Eduardo Reck Miranda Editor

Quantum Computing
in the Arts and
Humanities
An Introduction to Core Concepts,
Theory and Applications
Quantum Computing in the Arts and Humanities
Eduardo Reck Miranda
Editor

Quantum Computing
in the Arts and Humanities
An Introduction to Core Concepts, Theory
and Applications
Editor
Prof. Eduardo Reck Miranda
Interdisciplinary Centre for Computer
Music Research (ICCMR)
University of Plymouth
Plymouth, UK

ISBN 978-3-030-95537-3 ISBN 978-3-030-95538-0 (eBook)

https://doi.org/10.1007/978-3-030-95538-0

© Springer Nature Switzerland AG 2022

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface

Quantum Computing Beyond Science

Publications reporting research into quantum information processing started to

emerge around the middle of the 1970s; e.g. Holevo (1973), Poplavskii (1975) and
Ingarden (1976), to cite but three. Then, groundbreaking work by Feynman (1982)
glimpsed at the possibility of developing a quantum computing device: Feynman
proposed a mathematical quantum mechanical system capable of performing compu-
tations. This possibility commenced to break free from academic circles into the
realm of industry after Deutsch (1985) proposed an abstract Turing-like quantum
machine capable of universal computation.
The Turing machine is a widely accepted model of computation, which informed
the development of commercial classical computers, as we know them today. With
this in mind, the industry and investors gained confidence that it might be possible
to harness quantum mechanics to build a somewhat different type of computer. That
is, a computer that might be able to efficiently tackle problems that are not easily
tractable today, such as predicting how biological molecules interact and making
sense of colossal amounts of information. We will need to tackle these problems
more efficiently than ever to overcome contemporary and future challenges. These
include, for instance, climate monitoring and the development of vaccines to combat
new pathogens. Deutsch’s machine also paved the way for the development of what
is known as the ‘circuit model’ for programming quantum computers, which is the
norm nowadays.
Since then, large consumer electronics companies and many start-up ones have
made tremendous progress in the race to build quantum computers, in particular
in the last decade. But still, there is some leg to go before these machines become
mainstream, if at all. Nevertheless, research and development are progressing fast in
universities and companies all over the world.
Despite the hype surrounding this new computing technology, sceptics question
the benefits of quantum computers over classical ones. What will a quantum computer

v
vi Preface

be able to do that a standard computer would not be capable of? Today’s high-
performance supercomputers are rather powerful. And they will continue to improve.
Consider the Fugaku supercomputer, developed by Fujitsu and RIKEN in Japan: it
performs circa 600 quadrillion operations per second. To put this in perspective, a
top of the range desktop computer does approximately 100 billion operations per
second. That is, Fugaku is six million times faster than a decent desktop computer.
All the same, quantum computing is here to stay. What started as a desire to build
a machine for physicists to study quantum mechanics1 has been evolving towards
the ambition to develop general-purpose quantum computing technology for a wide
range of applications, some of which we have not even dreamed of yet. And more,
I advocate that processing speed is not the only benefit that such machines will
bring to society. They will bring new philosophies, new approaches to design, new
algorithms, new creative artefacts, new economies and so on. In the long run, using
an actual quantum computer for mundane applications may as well become a matter
of choice or luxury rather than necessity. Who knows?
Much ongoing research into quantum computing worldwide focuses on improving
hardware and developing solutions for science and technology. Nevertheless, there
also is growing activity within areas where quantum computing may impact beyond
science and technology, including the arts, humanities and fringe interdisciplinary
initiatives. These are exciting because they might influence developments in the
quantum computing industry; e.g., create unforeseen new markets.
This book brings a collection of chapters by scholars championing research in
their respective areas (such as Philosophy, Linguistics and Music, amongst others)
informed by quantum computing. The authors offer thought-provoking discussions
about cognition, perception, language, music, games, visualization of the sub-atomic
world, brain-machine communication and more.
I am indebted to all contributing authors who enthusiastically supported this book
project. They made possible what seemed impossible. We have here what is probably
the first book ever on Quantum Computing in the Arts and Humanities.

Plymouth, UK Eduardo Reck Miranda

October 2021

References
Deutsch, D. (1985). Quantum theory, the church-turing principle and the universal quantum
computer. Proceedings of the Royal Society of London A, 400, 97–117.
Feynman, R. P. (1982). Simulating physics with computers. International Journal of Theoretical
Physics, 21, 467–88.

1 The impetus to develop a quantum computer stemmed from the fact that some simulations to
study quantum mechanics would allegedly require computing power that even state of the art
supercomputers might take years to run. The idea then is to harness quantum mechanics to build
quantum computers rather than simulate quantum mechanics on standard computers.
Preface vii

Holevo, A. S. (1973). Bounds for the quantity of information transmitted by a quantum commu-
nication channel, Problemy Peredachi Informatsii, 9, 3–11. English translation in Problems of
Information Transmission, 9, 177–83.
Ingarden, R. S. (1976). Quantum information theory. Reports on Mathematical Physics, 10, 43–72.
Poplavskii, R. P. (1975). Thermodynamical models of information processing (in Russian), Uspekhi
Fizicheskikh Nauk, 115, 465–501.
Contents

From Digital Humanities to Quantum Humanities: Potentials

and Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
Johanna Barzen
Quantum Computing and Cognitive Simulation ..................... 53
Martha Lewis
The Philosophy of Quantum Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Michael E. Cuffaro
Quantum-Like Cognition and Rationality: Biological and Artificial
Intelligence Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
Andrei Khrennikov
Quantum Music, Quantum Arts and Their Perception . . . . . . . . . . . . . . . . 179
Volkmar Putz and Karl Svozil
Quanta in Sound, the Sound of Quanta: A Voice-Informed
Quantum Theoretical Perspective on Sound . . . . . . . . . . . . . . . . . . . . . . . . . . 193
Maria Mannone and Davide Rocchesso
Quantum Cinema and Quantum Computing . . . . . . . . . . . . . . . . . . . . . . . . . 227
Renate C.-Z. Quehenberger
How to Make Qubits Speak . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
Bob Coecke, Giovanni de Felice, Konstantinos Meichanetzidis,
and Alexis Toumi
A Quantum Computing Approach to Harnessing the Logic
of the Mind for Brain–Computer Interfacing . . . . . . . . . . . . . . . . . . . . . . . . . 299
Eduardo Reck Miranda
The History of Games for (Quantum) Computers . . . . . . . . . . . . . . . . . . . . . 345
James R. Wootton

ix
From Digital Humanities to Quantum
Humanities: Potentials and Applications

Johanna Barzen

Abstract Quantum computers are becoming real. Therefore, it is promising to use

their potentials in different application areas, which includes research in the human-
ities. Due to an increasing amount of data that needs to be processed in the digital
humanities the use of quantum computers can contribute to this research area. To
give an impression on how beneficial such involvement of quantum computers can
be when analysing data from the humanities, a use case from the media science is
presented. Therefore, both the theoretical basis and the tooling support for analysing
the data from our digital humanities project MUSE is described. This includes
a data analysis pipeline, containing e.g. various approaches for data preparation,
feature engineering, clustering, and classification where several steps can be realized
classically, but also supported by quantum computers.

Keywords Quantum computing · Quantum humanities · Machine learning ·

Quantum machine learning · Digital humanities · Data analysis · Artificial neural
networks · Pattern languages

1 Introduction

As quantum computers are becoming real there are several application areas where
the use of quantum computers is expected to be especially promising (National
Academies of Sciences, Engineering, and Medicine, 2019): one is simulation,
focusing on molecular properties used in material science or in pharma industry, and
the other is machine learning e.g. classification, clustering or dimension reduction
making heavy use of optimization.
The main difference between classical computers and quantum computers is the
different information unit used, namely bits for classical computers and qubits for
quantum computers (Nielsen & Chuang, 2010). Qubits are not restricted to the two
states 0 and 1 but can be in an infinite number of different combinations of these

J. Barzen (B)
Institute of Architecture of Application Systems (IAAS), University of Stuttgart, Stuttgart,
Germany
e-mail: [email protected]

© Springer Nature Switzerland AG 2022 1

E. R. Miranda (ed.), Quantum Computing in the Arts and Humanities,
https://doi.org/10.1007/978-3-030-95538-0_1
2 J. Barzen

states, referred to as superposition. Furthermore, individual qubits can be combined

into a quantum register resulting in exponentially growing number of data that such
a register can hold. As an example, a quantum register with 50 qubits corresponds
to 250 possible combinations of the states of the individual qubits which are again in
superposition corresponding to approximately a petabyte of classical data. All these
values can be manipulated by a single operation all at once, which is called quantum
parallelism.
Operations on single qubits or quantum registers, as a combination of qubits, are
unitary transformations. They can be used to process the data to achieve every-
thing classical computers can perform (Rieffel & Polak, 2011) but to also use
quantum inherent phenomena, e.g. entanglement, to—among other potentials—
solve a variety of problems that were practically not solvable before, some even
with exponential speedup (Nielsen & Chuang, 2010; Preskill, 2018). There are
different quantum computing models, e.g., gate-based (Michielsen et al., 2017),
measurement-based (Jozsa, 2006), and adiabatic quantum computing (Aharonov
et al., 2008), which represent quantum algorithms in various ways. As it can be
shown that the different models are formally equivalent (Aharonov et al., 2008;
Jozsa, 2006) this chapter restricts the considerations and implementations to the gate-
based quantum computing model. Also, quantum computers based on the gate-based
quantum computing model are universal quantum computers and most commercially
available quantum computers are based on this model (LaRose, 2019).
As different vendors, such as IBM or Rigetti, developed quantum computers in
recent years and offer access to them via the cloud (LaRose, 2019; Leymann et al.,
2020) quantum computing is becoming a lively research field and first real word
applications are developed in industry. When taking a closer look at the research
areas quantum computing is applied to, there are mainly applications in the natural
sciences that can be found, for example general approaches like (Bhaskar et al., 2015),
presenting quantum algorithms and circuits for modules implementing fundamental
functions (e.g. the square root, the natural logarithm, and arbitrary fractional powers)
or more specific approaches as in molecular simulation in the material sciences
(Kandala et al., 2017; McClean et al., 2017) or in artificial intelligence and machine
learning (Dunjko et al., 2016; Havlicek et al., 2018; Otterbach et al., 2017). Never-
theless, first applications can be identified in the humanities too, e.g. quantum social
science as interdisciplinary field which draws parallels between quantum physics and
the social sciences (Haven & Khrennikov, 2013), but this is still to be called rudi-
mentary in respect to the potential quantum computers can have in different areas of
the humanities.
The chapter is structured as follows: Sect. 2 presents the vision for quantum
humanities. Therefore, the potential benefits that can be expected but also the current
challenges are outlined. Section 3 introduces the digital humanities project MUSE
as use case for quantum humanities. As the use case strongly relies on analysing
data Sect. 4 focuses on introducing our data analysis pipeline, how to cope with cate-
gorical data, and how pattern languages based on data analysis can be detected. A
core concept of data analysis is artificial neural networks. Therefore, Sect. 5 provides
a mathematical definition of neurons, neuronal networks, and perceptrons, outlines
From Digital Humanities to Quantum Humanities … 3

how they are used for restricted Boltzmann machines and autoencoders, and sketches
their realizations on quantum computers. As many quantum algorithms are hybrid,
meaning they consist of both, a quantum part and a classical part, Sect. 6 focuses on
variational hybrid quantum–classical algorithms. The main idea is outlined as well
as an application for clustering, namely a quantum version for solving the maximum
cut problem. Therefore, techniques as quantum approximate optimization algorithm
(QAOA) and variational quantum eigensolver (VQE) are discussed. To support access
to the described techniques, Sect. 7 focuses on our quantum humanities analysis tool
(QHAna) that is used for pattern detection. In addition to the substantive goal of
the use case from the MUSE project to extract costume patterns in films, QHAna
provides easy access for people without quantum computing background to evaluate
the benefits of using quantum algorithms and allows them to gain initial application
knowledge in the field of quantum humanities. Since both classical analysis algo-
rithms and their quantum counterparts are included, the comparison of their results
allows a deeper understanding within quantum humanities. Section 8 concludes the
chapter and gives an outlook on future work in quantum humanities.
How to read this chapter: As quantum humanities brings together very heteroge-
nous disciplines (e.g. the respective humanities discipline, physics, mathematics,
computer science) and each has its own approaches, terminology, concepts and
research culture, a fundamental challenge of quantum humanities is to find a common
language to communicate between these disciplines. Thus, the chapter starts with an
easy-to-understand introduction to the overall topic, the vision and potentials that
does not require any previous knowledge (Sects. 1–4). The underlying ideas and
concepts get more and more refined, up to mathematical definitions (Sects. 5–6). We
use mathematical definitions as they give little room for interpretation and therefore
seems appropriate as long as there is no common language yet. As the mathematical
definitions may be challenging to some readers they are accomplished by descriptions
and pictures giving a high-level view on the core concepts, so some readers may skip
over the mathematical formalisms. Please note that algorithmic details themselves
are not the focus. Instead, the underlying concepts of the (classical and quantum)
algorithms are provided that are relevant for our approach, as well as the tooling
provided by QHAna. QHAna than outlines the practical usage of the introduced
concepts in a use case centric manner (Sect. 7).

2 Towards Quantum Humanities

How beneficial the use of computers, their practical support as well as techniques
and methods from computer science can contribute to research in the humanities has
been proven by the establishment of the digital humanities. With quantum computers
becoming available, it is promising to extend the usage of classical computers as
done in the digital humanities by the use of quantum computers. As stressed in the
introduction, working with quantum computers, programing algorithms and letting
them run on quantum hardware is very different to working with classical computers.
4 J. Barzen

Therefore, we coined the term quantum humanities to describe addressing research

in the humanities with methods and techniques from quantum computing (Barzen &
Leymann, 2019; QH, 2021). As the basis for quantum humanities the following
section will outline the core components defining digital humanities.

2.1 Digital Humanities

When speaking about digital humanities the term combines a broad variety of topics,
contents and contexts. This is reflected by the continuously growing amount of
literature on divergent approaches in this domain (Berry, 2012; Berry & Fagerjord,
2017; Brown, 2020; Jannidis et al., 2017; Terras et al., 2013). But there are three
core elements that can be identified when taking a closer look at approaches that
define digital humanities: (i) digital humanities is about bringing together humani-
ties research and information technology, (ii) it is a diverse field, and (iii) it is a still
emerging field (DHQ, 2021).
The first of those highlighted three core elements stresses the combination of
computer science and the humanities research. Digital humanities therefore are a
highly inter- and trans-disciplinary field operating between the traditional humanities
and the (in comparison) rather young computer science. The aim is to examine
existing and new questions from the humanities by using concepts, methods and
tools from computer science. Therefore, digital humanities is understood as a field
of methods that unites different topics, approaches and sub-disciplines and as such
fosters synergies between different scientific disciplines.
This is rather complex and hints to core element two: digital humanities are a
diverse field. ‘Humanities’ already encompasses a large variety of disciplines, such
as philosophy, linguistics, arts, cultural studies, archaeology, musicology and media
science, to name just a few. It gets even broader when taking into account that also
“computer science” covers quite different subject areas like databases, visualization,
or theoretical computer science. In the compilation of so many fields of interest,
digital humanities is a very heterogeneous area. It attempts to include the application
as well as the further development of tools and methods of computer science in
order to possibly obtain new and profound answers to research questions in the
humanities while performing a critical self-reflection of the methodological and
theoretical approaches.
Even though discussions on digital humanities are around for about 60 years and
core areas such as computational linguistics have established themselves through
their own courses of study, digital humanities is still a young field of research that
has only recently become an independent discipline (Rehbein & Sahle, 2013). This is
the third core element of many definitions of digital humanities: Digital humanities
is a young field whose own definition is constantly being developed (What is Digital
Humanities? 2021).
Nevertheless, a broad variety of research programs and projects, conferences
and competence centers, stress the benefits of using computers as tools as well as
From Digital Humanities to Quantum Humanities … 5

methods from computer science to support research in the humanities. Techniques

like databases, visualization, or distant reading have shown how the use of the clas-
sical computer can contribute to access knowledge and to broaden the approaches
for addressing already stated problems. Now, with quantum computers there is a
new type of computer just becoming reality and it allows to further improve the way
in which computers can support research in the humanities. Combining the power
quantum computer promises with research done in the humanities—based on the
achievements in digital humanities research—is what we call quantum humanities
(Barzen & Leymann, 2019).

2.2 Potential Benefits of Quantum Humanities

As quantum computers are superior to classical computer in various areas, their

use has a high potential contributing to research in the humanities. Especially, since
quantum computers are becoming more easily accessible—for example via the cloud
(Leymann et al., 2020)—their promising potentials are more and more applicable for
first ‘proof of concept’ application as outlined in Barzen and Leymann (2020) and
will be detailed in Sect. 7. In the following, seven potentials (Barzen & Leymann,
2019) will be outlined, which promise to be particularly interesting for supporting
digital humanities:
(I): The core benefit of quantum computers is an often stressed potential speedup
in several problem areas. Meaning, quantum computers solve certain types
of problems much faster (Rønnow et al., 2014) than classical computers (e.g.
in the decoding of passwords (Shor, 1995), the determination of global prop-
erties of Boolean functions (Deutsch, 1985) or unstructured search (Grover,
1996)). As the amount of data to be processed in the digital humanities is
continuously growing, the rapid evaluation of data is becoming increasingly
important. Since many algorithms developed for quantum computers can
make certain statements about global properties of functions much faster
than algorithms developed for classical computers, an enormous time saving
can be assumed here. In this context, time saving does not only mean a mere
gain in speed of the proper algorithms, but methods of dimensional reduc-
tion, for example, that reduce the amount of data to be processed by the
proper algorithms, can play a major role here—and the latter can also be
performed much more efficiently by quantum computers (e.g. by quantum
principal component analysis).
(II): Quantum computers can process large amounts of data in a single step
(Nielsen & Chuang, 2010). Due to the superposition of all possible states of
the quantum mechanical system, the quantum computer allows true paral-
lelism in the calculation (manipulations), which leads to a significantly
higher computing power. This in turn means that complex computational
6 J. Barzen

problems, such as those increasingly found in the digital humanities, can be

dealt with much more effectively.
(III): The results of quantum computers promise much higher precision in certain
areas than those of classical computers (Havlicek et al., 2018; Sierra-Sosa
et al., 2020). As the use of quantitative methods more and more leads to the
evaluation of data by means of data analytics, for example techniques from
the field of machine learning such as clustering and classification, the quality
of the results in terms of precision is becoming increasingly important.
Various new approaches, for example for support vector machines (SVM)
(Havenstein et al., 2018; Havlicek et al., 2018), or for variational quantum
classifier (VQC) (Sierra-Sosa et al., 2020), offer different promising fields
of application in digital humanities.
(IV): In addition, the use of quantum computers makes it possible to solve
problem classes that were previously considered practically unsolvable
(e.g. complexity class BQP (Nielsen & Chuang, 2010)). Hereby quantum
computers enable approximate solutions for otherwise algorithmically inac-
cessible problem classes. To what extent and for which fields of applications
in the digital humanities this can be beneficially used is still to be explored,
but it opens up a whole field of potential questions whose solutions now
deem to be possible. A fist promising use case to stress this potential is intro-
duced by (Palmer, 2020). Here natural language processing is performed on
a quantum computer using the power of quantum states to code complex
linguistic structures and novel models of meaning in quantum circuits to
understands both, grammatical structure and the meaning of words.
(V): Finally, there are certain types of problems that can only be solved by a
quantum computer, i.e. these types of problems can be proven not to be
solvable at all by a classical computer (Raz & Tal, 2018). Being able to solve
new types of problems, the digital humanities have the chance to work on
corresponding questions that have not yet been tackled at all, perhaps not
even identified or considered. To identify and explore these possible new
application areas is an exciting and promising task.
(VI): The use of a quantum computer promises to be significantly cheaper than that
of a conventional supercomputer: a price of about 200e per hour compute
time on a quantum computer can be assumed (Dickel, 2018). This can
be of benefit for the often financially strained humanities, especially for
the feasibility of smaller research projects by providing low-cost access to
high-performance hardware.
(VII): Furthermore, quantum computers are much more energy efficient than
computers with comparable performance (Nielsen & Chuang, 2010).
Without having a direct influence on digital humanities, the indirect influ-
ence should certainly be mentioned. Due to their energy efficiency quantum
computers contain a great potential for change, especially in times of climate
change.
From Digital Humanities to Quantum Humanities … 7

The above potentials reveal that it is of great importance to develop application

knowledge in the domain of quantum computing at an early stage, especially in a
field where the mathematical and physical basics required to use this technology
cannot be taken for granted.

2.3 Current Challenges of Quantum Humanities

The description of the potentials outlined above are from an abstract point of view
and must be considered much more comprehensively in its complexity, taking into
account various factors from related issues of quantum computing as well as from
digital humanities.
From a hardware perspective, the potentials stated above apply to an “ideal”
quantum computer, which is not yet available. Today’s quantum computers are error-
prone and have only limited capabilities (Preskill, 2018; Leymann & Barzen, 2020a).
They provide only a small number of qubits (LaRose, 2019) allowing only a limited
set of input data to be represented within the quantum computer. Besides this, noise
affects calculations on quantum computers (Knill, 2007; Preskill, 2018): The states
of the qubits are only stable for a certain amount of time—an effect referred to
as decoherence—due to unintended interactions between the qubits and their envi-
ronment (Leymann et al., 2020; Nielsen & Chuang, 2002). In addition, the opera-
tions that manipulate the qubits are also error-prone. Nevertheless, so-called NISQ
(Noisy Intermediate-Scale Quantum) machines (Preskill, 2018) allow first applica-
tions which can be used for prototypical implementations addressing small real-world
problems. This can contribute to establish application knowledge—also in the field of
quantum humanities. Taking a closer look at the recently published quantum roadmap
of IBM (Gambetta, 2020) it emphasizes the importance of starting to build application
knowledge now: Here IBM outlines its roadmap to 1 million error-corrected qubits
within a decade, providing more than 1,000 qubits in 2023, continuously increasing
the number of qubits for the coming years.
Next to the issues related to the hardware of quantum computers, from the software
perspective too several questions need to be addressed when working with quantum
computers. Even though quantum computers are becoming commercially available
(e.g. IBM, D-Wave, Rigetti, Honeywell, IONQ), questions regarding usability and
accessibility must be considered for each vendor. Also, more concrete questions like
the following need to be taken into account: How to encode your data properly based
on the processing demands of the chosen quantum algorithm (Weigold et al., 2021)?
How to expand oracles that many algorithms contain to keep the influence of such
an expansion on the number of qubits and operations required small? How to cope
with readout errors? (Leymann & Barzen, 2020a).
It is also necessary—even if the quantum computer is superior to classical
computers in certain areas as stated above—to address several open questions from
the application side. In a first step those existing and new problems from the digital
humanities need to be determined that are suitable to be considered being solved
8 J. Barzen

by quantum computers. Quantum humanities combines all the different disciplines

subsumed under the humanities with disciplines like computer science, mathematics,
and physics from the quantum computing site. This makes it an even more heteroge-
nous field than the digital humanities is already and demands a high degree of transla-
tion capabilities. Defining terms that are used differently in the various disciplines to
establish a shared language is one of the fundamental tasks for quantum humanities.
Also, when focusing on the more “pragmatical” benefits by using quantum
computers as a tool for answering concrete questing from the digital humanities,
e.g. how to run parts of a classification algorithm in a machine learning scenario
on quantum hardware, lots of questions still remain and need to be answered. For
example: How to choose the best algorithm to address the stated problem? How to
cope with categorical data while most algorithms require numerical data?
Based on the following use case we want to contribute reusable knowledge for
different applications in quantum humanities and introduce a toolchain supporting
data analysis by including quantum computers in the process.

3 Quantum Humanities Use Case: Project MUSE

In recent years, more and more research projects in the field of digital humanities
can be identified that are based on data and data analysis to provide new insights
into questions stated in the humanities. For example, when taking a closer look at the
books of abstracts of the DHd (Digital Humanites im deutschspachigen Raum (DHd,
2021)) conferences—as a well-established conference in the digital humanities—
over the last three years (Sahle, 2019; Schöch, 2020; Vogeler, 2018) the term data
(and its German translation, as well as all compound words containing the term)
occurs more than 5,600 times and hint to the significance of data in digital humanities
research. Also, in analogy, terms hinting to analysing this data via techniques from
machine learning (e.g. searching for: machine learning, artificial intelligence, un-
/supervised learning, clustering and classification) has about 1,000 counts, giving
an impression of the importance of this approach to analyse data gained during the
last years. Since quantum computers are expected to have substantial superiority
especially in the field of machine learning (Schuld & Petruccione, 2018), the use
case presented in the following is positioned in this area. Here, data from the digital
humanities project MUSE is analysed with the of help machine learning and quantum
machine learning techniques.
The project MUSE (2021) aims at identifying a pattern language for costumes in
films. It contains the method as well as a supporting toolchain to capture all relevant
information about costumes, to analyse them and to abstract the significant infor-
mation into costume patterns. Costume patterns capture those significant elements
of a costume (e.g. colour, material, way of wearing) that are used to communicate a
certain stereotype or a character trait, for example. When speaking of a “nerd” or a
“wild west outlaw” most recipients do have a rather clear idea of what these stereo-
types should look like to be recognized as “nerd” or “wild west outlaw” based on
From Digital Humanities to Quantum Humanities … 9

media socialization. Costume designers compose their characters by choosing each

part of a costume, its specifics and attributes with the explicit intent to support the
character, the story, and the actor in achieving a specific effect (Barzen & Leymann,
2014; Schumm et al., 2012). As there are often similar clothes, colours, or mate-
rials used to communicate the same stereotype there are conventions that have been
developed and knowledge about these conventions is contained in films.
To extract this knowledge the MUSE method consists of five steps (Barzen, 2018;
Barzen et al., 2018) which are supported by different tools and are outlined in the
following. The method has been proven to be generic by applying it in our parallel
project MUSE4Music (Barzen et al., 2016).

3.1 MUSE Ontology

Step 1: Define the domain by means of a comprehensive ontology which might

be based on several taxonomies (Barzen, 2013, 2018). These taxonomies structure
all relevant parameters that have a potential involvement in the effect a costume
might have. The MUSE ontology includes taxonomies of base elements (e.g. pants,
shirts, jumpers) and primitives (e.g. sleeve types, zipper, collar types), their materials,
designs, colours, way of wearing, conditions, and functions—to name just a few—
as well as so-called operators (e.g. worn above, wrapped around, attached) turning
base elements into a composed outfit. The hierarchical structure of the taxonomies
is important when analysing the data. As various algorithms require numerical data
the categorical data of MUSE must be transformed into numerical data based on the
structure of the taxonomies, for example (see Sect. 4.2, for more details see Barzen
et al. (2021)). With more than 3150 nodes, a comprehensive ontology to describe
costumes in a very detailed and structured manner has been developed.

3.2 MUSE Film Corpus

Step 2: Identify—based on well-defined criteria—those films that have a potential

“big” impact in terms of costumes on the recipients. Therefore, in a first step genres
were determined that promise a reoccurrence of quite similar characters and stereo-
types. For MUSE the genres high school comedy, western movies and fairy tales
were chosen. In a second step within each of these identified genres the 20 films
with the highest box office numbers and scores in rankings were chosen as part of
the initial film corpus to be analysed (for more details see Barzen (2018)).
10 J. Barzen

3.3 MUSE Data Set

Step 3: Capturing detailed information about the costumes of the initial film corpus.
To support capturing of all relevant parameters, a repository was designed and imple-
mented: The MUSE-repository (an open-source implementation is available under
(MUSE GitHub, 2021)) is based on the ontology described in Sect. 3.1 and assists to
collect all information about the films of the corpus, the characters and the costumes.
Currently (March 2021), the data set contains 5,231 costumes described from 58
films. These costumes contain 28,798 base elements and 63,711 primitives, 159,360
selections of colours and 180,110 selections of materials.

3.4 MUSE Data Analysis

Step 4: Analysing all captured information about costumes to determine those

costumes and costume attributes that achieve a similar effect in communicating with
the recipient hinting to costume patterns. The analysis consists of two main steps
(Falkenthal et al., 2016a): The first step applies data mining techniques, e.g. asso-
ciation rule mining, to determine hypotheses. These hypotheses could, for example,
identify those costume elements that are used to communicate a certain character
trait like “active person” or “shy person”. To refine and verify these hypotheses in
the second step, online analytical processing (OLAP) techniques (Falkenthal et al.,
2015) are used. As a result, indicators for costume patters are determined.
To improve the process of building hypotheses we are currently extending the
analysis of the MUSE data by various techniques from machine learning as well
as quantum machine learning (Barzen et al., 2021; Barzen & Leymann, 2020). A
more detailed discussion on the currently used methods and techniques is given in
the following Sects. 4–7.

3.5 MUSE Costume Patterns

Step 5: Abstracting the results of the analysis step 4 into costume patterns. Patterns
in the tradition of Alexander et al. (1977) are documents that follow a pre-defined
format to capture knowledge about proven solutions in an abstract way to make this
knowledge easily accessible and reusable for other applications. Patterns are related
to each other and compose a pattern language based on these relations. As stated
above, costume patterns capture the proven solutions about how costume designers
address the problem to communicate a certain stereotype like a “wild west sheriff” in
terms of all the significant elements of the costume. This contains e.g. base elements,
primitives, their relations, colours, ways of wearing, material, if they proved to be
significant in the analyse step. As the costume patterns are part of a costume pattern
From Digital Humanities to Quantum Humanities … 11

language, they support to solve complex problems by browsing through different

solutions related to each other.
To support the accessibility of the costume patterns, tooling support is provided
by our generic pattern repository called Pattern Atlas ((Leymann & Barzen, 2020b),
based on our former pattern repository PatternPedia (Fehling et al., 2014)).

