys. 29, 454 (1957) 20. J. von Neumann, Zeitschrift für Physik 57, 30 (1929)
5. J.A. Wheeler, Rev. Mod. Phys. 29, 463 (1957) 21. J. von Neumann, Eur. Phys. J. H 35, 201 (2010)
6. D. Deutsch, The Beginning of Infinity (Penguin Books, 22. X. Han, B. Wu, Phys. Rev. E 91, 062106 (2015)
New York, 2011) 23. D. Bohm, Quantum Theory (Prentice-Hall, New York,
7. H. Kragh, Ann. Sci. 66, 529 (2009) 1951)
8. M. Tegmark, Phys. Rev. D 85, 123517 (2012) 24. S. Osnaghi, F. Freitas, Jr., Stud. Hist. Philos. Mod.
9. L. Zehnder, Zeitschrift für Instrumentenkunde 11, 275 Phys. 40, 97 (2009)
(1891) 25. M. Jammer, The Philosophy of Quantum Mechanics
10. L. Mach, Zeitschrift für Instrumentenkunde 12, 89 (Wiley-Interscience, New York, 1974)
(1892) 26. S.W. Hawking, Phys. Rev. D 13, 191 (1976)
11. P. Hariharan, Basics of Interferometry (Elsevier, Ams- 27. M. Gell-Mann, in Complexity, Entropy and the Physics
terdam, 2007) of Information, ed. by J.B. Hartle, W.H. Zurek
12. R.B. Griffiths, Phys. Lett. A 178, 17 (1993) (Addison-Wesley, Reading, 1990), pp. 425–459
13. A. Zeilinger, Am. J. Phys. 49, 882 (1981) 28. S. Carroll, Something Deeply Hidden (Dutton, New
14. K.P. Zetie, S.F. Adams, R.M. Tocknell, Phys. Educ. 35, York, 2019)
46 (2000)
15. J. von Neumann, Mathematical Foundations of Quan-
tum Mechanics (Princeton University Press, Cambridge,
1955)
16. J.S. Bell, Physics 1, 195 (1964)
17. http://people.bss.phy.cam.ac.uk/∼mjd1014/evconja.
html

123