See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/284283974

Disproof of Bell’s Theorem: Illuminating the

Illusion of Entanglement (Second Edition)

Chapter · December 2014

CITATIONS READS

7 66

1 author:

Joy Christian
University of Oxford
59 PUBLICATIONS   486 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Mishra-Christian Theory of Quantum Gravity: A New Path View project

My Challenge to All Adherents of Bell's Theorem View project

All content following this page was uploaded by Joy Christian on 21 November 2015.

The user has requested enhancement of the downloaded file.

Disproof of Bell’s Theorem
Illuminating the Illusion of Entanglement

Second Edition

Joy Christian

BrownWalker Press
Boca Raton
 Contents
Preface to Second Edition . . . . . . . . . . . . . . . . . . . xiii
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . xv
1 On the Local-Realistic Origins of Quantum Correlations 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Local Origins of the EPR-Bohm Correlations . 5
1.3 Local Origins of ALL Quantum Correlations . 12
1.4 The Raison D’être of Quantum Correlations . . 17
1.5 Local Causality and the Division Algebras . . . 19
1.6 Concluding Remarks . . . . . . . . . . . . . . . . . 20
Appendix 1: A 2-D Analogue of EPRB Correlation 21
2 Disproof of Bell’s Theorem by Clifford Algebra Valued
Local Variables 29
3 Can Bell’s Prescription for the Physical Reality be
Considered Complete? 39
4 Failure of Bell’s Theorem and the Local Causality of
the Entangled Photons 49
5 Further Consolidations of the Disproof 59
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 59
5.1.1 Exact, Deterministic, Locally Causal Model for
the EPR - Bohm Correlations . . . . . . . . . . . 60
5.1.2 Eight Essential Requirements Satisfied by the Model 61
5.2 Operational Adequacy of the Local Beables
An (µ), and a Theory of their Measurement . . 64
5.3 Local Causality of the Model at the Level of
Microstates . . . . . . . . . . . . . . . . . . . . . . . 69
5.3.1 Remote Parameter Independence in the Local Model 72
5.3.2 Remote Outcome Independence in the Local Model 74
5.4 Statistical Interpretation of the Entangled
Singlet State . . . . . . . . . . . . . . . . . . . . . . 76

ix
 5.5 Derivation of the Malus’s Law for Sequential
Spin Measurements . . . . . . . . . . . . . . . . . . 78
5.6 Concluding Remarks . . . . . . . . . . . . . . . . . 80
6 Disproofs of Bell, GHZ, and Hardy Type Theorems
and the Illusion of Entanglement 83
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 83
6.2 EPR Elements of Reality are Points of a Unit
2-Sphere, not 0-Sphere as Bell Assumed . . . . 87
6.3 Bell’s Theorem, its Variants, and their
Spin-offs are All Non-Starters At Best . . . . . 92
6.4 Topologically Correct Local-Realistic
Framework for the EPR-Type Correlations . . 96
6.5 Exact, Local, and Realistic Completions of
the GHZ-3, GHZ-4, and Hardy States . . . . . . 110
6.5.1 Exact Local-Realistic Completion of the Hardy State 110
6.5.2 Exact Local-Realistic Completion of the Four-Particle
GHZ State . . . . . . . . . . . . . . . . . . . . . 117
6.5.3 Exact Local-Realistic Completion of the Three-Particle
GHZ State . . . . . . . . . . . . . . . . . . . . . 124
6.6 Exact Local-Realistic Completion of Any
Arbitrary Entangled State is Always Possible . 129
6.7 Concluding Remarks . . . . . . . . . . . . . . . . . 134
7 What Sets the Upper Bound on Quantum Correlations?141
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 142
7.2 Completing Bell’s Incomplete Accounting . . . 145
7.3 What Would Alice Observe at Her Detector? . 154
7.4 Upper Bound is Set by Maximum Possible
Torsion in the Set of All Possible Outcomes . . 158
7.4.1 When Codomain is an Arbitrary Manifold . . . 159
7.4.2 When Codomain is a Parallelized 3-Sphere . . 162
7.4.3 When Codomain is a Parallelized 1-Sphere . . 168
7.4.4 When Codomain is a Parallelized 0-Sphere . . 168
7.4.5 When Codomain is a Parallelized 7-Sphere . . 169

x
 7.5 Concluding Remarks . . . . . . . . . . . . . . . . . 180
8 Restoring Local Causality and Objective Reality to
the Entangled Photons 187
Appendix 2: Error Propagation Within a 3-Sphere . 204
9 Whither All the Scope and Generality of Bell’s
Theorem? 209
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 209
9.2 An Exact Local-Deterministic Model for the
EPR-Bohm Correlation . . . . . . . . . . . . . . . 210
9.2.1 A Complete Specification of the Singlet State . . . 210
9.2.2 Construction of the Measurement Functions . . . . 213
9.2.3 A Crucial Lesson from Basic Statistics . . . . . . 217
9.2.4 How Errors Propagate in a Parallelized 3-Sphere . 219
9.2.5 Derivation of Pair Correlations Among the Points
of S 3 . . . . . . . . . . . . . . . . . . . . . . . . 228
9.2.6 Derivation of Upper Bound Exceeding the CHSH
Bound . . . . . . . . . . . . . . . . . . . . . . . 229
9.3 Physical, Mathematical, and Conceptual
Fallacies of Weatherall’s Model . . . . . . . . . 234
9.3.1 What is Wrong with Weatherall’s Measurement
Ansatz? . . . . . . . . . . . . . . . . . . . . . . 234
9.3.2 Why did We Lose the Strong Correlation for SO(3)? 236
9.4 Concluding Remarks . . . . . . . . . . . . . . . . . 237
Appendix 3: Local-Realistic Violations of CHSH
Inequality as well as Clauser-Horne Inequality 238
A.3.1 Local-Realistic Violations of the CHSH Inequality
Verified in an Event-by-Event Simulation . . . . . 238
A.3.2 Computer Simulation also Confirms Local-Realistic
Violations of the Clauser-Horne Inequality . . . . 252
A.3.3 Another Explicit Simulation of the Model . . . 253
A.3.4 Exploring the World Beyond the Quantum World . 257
A.3.5 Elegant, Powerful, and Succinct Calculation of the
Correlation . . . . . . . . . . . . . . . . . . . . . 259

xi
 A.3.6 What Can We Learn from Weatherall’s Analysis? . 260
10 Refutation of Richard Gill’s Fallacious Argument Against
my Disproof of Bell’s Theorem 263
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 263
10.2 A Fallacy of Misplaced Concreteness1 . . . . . . 271
10.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . 275
11 Refutation of Some Fallacious Arguments Against my
Disproof of Bell’s Theorem 279
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . 279
11.2 Orientation of the 3-Sphere is Not a
Convention But a Hidden Variable . . . . . . . . 281
11.3 Handedness of the Bivectors is Not a
Convention But an Initial EPR State . . . . . . 289
11.4 Failure of Bell’s Theorem Implies
Vindication of the EPR Argument . . . . . . . . 290
11.5 The Local-Realistic Framework in Question is
Strictly Non-Contextual . . . . . . . . . . . . . . . 291
11.6 Non-Commuting Standard Scores are
Merely Calculational Tools . . . . . . . . . . . . . 295
11.7 A Simply-Connected Model Cannot be
Simulated by its Discretized Imitation . . . . . 296
11.8 Counting the Number of Trivectors has No
Bearing on the Integration . . . . . . . . . . . . . 297
11.9 Twist in the Clifford Parallels Cannot be
Revealed by an Incorrect Rotor . . . . . . . . . . 298
11.10 Null Bivector is a Bivector that Subtends No
Meaningful Area . . . . . . . . . . . . . . . . . . . 300
11.11 Correlation Between Raw Scores is Given by
Covariance Between Standard Scores . . . . . . 301
11.12 What is Asserted in Moldoveanu’s Preprint is Not
Necessarily What is True . . . . . . . . . . . . . . 302
11.13 Concluding Remarks . . . . . . . . . . . . . . . . 304
12 Replies to Critics 307
Appendix 4: The Meaning of a Geometric Product 334

xii
 Preface to Second Edition

In this edition I have included two new chapters, Chapters 9 and

10, bringing the total number of chapters to 12. Both chapters are
primarily my replies to specific criticisms, but they also bring fresh
and deeper insights into the main objective of the book—which is to
understand the origins of quantum correlations. In this regard the
appendix after Chapter 9 is of special interest, because it includes
a discussion of two event-by-event simulations of my 3-sphere model
for the EPR correlation, confirming it beyond any shadow of doubt.
I am indebted to Chantal Roth and Michel Fodje for the simulations.
Unfortunately there are some further results in the subject that
I have not been able to include in this edition. Some of these results
can be found on my blog at http://libertesphilosophica.info/blog/,
which is dedicated to discussions about the contents of this book.
The work on the two new chapters included in this edition was
funded by a grant from the Foundational Questions Institute (FQXi)
Fund, a donor advised fund of the Silicon Valley Community Founda-
tion on the basis of proposal FQXi-MGA-1215 to the Foundational
Questions Institute. It was carried out while I was an affiliate of
Theiss Research, USA, and of Wolfson College and the Department
of Materials of the University of Oxford, UK. I am grateful to Martin
Castell for his continued hospitality in the Department of Materials
during the course of this work, and to Fred Diether and Tom Ray for
their moral support in the last few years of my work on this subject.
A special thanks goes to my friend Lucien Hardy, who has spent
considerable amount of his valuable time trying to understand the
ideas discussed in this book. Finally, I thank the readers of the first
edition of the book for their comments and words of encouragement.

Joy J. Christian
Oxford, England
December, 2013

xiii
 Chapter 9
Whither All the Scope and
Generality of Bell’s Theorem?

Abstract: In a recent preprint James Owen Weatherall

has attempted a simple local-deterministic model for the
EPR-Bohm correlation and speculated about why his
model fails when my counterexample to Bell’s theorem
succeeds. Here I bring out the physical, mathematical,
and conceptual reasons why his model fails. In particu-
lar, I demonstrate why no model based on a tensor rep-
resentation of the rotation group SU(2) can reproduce
the EPR-Bohm correlation. I demonstrate this by cal-
culating the correlation explicitly between measurement
results A = ± 1 and B = ± 1 in a local and determinis-
tic model respecting the spinor representation of SU(2).
I conclude by showing how Weatherall’s reading of my
model is misguided, and bring out a number of miscon-
ceptions and unwarranted assumptions in his imitation
of my model as it relates to the Bell-CHSH inequalities.
In particular, I bring out the circularity in his reasoning
and unmask the straw-man he employes to argue his case.

9.1 Introduction

In a recent preprint [1] James Owen Weatherall has attempted a

local and deterministic model for the EPR-Bohm experiment, which

209
he affirms to have been inspired by my work on Bell’s theorem
[2][3][4][5][6]. Although his analysis mainly concerns his attempted
reconstruction of my model and why the attempt fails, his preprint
has been worded in a manner that has allowed some readers to em-
brace his discussion as a criticism of my work on Bell’s theorem.
Here I demonstrate that (as he himself stresses to some extent) the
analysis Weatherall presents in his preprint has nothing to do with
my model, or with the physics and mathematics of the EPR-Bohm
correlation [7]. In fact his analysis exhibits no understanding of how
my local-realistic framework works, nor of the reasons why it explains
the origins of all quantum correlations [3][8]. I show this by first cal-
culating the EPR-Bohm correlation in a successful local-deterministic
model based on the spinor representation of SU(2), and then reveal-
ing a number of misconceptions and unwarranted assumptions in
Weatherall’s reconstruction of my model as it relates to the Bell-
CHSH inequalities [6]. I conclude that, contrary to first impressions,
Weatherall’s thinly veiled criticism of my work is entirely vacuous.

9.2 An Exact Local-Deterministic Model for the

EPR-Bohm Correlation

In order to bring out the erroneous assumptions in Weatherall’s anal-

ysis, it would be convenient to assess it in the light of a successful
model for the EPR-Bohm correlation. This will allow us to unveil
his assumptions more easily.

9.2.1 A Complete Specification of the Singlet State

To this end, let us recall Bell’s definition of a complete theory [9].

He considered a physical theory to be complete just in case its
predictions for the EPR-Bohm experiment are dictated by local-
deterministic functions of the form

A (r, λ) : IR3 × Λ −→ S 0 ≡ {−1, +1}, (9.1)

where IR3 is the space of 3-vectors r, Λ is a space of the complete

states λ, and S 0 ≡ {−1, +1} is a unit 0-sphere. He then proved a
mathematical theorem concluding that no pair of functions of this
form can reproduce the correlation as strong as that predicted by

210
quantum mechanics for the rotationally invariant singlet state:
" n
#
1 X
E(a, b) = lim A (a, λk ) B(b, λk ) 6= − a · b . (9.2)
n≫1 n
k=1

As it stands, this conclusion of Bell is entirely correct and beyond

dispute, provided we accept prescription (9.1) as codifying a com-
plete specification of the singlet state. In his preprint, following Bell
and CHSH [9][10], Weatherall accepts (9.1) as codifying a complete
specification of the singlet state, whereas my work begins by recog-
nizing that (9.1) does not, and cannot, codify a complete specifica-
tion of the singlet state [3][8]. I have argued that Bell’s prescription
is based on an incorrect underpinning of both the EPR argument
[11] and the actual topological configurations involved in the EPR-
Bohm experiments [3], even if we leave the physics and mathematics
underlying the correlation aside. My argument is rather subtle and
requires a clear understanding of what is meant by both a func-
tion in mathematics and the geometry and topology of a parallelized
3-sphere. But the bottom line of the argument is that, for any two-
level system, the EPR criterion of completeness demands the correct
measurement functions to be necessarily of the form
± 1 = A (r, λ) : IR3 × Λ −→ S 3 ∼ SU(2), (9.3)
with the simply-connected codomain S 3 of A (r, λ) replacing the
totally disconnected codomain S 0 assumed by Bell. Thus A (r, λ)
= ± 1 now represents a point of a parallelized 3-sphere, S 3 . As
a function, it takes values from the domain IR3 × Λ and ends up
belonging to the codomain S 3 . Consequently, any correlation
between a pair of such results is a correlation between points of
a parallelized 3-sphere [8]. Unless based on a prescription of this
precise form, any Bell-type analysis simply does not get off the
ground, because without completeness there can be no theorem [3].
Here S 3 —which can be thought of as the configuration space
of all possible rotations of a rotating body (including spinorial sign
changes)—is defined as the set of all unit quaternions isomorphic to
a unit parallelized 3-sphere:
(   )
3 ψ  2
S := q(ψ, r) := exp β(r)  || q(ψ, r) || = 1 , (9.4)
2 

where β(r) is a bivector rotating about r ∈ IR3 with the rotation

angle ψ in the range 0 ≤ ψ < 4π. Throughout this paper I will fol-
low the concepts, notations, and terminology of geometric algebra