4 Analysing Data

Analysing data has two major purposes: discovery and prediction. The collection of
techniques focusing primarily on the first purpose is called data mining, and the set
of techniques focusing primarily on the second purpose is called machine learning.
Despite their different primary purposes, data mining and machine learning have a
large overlap in terms of the methods they use: Optimization and statistics are core of
both approaches, and data mining is even using selective techniques from machine
learning. This is why both disciplines are often subsumed by the term data science.
Another difference is how the suitability or appropriateness of a data mining or a
machine learning solution is assessed: The appropriateness of a data mining solution
is assessed by its capability to discover previously unknown facts, while the appropri-
ateness of a machine learning solution is assessed by correctly reproducing already
known knowledge. Once a machine learning solution reproduced known knowledge
the solution is considered to correctly deliver previously unknown knowledge.
Despite these differences, the development of a solution based on either of both
disciplines follows the same overall procedure which is described in the next section.

4.1 Data Analysis Pipeline

In the past, we used data mining technologies to analyse data about costumes in
films (see Sect. 3.4, for more details; see Falkenthal et al., 2016a; Barzen, 2018).
The general procedure shown in Fig. 1 has been used for this analysis, but it is also
applicable in analysing data with machine learning techniques (Sect. 5).
Data cleansing (Skiena, 2017) encompasses activities like data acquisition, format
conversion and so on. Data preparation (Skiena, 2017) deals with a proper treatment
and encoding of data, and feature engineering. Algorithm selection (Kubat, 2017)
determines the family of algorithms (e.g. classification, clustering), selects a proper

Data Data Algorithm Algorithm Result Hypothesis Hypothesis

Cleansing Preparation Selection Execution Evaluation Building Validation

Fig. 1 Major steps of the data analysis process chain

12 J. Barzen

family member, and choses hyperparameters of the selected algorithm. Result eval-
uation is often based on visualizing the results (Skiena, 2017) and the creation of
hypotheses, which are finally validated. Because these last steps are critical for the
success of a data analysis project they are depicted as separate steps in Fig. 1. If no
hypothesis can be built or successfully validated, other algorithms may be tried out to
finally succeed with at least one proper hypothesis. Successfully validated hypotheses
trigger further processing towards the final goal of the project, like finding patterns
(see Sect. 4.3).
Note that within a data analysis project, most time is typically spend in data
cleansing and data preparation (Skiena, 2017). Also, several algorithms, even several
algorithms of the same family of algorithms, are typically applied within an overall
data analysis project. And each algorithm applied involves performing several steps
of the process shown in Fig. 1. For example, in order to determine patterns of costumes
in films in MUSE, the mostly categorial data must be prepared, features must be engi-
neered, and the resulting data must be clustered (see the black shapes in Fig. 2). After
clusters have been determined, data about newly captured clothes can be classified
to identify which costume the clothes represents.
Each of these tasks can be achieved by a variety of algorithms (grey shapes below
black shapes). Data preparation, for example, involves one-hot encoding to turn
categorial data into binary vectors, or the set of data points representing clothes may
be turned into a distance space based on using Wu-Palmer similarity (see Sect. 4.2).
Feature engineering may be performed by training an autoencoder (see Sect. 5.5) or
multi-dimensional scaling after the transformation of the data into a distance space.

Data Algorithm Algorithm Result Data Algorithm Algorithm Result

Preparation Selection Execution Evaluation Preparation Selection Execution Evaluation

Data Feature
Clustering Classification
Preparation Engineering

Distance Space
Autoencoder MaxCut SVM
Computation

One-Hot Boltzmann Boltzmann

PCA
Encoding Machine Machine

Multi-Dim
OPTICS
Scaling

Fig. 2 Data analysis pipeline for determining patterns and succeeding classification
From Digital Humanities to Quantum Humanities … 13

Clustering may be based on solving the maximum cut problem (see Sect. 6.2), training
a corresponding Boltzmann machine (see Sect. 5.4), or using the k-means algorithm
(Arthur & Vassilvitskii, 2007). Classification can be done by means of a Boltzmann
machine too, or by using a SVM.
Furthermore, each of the tasks depicted as black shapes in Fig. 2 is in fact
performed as a process chain (see the small process chains refining two of the
black shapes). For example, feature engineering requires to select a corresponding
algorithm, e.g. multi-dimensional scaling and setting proper hyperparameters (see
Sect. 7.3). The corresponding embedding needs the data in a certain format which
has to be prepared. After executing the algorithm, the resulting data representation
must be evaluated; if it is not appropriate, the hyperparameters must be adapted, and
the algorithms must be run again. If the results based on this algorithm is not appro-
priate at all, another algorithm (e.g. principal component analysis (PCA)) must be
tried out. Once the features are properly engineered, clustering is performed, which
means that the corresponding process chain is executed. Thus, a pipeline results that
consists of a sequence of several process chains.

4.2 Categorical Data

Many data in the Humanities are categorical data, i.e. data with a finite set of
values, often strings, that cannot be used for calculations. For example, computing
the maximum of a set of textures or the average of jobs is meaningless. But most
machine learning algorithms presume numerical data or even metrical data. Thus,
categorial data must be transformed into metrical data to be able to be processed by
such algorithms.
Barzen et al. (2021) discussed a mechanism of how to turn tuples of categorical
data the domains of which are tree structured into an approximately equivalent set
of vectors in Rn , i.e. into a set of metrical data. Our mechanism is (i) based in the
Wu-Palmer similarity (Wu & Palmer, 1994), (ii) the fact that similarities can be
turned into distance measures, and (iii) that a finite set with a distance measure can
be embedded into an appropriate Rn by means of multidimensional scaling (MDS)
(Cox & Cox, 2001): see Barzen et al. (2021) for all the details.
Data elements the values of which are given by means of a taxonomy are such data
with a tree structured domain. In our application area of costumes in films (Barzen,
2018), most data types have domains defined by taxonomies. Thus, based on our
mechanism this data can be embedded into a vector space and, consequently, can be
processed by machine learning algorithms.
Figure 3 shows a small fraction of one of the taxonomies of our costume data set.
To compute the similarity of two nodes in a tree (e.g. a “swimming shorts" and a
“cycling shorts” in the figure) the lowest common ancestor of these two nodes must
be determined: in our example, this is the node “short pants”. Next, the length of the
paths (in terms of number of edges) from each of these nodes to their lowest common
ancestor must be determined: in our example, these lengths are L 1 = 2 and L 2 = 1.
14 J. Barzen

L1 = 2

L3 = 4

L2 = 1

Fig. 3 Example for using the Wu-Palmer similarity measure

Then, the length of the path from their lowest common ancestor to the root of the tree
must be determined: in our example, this length is L 3 = 4. Finally, the Wu-Palmer
similarity is defined as

2L 3 8 8
ω(swimming shorts, cycling shorts) = = = ≈ 0.72.
L 1 + L 2 + 2L 3 2+1+8 11

The similarity between two nodes is between 0 and 1, i.e. “swimming shorts” and
“cycling shorts” are quite similar. A first quick overview of the similarities between
the data elements of a data set can be given by a similarity matrix that presents the
similarities of pairs of data elements and may render these values by means of a
colour code (see Sect. 7.1 for an example).
The method to determine the similarity ω between two values of a tree structured
domain (part 1 in Fig. 4) can be extended to determine the similarity of two sets of
values of the same tree structured domain: in a nutshell, the similarities ω of all pairs is
determined whose first component is from the first set and whose second component
is from the second set; a certain mean value of these similarities is computed, resulting
in a similarity measure σ between two sets (part 2 of Fig. 4). Based on the similarity
measure σ the similarity of tuples of the same dimension can be determined whose
components are sets of values of the same tree structured domain (part 3 in Fig. 4): the
similarities between each components of the tuples is determined (e.g. the similarity
of the colours of two tuples representing costume A and costume B, the similarity
of their materials and so on); the mean value of these similarities is computed which
results in a similarity measure μ between two tuples. See Barzen et al. (2021) for
the details.
From Digital Humanities to Quantum Humanities … 15

σ
σ

Fig. 4 Computing similarities of sets and tuples

The similarity ω(i, j) between two nodes i and j (and the similarity σ (i, j) of
two sets or the similarity μ(i, j) between two tuples i and j, respectively) can be
transformed into the dissimilarity or distance δ(i, j) between i and j, respectively, by
means of

δ(i, j) = ω(i, i) + ω( j, j) − 2ω(i, j)

with ω(i, i) = ω(j, j) = 1 (substituting ω by σ or μ in the formula accordingly).

This way, a set of data points the tuples of which have components whose domains
are define by taxonomies can be transformed into a set with a distance measure. In
analogy to a similarity matrix the distances between the data elements are used to
build a distance matrix.
Figure 5 depicts this situation: A set M with a distance measure δ is given, and
the distances of the elements of M are indicated by the length of the lines between
the points. An embedding ε maps the set M into Rn , i.e. the elements of M become
vectors in Rn (often referred to as feature space). But the vectors are not chosen
arbitrarily but the mapping ε tries to keep the distances in M approximately the same
than the distances in Rn , i.e. the distances of the corresponding vectors (indicated by
the length of the dashed lines between the vectors in Rn ) are approximately the same
as the distances in M; the distances in Rn are measured by some norm  ·  p . Here,

Fig. 5 An embedding maps

a set with a distance measure
into a Euclidian vector space
16 J. Barzen

“n” and “p” are hyperparameters that must be chosen based on former experience.
Embeddings can be computed by multidimensional scaling (MDS) (Cox & Cox,
2001).
This way, a set of data points M with components whose domains are defined by
taxonomies can be transformed into a set of data points in Rn such that the distances
in M are approximately the distances in Rn . After this transformation, the data points
can be further processed by most machine learning algorithms. Our costume data set
is transformed in this manner (see Sect. 7.1) and is, thus, available to be analysed by
several machine learning algorithms.

4.3 Creating Pattern Languages Based on Data Analysis

The main purpose of our project MUSE is the creation of pattern languages of
costumes in films. Pattern languages can be built in several ways. Often, they are
built from experience: practitioners recognize that a particular problem occurred
repeatedly in the past perhaps in various contexts and that the solutions applied
to solve the problem are very similar. Thus, these solutions can be abstracted to
identify their underlying principles such that these principles can be applied to solve
that particular problem even in new contexts. This way, a new pattern has been
identified. Similarly, relations between patterns are established based on experience.
Those relations between patterns become links with an associated semantics.
What results is a pattern language: A pattern language is a collection of patterns
within a particular domain and their relations. From an abstract point of view, a
pattern language is a weighted, directed graph. The nodes of this graph are the
patterns, the edges are the links between the patterns, and the weights of the edges
are the semantics of the links.
In contrast to derive patterns from experience, our project MUSE strives towards
deriving patterns by means of analysing data. The method we developed in MUSE
is generally applicable, and its use has already been initially verified also in the
domain of music (Barzen et al., 2016) and is envisioned by other aspects of films (see
Falkenthal et al., 2016b). Applying the MUSE method in other areas of the humanities
successfully seems to be possible. After having used data mining techniques in our
method first, we are now exploiting machine learning techniques and the initial results
we get seem to be more promising than using mining techniques.
To identify a pattern language based on data analysis the process sketched in Fig. 1
in Sect. 4.1 must be extended (see Fig. 6). If the hypothesis validation step resulted

Pattern
Hypothesis Pattern Pattern Pattern
Candidate
Validation Verification Relations Language
Identification

Fig. 6 Creating a pattern language

From Digital Humanities to Quantum Humanities … 17

in a verified hypothesis, this hypothesis must be turned into a pattern document. That
is the various sections of the pattern document corresponding to the hypothesis must
be authored. Thus, a pattern candidate is created: the resulting document is “only”
a candidate but not yet a pattern because its general applicability still needs to be
verified. Such a verification is performed in the succeeding pattern verification step.
Reiners (2014) proposes a system based on which a community e.g. can discuss
the candidate, exchange experiences with it, and can jointly decide to promote the
candidate to a pattern. The new pattern must be related to the other patterns of the
domain in the pattern relations step. This way a pattern language results. Note that
a pattern language is a “living” entity, i.e. over the time patterns will be added to
the pattern language. Typically, a first version of a pattern language will be made
available or published when “major” problems of the underlying domain can be
solved with it.

5 Artificial Neural Networks

An artificial neural network (or neural network for short) is a mathematical model that
mimics the biological neural networks of brains of animals. Such a neural network
is represented as a directed graph the nodes of which abstract neurons and the edges
abstract synapses of a brain. In this section, neurons and neural networks are defined,
perceptrons and their use for classification are sketched, and restricted Boltzmann
machines, as well as autoencoders are described.

5.1 Neurons

The first model of an artificial neuron roots back to McCulloch and Pitts (1943),
nowadays called a McCulloch-Pitts neuron. Such a neuron accepts binary data as
input and produces a binary output if a predefined threshold is exceeded. This model
was extended and called perceptron (see Sect. 5.3) in Rosenblatt (1958). Today a
neuron is abstracted as a mathematical function (Zurada, 1992): it receives multiple
values as input and produces a single value as output. What characterizes functions
that represent neurons is the way in which the output of a neuron is computed, i.e.
the ingredients of such functions and how they interact.
Definition Let w1 , . . . , wn ∈ R be real numbers called weights; the function


n
 : Rn → R, x → wi xi
i=1

is called propagation function. Furthermore, a function α : R → R is called

activation function. Then, a neuron is a map
18 J. Barzen

Fig. 7 Ingredients of an artificial neuron

 n 

ν : Rn → R, x → α wi xi .
i=1


Figure 7 shows a graphical representation of an artificial neuron: the left side of
the figure represents the details of the definition before, the right side of the figure
depicts a more condensed version of this graphical representation in which the two
nodes representing Π and α are combined into a single node denoted with Π|α. The
input values x1 , . . . , xn of the neuron ν are connected via directed edges with the
neuron’s propagation function Π; the weight wi of the input x i is shown as a number
associated with the corresponding edge. The purpose of the propagation function Π
is to aggregate the input of the neuron considering the corresponding weights of the
input values. The purpose of the activation function α is (i) to decide whether its
input suffice to pass on a non-zero output and (ii) what the output value will be.
An important variant defines a neuron as a map ν : {1} × Rn ⊆ Rn+1 → R, with
 

n
ν(1, x1 , . . . , xn ) = α w0 + wi xi .
i=1

Thus, a neuron has one input x 0 of constant value “1” (see Fig. 8). The corre-
sponding weight w0 is called bias, denoted as “b”. The bias influences the firing of
the neuron: a negative bias results in the neuron firing less frequently, a positive bias
results in the neuron firing more frequently.

Fig. 8 A neuron with a bias x0 = 1

w0=b
x1
w1
x2
. w2 o
.
.
wn
xn
From Digital Humanities to Quantum Humanities … 19

One problem with realizing neurons on a quantum computer is that quantum algo-
rithms are unitary transformation, thus, linear, but activation functions of neurons,
are typically non-linear. Cao et al. (2017) and Wan et al. (2017) present quantum
neurons with different activation functions. A special neuron (a so-called percep-
tron—see Sect. 5.3) on a quantum computer of the IBM Quantum Experience family
has been described in Tacchino et al. (2018). This implementation proves the expo-
nential advantage of using quantum computers in terms of requiring only N qubits
in order to represent input bit-strings of 2N bits. But in order to prepare this input an
exponential number of 1-qubit gates and CNOTs are needed, in general (Leymann &
Barzen, 2020a).

5.2 Neural Networks

Neural networks are directed graphs the nodes of which are neurons (Kubat, 2017;
Skiena, 2017). The edges of the graph specify which neuron passes its output as
input to which other neuron. The neurons are partitioned into disjoint sets L1 ,…, Ln
(so-called layers) such that neurons of layer Li are connected with neurons of layer
Li+1 . The neurons of layer L1 get their input from the outside, i.e. not from other
neurons; this layer is called the input layer of the neural net. Neurons of layer Ln
pass their output to the outside, i.e. not to other neurons; this layer is called the output
layer of the neural network. All other layers are referred to as hidden layers. A neural
network with at least one hidden layer is sometimes called a deep neural network.
Definition A neural network is a tuple (N, E, L , X, Y) with the following
properties:
• G = (N, E) is a directed graph,
• L = {L1 ,…, Ln } ⊆ P(N) is a partition of the set of nodes, i.e. ∀i: Li = ∅, Li ∩ Lj
= ∅ for i = j, and ∪ Li = N; Li is called layer, L1 input layer, Ln output layer,
and Li (1 < i < n) hidden
 layer,
• each node νi = w ji , i , αi ∈ N is a neuron with a set of weights {wji }, a
propagation function Π i , and an activation function αi ,
• X = {x1 ,…, xm } is the set of input values,
• Y = {y1 ,…, yk } is the set of output values,
• the set of edges E ⊆ (N × N) ∪ (X × N) ∪ (N × Y) connects two neurons, or an
input value and a neuron, or a neuron and an output value,
• for ν ∈ Li and (ν, ν’) ∈ E ∩ (N × N) it is ν’ ∈ Li+1 ,
• for ν ∈ Li , x ∈ X, and (x, ν) ∈ E it is i = 1,
• for ν ∈ Li , y ∈ Y, and (ν, y) ∈ E it is i = n, and y is the output of ν,
• for (νj , νi ) ∈ E ∩ (N × N) the output of νj is passed to νi where it is processed by
νi ’s propagation function Π i weighted by wji ,
• for ν ∈ L1 and (x, ν) ∈ E, ν processes x by its propagation function Π weighted
by wji .

20 J. Barzen

Effectively, a neural network that consumes m input values and that produces k
output values is a function Rm → Rk . The computational power of neural networks
stems from the fact that any continuous function on a compact set K ⊇ Rm → Rk
can be approximated with arbitrary precision by a neural network (see Kidger and
Lyons (2020) for details).
Figure 9 shows the principle structure of a neural network according to our defi-
nition before. Note that there are some variants of our definition: sometimes, edges
between neurons with the same layer are supported, sometimes edges are allowed
between neurons of any layer, sometimes a neuron may be connected to itself, and
so an. However, such variations are for use in special cases.
The structure of a neural network, i.e. its set of layers and its edges, is referred
to as its topology. Defining the topology of a neural network, specifying the activa-
tion functions and propagation functions of its neurons is called modelling a neural
network. Note that most often the propagation function from our definition in Sect. 5.1
is used, but activation functions change across problem domains; even different
activation functions for different neurons of a given neural network may be defined.
Training a neural network consists of choosing appropriate weights and biases. For
this purpose, the neural network
 is considered as a function F : Rm → Rk . Further-
more, a set of input vectors xj ∈ R is given together with the corresponding
m
 
known correct output vectors y j ∈ Rk ; note that the set of pairs (x j , y j ) is
called training data. Ideally, the neural network will output yj for each input xj ; but
the neural network is not (yet) realizing the ideal function, i.e. it is just approximating

Fig. 9 Principle structure of a neural network

From Digital Humanities to Quantum Humanities … 21

it: it will output Y j = F(x j ). Thus, the neural network will produce an error |y j −Y j |
in computing yj . The goal of training the neural network is to minimize these errors
considering all training data by adapting the weights and biases.
For this purpose, the so-called loss function
 2  2
L(w, b) = yj − Yj = y j − F(x j )
j j

is minimized (as usual, the square of errors is used instead of their modulus). Here,
“w” denotes all weights of the neural network and “b” all its biases. Training now
means to choose w and b such that L(w, b) is minimized. Lots of mathematical proce-
dures are known to minimize L, e.g. gradient-based methods (e.g. stochastic gradient
descent) or derivative-free methods (e.g. Nelder-Mead) can be chosen (Nocedal &
Wright, 2006). Note that gradient-based methods require differentiability of the loss
function which in turn requires differentiability of F which in turn requires differ-
entiability of the activation functions of the neurons composing the neural network
implementing F (see Sect. 5.1). After this training phase, the neural network is ready
for use. It is said that the neural network has learned. The use of training data, i.e. input
values with corresponding known output values, is referred to as supervised learning:
the processing of the input data by the neural network is supervised by comparing
its output with the known given results associated with the input. Learning without
supervision is referred to as unsupervised learning: this kind of learning is discussed
in Sect. 5.5 in the context of autoencoders.
An implementation of quantum neural network that requires a single qubit for
each neuron (plus additional ancillae) has been proposed by Cao et al. (2017). The
training of this quantum neural network is performed in a hybrid quantum–classical
manner, i.e. the optimization is executed by classical software using the Nelder-
Mead algorithm. In contrast to training classical neural networks where the individual
training data is processed sequentially, the training of this quantum neural network
can be done based on a superposition of the input/output pairs of the training data—
something impossible for classical neural networks. A quantum neural network has
been realized on a near-term quantum processor as described by (Farhi & Neven,
2018); their neural network has been successfully used as a binary classifier on the
MNIST data set to distinguish two different handwritten digits. A set of requirements
on implementations of quantum algorithms that represent “meaningful” quantum
neural networks has been posed in Schuld et al. (2014).

5.3 Perceptrons

A perceptron can decide whether a given data point represented as input is left or
right of a given straight line in the plane (or in higher dimensions, right or left of a
hyperplane). Thus, a perceptron can be used as a binary classifier of linear separable
22 J. Barzen

H H
x

Fig. 10 Binary classifier for linear separable sets

data sets: the left side of Fig. 10 shows two kinds of data points (grey points and
black points) in the plane. They are linear separable because a straight line H can be
found that splits these two data sets such that on one side of H only the black data
points are located, on the other side only grey data points reside.
Any straight line H (or hyperplane in higher dimensions) can be defined by means
of a vector w that is orthogonal to H, and whose length w = θ is the distance of H
from the origin (so-called Hesse normal form—see the right side of the figure). For
any point x, the scalar product w, x is the length of the projection of x onto w, and
w, x · x itself is the projection of x onto the straight line defined by w. Thus, a point
x is on H if and only if w, x − θ = 0. Furthermore, x is “on the right side” of H if
the length w, x of the projection of x onto w is greater than θ (i.e. w, x − θ > 0),
and x is “on the left side” of H if the length w, x of the projection of x onto w is
less than θ (i.e. w, x − θ < 0). Consequently, the value of w, x − θ determines
whether a point x is right or left of the hyperplane H defined by w, i.e. it serves as
is a binary classifier: a positive value classifies x as a black point, a negative value
classifies x as a gray point.
Definition A perceptron is a neuron ν with Heaviside-function θ as activation
function, where θ := (w1 , . . . , wn ) for the weights w1 , . . . , wn of the neuron.

The vector w = (w1 , . . . , wn ) of the weights of the perceptron ν determines a
hyperplane H (with w = θ ). When a point x = (x1 , . . . , xn ) is passed as input to
the propagation function Π of ν, the scalar product of w and x results:

n
(x) = wi xi = w, x.
i=1

With
From Digital Humanities to Quantum Humanities … 23
 n 

ν(x) = θ wi xi ∈ {0, 1},
i=1

it is ν(x) = 1 ⇐⇒ θ (w, x) = 1 ⇐⇒ w, x ≥ θ ⇐⇒ w, x − θ ≥ 0.

Thus, ν(x) = 1 classifies x as “right of H” (considering x on H, i.e. w, x−θ = 0, as
“right of H”), and ν(x) = 0 classifies x as “left of H”. This reveals that a perceptron
is a binary classifier.
Given a linear separable set of data, how are the weights w = (w1 , . . . , wn )
determined the corresponding hyperplane of which splits the data set into two disjoint
parts? In other words: how does a perceptron learn to split the data set correctly? For
this purpose, the following procedure is used (e.g. da Silva et al. (2017)):
• Assume a set T = {(t1 , c1 ), · · · , (tm , cm )} of training data is given, i.e. t j =
1, x1 j , · · · , xn j , is a data point (with its first component set to “1” to introduce
its weight as a bias) and c j ∈ {0, 1} classifies the data point as member of the class
“0” or member of the class “1”. For example the class “0” may indicate a grey
point and the class “1” a black point in Fig. 10. Since the training data specifies the
correct result of the classification, the algorithm sketched is a supervised learning
algorithm.
• The algorithm is performed in several steps τ; the tuple w(τ ) =
(w0 (τ ), w1 (τ ), · · · , wn (τ )) denotes the bias b(τ ) = w0 (τ ) and the weights of
the perceptron at step τ.
• At step τ = 0, w(0) is chosen randomly (“small” values have been shown to be a
good first choice).
• At any step τ > 0 the following is computed for 1 ≤ j ≤ m, i.e. for the whole set
of training data:
 n 
 

y j (τ ) = b(τ ) wi (τ )xi j + b(τ ) = b(τ ) w(τ ), t j ∈ {0, 1},
i=1

i.e. y j (τ ) is the classification of the data point t j at step τ.

 
• The value c j − y j (τ ) measures the error of classifying the data point t j at step
τ as y j (τ ), while c j is the correct known classification. The goal is to minimize
the overall classification error

1 
m
e(τ ) = |dk − yk (τ )|,
m k=1

i.e. e(τ ) < γ , where γ is a predetermined error threshold. If the error is below this
threshold the algorithms stops.
• Otherwise, a new bias and new weights are determined: chose an arbitrary data
point t j (i.e. 1 ≤ j ≤ m) and compute
24 J. Barzen

wi (τ + 1) = wi (τ ) + r · c j − y j (τ ) · xi j ,

for 0 ≤ i ≤ n. Thus, if the computed classification y j (τ ) for data point t j is

correct, i.e. it is the known classification c j , w(τ ) is not changed “in direction
of t j ”, i.e. w(τ + 1) = w(τ ). Otherwise, w(τ ) is modified “in direction of t j ”,
i.e. w(τ + 1) = w(τ ) ± r · t j . Here, r is a predetermined constant, the so-called
“learning rate”.
Assuming linear separable training data, the above algorithm converges and sepa-
rates the training data correctly (Novikoff, 1962). Typically, the algorithm converges
fast, i.e. the condition e(τ ) < γ is met fast. But there are situations in which the
convergence is slow and the algorithm should be stopped even if the result is not quite
precise. This is covered by the additional condition “τ = N” limiting the number of
iterations even if e(τ ) ≥ γ , i.e. the result of the learning process is above the error
threshold. Note that the variables γ , r, and N are so-called hyperparameters and
must be properly set prior to the learning process, e.g. based on former experience.
The separating hyperplane learned by the perceptron depends on the initial values
of the bias and the weights w(0) chosen during the initial step τ = 0. This is depicted
on the left side of Fig. 11: different initial values w(0) result in different hyperplanes
H1 , H2 , H3 . These hyperplanes are differently suited in correctly classifying new
data points: two new data points are shown as a triangle and a square. The triangle
is classified as a black point by H1 and H3 (being “right” of the hyperplanes), and as
a grey point by H2 (being left of it). The square is classified by H3 as a black point,
but as a grey point by H1 and H2 . This non-determinism of a perceptron learning a
separating hyperplane is in contrast to another machine learning technique (which
is not based on neural networks at all): a support vector machine (SVM) (Burges,
1998). A support vector machine determines a unique separating hyperplane based
on linear separable training data: this is achieved by computing the hyperplane with
the maximum margin that does not contain any of the test data (see the right side
of Fig. 11—the grey shaded rectangle represents the margin of the hyperplane). A
support vector machine would classify the square as a grey point, and the triangle as a

H1 H2
H

H3

Fig. 11 Computing separating hyperplanes via perceptrons and support vector machines, and using
a computed hyperplane as classifier for new data
From Digital Humanities to Quantum Humanities … 25

black point. Furthermore, support vector machines can even be used to classify non-
linear separable data sets (Bennett & Campbell, 2000) by embedding the data into a
high-dimensional vector space (which already hints that quantum implementations
of support vector machines are advantageous because the state spaces of quantum
computers are extremely huge).
Wiebe et al. (2016) proposed two quantum perceptrons: the first one can be trained
with a quadratic speedup in size of the training data, and the second one quadratically
reduces the training time required to achieve a given precision when compared with
the algorithm presented in this section. A quantum perceptron has been suggested in
Tacchino et al. (2018) that requires exponentially less storage resources for training;
it has been implemented and validated on a near-term quantum computer of the IBM
Q family of systems.

5.4 Restricted Boltzmann Machines

Restricted Boltzmann machines (RBMs) are attributed to Smolensky (1986). While

for general Boltzmann machines learning is not feasible under practical condi-
tions, restricted Boltzmann machines can learn efficiently (see Hinton (2012) for
an overview). A restricted Boltzmann machine is a neural network with two layers:
an input layer and a hidden layer (see Fig. 12). Each neuron of the input layer has a
single input value, and the weight associated with such an input is “1”. Input values
as well as the output of each neuron are Boolean values; especially the output of a
restricted Boltzmann machine, i.e. the output of the neurons of the hidden layer, are
Boolean values. Each neuron has a bias (not shown in Fig. 12).

Fig. 12 Structure of a weights = 1 weights wij

restricted Boltzmann
machine; biases are not
shown x1

x2
y1
x3 .
.
. .
.
yk
.

xm
Input Visible Hidden Output
Values Layer {vi} Layer {hj} Values
26 J. Barzen

Definition A restricted Boltzmann machine is a two layer neural network (N, E,

L , X, Y) with following properties:
• the output of each neuron is a Boolean value (especially, y ∈ {0, 1} for y ∈ Y),
• the input values are Boolean values, i.e. x ∈ {0, 1} for x ∈ X,
• each neuron νi ∈ N has a bias; for νi ∈ L1 the bias is denoted by ai , for νi ∈ L2
the bias is denoted by bi ,
• for νi ∈ L1 there exists exactly one xi ∈ X with (xi , νi ) ∈ E (ignoring the input “1”
for the bias),
• for νi ∈ L1 the weight wii of (xi , νi ) ∈ E is “1”,
• L1 × L2 ⊆ E.
L1 is called visible layer, L2 is called hidden layer, i.e. the last property enforces
that each neuron of the visible layer is connected to each neuron of the hidden layer.

Training of a restricted Boltzmann machine (see Hinton (2012) for all details) is
different from the general mechanism for training a neural network (see Sect. 5.2).
The input values of a restricted Boltzmann machine are Boolean values. But data to
be processed is typically not Boolean. For this purpose, each neuron ν of the input
layer represents an observed value o: the value o is observed if and only if the input
of ν is “1”. Especially, o maybe of any type of data, i.e. not only numerical data but
also categorical data. This way, restricted Boltzmann machines can be used with any
type of data as input; and similar for output data.
The application areas of restricted Boltzmann machines include, for example,
classification (e.g. Chen and Srihari (2015), Larochelle et al. (2012)) and feature
learning (e.g. Zheng et al. (2013), Tomczak (2016)). For classification purposes, the
training data T = {(t, c)} are one-hot encoded, i.e. the class indicator c becomes
a vector c = (k1 , ..., kr ) ∈ {0, 1}r (where r is the number of classes) with k j = 1
if t is of class j, and k i = 0 otherwise (see the left side of Fig. 13). Note that this
is multi-class classification, i.e. data can be associated with more than two classes.
For feature learning (right side of Fig. 13), the properties of the data correspond to

Fig. 13 Using restricted Boltzmann machines for classification and feature learning
From Digital Humanities to Quantum Humanities … 27

the input and the output are the features learned. Typically, these features are used
by another algorithm for further processing, e.g. by another restricted Boltzmann
machine (so-called “stacked” restricted Boltzmann machines).
A special kind of quantum restricted Boltzmann machines that can approximate
any quantum circuit has been specified in Wu et al. (2020). Also, an implementation of
such kind of a quantum restricted Boltzmann machine on a NISQ device of the IBM Q
family has been provided. Quantum Boltzmann machines (i.e. a general Boltzmann
machine which supports also connections between neurons of the same layer) as
well as quantum restricted Boltzmann machines have been investigated by Amin
et al. (2018) with the result that their quantum Boltzmann machine outperforms a
classical Boltzmann machine on a specific learning task. A circuit realizing a quantum
restricted Boltzmann machine is given in Zhang and Zhou (2015). No implementation
on a real quantum computer has been evaluated but a simulation: it turned out that
the quantum restricted Boltzmann machine performed a classification task faster and
with higher precision than a classical machine. An implementation of a restricted
Boltzmann machine to be realized on a D-Wave quantum annealer is described in
Denil and Freitas (2011); obstructions for a real implementation are discussed.

5.5 Autoencoders

Today’s concept of an autoencoder goes back to Hinton and Zemel (1994). At a

first glance, it can be considered as two restricted Boltzmann machines merged
together the purpose of which is to reconstruct their input as output. After training,
the autoencoder may be split again into two restricted Boltzmann machines, each of
which may be used in combination with further machine learning algorithms.
Definition An autoencoder is a three layer neural network (N, E, L , X, Y) with
following properties:
• let F be the function represented by the autoencoder, then: F ≈ id (i.e. the
autoencoder’s output approximates its input),
• card L1 = card L3 (i.e. input layer and output layer contain the same number of
neurons).
The hidden layer L2 is called code (layer). Input layer and code layer are collec-
tively referred to as encoder, code layer and output layer are collectively referred to
as decoder.

Note that there are variants of autoencoders that support more than one hidden
layer, which is not discussed here. Figure 14 depicts on its left side the principle
structure of an autoencoder consisting of an encoder and a decoder. The right side of
the figure shows its structure as a neural network with three layers, the hidden layer
named “code”. Roughly, it can be considered as two restricted Boltzmann machines
with the code layer being the hidden layer of the encoder and the input layer of the
decoder (“stacked machines”).
28 J. Barzen

Fig. 14 Principle structure of an autoencoder and its structure as a neural network

The encoder transforms its input into a code (also called a representation of the
input). If the input is a data point in Rm and the code is a data point in Rk , the encoder
is a map ϕ : Rm → Rk . Because the output layer must have the same number of
neurons as the input layer, the output is again a data point in Rm , i.e. the decoder is a
map ψ : Rk → Rm . By definition, it is ψ ◦ϕ ≈ id. That is, for any input x, the output
is ψ ◦ ϕ(x) ≈ id(x) = x; thus, the difference (or error) between x and ψ ◦ ϕ(x)
 square of the error (x − ψ ◦ ϕ(x)) is
2
should be as small as possible (as usual  kthe
taken). For a set of training data T = x , the sum of these squared errors should
be minimal. Now, ϕ and ψ perform their mapping based on their weights wij and wi j
as well as their biases bi and bi , respectively. Consequently, the sum of the squared
errors is a function L (“loss function”) of these weights and biases:
  2
L(w, b) = xk − ψ ◦ ϕ xk .
k

Training an autoencoder means minimizing this loss function based on

some optimization algorithm, i.e. to determine wi j , bi , wi j , bi such that
 
L wi j , bi , wi j , bi ≈ 0. Note that the training of an autoencoder is unsuper-
vised because no set of training data is required that is labeled to indicate the output
for a given input.
After training, the encoder may be split from the decoder. The encoder will then
transform a data point x ∈ Rm into a data point ϕ(x) ∈ Rk , and it is known that
this ϕ(x) represents x faithful, i.e. it contains “the essence of x” because it includes
all information to reconstruct x (by means of ψ): ϕ(x) represents the features of x.
This is the reason for using the term “feature learning” for training an encoder. If k
< m, i.e. if the code has less components than the input, a dimension reduction is
achieved. Reducing the dimension of data is key because many algorithms are very
sensitive to it, i.e. they are much more efficient if a data point to be processed has
less components.
Note the similarity to embeddings as discussed in Sect. 4.2 which also reduce
dimensions of data. In contrast to encoders, embeddings try to keep distances between
From Digital Humanities to Quantum Humanities … 29

data points intact. While some other algorithms that might be used to process data
after dimension reduction require similar distances between the high-dimensional
data points and the low-dimensional data points, some other do not consider these
distances. Thus, the next algorithm used to process the data influences the decision
whether an embedding has to be used or whether an encoder suffice.
Based on the work of Lamata et al. (2019), a quantum autoencoder requiring
for each neuron a separate qubit has been worked out in Bondarenko and Feld-
mann (2020). Each neuron has a unitary operation associated with it that operates
on the neuron and all neurons of the preceding layer. The number of such operations
needed to apply the autoencoder grows exponentially with the maximum number of
neurons per layer. Training of their autoencoder has been simulated in MATLAB. An
implementation of a (small) autoencoder on a Rigetti Quantum Computer has been
described in Ding et al. (2019); the corresponding circuits are provided. This autoen-
coder was successfully used for dimension reduction. Similar has been achieved in
Pepper et al. (2019) based on a photonic quantum computer. Romero et al. (2017)
described a variational quantum–classical way to implement a quantum autoencoder.
Figure 15 depicts (for illustration only, not for reading) our implementation of the
quantum part of the algorithm based on the simulator of IBM Q und Qiskit (2021).
The input data is prepared in four qubits that are transformed by the encoder. Once
the encoder finished its processing, the first two qubits are set to |0 while the last
two qubits are kept in the state the encoder produced. Effectively, the code of the
autoencoder consists of these two last qubits, i.e. a dimension reduction by 50% is
achieved. Next, the decoder tries to reconstruct the original input from the code. The
decoding result is measured and analysed by a classical part (not shown in the figure);
based on this analysis another iteration may be required. Once the autoencoder is
successfully trained, the encoder can be used for dimension reduction.

Fig. 15 Example circuit of a quantum autoencoder

30 J. Barzen

6 Variational Hybrid Quantum–Classical Algorithms

As described in Sect. 2.3, near-term quantum computers support “short” computa-

tions only due to decoherence of their qubits and lack of fidelity of their operations.
However, many problems require more complex computations than supported on
such NISQ devices. In this section, the principle mechanisms of how to support such
complex computations on NISQ devices is described.

6.1 The Main Idea

Some quantum algorithms consist of both, a pure quantum part and a pure classical
part (e.g. Shor’s algorithm for factorization): its quantum part is performed on a
quantum computer, and its classical part is performed on a classical computer. That
is, such an algorithm inherently requires classical pre- or post-processing to achieve
its goal, it would not succeed without its classical part. Such a split algorithm is
referred to as a hybrid quantum–classical algorithm.
Instead of such algorithms that are inherently hybrid, algorithms might be
designed to be hybrid from the outset to limit their processing on a quantum computer.
For example, this is enforced by today’s near-term quantum devices that restrict the
amount of processing that can be performed with sufficient precision. Such an algo-
rithm requires additional classical processing to compensate for the restricted amount
of work performed by the quantum computer: the algorithm is not inherently hybrid
quantum–classical but it is so by design.
Two main problem domains are addressed by such kind of hybrid algorithms by
design: problems that at their core can be solved based on the Raleigh-Ritz principle
(Yuan et al., 2019), and problems that can be reduced to combinatorial optimiza-
tion problems. Both kind of algorithms make use of a quantum part that consists
of parameterized quantum circuits that prepare a state on a quantum computer and
measure it, and a classical part that optimizes the corresponding parameters in itera-
tions in dependence of the measured results (see Fig. 16). The parameterized unitary
operator U ( p1 , ..., pk ) in Fig. 16 is called an “ansatz”: its goal is to prepare states that
deliver measurement results m that are good candidates for arguments optimizing a
given function F. Thus, the value F(m) is dependent on the parameters p1 , ..., pk . By

Fig. 16 Principle structure

of variational hybrid
quantum–classical
algorithms
From Digital Humanities to Quantum Humanities … 31

varying these parameters based on classical optimization algorithms a corresponding

m may be found (Shaydulin et al., 2019). Finding an appropriate ansatz is hard. Note
that a cloud environment is particularly suited to perform hybrid quantum–classical
algorithms (Karalekas et al., 2020; Leymann et al., 2020).
Varying the parameters iteratively gives rise to the name variational hybrid
quantum–classical algorithm (or variational algorithms for short) for this kind of
approach. The application areas of variational algorithms span a broad spectrum
including factoring of numbers (Anschuetz et al., 2018), solving linear equations
(Bravo-Prieto et al., 2020), training autoencoders (Romero et al., 2017), or solving
non-linear partial differential equations (Lubasch et al., 2020). A general approach
towards a theory of variational hybrid quantum–classical algorithm is proposed in
McClean et al. (2016).

6.2 Maximum Cut: A Combinatorial Optimization Problem

The maximum cut problem is to partition the node set of a graph in two sets such
that the number of edges between nodes of the different sets is maximal. In Fig. 17,
part 1, a graph with four edges {A,B,C,D} is shown where each node is connected
to each other node. Part 2 and part 3 of the figure depict two different maximum
cuts: the members of one node set are coloured black while the members of the other
node set are left white. Dashed edges connect nodes of the same set, i.e. they do not

Fig. 17 Maximum cut of a graph: examples

32 J. Barzen

contribute to the cut, while solid lines connect nodes of the two different set, i.e. they
contribute to the cut. The number of edges of each cut is four.
Part 4 of the figure adds a weight to each edge, e.g. the weight of an edge indicates
the distance between the nodes connected by the edge; such a distance, for example,
might be determined by means of a distance measure (see Sect. 4.2). The weighted
maximum cut problem strives towards finding a partition of the node set such that
the sum of the weights of the edges between nodes of different sets is maximal. Note
that by setting each weight to “1”, the maximum cut problem is seen to be a special
case of the weighted maximum cut problem. Part (4) shows a weighted maximum
cut: the node set is partitioned into {A,B,C} and {D}, and the sum of the weights of
the corresponding edges is 20. If the weights are distances, the weighted maximum
cut is nicely interpreted as determining clusters: the two node sets consist of nodes
that are close to each other within each set, but nodes of different sets are far apart.
For example, D has distance 6, 9, 5 to A, B, C, respectively, but A, B, C are quite
close to each other (distances 1, 1, 2). Thus, an algorithm for determining a weighted
maximum cut can be used for clustering.
Let G = (N, E) be a graph with n nodes (i.e. card(N) = n), and let S, T be a cut of
N (i.e. S ∩ T = ∅, S ∪ T = N, S = ∅, and T = ∅). The function z : N → {−1, +1}
indicates membership to S or T, i.e. z(u) = +1 ⇐⇒ u ∈ S and z(u) = −1 ⇐⇒
u ∈ T (elements of S are labelled +1, elements of T are labelled −1—or “black”
and “white” in the figure). For brevity, zu : = z(u).
For i, j ∈ S, it is zi = zj = + 1, thus zi · zj = 1; and for i, j ∈ T, it is zi = zj −1,
thus zi · zj = 1. Consequently, it is i, j ∈ S ∨ i, j ∈ T ⇔ zi · zj = 1, i.e. (i, j) is not a
cut edge. Similarly, (i ∈ S ∧ j ∈ T) ∨ (j ∈ S ∧ i ∈ T) ⇔ zi zj = − 1, i.e. (i, j) is a cut
edge. Reformulation results in:
• (i, j) ∈ E is not a cut edge ⇔ 1 − zi zj = 0 ⇔ ½ · (1 − zi zj ) = 0
• (i, j) ∈ E is a cut edge ⇔ 1 − zi zj = 2 ⇔ ½ · (1 − zi zj ) = 1
Thus, a maximum cut maximizes the following cost function:

1  
C(z) = 1 − zi z j .
2 (i, j)∈E

This is because each cut edge contributes a “1” to the sum, and a non-cut edge
contributes a “0”: the more cut edges exist, the higher is C(z). Such kind of a formula,
i.e. a formula with parameters with values from {−1, +1} is called an Ising formula,
which is important for solving quantum problems on so-called quantum annealers.
In this chapter it is assumed that algorithms are formulated in the gate model. For
this purpose, parameters from {0, 1} are better suited, i.e. the above cost formula has
been transformed. To achieve this, a function x : N → {0, 1} indicates membership
to S or T, i.e. x(u) = 1 ⇐⇒ u ∈ S and x(u) = 0 ⇐⇒ u ∈ T . For brevity, x u : =
x(u). This implies:
• (i, j) ∈ E is not a cut edge ⇔ i, j ∈ S ∨ i, j ∈ T ⇔ xi · (1 − xj ) + xj · (1 − xi ) = 0
From Digital Humanities to Quantum Humanities … 33

• (i, j) ∈ E is a cut edge ⇔ (i ∈ S ∧ j ∈ T) ∨ (j ∈ S ∧ i ∈ T) ⇔ xi · (1 − xj ) + xj · (1

− xi ) = 1
The cost formula with binary parameters for a maximum cut is then:

1  
C(x) = xi · (1 − x j ) + x j · (1 − xi ) .
2 (i, j)∈E

The term xi · (1 − x j ) + x j · (1 − xi ) is a map {0, 1}n → {0, 1}, which is also

called a clause. A combinatorial optimization problem is to determine a z ∈ {0, 1}n
(i.e. binary parameters) that maximizes (or minimizes) a cost function

m
C(z) = Cα ,
α=1

where each Ci : {0, 1}n → {0, 1} (1 ≤ i ≤ m) is a clause with n binary parameters.

This reveals that the maximum cut problem is a (binary) combinatorial optimiza-
tion problem. Many other practically relevant problems like the travelling salesman
problem are such problems. Next, it is shown how such problems can be solved with
quantum computers.

6.3 QAOA

The concept of the quantum approximate optimization algorithm (QAOA) has been
introduced by Farhi et al. (2014). It is used to approximately solve combinatorial
optimization problems on near-term quantum devices and has been first applied to the
maximum cut problem in Farhi et al. (2014). Crooks (2018) provided a QAOA-based
implementation of a maximum cut algorithm on a near-term quantum computer of
Rigetti and showed that this implementation is faster than the well-known classical
Goemans–Williamson algorithm. The results can even be improved following the
warm starting idea for QAOA (Egger et al., 2020).
The basic idea is as follows: The cost function C of a combinatorial optimiza-
tion problem induces a map C : Hn → Hn by interpreting z ∈ {0, 1}n as the
computational basis vector |z ∈ Hn and defining

m
C|z  := Cα (z) · |z  = f (z) · |z .
α=1

Thus, the matrix of C in the computational basis is a diagonal matrix, i.e. the map
C is Hermitian and can be used to measure the states of a quantum computer (C is an

observable). Let z ∈ {0, 1}n such that f (z ) = max{f (z)} = C max . The expectation
value Cψ of measuring C in any state |ψ = x z |z is Cψ = ψ|Cψ =
34 J. Barzen

  
x z |z|x z f (z)|z = |x z |2 f (z) ≤ |x z |2 f (z ) = f (z )|x z |2 = f (z ) =

Cmax , i.e. the expectation value is less than or equal Cmax . If |ψ is close to |z , Cψ
will be close to Cmax . This means that with
   

|ψ  = x z |z  = x z |z  + x z  z ,
z∈{0,1}n z =z 

it must be achieved that |x z  |2 gets close to 1 (so-called amplitude amplification), i.e.


a measurement will result with high probability in z . This is realized by a unitary
transformation (the “ansatz”—see Sect. 6.1) the construction of which involves the
cost function and rotations that emphasize components x z of the state ψ the more z
contributes to the cost; these angles are the parameters of the ansatz that are classically
optimized. For all the details see Farhi et al. (2014).

6.4 Computing Maximum Cuts via Eigenvalues

Let G = (N, E) be a graph with n nodes (i.e. card(N) = n), and let S, T be a partition
of N. Such a partition can be described by a bit vector x ∈ {0, 1}n as follows:
xi = 1 ⇐⇒ i ∈ S and xi = 0 ⇐⇒ i ∈ T . There are 2n different bit vectors, i.e.
partitions. For each partition the number wd of edges between nodes of different sets
of the partition as well as the number ws of edges between nodes of the same set of the
partition is determined. Next, x ∈ {0, 1}n is considered as the binary representation
of a natural number x ∈ N, and wx := 21 (ws − wd ) is associated with this number x.
Finally, the matrix M ∈ C2 ×2 is defined where m i j = 0 ⇐⇒ i = j and m ii = wi ,
n n

i.e. the i-th row of M has as diagonal element the number wi corresponding to the
partition i ∈ {0, 1}n , and all other elements of the row are zero.
M is a diagonal matrix and, thus, Hermitian, i.e. all eigenvalues are real numbers.
Each vector of the computational basis is an eigenvector of M. It can be proven that
each vector of the computational basis that is an eigenvector of the lowest eigenvalue
defines a maximum cut. Thus, determining the lowest eigenvalue of the matrix M
and one of its eigenvectors means to determine a maximum cut of the corresponding
graph. Note that an analogue construction can be made for arbitrary weighted graphs.
Using eigenvalues to solve the maximum cut problem is based on Poljak and Rendl
(1995).

6.5 VQE

A variational hybrid algorithm (called variational quantum eigensolver VQE) to

determine the lowest eigenvalue of an operator has been first developed by Peruzzo
et al. (2014). This work has been extended towards a variational hybrid algorithm
From Digital Humanities to Quantum Humanities … 35

to approximate all eigenvalues of an operator (Higgott et al., 2019). As a conse-

quence of Sect. 6.4, the maximum cut problem can be solved by using VQE. In addi-
tion, dimension reduction based on principal component analysis (PCA) requires the
computation of all eigenvalues of a certain matrix, i.e. this algorithm is key for our
work.
The basic idea of VQE is as follows: according to the Raleigh-Ritz principle
(Yuan et al., 2019), the lowest eigenvalue λmin of a Hermitian operator O satisfies
the following equation:

ψ|O|ψ|
λmin ≤ for any |ψ = 0.
ψ, ψ

With ψ|O|ψ = O|ψ for each state |ψ (i.e. ψ = 1), the expectation value
of O provides an upper bound of the lowest eigenvalue of O: λmin ≤ O|ψ . Thus, if
a state |ψ can be found that minimizes the expectation value of O, O|ψ is close to

the lowest eigenvalue of O. (Note that O|ψ   = λmin holds for an eigenvector |ψ 
of λmin ).  
 
In order to find such a ψ̂ , series of states |ψ( p1 , ..., pk ) are iteratively
constructed that depend on parameters p1 , ..., pk ; in Fig. 16 the unitary operator
U ( p1 , ..., pk ) prepares this state. The expectation value O|ψ( p1 ,..., pk ) is measured
for each parameterized state, and the measurement result is subject to a clas-
sical optimization algorithm that determines new parameters p1 , ..., pk that reduce
the expectation value further; in Fig. 16 the function F(m) is this expectation

value. These iterations approximate |ψ , i.e. the final set of parameters determine
O|ψ( p1 ,..., pk ) ≈ λmin and |ψ( p1 , ..., pk ) is an approximation of the corresponding
eigenvector.
An efficient measurement of the expectation values is done as follows: Each
Hermitian operator O can be written as a linear combination of “simpler” Hermitian
operators Oi , i.e. O = xi Oi . Here, “simpler” means that Oi is a combination of
Pauli operators. The expectation value Oi v of such a combination can be efficiently
measured in a quantum computer (Hamamura & Imamichi, 2019), and it is Ov =
xi Oi v , i.e. the expectation value of O can be easily computed classically based
on the measured Oi v —i.e. in Fig. 16 the function F computes xi Oi . For all the
details see Peruzzo et al. (2014).

7 QHAna: A Quantum Humanities Analysis Tool

As described in Sects. 5 and 6 there are several machine learning techniques for which
first implementations are available on quantum computers. Some can be used to just
gain initial experiences, others already show improvements in terms of speedup or
precision. Bringing together the aspects and concepts outlined in the above sections,
QHAna (2021) was developed. QHAna is our quantum humanities data analysis
36 J. Barzen

tool that aims for several goals: (i) There is the content-related goal of supporting
the identification of patterns by data analysis. A feasibility study is provided by the
use case (Sect. 3) that focuses on analysing costume data to improve the under-
standing of vestimentary communication in films. (ii) By performing this analysis
by classical and hybrid quantum–classical algorithms a comparison of both methods
is supported, allowing to assess the benefits of using quantum algorithms. (iii) This
comparison supports the goal of improving the understanding of potential benefits
quantum computing may provide for digital humanities research (see Sect. 2.2). (iv)
Additionally, QHAna allows the integration of heterogeneous tools from different
quantum computing vendors in a single analysis pipeline. (v) Thus, one of the overar-
ching goals of QHAna is to provide easy access to people without quantum computing
background to gain first application knowledge in the domain of quantum humanities.
In the following QHAna will be introduced by this core objectives and functions.

7.1 Support the Identification of (Costume) Patterns

The primary goal of QHAna is to support analysing data when aiming at the iden-
tification of patterns based on the data analysis pipeline defined in Sect. 4.1. By
providing easy access to the analysis methods and techniques outlined above, QHAna
is designed to e.g., improve the understanding of vestimentary communication in
films by supporting the identification of costume patterns in the MUSE dataset (see
Sect. 3.3). Note that QHAna is data independent and therefore, not limited to the
data of our use case. Data from other application areas like from the MUSE4Music
project (Barzen et al., 2016) is planned to be analysed.
Based on initial question like (i) “which base elements are used to communicate
specific stereotypes?”, (ii) “do certain age impression and attributes like colour or
condition communicate certain character traits within one genre?”, (iii) “and if this
is true, how to group new costumes to the suitable category when being captured?”
several analysis steps need to be performed. The upper part of Fig. 18 gives an
impression of the main menu that guides through the pipeline for analysing data. To
approach an answer to these sample questions, for example, question (ii) is addressed
by clustering algorithms, focusing on finding those costumes that achieve the same
effect, while the question (iii) is addressed by classification algorithms. For both, the
first step is about preparing the data depending on the requirements of the algorithms.
As this is the basis for all the other following steps, it will be described in more detail
how QHAna supports this step.
As Fig. 18 depicts, the tab “Data Preparation” allows to prepare the data depending
on the requirements of specific analysis algorithms. As described in Sect. 4.2, several
algorithms require numerical data and the MUSE data is mostly categorical data.
Therefore, the subtabs “Distance Space Calculation” and “One-Hot Encoding” as
shown in Fig. 18 provide different options to transform categorical data into numer-
ical ones. For our example we chose the “Distance Space Calculations” that supports
the approach described in Sect. 4.2: Here, a dropdown menu allows to choose the
From Digital Humanities to Quantum Humanities … 37

Fig. 18 Screenshot of QHAna “Data Preparation”

domain of the distance space. In our MUSE use case, the “Costume” data set is
chosen. Figure 18 depicts how to specify a view for analysing a defined subset of
the attributes of the MUSE data set. For this purpose, the attributes of interest of
our example question are chosen, e.g., “Dominant Color”, “Dominant Condition”,
“Dominant Character Trait”, “Dominant Age Impression”, and “Genre” are selected.
Furthermore, the user can define which element comparer (that allow to compare two
elements within an attribute category) and attribute comparer (that allow to compare
sets of elements within an attribute), or which aggregator type (that define how data
points are aggregated) and transformer type (that describe the function used to trans-
form similarity measures into distance measures), are most suitable for the use case,
as well as how empty attributes are to be treated (e.g. if a costume has no value for
the attribute category “Color” the missing element should be “ignored”). The table
at the top provides an overview of all selections made.
The tap “Documentation” (right of “Data Preparation”) allows to retrieve general
information about the currently processed data and its structure (e.g. attribute
lists, taxonomies), explanations about the components already implemented in
QHAna (e.g. different cluster or classification algorithms), references to software
dependencies as well as all relevant related papers used to implement the tool.
After computing the distance space, QHAna supports calculating corresponding
distance matrices (see Sect. 4.2). Here, the user can choose how many costumes
should be analysed and whether these costumes should be selected randomly from
the whole data set or custom specific. Figure 19 gives an example of such a distance
38 J. Barzen

Fig. 19 Screenshot of a
distance matrix

matrix of 25 costumes. Note that the visualization of the distance may become very
complex with an increasing number of input data and the resulting visualization may
become difficult to comprehend.
Distances between pairs of data elements (e.g. costumes) are represented by a
colour code: The distance value of two costumes becomes a coefficient of the distance
matrix and this value is turned into a colour: dark blue corresponds to no distance at
all, while dark red indicates maximum distance. Each costume is represented by a
number (0–24 in Fig. 19). As the distance matrix depicts, some of the costumes are
very close to each other based on the selected attributes and others are very far away
(i.e. they are not really similar). For example, costume 5 is very close to costume
9 and costume 15, while it has hardly any similarity with (i.e. it is far away from)
costume 11 and 22. For a more detailed comparison of these costumes the “Entity
Table” subtab of the “Overview” tab (see Fig. 20) allows to compare the costumes of
interest in detail by their attributes chosen. Figure 20 depicts costumes 5, 9, 11, 15
and 22 stressing that 5, 9 and 15 have the same attribute values, while costume 11 and
15 differ in their attribute values. As most costumes have more than just 5 attributes,
changing the input parameters in the costume distance space definition may have a
deep impact on the similarity measures of these costumes. Also, experts sometimes
need to verify the concrete costumes manually. For this purpose, the entity table
provides for each costume links to the MUSE-repository (the film, the character, the
concrete costume) allowing further investigation of the costumes with all the detailed
information, in particular it provides the visual representations of the costumes from
the MUSE-repository.
After the data is prepared the next step allows to—if needed—perform “Feature
Engineering” to reduce the number of relevant attributes. The tab “Feature Engi-
neering” provides several techniques for dimension reduction and feature extraction
such as autoencoders, embeddings, or PCA. This provides the basis to use clustering
From Digital Humanities to Quantum Humanities … 39

Entity Table
[5,9,11,15,22] Table Entities of List

Reference Reference Reference Dominant Dominant Dominant Age

ID Dominant Color Genre
Film Character Costume Condition Character Trait Impression
['High school
5 Link Link Link ['Light Blue'] ['Ironed'] ['Active', 'Strong'] ['40s'] comedy',
'Love movie']
['High school
9 Link Link Link ['Light Blue'] ['Ironed'] ['Active', 'Strong'] ['40s'] comedy',
'Love movie']
['Drama', 'High
11 Link Link Link ['Dark Brown'] ['Worn Out'] ['Good', 'Passive'] ['Youth'] school comedy']

['High school
15 Link Link Link ['Light Blue'] ['Ironed'] ['Active', 'Strong'] ['40s'] comedy',
'Love movie']
['Good', 'Passive', ['High school
22 Link Link Link ['Dark Brown'] ['Worn Out'] ['Youth']
Weak'] comedy']

Fig. 20 Screenshot of “Entity Table”

techniques that allow to group those costumes together that have the same effect,
e.g. communicating a Sheriff by certain base elements like a Sheriff star, Cowboy
boots and so on (Barzen, 2018). As there are new costumes captured on a daily
basis and they need to be classified in terms of being mapped to the costume pattern
they contribute to, running classification algorithms is very promising for the MUSE
use case. All those steps supported by QHAna aim at improving the understanding
of vestimentary communication in films. How feature engineering, clustering, and
classification are supported is unfolded in what follows.