211
[12][13]. Accordingly, β(r) ∈ S 2 ⊂ S 3 can be parameterized by a
unit vector r = r1 e1 + r2 e2 + r3 e3 ∈ IR3 as

β(r) := ( I · r )
= r1 ( I · e1 ) + r2 ( I · e2 ) + r3 ( I · e3 )
= r1 e2 ∧ e3 + r2 e3 ∧ e1 + r3 e1 ∧ e2 , (9.5)

with β 2 (r) = −1. Here the trivector I := e1 ∧ e2 ∧ e3 (which also

squares to −1) represents a volume form of the physical space [12][13].
Each configuration of the rotating body can thus be represented by
a quaternion of the form
ψ ψ
q(ψ, r) = cos + β(r) sin , (9.6)
2 2
which in turn can always be decomposed as a product of two bivec-
tors belonging to an S 2 ⊂ S 3 ∼ SU(2),
ψ ψ
β(r′′ ) β(r′ ) = cos + β(r) sin , (9.7)
2 2
with ψ being its rotation angle from q(0, r) = 1. Note also that
q(ψ, r) reduces to ± 1 as ψ → 2κπ for κ = 0, 1, or 2.
It is of paramount importance to note here that our topologically
corrected prescription (9.3) does not alter the actual measurement
results. For a given vector r and an initial state λ, both operationally
and mathematically we still have

A (r, λ) = + 1 or − 1 (9.8)

as the image points of the function A (r, λ) as demanded by Bell,

but now the topology of its codomain has changed from a 0-sphere
to a 3-sphere, with the latter embedded in IR4 in such a manner that
the prescriptions (9.1) and (9.3) are operationally identical [3][8]. On
the other hand, without this topological correction it is impossible to
provide a complete account of all possible measurement results in the
sense specified by EPR. Thus the selection of the codomain S 3 ֒→ IR4
in prescription (9.3) is not a matter of choice but necessity. What is
responsible for the EPR correlation is the topology of the set of all
possible measurement results. But once the codomain of A (r, λ) is
so corrected, the proof of Bell’s theorem (as given in Refs. [9]) simply
falls apart. Moreover, it turns out that the strength of the correlation
for any physical system is entirely determined by the torsion within
the codomain of the local functions A (r, λ).

212
 Returning to Weatherall’s analysis, it should be clear now that,
because it is based on a prescription other than (9.3), it too is a
non-starter [3]. More importantly, once we recognize that the only
way of providing a complete account of all possible measurement
results for the singlet state is by means of prescription (9.3), the sta-
tistical procedure for analyzing the correlation must be consistently
customized for the set S 3 of all unit quaternions, which Weatherall
fails to do. Since this procedure can be appreciated more readily
by studying the explicit construction of my model from my book
and elsewhere, I now proceed to reproduce the model in some detail
in the following subsections. Doing so will also dispel a persistent
but gravely disingenuous charge that my model encounters “certain
technical complications” [4][5].

9.2.2 Construction of the Measurement Functions

Once the measurement results are represented by functions of the

form (9.3), it is easy to reproduce the EPR-Bohm correlation in a
manifestly local, realistic, and deterministic manner. This is because
a parallelized 3-sphere has quite a unique and distinctive topological
structure [6][7]. It is one of the only two parallelizable spheres with
non-vanishing torsion—the other one, with variable torsion, being
the 7-sphere. Once parallelized by a constant torsion, the 3-sphere
remains closed under multiplication, and forms one of the only four
possible normed division algebras [7][8]. These are profound concepts
underlying the very existence and strength of quantum correlations
[3]. By ignoring them and dismissing them as irrelevant, Weatherall
is ignoring the physics and mathematics of the quantum correlations.
More importantly, because of the unique and distinctive topological
characteristics of the 3-sphere [7], the measurement functions such as
A (r, λ) for the correlation have to be constructed in a very specific
manner for any model to be successful [3]. To this end, let us begin
with the following definition of the orientation of a vector space:

Definition 9.2.1 An orientation of a finite dimensional vector space

Vd is an equivalence class of ordered basis, say {f1 , . . . , fd }, which
determines the same orientation of Vd as the basis {f1′ , . . . , fd′ } if
fi′ = ωij fj holds with det(ωij ) > 0, and the opposite orientation of
Vd as the basis {f1′ , . . . , fd′ } if fi′ = ωij fj holds with det(ωij ) < 0.

(Here repeated indices are summed over.) Thus each positive di-
mensional real vector space has precisely two possible orientations,

213
which (rather suggestively) can be denoted as λ = +1 or λ = −1.
More generally an oriented smooth manifold such as S 3 consists of
that manifold together with a choice of orientation for each of its
tangent spaces.
It is important to note that orientation of a manifold is a relative
concept [14]. In particular, the orientation of a tangent space Vd of
a manifold defined by the equivalence class of ordered basis such as
{f1 , . . . , fd } is meaningful only with respect to that defined by the
equivalence class of ordered basis {f1′ , . . . , fd′ }, and vice versa. To
be sure, we can certainly orient a manifold absolutely by choosing
a set of ordered bases for all of its tangent spaces, but the resulting
manifold can be said to be left or right oriented only with respect of
another such set of ordered basis [14].
Now the natural configuration space for an EPR-Bohm type
experiment is a unit parallelized 3-sphere, which can be embedded
in IR4 with a choice of orientation, say λ = +1 or −1. This choice of
orientation can be identified with the initial state of the particle pair
in the singlet state with respect to the orientation of the detector
basis as follows. We first characterize the embedding space IR4 by
the graded basis
{ 1, L1 (λ), L2 (λ), L3 (λ) } , (9.9)
3
with λ = ± 1 representing the two possible orientations of S and
the basis elements Lµ (λ) satisfying the algebra
Lµ (λ) Lν (λ) = − gµν − ǫµνρ Lρ (λ) , (9.10)
with an arbitrary metric gµν on S 3 . Here the bivectors { aµ Lµ (λ) }
will represent the spin angular momenta of the particles, with µ =
1, 2, 3 and the repeated indices summed over. These momenta can
be assumed to be detected by the detector bivectors, say { aµ Dµ },
with the corresponding detector basis { 1, D1 , D2 , D3 } satisfying
the algebra
Dµ Dν = − gµν − ǫµνρ Dρ (9.11)
and related to the spin basis { 1, L1 (λ), L2 (λ), L3 (λ) } as
    
1 1 0 0 0 1
 L1 (λ)  0 λ 0 0  
     D1 .
 L2 (λ)  =  0 0 λ 0   D2 
(9.12)
L3 (λ) 0 0 0 λ D3
Evidently, the determinant of this matrix works out to be det(ωij ) = λ.
Since λ2 = +1 and ω 2 is a 4 × 4 identity matrix, this relation can be

214
more succinctly written as

Lµ (λ) = λ Dµ and Dµ = λ Lµ (λ) , (9.13)

or equivalently as

{ 1, L1 (λ), L2 (λ), L3 (λ) } = {1, λ D1 , λ D2 , λ D3 } (9.14)

and

{ 1, D1 , D2 , D3 } = {1, λ L1 (λ), λ L2 (λ), λ L3 (λ)} . (9.15)

These relations reiterate the fact that orientation of any manifold

is a relative concept. In particular, orientation of S 3 defined by the
spin basis { 1, Lµ (λ) } is meaningful only with respect to that defined
by the detector basis { 1, Dµ } with the orientation λ = +1, and vice
versa. Thus the spin basis are said to define the same orientation of
S 3 as the detector basis if Lµ (λ = +1) = +Dµ , and the spin basis
are said to define the opposite orientation of S 3 as the detector basis
if Lµ (λ = −1) = − Dµ . Note also that the numbers 1 and L(r, λ)
are treated here on equal footing.
We are now in a position to define the functions A (a, λ) and
B(b, λ) as results of measurement interactions (or Clifford products)
between detector bivectors − D(a) and + D(b) and spin bivectors
L(a, λ) and L(b, λ) as follows:

SU(2) ∼ S 3 ∋ ± 1 = A (a, λk ) = − D(a) L(a, λk )

= {− aµ Dµ } { aν Lν (λk ) }
(
+ 1 if λk = + 1
= (9.16)
− 1 if λk = − 1

and

SU(2) ∼ S 3 ∋ ± 1 = B(b, λk ) = + D(b) L(b, λk )

= {+ bµ Dµ } { bν Lν (λk ) }
(
− 1 if λk = + 1
= (9.17)
+ 1 if λk = − 1 ,

where the relative orientation λ is now assumed to be a random

variable, with 50/50 chance of being +1 or − 1 at the moment of
creation of the singlet pair of spinning particles. In what follows, I
will assume that the orientation of S 3 defined by the detector basis

215
{ 1, Dν } has been fixed before hand [3][6]. Thus the spin bivector
{ aµ Lµ (λ) } is a random bivector with its handedness determined
relative to the detector bivector { aν Dν }, by the relation
L(a, λ) ≡ { aµ Lµ (λ) } = λ { aν Dν } ≡ λ D(a), (9.18)
where, as a direct consequence of the algebra (9.10) with gµν = δµν ,
the bivectors L(a, λ) satisfy the following identity:
L(a, λ) L(a′ , λ) = − a · a′ − L(a × a′ , λ). (9.19)
Using these relations the spin detection events (9.16) and (9.17) fol-
low at once from the algebras defined in (9.10) and (9.11).
Evidently, the measurement results A (a, λ) and B(b, λ) as
defined above, in addition to being manifestly realistic, are strictly
local and deterministically determined numbers. In fact, they are
not even contextual. Alice’s measurement result A (a, λ)—although
it refers to a freely chosen direction a—depends only on the initial
state λ; and likewise, Bob’s measurement result B(b, λ)—although
it refers to a freely chosen direction b—depends only on the initial
state λ. Let us also not overlook the fact that, as binary numbers,
A (a, λ) = ± 1 and B(b, λ) = ± 1 are still points of a parallelized
3-sphere. To confirm this, recall that a parallelized 3-sphere is a set
of unit quaternions of the form
 
ψ
ψ
qk (ψ, r, λ) := λk cos + L r, λk sin , (9.20)
2 2
and a measurement result such as A (a, λ) = ±1 is a limiting case
of such a quaternion constituting the 3-sphere:
S 3 ∋ ± 1 = A (a, λ) = lim
′
A (a, a′ , λ)
a →a
= lim
′
{ − D(a) L(a′ , λ) }
a →a
= lim
′
{ ( − I · a)( λ I · a′ ) }
a →a
= lim { λ a · a′ + λ I · (a × a′ ) }
a′ → a
 
ψ ψ
= lim λ cos + L(c, λ) sin
ψ→ 2κπ 2 2
= lim { q(ψ, c, λ) } . (9.21)
ψ→ 2κπ

Here I = e1 ∧ e2 ∧ e3 is the volume form, limit a′ → a is equiva-

lent to the limit ψ → 2κπ for κ = 0, 1, or 2, ψ = 2 ηa a′ is the rota-
tion angle about the axis c := a × a′ /|a × a′ |, and ηa a′ is the angle
between a and a′ [6].

216
9.2.3 A Crucial Lesson from Basic Statistics

It is important to note that the variables A (a, λ) and B(b, λ)

defined in equations (9.16) and (9.17) are generated with different
bivectorial scales of dispersion (or different standard deviations) for
each measurement direction a and b. Consequently, in statistical
terms these variables are raw scores, as opposed to standard scores
[15]. Recall that a standard score, z, indicates how many standard
deviations an observation or datum is above or below the mean. If
x is a raw (or unnormalized) score and x is its mean value, then the
standard (or normalized) score, z(x), is defined by
x−x
z(x) = , (9.22)
σ(x)

where σ(x) is the standard deviation of x. A standard score thus

represents the distance between a raw score and population mean in
the units of standard deviation, and allows us to make comparisons
of raw scores that come from very different sources [3][15]. In other
words, the mean value of the standard score itself is always zero, with
standard deviation unity. In terms of these concepts the correlation
between raw scores x and y is defined as
" n
#
1 X k
lim (x − x ) (yk − y )
n≫1 n
k=1
E(x, y) = (9.23)
σ(x) σ(y)
" n
#
1 X
= lim z(xk ) z(yk ) . (9.24)
n≫1 n
k=1

It is vital to appreciate that covariance by itself—i.e., the numerator

of equation (9.23) by itself—does not provide the correct measure of
association between the raw scores, not the least because it depends
on different units and scales (or different scales of dispersion) that
may have been used in the measurements of such scores. Therefore,
to arrive at the correct measure of association between the raw scores
one must either use equation (9.23), with the product of standard
deviations in the denominator, or use covariance of the standardized
variables, as in equation (9.24).
Now, as discussed above, the random variables A (a, λ) and
B(b, λ) are products of two factors—one random and another non-
random. Within the variable A (a, λ) the bivector L(a, λ) is a ran-
dom factor—a function of the orientation λ, whereas the bivector

217
− D(a) is a non-random factor, independent of the orientation λ :
A (a, λ) = − D(a) L(a, λ) (9.25)
and B(b, λ) = + D(b) L(b, λ) (9.26)
Thus, as random variables, A (a, λ) and B(b, λ) are generated with
different standard deviations—i.e., different sizes of the typical error.
More specifically, A (a, λ) is generated with the standard deviation
− D(a), whereas B(b, λ) is generated with the standard deviation
+ D(b). These two deviations can be calculated as follows. Since er-
rors in the linear relations propagate linearly, the standard deviation
σ(A ) of A (a, λ) is equal to − D(a) times the standard deviation of
L(a, λ) [which I will denote as σ(A) = σ(La )], whereas the standard
deviation σ(B ) of B(b, λ) is equal to + D(b) times the standard
deviation of L(b, λ) [which I will denote as σ(B) = σ(Lb )]:
σ(A ) = − D(a) σ(A) (9.27)
and σ(B ) = + D(b) σ(B). (9.28)
But since the bivector L(a, λ) is normalized to unity, and since its
mean value m(La ) vanishes on the account of λ being a fair coin, its
standard deviation is easy to calculate, and it turns out to be equal
to unity:
v
u n  2
u1 X  
σ(A) = t  A(a, λk ) − A(a, λk ) 
n
k=1
v
u n
u1 X 2
= t || L(a, λk ) − 0 || = 1, (9.29)
n
k=1

where the last equality follows from the normalization of L(a, λ).
Similarly, it is easy to see that σ(B) is also equal to 1. Conse-
quently, the standard deviation of A (a, λ) = ± 1 works out to be
− D(a), and the standard deviation of B(b, λ) = ± 1 works out to
be + D(b). Putting these two results together, we arrive at the
following standard scores corresponding to the raw scores A and B:
A (a, λ) − A (a, λ) − D(a) L(a, λ) − 0
A(a, λ) = = = L(a, λ)
σ(A ) − D(a)
(9.30)
and
B(b, λ) − B(b, λ) + D(b) L(b, λ) − 0
B(b, λ) = = = L(b, λ),
σ(B) + D(b)
(9.31)