7.2 Comparing Classical and Quantum Machine Learning

Algorithms

To get the optimal results when approaching an answer to a stated question the
different analysis steps are performed based on different algorithms, often in parallel
by different classical implementations as well as by different quantum–classical
implementations. This allows comparing the different results of classicals machine
learning algorithms amongst each other, comparing different results of quantum
machine learning algorithms amongst each other, and comparing results from classi-
cals machine learning algorithms with results from quantum machine learning algo-
rithms. This enables finally selecting the most suitable algorithm for the problem
at hand as well as improving the application of an algorithms when comparing the
results and optimizing iteratively the hyperparameters. Please note that due to the
limitations of today’s quantum computers only small sets of data can be processed
when using quantum machine learning algorithms.
40 J. Barzen

To outline how such a comparison is enabled by QHAna, two examples are

provided, one from “Feature Engineering” where a different technique is used to
get so a distant matrix and one from “Clustering” where the comparison of different
implementations of maximum cut algorithms (see Sect. 6.2–6.5) is outlined.
Classical and Quantum-based Approaches to Distance Matrices
As described in Sects. 5.4 and 5.5, autoencoders can be used for dimension reduc-
tion. Therefore, four implementations of autoencoders have been used to reduce
the dimension of MUSE data. We realized two autoencoders classically (based on
PyTorch (2021) and TensorFlow (2021)) and two autoencoders in a hybrid manner:
in the latter case, a classical autoencoder is first used to reduce all input data to a
three-dimensional feature space, while in the succeeding step (i) a quantum inspired
autoencoders (using TensorFlow Quantum (2021)) and (ii) a quantum autoencoders
(using Qiskit, see Fig. 15 for an impression of the circuit) are used to further reduce the
dimension to two (for more details see Barzen et al. (2021). Please note that those
autoencoders are currently being integrated into QHAna). By pairwise measuring
the fidelity of the quantum states and applying the Fubini-Study metric (Biamonte,
2020) to the measured data, the distances between the resulting quantum states can
be computed.
The left side of Fig. 21 gives an example of the distance values of 10 costumes
(with the same 5 attributes selected as in Fig. 18) as result of a hybrid quantum
inspired autoencoder. As before, 0 (dark blue) corresponds to the smallest distance
and 1 (bright red) to the largest possible distance (note that based on the actual
data, the largest distance is about 0.30). What can be seen is that the costumes 0
to 4 are highly similar to each other but rather different from costumes 5 to 9 and
that the costumes 5 to 9 are again highly similar to each other. The right side of
Fig. 21 shows the distances of the same 10 costumes determined by using the Wu
and Palmer similarity measure (see 4.2 and 7.1). What can be seen is that quite similar

Fig. 21 Distance matrix as result of a hybrid autoencoder (left side) and computed via the Wu and
Palmer similarity (right side)
From Digital Humanities to Quantum Humanities … 41

results have been achieved by using these two different techniques: The statistical
approach (autoencoder) on the one hand and the approach using taxonomies (Wu-
Palmer similarities) on the other hand, identified the same costumes as being close
to each other. This hints to a potential costume pattern. Also, it provides first insights
on comparing a classical and a quantum-based approach to compute distance values.
Classical and Quantum-based Approaches to Clustering
Implementations of both, classical clustering algorithms and first quantum–clas-
sical clustering algorithms are available under the tab “Clustering” in QHAna.
Currently, this includes an implementation of the OPTICS cluster algorithm (Ankerst
et al., 1999), four different implementations of the maximum cut (a.k.a. maxcut)
algorithm (see Sect. 6.2) and four different implementations of the k-means algo-
rithm (Arthur & Vassilvitskii, 2007; Khan et al., 2019). The implementations of
the maximum cut range from a classical naive implementation (“classicNaiveMax-
Cut”), to two approximative implementations (“sdpMaxCut” based on semidefinite
programming (Boumal, 2015) and “bmMaxCut” using a Bureir-Monteiro solver
(Boumal et al., 2016)), to a quantum–classical implementation (“qaoaMaxCut”) that
is based on QAOA (see Sect. 6.3) and implemented in Qiskit (see Max Cut, 2021).
Figure 22 allows a comparison of the results of using different maximum cut
implementations. Input are the same 10 costumes as in Fig. 21, using the default
values of the hyperparameters per algorithm provided by QHAna. Diagram 1 of
the figure shows the costumes and their distances (upper left corner) in an MDS-
based embedding. The subtab “Embedding” of “Feature Engineering” supports to
use MDS as one implemented approach to map the distance measures of the chosen
costumes with their chosen attributes to a metrical feature space. As a result of the
embedding via MDS the data is mapped into Rn in such a way that the original
distances of the data elements are nearly the distances in the feature space (see
Sect. 4.2). The other three diagrams present clustering results achieved by different
maximum cut implementations namely “qaoaMaxCut” (Diagram 2), “sdpMaxCut”
(Diagram 3), and “bmMaxCut” (Diagram 4). It can be seen that the two approximative
implementations (Diagrams 3 and 4) have identical results: The first cluster (red
circles) contains the costumes 0–4, while the second cluster (blue circles) contains
the costumes 5–9. The result of the quantum–classical implementation (part 2) is
rather different: cluster one (red circles) contains the costumes 3, 4, 7, and 9 while
the second cluster (blue circles) contains the costumes 0, 1, 2, 5, 6, and 8. In providing
easy access to several implementation the comparison of such results is supported
by QHAna. In addition, the results can be manually verified using the entity table of
QHAna (see Fig. 20) as outlined in Sect. 7.2. The entity table allows the evaluation
of the attributes of the costumes that are part of the clusters and provides all the
details of the costumes by linking them to the MUSE-repository. This can be used
to improve the understanding of the benefits quantum computing can have, e.g. for
quantum humanities research.
42 J. Barzen

Fig. 22 Values of 10 costumes embedded via MDS (1) and their clustering results of different
maximum cut implementations (2–4)

7.3 Improve the Understanding of the Benefits of Quantum

Computing

In a fist attempt when taking a closer look at the comparison of Fig. 22 one might
decide for the classical implemented clustering for further analysis as it seems more
precise. But QHAna also allows to improve the results by adjusting the hyperpa-
rameters of the algorithms. For the results of Fig. 22 this would include to poten-
tially improve the results of the quantum–classical implementation of the algo-
rithm as shown in Fig. 22 to equal the results of the classical implementation.
Thus, QHAna lists the selectable parameters (a.k.a. hyperparameters) specific to
the chosen algorithms, together with descriptions that support selecting the right
parameters. Figure 23 gives an impression of the hyperparameters to be selected for
the quantum–classical implementation of a maximum cut algorithm including the
user token required to run the algorithm on the IBM Q systems.
Figure 24 outlines the influence of the maximum number of iterations performed
by the algorithm, while all other parameters keep the default settings of QHAna.
As can be seen, increasing the number of iterations seems to improve the results:
Diagram 1 is performed with only one iteration, while the number of maximum
From Digital Humanities to Quantum Humanities … 43

Fig. 23 Screenshot of “Clustering”

iterations is increased by 50 each time (Diagram 2: 50 trials, Diagram 3: 100 trials,

and Diagram 4: 150 trials). Thus, the results approach those achieved by the classical
maximum cut implementations up to the identification of the same clusters when
performed with a maximum of 150 trials (Diagram 4).
By enabling easy comparison of different techniques and implementations of
classical and quantum–classical algorithms during the data analysis process, QHAna
contributes to a better understanding of the potential benefits (Sect. 2.2) that quantum
computing can provide for quantum humanities research. This includes as described
above identifying optimal hyperparameters, e.g., approximating the results of the
quantum–classical implementation and the classical ones, which then can be applied
to larger use cases. But it also includes identifying already available advantages of
applying quantum machine learning. An example of these potential advantages of
quantum machine learning, is the precision that the quantum kernel estimation (QKE)
method for SVM (Havlicek et al., 2018) can achieve.
Currently the tab “Classification” of QHAna provides three implementation of
SVM (see Sect. 5.3) methods (“classicSklearnSVM” an implementation of classical
support vector machines, “qkeQiskitSVM” an implementation using the quantum
kernel estimation (QKE) variant of the SVM, and an implementation of an variational
quantum SVM (Havlicek et al., 2018)) and two classifiers based on neural networks
(an implementation based on a classical neural network, an implementation based
on a hybrid quantum–classical neural network). Figure 25 gives an example of the
44 J. Barzen

Fig. 24 Clustering results of the quantum–classical maximum cut implementations when changing
the number of maximum iterations performed

Fig. 25 Classification results of the classically implemented SVM (left) and of the QKE based
implemented SVM (right)
From Digital Humanities to Quantum Humanities … 45

results of the classically implemented SVM (Diagram 1) and of the QKE based SVM
(Diagram 2). Input for this example are 30 costumes and their distance values based
on the attributes stereotype and role relevance embedded by MDS, and classified as
“positive” or “negative” based on another training set for the classifier. As can be
seen in Fig. 25 these classes are not linear separable. Thus, finding the right classifier
is a challenging task. Figure 25 depicts that (by using the default parameters QHAna
provides) the result of the QKE variant (Diagram 2) is much more precise (accuracy
of 94% versus 89%) than the result of the classical variant (Diagram 1).

7.4 Integration of Heterogeneous Tools

The prototype of QHAna has a modular structure. Thus, it can easily be extended by
further algorithm implementations. Various implementations are already available
(see Sect. 7.1–7.3). Because QHAna can also serve REST APIs the effective inte-
gration of components accessible via such APIs is straightforward; this way, further
implementations of techniques described in the theoretical sections of this contribu-
tion into QHAna is achieved: This includes—but is not restricted to—extending the
feature engineering step by quantum principal component analysis, for example, or
add quantum restricted Boltzmann machines, using more clustering algorithms and
integrating more classification algorithms. Also, algorithms used in other domains
(like HHL, VQE and so on) can be integrated into QHAna.
Another significant advantage of QHAna is that it provides comfortable access
to different backends to run the algorithms on. Figure 23 shows, for example, that
there are different simulators and IBM quantum computing backends integrated
and selectable already. Currently we are integrating more backends like PennyLane
(PennyLane, 2021) and TensorFlow.

7.5 Provide Easy Access to Quantum Computing

Working with quantum computers still requires advanced knowledge in physics and
mathematics and is therefore a great challenge for research in the field of digital
humanities. As there are promising potentials in using the quantum computer in this
area (see Sect. 2.2), it is of interest to provide a simple, straightforward access to
this new technology without having to deal with all the mathematical, physical, and
algorithmic details. This is what QHAna aims for. As Fig. 23 depicts, even very
complex quantum–classical algorithms are provided in an abstract manner allowing
to adjust their hyperparameters, but default values of these parameters are set. Thus,
these algorithms and their benefits can be applied by someone not familiar with qubits,
QPU connectivity, quantum circuits, and so on. In Sects. 5 and 6 we provided the
core concepts to enable an understanding of the underlying ideas of the (classical and
quantum) algorithms used in QHAna, but no details about classical algorithms nor
46 J. Barzen

quantum algorithms are given. These algorithms and their implementations evolve
rapidly and potentially change quite frequently. Their details are not the focus of
QHAna. Instead, providing access to a complex new technology to learn about and
participate in the advantages given for further research is at the heart of QHAna.

8 Conclusion and Outlook

As outlined in this contribution there are a lot of potentials in applying quantum

computers to research done in the humanities but there is also a long way to go to
fully benefit from them. Thus, the benefits and challenges of quantum computers
were stressed and—based on our use case MUSE—first application knowledge for
further quantum humanities is provided.
As the use case is focused on machine learning, a data analysis pipeline for
machine learning tasks was introduced (that is especially useful for detecting patterns
in a domain) and core concepts that are promising to be applied in the fields of the
humanities were introduced and discussed. Therefore, artificial neural networks were
described by providing a mathematical definition of neurons, neural networks, and
perceptrons. Also, their use for restricted Boltzmann machines and autoencoders as
well as first realizations of these on quantum computers were described. Quantum
algorithms are often hybrid; therefore, the main idea of variational hybrid quantum–
classical algorithms was sketched as well as an application for clustering based on
hybrid techniques like quantum approximate optimization algorithm (QAOA) and
variational quantum eigensolver (VQE) were discussed.
To provide a straightforward access to the described techniques an analysis tool
called QHAna has been introduced. The presented prototype has a modular architec-
ture and implementation; thus, it can be extended easily. We plan to realize an import
and export functionality such that—after key steps of the supported pipeline—the
intermediate results can be exported, and results from other steps can be imported.
Especially, this will allow to consider results from external algorithms (i.e. algorithms
not available via QHAna) in the pipeline as well as to pass results from QHAna for
further processing to external algorithms. We also plan to integrate workflow features
such that users may define their own sequence of invoking algorithm implementations
(those provided by QHAna as well as external ones) to support custom pipelines of
data analysis in a multitude of domains: this will significantly improve the flexibility
of QHAna.
As quantum computers are continuously improving and the number of qubits is
expected to increase constantly and significantly (Gambetta, 2020), it is of great
importance to develop application knowledge in the domain of quantum computing
at an early stage. Therefore, we envision to generalize the application knowledge
that can be extracted from our approach to quantum humanities patterns to make
it reusable for different use cases in other application domains of the humanities.
Providing solution knowledge on such a promising and innovative topic as quantum
computing is already relevant in itself. First samples for pattern languages in this
From Digital Humanities to Quantum Humanities … 47

domain can be seen (Leymann, 2019; Weigold et al., 2020, 2021). It is of particular
relevance to provide application knowledge in a field where the mathematical and
physical basics required to use this technology cannot be taken for granted, but fields,
which are essential for the critical reflection of digital methods, such as the digital
humanities. Especially, if a quantum computer is not only to be used as a tool like for
quantum machine learning but for thinking about totally new questions that have not
yet been tackled at all, perhaps not even identified or considered, knowledge about
the corresponding concepts, methods, and application potentials is key. To identify
and explore these possible new application areas is an exciting task and needs to
be started by building knowledge on how quantum computers can be applied to
problems stated in the humanities.

Acknowledgements I am very grateful to Frank Lemann for discussing several subjects of this
chapter. Also, I would like to thank Felix Truger, Philipp Wundrack, Marcel Messer, Daniel Fink
and Fabian Bühler for their valuable input and implementing several aspects of our use case.
This work was partially funded by the BMWi project PlanQK (01MK20005N) and the Terra
Incognita project Quantum Humanities funded by the University of Stuttgart.

References

Aharonov, D., van Dam, W., Kempe, J., Landau, Z., Lloyd, S., & Regev, O. (2008). Adiabatic
quantum computation is equivalent to standard quantum computation. SIAM Review, 50(4), 755–
787.
Alexander, C., Ishikawa, S., Silverstein, M., Jacobson, M., Fiksdahl-King, I., & Angel, S. (1977).
A pattern language: Towns, buildings, constructions. Oxford University Press.
Amin, M. A., Andriyash, E., Rolfe, J., Kulchytskyy., B., & Melko, R. (2018). Quantum Boltzmann
machine. Physical Review X, 8, 021050.
Ankerst, M., Breunig, M. M., Kriegel, H., & Sander, J. (1999). OPTICS: ordering points to identify
the clustering structure. In ACM SIGMOD international conference on Management of data.
ACM Press.
Anschuetz, E. R., Olson, J. P., Aspuru-Guzik, A., & Cao, Y. (2018). Variational quantum factoring.
arXiv:1808.08927
Arthur, D., & Vassilvitskii, S. (2007). “k-means++: The advantages of careful seeding”. In Proceed-
ings of the eighteenth annual ACM-SIAM symposium on Discrete algorithms (pp. 1027–1035).
Society for Industrial and Applied Mathematics.
Barzen, J. (2013). Taxonomien kostümrelevanter Parameter: Annäherung an eine Ontologisierung
der Domäne des Filmkostüms. Technical Report No. 2013/04, Universität Stuttgart. https://www.
iaas.uni-stuttgart.de/publications/TR-2013-04_Taxonomien-_kostuemrelevanter_Parameter.pdf
Barzen, J. (2018). Wenn Kosüme sprechen—Musterforschung in den Digital Humanities am
Beispiel vestimentärer Kommunikation im Film. Dissertation, University of Cologne. https://
kups.ub.uni-koeln.de/9134/
Barzen, J., Breitenbücher, U., Eusterbrock, L., Falkenthal, M., Hentschel, F., & Leymann, F. (2016).
The vision for MUSE4Music. Applying the MUSE method in musicology. In: Computer science—
Research and development. Advancements of service computing: Proceedings of summer SoC
2016 (Vol. 32, pp. 3–4).
Barzen, J., Falkenthal, M., & Leymann, F. (2018). Wenn Kostüme sprechen könnten: MUSE—Ein
musterbasierter Ansatz an die vestimentäre Kommunikation im Film. In P. Bockwinkel, B. Nickel,
G. Viehhauser (Eds.), Digital humanities. Perspektiven der Praxis, Frank & Timme (pp. 223–241).
48 J. Barzen

Barzen, J., & Leymann, F. (2014). Costume languages as pattern languages. In Proceedings of
Pursuit of Pattern Languages for Societal Change—Preparatory Workshop.
Barzen, J., & Leymann, F. (2019). Quantum humanities: A vision for quantum computing in digital
humanities. In SICS software-intensive cyber-physical systems (pp. 1–6). https://doi.org/10.1007/
s00450-019-00419-4
Barzen, J., & Leymann, F. (2020). Quantum humanities: A first use case for quantum-ML in media
science. In ISAAI’19 Proceedings—Artificial Intelligence, Digitale Welt (Vol. 4(1), pp. 102–103).
Barzen, J., Leymann, F., Falkenthal, M., Vietz, D., Weder, B., & Wild, K. (2021). Relevance of
near-term quantum computing in the cloud: A humanities perspective. In Cloud Computing and
Services Science, Communications in Computer and Information Science. Springer (to appear).
Bennett, K. P., & Campbell, C. (2000). Support vector machines: Hype or Hallelujah? ACM
SIGKDD, 2(Issue 2).
Berry, D. M. (2012). Understanding digital humanities. Palgrave Macmillan.
Berry, D. M., & Fagerjord, A. (2017). Digital humanities: Knowledge and critique in a digital age.
Polity Press.
Bhaskar, M. K., Hadfield, S., Papageorgiou, A., & Petras, I. (2015). Quantum algorithms and circuits
for scientific computing. arXiv:1511.08253
Biamonte, J. (2020). Lectures on quantum tensor networks. arXiv:1912.10049
Bondarenko, D., & Feldmann, P. (2020). Quantum autoencoders to denoise quantum data. Physical
Review Letters, 124, 130502.
Boumal, N. (2015). A Riemannian low-rank method for optimization over semidefinite matrices
with block-diagonal constraints. arXiv:1506.00575
Boumal, N., Voroninski, V., & Bandeira, A. S. (2016). The non-convex Burer-Monteiro approach
works on smooth semidefinite programs. In Advances in Neural Information Processing Systems
29 (NIPS 2016).
Bravo-Prieto, C., LaRose, R., Cerezo, M., Subasi, Y., Cincio, L., & Coles, P. J. (2020). Variational
quantum linear solver. arXiv:1909.05820
Brown, K. (Ed.) (2020). The Routledge companion to digital humanities and art history. Routledge.
Burges, Ch. J. C. (1998). A tutorial on support vector machines for pattern recognition. Data Mining
and Knowledge Discovery, 2, 121–167.
Cao, Y., Guerreschi, G. G., & Aspuru-Guzik, A. (2017). Quantum neuron: an elementary building
block for machine learning on quantum computers. arXiv:1711.11240
Chen, G., & Srihari, S. H. (2015). Restricted Boltzmann machine for classification with hierarchical
correlated prior. arXiv:1406.3407v2
Cox, T. F., & Cox, M. A. A. (2001). Multidimensional scaling. Chapman and Hall.
Crooks, G. E. (2018). Performance of the quantum approximate optimization algorithm on the
maximum cut problem. arXiv:1811.08419v1
da Silva, I. N., Spatti, D. H., Flauzino, R. A., Liboni, L. H. B., & dos Reis Alves, S. F. (2017).
Artificial neural networks: A practical course. Springer International Publishing.
Denil, M., & de Freitas, N. (2011). Toward the implementation of a quantum RBM. Oxford University
Research Archive.
Deutsch, D. (1985). Quantum theory, the Church-Turing principle, and the universal quantum Turing
machine. Proceedings of the Royal Society of London, A400, 97117.
DHd. (2021). Digital Humanites im deutschspachingen Raum. https://dig-hum.de/
DHQ. (2021). Digital humanities quarterly. http://www.digitalhumanities.org/dhq/.
Dickel, C. (2018). A cloud quantum computer business plan. In: QuTech. http://blog.qutech.nl/
index.php/2018/07/18/a-cloud-quantum-computer-business-plan/
Ding, Y., Lamata, L., Sanz, M., Chen, X., & Solano, E. (2019). Experimental implementation of a
quantum autoencoder via quantum adders. Advanced Quantum Technologies, 1800065, 2019.
Dunjko, V., Taylor, J. M., & Briegel H. J. (2016). Quantum-enhanced machine learning. arXiv:
1610.08251
Egger, D. J., Marecek, J. & Woerner, S. (2020). Warm-starting quantum optimization. arXiv:2009.
10095
From Digital Humanities to Quantum Humanities … 49

Falkenthal, M., Barzen, J., Breitenbücher, U., Brügmann, S., Joos, D., Leymann, F., & Wurster, M.
(2016a). Pattern research in the digital humanities: How data mining techniques support the iden-
tification of costume patterns. In Computer Science—Research and Development. Advancements
of Service Computing: Proceedings of SummerSoC 2016 (Vol. 32, pp. 3–4).
Falkenthal, M., Barzen, J., Breitenbücher, U., Fehling, C., Leymann, F., Hadjakos, A., Hentschel,
F., & Schulze, H. (2016b). Leveraging pattern applications via pattern refinement. In Proceed-
ings of the International Conference on Pursuit of Pattern Languages for Societal Change
(PURPLSOC).
Falkenthal, M., Barzen, J., Dörner, S., Elkind, V., Fauser, J., Leymann, F, & Strehl, T. (2015).
Datenanalyse in den Digital Humanities—Eine Annäherung an Kostümmuster mittels OLAP
Cubes. In Datenbanksysteme für Business, Technologie und Web (BTW) (Vol. 16). Fachtagung
des GI-Fachbereichs “Datenbanken und Informationssysteme” (DBIS).
Farhi, E., Goldstone, J., & Gutmann, S. (2014). A quantum approximate optimization algorithm.
In MIT-CTP/4610.
Farhi, E., & Neven, H. (2018). Classification with quantum neural networks on near term processors.
arXiv:1802.06002v2
Fehling, C., Barzen, J., Falkenthal, M., & Leymann, F. (2014). PatternPedia—Collaborative Pattern
identification and authoring. In Proceedings of Pursuit of Pattern Languages for Societal
Change—Preparatory Workshop.
Gambetta, J. (2020). IBM’s roadmap for scaling quantum technology. In IBM Research Blog.
Retrieved September 15, 2020, from https://www.ibm.com/blogs/research/2020/09/ibm-qua
ntum-roadmap/
Grover, L. K. (1996). A fast quantum mechanical algorithm for database search. In Proceedings of
the twenty-eighth annual ACM symposium on Theory of computing (pp. 212–219).
Hamamura, I., & Imamichi, T. (2019). Efficient evaluation of quantum observables using entangled
measurements. arXiv:1909.09119v2
Haven, E., & Khrennikov, A. (2013). A Quantum social science. Cambridge University Press.
Havenstein, C., Thomas, D., & Chandrasekaran, S. (2018). Comparisons of performance between
quantum and classical machine learning. SMU Data Science Review, 1(4), Article 11.
Havlicek, V., Córcoles, A. D., Temme, K., Harrow, A. W., Kandala, A., Chow, J. M., & Gambetta,
J. M. (2018). Supervised learning with quantum enhanced feature spaces. arXiv:1804.11326
Higgott, O., Wang, D., & Brierley, S. (2019). Variational quantum computation of excited states.
Quantum, 3, 156.
Hinton, G. E., (2012). A practical guide to training restricted Boltzmann machines. In G. Montavon,
G. B. Orr, K. R. Müller (Eds.), Neural networks: Tricks of the trade. Lecture Notes in Computer
Science (Vol. 7700). Springer.
Hinton, G. E., & Zemel, R. S. (1994). Autoencoders, minimum description length and Helmholtz
free energy. In Advances in neural information processing systems (Vol. 6, pp. 3–10).
Jannidis, F., Kohle, H., & Rehbein, M. (Eds.) (2017). Digital humanities: Eine Einführung. J.B.
Metzler.
Jozsa, R. (2006). An introduction to measurement based quantum computation. NATO Science
Series, III: Computer and Systems Sciences. Quantum Information Processing-From Theory to
Experiment, 199, 137–158.
Kandala, A., Mezzacapo, A., Temme, K., Takita, M., Brink, M., Chow, J. M., & Gambetta, J. M.
(2017). Hardware-efficient variational quantum Eigensolver for small molecules and quantum
magnets. arXiv:1704.05018
Karalekas, P. J., Tezak, N. A., Peterson, E. C., Ryan, C. A., da Silva, M. P., & Smith, R. S. (2020). A
quantum-classical cloud platform optimized for variational hybrid algorithms. Quantum Science
and Technology, 5, 024003.
Khan, S. U., Awan, A. J., & Vall-Llosera, G. (2019). K-means clustering on noisy intermediate
scale quantum computers. arXiv:1909.12183
50 J. Barzen

Kidger, P., & Lyons, T. (2020). Universal approximation with deep narrow networks. In Machine
learning research (Vol. 125, pp. 1–22). Proc. 33rd Annual Conference on Learning Theory (COLT
2020).
Knill, E. (2007). Quantum computing with very noisy devices. arXiv:quant-ph/0410199 (preprint)
Kubat, M. (2017). An introduction to machine learning. Springer International Publishing AG.
Lamata, L., Alvarez-Rodriguez, U., Martın-Guerrero, J. D., Sanz, M., & Solano, E. (2019). Quantum
autoencoders via quantum adders with genetic algorithms. Quantum Science and Technology, 4,
014007.
Larochelle, H., Mandel, M., Pascanu, R., & Bengio, Y. (2012). Learning algorithms for the
classification restricted Boltzmann machine. Journal of Machine Learning Research, 13,
643–669.
LaRose, R. (2019). Overview and comparison of gate level quantum software platforms. Quantum,
3, 130. arXiv:1807.02500
Leymann, F. (2019). Towards a pattern language for quantum algorithms. In First international
workshop (QTOP 2019).
Leymann, F., & Barzen, J. (2020a). The bitter truth about gate-based quantum algorithms in the
NISQ era. Quantum Science and Technology, 5, 044007.
Leymann, F., & Barzen, J. (2020b). Pattern atlas. arXiv:2006.05120
Leymann, F., Barzen, J., Falkenthal, M., Vietz, D., Weder, B., & Wild, K. (2020). Quantum in the
cloud: Application potentials and research opportunities. In Proceedings of the 10 th International
Conference on Cloud Computing and Services Science (pp. 9–24). SciTePress.
Lubasch, M., Joo, J., Moinier, P., Kiffner, M., & Jaksch, D. (2020). Variational quantum algorithms
for nonlinear problems. Physical Review A, 101, 010301.
McClean, J. R., et al. (2017). OpenFermion: The electronic structure package for quantum
computers. arXiv:1710.07629
McClean, J. R., Romero, J., Babbush, R., & Aspuru-Guzik, A. (2016). The theory of variational
hybrid quantum-classical algorithms. New Journal of Physics, 18, 023023.
McCulloch, W. S., & Pitts, W. (1943). A logical calculus of the ideas immanent in nervous activity.
Bulletin of Mathematical Biophysics, 5, 115–133.
Michielsen, K., Nocon, M., Willsch, D., Jin, F., Lippert, T., & De Raedt, H. (2017). Benchmarking
gate-based quantum computers. Computer Physics Communications, 220, 44–55.
MUSE. (2021). Muster Suchen und Erkennen. https://www.iaas.uni-stuttgart.de/en/projects/muse/
MUSE GitHub. (2021). https://github.com/Muster-Suchen-und-Erkennen/muse-docker
National Academies of Sciences, Engineering, and Medicine. (2019). Quantum computing: progress
and prospects. The National Academies Press. https://doi.org/10.17226/25196
Nielsen, M., & Chuang, I. L. (2010). Quantum computation and quantum information. Cambridge
University Press.
Nocedal, J., & Wright, S. J. (2006). Numerical optimization. Springer Science + Business Media,
LLC.
Novikoff, A. (1962). On convergence proofs for perceptrons. In Proc. Mathematical Theory of
Automata, Vol. 12, pp. 615–622.
Otterbach, J. S., et al. (2017). Unsupervised machine learning on a hybrid quantum computer. arXiv:
1712.05771v1
Palmer, M. (2020). Quantum computers will create better versions of Alexa and Siri. In: Sifted.
Retrieved December 10, 2020, from https://sifted.eu/articles/quantum-computers-nlp/
PennyLane. (2021). https://pennylane.ai/
Pepper, A., Tischler, N., & Pryde. G. J. (2019). Experimental realization of a quantum autoencoder:
The compression of qutrits via machine learning. Physical Review Letters, 122, 060501.
Peruzzo, A., McClean, J., Shadbolt, P., Yung, M.-H., Zhou, X.-Q., Love, P. J., AspuruGuzik,
A., & O’Brien, J. L. (2014). A variational eigenvalue solver on a quantum processor. Nature
Communications, 5, 4213.
Poljak, S., & Rendl, F. (1995). Solving the max-cut problem using eigenvalues. Discrete Applied
Mathematics, 62, 249–278.
From Digital Humanities to Quantum Humanities … 51