218
where I have used identities such as − D(a)D(a) = +1. Needless to
say, these standard scores are pure bivectors:
SU(2) ∼ S 3 ⊃ S 2 ∋ L(a, λ) = ± 1 about a ∈ IR3 , (9.32)
3 2 3
and SU(2) ∼ S ⊃ S ∋ L(b, λ) = ± 1 about b ∈ IR . (9.33)

9.2.4 How Errors Propagate in a Parallelized 3-Sphere

As noted towards the end of subsection 9.2.1, one of several oversights

in Weatherall’s reading of my model concerns his failure to recog-
nize the necessity of applying the correct statistical procedure for
analyzing the correlation between the measurement results defined
by (9.16) and (9.17). This is surprising, because I pointed out this
oversight to him in a private correspondence more than eighteen
months ago. As we noted above, a parallelized 3-sphere is a set
of unit quaternions. Each point of a parallelized 3-sphere is thus
represented by a unit quaternion. As a result, the correct statisti-
cal procedure within my model must take into account how errors
propagate in a 3-sphere of unit quaternions.
Accordingly, let a probability density function P (q) : S 3 → [ 0, 1]
of random quaternions over S 3 be defined as:
( )
2
1 || q − m(q)||
P (q) = q exp − 2 , (9.34)
2π || σ(q)||
2 2 || σ(q)||

where the square root of q = p p, p ∈ S 3 , is defined as

√ √
q = p p := ± p† ( p p ) = ±( p† p ) p = ± p . (9.35)
It is a matter of indifference whether the distribution of q ∈ S 3 so
chosen happens to be normal or not. Here q is an arbitrary quater-
nion within S 3 (λ) of the form (9.20), which is a sum of a scalar
and a bivector (treated on equal footing), with 0 ≤ ψ ≤ 4π being
the double-covering rotation angle about r-axis. The mean value of
q is defined as
n
1 X k
m(q) = q , (9.36)
n
k=1
and the standard deviation of q is defined as
v
u n
u1 X †
σ[q(ψ, r, λ)] := t { qk (ψ) − m(q)} { qk (2π − ψ) − m(q)} .
n
k=1
(9.37)

219
 Note that in this definition q(ψ) is coordinated by ψ to rotate
from 0 to 2π, whereas the conjugate q† (2π − ψ) is coordinated by ψ
to rotate from 2π to 0. Thus, for a given value of λ, both q(ψ) and
q† (2π − ψ) represent the same sense of rotation about r (either both
represent clockwise rotations or both represent counterclockwise
rotations). This is crucial for the calculation of standard deviation,
for it is supposed to give the average rotational distance within S 3
from its mean, with the average being taken, not over rotational dis-
tances within a fixed orientation of S 3 , but over the changes in the
orientation λ of S 3 itself. Note also that, according to the definition
(9.20), q(ψ) and its conjugate q† (ψ) satisfy the following relation:

q† (2π − ψ) = − q(ψ) . (9.38)

Consequently, the standard deviation of both q† (2π − ψ) and − q(ψ)

must, with certainty, give the same number:

σ[ q† (2π − ψ)] ≡ σ[ − q(ψ)] . (9.39)

It is easy to verify that definition (9.37) for the standard deviation of

q(ψ) does indeed satisfy this requirement, at least when m(q) = 0.
What is more, from equation (9.38) we note that the quantity being
averaged in the definition (9.37) is essentially − q q. This quantity is
insensitive to spinorial sign changes such as q → − q, but transforms
into the quantity − q† q† under orientation changes such as λ → −λ.
By contrast, the quantity − q q† would be insensitive to both
spinorial sign changes as well as orientation changes. Thus σ[q(λ)],
as defined in (9.37), is designed to remain sensitive to orientation
changes for correctly computing its averaging function on q(λ) in
the present context.
Now, in order to evaluate σ(A ) and σ(La ), we can rewrite the
quaternion (9.20) rotating about r = a as a product

q(ψ, a, λ) = p(ψ, a) L(a, λ) (9.40)

of a non-random, non-pure quaternion

      
ψ ψ π−ψ
p(ψ, a) := sin − D(a) cos = exp − D(a)
2 2 2
(9.41)
and a random, unit bivector L(a, λ) satisfying
n
1 X
L(a, λk ) L† (a, λk ) = 1 . (9.42)
n
k=1

220
Note that p(ψ, a) reduces to the unit bivector ± D(a) for rotation
angles ψ = 0, ψ = 2π, and ψ = 4π. Moreover, using the relations
L(a, λ) = λ D(a) and D2 (a) = − 1 it can be easily checked that the
product in (9.40) is indeed equivalent to the quaternion defined in
(9.20) for r = a. It is also easy to check that the non-random quater-
nion p(ψ, a) satisfies the following relation with its conjugate:

p† (2π − ψ, a) = p(ψ, a) . (9.43)

Consequently we have
†
q† (2π − ψ, a, λ) = {p(2π − ψ, a) L(a, λ)}
= L† (a, λ) p† (2π − ψ, a) = L† (a, λ) p(ψ, a) .
(9.44)

Thus, substituting for q(ψ, a, λ) and q† (2π − ψ, a, λ) from Eqs.

(9.40) and (9.44) into Eq. (9.37), together with
n
( n
)
1 X k 1 X k
m(qa ) = qa = p(ψ, a) L(a, λ )
n n
k=1 k=1
( n
)
1 X k
= p(ψ, a) λ D(a)
n
k=1
= 0, (9.45)

we have
v
u n
u1 X
σ[q(ψ, a, λ)] = t { p(ψ, a) L(a, λk ) } { L† (a, λk ) p(ψ, a) }
n
k=1
v ( )
u n
u
t 1 X k † k
= p(ψ, a) L(a, λ ) L (a, λ ) p(ψ, a)
n
k=1
p
= p(ψ, a) p(ψ, a)
= ± p(ψ, a) . (9.46)

Here I have used the normalization of L(a, λ) as in (9.42), and the

last equality follows from the definition (9.35). It can also be deduced
from the polar form of the product

p(ψ, a) p(ψ, a) = cos (π − ψ) − D(a) sin (π − ψ)

= exp { − D(a) (π − ψ)} . (9.47)

221
 The result for the standard deviation we have arrived at, namely

σ[ q(ψ, a, λ)] = ± p(ψ, a) , (9.48)

is valid for all possible rotation angles ψ between the detector bivec-
tor − D(a) and the spin bivector L(a, λ). For the special cases when
ψ = 0, π, 2π, 3π, and 4π, it reduces to the following set of standard
deviations:

σ[ q(ψ = 0, a, λ)] = σ(A ) = ± D(a)

σ[ q(ψ = π, a, λ)] = σ(La ) = ± 1
σ[ q(ψ = 2π, a, λ)] = σ(A ) = ± D(a)
σ[ q(ψ = 3π, a, λ)] = σ(La ) = ± 1
and σ[ q(ψ = 4π, a, λ)] = σ(A ) = ± D(a) . (9.49)

To understand the physical significance of these results, let us first

consider the special case when ψ = π. Then

q(ψ = π, a, λ) = + L(a, λ) , (9.50)

which can be seen as such from the definition (9.20) above. Similarly,
for the conjugate of q(ψ = π, a, λ) we have

q† (ψ = π, a, λ) = − L(a, λ) = + L† (a, λ) . (9.51)

Moreover, we have m(La ) = 0, since L(a, λ) = + λ D(a) with λ = ± 1

being a fair coin. Substituting these results into definition (9.37) of
standard deviation—together with ψ = π—we arrive at
v
u n
u1 X
σ(La ) = t L(a, λk ) L† (a, λk ) = ± 1 , (9.52)
n
k=1

since L(a, λ) L† (a, λ) = 1. Similarly, we can consider the case when

ψ = 3π and again arrive at σ(La ) = ± 1.
Next, we consider the three remaining special cases, namely
ψ = 0, 2π, or 4π. These cases correspond to the measurement results,
as defined, for example, in equation (9.16). To confirm this, recall
from equation (9.21) that a measurement result such as A (a, λ) = ±1
is a limiting case of the quaternion (9.20). If we now rotate the bivec-
tor L(c, λ) to L(a, λ) as

D(r) L(c, λ) D† (r) = L(a, λ) (9.53)

222
using some D(r), and multiply Eq. (9.21) from the left by D(r) and
from the right by D† (r), then we arrive at

A (a, λ) = lim D(r) q(ψ, c, λ) D† (r)
a′ → a
 
ψ ψ
= lim λ cos + L(a, λ) sin
ψ→ 2κπ 2 2
= lim { q(ψ, a, λ) }
ψ→ 2κπ

= lim { p(ψ, a) L(a, λ) } , (9.54)

ψ→ 2κπ

where p(ψ, a) is defined in equation (9.41). The limit a′ → a is thus

physically equivalent to the limit ψ → 2κπ for κ = 0, 1, or 2. We
therefore have the following relation between A (a, λ) and q(ψ, a, λ):

q(ψ = 2κπ, a, λ) = ± D(a) L(a, λ) = ± A (a, λ) , (9.55)

and similarly between A † (a, λ) and q† (ψ, a, λ):

†
q† (ψ = 2κπ, a, λ) = { ± D(a) L(a, λ) } = ± A † (a, λ) . (9.56)

For example, for ψ = 0 the definition (9.20) leads to

q(ψ = 0, a, λ) = − D(a) L(a, λ) = + A (a, λ) . (9.57)

This tells us that in the ψ → 0 limit the quaternion q(ψ, a, λ)

reduces to the scalar point − D(a) L(a, λ) of S 3 . Moreover, we have
m(A ) = 0, since m(La ) = 0 as we saw above. On the other hand,
from definition (9.20) of q(ψ, a, λ) we also have the following relation
between the conjugate variables A † (a, λ) and q† (ψ = 2π, a, λ):
†
q† (ψ = 2π, a, λ) = + { D(a) L(a, λ) }
= + L† (a, λ) D† (a)
= − L† (a, λ) D(a)
= − A † (a, λ) . (9.58)

This tells us that in the ψ → 2 π limit the quaternion q† (ψ, a, λ)

reduces to the scalar point − L† (a, λ) D(a) of S 3 . Thus the case
ψ = 0 does indeed correspond to the measurement events. The phys-
ical significance of the two remaining cases, namely ψ = 2π and 4π,
can be verified similarly, confirming the set of results listed in (9.49):

σ(A ) = ± D(a) . (9.59)

223
Substituting this and σ(B) = ± D(b) into Eqs. (9.30) and (9.31)
then immediately leads to the standard scores:

± A (a, λ) − A (a, λ)
A(a, λ) =
σ(A )
± D(a) L(a, λ) − 0
=
σ(A )
 
± D(a)
= L(a, λ) = L(a, λ) (9.60)
σ(A )
and
± B(b, λ) − B(b, λ)
B(b, λ) =
σ(B)
± D(b) L(b, λ) − 0
=
σ(B)
 
± D(b)
= L(b, λ) = L(b, λ) . (9.61)
σ(B)

This confirms the standard scores derived in the equations (9.30) and
(9.31) of the previous subsection.
So far I have assumed that randomness in the measurement
results A (a, λ) and B(b, λ) originates entirely from the initial state
λ representing the orientation of the 3-sphere. In other words, I have
assumed that the local interaction of the fixed detector D(a) with
the random spin L(a, λ) does not introduce additional randomness
in the measurement result A (a, λ). Any realistic interaction be-
tween D(a) and L(a, λ), however, would inevitably introduce such
a randomness, of purely local, experimental origin. We can model
this randomness by introducing an additional random variable, say
ra ∈ [ 0, 1], not dependent on λ. Physically we can think of ra
as an alignment parameter between the detector bivector D(a) and
the spin bivector L(a, λ), with ra = 1 representing the perfect align-
ment. Clearly, introduction of this additional random parameter will
make all the bivectors and quaternions unnormalized, and the corre-
sponding probability density function (9.34) would then represent a
Gaussian distribution—provided we also assume that the orientation
λ = ± 1 of S 3 itself is distributed non-uniformly between its values
+1 and −1. Moreover, although the measurement results would then
fall within the range − 1 ≤ A (a, λ) ≤ + 1, their mean value would
be zero for a uniformly distributed λ, since the mean value of the
product of the independent random variables ra and λ would then

224
be the product of their mean values:

m(ra λ) = m(ra ) m(λ). (9.62)

More importantly, the standard scores computed above in equa-

tions (9.60) and (9.61) would not be affected by this more realis-
tic random process ra λ—at least for the special case of uniformly
distributed λ—because they involve the ratios of the corresponding
raw scores and standard deviations centered about the mean values
m(ra λ) = 0 = m(rb λ):

± A (a, λ) − A (a, λ)
A(a, λ) =
σ(A )
± ra D(a) L(a, λ) − 0
=
σ(A )
 
± ra D(a)
= L(a, λ) = L(a, λ) (9.63)
σ(A )

and

± B(b, λ) − B(b, λ)
B(b, λ) =
σ(B)
± rb D(b) L(b, λ) − 0
=
σ(B)
 
± rb D(b)
= L(b, λ) = L(b, λ) . (9.64)
σ(B)

Let us now try to understand the propagation of error within

S 3 from this physically more realistic perspective. To this end, let
the random variable q(ψ, a, λ) ∈ S 3 (λ) be such that the measure-
ment results A (a, λ) ∈ [−1, +1] remain as before, but the bivectors
L(a, λ) are subject to a random process ra λ such that S(a, λ, ra ) =
ra L(a, λ) with ra ∈ [ 0, 1]. Then the mean value m(S) and standard
deviation σ(S) of S would be a bivector and a scalar:

m(S) = a bivector
and σ(S) = a scalar. (9.65)

If we now take the detector bivector to be D(a) = I · a as before, then

the measurement results can be identified as −1 ≤ A = D S ≤ +1
so that m(A ) ≥ 0. Since D is a non-random bivector, errors gen-
erated within A by the random process ra λ would stem entirely

225
 q(f )

m(A ) + σ(A ) • f (S)

m(A ) •

m(A ) − σ(A ) •

•
• •
q
m(S) − σ(S) m(S) m(S) + σ(S)

Figure 9.1: A propagation of the 68% probability interval from a

random bivector S to a scalar A within the parallelized 3-sphere.

from the random bivector S, and propagate linearly. In other words,

the standard deviations within the random number A due to the
random process ra λ would be given by

σ(A ) = D σ(S). (9.66)

But since σ(S), as we noted, is a scalar, the typical error σ(A )

generated within A due to the random process ra λ is a bivec-
tor. The standardized variable (which must be used to compare
the raw scores A with other raw scores B ) is thus also a bivector:
A := A /σ(A ) = scalar × S.
As straightforward as it is, the above conclusion may seem un-
usual. It is important to recall, however, that in geometric algebra
both scalars and bivectors are treated on equal footing [12][13]. They
both behave as real-valued c-numbers, albeit of different grades. To
appreciate the consistency and naturalness of the above conclusion,
let
A = f (S) = D S (9.67)
be a continuous function generated by the geometric product of two
bivectors D(a) and S(a, λ, ra ) as before. The natural question then
is: How does a typical error in S governed by the probability density

226
(9.34)—which can be represented by the 68% probability interval
[ m(S) − σ(S), m(S) + σ(S) ] (9.68)
as shown in the Fig. 9.1—propagate from the random bivector S to
the random scalar A , through the function f (S) = D S? To answer
this question we note that the two end points of the interval (9.68)
represent two points, say q− and q+ , of the 3-sphere, which is a
Riemannian manifold. The geometro-algebraic distance between the
points q− and q+ can therefore be defined, say, as




d q−, q+ = q− − q+ × sign q− − q+ . (9.69)
Moreover, from definition (9.67) of A and a first-order Taylor ex-
pansion of the function f (S) about the point S = m(S) we obtain

∂f 
A = f (m(S)) + (S − m(S)) + . . . (9.70)
∂S S = m(S)

Now it is evident that the slope ∂f /∂S = D of this line is a constant.

Therefore the mean m(A ) and the standard deviation σ(A ) of the
distribution of A can be obtained by setting S = m(S) and S = σ(S):
m(A ) = f (m(S)) = D m(S) = a scalar (9.71)
∂f
and σ(A ) = σ(S) = D σ(S) = a bivector. (9.72)
∂S
The probability distribution of A is thus represented by the 68%
interval
[ m(A ) − σ(A ), m(A ) + σ(A ) ] . (9.73)
If we now set ra = 1 and thereby assume that S is in fact the unit
bivector L with a vanishing mean, then we have m(A ) = 0 and
σ(A ) = ± D, as in equation (9.59) above.
Finally, it is instructive to note that, geometrically, the prop-
agation of error within S 3 is equivalent to a simple change in the
perspective (cf. Fig. 9.1):
bivector scalar scalar bivector
z }| { z }| { f (S) z }| { z }| {
3
S ∋ m(S) ± σ(S) −−−−−−−A m(A ) ± σ(A ) . (9.74)
| {z } | {z }
quaternion quaternion

In particular, the probability density of the scalar A over S 3 corre-

sponding to interval (9.73) is equivalent to that of the bivector S over
S 3 corresponding to interval (9.68). With this, we are now ready to
derive the EPR-Bohm correlations.

227
9.2.5 Derivation of Pair Correlations Among the Points of S 3

We begin by noting that, according to my model, EPR-Bohm corre-

lations are correlations among the points of a parallelized 3-sphere
[3]. Now, since we have assumed that initially there was 50/50 chance
between the right-handed and left-handed orientations of the 3-sphere
(i.e., equal chance between the initial states λ = + 1 and λ = − 1),
the expectation values of the raw scores A (a, λ) and B(b, λ) vanish
identically. On the other hand, as discussed above, the correlation
between these raw scores (or their first product moment coefficient
à la Pearson [15]) can be obtained only by computing the covari-
ance between the corresponding standardized variables A(a, λ) and
B(b, λ), which gives
" n
#
1 X
E(a, b) = lim A(a, λk ) B(b, λk )
n≫1 n
k=1
" n
#
1 X  
k k
= lim aµ Lµ (λ ) bν Lν (λ )
n≫1 n
k=1
" n
#
1 X  k

= − gµν aµ bν − lim ǫµνρ aµ bν Lρ (λ )
n≫1 n
k=1
" n
#
1 X k
= − gµν aµ bν − lim λ { ǫµνρ aµ bν Dρ }
n≫1 n
k=1
= − gµν aµ bν − 0 , (9.75)

where I have used algebra defined in (9.10) and the relation (9.18).
Consequently, as explained in the paragraph just below Eq. (9.24),
when the raw scores A = ± 1 and B = ± 1 are compared, their prod-
uct moment will inevitably yield
" n
#
1 X
E(a, b) = lim A (a, λk ) B(b, λk ) = − gµν aµ bν ,
n≫1 n
k=1
(9.76)
since the correlation between the raw scores A and B is equal to
covariance between the standard scores A and B.
So far in this section we have put no restrictions on the metric
tensor, which, in the normal coordinates centered at a point of S 3
would be of the form
1

gµν (x) = δµν − R α µ ν γ xα xγ + O |x|3 . (9.77)
3

228
In other words, the algebra (9.10) we have used in the derivation
of correlation (9.76) is a general Clifford algebra, with no restric-
tions placed on the quadratic form [16]. On the other hand, if the
codomain of the measurement functions A (a, λ) is taken to be a
parallelized 3-sphere, then the above metric tensor specializes to
the Euclidean metric δµν , because the Riemann curvature tensor
of a parallelized 3-sphere vanishes, inducing a non-vanishing torsion
[6]. This case corresponds to the geometry of the group SU(2) and
specializes the correlation (9.76) to exhibit maximum strength:

E(a, b) = − gµν aµ bν −→ − δµν aµ bν = − cos ηab , (9.78)

which in turn manifests the sensitivity of A (a, λ) and B(b, λ) to

spinorial sign changes. To appreciate the significance of these changes
[6], recall from subsection 9.2.1 that a parallelized 3-sphere is a set
of unit quaternions of the form
ψ ψ
q(ψ, r) = cos + β(r) sin , (9.79)
2 2
with ψ being the rotation angle. It is easy to check that q(ψ, r)
respects the following rotational symmetries:

q(ψ + 2κπ, r) = − q(ψ, r) for κ = 1, 3, 5, 7, . . . (9.80)

q(ψ + 4κπ, r) = + q(ψ, r) for κ = 0, 1, 2, 3, . . . (9.81)

Thus q(ψ, r) correctly represents the state of a body that returns to

itself only after even multiples of a 2π rotation.
It is very important to appreciate that the strong correlation
derived in (9.78) are correlation among the points of a parallelized
3-sphere, S 3 , taken as the codomain of the measurement functions
A (r, λ). Thus the strength and the very existence of the EPR-
Bohm correlation (or of any correlation for that matter) stem entirely
from the topological properties of the codomain of the measurement
functions A (r, λ). Had we chosen the codomain of A (r, λ) to be any
manifold other than a parallelized 3-sphere, the resulting correlation
would not have been as strong as − cos ηab .

9.2.6 Derivation of Upper Bound Exceeding the CHSH Bound

Returning to the expectation value (9.76) in its most general form

we can now proceed to derive the Bell-CHSH-type bound on possible
correlations [3][17]. To this end, consider four observation axes, a,

229
a′ , b, and b′ , for the standard EPR-Bohm experiment. Then the
corresponding CHSH string of expectation values [3], namely the
coefficient
S(a, a′ , b, b′ ) := E(a, b) + E(a, b′ ) + E(a′ , b) − E(a′ , b′ ) , (9.82)
√
would be bounded by the constant 2 2, as discovered by Tsirel’son
within the setting of Clifford algebra applied to quantum mechanics
in general [3][17]. Here each of the joint expectation values of the
raw scores A (a, λ) = ± 1 and B(b, λ) = ± 1 are defined as
" n
#
1 X
E(a, b) = lim A (a, λk ) B(b, λk ) , (9.83)
n≫1 n
k=1

with the binary numbers such as A (a, λ) defined by the limit

S 3 ∋ ± 1 = A (a, λ) = lim { q(ψ, a, λ) } = − D(a) L(a, λ).
ψ→ 2κπ
(9.84)
Thus A (a, λ) and B(b, λ) are points of a parallelized 3-sphere and
E(a, b) evaluated in (9.83) gives correlation between such points
of the 3-sphere [3]. The correct value of the correlation, however,
cannot be obtained without appreciating the fact that the number
A (a, λ) = ± 1 is defined as a product of a λ-in-dependent constant,
namely − D(a), and a λ-dependent random variable, namely L(a, λ).
Thus the correct value of the correlation is obtained by calculating
the covariance of the corresponding standardized variables
Aa (λ) ≡ A(a, λ) = L(a, λ) (9.85)
and Bb (λ) ≡ B(b, λ) = L(b, λ) , (9.86)
as we discussed just below equation (9.24). In other words, corre-
lation between the raw scores A (a, λ) and B(b, λ) is obtained by
calculated the covariance between the standard scores A(a, λ) and
B(b, λ), as in equation (9.75) above:
" n
#
1 X k k
E(a, b) = lim A(a, λ ) B(b, λ ) = − gµν aµ bν . (9.87)
n≫1 n
k=1

The correlation between the raw scores is thus necessarily equal to

the covariance between the standard scores:
" n
#
1 X
E(a, b) = lim A (a, λk ) B(b, λk )
n≫1 n
k=1
" n
#
1 X k k
= lim A(a, λ ) B(b, λ ) = − gµν aµ bν . (9.88)
n≫1 n
k=1

230
Using this identity we can now rewrite the CHSH string of expecta-
tion values (9.82) in two equivalent expressions,
S(a, a′ , b, b′ ) =
" n
# " n
#
1 X k k 1 X k k
lim Aa (λ ) Bb (λ ) + lim Aa (λ ) Bb′ (λ )
n≫1 n n≫1 n
k=1 k=1
" n
# " n
#
1 X 1 X
+ lim Aa′ (λk ) Bb (λk ) − lim Aa′ (λk ) Bb′ (λk )
n≫1 n n≫1 n
k=1 k=1
(9.89)
and
S(a, a′ , b, b′ ) =
" n
# " n
#
1 X k k 1 X k k
lim Aa (λ ) Bb (λ ) + lim Aa (λ ) Bb′ (λ )
n≫1 n n≫1 n
k=1 k=1
" n
# " n
#
1 X k k
1 X k k
+ lim Aa′ (λ ) Bb (λ ) − lim Aa′ (λ ) Bb′ (λ ) .
n≫1 n n≫1 n
k=1 k=1
(9.90)
Our goal now is to find the upper bound on these strings of expec-
tation values. To this end, we first note that the four pairs of mea-
surement results occurring in the above expressions do not all occur
at the same time. Let us, however, conform to the usual assumption
of counterfactual definiteness and pretend that they do occur at the
same time, at least counterfactually, with equal distribution. This
assumption allows us to simplify the above expressions as
" n
′ ′ 1 X 
S(a, a , b, b ) = lim Aa (λk ) Bb (λk ) + Aa (λk ) Bb′ (λk )
n≫1 n
k=1
#
k k k k

+ Aa′ (λ ) Bb (λ ) − Aa′ (λ ) Bb′ (λ )

(9.91)
and
" n
′ ′ 1 X 
S(a, a , b, b ) = lim Aa (λk ) Bb (λk ) + Aa (λk ) Bb′ (λk )
n≫1 n
k=1
#
k k k k

+ Aa′ (λ ) Bb (λ ) − Aa′ (λ ) Bb′ (λ ) .