Preskill, J. (2018). Quantum computing in the NISQ era and beyond. Quantum, 2, 79.
PyTorch. (2021). https://pytorch.org/
QH. (2021). Quantum humanities. https://www.iaas.uni-stuttgart.de/forschung/projekte/quantum-
humanities/
QHAna. (2021). Quantum humanities data analysis tool. https://github.com/UST-QuAntiL/qhana
Qiskit. (2021). https://qiskit.org/
Qiskit Max Cut. (2021). https://qiskit.org/documentation/locale/de_DE/tutorials/optimization/6_e
xamples_max_cut_and_tsp.html
Raz, R., & Tal, A. (2018). Oracle separation of BQP and PH. Iin: Electronic colloquium on
computational complexity, Report No. 107.
Rehbein, M., & Sahle, P. (2013). Digital Humanities lehren und lernen. Modelle, Strategien,
Erwartungen. In: H. Neuroth, N. Lossau, A. Rapp (Eds.), Evolution der Informationsinfrastruktur
Kooperation zwischen Bibliothek und Wissenschaft. Verlag Werner Hülsbusch.
Reiners, R. (2014). An evolving pattern library for collaborative project documentation. Disserta-
tion, University of Aachen.
Rieffel, E., & Polak, W. (2011). Quantum computing: A gentle introduction. The MIT Press.
Romero, J., Olson, J. P., & Aspuru-Guzik, A. (2017). Quantum autoencoders for efficient
compression of quantum data. Quantum Science Technology, 2, 045001.
Rønnow, T., Wang, Z., Job, J., Boixo, S., Isakov, S. V., Wecker, D., Martinis, J. M., Lidar, D. A., &
Troyer, M. (2014). Defining and detecting quantum speedup. Science, 345(6195).
Rosenblatt, F. (1958). The perceptron: A probabilistic model for information storage and
organization. Psychological Review, 65(6).
Sahle, P. (Ed.) (2019). DHd 2019. Digital humanities: Multimedial & multimodal. Konferenzab-
stracts. Mainz/Frankfurt a M.
Schöch, C. (Ed.) (2020). DHd 2020. Spielräume Digital Humanities zwischen Modellierung und
Interpretation. Konferenzabstracts. https://doi.org/10.5281/zenodo.3666690
Schuld, M., & Petruccione, F. (2018). Supervised learning with quantum computers. Springer.
Schuld, M., Sinayskiy, I., & Petruccione, F. (2014). The quest for a Quantum Neural Network.
arXiv:1408.7005v1
Schumm, D., Barzen, J., Leymann, F., & Ellrich, L. (2012). A pattern language for costumes in films.
In Proceedings of the 17th European Conference on Pattern Languages of Programs (EuroPLoP
2012).
Shaydulin, R., Safro, I., & Larson, J. (2019). Multistart methods for quantum approximate
optimization. arXiv:1905.08768
Shor, P. W. (1995). Polynomial-time algorithms for prime factorization and discrete logarithms on
a quantum computer. arXiv:quant-ph/9508027
Sierra-Sosa, D., Arcila-Moreno, J., Garcia-Zapirain, B., Castillo-Olea, C., & Elmaghraby, A. (2020).
Dementia prediction applying variational quantum classifier. arXiv:2007.08653
Skiena, S. S. (2017). The data science design manual. Springer International Publishing AG.
Smolensky, P. (1986). Information processing in dynamical systems: Foundations of harmony
theory. In D. E. Rumelhart & J. L. McLelland (Eds.), Parallel distributed processing: explorations
in the microstructure of cognition, Volume 1: Foundations (pp. 194–281). MIT Press.
Tacchino, F., Macchiavello, C., Gerace, D., & Bajoni, D. (2018). An artificial neuron implemented
on an actual quantum processor. arXiv:1811.02266v1
TensorFlow. (2021). https://www.tensorflow.org/
TensorFlow Quantum. (2021). https://www.tensorflow.org/quantum
Terras, M., Nyhan, J., & Vanhoutte, E. (2013). Defining digital humanities. A reader.
Tomczak, J. M. (2016). Learning informative features from restricted Boltzmann machines. Neural
Processing Letters, 2016(44), 735–750.
Vogeler, G. (Ed.) (2018). DHd 2018. Kritik der digitalen Vernunft. Konferenzabstracts.
Wan, K. H., Dahlsten, O., Kristjánsson, H., Gardner, R., & Kim, M. S. (2017). Quantum
generalisation of feedforward neural networks. npj Quantum Information, 3, 36.
52 J. Barzen

Weigold, M., Barzen, J., Breitenbücher, U., Falkenthal, F., Leymann, F., & Wild, K. (2020). Pattern
views: Concept and tooling for interconnected pattern languages. In S. Dustdar (Ed.), Service-
oriented computing. SummerSOC 2020. Communications in computer and information science
(Vol. 1310). Springer, Cham. https://doi.org/10.1007/978-3-030-64846-6_6
Weigold, M., Barzen, J., Leymann, F., & Salm, M. (2021). Data encoding patterns for quantum algo-
rithms. In The Hillside Group (Ed.), Proceedings of the 27th Conference on Pattern Languages
of Programs (PLoP’20) (to appear).
What is Digital Humanities? (2021). http://whatisdigitalhumanities.com/
Wiebe, N., Kapoor, A., & Svore, K. M. (2016). Quantum perceptron models. arXiv:1602.04799
Wu, Y., Wei, C., Qin, S., Wen, Q., & Gao, F. (2020). Quantum restricted Boltzmann machine
universal for quantum computation. arXiv:2005.11970v3
Wu, Z., & Palmer, M. (1994). Verb semantics and lexical selection. In ACL’94 Proceedings of the
32nd annual meeting on Association for Computational Linguistics.
Yuan, X., Endo, S., Zhao, Q., Li, Y., & Benjamin, S. (2019). Theory of variational quantum
simulation. Quantum, 3, 191.
Zhang, P., Li, S., & Zhou, Y. (2015). An algorithm of quantum restricted Boltzmann machine
network based on quantum gates and its application. Shock and Vibration Volume. Article ID
756969.
Zheng, X., Wu, Z., Meng, H., Li, W., & Cai, L. (2013). Feature learning with Gaussian restricted
Boltzmann machine for robust speech recognition. arXiv:1309.6176
Zurada, J. M. (1992). Introduction to artificial neural systems. West Publishing Company.
(All links have been last followed March 11, 2021)
Quantum Computing and Cognitive
Simulation

Martha Lewis

Abstract Cognitive Science is the study of the mind and how it relates to human
behaviour. As a field, it is highly interdisciplinary, bringing together theories from
Psychology, Neuroscience, Biology, and others. The field of cognitive science oper-
ates at a number of levels. On the one hand, some aspects of cognitive science look
at the behaviour of neurons, whereas other aspects look to explain human behaviour
at a more abstract level, seeking to explain human decision making or reasoning.
A key area of research in cognitive science is how to formalise human behaviours
around judgements of similarity, categorization, and decision making. In the field
of Physics, Quantum Mechanics has fundamentally altered our understanding of the
way in which particles behave. Quantum mechanics has a number of unintuitive phe-
nomena, some of which can be used model unusual aspects of human behaviour. The
application of quantum theory to model human behaviour is wide-ranging. In this
chapter we will look at three main areas in which it has been applied. One key area
is in how similarity judgements can be modelled. There are a number of phenomena
around similarity judgements that are not well modelled using a view of concepts
that does not take the state of the observer into account. These include the asymmetry
of similarity judgements and the fact that similarity can change depending on other
exemplars that are present. Another key area is in judgement and decision-making.
Again, puzzling phenomena have been observed regarding the fact that judgements
do not follow classical probability theory, known as the conjunction fallacy, can
depend crucially on the order in which questions are presented or on the amount of
knowledge about the world. Finally, approaches to modelling cognitive phenomena
at the neural level will also be discussed within a quantum-theoretic framework.

Keywords Quantum computing · Quantum humanities · Cognitive modelling ·

Similarity judgements · Decision-making

M. Lewis (B)
Department of Engineering Mathematics, University of Bristol, Bristol, UK
e-mail: [email protected]

© Springer Nature Switzerland AG 2022 53

E. R. Miranda (ed.), Quantum Computing in the Arts and Humanities,
https://doi.org/10.1007/978-3-030-95538-0_2
54 M. Lewis

1 Introduction

Cognitive science is the study of the mind and how it relates to human behaviour.
As a field, it is highly interdisciplinary, bringing together theories from Psychol-
ogy, Neuroscience, Biology, and others. The field of cognitive science operates at
a number of levels. On the one hand, some aspects of cognitive science look at the
behaviour of neurons, whereas other aspects look to explain human behaviour at a
more abstract level, seeking to explain human decision-making or reasoning. There
is a long-standing history of vector space models in cognitive science. Theories of
categorization such as those developed by Ashby and Gott (1988), Nosofsky (1986),
Rosch (1975) utilize notions of distance and similarity that can readily be incor-
porated in vector space models of meaning. Hampton (1987), Smith and Osherson
(1984), Tversky (1977) encode meanings as feature vectors, and models of high-level
cognitive reasoning have been implemented within vector symbolic architectures
(Gayler, 2003; Plate, 1995; Smolensky, 1990).
Another field in which vector space models play an important role is Physics,
and especially quantum theory. Though seemingly unrelated to language, intriguing
connections have recently been uncovered. The link between Physics and natural lan-
guage semantics that vector space models provide has been successfully exploited,
providing novel solutions and a fresh perspective for a number of problems related
to cognitive science, such as modelling logical aspects in vector spaces (Widdows
& Peters, 2003). Methods from quantum logic have also been applied to cognitive
processes related to the human mental lexicon, such as word association (Bruza
et al., 2009), decision-making (Pothos & Busemeyer, 2013), and human probability
judgements (Busemeyer et al., 2011). Furthermore, the categorical model of Coecke
et al. (2010), inspired by Quantum Mechanics, has provided a convincing account
of compositionality in vector space models and an extensible framework for linguis-
tically motivated research on sentential semantics. More recently, the link between
Physics and text meaning was made more concrete by a number of proposals that aim
at replacing the traditional notion of a word vector with that of a density matrix—a
concept borrowed from Quantum Mechanics which can be seen as a probability dis-
tribution over vectors (Bankova et al., 2019; Piedeleu et al., 2015; Sadrzadeh et al.,
2018).
A key area of research in cognitive science is how to formalize human behaviours
around judgements of similarity, categorization, and decision-making. However, for-
malizing concepts within a vector space, and relying on distance within vector space
to give a notion of similarity, can lead to some puzzling features, at least if we
assume that distance and similarity behave classically. For example, it has been
shown (Tversky, 1977) that judgements of similarity are not always symmetric, as
you would expect if they are just based on distance in a vector space.
Quantum Computing and Cognitive Simulation 55

In the field of Physics, Quantum Mechanics has fundamentally altered our under-
standing of the way in which particles behave. Quantum mechanics has a number
of unintuitive phenomena, some of which can be used to model unusual aspects
of human behaviour. This is not to say that a description of the brain and mind is
provided in terms of quantum phenomena at a small scale, although this sort of mod-
elling has been proposed (Hameroff, 2014; Penrose, 1990). Instead, the idea is that
the formalism arising from quantum theory provides the right sort of mathematical
tools to model these puzzling aspects of human behaviour.
The application of quantum theory to model human behaviour is wide-ranging. In
this chapter, we will look at three main areas in which it has been applied. One key
area is how similarity judgements can be modelled. There are a number of phenom-
ena around similarity judgements that are not well modelled using a view of concepts
that does not take the state of the observer into account. These include the asymmetry
of similarity judgements and the fact that similarity can change depending on other
exemplars that are present (Tversky, 1977). These phenomena have been addressed
in Pothos et al. (2013). Another key area is in judgement and decision-making. Again,
puzzling phenomena have been observed regarding the fact that judgements do not
follow classical probability theory, known as the conjunction fallacy (Tversky &
Kahneman, 1983), and can depend crucially on the order in which questions are
presented (Moore, 2002) or on the amount of knowledge about the world (Tversky
& Shafir, 1992). These phenomena are modelled within a quantum framework in
Busemeyer et al. (2011), Franco (2009), Khrennikov and Haven (2009), Pothos and
Busemeyer (2009), Wang and Busemeyer (2013). Related to this area is the phe-
nomenon of contextuality. This can be summarized as the idea that there may be
sets of random variables (in Psychology, results of an experiment) that have pairwise
joint distributions but for which no joint distribution across the whole set of random
variables can be found. This is one of the key aspects of quantum theory. Amazingly,
the same theory was also developed in Psychology, and has been formalized in Dzha-
farov and Kujala (2014, 2016). Finally, quantum theory has been applied to describe
categorization and concept combination. Again, the ways in which humans use con-
cepts has been shown not to be well modelled by classical views of combination like
fuzzy set theory, where problems known as over- and underextension are observed
(Hampton, 1987, 1988a, 1988b, 1997; Smith & Osherson, 1984). Approaches to
answer these phenomena have been proposed in Aerts (2009), Aerts and Gabora
(2005a, 2005b), Aerts et al. (2015), Sozzo (2014, 2015). The problem of modelling
how concepts compose can also be addressed via a linguistic route. The problem of
modelling concept composition by means of the grammatical structure in language
has been addressed in Coecke et al. (2010), and applications of this to the problem
of overextension was proposed in Coecke and Lewis (2015).
Approaches to modelling cognitive phenomena at the neural level have also been
considered within a quantum-theoretic framework. Fuss and Navarro (2013) show
that a quantum random walk approach to modelling choice tasks better simulates
human reaction times. Consideration of neuronal activation within a quantum frame-
work is also discussed in Pothos and Truebloodm (2015). The question of modelling
how concepts can ‘bind’ together (for example, the combination of an adjective and
56 M. Lewis

a noun) was addressed at a vector-based level in Smolensky (1990) and has been
investigated within a cognitive science context in Martin and Doumas (2020). The
compositional distributional framework of Coecke et al. (2010) has the potential to
model these aspects of neuronal activity well, and combining this theory with the
tensor product binding of Smolensky (1990) is an area for future research.
In the remainder of the chapter, we cover the following. We provide a short section
on the mathematical notation we will use. In Sect. 2, we will summarize the cognitive
phenomena that have been described by quantum theory. In Sect. 3, we describe
how quantum theory has been used to address each of these phenomena, as well as
discussing quantum-theoretic approaches to modelling neuronal-level phenomena. In
Sect. 4, we discuss the implications for using quantum computing to model cognitive
phenomena and Artificial Intelligence more generally.

1.1 Mathematical Notation

We assume that the reader has a general understanding of linear algebra. We use
Dirac’s bra-ket notation to represent vectors, their duals, and inner products.
• A ket |v is a column vector in a Hilbert space H. We will always consider the
space to be finite-dimensional, so H = Cn .
• A bra v| is the vector dual to |v. It is the conjugate transpose of |v and can
be thought of as a row vector whose elements are the complex conjugates of the
elements in |v.
• The inner product of two vectors |v, |w is represented by v|w.
• If a matrix is represented by M, its multiplication with a vector |v is given by
M|v and its multiplication with a bra v| is given by v|M. Often, these will be
combined to give a scalar v|M|v.
• The
√ absolute value of a complex number α = a + ib is represented as |α| =
a 2 + b2 . √
• The Euclidean norm of a vector is written as ||v|| = v|v.

2 Cognitive Phenomena

One of the cornerstones of human cognition is the ability to categorize. From an

evolutionary perspective, it is essential to be able to categorize objects into edi-
ble/inedible, or animals into dangerous/safe. Central to the notion of categorization
is the notion of similarity. An animal can be classified as safe based on how similar
it is to another animal known to be safe. The study of categorization and similarity
is therefore central to cognitive science, and has been widely researched (Ashby &
Gott, 1988; Nosofsky, 1986; Rosch, 1975). How can we determine whether one item
is similar to another or not? One of the approaches to studying the notion of similar-
Quantum Computing and Cognitive Simulation 57

ity and categorization is to view objects as represented by points in a feature space,

and then determining similarity within that space. Categories can then be viewed as
forming regions within the feature space. However, under classical assumptions this
kind of model does not necessarily model human behaviour well. We describe here
some of these key phenomena.

2.1 Asymmetry of Similarity Judgements

In Tversky (1977), the question is addressed on whether objects can be represented

as points in a geometric space together with the distance between the points being
measured via a metric on the space. The similarity between objects is then viewed
as a function of the distance between the objects in the space.
Within this model, a key assumption is that distance is viewed as a metric, meaning
that the following hold: given objects x, y, and z, and distance metric d,

d(x, x) = 0 (1)
d(x, y) = d(y, x) (symmetry) (2)
d(x, z) ≤ d(x, y) + d(y, z) (triangle inequality) (3)

Tversky (1977) show that these assumptions do not necessarily hold. In a series of
experiments, Tversky showed that when one object is considered more prominent
than another, the less prominent object is considered more similar to the more promi-
nent object and vice versa. Firstly, pairs of countries are assessed (by Israeli college
students) for which country is most prominent. The pairs of countries are, for exam-
ple, China and Vietnam,1 the USA and Mexico, or Belgium and Luxembourg. A
separate group of students were then asked to judge which phrase they preferred to
use when describing the similarity of the two countries out of “country a is similar to
country b” and “country b is similar to country a”. Across all pairs, the majority of
students chose the ordering in which the more prominent country was given second,
for example “Vietnam is similar to China” rather than “China is similar to Vietnam”.
This kind of effect was seen across a range of topics and modalities, for example,
when judging the similarity of shapes, letters, and sequences of sounds.

2.2 Diagnosticity

Tversky (1977) further show that judgements of similarity can be altered by the
presence of comparison objects. According to a geometric model of similarity, the
similarity of two objects should not be affected by the presence of other objects.

1 In fact, the countries assessed are Red China and North Vietnam.
58 M. Lewis

However, this is shown not to hold for human similarity judgements in the following
manner. Pairs of quadruples of countries were designed that differ in only one entry,
i.e. there are pairs of sets {a, b, c, p} and {a, b, c, q}. Participants are asked to say
which of b, c and q or p country a is most similar to. For example, one such pair is the
sets of countries {Austria, Sweden, Poland, Hungary} and {Austria, Sweden, Norway,
Hungary}. Participants see only one quadruple and are asked to judge which country
Austria is most similar to. Tversky (1977) find that in the set {Austria, Sweden,
Poland, Hungary}, most participants judge Austria to be most similar to Sweden, but
in the quadruple {Austria, Sweden, Norway, Hungary} participants judge Austria to
be most similar to Hungary.

2.3 Conjunction and Disjunction Fallacies

In modelling human decision-making, one assumption is that human behaviour can

be expressed by means of probabilities. A key facet of classical probability is that
the probability of a conjunction of events will always be less than or equal to the
probability of one of its constituents: P(A&B) ≤ P(A). Similarly, the probability
of a disjunction of events will always be greater than or equal to the probability
of one of its constituents: P(A) ≤ P(A or B). However, experiments in Tversky
and Kahneman (1983) showed that these rules are not adhered to when humans
make decisions about category membership. The experiments run as follows. Human
participants are given a story about a person, and asked to make certain judgements
about them. The most frequently cited is the following:
Linda is 31 years old, single, outspoken and very bright. She majored in philosophy. As a
student, she was deeply concerned with issues of discrimination and social justice, and also
participated in anti-nuclear demonstrations.

The description is followed by eight statements about the person’s career or per-
sonality:
• Linda is a teacher in elementary school.
• Linda works in a bookstore and takes Yoga classes.
• Linda is active in the feminist movement (F).
• Linda is a psychiatric social worker.
• Linda is a member of the League of Women Voters.
• Linda is a bank teller (T ).
• Linda is an insurance salesperson.
• Linda is a bank teller and is active in the feminist movement (T &F).
Participants are asked to rank the eight statements associated with each description
by how probable each statement is. A large majority (between 85 and 90%, depending
on the participant pool) rank the statements in the order T &F > T , that is, they
judge that the probability that Linda is both a feminist and a bank teller is higher than
the probability that Linda is a bank teller. This phenomenon still holds (to a lesser
Quantum Computing and Cognitive Simulation 59

extent) in other contexts where the participants are asked to place a bet on the two
statements T and T &F, in a domain of medical experts (where the story that is told
is about symptoms and the task is to diagnose the patient), and in a range of other
situations designed to reduce the incidence of the conjunction fallacy. Experiments
where fallacy does not hold to such a great extent are when the participant pool is
with a range of statistically sophisticated participants, and when the experiment is
phrased in terms of numbers of people.

2.4 Question Order Models

In Moore (2002), the effect of order on answers to questions is addressed. If you are
asked “Do you try to eat healthily?” and “Do you like McDonalds?”, your answer
to each question may change depending on which question is asked first. There are
a number of ways in which the ordering could affect the answers. Moore (2002)
identifies four types of such effect, termed contrast, consistency, additive, and sub-
tractive. The consistency effect runs as follows. A group of participants is asked
a pair of questions. For a concrete example, consider the questions “Do you think
Bill Clinton is trustworthy” and “Do you think Al Gore is trustworthy”. For context,
these questions were asked in 1997. If the question is asked first, it is considered
to be asked in a non-comparative context, meaning that there is nothing to imme-
diately compare the question to. If the question is asked second, it is considered to
be asked in a comparative context, because it can be considered in comparison to
the first. The answers to the questions were distributed as in Table 1. The effect of
the question ordering is that the answer to the question in the comparative context is
altered to make it more consistent with the answer in the non-comparative context.
So, if a participant is first asked whether Gore is honest and trustworthy, and they
answer positively, then they are more likely to answer positively to the question of
whether Clinton is honest and trustworthy. On the other hand, if a participant is first
asked whether Clinton is honest and trustworthy, and answers negatively, then they
are more likely to answer negatively to the question of whether Gore is honest and
trustworthy. The effect is to make the answers to the two questions more consistent.

Table 1 Example of the consistency effect. ∗ indicates significant difference. Figures from Moore
(2002)
Per cent saying yes
Context Clinton Gore Gap
First 50 68 +18∗
Second 57 60 +3
Difference +7∗ −8∗ −15∗
60 M. Lewis

Table 2 Example of the contrast effect. ∗ indicates significant difference. Figures from Moore
(2002)
Per cent saying yes
Context Gingrich Dole Gap
First 41 60 +19∗
Second 33 64 +31∗
Difference −8∗ +4∗ +12∗

Table 3 Example of the additive effect. ∗ indicates significant difference. Figures from Moore
(2002)
Per cent saying ‘All’ or ‘Many’
Context White Black Gap
First 41 46 +5
Second 53 56 +3
Difference +12∗ +10∗ −2

The contrast effect occurs when people alter their answers to make the answers
contrast with each other. For example, in a Gallup poll in 1995 when asked the
question of whether the label ‘honest and trustworthy’ applies to either Bob Dole or
Newt Gingrich, people’s answers in the comparative context altered to increase the
gap between them, emphasizing the differences between the two. Figures are shown
in Table 2.
Under the additive effect, considering either question first has the same effect. In
a series of racial hostility polls in 1996, people were asked the questions “Do you
think that only a few white people dislike blacks, many white people dislike blacks,
or almost all white people dislike blacks?” and “Do you think that only a few black
people dislike whites, many black people dislike whites, or almost all black people
dislike whites?”. In both cases, the percentage responding ‘All’ or ‘Many’ increased
in the comparative context; see Table 3.
Lastly, in the subtractive context considering either question first again has the
same effect, but in this case the effect is to decrease the number of positive answers.
The example given in Moore (2002) is the question of whether baseball players Pete
Rose and, separately, Shoeless Joe Jackson should be eligible for admission to the
Baseball Hall of Fame. In each case, the players are ineligible, and the participants
are told why. In each case, the number of people answering favourably to the question
of whether they should be allowed decreased in the comparative context (Table 4).
These four examples show clearly the different kinds of effects that question order
can have on responses, and that the context in which a question is asked is crucial
for its response. These effects are addressed in Wang and Busemeyer (2013), Wang
et al. (2013).
Quantum Computing and Cognitive Simulation 61

Table 4 Example of the subtractive effect. ∗ indicates significant difference. Figures from Moore
(2002)
Per cent saying ‘Favourable’
Context Rose Jackson Gap
First 64 45 +19∗
Second 52 33 +19∗
Difference −12∗ −12∗ 0

2.5 The ‘Sure Thing’ Principle

A counter-intuitive aspect of human decision-making is the violation of the ‘sure

thing’ principle. The sure thing principle was introduced by Savage (1954). This
says that if we prefer x to y given any possible state of the world, then we should
prefer x to y even when the exact state of the world is unknown. However, people do
not always make decisions according to the sure thing principle. Tversky and Shafir
(1992) present the following experiment:
Imagine that you have just taken a tough qualifying examination. It is the end of the fall
quarter, you feel tired and run-down, and you are not sure that you passed the exam. In
case you failed you have to take the exam again in a couple of months-after the Christmas
holidays. You now have an opportunity to buy a very attractive 5-day Christmas vacation
package to Hawaii at an exceptionally low price. The special offer expires tomorrow, while
the exam grade will not be available until the following day. Would you:

1. buy the vacation package

2. not buy the vacation package
3. pay a $5 non-refundable fee in order to retain the rights to buy the vacation package at
the same exceptional prices the day after tomorrow—after you find out whether or not
you have passed the exam.

31% of respondents said that they would buy the vacation package, 7% said that they
would not buy the vacation package, and 60% said that they would reserve the low
price for the small fee-paying to wait until the uncertainty was resolved.
However, when the participants were asked to imagine they knew the outcome of
the exam, a majority said that they would buy the package:
Imagine that you have just taken a tough qualifying examination. It is the end of the semester,
you feel tired and run-down, and you find out that you [passed the exam/failed the exam.
You will have to take it again in a couple of months-after the Christmas holidays]. You now
have an opportunity to buy a very attractive 5-day Christmas vacation package to Hawaii at
an exceptionally low price. The special offer expires tomorrow. Would you

1. buy the vacation package

2. not buy the vacation package
3. pay a $5 non-refundable fee in order to retain the rights to buy the vacation package at
the same exceptional prices the day after tomorrow.
62 M. Lewis

Table 5 Matrix of payoffs for the prisoner’s dilemma situation

Other D Other C
You D You 10, other 10 You 25, other 5
You C You 5, other 25 You 20, other 20

In the context where the participant had passed the exam, 54% of respondents said
that they would buy the vacation package, 16% of respondents said that they would
not buy the vacation package, and 30% of students said that they would pay the $5
waiting fee. Very similarly, in the case where the participant had failed the exam, 57%
of respondents said that they would buy the vacation package, 12% of respondents
said that they would not buy the vacation package, and 31% of students said that they
would pay the $5 waiting fee. Essentially, in the context where the outcome of the
exam is known, a majority of participants decide to book the holiday—whether or
not it is pass or fail. According to the sure thing principle, those participants should
decide to book the holiday in that case where the outcome of the exam is not known.
But, as can be seen, this does not happen.
A similar experiment involves gambling:
Imagine that you have just played a game of chance that gave you a 50% chance to win $200
and a 50% chance to lose $100. The coin was tossed and you have [won $200/lost $100].
You are now offered a second identical gamble
50% chance to win $200 and 50% chance to lose $100. Would you:

1. Accept the second gamble

2. Reject the second gamble

In the context that the participants had won the first gamble, 69% accepted and 31%
rejected the second. In the context that participants had lost the first gamble, the
split was 59% accept, 41% reject. Either way, a majority of participants accepted the
second gamble.
However, in a context where they did not know the outcome of the first gamble,
a majority of the same set of participants rejected the second gamble. The split was
36% accept, 64% reject.
The explanation given in Tversky and Shafir (1992) is that the presence of uncer-
tainty makes it more difficult for people to focus on the implications of each outcome.
A further example is given in a prisoner’s dilemma situation. In the prisoner’s
dilemma, a hypothetical prisoner has the choice either to cooperate with the other
prisoner, i.e. keep quiet about their activities, or defect, i.e. tell the authorities what
they know. The possible outcomes of cooperating or defecting are presented in the
following payoff matrix (Table 5).
In an experimental situation, if the participant is told that the other has cooperated,
the majority of participants choose to defect. Also, if the participant is told that the
other has defected, the majority of participants choose to defect. However, if the
Quantum Computing and Cognitive Simulation 63

participant is not told what the choice of the other is, the majority of participants
choose to cooperate.

2.6 Categorization

One of the most important things that humans can do is to categorize their experi-
ences. Is this fruit good to eat, or poisonous? Is that large striped animal dangerous?
Categorization can be thought of in terms of similarity to other experiences, and to
help us do this we use concepts. There have been a number of theories of concepts.
The classical view of concepts dates back to Plato and Aristotle. Under the classical
view of concepts, a concept is viewed as a set of necessary and sufficient conditions,
formulated as a list of properties. Such a view of concepts is inadequate. Firstly,
for many natural concepts it is impossible to give a set of necessary and sufficient
conditions (Wittgenstein, 1953). Instead, instances of a category can be thought of
as sharing a set of ‘family resemblances’. For example: what makes the images in
Fig. 1 pictures of chairs, and the images in Fig. 2 not pictures of chairs? It is difficult
to decide on a set of necessary and sufficient conditions that rules the kneeling chair
to be a chair, but rules out a bar stool.2,3,4,5
Furthermore, the classical view does not take into account the graded and context-
dependent nature of concepts. Consider the much simpler concept tall. Whether or
not to describe something as tall will firstly be dependent on the set of things that
form its context, for example, 4-year-old children, or Dutch women, or giraffes, or
mountains. Secondly, even if we have fixed a context, it is not a crisp question of

Fig. 1 Images of chairs

2 https://commons.wikimedia.org/wiki/File:Philipp_Rumpf_Studienblatt_eines_Stuhls.jpg.
3 https://commons.wikimedia.org/wiki/File:Sleekform_Kneeling_Chair.jpg.
4 https://commons.wikimedia.org/wiki/File:Shaker_no._7_rocking_chair_Rocking_Chair,_1878

%E2%80%931910_(CH_18460985).jpg.
5 https://commons.wikimedia.org/wiki/File:LibertyChair.jpg.
64 M. Lewis

Fig. 2 Non-chairs

Fig. 3 Heights of male celebrities

whether we can divide some people as tall. In Fig. 3, how are we to decide what is
the necessary and sufficient condition for being tall? Certainly, anyone to the right of
Jeremy Clarkson is tall, and anyone to the left of Nikita Khrushchev is not tall, but
should we state the condition to be 180 cm? 184 cm? If we say 184 cm and someone
is 183.5 cm, are they really not tall?6,7,8,9
These sorts of considerations led to the formulation of prototype theory (Rosch,
1975). In prototype theory, concepts are assumed to have a prototype, and members
of the concept are judged as being more or less similar to the prototype. For example,
in the concept bird, a robin is judged as a very prototypical bird, whereas an ostrich
or a penguin would be judged as much less typical. This allows the consideration

6 https://commons.wikimedia.org/wiki/File:Bar_Stools_(49907001456).jpg.
7 https://commons.wikimedia.org/wiki/File:52.8_Inch_Sofa_Couch_Loveseat,_Grey.jpg.
8 https://commons.wikimedia.org/wiki/File:TabouretAFDB.jpg.
9 https://commons.wikimedia.org/wiki/File:Wooden_Rocking_Horse_with_Weels.jpg.
Quantum Computing and Cognitive Simulation 65

of concepts in terms of family resemblances, and accounts for the graded nature of
concepts.
Further, the exemplar theory of concepts views a category as based on specific
instances, or exemplars, that are stored in memory. Categorization judgements are
based on judgements of similarity to the exemplars, which are easily called to mind.
Exemplar theory and prototype theory are both backed by experimental evidence
(Nosofsky, 1986; Rosch, 1975). Key characteristics of these theories are that they can
account for the graded nature of concepts, and in both cases, we represent concept
instances within some kind of feature space, equipped with a distance metric.
Another key feature of human concept use is that we are able to combine concepts
to create new ones. This can be done in very simple ways, for example, combining
the concept blue and the concept book, we obtain a blue book. How can we formalize
this? One approach to formalizing concept combination used fuzzy set theory. Under
this view, a concept is a fuzzy set, that is, a set whose members have membership
values in the interval [0, 1] rather than in the set {0, 1}. We write

μ A : X → [0, 1] (4)

and use the notation μ A (x) to describe the membership of the item x in the concept
A. In fuzzy set theory, we can form the conjunction or disjunction of two sets as
follows:

μA and B (x) = min(μ A (x), μ B (x) (5)

μA or B (x) = max(μ A (x), μ B (x)) (6)

Under these rules, combinations quickly become problematic. The classic exam-
ple is that of a pet fish (Smith & Osherson, 1984). Consider a fish like a goldfish. A
goldfish is not a particularly typical fish, nor a very typical pet, so we might assign
low memberships to goldfish in each of these concepts:

μpet (goldfish) = 0.3 (7)

μfish (goldfish) = 0.25 (8)

However, it is a very typical pet fish, i.e. we would assign something like

μpet fish (goldfish) = 0.9 (9)

The question then is what is going on in the combination of these concepts to

allow the typicality of the instance in the combined concept to be higher than it is in
either of the individual concepts.
This kind of phenomenon was examined in Hampton (1988a, 1988b). In the ‘pet
fish’ kind of example, where the typicality of an item in a conjunction of concepts is
higher than the typicality of an item in one or more of its constituent concepts, the
66 M. Lewis

phenomenon is called ‘overextension’. On the other hand, in the case of disjunction

of concepts, a phenomenon termed underextension is also seen. In this case, the
typicality of an item in the disjunction of two concepts is lower than typicality in one
or more of the constituent concepts.