(9.92)

231
 The obvious question now is: Which of these two expressions
should we evaluate to obtain the correct bound on S(a, a′ , b, b′ )?
Clearly, in view of the identity (9.88) both expressions would give
one and the same answer [3]. Thus it should not matter which of the
two expressions we use to evaluate the bound. But it is also clear
from the discussion in subsections 9.2.3 and 9.2.4 that the correct
bound on the expression (9.91) involving the raw scores A and B
can only be obtained by evaluating the expression (9.92) involving
the standard scores A and B. Stated differently, if we tried to obtain
the bound on S(a, a′ , b, b′ ) by disregarding how the measurement
results have been generated in the model statistically, then we would
end up getting a wrong answer. By following the Bell-CHSH rea-
soning blindly Weatherall ends up making such a mistake. In the
end S(a, a′ , b, b′ ) is a functional of a random variable, and as such
proper statistical procedure tailored to my model must be employed
for its correct evaluation. This is an important point and the reader
is urged to review the discussions in subsections 9.2.3 and 9.2.4 once
again to appreciate its full significance.
With these remarks in mind we proceed to obtain the upper
bound on S(a, a′ , b, b′ ) by evaluating the expression (9.92) as fol-
lows. Since the standard scores Aa (λ) = L(a, λ) and Bb (λ) = L(b, λ)
appearing in this expression represent two independent equatorial
points of the 3-sphere, we can take them to belong to two discon-
nected “sections” of S 3 (i.e., two disconnected 2-spheres within S 3 ),
satisfying

[ Ar (λ), Br′ (λ) ] = 0 ∀ r and r′ ∈ IR3 , (9.93)

which is equivalent to anticipating a null outcome along the direction

r × r′ exclusive to both r and r′ . If we now square the integrand
of equation (9.92), use the above commutation relations, and use
the fact that all bivectors square to −1, then the absolute value of
S(a, a′ , b, b′ ) leads to the following variance inequality [3]:

|S(a, a′ , b, b′ )| = |E(a, b) + E(a, b′ ) + E(a′ , b) − E(a′ , b′ )|

v " #
u n
u 1 X 
6 t lim 4 + 4 T a a′ (λk ) T b′ b (λk ) ,
n≫1 n
k=1
(9.94)
where the classical commutators
1
T a a′ (λ) := [ Aa (λ), Aa′ (λ)] = − Aa×a′ (λ) (9.95)
2

232
and
1
T b′ b (λ) := [ Bb′ (λ), Bb (λ)] = − Bb′ ×b (λ) (9.96)
2
are the geometric measures of the torsion within S 3 . Thus, it is the
non-vanishing torsion T within the parallelized 3-sphere—the paral-
lelizing torsion which makes its Riemann curvature tensor vanish—
that is ultimately responsible for the strong quantum correlation
[3][7]. We can see this at once from Eq. (9.94) by setting T = 0,
and in more detail as follows: Using definitions (9.85) and (9.86) for
Aa (λ) and Bb (λ) and making a repeated use of the bivector identity

L(a, λ) L(a′ , λ) = − a · a′ − L(a × a′ , λ) (9.97)

specialized for the metric gµν = δµν on S 3 , the above inequality for
S(a, a′ , b, b′ ) can be further simplified to

|S(a, a′ , b, b′ )|
v " #
u n
u 1 X
6 t4 − 4 (a × a′ ) · (b′ × b) − 4 lim L(z, λk )
n≫1 n
k=1
v " #
u n
u 1 X
6 t4 − 4 (a × a′ ) · (b′ × b) − 4 lim λk D(z)
n≫1 n
k=1
p
′ ′
6 2 1 − (a × a ) · (b × b) − 0 , (9.98)

where z = (a × a′ ) × (b′ × b), and—as before—I have used the rela-

tion (9.18) between L(z, λ) and D(z) from subsection 9.2.2. Finally,
by noticing that (a × a′ ) · (b′ × b) is bounded by trigonometry as

−1 6 (a × a′ ) · (b′ × b) 6 +1 , (9.99)

the above inequality can be reduced to the form

√ √
− 2 2 6 E(a, b) + E(a, b′ ) + E(a′ , b) − E(a′ , b′ ) 6 + 2 2 ,
(9.100)
which exhibits an extended upper bound on possible correlations.
Thus, when in an EPR-Bohm experiment raw scores A = ± 1 and
B = ± 1 are compared by coincidence counts [18], the normalized
expectation value of their product
h i
C++ (a, b) + C−− (a, b) − C+− (a, b) − C−+ (a, b)
E(a, b) = h i
C++ (a, b) + C−− (a, b) + C+− (a, b) + C−+ (a, b)
(9.101)

233
is predicted by my model to respect, not the √ Bell-CHSH upper
bound 2, but the Tsirel’son upper bound 2 2, where C+− (a, b)
etc. represent the number of joint occurrences of detections + 1 along
a and − 1 along b etc.
This completes the presentation of my local, realistic, and deter-
ministic model for the EPR-Bohm correlation.

9.3 Physical, Mathematical, and Conceptual

Fallacies of Weatherall’s Model

9.3.1 What is Wrong with Weatherall’s Measurement

Ansatz?

With the successful model firmly in place, we are now in a position to

understand why Weatherall’s model fails. To begin with, his model is
based on a different representation of the rotation group. It is in fact
based, not on the spinorial rotation group SU(2), but something akin
to its tensorial cousin SO(3), which is a group of all proper rotations
in IR3 , insensitive to spinorial sign changes. In fact, Weatherall takes
a rather odd space, namely IR3 ∧ IR3 , for the codomain of his mea-
surement functions A (r, λ), and then introduces another projection
map to arrive at the measurement results {−1, +1}. Compared to
my measurement functions (9.16) obtained through a smooth lim-
iting process (9.21), his two-step measurement process is rather ar-
tificial. Moreover, since his is not a simply-connected, parallelized
codomain such as S 3 that remains closed under multiplication, it
cannot possibly satisfy the completeness criterion of EPR [3][7]. It
is therefore not surprising why Weatherall is unable to find strong
correlation among its points. Moreover, at the end of his two-step
process his measurement results {−1, +1} are no longer the image
points within the codomain IR3 ∧ IR3 of his measurement functions.
This is in sharp contrast with the situation in my model, where
my measurement results {−1, +1} remain very much a part of the
codomain S 3 of my measurement functions. The reason why this
comes about naturally within my model is because S 3 , the set of unit
quaternions, is a simply-connected surface embedded in IR4 that is
equipped with a graded basis made of both scalars and bivectors:
{1, e2 ∧ e3 , e3 ∧ e1 , e1 ∧ e2 } . (9.102)
Thus the scalars {−1, +1} are as much a part of S 3 as the bivectors
L(r, λ) are, regulated by these unified basis [3]. I am tempted to

234
E (a, b)

+1

0 ηab

−1
0 90 180 270 360

Figure 9.2: Local-deterministic correlations among the points of a

parallelized 3-sphere can be stronger-than-classical but not quantum.

quip: What Nature has joined together, let no man put asunder. By
contrast Weatherall’s codomain is disconnected between the space
IR3 ∧ IR3 of “bivectors” and the set {−1, +1} of scalars. His image
points can thus be at best either bivectors or scalars, but not both.
It is a disjoint world, more like the world of quantum mechanics.
Let us, however, be more charitable to Weatherall. Let us grant
him the codomain of his measurement functions to be the connected
real projective space IRP3 , which is homeomorphic to the rotation
group SO(3). After all, he does mention the Lie algebra so(3) in one
of his footnotes. So let us grant him the smooth one-step measure-
ment process

± 1 = A (r, λ) : IR3 × Λ −→ IRP3 ∼ SO(3) (9.103)

to reach the image subset {−1, +1}. This smooth map is well-defined
within my model, since IRP3 is simply the set S 3 of unit quaternions
[cf. Eq. (9.4)] with each point identified with its antipodal point
[6]. The measurement results ± 1 ∈ IRP3 ∼ SO(3) are thus limiting
points of a quaternion, just as in equation (9.21). The map that
takes us from S 3 to IRP3 can now be used to project the metric δµν

235
on S 3 onto IRP3 to obtain the following induced metric on IRP3 :

 2
− 1 + π ηab if 0 ≤ ηab ≤ π
− Jµν aµ bν = (9.104)

 2
+ 3 − π ηab if π ≤ ηab ≤ 2π .

Further details of how this metric is obtained from the metric on S 3

can be found in section III of Ref. [6]. The two metrics δµν and Jµν
thus provide relative measures of geodesic distances on the manifolds
S 3 and IRP3 , respectively. Substituting the metric on IRP3 into
equation (9.75) the correlation between the points of SO(3) then
works out to be

 2
− 1 + π ηab if 0 ≤ ηab ≤ π
E(a, b) = − Jµν aµ bν =


+ 3 − π2 ηab if π ≤ ηab ≤ 2π .
(9.105)

The two sets of correlations, (9.78) and (9.105), are compared in

Fig. 9.2. The general correlation function E(a, b) derived in equation
(9.75) can thus serve to distinguish the geodesic distances D(a, b)
on the groups SU(2) and SO(3) [6].

9.3.2 Why did We Lose the Strong Correlation for SO(3)?

It is crucial to appreciate that even when we choose SO(3) as the

codomain of the function A (r, λ) the correct statistical procedure
that must be followed is the one described in subsections 9.2.3 and
9.2.4 above. This is because, as Weatherall himself notes, the Lie
algebras of SU(2) and SO(3) are isomorphic to each other. In other
words, the local algebraic or tangent space structures on SU(2) and
SO(3) are identical, but not their metrical structures in the sense
of geodesic distances. Thus the above statistical procedure, tailored
to the graded basis (9.102), leading up to the general expression
(9.76) for correlations and beyond, is equally inevitable in the case
of SO(3). Comparing the two sets of correlations resulting from
this procedure—one for the prescription (9.3) and other for the pre-
scription (9.103)—it is then easy to see why we have lost the strong
correlation in the second case. We started out with S 3 as a codomain
of A (r, λ) and then, for the case of SO(3), we identified each point
of S 3 with its antipodal point. But in doing so we lost the follow-
ing spinorial rotation symmetry satisfied by q(ψ, r), as described in

236
equations (9.80) and (9.81) above:

q(ψ + 2κπ, r) = − q(ψ, r) for κ = 1, 3, 5, 7, . . . (9.106)

In other words, by identifying the antipodal points of S 3 we lost

the sensitivity to spinorial sign changes. As a result, q(ψ, r) now
represents the state of a rotating body that returns to itself after
any and all multiples of 2π rotation:

q(ψ + 2κπ, r) = + q(ψ, r) for any κ = 0, 1, 2, 3, . . . (9.107)

This is the real reason why we lost the strong correlation for the
SO(3) case. The reason Weatherall has argued for is an artifact of
his bad choice of measurement functions. It stems from a failure
to appreciate the unified nature of the graded basis (9.102) and the
associated fact that the scalars {−1, +1} and the bivectors L(r, λ)
occur as image points within the same codomain S 3 in my model.
Thus the loss of correlation has nothing to do with the fact that
ultimately the measurement functions must map to the image subset
{−1, +1}. They manifestly do in my model [cf. Eqs. (9.16), (9.17),
and (9.21)]. The raw scalar numbers A = ± 1 and B = ± 1 mapped
to the image subset {−1, +1}—according to my model—are indeed
the numbers used by Alice and Bob for calculating the correlation
in the usual manner. And when, at the end of their experiment,
they evaluate the statistical quantity S(a, a′ , b, b′ ) involving these
numbers as
" n
′ ′ 1 X 
S(a, a , b, b ) = lim Aa (λk ) Bb (λk ) + Aa (λk ) Bb′ (λk )
n≫1 n
k=1
#
k k k k

+ Aa′ (λ ) Bb (λ ) − Aa′ (λ ) Bb′ (λ ) ,

(9.108)

they will inevitably

√ find that it exceeds the bound of 2 and extends
to the bound of 2 2. This conclusion may seem odd from the per-
spective of Bell-type reasoning, but the evidence presented for it in
subsection 9.2.6 is incontrovertible.

9.4 Concluding Remarks

In addition to the main issue discussed above, it is instructive to

reflect on the broader reasons why Weatherall’s model fails. We can

237
in fact identify at least six erroneous steps which engender the failure
of his model from the start:

1. Choice of incomplete codomain of the measurement functions

A (r, λ) (although with correct image points ± 1).

2. Neglect of putting scalars and bivectors on equal footing within

a single, comprehensive, real number system.

3. Failure to implement a spinor representation of SU(2) by rec-

ognizing the significance of spinorial sign changes.

4. Lack of appreciation of the role played by the parallelizing tor-

sion within S 3 for the existence and strength of strong correla-
tions (or not recognizing the discipline of parallelization as the
true source of strong correlations).

5. Failure to appreciate how errors propagate within S 3 , when

taken as a codomain of the measurement functions.

6. Neglect of the correct statistical procedure in √

the derivations
of both the correlation and the upper bound 2 2.

Although interconnected, any one of these reasons is sufficient for the

failure of Weatherall’s model. Recognizing this, I must conclude that,
contrary to first impressions, Weatherall’s thinly veiled criticism of
my work is entirely vacuous.

Appendix 3: Local-Realistic Violations of CHSH

Inequality as well as Clauser-Horne Inequality

A.3.1 Local-Realistic Violations of the CHSH Inequality

Verified in an Event-by-Event Simulation

In this appendix I demonstrate violations of both the CHSH and the

Clauser-Horne inequalities within the local model described in the
previous pages. This will further lay bare the vacuity of Weatherall’s
analysis. To this end, let us first confirm the following probabilistic

238
predictions of quantum mechanics within our local model:

P ++ := P {A = +1, B = +1; A B = +1 | ηab }

1 η 
ab
= sin2 , (A.9.1)
2 2
P −− := P {A = −1, B = −1; A B = +1 | ηab }
1 η 
ab
= sin2 , (A.9.2)
2 2
P −+ := P {A = −1, B = +1; A B = −1 | ηab }
1 η 
ab
= cos2 , (A.9.3)
2 2
and P +− := P {A = +1, B = −1; A B = −1 | ηab }
1 η 
ab
= cos2 . (A.9.4)
2 2
Here ηab is the angle between a and b. As is well known, confirming
these probabilities is equivalent to confirming
" n
#
1X i i
E(a, b) = lim A (a, λ ) B(b, λ )
n≫1 n
i=1
P ++ × (A B = +1) + P −− × (A B = +1)
=
P ++ + P −− + P −+ + P +−
P −+ × (A B = −1) + P +− × (A B = −1)
+
P ++ + P −− + P −+ + P +−
++ −−
P +P − P −+ − P +−
= ++
P + P −− + P −+ + P +−
1 η  1 η 
ab ab
= sin2 + sin2
2 2 2 2
1 η  1 η 
ab ab
− cos2 − cos2
2 2 2 2
= − cos ηab , (A.9.5)

which is the strong correlation (9.78) predicted by our local model

(see also Ref. [18]). To confirm this result explicitly, suppose ini-
tially a pair of spins, −L(eo , λ) and L(eo , λ), is created in a sin-
glet configuration, where eo is a random direction about which the
spins are rotating at the moment of their creation, and λ = ± 1 is
their initial state, which, in our model, is taken to be the orienta-
tion of a parallelized 3-sphere within which the events A and B are
occurring. At a later time, t, let Alice and Bob detect the spins with
detectors D(a) and D(b), respectively. Thus, at time t, the spin

239
bivectors −L(et , λ) and L(et , λ) are projected by them onto the
detector bivectors D(a) and D(b), respectively. Then, as we defined
in Eqs. (9.16) and (9.17) above [cf. also Eq. (9.21)], we have the results

S 3 ∋ ± 1 = A (a, λ) = lim A (a, eo , λ)

eo → a

= lim { − D(a) L(eo , λ) }

eo → a

= lim { ( − I · a)( λ I · eo ) }
eo → a

= lim { +λ a · eo + λ I · (a × eo ) }
eo → a
(A.9.6)

and

S 3 ∋ ± 1 = B(b, λ) = lim B(b, eo , λ)

eo → b

= lim { + D(b) L(eo , λ) }

eo → b

= lim { ( + I · b)( λ I · eo ) }
eo → b

= lim { −λ b · eo − λ I · (b × eo ) }.
eo → b
(A.9.7)

Let me stress once again that the measurement results A (a, λ) and
B(b, λ) as defined above, in addition to being manifestly realistic,
are strictly local and deterministically determined numbers within
the 3-sphere. In fact, they are not even contextual. Alice’s measure-
ment result A (a, λ)—although it refers to a freely chosen direction
a—depends only on the initial state λ; and likewise, Bob’s measure-
ment result B(b, λ)—although it refers to a freely chosen direction
b—depends only on the initial state λ. In other words, they are
statistically independent events within S 3 .
Now, for the purposes of our calculations below, it is convenient
to make a note of the following useful identities:

L(eo , λ) = λ D(eo ) = ± D(eo ) = (± I) · eo = I · (± eo )

= D(± eo ) = D(λ eo )
and L (eo ) = D2 (eo ) = −1.
2
(A.9.8)

Thus, with eo uniformly distributed in IR3 , the measurement results

240
 S3
B = ±1
A = ±1

A B = ±1

Figure 9.3: The measurement results A and B observed by Alice

and Bob are points of a parallelized 3-sphere, S 3 . Since S 3 remains
closed under multiplication, their product A B is also a point of the
same 3-sphere, with its value, ±1, dictated by the topology of S 3 .