2.7 Negation

Another problematic phenomenon in human concept use is the use of negation. In

Alxatib and Pelletier (2011), an experiment is performed where the membership of
humans in the category ‘tall’ is measured. Imagine we have five suspects in a robbery,
all men, with heights 5 4 , 5 7 , 5 11 , 6 2 , and 6 6 . Participants are asked to
judge, of each suspect, whether they are (a) tall, (b) not tall, (c) tall and not tall, or
(d) neither tall nor not tall. The percentages of people agreeing to these statements
are given in the table below. Note that in the experiment, the suspects were shown in
a randomized order; we present them here in height order for ease of presentation.

Suspect Tall Not tall Tall and not tall Neither tall nor not tall
True False True False True False True False
5 4 1.3 98.7 94.7 3.9 14.5 76.3 27.6 65.8
5 7 5.3 93.4 78.9 17.1 21.1 65.8 31.6 57.9
5 11 46.1 44.7 25 67.1 44.7 40.8 53.9 42.1
6 2 80.3 10.5 9.2 82.9 28.9 56.6 36.9 55.3
6 6 98.7 1.3 0 100 5.3 81.6 6.6 89.5

There is a strong correlation between the values for the borderline statements ‘tall
and not tall’ and ‘neither tall nor not tall’. Participants were willing to agree to both
of these statements. In the first statement ‘tall and not tall’, a participant would seem
to ascribe the value true both to ‘tall’ and to ‘not tall’, whereas in the second, the
participant would also judge that he is neither.
The problem of negation is also considered by Hampton, in a setting combined
with conjunction (Hampton, 1997). In this setting, Hampton considers the typicality
of items to conjunctions of concepts. For example, we can consider the typicality
of parrot to the conjunction Pets which are also birds and Pets which are not birds.
Of course, we would consider the typicality of parrot in the first to be high and in
the second to be low. However, if concepts are being combined in line with log-
ical reasoning, then the sum of the typicality in Pets which are also birds and in
Pets which are not birds should not exceed the typicality of parrot in Pets. This is
another form of overextension, as we have seen in previous descriptions.
Quantum Computing and Cognitive Simulation 67

3 Quantum Models

The problems for modelling cognitive phenomena outlined in Sect. 2 above have all
been tackled using methods from quantum theory. Below, in Sect. 3.1 we give an
overview of the key ideas before explaining the models in detail. After this overview,
we discuss approaches to modelling similarity, diagnosticity, and the conjunction
fallacy. These approaches have been developed over a research programme laid out
in Busemeyer et al. (2011), Franco (2009), Pothos et al. (2013) and other papers,
and crucially depend on the notion of interference. Following this, we consider the
sure thing principle and question order models. Again, these models depend on the
notion of interference. We go on to discuss models of concept combination and
negation. These approaches use the quantum phenomenon of entanglement as well
as interference to describe how humans combine concepts. Linked to these is the
notion of contextuality. Finally, we talk about quantum models that are applied at a
neuronal level.

3.1 Overview of Quantum Models in Psychology

The following gives an overview of some of the key quantum concepts used in
approaches such as those in Busemeyer et al. (2011), Pothos et al. (2013). Within
applications of quantum theory to Psychology, questions are considered from the
point of view of an agent. We consider an agent as having a certain state |ψ, and as
questions are asked of the agent, or the agent is exposed to a stimulus, the state of
the agent changes. Within quantum theory, a question about a system is termed an
observable: a self-adjoint linear operator. The eigenvectors {|ai }i of an observable
A form an orthonormal basis of the Hilbert space, and the eigenvalues {ai } associ-
ated with the {|ai }i form the values that the observable can take, i.e. the answers
to the questions. To give a simple example: a yes/no question such as “is the sky
blue?” might be encoded in a two-dimensional vector space. An observable B on the
space will have two orthonormal eigenvectors |blue and |not blue. We can set the
eigenvalues associated with |blue and |not blue to be 1 and 0, respectively.
Now, we represent an agent’s belief about the state of the sky as a vector |s.
Given that the agent is ambivalent about what the weather is (possibly clear and
blue, possibly overcast and grey), the probability that we receive the answer ‘yes!’
to the question “Is the sky blue” is given by the following calculation:

P(yes) = |blue|s|2 (10)

This quantity is the square of the projection onto the blue axis, illustrated in Fig.
4.
This notion can be extended. Firstly, there may be more than one basis vector
corresponding to a given answer, so that that answer corresponds to a subspace of
68 M. Lewis

Fig. 4 A two-dimensional
space with orthonormal basis blue
|blue, |not blue. The state
|s0  represents belief in a
| s0⟩
mostly blue sky with a ⟨blue | s0⟩
couple of clouds, and the
state |s1  represents belief in
a mostly grey sky. The
projection onto the blue axis
is high for |s0  and low for
|s1 , whereas the projection
onto the not blue axis is low | s1⟩
for |s0  and higher for |s1 .
⟨blue | s1⟩

not blue
⟨not blue | s1⟩
⟨not blue | s0⟩

the vector space of dimension greater than 1. For example, we could have a three-
dimensional vector space with one dimension |blue corresponding to blue sky, one
dimension |cloudy corresponding to cloudy sky, and one dimension |night corre-
sponding to nighttime, and suppose we define an observable D to ask the question
“is it daytime?” D has eigenvectors |blue and |cloudy with eigenvalue 1 (corre-
sponding to yes) and eigenvector |night with eigenvalue 0 (corresponding to no).
Then,

D = 1 · |blueblue| + 1 · |cloudycloudy| + 0 · |nightnight| = |blueblue| + |cloudycloudy|

(11)
The probability of an answer ‘yes’ to the question “is it daytime?” is again the
square of the probability amplitude, which is the magnitude of the projection onto
the ‘daytime’ subspace spanned by the eigenvectors |blue and |cloudy. If the
current state of affairs is given by |s, then this is calculated by

P(yesday ) = s|D|s (12)

illustrated in Fig. 5.
More generally again, the current state of affairs may be uncertain. Quantum
theory models this as a mixture of states

ρ= si |si si | (13)
i


where i si = 1.
Then the probability of the answer ‘yes!’ to the question “is it daytime?” is given
by
T r (Dρ) (14)
Quantum Computing and Cognitive Simulation 69

Fig. 5 This figure shows a night

three-dimensional space with
orthonormal basis
|blue, |cloudy, and
|night. The state |s0 
represents belief that the sky
is slightly cloudy and just
starting to get dark. The
probability amplitude of the
answer ‘yes!’ to the question cloudy
“is it daytime?” is given by
the projection of |s0  onto
the plane spanned by
|blue, |cloudy | s0⟩

blue

where T r stands for matrix trace. If we are not at all uncertain about the state, then
ρ = |ss| and this is equivalent to the formula (12).
A key difference between classical probability theory and quantum probability
theory is the treatment of conjunctions of events. In classical probability theory, the
probability P(AB) conjunction of two events A and B is equal to the probability
P(B A) of the two events in the opposite order. In quantum probability, this does not
hold, due to the fact that in general the matrices corresponding to the observables Â
and B̂ will not commute.
This means that conjunction in quantum probability needs to be defined in sequen-
tial terms. Specifically, suppose we want to calculate the probability of an agent in
state |s judging A and then B. We have

P(A and then B) = P(A) · P(B|A) (15)

The first term P(A) is expressed as

P(A) = ||A|s||2 (16)

and the second term P(B|A) is given by

P(B|A) = ||B|s A ||2 (17)

where |s A  is the state of knowledge on the agent after judging that A holds, i.e.

A|s
|s A  = (18)
||A|s||

Putting this all together, we see that

70 M. Lewis

P(A and then B) = ||A|s||2 ||B|s A ||2 (19)

 
B A|s 2
= ||A|s||2 = ||B A|s||2 (20)
||A|s||

i.e. the agent projects first to subspace A, and then to subspace B. In general, this is
not equal to ||AB|s||2 , as it would be in classical probability.

3.2 Similarity Judgements

In Pothos et al. (2013), the model described above is used to account for the asym-
metry of similarity judgements. Recall that Tversky (1977) showed that when asked
to consider the similarity of one entity, such as a country, to another, the judgements
of similarity are higher if the entity that is more salient is the one being compared
to. So, for example, more participants agreed with the statement “Korea is similar
to China” than “China is similar to Korea” (the participants were all Israeli college
students).
Pothos et al. (2013) explicate this as follows. The entities being judged, for exam-
ple, China and Korea are modelled as subspaces of a given vector space. Each of
these subspaces has an associated projector PChina , PKorea . The state of belief of the
agent is modelled as a quantum state |ψ. The probability that the agent’s belief state
is consistent with the concept China or the concept Korea is given by the square of
the projection onto the relevant subspace: ||PChina |ψ||2 = ψ|PChina ψ. The initial
belief state | psi is set up so that ψ|PChina ψ = ψ|PKorea ψ.
Then, the similarity of Korea to China is calculated but first projecting | psi onto
the Korea subspace, and then projecting the resulting vector onto the China subspace:

Sim(Korea, China) = ||PChina PKorea |ψ||2 (21)

This quantity calculates the probability that the original state |ψ is consistent with
Korea, multiplied by the probability that this new state is consistent with China.
The agent projects their belief state onto Korea, after which their new belief state is
|ψ
normalized, so that |ψKorea  = ||PPKorea
Korea |ψ||
. This new state is projected onto China, and
is consistent with the concept China with probability ||PChina |ψKorea . So, we have
that
||PChina |ψKorea ||2 · ||PKorea |ψ||2 ||PChina PKorea |ψ||2 (22)

Now, as long as the projectors PChina and PKorea do not commute, we have that
Sim(Korea, China) = Sim(China, Korea). Moreover, Pothos et al. (2013) show that
the similarity of Korea to China can be made to be higher than the similarity of China
to Korea by modelling the more salient entity, in this case China, as a larger (higher
dimensionality) subspace.
Quantum Computing and Cognitive Simulation 71

3.3 Diagnosticity

Pothos et al. (2013) also show that the phenomenon of diagnosticity (Sect. 2.2) can be
modelled. In this model, each country is modelled as a single-dimensional subspace.
The diagnosticity effect is produced by the order in which the projectors for the
countries are applied to the agent’s state vector |ψ.
Recall from Sect. 2.2 that the diagnosticity effect says that similarity judgements
can be affected by the context they are made in. An example from Tversky (1977)
is the following. Take the sets of countries {Austria, Sweden, Poland, Hungary} and
{Austria, Sweden, Norway, Hungary}. Participants see only one quadruple and are
asked to judge which country Austria is most similar to. In the quadruple {Austria,
Sweden, Poland, Hungary}, Austria is judged most similar to Sweden by a majority
of participants. However, in {Austria, Sweden, Norway, Hungary}, Austria is judged
most similar to Hungary. Given that both are available in each set of countries, why
does the similarity change?
Pothos et al. (2013) set each country to be represented as a ray. The similarity
of Austria to any of the other countries in the set is modelled by saying that the
agent’s belief state |ψ first projects onto the subspaces corresponding to the other
countries. There are two orders in which this can happen, since there are two other
countries. Therefore, these orders are averaged. Then, depending on the placing of
the rays in the Hilbert space, the required effect can be produced. Pothos et al. (2013)
show that with some reasonable assumptions about the similarities between the pairs
of countries (i.e. we assume that Sweden is similar to Norway, and that Hungary is
similar to Poland), the effect can be reliably reproduced over a number of simulations.

3.4 Conjunction Fallacy and Over/Underextension

Franco (2009) provide an account of the conjunction fallacy using the notion of
interference effect. The model works as follows. Recall that in the conjunction fallacy
(Sect. 2.3), the problem is that the conjunction of two events (that Linda is a bank
teller and a feminist) is judged more probable than the single event that Linda is a
feminist. According to classical probability, this cannot hold.
Busemeyer et al. (2011), Franco (2009) model this using the quantum formalism
described in Sect. 3.1. They consider the two questions A: “is Linda a feminist?”
and B: “is Linda a bank teller?”. The question A is represented by an observable
on a two-dimensional Hilbert space with eigenvectors |a0  and |a1  which have the
eigenvalues a0 = 0 (Linda is not a feminist) and a1 = 1 (Linda is a feminist), respec-
tively. Question B is similarly represented. Furthermore, Franco (2009) represent the
probability P(b j |ai ) as the probability that the answer to question B is b j , given that
the answer to A has already been determined to be ai . If this latter condition is the
case, then within this two-dimensional setting this means that the current state we are
in is |ai  itself, and therefore P(b j |ai ) = |b j |ai |2 , following Eq. (10). Notice that
72 M. Lewis

Fig. 6 Orthonormal bases in | b1⟩ bank teller

a two-dimensional space
representing the questions
“Is Linda a bank teller?” and
“Is Linda a feminist?”
| a1⟩ feminist

| b0⟩ not bank teller

| a0⟩ not feminist

this means that P(b j |ai ) = |b j |ai |2 = |ai |b j |2 = P(ai |b j ). Now as in Eq. (20),
the probability of the answer that Linda is a feminist and a bank teller, P(a1 -and-b1 )
is modelled as P(a1 )P(b1 |a1 ).
Franco (2009) then lays out a specific relation between the eigenvectors of  and
the eigenvectors of B̂, as follows.
The two observables  and B̂ both operate on the same two-dimensional Hilbert
space, and the eigenvectors of  can be transformed into the eigenvectors of B̂ via
a unitary transformation
 √ √ 
e−iξ√ P(b1 |a1 ) e−iφ √1−P(b1 |a1 )
U=
−eiφ 1−P(b1 |a1 ) eiξ P(b1 |a1 )

Visually, we have a pair of orthonormal bases as given in Fig. 6, where |bi  = U |ai .
Now, given this representation, the state of knowledge of the agent is represented
as a vector:  
|s = P(a0 )|a0  + eiφa P(a1 )|a1 

which gives the correct probabilities for the agent to judge whether or not Linda is a
feminist.
Expressing this vector in the basis of B̂, we obtain
Quantum Computing and Cognitive Simulation 73

| b1⟩ bank teller

| a1⟩ feminist
P( feminist and then bank teller)

P(bank teller) | s⟩

| b0⟩ not bank teller

| a0⟩ not feminist

Fig. 7 Starting in the state |s, the probability that Linda is judged to be a bank teller is less than
that the probability that Linda is judged first to be a feminist and then to be a bank teller

 
|s = P(a0 )U T |b0  + P(a1 )U T |b1  (23)
    
= P(a0 ) P(b1 |a1 )eiξ − P(a1 ) P(b1 |a0 )eiφa −φ |b0  (24)
    
+ P(a0 ) P(b1 |a0 )eiφ + P(a1 ) P(b1 |a1 )eiφa −ξ |b1  (25)

Now, calculating P(b1 ), i.e. the probability that the agent judges that Linda is a bank
teller, we get
   
P(b1 ) = | P(a0 ) P(b1 |a0 )eiφ + P(a1 ) P(b1 |a1 )eiφa −ξ |2 (26)

= P(a0 )P(b1 |a0 ) + P(a1 )P(b1 |a1 ) + 2 P(a0 )P(a1 )P(b1 |a0 )P(b1 |a1 )cos(φ + ξ + φa )
(27)

The last term in this sum is called the interference term I (s, A). If P(a0 )P(b1 |a0 ) +
I (s, A) is negative, then we obtain that P(b1 ) < P(a1 and then b1 ), i.e. that the prob-
ability that Linda is a bank teller is judged less likely than the probability that Linda
is a feminist and a bank teller. This can be understood visually in Fig. 7.
This calculation is expressed succinctly in Busemeyer et al. (2011) as follows.
Consider that the event of answering ‘yes’ to “is Linda a feminist?” is represented by
the projector PF , and that the event of answering ‘yes’ to “is Linda a bank teller?” is
given by PB . The projectors corresponding to the answer ‘no’ to each question are
given by P¬F = 1 − PF and P¬B = 1 − PB , respectively.
Now, as in Eq. (20), given a state |s we have

P(F and then B) = ||PB PF |s||2 (28)

74 M. Lewis

We can compare this value with the value P(B) = ||PB |s||2 as follows:

||PB |s||2 = ||PB · I |s||2 = ||PB (PF + P¬F |s||2 (29)

= ||PB PF |s + PB P¬F |s|| 2
(30)
= ||PB PF |s|| + ||PB P¬F |s|| + s B¬F |s B F  + s B F |s B¬F 
2 2
(31)

where |s B F  = PB PF |s and |s B¬F  = PB P¬F |s.

This term s B¬F |s B F  + s B F |s B¬F  is the interference term that we called I (s, A)
above, and we can construct this term to be negative. Again, if s B¬F |s B F  +
s B F |s B¬F  + ||PB P¬F |s||2 is negative overall, then we have

||PB |s||2 < ||PB PF |s||2 (32)

and the conjunction fallacy is fulfilled.

A related phenomenon to the conjunction fallacy is that a similar phenomenon
is observed when considering disjunctions. Morier and Bordiga (1984) show that
in a setup where participants are asked to judge the probability that Linda is (a) a
feminist, (b) a bank teller, (c) a feminist and a bank teller, and (d) a feminist or a
bank teller, the order from least probable to most probable is given by

P(bank teller) < P(feminist and bank teller) < P(feminist or bank teller) < P(feminist)

Busemeyer et al. (2011) show that their probability model can also account for
this phenomenon. They consider the probability of answering ‘no’ to the question
“is Linda neither a bank teller nor a feminist?”. This is modelled as the probability
1 − ||P¬F P¬B ||2 . Notice that here, the order of the projectors is such that the state is
first projected onto P¬B —this is because this subspace is more probable. Consider

||P¬F |s||2 = ||P¬F P¬B ||2 + ||P¬F PB ||2 + s¬F B |s¬F¬B  + s¬F¬B |s¬F B  (33)

Again, if the interference term s¬F B |s¬F¬B  + s¬F¬B |s¬F B  is sufficiently negative,
we have ||P¬F PB ||2 + s¬F B |s¬F¬B  + s¬F¬B |s¬F B  < 0 and then ||P¬F |s||2 <
||P¬F P¬B ||2 , as needed for the disjunction fallacy.

3.4.1 Double Overextension

Whilst this model covers a number of kinds of probability judgement error, one phe-
nomenon that is not covered is the phenomenon of double overextension. This occurs
when the probability of an item belonging to a conjunction of concepts is judged
to be higher than the probability of the item belonging to either of the constituent
concepts. In the Linda story, this would mean that the probability of Linda being
judged to be a feminist and a bank teller would not only be judged higher than the
Quantum Computing and Cognitive Simulation 75

probability of her being a bank teller, but also higher than her being judged to be a
feminist.
A quick look at the formalism shows that this cannot hold, at least as long as we
assume the order of conjunction is fixed.

P(F and then B) = P(F) · P(B|F) = ||PB PF |s||2 (34)

= ||PB |s F || · ||PF |s|| < ||PF |s|| = P(F)
2 2 2
(35)

As long as we hold the order fixed, we cannot have that both P(F) and P(B) are
lower than P(F and then B).
However, this kind of double overextension has been empirically observed, as
described in Sect. 2.6. In Sect. 3.8, we will describe an approach in which the problem
of double overextension can be modelled.

3.5 Violation of the ‘Sure Thing’ Principle

The ‘sure thing’ principle states that if an agent would take a specific course of action
C in all possible states of the world, then they should take action C when they do
not know what the state of the world is. However, as described in Sect. 2.5, humans
usually do not behave like this. This is seen in a prisoner’s dilemma situation, the
situation where we are choosing whether to book a holiday, or a situation where we
are choosing whether to make a bet, as well as many more. Pothos and Busemeyer
(2009) show that this phenomenon can be modelled using the same kind of quantum
similarity model described in Sect. 3.2. Other approaches include that in Khrennikov
and Haven (2009).
The model is applied to a prisoner’s dilemma situation. The situation can be
modelled as the tensor product of one space modelling our opponents’ actions, and
one space modelling our actions. Each space is two-dimensional, with one dimension
corresponding to the ‘cooperate’ (C) action, and the other corresponding to the
‘defect’ (D) action. The agent’s state at the start of the game is therefore represented
as the state
1
|ψ0  = |D D + |DC + |C D + |CC (36)
2
If the opponent is known to defect, then the state changes to

|ψ D  = α D D |D D + α DC |DC (37)

where α2D D + α2DC = 1. On the other hand, if the opponent is known to cooperate,
the state changes to
|ψC  = αC D |C D + αCC |CC (38)
76 M. Lewis

where αC2 D + αCC

2
= 1. Our decision whether or not to defect is modelled as a
projection M onto our defect axis, i.e.

M = |D DD D| + |C DC D| (39)

Now, if we know that our opponent defected, then our probability of defection is

ψ D |M|ψ D  = (α∗D D D D| + α∗DC DC|)(|D DD D| + |C DC D|)(α D D |D D + α DC |DC)

(40)
= |α D D | 2
(41)

where |α| indicates the absolute value of α. However, if we know that our opponent
cooperated, then our probability of defection is

ψC |M|ψC  = (αC∗ D C D| + αCC

∗
CC|)(|D DD D| + |C DC D|)(αC D |C D + αCC |CC)
(42)
= |αC D | 2
(43)

We model not knowing whether the opponent has defected or not as the superposition
|ψ = √12 (|ψ D  + |ψC ) When we are in this state of ignorance, we have

1 ∗
ψ|M|ψ = (α D D| + α∗DC DC| + αC∗ D C D| + αCC
∗
CC|)(|D DD D| + |C DC D|)
2 DD
(44)
(α D D |D D + α DC |DC + αC D |C D + αCC |CC) (45)
1 2
= (α D D + α2DC ) + Re(α D D D D|α DC DC) (46)
2

Again, this last term is an interference term, and may be negative. This means that
we can obtain the situation that the probability of defecting is smaller in the state of
ignorance than it is in the case that we know what the action of our opponent is.

3.6 Question Order Models and Contextuality

The quantum probability theory of judgement proposed in Busemeyer et al. (2011)

and described in Sect. 3.1 is applied to question order effects in Wang et al. (2013).
The question order effects are those taken from Moore (2002), outlined in Sect.
2.4. Consider the consistency effect exemplified by the questions “Do you think
Clinton is trustworthy?” and “Do you think Gore is trustworthy?”. The consistency
effect means that when the question is asked second (the comparative context), people
change their responses to make them more similar to the response to the first question,
in comparison to when the question is asked first. Wang et al. (2013) model this as
follows.
Quantum Computing and Cognitive Simulation 77

A person’s belief is represented by a state vector |s in an n-dimensional Hilbert

space. Each potential response x to a question Q corresponds to a subspace Qx that
has a unique orthogonal  projector PQx . The projectors to all responses to question
Q sum to the identity: x PQx = I . The probability of responding x to the question
Q is given by ||PQx ||2 , and the updated belief state after deciding on an answer to
P |s
the question is |s Qx  = ||PQx
Qx |s||
. As in Sect. 3.1, the probability of responding x to
question A and then y to question B is

p(Ax and then By) = ||PAx |s||2 ||PBy |s Ax ||2 = ||PBy PAx |s||2 (47)

An order effect for a given question occurs when the probability of giving an
answer to a question in the comparative context differs from the probability of giv-
ing that answer in the non-comparative context. In the Clinton/Gore setting, we
have that the total probability T PC y of answering ‘yes’ to the question “is Clinton
trustworthy?” in the comparative context is given by

T PC y = p(Gy and then C y) + p(Gn and then C y)

= ||PC PG |s||2 + ||PC P¬G |s||2 (48)

However, as we saw in Sect. 3.4, the probability of answering ‘yes’ in the non-
comparative context is given by

p(C y) = ||PC |s||2 = ||PC PG |s + PC P¬G )|s||2 (49)

= ||PC PG |s||2 + ||P¬G )|s||2 + 2Res|PG Pc (1 − PG )|s
= T PC y + 2Res|PG Pc (1 − PG )|s (50)

The order effect CC is therefore given by

T PC y − p(C y) = −2Res|PG PC (1 − PG )|s = 2||PC PG |s||2 − 2Res|PG PC |s

(51)
= 2||PC PG |s||2 − 2|s|PG PC |s|cosφ (52)
= 2||PC PG |s||2 − 2θ||PC |s|| · ||PG |s|| (53)

where θ = cosφ|s|PG PC |s|/(||PC |s|| · ||PG |s||).