(A.9.6) and (A.9.7) can be expressed equivalently as

S 3 ∋ ± 1 = A (a; eo , λ) = − D(a) D(λ eo )

(
+ 1 if λ eo = + a
= (A.9.9)
− 1 if λ eo = − a

and

S 3 ∋ ± 1 = B(b; eo , λ) = + D(b) D(λ eo )

(
− 1 if λ eo = + b
= (A.9.10)
+ 1 if λ eo = − b .

Note again that measurement functions A (a; eo , λ) and B(b; eo , λ)

are manifestly local, precisely as demanded by Bell. Apart from
the common causes λ and eo , their values depend only on a and
b, respectively, and nothing else. Moreover, since eo is uniformly
distributed within IR3 and λ has 50/50 chance of being ± 1, their
averages vanish:

A (a; eo , λ) = 0 = B(b; eo , λ) . (A.9.11)

241
 Let us confirm this averaging conditions in some more detail by
explicitly rewriting equations (A.9.6) and (A.9.7) as

S 3 ∋ ± 1 = A (a; eo , λ)
= lim { + cos( ηaλeo ) + I · ca (λeo ) sin( ηaλeo )}
ηaλeo → κπ
(A.9.12)

and

S 3 ∋ ± 1 = B(b; eo , λ)
= lim { − cos( ηbλeo ) − I · cb (λeo ) sin( ηbλeo )}.
ηbλeo → κπ
(A.9.13)

Here κ = 0, 1, 2, . . .; ηaλeo is the angle between a and λ eo ; ca (eo )

:= a × eo /|a × eo |; and cb (eo ) := b × eo /|b × eo |. Now, since λeo is
a random unit vector in IR3 , the quantity C(a; eo , λ) := cos( ηaλeo ),
being a function of this random variable, acts as a loaded die, with
“mass” distribution (or probability density) | cos( ηaλeo )|, taking val-
ues in the interval [−1, +1], with 50% chance of landing on its face
marked +1:
−1 ≤ C(a; eo , λ) ≤ +1 , (A.9.14)
with 50% chance for the occurrence1

C(a; eo , λ) = +1 ,

1 To understand these odds, let p(a; eo , λ) = | cos( ηaλeo )| be the probability

density function defined over a 2-sphere of unit radius with respect to a fixed
reference vector a = x, for observing a measurement result A = +1 or A = −1
about a:
p(a; eo , λ) : S 2 → [0, 1] .
For the purposes of our calculation a particular vector λeo on the 2-sphere can
be specified by the angle ηaλeo in the range [ 0, π ] with respect to the fixed
vector a. Consequently, we can calculate the probability P (A , a) of occurrence
of a measurement result A = +1 or −1 by integrating over S 2 as follows:
1
Z
P (A , a) = p(a; eo , λ) dΩ
4π S 2
Z π Z 2π
1
= | cos( ηaλeo )| sin( ηaλeo ) dξyλeo dηaλeo ,
4π 0 0
where dΩ = sin( ηaλeo ) dξyλeo dηaλeo is the differential solid angle on the
2-sphere, with ξyλeo being the azimuthal angle from the y-axis and ηaλeo
being the polar angle from a = x. Now it is evident from Eq. (A.9.12) that for
the occurrence of the result A = +1 the angle ηaλeo must fall in the range
[ 0, π/2 ] for λ = +1, and in the range [ π/2, π ] for λ = −1. Consequently, the

242
and similarly for C(a; eo , λ) = −1. Note that for a fixed orienta-
tion λ, say λ = +1, C(a; eo , +1) has only 25% chance of landing on
its face marked +1. But for λ = +1 or λ = −1, which is equiva-
lent to the possibilities ηaeo or π − ηaeo in the cos( ηaλeo ) part of
A (a; eo , λ), C(a; eo , λ) has further 25% chance of landing on its
face +1, because λ = +1 or λ = −1 are two mutually exclusive pos-
sibilities. Consequently, the die C(a; eo , λ) has exactly 50/50 chance
of landing on its faces +1 or −1 in general. Thus, the measurement
outcomes A = +1 or −1 observed by Alice (regardless of Bob) is
a result of a throw of the die C(a; eo , λ), which has 50/50 chance
of landing on its face marked +1 or −1, depending on the values
of ηaλeo . One can confirm these odds for C(a; eo , λ) in a computer
simulation, as has been done in the simulation described in Fig. 9.4.

probability of observing the result A = +1 is

Z π Z 2π
1 2
P (A = +1, a) = cos( ηaeo ) sin( ηaeo ) dξyeo dηaeo
4π 0 0
Z π Z 2π
1
+ cos( π − ηaeo ) sin( π − ηaeo ) dξyeo dηaeo
4π π2 0
Z π
1 2
= cos( ηaeo ) sin( ηaeo ) dηaeo
2 0
1 π
Z
− cos( ηaeo ) sin( ηaeo ) dηaeo
2 π2
π π
1 sin2 ( ηaeo )  2 1 sin2 ( ηaeo )  1 1 1
= ×  − ×  = + = .
2 2  2 2 π 4 4 2
0 2

Similarly, it is evident from Eq. (A.9.12) that for the occurrence of a result
A = −1 the angle ηaλeo must fall in the range [ π/2, π ] for λ = +1, and in the
range [ 0, π/2 ] for λ = −1, with the probability density remaining non-negative:
0 ≤ p(a; eo , λ) ≤ 1. Consequently, the probability of observing A = −1 is
Z π Z 2π
1 2
P (A = −1, a) = − cos( π − ηaeo ) sin( π − ηaeo ) dξyeo dηaeo
4π 0 0
Z π Z 2π
1
− cos( ηaeo ) sin( ηaeo ) dξyeo dηaeo
4π π2 0
Z π
1 2
= cos( ηaeo ) sin( ηaeo ) dηaeo
2 0
1 π
Z
− cos( ηaeo ) sin( ηaeo ) dηaeo
2 π2
π π
1 sin2 ( ηaeo )  2 1 sin2 ( ηaeo )  1 1 1
= ×  − ×  = + = .
2 2  2 2 π 4 4 2
0 2

These probabilities are confirmed in the simulation described in Fig. 9.4.

243
 The question now is: How do the actual values A and B of
the functions A (a; eo , λ) and B(b; eo , λ) come about within the
3-sphere when they are observed simultaneously by Alice and Bob?
To answer this question, recall that 3-sphere remains closed under
multiplication, with its points represented by quaternions of the form
(9.6), and A and B are limiting values of such quaternions. There-
fore the values A and B observed by Alice and Bob are constrained
by the value A B(a, b; eo ) of the product A (a; eo , λ)B(b; eo , λ),
which is a limiting value of the product quaternion:

S 3 ∋ ± 1 = A (a; eo , λ)B(b; eo , λ)
= lim [{ + a · eo + I · (a × eo ) } { − b · eo − I · (b × eo )}] ,
eo → ±a
eo → ±b
(A.9.15)

where I have omitted explicit reference to λ in the RHS because

it drops out from the product. Let me stress again that, since S 3
remains closed under multiplication, a limit like this one is the only
legitimate way of determining the correct values of A , B, and A B,
since all three of them are constrained to be what they possibly
could be by the very geometry and topology of the 3-sphere. In
other words, by taking the orientation λ of S 3 as the initial state
of the spin system, we have in fact taken the entire geometry and
topology of S 3 (λ) as the common cause that determines the actual
values A and B, which are thus dictated to be what they turn out
to be by the product A B resulting in the limit (A.9.15). Hence our
goal is to find the correct product quaternion whose limit is the value
A B = +1 or −1.
To this end, let us work out the product appearing on the RHS
of the above limit equation explicitly, which gives

{ + a · eo + I · (a × eo ) } { − b · eo − I · (b × eo )}
= − (a · eo )(b · eo ) + (a × eo ) · (b × eo )
− I · { (a · eo )(b × eo ) + (b · eo )(a × eo ) − (a × eo ) × (b × eo ) }
−→ A B(a, b; eo ) = ± 1 ∈ S 3 . (A.9.16)

Although it may not be obvious at first sight, since S 3 remains closed

under multiplication, this product is necessarily a unit quaternion of

244
Figure 9.4: A simulation of outcome probabilities for simultaneously
occurring measurement events A = ±1 and B = ±1 within a paral-
lelized 3-sphere. The probabilistic predictions of the manifestly local
model described in the text match exactly with the corresponding
predictions of quantum mechanics [6]. The x-axis in the chart spec-
ifies the values of the relative angle ηab , and the y-axis records the
rate of simultaneous occurrences of events A = ±1 and B = ±1 as
a function of ηab . Thus, for example, the blue curve depicts the
ratio of the number of simultaneous occurrences of events A = +1
and B = +1 (or A = −1 and B = −1) over the total number of
occurrences and non-occurrences of such events. The straight line
at y = 0.5, on the other hand, depicts the number of occurrences of
events such as A = +1 at one station (with no detector present at
the other station) over the total number of occurrences and non-
occurrences of such events. The computer code for the simula-
tion is available at https://github.com/chenopodium/JCS. Further
discussion on the relevant issues can be found also on my blog at
http://libertesphilosophica.info/blog/. I am grateful to Chantal Roth
for kindly writing the code and for insightful discussions. [It is worth
noting that two other investigators have independently reproduced
the above simulation, writing codes in languages other than Java
used by Chantal Roth. Austin Smith used Excel Visual Basic to
reproduce the simulation, whereas John Reed used Mathematica.
Moreover, using Python Michel Fodje has built a different simula-
tion, which is available at https://github.com/minkwe/epr-simple/.]

245
the form (9.6):
ψ ψ
q(ψ, r) = cos + (I · r) sin
2 2
⇐⇒ C(a, b; eo ) + I · ck (eo ) S(a, b; eo ) ,
with C 2 + S 2 = 1 . (A.9.17)

In other words, the product (A.9.16) is a sum of a scalar part and a

bivector part, and when its scalar part, C, reduces to ±1, its bivec-
tor part reduces to zero, because the coefficient, S, of the bivector
I · ck (eo ) reduces to zero. In view of (A.9.15), (A.9.16), and (A.9.17)
it is thus clear that the limits A B(a, b; eo ) → +1 or −1 are equiv-
alent to the limits

C(a, b; eo ) := − (a·eo )(b·eo ) + (a×eo )·(b×eo ) −→ + 1 (A.9.18)

or

C(a, b; eo ) := − (a·eo )(b·eo ) + (a×eo )·(b×eo ) −→ − 1 . (A.9.19)

But since eo is a uniformly distributed random unit vector in IR3 ,

the quantity C(a, b; eo ) is also a random variable, taking values from
the interval [−1, +1]:

−1 ≤ C(a, b; eo ) ≤ +1 . (A.9.20)

Thus, just as C(a; eo , λ) in Eq. (A.9.14), C(a, b; eo ) is a die with

certain propensities for landing on its faces +1 or −1. In particular,
for a sufficiently large number of random vectors eo , the random
variable C(a, b; eo ) has 47% chance of landing on its face marked
−1 for b = +a, and 47% chance of landing on its face marked +1 for
b = −a. What is more, from the LHS of the conditions (A.9.18) and
(A.9.19), using the limits eo → a and eo → b , it is easy to verify
that

A (a; eo , λ)B(b; eo , λ) = A B(a, b; eo ) −→ −1 when b = +a,

(A.9.21)
whereas

A (a; eo , λ)B(b; eo , λ) = A B(a, b; eo ) −→ +1 when b = −a ,

(A.9.22)
where the second result is due to the spinorial sign change that the
quaternion A B(a, b = −a) undergoes with respect to the quater-
nion A B(a, b = +a) upon rotation by angle ηab = π. [ It is worth

246
Figure 9.5: A simulation of the correlation between simultaneously
occurring measurement events A = ±1 and B = ±1 within the par-
allelized 3-sphere shown in Fig. 9.3. As we noted in the caption of
Fig. 9.4, the probabilistic predictions of the manifestly local model
described in the text match exactly with the corresponding predic-
tions of quantum mechanics [3][6][8]. The correlation generated in
this simulation is essentially an “addition” of the coincident probabil-
ities given in Eqs. (A.9.37) to (A.9.40), as computed in Eq. (A.9.41).

noting here that Weatherall’s model is incapable of exhibiting such a

change from A B = −1 to A B = +1 in the product of the observed
numbers A and B. His model has been deliberately manufactured
to fail this first test, in anticipation of his intended goal.]
The above results are of course two very special cases. They
are the cases when the measurement axes chosen by Alice and Bob
happen to be aligned or anti-aligned with each other. In more gen-
eral cases, when the measurement axes of Alice and Bob happen to
subtend angles different from ηab = 0 or π, it takes a bit more effort
to deduce what values the results A and B and their product A B
within S 3 would turn out to have. In fact, because C(a, b; eo ) is a
random variable, in more general cases statistical considerations are
inevitable. The question we must ask in the general cases is the fol-
lowing: Given the random variable C(a, b; eo ) taking values in the

247
interval [−1, +1], what are the chances—as functions of the angle
ηab between a and b—for the occurrences of the events A B = +1
and A B = −1? The answer to this question can be obtained by
computing the following set of conditional probabilities:

P { A = +1, B = +1; A B = +1 | ηab }

= P { C(a, b; eo , φpo , φqo , φro , φso ) = +1 | ηab } , (A.9.23)

P { A = −1, B = −1; A B = +1 | ηab }

= P { C(a, b; eo , φpo , φqo , φro , φso ) = +1 | ηab } , (A.9.24)