A similar calculation gives the order effect for Gore, C G to be

C G = 2||PG PC |s||2 − 2θ||PC |s|| · ||PG |s|| (54)

Now, since θ is the same in both cases, we have

78 M. Lewis

0 = 2 p(Gy and then C y) − CC − (2 p(C y and then Gy) − C G ) (55)

= 2 p(Gy and then C y) − p(Gy and then C y) − p(Gn and then C y) + p(C y) (56)
− (2 p(C y and then Gy) − p(C y and then Gy) − p(Cn and then Gy) + p(Gy)) (57)
= p(Gy and then C y) − p(Gn and then C y) + p(C y and then Gy) + p(C y and then Gn)
(58)
− p(C y and then Gy) + p(Cn and then Gy) − p(Gy and then C y) − p(Gy and then Cn)
(59)
= p(C y and then Gn) + p(Cn and then Gy) − p(Gy and then Cn) − p(Gn and then C y)
(60)

This last quantity p(C y and then Gn) + p(Cn and then Gy) − p(Gy and then Cn)
− p(Gn and then C y) is termed q.
More generally, take the quantity p AB = p(Ay and then Bn) + p(An and then
By) to be the probability that the answers to the two questions are different in the AB
order, and take p B A = p(By and then An) + p(Bn and then Ay) in the B A order.
Then the quantum model predicts that

q = p AB − p B A = 0 (61)

Wang et al. (2013) test this theory on the data from Moore (2002) (described in
Sect. 2.4) using the sample frequencies to compute the values p AB and p B A , and then
using a z-test to compute whether q is significantly different from 0.
The conclusions are that the quantum question order model fits the data extremely
well. The three datasets exemplifying the consistency, contrast, and additive effects
have q-values that are not at all significantly different from 0. The fourth dataset
(Rose-Jackson) was not well modelled by the QQ model. However, the authors
argue that there is a violation of the assumptions in this model. Specifically, in this
case, additional information is supplied about each of the characters, and this entails
that the value θ is not the same in both cases. Hence, the cancellations worked out
in Eq. (56) do not hold in this case.
A crucial part of the argument in this paper is also that other probabilistic models
cannot account for both the question order effects seen and the QQ equality. In the
case of a Bayesian probability model, the assumption is that the event Ay is a set
contained in a sample space, and that By is another set contained in the same sample
space. The probability of the conjunction is then defined by the intersection of the
two events, which is commutative. In order to introduce order effects, two new events
must be introduced: an event O1: “A was asked before B”, and an event O2: “B was
asked before A”. However, in general, introducing these two events does not predict
the QQ equality to obtain.
Markov models can incorporate question order effects, since they can encode the
memory of previous events into the state representation. However, the authors argue
that in general, QQ equality is not upheld.
Quantum Computing and Cognitive Simulation 79

3.7 Contextuality

A fundamental aspect of Quantum Mechanics is contextuality. The phenomenon

of contextuality is as follows. Suppose we have a system with a number of agents
who can perform measurements on the system, and suppose that we have determined
joint distributions on the outcomes of pairs of measurements, by means of redoing the
experiment a number of times. Contextuality occurs when it is not possible to form
a joint distribution across all measurement outcomes such that this joint distribu-
tion agrees with the pairwise distributions. This means that before the measurement
happens, the outcome of the measurement is undetermined. Contrast this with the
outcomes of measurements in normal life. Suppose you wake up, and you wonder
about the weather outside. You can take a measurement (open your curtains) and
see that it is sunny, or cloudy. However, whether or not it was sunny or cloudy was
determined before you opened the window.
We explain this below following the explanation in Nielsen and Chuang (2010).
To show that the joint distribution does not exist, an analysis of the situation outside
of quantum theory is made, and a Bell inequality is formed. Suppose that we have
a system and two observers Alice and Bob. Alice can make two measurements, and
each has the possibility of returning 1 or −1. Bob can also make two measurements,
and each of these has the possibility of returning 1 or −1. The measurements are
each modelled as random variables A, A , B, and B  . We now consider the quantity

AB + A B + A B  − AB  = (A + A )B + (A − A)B  (62)

Since A, A = ±1, either (A + A )B = 0 or (A − A)B  = 0. And then we have

that AB + A B + A B  − AB  = ±2. We now consider what is termed a hidden
variable. We suppose that it is possible to determine the outcome of each of the
random variables individually so that A = a, A = a  , B = b, and B  = b , giving
us a joint probability p(a, a  , b, b ). Now, we can calculate the expected value of
AB + A B + A B  − AB  as

E(AB + A B + A B  − AB  ) = p(a, a  , b, b )(ab + a  b + a  b − ab )
a,a  ,b,b
(63)

 
≤2 p(a, a , b, b ) = 2 (64)
a,a  ,b,b

However, within quantum theory this inequality is violated. Consider the quantum
state |ψ = |01+|10
√
2
, and suppose Alice can perform measurements A = Z 1 or A =
X 1 on the first qubit and Bob can perform measurements B = −Z√2 −X
2
2
or B  = Z 2√−X
2
2

on the second qubit. Calculating the average value M of an observable M as

10

ψ|M|ψ, we find that

10 Z and X are the Pauli matrices.

80 M. Lewis

1 1 1 1
AB = √ , A B = √ , A B   = √ , AB   = − √ (65)
2 2 2 2
√
meaning that E(AB + A B + A B  − AB  ) = 2 2, and violating the inequality in
(64).
The violation of this inequality has been demonstrated experimentally, and shows
that either there is no set of definite values of the random variables A, A , B, and
B  that exist independent of our observation, or that the act of Alice’s measurement
affects the act of Bob’s measurement.
The notion of contextuality also arises within cognitive science. Contextual-
ity is crucially different from context-dependence. The phenomenon of context-
dependence arises when the answer to a question depends on the context that the
question is asked. This kind of context-dependence is called direct influences (Dzha-
farov & Kujala, 2016). However, another kind of context-dependence that is analo-
gous to quantum contextuality may be present. In Basieva et al. (2019), the following
example is given. Suppose we have the three questions
1. Do you like chocolate?
2. Are you afraid of pain?
3. Do you see your dentist regularly?
Imagine that you are asked a pair of these questions. It may well be that the context in
which you are asked a given question directly influences your answer to that question.
So if you are asked (1) and (2) together, you may respond positively to (1), but if you
are asked (1) and (3) together, the thought of going to the dentist may incline you to
answer negatively to (1). This is the notion of direct influence. Basieva et al. (2019),
following Dzhafarov and Kujala (2016), use the following notation: a response to
j
question qi asked in context c j is a random variable denoted Ri , and described by

j Y es N o response
Ri : j j (66)
pi 1 − pi probability

These random variables can be laid out as follows (Dzhafarov & Kujala, 2016):

R11 R21 c1
R22 R32 c2
(67)
R13 R33 c3
q1 q2 q3 system R3

This says that in context 1 (c1 ), questions 1 and 2 (q1 and q2 ) are asked, and so on.
Now, we might assume that all direct influences have somehow been eliminated,
so that you have the same probability of answering yes or no to a given question no
matter what the context. We then have that
Quantum Computing and Cognitive Simulation 81

R11 = R13 (68)

R21 = R22 (69)
R32 = R33 (70)

However, we can also imagine a setup in which, if you answer yes to one question
in the context, you always answer no to the other question. Representing yes and no
by 1 and −1, we then have

R11 = −R21 (71)

R22 = −R32 (72)
R13 = −R33 (73)

We can then easily derive a contradiction by following a chain of equalities:

R11 = R13 = −R33 = −R32 = R22 = R21 = −R11 (74)

This is saying that the pairs of random variables R11 and R13 , R21 and R22 , R32 and
R33 cannot be identical. This kind of context-dependence is not reducible to direct
influences.
Within quantum theory, contextuality applies only to systems in which there is
no notion of direct influences. Measurements of a property in different conditions
are described by the same random variable. The theory developed in Dzhafarov
and Kujala (2016), called contextuality-by-default or CbD, can also take account
of the presence of direct influences, and is therefore an extension of the quantum-
theoretic notion. CbD is able to quantify how much the distributions of responses
to different questions differ in different contexts, and how much the identities of the
random variables differ in different contexts. By comparing these two quantities, it
is possible to show how much contextuality is present over and above the effect of
direct influences.
In previous work examining the contextuality of human judgements, the impact
of direct influences was not always considered (Aerts et al., 2013; Dzhafarov &
Kujala, 2014; Khrennikov et al., 2014), and so it was not clear the extent to which
true contextuality was actually present in human decision-making. For example, in
Bruza et al. (2012, 2015), whilst the effect of direct influences was taken into account,
the presence of true contextuality was only minimally detected. However, in Basieva
et al. (2019), Cervantes and Dzhafarov (2019), Dzhafarov (2019), experiments have
been carried out to show that this phenomenon is indeed present in human judgement.
82 M. Lewis

3.8 Concept Combination and Overextension

As described in Sect. 2.6, the phenomenon of overextension and underextension has

been observed in experiments in Hampton (1987, 1988a, 1988b), Smith and Osherson
(1984).
In Aerts and Gabora (2005a, 2005b), the authors set out a quantum model of
concepts and their combination. The model of a concept is called the State-Context-
Property or SCOP formalism. The idea behind this is that typicality ratings of exem-
plars in a concept can depend on the context in which they are considered. For exam-
ple, in the context “The pet is chewing a bone”, a set of examples might be ordered as
dog, cat, rabbit, hamster, guinea pig, mouse, hedgehog, bird, parrot, snake, canary,
goldfish, spider, guppy, whereas under the context “The pet is being taught”, the same
set of exemplars would be ordered dog, parrot, cat, bird, hamster, canary, guinea
pig, rabbit, mouse, hedgehog, snake, goldfish, guppy, spider. The idea is that each
context acts on the concept to bring it into another state.
A SCOP consists of a set  of states pi , a set M of contexts ei , such as “the pet is
chewing a bone”, and a set L of properties ai , such as ‘is feathered’. Together with
these sets, there is a function

μ :  × M ×  → [0, 1] (75)

which is a probability function that describes how a state p under the influence of
context e changes into state q.

μ(q, e, p) =  p|Pe | p (76)

There is also a function

ν :  × L → [0, 1] (77)

which gives the weight of a property a for the concept in the state p. So, for example,
after considering the concept ‘pet’ in the context “the pet is chewing a bone”, the
weight of the property ‘is furry’ might increase.

ν( p, a) =  p|Pa | p (78)

In Aerts and Gabora (2005a), a Hilbert space representation of the theory is given.
Concepts are represented in a Hilbert space in which the basis dimensions of the space
represent basic states of the concept. The basic states of a concept are defined via
basic contexts of a concept, which roughly means that the basic concept contexts are
the most specific. For example, we might have the contexts a1 = “the pet is chewing
a bone”, and a2 = “the pet is chewing a bone in the garden”, and we would say that
a2 is more specific than a1 . The set of atomic contexts is denoted by X , and these are
considered to be rank-1 projectors. The basic states are then the states corresponding
to these rank-1 projectors. The ground state p̂ of a concept is given by the following
Quantum Computing and Cognitive Simulation 83

superposition of the basis states of the Hilbert space:

 1
| p̂ = |u (79)
u∈X
|X |

so that the ground state has an equal probability of changing to any basic state under
one of the basic contexts.
General contexts are represented by projectors that are not merely of rank 1. So
suppose that we have 100 basic contexts, and 25 of these basic contexts are more
specific than ebone = the pet is chewing a bone, i.e. they are more detailed contexts
in which the pet is chewing a bone.11 Then we would build a projector corresponding
to the context ebone by summing together the projectors for the basic contexts in the
set E bone = {u ∈ X |u is more specific than ebone }:

Pbone = |uu| (80)
u∈E bone

The new state under the influence of this context is calculated in the standard way,
i.e. in the same way that the knowledge state of the agent is evolved in Sect. 3.1,
Eq. (18):
Pbone | p̂
| pbone  = (81)
||Pbone | p̂||

In our example with 100 basic contexts and 25 of these basic contexts are more
specific than e1 , we have that
  1
Pbone | p̂ = |uu|| p̂ = √ |u (82)
u∈E bone u∈E bone
100
  1 25
||Pbone | p̂||2 =  p̂|Pbone | p̂ = | p̂|u|2 = = (83)
u∈E bone u∈E bone
100 100

so that
 1  1
| pbone  = √ |u = |u (84)
u∈E bone 25 u∈E bone
5

Recall that in the pet fish problem, the problem is that the exemplar ‘goldfish’ is
viewed as more typical to the ‘pet fish’, conceived as the conjunction of two concepts
‘pet’ and ‘fish’ than it is to either of the constituent concepts. To describe the notion
of the typicality of an exemplar to a concept, Aerts and Gabora (2005a) use the
function μ described in Eq. (75). Aerts and Gabora (2005a) explicate the notion

11 Note that there might be basic contexts that are unrelated to “the pet is chewing a bone”, for
example, “the pet is chewing a cracker in a cage and scratching its wing”—although this might be
a basic context, it is not related to the context “the pet is chewing a bone”.
84 M. Lewis

Table 6 Numbers of basic states assigned to each exemplar in the concepts pet and fish
Pet Fish
Dog 50 0
Cat 35 0
Goldfish 10 10
Shark 5 20
Tuna 0 20

of typicality as follows. Each exemplar is assigned to a context, so if we want to

consider the exemplar pgf = goldfish, we form the context egf = the pet is a goldfish.
Now, consider the concept pet in the context ebone = the pet is chewing a bone. As
described by Eqs. (18) and (84) in the example above, we apply the context to the
state pet to obtain the state pbone .
To calculate the weight of the exemplar goldfish in the context
the pet is chewing a bone, we calculate the probability that pbone changes to
pgf in the context egf , i.e. we calculate

μ( pgf , egf , pbone ) =  pbone |Pgf | pbone  (85)

Earlier, we said that there are 25 out of 100 basic contexts in which the pet is
chewing a bone. Suppose there are 10 contexts where the pet is a goldfish, but only
1 context where the pet is chewing a bone and is a goldfish. Then we will have
   1
μ( pgf , egf , pbone ) =  pbone |Pgf | pbone  = v|uu|w (86)
u∈E gf v∈E bone w∈E bone
25
 1 1
= = (87)
u∈E bone ∩E gf
25 25

So the weight of the exemplar goldfish in the context the pet is chewing a bone is
1
25
.This quantity is what Aerts et al. (2015) use to formalize the notion of typicality
of an exemplar to a concept.
To examine the pet fish situation, we will first set up an example situation. In Aerts
and Gabora (2005a), these figures were generated from a psychological experiment
involving asking humans for the typicality figures. In this chapter, we simplify the
figures for explanatory purposes.
Suppose that we have a pool of 100 basic states for pet and a pool of 50 basic
states for fish. The basic states are distributed amongst the following exemplars as in
Table 6.
For each concept, we outline how many basic contexts are included in each of the
x
contexts epet = the x is a pet, efish
x
= the x is a fish, efurry
x
= the x is furry, and eswim
x
=
the x is swimming around in a glass tank in the house.
Quantum Computing and Cognitive Simulation 85

Table 7 Number of basic contexts assigned to each context for each exemplar in the concept pet
pet pet pet pet
epet efish efurry eswim
Dog 50 0 45 0
Cat 35 0 30 0
Goldfish 10 10 0 9
Shark 5 5 0 5
Tuna 0 0 0 0

Table 8 Numbers of basic contexts assigned to each context for each exemplar in the concept fish
fish fish fish fish
epet efish efurry eswim
Dog 0 0 0 0
Cat 0 0 0 0
Goldfish 10 10 0 9
Shark 5 20 0 5
Tuna 0 20 0 0

The typicality of goldfish in the concept pet is then


μ(goldfish, goldfish, pet) =  pet|Pgoldfish | pet =  pet| |uu|| pet (88)
u∈E goldfish
1    10
= v|uu|w = (89)
100 u∈E v∈E pet w∈E pet
100
goldfish

20
A similar calculation gives us that the typicality of goldfish in the concept fish is 100 .
pet fish pet fish
The shared contexts of pet and fish are efish , epet , eswim , and eswim . Now, to form com-
binations of concepts, such as pet fish, the concept is represented in the tensor product
space of the two concepts pet and fish. The basic contexts of the concept pet fish are
denoted as E pet fish which is the intersection of the set E pet of the basic contexts of pet,
and the set E fish of the basic contexts of fish. The assumption is made for the exam-
pet fish pet pet
ple that E pet fish = E fish E pet , where E fish = {u ∈ E pet |u is more specific than efish },
i.e. it is the set of basic contexts of the concept pet that are more specific than
the pet is a fish. Note that in Tables 7 and 8, the cardinality of that set is 15.
The concept pet fish is modelled as
 1
|pet fish = √ |u ⊗ |u (90)
u∈E pet fish
15
86 M. Lewis

To look at the membership of goldfish in pet fish, we consider the context

the pet is a goldfish, modelled as

pet

G = Pgoldfish ⊗ 1fish = |uu| ⊗ I (91)
pet
u∈E goldfish

Then the typicality of goldfish in pet fish is given by

μ(goldfish, goldfish, pet fish)

= pet fish|G|pet fish (92)
1   
= v| ⊗ v| |uu| ⊗ I |w ⊗ |w (93)
15 pet fish pet
v∈E u∈E goldfish w∈E pet fish

1    10
= v|uu|vv|w = (94)
15 pet
15
v∈E pet fish u∈E goldfish w∈E pet fish

We can see that the typicality of goldfish in pet fish at 23 is much higher than the
1 2
typicality of goldfish in either pet ( 10 ) or fish ( 10 ).
Moreover, although we have only applied the context goldfish to the pet subsystem,
the fish subsystem has changed as well:

ν(goldfish, pet fish)

√
G| pet f ish 15   1
= √ = √ |uu| ⊗ I √ |w ⊗ |w (95)
pet fish|G|pet fish 10 pet 15
u∈E goldfish w∈E pet fish

1   1 
= √ |uu|w ⊗ |w = √ |u ⊗ |u (96)
10 pet pet fish 10 pet
u∈E goldfish w∈E u∈E goldfish

If we constructed a similar context saying that the fish is a goldfish, the concept would
collapse to the same state. In the pet fish concept, saying that the pet is a goldfish has
the same effect as saying that the fish is a goldfish—and this is the desired outcome.
In contrast, if the concept pet fish was modelled without entanglement, for example,
as

|pet fish  = |petfish  ⊗ |fishpet  (97)

 
= |u ⊗ |v (98)
pet fish
u∈E fish v∈E pet
Quantum Computing and Cognitive Simulation 87

Then
  
G|pet fish  = |uu| ⊗ I |v ⊗ |w (99)
pet pet fish
u∈E goldfish v∈E fish w∈E pet
 
= |u ⊗ |w (100)
pet fish
u∈E goldfish w∈E pet

where although the pet subsystem has changed to a goldfish, the fish subsystem has
not. If we constructed a similar context saying that the fish was a goldfish, we would
not obtain the same state. We would instead have a state consisting of a generic
pet, and a goldfish. As such, the entanglement is essential to correctly model this
problem.
Aerts and Gabora (2005b) go on to describe that this approach could be extended
to model the composition of concepts in more complicated structures, up to the
sentence level.
Whilst the approach given in Aerts and Gabora (2005b) is able to model the
phenomenon of overextension, it suffers from the problem that the space in which
concepts must be modelled gets longer dependent on the phrase or sentence length.
In Coecke et al. (2010), an approach to modelling language was proposed that com-
bines the very successful distributional models of meaning with the compositional
approaches from formal semantics. In brief, this approach models words in differing
vector spaces based on their grammatical type. Nouns are modelled in a vector space
N for nouns, and sentences are modelled in a vector space S for sentences. Adjectives
are modelled as linear maps on the noun space, i.e. matrices in N ⊗ N , and transitive
verbs are bilinear maps from two copies of the noun space to the sentence space, or
tensors in N ⊗ S ⊗ N . The benefit of this approach is that arbitrarily long sentences
can all be mapped into the same vector space S. Whilst this approach was originally
developed with an eye to natural language processing, it has also been used to tackle
problems in cognitive science.
In Coecke and Lewis (2015), the compositional distributional model is used to
examine the pet fish problem, and more generally the problem of overextension.
The first approach to tackling the pet fish problem is to view the word pet as an
adjective, rather than a noun. Then the parameters of the matrix can be tuned so
that the similarity of the vector goldfish to pet fish is greater than the similarity to
either pet or fish. The sentence can further be expanded out to look at the conjunc-
tions pet which is a fish and fish which is a pet, and this format is used to model the
importance of attributes in the conjunction of concepts. In Hampton (1987), data is
collected on the importance of attributes in concepts and their combination. Con-
cepts are considered in pairs that are related to some degree, for example, ‘Pets’ and
‘Birds’. Six pairs are considered in total. Participants are asked to generate attributes
for each concept and for their conjunctions, where conjunction in this case is ren-
dered as “Pets which are also Birds”, or “Birds which are also Pets”. For example,
attributes such as ‘lives in the house’, ‘is an animal’, and ‘has two legs’ are generated
for ‘Pets’, ‘Birds’. For each pair of concepts and their conjunction, attributes that
88 M. Lewis

had been generated by at least 3 out of the 10 participants were collated. Participants
were then asked to rate the importance of each attribute to each concept and to each
conjunction. Importance ratings were made on a 7-point verbal scale ranging from
“Necessarily true of all examples of the concept” to “Necessarily false of all exam-
ples of the concept”. Numerical ratings were subsequently imposed ranging from 4
to −2, respectively.
The question then arises of how the importance of attributes in the conjunction
of the concepts is related to the importance of attributes in the constituent concepts.
Key phenomena are that conjunction is not commutative, that inheritance failure can
occur (i.e. an attribute that is important in one of the concepts is not transferred to the
conjunction), that attribute emergence can occur, where an attribute that is important
in neither of the conjuncts becomes important in the conjunction, and further, that
necessity and impossibility are preserved. In order to model this data, Hampton uses
a multilinear regression.
In Coecke and Lewis (2015), the phrase A which is a B is modelled as

|A which is a B = |A is|B

where is is a matrix modelling the word is as a transitive verb.

To model the data from Hampton (1987), parameters for the verb is are found
using a numerical solver. Results are that it is essentially possible to model the data
exactly. We should note that since a large matrix is used, the numbers of parameters
are very large, however, as a demonstration of how the compositional model can be
applied to psychological data it is still useful.

3.9 Modelling of Over- and Underextension in Psychological

Data

The phenomenon of over- and underextension in human concept combination has

been extensively investigated in Hampton (1988a, 1988b). In Hampton (1988a),
Hampton looks at how the membership of exemplars in conjunctions of concepts
compares that the membership of exemplars in the individual concepts. In Hampton
(1988b), he examines how the membership of exemplars in disjunctions of concepts
compares to that in the individual concepts. The overall findings are that in the con-
junctive case, overextension is often seen, and in the disjunctive case, underextension
is often seen. Moreover, membership in the conjunction is dependent on the order in
which the conjunction is made. As we saw in Sect. 3.2, if concepts are modelled as
regions of a feature space, the order of conjunction should not matter.
These phenomena are given a quantum modelling in Aerts (2009). The model is
based on a Fock space. The combination of two concepts is considered in two differing
ways. Concepts A and B are modelled as states |A and |B of a Hilbert space H.
Firstly, the disjunction of two concepts A and B is considered as the superposition
of their two states: A or B = √12 (|A + |B). The membership μA or B (X ) of an item
X in the concept A or B is given by an orthogonal projection M X , i.e.
Quantum Computing and Cognitive Simulation 89

1
μA or B (X ) = (A| + B|)M X (|A + |B)
2
1
= (A|M X |A + B|M X |B + Re(A|M X |B)
2
1
= (μA (X ) + μB (X )) + Re(A|M X |B)
2
where this last term is the interference term, as in Sect. 3.4.
On the other hand, the disjunction of the two concepts can be modelled in two
copies of H, as |A ⊗ |B. The operator M X ⊗ M X is a projection operator over the
space that asks whether X is a member of A and whether X is a member of B. To
determine the membership of X in A or B, we therefore calculate

μA or B (X ) = 1 − (A| ⊗ B|)(I − M X ) ⊗ (I − M X )(|A ⊗ |B)

= 1 − (1 − μ A (X ))(1 − μ B (X )) = μ A (X ) + μ B (X ) − μ A (X )μ B (X )

which gives us a classical-style disjunction.

For the case of conjunction, a classical-style conjunction can be modelled in
H ⊗ H simply by taking

μA and B (X ) = A| ⊗ B|)(M X ⊗ M X )(|A ⊗ |B)

= μ A (X )μ B (X )

and the conjunction in one copy of H is modelled as √12 (|A + |B) again. This
means that in the single space, conjunction and disjunction are the same.
The single space and the double space are combined in a Fock space H ⊗ H ⊕ H.
The concept overall is represented as

neiφ
ψ(A, B) = meiθ |A ⊗ |B + √ (|A + |B)
2

where m 2 + n 2 = 1. Then,

μA and B (X ) =

neiφ neiφ
meiθ A| ⊗ B| + √ (A| + B|) M X ⊗ M X ⊕ M X meiθ |A ⊗ |B + √ (|A + |B)
2 2
1
= m 2 μ A (X )μ B (X ) + n 2 ( (μ A (X ) + μ B (X )) + Re(A|M X |B))
2

A similar calculation gives us that

1
μA or B (X ) = m 2 (μ A (X ) + μ B (X ) − μ A (X )μ B (X )) + n 2 ( (μ A (X ) + μ B (X )) + Re(A|M X |B))
2
90 M. Lewis

Essentially, the part of the concept that is modelled as a superposition contributes to

the combinations A and B or A or B to move the combination away from a classical
combination of concepts.
Aerts (2009) then shows that the data from Hampton (1988a, 1988b) can be
modelled in a space R2 ⊗ R2 ⊕ R4 , in which the angles θ and φ are tuned.
In Sozzo (2014), the same sort of Fock space modelling is used to describe
problematic borderline cases in negation. Sozzo (2014) models the data from Alx-
atib and Pelletier (2011) by modelling the concepts tall and not tall as separate
states in the one-system part of the space, H. The combinations tall and not tall and
neither tall nor not tall are then modelled in the two-system part of the space H ⊗ H,
and combined with the superposition of |A and|A . As such, the superposition of
the two states combines with their logical combination to produce the effects seen
in Alxatib and Pelletier (2011).
In Aerts et al. (2015), Sozzo (2015), this Fock space modelling of conjunction and
negation are combined. If a concept A is modelled by |A, then the negation of that
concept is modelled by another state |A . Within the first part of the Fock space, in
which the concepts A and B are modelled within a single Hilbert space H, negations
are modelled using |A  and |AB. However, within the two-system part of the Fock
space, the negations of concepts can be represented classically, as A and B or A or B
are. Specifically, the membership of X in the negation of A is modelled as

A|I − M X |A = 1 − μ A (X )

This system is able to successfully model data from Hampton (1997) and from Sozzo
(2015).

3.9.1 Ambiguity, Entanglement, and Concept Combination

A key part of human concept use is how we express these concepts in language.
However, when concepts are expressed in language, the language can have one word
which refers to the same concept. A well-used example is the word bank. If I say
“I’m going to the bank to deposit some money”, then bank has a different meaning
than if I say “I’m going to the bank to fish”. In Bruza et al. (2012), an experiment
is performed that looks into pairs of such ambiguous words and their combinations.
An example is apple chip. This could mean a part of a Macintosh computer, or a
slice of dried fruit. If I interpret apple in the sense computer, then it is very unlikely
that I should interpret chip as a slice, or part, or the computer. Similarly, if I interpret
apple in the sense fruit, then it is very unlikely that I will interpret chip in the sense
of a computer chip.
A quantum modelling of this phenomenon would seem very appropriate: if we
view each individual concept as modelled in a Hilbert space H, and I disambiguate
the meanings so that I have 4 vectors |applecomputer , |applefruit , |chipcomputer , and
|chipslice , then we can model the combined concept in H ⊗ H as
Quantum Computing and Cognitive Simulation 91

|applecomputer |chipcomputer  + |applefruit |chipfruit 

apple chip = √
2

Given a state |ψ that the agent is in, the interpretation of the combination will
collapse to either the computer sense or the fruit sense.
However, another form of combination is as follows. We could model
apple as |apple = √12 |applecomputer  + |applefruit , and |chi p = √12 |chipcomputer  +
|chipslice 

apple chip = |apple ⊗ |chi p (101)

1
= (|applecomputer |chipcomputer  + |applecomputer |chipfruit  (102)
2
+ |applefruit |chipcomputer  + |applefruit |chipfruit ) (103)

In this case, a participant could interpret the combination apple chip in any of the
four possible combinations.
This kind of phenomenon is investigated in Bruza et al. (2015) In these experi-
ments, the participant is first primed for one of the senses of the combination—for
example, the word banana is shown to them before they are asked to make a decision
on the interpretation of the compound. The participants are then asked to describe
the interpretation they have made. These interpretations are used to build a set of
probabilities for the interpretations of the compounds, and these probabilities are
analysed to determine whether the compound is separable or entangled. There are
a number of factors which need to be accounted for in this, which are discussed in
Dzhafarov and Kujala (2016) and which we discussed in Sect. 3.7.
The results are that in general, there is no clear evidence for the entanglement of
these concepts in the human mental lexicon. However, in a number of cases, there
is some evidence of phenomena occurring which are not completely explained by a
classical probabilistic model.
Another approach to modelling the ambiguity of words is to model words as
density matrices (Meyer & Lewis, 2020; Piedeleu et al., 2015). This was proposed
as an extension of the compositional distributional model of Coecke et al. (2010). As
mentioned in Sect. 3.1, a density matrix is a probability distribution over a number
of quantum states: 
ρ= αi |vi vi | (104)
i


where ∀i, αi > 0 and i αi = 1. The idea of using density matrices to model ambi-
guity runs as follows. Each sense of a word is modelled as a vector, and then these
senses are mixed together following Eq. (104) above. Crucially, the same composi-
tional framework as in the standard vector-based model is available. This means that
as words are used in context, their meaning can be disambiguated, in the same way
that humans are able to. In Meyer and Lewis (2020), a neural approach to building
density matrices for words is proposed and tested on a set of sentence similarity
92 M. Lewis

data that requires words to be disambiguated correctly. The representations learnt

by Meyer and Lewis (2020) perform very well, outperforming some of the strongest
neural network systems for language processing currently available.

3.10 Cognitive and Neuronal Structures

In the foregoing sections, the level of modelling was at a high, cognitive level. There
is no attempt to form any match between these models and actual processes in the
brain. However, there has been a range of research into using quantum theory to
model cognitive processes at a level that is closer to a neuronal level.
Fuss and Navarro (2013) describe a quantum model of two-alternative forced
choice (2AFC) experiments. In this sort of experiment, the participant is asked to
make a choice about a stimulus. For example, there may be two shapes on a computer
screen and the participant is asked which one is a square. The participant must choose
one or the other, and does not have to choose to say neither. Information about the
nature of processing is inferred from reaction times. For example, the average reaction
time to the image in Fig. 8a may be faster than the average reaction time to the image
in Fig. 8b.
A standard way of modelling reaction times in this sort of experiment is to say that
a pair of neurons, or groups of neurons, is responsible for accumulating evidence for
one choice or the other, and that when the evidence hits a threshold, the participant
reacts. The two accumulators are in a race to get to the threshold. So in the examples in
Fig. 8, evidence is being accumulated by one accumulator for left and one for right.
In Fig. 8a, evidence for right would only accumulate slowly, and so the reaction
left would happen more quickly, whereas in Fig. 8b, more of the evidence would
be added to the right accumulator, and so overall the threshold would take longer to
reach. This is usually modelled as a random walk with drift. An extension to a simple
random walk model is to consider that multiple random walks could be happening in
parallel. However, having multiple random walks happening in parallel can still only
produce a model where evidence is accumulated. On the other hand, a model that
explicitly models neurons and the connections between them (connectionist models)
can account for inhibitory connections, i.e. that evidence for one choice can inhibit
the accumulation of evidence for the other choice.

Fig. 8 Differing stimuli in a 2AFC experiment

Quantum Computing and Cognitive Simulation 93

In Fuss and Navarro (2013), a parallel random walk model using quantum the-
ory is proposed. This allows for the kinds of inhibition we just mentioned to be
modelled, by means of interference. This kind of model can produce qualitatively
different probability distributions over the search space, allowing for more efficient
decision-making to be carried out. The authors show that this kind of model can fit
2AFC experimental data more closely than a simple standard random walk with drift
(although there are more complex models which may do better).
More recently, Khrennikov et al. (2018) have begun to formulate a description
of neuronal substrates within quantum terms. This does not claim that neuronal and
cognitive activities are being modelled at the quantum level (although models to
explain cognitive activity through actual quantum modelling has been proposed in
Hameroff (2014), Penrose (1990)). Rather, the idea is that quantum theory can be
used to represent the information and activity present in neuronal assemblies.