P { A = +1, B = −1; A B = −1 | ηab }

= P { C(a, b; eo , φpo , φqo , φro , φso ) = −1 | ηab } , (A.9.25)

and

P { A = −1, B = +1; A B = −1 | ηab }

= P { C(a, b; eo , φpo , φqo , φro , φso ) = −1 | ηab } , (A.9.26)

where the random variable C(a, b; eo )—which (as we saw above) can
be thought of as a loaded die—is now given by

C(a, b; eo , φpo , φqo , φro , φso )

= { − cos(ηaeo + φpo ) cos(ηbeo + φro )
+ cos ηca cb sin(ηaeo + φqo ) sin(ηbeo + φso ) }/Na Nb . (A.9.27)

In the form of Eq. (A.9.17), this is the scalar part of the product of
the following two manifestly local quaternions:

q(a; eo , λ, φpo , φqo )

= { + cos( ηaλeo + φpo ) + I · ca (λeo ) sin( ηaλeo + φqo ) }/Na
≡ { Ca + (I · ca ) Sa } (A.9.28)

and

q(b; eo , λ, φro , φso )

= { − cos( ηbλeo + φro ) − I · cb (λeo ) sin( ηbλeo + φso ) }/Nb
≡ { Cb + (I · cb ) Sb }, (A.9.29)

where q
Na = cos2 (ηaeo + φpo ) + sin2 (ηaeo + φqo )

248
t

A = ±1 B = ±1
S3
a b

λeo

Figure 9.6: The measurement results A (a; eo , λ) and B(b; eo , λ)

observed independently by Alice and Bob are manifestly local.

and q
Nb = cos2 (ηbeo + φro ) + sin2 (ηbeo + φso )
are normalizing factors. Needless to say, C(a, b; eo , φpo , φqo , φro , φso )
specified in Eq. (A.9.27) is the same as that specified in Eqs. (A.9.18)
and (A.9.19) apart from the phase shifts φpo , φqo , φro , and φso . These
phase shifts contribute to the rotations of the product quaternion
about the random vector ck (eo ) [which is different in general from
both ca (λeo ) and cb (λeo ) ]. They are simply numerical constants2 ,
independent of the parameter vectors a and b, or of the random
vector eo . They are thus parts of the geometry of the 3-sphere. To-
gether with eo , they form what Bell referred to as the “past causes.”
In terms of the product of the quaternions (A.9.28) and (A.9.29),
we are now in a position to express the results A (a; eo , λ) and
B(b; eo , λ), as well as their product A B(a, b; eo ), in the following
manifestly local-realistic forms:
S 3 ∋ ± 1 = A (a; eo , λ)
= lim { C(a; eo , λ, φpo , φqo ) + I · ca (λeo ) S(a; eo , λ, φpo , φqo )},
Ca → ±1
(A.9.30)
3
S ∋ ± 1 = B(b; eo , λ)
= lim { C(b; eo , λ, φro , φso ) + I · cb (λeo ) S(b; eo , λ, φro , φso )},
Cb → ±1
(A.9.31)

2 In physical terms, the constant phase shifts ensure that Alice and Bob do not
end up detecting the same particle of either spin in a given run of experiments.

249
and

S 3 ∋ ± 1 = A B(a, b; eo )
= lim { C(a, b; eo , φpo , φqo , φro , φso )
Cab → ±1

+ I · ck (eo ) S(a, b; eo , φpo , φqo , φro , φso )}.

(A.9.32)

Here the vector ck (eo ) depends on both ca (λeo ) and cb (λeo ), as well
as on the other past causes. Although we do not require its explicit
expression, it can be easily worked out by evaluating the geometric
product of the quaternions q(a; eo , λ, φpo , φqo ) and q(b; eo , λ, φro , φso ).
The 3-sphere analogue of Bell’s locality condition then takes the form

{ Cab + (I · ck ) Sab } = { Ca + (I · ca ) Sa } { Cb + (I · cb ) Sb }
( with all C 2 + S 2 = 1 ),
i.e., S 3 ∋ qab = qa qb for qa and qb ∈ S 3 , (A.9.33)

which, in the measurement limits (A.9.30) and (A.9.31), reduces to

factorization of the corresponding scalar values (cf. Eq. (3.4) of [20]):

A B(a, b; eo ) = A (a; eo , λ) × B(b; eo , λ) (A.9.34)

giving S 3 ∋ ± 1 = {± 1} × {± 1} = ± 1 ∈ S 3 .

Needless to say that the constraints (A.9.33) and (A.9.34), which

are manifestations of the geometry and topology of the 3-sphere,
put strong restrictions on the possible values of the phase shifts
φpo , φqo , φro , and φso . One possible set of their values in degrees is
{ 0, 0, −86.92, and 37.99 }. With these values, the scalar parts of
A (a; eo , λ) and B(b; eo , λ) are given by

C(a; eo , λ, φpo , φqo ) = { + cos(ηaλeo + φpo )}/Na

and
C(b; eo , λ, φro , φso ) = { − cos(ηbλeo + φro )}/Nb ,
respectively, whereas the scalar part C(a, b; eo , φpo , φqo , φro , φso ) of
A B(a, b; eo ) is the one stated in equation (A.9.27).
As depicted in Fig. 9.3, (A.9.30), (A.9.31), and (A.9.32) are the
explicit expressions of the three points of the 3-sphere we have been
looking for. Using these expressions we can now compute the prob-
abilities for the occurrences of simultaneous events at the two ends of
the EPR experiment: P {A = +1, B = +1}, P {A = −1, B = −1},

250
P {A = −1, B = +1}, and P {A = +1, B = −1}. From the local-
ity conditions (A.9.33) and (A.9.34) it is easy to see, however, that in
the measurement limits these probabilities are causally constrained
by the respective values of C(a, b; eo , φpo , φqo , φro , φso ), as indicated in
Eqs. (A.9.23) to (A.9.26). Thus, for example, the probability of the
simultaneous occurrence of the results A = −1 and B = +1 is given
by the probability of the die C(a, b; eo , φpo , φqo , φro , φso ) landing on
its face marked −1. The results of numerical computations of such
probabilities are illustrated in Fig. 9.4. They turn out to be
1
P1+ (a) = P { Ca = +1 | no detector at b } = , (A.9.35)
2
1
P2− (b) = P { Cb = −1 | no detector at a } = , (A.9.36)
2
1 η 
++ ab
P12 = P { Cab = +1 | ηab } = sin2 , (A.9.37)
2 2
1  ηab 
−−
P12 = P { Cab = +1 | ηab } = sin2 , (A.9.38)
2 2
1  ηab 
−+
P12 = P { Cab = −1 | ηab } = cos2 , (A.9.39)
2 2
1  ηab 
+−
and P12 = P { Cab = −1 | ηab } = cos2 . (A.9.40)
2 2
As described in Eq. (A.9.5), the correlation predicted by our local
model can now be readily calculated as follows:
" n
#
1X i i
E(a, b) = lim A (a, λ ) B(b, λ )
n≫1 n
i=1
++ −− −+ +−
P12 + P12 − P12 − P12
= ++ −− −+ +−
P12 + P12 + P12 + P12
1  ηab  1  ηab  1 η  1 η 
ab ab
= sin2 + sin2 − cos2 − cos2
2  2 2  2 2 2 2 2
ηab ηab 
= sin2 − cos2
2 2
= − cos ηab . (A.9.41)

This correlation can also be computed directly in the simulation

described in Fig. 9.4. The result is depicted in Fig. 9.5.
Let me stress that the above correlation is produced entirely
within a strictly local and deterministic model of the physical reality.
As we have already noted several times, the measurement results
A (a; eo , λ) and B(b; eo , λ) defined in (A.9.30) and (A.9.31) are
manifestly local. Moreover, the product quaternions qab , and hence

251
their scalar limits A (a; b; eo , λ), are intrinsically factorizable, as
shown in Eqs. (A.9.33) and (A.9.34). The values of the functions
A (a; eo , λ) and B(b; eo , λ) are thus determined by a common cause;
namely, the geometry and topology of the 3-sphere—and the local
vectors a and b. In practical terms, this common cause is translated
into the set {λeo , φpo , φqo , φro , φso }. The actual measurement results
A and B observed by Alice and Bob are thus only dependent on
these common causes and the local vectors.
For completeness, let me also note that the above correlation
gives rise to the following string of expectation values:

E(a, b) + E(a, b′ ) + E(a′ , b) − E(a′ , b′ )

= − cos ηab − cos ηab′ − cos ηa′ b + cos ηa′ b′ .
(A.9.42)

It is well known that the RHS of the above equation violates the
celebrated Bell-CHSH inequality for certain choices of angles [10].

A.3.2 Computer Simulation also Confirms Local-Realistic

Violations of the Clauser-Horne Inequality

It is worth noting here that the local-realistic model described above

violates not only the Bell-CHSH inequality, but also the probabilistic
Clauser-Horne inequality3 [20]:
++ ++ ++ ′
−1 ≤ P12 (a, b) − P12 (a, b′ ) + P12 (a , b)
++ ′
+ P12 (a , b′ ) − P1+ (a′ ) − P2+ (b) ≤ 0 .
(A.9.43)

To appreciate this, recall the following predictions of our local model:

1 1 η 
ab
P1+ (a′ ) = P2+ (b) = and P12 ++
(a, b) = sin2 .
2 2 2
(A.9.44)
These predictions are the same as the ones stated in Eqs. (A.9.35)
to (A.9.37)—which, in turn, are the same as those predicted by
quantum mechanics. Consequently, Clauser-Horne inequality is also
inevitably violated by our local model.
It is also worth recalling here that once the topology of the
codomain of the measurement functions is correctly specified, not
only the EPR-Bohm correlation, but also the correlations predicted
3 I am grateful to Lucien Hardy for raising the question about the violations of
Clauser-Horne inequality in the present context.

252
by the rotationally non-invariant quantum states such as the GHZ
states and the Hardy state—and in fact those predicted by ALL
quantum states—can be reproduced exactly in a purely classical,
local-realistic manner [7][8][19]. Thus, contrary to the widespread
belief, the correlations exhibited by such states are not irreducible
quantum effects, but purely local-realistic, topological effects [3][6][8].
Needless to say, this vindicates Einstein’s suspicion that quantum
state merely describes statistical ensemble of physical systems, and
not the individual physical system. It is this inevitable conclusion
that Weatherall is resisting.

A.3.3 Another Explicit Simulation of the Model

As mentioned in the caption under Fig. 9.4, Michel Fodje has built
another explicit, event-by-event simulation of the model discussed
above that is worth elaborating on. While the simulation by Chantal
Roth is based on the joint probability density function |C(a, b; eo )|
defined in (A.9.27), the simulation by Michel Fodje is based on the
individual probability density functions |C(a; eo )| and |C(b; eo )|. We
saw in footnote 1 on pages 242 and 243 that the individual probability
density functions p(a; eo , λ) = | cos( ηaλeo )| satisfy the relation
Z  π2
1 1  1
| cos( ηaλeo )| dΩ = sin ( ηaeo ) =
2
(A.9.45)
4π S 2 2 0 2

with respect to any fixed vector a. This suggests that the constraints
1
| cos( ηaeo )| ≥ sin2 ( θo ) ≤ | cos( ηbeo )| (A.9.46)
2
for arbitrary angles ηaeo ∈ [0, 2π) and ηbeo ∈ [0, 2π) should play a
crucial role in dictating the strength of the correlation between the
results A (a; eo , θo ) and B(b; eo , θo ) for a given angle θo ∈ [0, π/2].
According to this constraint the probability densities | cos( ηaeo )| and
| cos( ηbeo )| for observing the measurement results A (a; eo , θo ) and
B(b; eo , θo ) depend on the common angle θo , just as they depend on
the common vector eo [9]. This in turn suggests that we may treat
θo ∈ [0, π/2] as an additional random parameter, and take the set
  
 1
Λ := (eo , θo )  | cos( ηxeo )| ≥ sin2 ( θo ) ∀ x ∈ IR3 (A.9.47)
2

as a set of initial or complete states of our physical system [9]. Given

one such state, the outcomes of measurements are deterministically
determined by the topological constraints within the 3-sphere. From

253
Figure 9.7: Another explicit simulation of the correlation between
simultaneously occurring measurement events A = ±1 and B = ±1
within a parallelized 3-sphere. The code for this simulation is written
by Michel Fodje, in Python. Along with other relevant information,
it can be downloaded from https://github.com/minkwe/epr-simple/.

a geometrical point of view, the parameter θo links two disconnected

“sections” of S 3 (i.e., two “orthogonal” 2-spheres within S 3 ) defined
by the bivectors I · a and I · b, by means of the constraints (A.9.46).
With these considerations, we define the measurement functions
as

A (a; eo , θo ) = sign{ − cos( ηaeo )}, for a given θo ∈ [0, π/2] ,

(A.9.48)
and

B(b; eo , θo ) = sign{ + cos( ηbeo )}, for a given θo ∈ [0, π/2] ,

(A.9.49)
where the vectors a and b are specific instances of the vector x. For
θo = 0 this prescription reduces to that of Bell’s own local model [9].
Once again let me stress the obvious that these functions define
manifestly local measurement results. What is more, given the initial
state (eo , θo ), the local outcomes A (a; eo , θo ) and B(b; eo , θo ) are
deterministically and ontologically determined to be either +1 or −1,

254
for any freely chosen vectors a and b, where eo is a random vector
on S 2 as defined before, and θo is a random angle, chosen from the
interval [0, π/2]. The correlation is then calculated quite simply as
" n
#
1X
E(a, b) = lim A (a; eio , θoi ) B(b; eio , θoi ) = − a · b .
n≫1 n
i=1
(A.9.50)

For the measurement functions defined in (A.9.48) and (A.9.49),

the probabilities of observing the specific outcomes +1 or −1 turn
out to be exactly 1/2, with 100% detector efficiency. In other words,
every particle that emerges in a state (eo , θo ) ends up being detected
by the detector, just as in the previous simulation. On the other
hand, the probabilities of jointly observing the results A (a; eo , θo )
and B(b; eo , θo ) turn out to be exactly those predicted by quantum
mechanics [cf. Eqs. (A.9.37) to (A.9.40)]. Consequently, not only the
correlations between the results turn out to be those predicted by
quantum mechanics [cf. Eq. (A.9.41)], but also the Clauser-Horne
inequality is necessarily violated in this simulation [cf. Eq. (A.9.44)].
It is important to recognize that the strength of the correlation
exhibited in this simulation does not stem from exploiting any known
or unknown loopholes—such as the detection loophole. Needless to
say, a measurement event cannot occur if there does not exist a state
that can bring about that event. If there are no clouds in the sky in
the first place, then there can be no rain, no matter where one goes.
As we noted, the state of the spin system is specified by the pair
(eo , θo ) defined by the set (A.9.47)—not just by eo . In other words,
the initial distribution of the physical states is defined by the set
  
 1
Λ := (eo , θo )  | cos( ηxeo )| ≥ sin2 ( θo ) ∀ x ∈ IR3 , (A.9.51)
2
which reduces to that of Bell’s local model [9] for θo = 0 = constant.
Accordingly, since there are no states of the physical system for which
| cos( ηxeo )| < 12 sin2 ( θo ) holds true for any x, a measurement event
cannot possibly occur for | cos( ηxeo )| < 12 sin2 ( θo ), no matter what
x is [9]. If x happens to be equal to b, for example, then there is no
reason for the detector at b to click when | cos( ηbeo )| < 12 sin2 ( θo ),
even for the non-vanishing random angles θo in the interval [0, π/2].
Independently of this physical picture, it is also instructive to
view the simulation purely from the perspective of computability.
From this perspective the correlation is calculated as follows: Alice
freely chooses a vector a on S 2 . She is then given four scalar numbers,
(θo , exo , eyo , ezo ) , (A.9.52)

255
t

A = ±1 B = ±1
S3
a b

(eo , θo )

Figure 9.8: The measurement results A (a; eo , θo ) and B(b; eo , θo )

are deterministically brought about by the common cause (eo , θo ).

or a pair (eo , θo ), represented by a randomly chosen vector eo on S 2

and a randomly chosen scalar θo from the interval [0, π/2]. Similarly,
Bob freely chooses a vector b on S 2 , and is also given the same four
scalar numbers (eo , θo ). Using these scalars Alice and Bob compute
the numbers A (a; eo , θo ) and B(b; eo , θo ), respectively, as follows:
(
sign{− cos( ηaeo )} if | cos( ηaeo )| ≥ 21 sin2 ( θo )
A (a; eo , θo ) =
0 if | cos( ηaeo )| < 21 sin2 ( θo )
(A.9.53)
and
(
sign{+ cos( ηbeo )} if | cos( ηbeo )| ≥ 1
2 sin2 ( θo )
B(b; eo , θo ) =
0 if | cos( ηbeo )| < 1
2 sin2 ( θo ).
(A.9.54)

This prescription simply partitions the outcomes A (a; eo , θo ) and

B(b; eo , θo ) into three sets of numbers, +1, −1, and 0, which are
then summed over the full ranges of the possible vectors eo and
possible scalars θo , giving the vanishing local averages:
" n
# " n
#
1X i i 1X i i
lim A (a; eo , θo ) = 0 = lim B(b; eo , θo ) ,
n≫1 n n≫1 n
i=1 i=1
(A.9.55)
where n is the total number of non-vanishing outcomes in the sets.
Alice and Bob then multiply the outcomes A (a; eo , θo ) and
B(b; eo , θo ) for each pair (eo , θo ), add all of the products together,

256
and divide the sum by the total number of non-vanishing products
(i.e., “coincidences”) they have added. The result is the correlation
" n
#
1X
E(a, b) = lim A (a; eio , θoi ) B(b; eio , θoi ) = −a · b.
n≫1 n i=1
(A.9.56)

Thus, regardless of the interpretation given to numbers A (a; eo , θo )

and B(b; eo , θo ), we have arrived at a refutation of Bell’s theorem4 .

A.3.4 Exploring the World Beyond the Quantum World

It turns out that the simulation discussed above can be generalized

to generate correlations of any strength—from the weakest possible
(Bell’s model) to the strongest possible (the box model), provided
the distribution of the complete states is generalized from (A.9.51) to
  

Λ := (eo , θo , lo )  | cos( ηxeo )| ≥ lo sin2 ( θo ) ∀ x ∈ IR3 ,
(A.9.57)
with the scalar lo ∈ [0, 1] being an additional, non-random common
cause [cf. Fig. 9.9]. The two measurement functions are then given by

A (a; eo , θo , lo ) = sign{ − cos( ηaeo )}, for a given pair (θo , lo ) ,

(A.9.58)
and

B(b; eo , θo , lo ) = sign{ + cos( ηbeo )}, for a given pair (θo , lo ) ,

(A.9.59)
where the freely chosen vectors a and b are specific instances of the
vector x, just as before. The correlation between the measurement
results A (a; eo , θo , lo ) and B(b; eo , θo , lo ) can then be determined
as usual by computing the expectation value of their scalar product:
" n
#
1X i i i i i i
E(a, b) = lim A (a; eo , θo , lo ) B(b; eo , θo , lo ) . (A.9.60)
n≫1 n i=1

4 The original simulation by Michel Fodje confirming the above results has been
translated by John Reed from Python to Mathematica. It can be found in PDF
format at this page: http://libertesphilosophica.info/Minkwe Sim J Reed.pdf.

257
t

A = ±1 B = ±1
S3
a b

(eo , θo , lo )

Figure 9.9: The measurement results A (a; eo , θo ) and B(b; eo , θo )

are deterministically brought about by the initial state (eo , θo , lo ).

Special Case (1): Weakest Possible Correlation—Bell’s Model:

If we now fix either θo = 0 or lo = 0 (or both = 0) in the distribution

(A.9.57), for all eo , it reduces to that of Bell’s own local model [9],
  

Λ = (eo , θo = 0 or lo = 0)  | cos( ηxeo )| ≥ 0 ∀ x ∈ IR3 ,
(A.9.61)
generating the weakest possible (or “classical”) correlation:

 2
− 1 + π ηab if 0 ≤ ηab ≤ π
E(a, b) = (A.9.62)

 2
+ 3 − π ηab if π ≤ ηab ≤ 2π .

Special Case (2): Strongest Possible Correlation—the Box Model:

On the other hand, if we fix θo = π/2 and lo = 1 in the distribution

(A.9.57), again for all eo , it reduces to that of the “box” model,
  


Λ = (eo , θo = π/2, lo = 1)  | cos( ηxeo )| ≥ 1 ∀ x ∈ IR , 3

(A.9.63)
generating the strongest possible (albeit unphysical) correlation:


− 1 if 0 ≤ ηab < π/2 or 3π/2 < ηab ≤ 2π
E(a, b) =


+ 1 if π/2 < ηab < 3π/2 .
(A.9.64)

258
Special Case (3): “Quantum” Correlation—the 3-Sphere Model:

Finally, if we set lo = 1/2 in the distribution (A.9.57), again for all

eo , but keep θo random, then we are led back to the distribution
   
1  1 2 3
Λ = eo , θo , lo = | cos( ηxeo )| ≥ sin ( θo ) ∀ x ∈ IR ,
2  2
(A.9.65)
generating the “quantum” correlation discussed above. Thus, if we
view lo = 1/2 as the average of the extremes lo = 0 and lo = 1, then
the key feature that generates precisely the quantum correlation is
the randomness of θo , channeled through the geometrical constraint
1
| cos( ηxeo )| ≥ sin2 ( θo ) ∀ x ∈ IR3 . (A.9.66)
2

A.3.5 Elegant, Powerful, and Succinct Calculation of the

Correlation

The above simulations once again confirm the fact that EPR-Bohm
correlations are local-realistic correlations among the binary points of
a parallelized 3-sphere [6]. As we saw in section 9.2, however, this fact
can be expressed more elegantly by understanding how random errors
propagate within a parallelized 3-sphere. In particular, we saw that
EPR-Bohm correlations can be derived by recognizing that the raw
scores A (a, λ) and B(b, λ) are generated within S 3 with different
bivectorial scales of dispersion, and hence the correct correlation
between them can be determined only by calculating the covariation
of the corresponding standardized variables L(a, λ) and L(b, λ):
" n
#
1X i i
E(a, b) = lim A (a, λ ) B(b, λ )
n≫1 n
i=1
" n
#
1X i i
= lim L(a, λ )L(b, λ )
n≫1 n
i=1
= −a · b, (A.9.67)
where
L(a, λi ) L(b, λi ) ≡ − a · b − L(a × b, λi )
≡ − a · b − λi D(a × b), (A.9.68)
and the standardized variables are defined as
q(ψ, a, λ) − m(q) A (a, λ) − m(A )
L(a, λ) := = . (A.9.69)
σ[ q(ψ, a, λ)] σ[ A (a, λ)]

259
Note that this definition holds for any point q(ψ, a, λ) of S 3 , and
not just for the limiting point q(ψ, a, λ) = A (a, λ) = ±1 obtained
in equation (9.21). The EPR-Bohm correlations − a · b are thus
correlations between any two points q(ψ, a, λ) and q(ψ, b, λ) of
S 3 , with the scalar points A (a, λ) = ±1 and B(b, λ) = ±1 being
only a special case.
It is also noteworthy that the correlation between the raw scores
A (a, λ) and B(b, λ) is determined in Eq. (A.9.67) by calculating
their covariance divided by the product of their standard deviations
σ[ A (a, λ) ] and σ[ B(b, λ) ]:

σ[ A (a, λ) ] σ[ B(b, λ) ] = − D(a)D(b)

= (−I · a)(+I · b)
= ab
= a · b + a ∧ b. (A.9.70)

This product, however, is precisely the rotor that quantifies the twist
in the Hopf fibration of the 3-sphere [3][6]. As discussed in detail in
Chapter 8, its value varies from +1 for b = a to −1 for b = − a and
back, producing the correct combination of observed probabilities.

A.3.6 What Can We Learn from Weatherall’s Analysis?

It is evident from the above results that there is a valuable lesson

to be learned from Weatherall’s analysis: An explicit, constructive,
quantitatively precise physical model X cannot be undermined by
repudiating its distorted misrepresentation Y, even by appealing
to a formal theorem (especially when that theorem is grounded on
unphysical assumptions). Such a strategy only serves to exemplify
an elementary logical fallacy—namely, the straw-man fallacy.

Acknowledgments
I am grateful to Lucien Hardy for several months of correspondence
which led to improvements in section 9.2.4. I am also grateful to
Martin Castell for his hospitality in the Materials Department of
the University of Oxford. This work was funded by a grant from
the Foundational Questions Institute (FQXi) Fund, a donor advised
fund of the Silicon Valley Community Foundation on the basis of
proposal FQXi-MGA-1215 to the Foundational Questions Institute.

260
References
[1] J. O. Weatherall, The Scope and Generality of Bell’s Theorem,
arXiv:1212.4854 (this preprint is now published).
[2] J. Christian, Disproof of Bell’s Theorem, arXiv:1103.1879. See
also http://libertesphilosophica.info/blog/.
[3] J. Christian, Disproof of Bell’s Theorem: Illuminating the Illu-
sion of Entanglement (BrownWalker Press, Florida, 2012).
[4] J. Christian, Refutation of Some Arguments Against my Disproof
of Bell’s Theorem, arXiv:1110.5876.
[5] J. Christian, Refutation of Richard Gill’s Argument Against my
Disproof of Bell’s Theorem, arXiv:1203.2529.
[6] J. Christian, Macroscopic Observability of Spinorial Sign
Changes under 2π Rotations, arXiv:1211.0784.
[7] J. Christian, What Really Sets the Upper Bound on Quantum
Correlations?, arXiv:1101.1958.
[8] J. Christian, On the Origins of Quantum Correlations,
arxiv:1201.0775. See also http://libertesphilosophica.info/blog/.
[9] J. S. Bell, Physics 1, 195 (1964); see also Rev. Mod. Phys. 38,
447 (1966), and Dialectica 39, 86 (1985).
[10] J. F. Clauser, M. A. Horne, A. Shimony, and R. A. Holt, Phys.
Rev. Lett. 23, 880 (1969).
[11] A. Einstein, B. Podolsky, and N. Rosen, Phys. Rev. 47, 777
(1935).
[12] C. Doran and A. Lasenby, Geometric Algebra for Physicists
(Cambridge University Press, Cambridge, 2003).
[13] D. Hestenes, Am. J. Phys. 71, 104 (2003). For more general
discussion see also http://libertesphilosophica.info/blog/.
[14] J. W. Milnor, Topology from the Differentiable Viewpoint
(Princeton University Press, Princeton, NJ, 1965).
[15] J. L. Rodgers and W. A. Nicewander, The American Statistician
42, 59 (1988).
[16] T. Frankel, The Geometry of Physics: An Introduction (Cam-
bridge University Press, 1997), p 501.
[17] B. S. Cirel’son, Lett. Math. Phys. 4, 93 (1980); L. J. Landau,
Phys. Lett. A 120, 54 (1987).
[18] A. Aspect et al., Phys. Rev. Lett. 49, 91 (1982); G. Weihs et
al., Phys. Rev. Lett. 81, 5039 (1998).
[19] J. Christian, Disproofs of Bell, GHZ, and Hardy Type Theorems
and the Illusion of Entanglement, arXiv:0904.4259.
[20] J. F. Clauser and A. Shimony, Rep. Prog. Phys. 41, 1881 (1978).

261
 Disproof of Bell’s Theorem:
Illuminating the Illusion of Entanglement

Copyright c 2012, 2014 Joy Christian

All rights reserved.
No part of this book may be reproduced or transmitted in any form
or by any means, electronic or mechanical, including photocopying,
recording, or by any information storage and retrieval system,
without written permission from the publisher.

BrownWalker Press
Boca Raton, Florida
USA · 2014

First published 2012

Second edition 2014

ISBN-10: 1-61233-724-4 / ISBN-13: 978-1-61233-724-1

www.brownwalker.com

Cover design by Shereen Siddiqui

Cover photo @Cutcaster.com/Michael Brown

Library of Congress Cataloging-in-Publication Data

Christian, Joy, author.

Disproof of Bell’s theorem : illuminating the illusion of
entanglement / Joy Christian. – Second edition.
pages cm
Includes bibliographical references and index.
ISBN 978-1-61233-724-1 (pbk. : alk. paper) – ISBN 1-61233-724-4
(pbk. : alk. paper)
1. Bell’s theorem. 2. Quantum theory. I. Title.
QC174.17.B45C47 2014
530.12–dc23
2013040705

View publication stats