3.11 Tensor Product Representations

Whilst not explicitly quantum-inspired, the tensor product representations of Smolen-

sky (1990) use structures which are native to quantum systems. The aim in Smolensky
(1990) is to describe a means of representing symbolic structures and their compo-
sition within a connectionist substrate. Connectionist networks (nowadays usually
called neural networks) are based on the idea that cognitive systems can be modelled
by the interconnection of small artificial neurons, and indeed the motivation for this
is clear—as our cognitive system is based on the interconnection of many interacting
neurons. There are other aspects to the basis of our cognitive system, such as the
physical bodies that each of us inhabits, but undeniably our (actual) neural network
plays a part in cognition. Also, there is no claim here that artificial neural networks
bear any resemblance to real neural networks. However, artificial neural networks
are extremely successful at various tasks.
An aspect of cognition that humans excel at is symbolic manipulation. This
doesn’t mean complex mathematics, but the ability that if we understand the sentence
Jen is Junpa’s mother, we can also understand the sentence Junpa is Jen’s mother,
even if we judge it to be false.
Artificial neural networks don’t have an obvious means of representing symbols
so that we can manipulate them in this way, and in Smolensky (1990), Smolensky
and Legendre (2006) a method for doing so is proposed.
Vectors and tensor products are used to represent objects, roles, and structures.
The underlying idea is that roles, such as subject, or object, are represented as vectors,
and fillers, also represented as vectors, can be bound to these roles using the tensor
product, and collections of roles and fillers are combined using vector addition.
Recursive structures such as trees can be represented. The contents of the leaves are
encoded in the fillers, and role vectors encode the tree structure (Table 9).
To be able to represent trees, we view the binding of a role and a filler as itself a
filler, i.e. the filler is itself a complex structure, and not an atomic symbol. The binding
94 M. Lewis

Table 9 Space of descriptions

Structure Symbolic Connectionist
Set f f ∈ VF
String fi /ri fi ⊗ ri

Tree s = {fi /ri } s = i fi ⊗ ri ∈ S ∗

f/r of a filler f to a role r is represented

 as a vector f ⊗ r. A sentence s is represented
as a sum of filler/role bindings i fi ⊗ ri , and these can be applied recursively. For
example, let s = [NP VP] be a binary tree with left and right subtrees NP and VP.
Let s, v1 , v2 be the vectors representing s, NP, VP. The connectionist representation
of s is

s = v1 ⊗ r0 + v2 ⊗ r1

If VP is a tree rather than an atomic symbol, it can be expressed in terms of its

left and right subtrees VP = [Vt NP]. If v3 , v4 represent the trees Vt, NP, then the
structure s = [NP [Vt NP]] has the following representation:

s = v1 ⊗ r0 + (v3 ⊗ r0 + v4 ⊗ r1 ) ⊗ r1
= v1 ⊗ r0 + v3 ⊗ (r0 ⊗ r1 ) + v4 ⊗ (r1 ⊗ r1 )
≡ v1 ⊗ r0 + v3 ⊗ r01 + v4 ⊗ r11

A notable feature of this representation is that the vector space in which concepts
live must be arbitrarily large, depending on the size of the structure to be represented.
Symbols at depth d in a binary tree are realized by S(d) , the F R d -dimensional vector
space formed from vectors of the form f ⊗ ri ⊗ r j ⊗ · · · ⊗ rk with d role vectors,
where F is the dimension of the filler vectors f and R is the dimension of the individual
role vectors ri . A vector space containing all vectors in S(d) for all d is

S ∗ ≡ S(0) ⊕ S(1) ⊕ S(2) ⊕ · · ·

Vectors s(i) are embedded into this space, meaning that the normal operation of vector
addition can be used to combine sentence components.
Once a set of concepts (e.g. Jen, loves, and Junpa) have been combined into
one vector representation, it is also possible to pull the representation apart, and
recombine it. This is done by means of an unbinding mechanism. Suppose we form
the sentence representation

|S = |J en ⊗ |subj + |r uns ⊗ |ver b

Quantum Computing and Cognitive Simulation 95

we can ask “who is the subject of the sentence” by unbinding |subj from the
sentence. If role vectors are all orthogonal, then the unbinding is simply done by
taking the inner product of |S with the role vector: S|subj = |J en.
Tensor product networks have recently been used extensively to examine the
structure of recurrent neural networks (McCoy et al., 2019), and also as a potential
substrate for quantum language processing (Wiebe et al., 2019). Moreover, very
similar structures have been used in the neural-based computing framework Nengo
(Bekolay et al., 2014). These latter structures are different in that they use circular
convolution (Plate, 1995) rather than tensor product to bind roles and fillers. The
benefit of using circular convolution rather than the tensor product is that all vectors
are mapped into one shared space. An approximate unbinding mechanism can be
employed by using circular correlation with a query vector.
Tensor product representations have also been used in Pothos and Truebloodm
(2015) to model how the similarity of objects can depend on where the similarity
is to be found. For example, given two birds with a spotted pattern, if the spotted
pattern is found on the wings of one bird and the back of the other, they will not be
judged as similar as if they both have the spotted pattern on their wings. The use of
role vectors in the tensor product representations of Smolensky (1990), Smolensky
and Legendre (2006) is leveraged to account for this phenomenon.
The compositional distributional model Coecke et al. (2010), mentioned in Sect.
3.8, has the property that all representations are mapped into one shared space, and
shares some structures similar to the ICS model. In Al-Mehairi et al. (2017), it
is shown that the ICS model can be mapped onto the compositional distributional
model, to form a categorical compositional model of cognition.
A further mapping from ICS to the compositional distributional model can be
considered. One of the most useful aspects of the compositional distributional model
is the way that functional words such as adjectives and verbs modify the nouns
they are applied to, either by mapping them to another vector in the noun space,
or by mapping them into a sentence space. Within the ICS model, this does not
happen in the same way. Consider the following problem, as investigated in Martin
and Doumas (2020). Under the assumption that the concepts we use are expressed
by the language we use, the way that we compose concepts can be examined by
looking at the properties of words and phrases. Within the field of neural modelling,
a key question is how neural representations of two concepts can be composed into
a single combined representation. Neural representations of concepts are commonly
modelled as (normalized) vectors, and similarity between concept representations
is calculated by the inner product. Two common proposals for how to combine
neural representations are via firstly tensor product binding or holographic reduced
representations, as described above, or secondly vector addition (Hummel, 2011).
The problem with tensor product binding is that it does not transform the argument
of a relation in the necessary way. Consider vectors |fluffy, |cat, and |dog. If the
combinations fluffy dog and fluffy cat are represented by the tensor product of the
two component vectors:
96 M. Lewis

Fig. 9 Fluffy dogs and

fluffy cats are more similar
than dogs and cats

fluffy dog = |fluffy ⊗ |dog

fluffy cat = |fluffy ⊗ |cat

then the similarity between fluffy cat and fluffy dog is just the similarity between cat
and dog, since

(fluffy| ⊗ dog|)(|fluffy ⊗ |cat) = fluffy|fluffy ⊗ dog|cat

= dog|cat (105)

This is undesirable, as can be seen in Fig. 9.

Quantum Computing and Cognitive Simulation 97

On the other hand, vector addition has been shown to be successful in modelling
concept combination Hummel (2011), Martin and Doumas (2020), and has a clear
neural correlate, termed synchronous binding.12,13,14,15
However, it is arguably not able to express gradations in meaning that are needed
in terms like red car versus red wine. For example, given |red, |wine, and |car,
suppose that wine and car are not very similar, say wine|car = 0.1, and they are
each fairly similar to red, say red|car = red|wine = 0.5. Then, adding vectors
and normalizing:

|red + |car
red car = √
3
|red + |wine
red wine = √
3

and

red car|red wine

1
= (red|red + red|car + red|wine
3
+ wine|car)
2 wine|car
= +
3 3
i.e. the similarity has increased from 0.1 to just over two-thirds. However, this again
does not accord with our intuitions: the similarity between wine and car should
not increase dramatically when the adjective red is applied, at least in part because
applying this adjective does not even make the colours very similar. It might be
argued that this can be overcome by downweighting the importance of the adjective
in an adjective-noun combination. However, there are situations when we would like
similarity to increase when the adjective is applied: the adjective red in red car and
red van should increase similarity.
A approach than can address both these aspects is the quantum-inspired approach
of Coecke et al. (2010). Since adjectives and verbs are represented as matrices rather
than vectors, the combination of an adjective and a verb transforms the argument so

12 https://commons.wikimedia.org/wiki/File:Wayfield%27s_Young_Argos,
_the_Labrador_Retriever.jpg. Attribution: Andrew Skolnick, en:User:Askolnick, CC BY-SA
3.0. http://creativecommons.org/licenses/by-sa/3.0/, via Wikimedia Commons.
13 https://commons.wikimedia.org/wiki/File:Cat_November_2010-1a.jpg. Attribution: Alvesgas-

par, CC BY-SA 3.0. https://creativecommons.org/licenses/by-sa/3.0, via Wikimedia Commons.

14 https://commons.wikimedia.org/wiki/File:Cute_dog_on_beach,_Los_Angeles,

_California_LCCN2013634674.tif. Attribution: Carol M. Highsmith, Public domain, via Wikime-

dia Commons.
15 https://commons.wikimedia.org/wiki/File:White_Persian_Cat.jpg. Attribution: Optional at the

Persian language Wikipedia, GFDL. http://www.gnu.org/copyleft/fdl.html, via Wikimedia Com-

mons.
98 M. Lewis

that fluffy cats can be moved closer to fluffy dogs, i.e. the problem in Eq. (105) is
not encountered. On the other hand, the application of an adjective like red does not
force red car to become similar to red wine. How? Essentially, the use of a matrix
gives enough ‘space’ for different uses of words to be represented.

3.11.1 Reasoning in Vector Spaces

Another line of work looks at using vector space methods to model structure in
concepts. A key characteristic of the concepts we use is that they are hierarchical.
Besides categorizing individual entities, we want to talk about types of entities—for
example, we want to simply say that tigers are dangerous, rather than “that tiger
is dangerous. And that tiger is dangerous. And that one. And...”. Within a quantum
model, we can start doing this by considering concepts as subspaces of a vector space,
and using the lattice structure of the subspaces, or equivalently their projectors, to
reason with. This was originally proposed by Birkhoff and Von Neumann (1936),
and was applied in an information retrieval setting by Widdows and Peters (2003).
In that work, the authors describe a means of applying negation and conjunction in
document queries, based on quantum logic.
The density matrix variant of the compositional distributional model of Coecke
et al. (2010) forms an extension of the quantum logic formed by the lattice of projec-
tors. Words are modelled as positive operators, and hierarchy in concepts is modelled
via the Löwner order. In Bankova et al. (2019), Sadrzadeh et al. (2018), it is shown
that the Löwner order is compatible with the compositional model used. In Lewis
(2019), a concrete modelling of words as positive operators is given which is shown
to model human judgements of entailment very well. Further work into modelling
negation is underway (Lewis, 2020), and work on modelling other forms of logical
structure such as generalized quantifiers has also been undertaken by Hedges and
Sadrzadeh (2019), Dostál et al. (2021).

4 Concluding Remarks

There are clearly a number of applications of quantum theory to problems in cognitive

science. We have seen that human use of concepts has some unusual features, which
seem to be well modelled using a quantum framework.
In the approach proposed by Busemeyer et al. (2011), Pothos et al. (2013), Wang
et al. (2013) and used in numerous other papers, a concept is modelled as a subspace
of a Hilbert space, and the knowledge state of an agent is modelled as a state in
the Hilbert space. The theory therefore crucially incorporates the way in which
decisions are dependent on the state of the agent, and how the presence or absence of
information and the order in which stimuli occur can affect judgements. Within this
model, the phenomenon of interference does a lot of work. In the modelling of the
asymmetry of similarity judgements, ||PChina |ψKorea ||2 · ||PKorea |ψ||2 is interpreted
Quantum Computing and Cognitive Simulation 99

as the similarity of Korea to China. However, in the model of the conjunction fallacy,
this is interpreted as the conjunction of A and then B, as it also is in the question
order model.
The conjunction and negation model proposed by Aerts and Gabora (2005b) also
models concepts as subspaces of a vector space. Here, the subspaces have very high
dimensionality and are based on very specific contexts that objects are seen in. The
notion of entanglement plays a key part in modelling the combination of concepts
such as pet fish. Aerts (2009), Aerts et al. (2015), Sozzo (2015) model concepts as
states within a vector space, and again use the notion of entanglement. In this latter
work, the vector spaces used are smaller—for example, around eight-dimensional.
A drawback of these compositional methods is that in all cases, the size of the
underlying space grows as more concepts are added to the combination. This has at
least two drawbacks: for more complex concepts, spaces may grow unfeasible large,
and furthermore it means that combinations of unequal numbers of concepts are not
easy to compare.
Various papers have examined whether a form of contextuality is present in human
concept use. In some cases, a claim of contextuality is erroneous, as argued by
Dzhafarov and Kujala (2016). On the other hand, Basieva et al. (2019), Dzhafarov
(2019) have shown that experiments can be designed in which human decision-
making does show contextuality. If human decision-making does indeed display true
contextuality, then the use of quantum computing to model these forms of decisions
may provide a very useful substrate. However, in many cases there is no evidence of
contextuality over and above the effect of direct influences, a more straightforward
form of context-dependence.
Vector-based models of meaning and compositionality have been extremely suc-
cessful in modelling language (Coeckeetal., 2010; Kartsaklis & Sadrzadeh, 2016;
McCoy et al., 2019; Sadrzadeh et al., 2018; Smolensky, 1990). These models use the
same fundamental structures as quantum theory, and extensions of the compositional
distributional model have incorporated crucially quantum phenomena such as den-
sity matrices Bankova et al. (2019), Meyer and Lewis (2020), Piedeleu et al. (2015),
Sadrzadeh et al. (2018). The models proposed in Smolensky (1990) can be argued
to have relevance for realistic neuronal modelling (Bekolay et al., 2014). Moreover,
theory from Smolensky (1990) has been developed for potential implementation on
quantum computers in Wiebe et al. (2019), and the model from Coecke et al. (2010)
has actually been implemented on a modern quantum computer (Lorenz et al., 2021).
This implementation takes advantage of the fact that the fundamental structures in
this model are the same as those used in quantum theory. In Lorenz et al. (2021),
experiments on sentence classification are run, one to classify sentences as belonging
to one of two topics, and a second to classify noun phrases as containing either a
subject relative clause or an object relative clause. The general methodology of the
experiments is to first generate a linguistic parse of a given sentence, and to express
the sentence in terms of a DisCoCat diagram (as in Coecke et al. (2010)). The diagram
is rewritten to eliminate complex aspects of the diagram—namely by straightening
wires in the diagram in a certain way. Following this step, it is then possible to build a
quantum circuit whose structure matches the grammatical structure of the sentence.
100 M. Lewis

A quantum compiler translates the circuit into instructions for a physical quantum
computer, which runs the circuit a certain number of times, and the relative frequency
of given answers can be recorded.

4.1 Potential for Quantum Computing to Contribute

to Cognitive Science

So what is the potential for cognitive science within the realm of quantum computing?
On the one hand, we can simply say that human cognitive faculties are not computers,
whether classical or quantum. However, this is unnecessarily negative. There are a
wide range of cognitive processes that are well modelled using quantum theory.
Models of similarity, judgement, and decision-making moreover seem to be well
modelled even in fairly low-dimensional spaces.
At the same time, models that incorporate compositionality at a more complex
level, using the grammar of natural language, have been implemented on quantum
computers, and the potential for the benefits of this are large. The successes in Lorenz
et al. (2021) have the potential to be extended to the cognitive domain. Drawing
together some of the theory from the similarity and decision-making models to
form a unified whole with models of language is an important way forward for this
endeavour. One way to progress this would be to consider a form of compositionality
that can be applied at the cognitive, rather than the language level, like a language
of thought (Fodor, 1975).
Fodor’s language of thought hypothesis is summarized (by himself (Fodor &
Pylyshyn, 1988)) as the idea that cognition is reliant on mental representations, and
that those mental representations have a combinatorial syntax and semantics. This
argument is based on the fact that human cognitive abilities have three key properties:
• Productivity: We are able to entertain a large number of thoughts and to generate
these on the fly, as combinations of representations that we already have.
• Systematicity: This is the property that if, for example, “Junpa loves Jen” can be
understood, then “Jen loves Junpa” should also be understood, whether or not it
is true.
• Compositionality: This is the property that the meaning of complexes of repre-
sentations should be understandable from the meanings of the individual repre-
sentations and how they are combined.
Whilst natural language arguably has these properties, and furthermore is intrinsically
related to our thoughts, we do not have a ‘grammar’ that we can apply to thoughts to
combine them in the same way that grammar exists for natural languages. Developing
such a model together with the right sort of representations and notions of similarity
and decision-making has the potential to form a very powerful model of cognition.
Utilizing the compositional approach proposed in Coecke et al. (2010) may provide
a way forward for a form of cognitive architecture that can be implemented on a
quantum computer.
Quantum Computing and Cognitive Simulation 101

In this chapter, we have covered a range of quantum approaches to cognitive tasks.

These all have the capacity to model certain aspects of human cognitive processes
more faithfully than other approaches. The compositional models of Coecke et al.
(2010) and Smolensky (1990) may be able to extend the purely cognitive approaches
to model more complex phenomena in an intrinsically quantum way. Moreover,
these compositional models are already being implemented on quantum computers.
Hence, using quantum computing for cognitive simulation, and AI more generally,
has a bright future ahead of it.

References

Aerts, D. (2009, October). Quantum structure in cognition. Journal of Mathematical Psychology,

53(5), 314–348. ISSN 00222496. https://doi.org/10.1016/j.jmp.2009.04.005. https://linkinghub.
elsevier.com/retrieve/pii/S0022249609000558
Aerts, D., & Gabora, L. (2005a, January). A theory of concepts and their combinations I: The
structure of the sets of contexts and properties. Kybernetes, 34(1/2), 167–191. ISSN 0368-492X.
https://doi.org/10.1108/03684920510575799
Aerts, D., & Gabora, L. (2005b, January). A theory of concepts and their combinations II:
A Hilbert space representation. Kybernetes, 34(1/2), 192–221. ISSN 0368-492X. https://
doi.org/10.1108/03684920510575807. https://www.emerald.com/insight/content/doi/10.1108/
03684920510575807/full/html
Aerts, D., Gabora, L., & Sozzo, S. (2013). Concepts and their dynamics: A quantum-theoretic
modeling of human thought. Topics in Cognitive Science, 5(4), 737–772. ISSN 1756-8765. https://
doi.org/10.1111/tops.12042. https://onlinelibrary.wiley.com/doi/abs/10.1111/tops.12042
Aerts, D., Sozzo, S., & Veloz, T. (2015). Quantum structure of negation and conjunction in human
thought. Frontiers in Psychology, 6. ISSN 1664-1078. https://doi.org/10.3389/fpsyg.2015.01447.
https://www.frontiersin.org/articles/10.3389/fpsyg.2015.01447/full
Al-Mehairi, Y., Coecke, B., & Lewis, M. (2017). Categorical compositional cognition. In J. A. de
Barros, B. Coecke, & E. Pothos (Eds.), Quantum interaction. Lecture notes in computer science
(pp. 122–134). Springer International Publishing. ISBN 978-3-319-52289-0. https://doi.org/10.
1007/978-3-319-52289-0_10
Alxatib, S., & Pelletier, F. J. (2011). The psychology of Vagueness: Borderline cases and contradic-
tions. Mind & Language, 26(3), 287–326, 2011. ISSN 1468-0017. https://doi.org/10.1111/j.1468-
0017.2011.01419.x. https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1468-0017.2011.01419.x
Ashby, F. G., & Gott, R. E. (1988). Decision rules in the perception and categorization of multi-
dimensional stimuli. Journal of Experimental Psychology: Learning, Memory, and Cognition,
14(1), 33–53. ISSN 1939-1285 (Electronic), 0278-7393 (Print). https://doi.org/10.1037/0278-
7393.14.1.33
Bankova, D., Coecke, B., Lewis, M., & Marsden, D. (2019, March). Graded hyponymy for com-
positional distributional semantics. Journal of Language Modelling, 6(2), 225. ISSN 2299-
8470, 2299-856X. https://doi.org/10.15398/jlm.v6i2.230. http://jlm.ipipan.waw.pl/index.php/
JLM/article/view/230
Basieva, I., Cervantes, V. H., Dzhafarov, E. N., & Khrennikov, A. (2019, November). True contextu-
ality beats direct influences in human decision making. Journal of Experimental Psychology: Gen-
eral, 148(11), 1925–1937. ISSN 1939-2222, 0096-3445. https://doi.org/10.1037/xge0000585.
http://doi.apa.org/getdoi.cfm?doi=10.1037/xge0000585
Bekolay, T., Bergstra, J., Hunsberger, E., DeWolf, T., Stewart, T. C., Rasmussen, D., Choo, X.,
Voelker, A. R., & Eliasmith, C. (2014). Nengo: A Python tool for building large-scale functional
102 M. Lewis

brain models. Frontiers in Neuroinformatics, 7. ISSN 1662-5196. https://doi.org/10.3389/fninf.

2013.00048. http://journal.frontiersin.org/article/10.3389/fninf.2013.00048/abstract
Birkhoff, G., & Von Neumann, J. (1936). The logic of quantum mechanics. Annals of Mathematics,
37(4), 823–843, 1936. ISSN 0003-486X. https://doi.org/10.2307/1968621. https://www.jstor.
org/stable/1968621
Bruza, P. D., Kitto, K., Ramm, B., Sitbon, L., Song, D., & Blomberg, S. (2012, April). Quantum-like
non-separability of concept combinations, emergent associates and abduction. Logic Journal of
the IGPL, 20(2), 445–457. ISSN 1367-0751. https://doi.org/10.1093/jigpal/jzq049
Bruza, P., Busemeyer, J. R., & Gabora, L. (2009, October). Introduction to the special issue
on quantum cognition. Journal of Mathematical Psychology, 53(5), 303–305. ISSN 0022-
2496. https://doi.org/10.1016/j.jmp.2009.06.002. http://www.sciencedirect.com/science/article/
pii/S0022249609000637
Bruza, P. D., Kitto, K., Ramm, B. J., & Sitbon, L. (2015, August). A probabilistic framework for
analysing the compositionality of conceptual combinations. Journal of Mathematical Psychol-
ogy, 67, 26–38. ISSN 00222496. https://doi.org/10.1016/j.jmp.2015.06.002. https://linkinghub.
elsevier.com/retrieve/pii/S002224961500036X
Busemeyer, J. R., Pothos, E. M., Franco, R., & Trueblood, J. S. (2011). A quantum theo-
retical explanation for probability judgment errors. Psychological Review, 118(2), 193–218.
ISSN 1939-1471, 0033-295X. https://doi.org/10.1037/a0022542. http://doi.apa.org/getdoi.cfm?
doi=10.1037/a0022542
Cervantes, V. H., & Dzhafarov, E. N. (2019, August). True contextuality in a psychophysical exper-
iment. Journal of Mathematical Psychology, 91, 119–127. ISSN 00222496. https://doi.org/10.
1016/j.jmp.2019.04.006. https://linkinghub.elsevier.com/retrieve/pii/S002224961830186X
Coecke, B., & Lewis, M. (2015, September). A compositional explanation of the pet fish phe-
nomenon. arXiv:1509.06594 [cs, math]. http://arxiv.org/abs/1509.06594
Coecke, B., Sadrzadeh, M., & Clark, S. (2010, March). Mathematical foundations for a compo-
sitional distributional model of meaning. arXiv:1003.4394 [cs, math]. http://arxiv.org/abs/1003.
4394
Dostál, M., Sadrzadeh, M., & Wijnholds, G. (2021). Fuzzy generalised quantifiers for natural lan-
guage in categorical compositional distributional semantics. In M. Mojtahedi, S. Rahman, & M. S.
Zarepour (Eds.), Mathematics, logic, and their philosophies: Essays in honour of Mohammad
Ardeshir. Logic, epistemology, and the unity of science (pp. 135–160). Springer International
Publishing. ISBN 978-3-030-53654-1. https://doi.org/10.1007/978-3-030-53654-1_6
Dzhafarov, E. N. (2019, November). Contextuality and probability in quantum mechanics and
beyond: A preface. Philosophical Transactions of the Royal Society A: Mathematical, Physical
and Engineering Sciences, 377(2157), 20190371. ISSN 1364-503X, 1471-2962. https://doi.org/
10.1098/rsta.2019.0371. https://royalsocietypublishing.org/doi/10.1098/rsta.2019.0371
Dzhafarov, E. N., & Kujala, J. V. (2014). On selective influences, marginal selectivity, and
bell/CHSH inequalities. Topics in Cognitive Science, 6(1), 121–128. ISSN 1756-8765. https://
doi.org/10.1111/tops.12060. https://onlinelibrary.wiley.com/doi/abs/10.1111/tops.12060
Dzhafarov, E. N., & Kujala, J. V. (2016). Context–content systems of random variables: The
contextuality-by-default theory. Journal of Mathematical Psychology, 23.
Fodor, J. A. (1975). The language of thought. Harvard University Press.
Fodor, J. A., & Pylyshyn, Z. W. (1988, March). Connectionism and cognitive architecture:
A critical analysis. Cognition, 28(1), 3–71. ISSN 0010-0277. https://doi.org/10.1016/0010-
0277(88)90031-5. https://www.sciencedirect.com/science/article/pii/0010027788900315
Franco, R. (2009). The conjunction fallacy and interference effects. Journal of Mathematical Psy-
chology, 53(5), 415–422. ISSN 1096-0880 (Electronic), 0022-2496 (Print). https://doi.org/10.
1016/j.jmp.2009.02.002
Fuss, I. G., & Navarro, D. J. (2013, September). Open parallel cooperative and competitive decision
processes: A potential provenance for quantum probability decision models. Topics in Cognitive
Science, n/a–n/a. ISSN 17568757. https://doi.org/10.1111/tops.12045
Quantum Computing and Cognitive Simulation 103

Gayler, R. W. (2003, July). Vector symbolic architectures answer Jackendoff’s challenges for cog-
nitive neuroscience. In Joint International Conference on Cognitive Science (p. 6). https://arxiv.
org/abs/cs/0412059
Hameroff, S. (2014, January). Quantum walks in brain microtubules-a biomolecular basis for quan-
tum cognition? Topics in Cognitive Science, 6(1), 91–97. ISSN 17568757. https://doi.org/10.
1111/tops.12068
Hampton, J. A. (1987, January). Inheritance of attributes in natural concept conjunctions. Memory
& Cognition, 15(1), 55–71. ISSN 1532-5946. https://doi.org/10.3758/BF03197712
Hampton, J. A. (1988a, November). Disjunction of natural concepts. Memory & Cognition, 16(6),
579–591. ISSN 1532-5946. https://doi.org/10.3758/BF03197059
Hampton, J. A. (1988b). Overextension of conjunctive concepts: Evidence for a unitary model of
concept typicality and class inclusion. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 14(1), 12–32. ISSN 1939-1285(Electronic),0278-7393(Print). https://doi.org/10.
1037/0278-7393.14.1.12
Hampton, J. A. (1997, November). Conceptual combination: Conjunction and negation of natural
concepts. Memory & Cognition, 25(6), 888–909. ISSN 0090-502X, 1532-5946. https://doi.org/
10.3758/BF03211333. http://link.springer.com/10.3758/BF03211333
Hedges, J., & Sadrzadeh, M. (2019, June). A generalised quantifier theory of natural lan-
guage in categorical compositional distributional semantics with bialgebras. Mathemat-
ical Structures in Computer Science, 29(6), 783–809. ISSN 0960-1295, 1469-8072.
https://doi.org/10.1017/S0960129518000324. https://www.cambridge.org/core/journals/
mathematical-structures-in-computer-science/article/abs/generalised-quantifier-theory-of-
natural-language-in-categorical-compositional-distributional-semantics-with-bialgebras/
9738AE3A40B9A51AE2134D13E4FD3C5B
Hummel, J. E. (2011, June). Getting symbols out of a neural architecture. Connection Science,
23(2), 109–118. ISSN 0954-0091. https://doi.org/10.1080/09540091.2011.569880
Kartsaklis, D., & Sadrzadeh, M. (2016, December). Distributional inclusion hypothesis for tensor-
based composition. In Proceedings of COLING 2016, the 26th International Conference on
Computational Linguistics: Technical Papers, Osaka, Japan (pp. 2849–2860). The COLING
2016 Organizing Committee. https://www.aclweb.org/anthology/C16-1268
Khrennikov, A., Basieva, I., Dzhafarov, E. N., & Busemeyer, J. R. (2014, October) Quantum models
for psychological measurements: An unsolved problem. PLoS ONE, 9(10), e110909. ISSN 1932-
6203. https://doi.org/10.1371/journal.pone.0110909
Khrennikov, A., Basieva, I., Pothos, E. M., & Yamato, I. (2018, November). Quantum probabil-
ity in decision making from quantum information representation of neuronal states. Scientific
Reports, 8(1), 16225. ISSN 2045-2322. https://doi.org/10.1038/s41598-018-34531-3. https://
www.nature.com/articles/s41598-018-34531-3
Khrennikov, A. Yu., & Haven, E. (2009, October). Quantum mechanics and violations of the sure-
thing principle: The use of probability interference and other concepts. Journal of Mathematical
Psychology, 53(5), 378–388. ISSN 0022-2496. https://doi.org/10.1016/j.jmp.2009.01.007. http://
www.sciencedirect.com/science/article/pii/S0022249609000182
Lewis, M. (2019, September). Compositional hyponymy with positive operators. In Proceedings
of the International Conference on Recent Advances in Natural Language Processing (RANLP
2019), Varna, Bulgaria (pp. 638–647). INCOMA Ltd. https://doi.org/10.26615/978-954-452-
056-4_075. https://www.aclweb.org/anthology/R19-1075
Lewis, M. (2020, September). Towards logical negation for compositional distributional semantics.
IfCoLog Journal of Applied Logics, 7(5), 771–794.
Lorenz, R., Pearson, A., Meichanetzidis, K., Kartsaklis, D., & Coecke, B. (2021, February). QNLP in
practice: Running compositional models of meaning on a quantum computer. arXiv:2102.12846
[quant-ph]. http://arxiv.org/abs/2102.12846
Martin, A. E., & Doumas, L. A. A. (2020, February). Tensors and compositionality in
neural systems. Philosophical Transactions of the Royal Society B: Biological Sciences,
104 M. Lewis

375(1791), 20190306. https://doi.org/10.1098/rstb.2019.0306. https://royalsocietypublishing.

org/doi/10.1098/rstb.2019.0306
McCoy, R. T., Linzen, T., Dunbar, E., & Smolensky, P. (2019, May). RNNs implicitly implement
tensor product representations. In ICLR 2019—International Conference on Learning Represen-
tations. https://hal.archives-ouvertes.fr/hal-02274498
Meyer, F., & Lewis, M. (2020, November). Modelling lexical ambiguity with density matrices. In
Proceedings of the 24th Conference on Computational Natural Language Learning (pp. 276–290)
(Online). Association for Computational Linguistics. https://doi.org/10.18653/v1/2020.conll-1.
21. https://www.aclweb.org/anthology/2020.conll-1.21
Moore, D. W. (2002). Measuring new types of question-order effects. Public Opinion Quarterly,
66(1), 80–91. ISSN 0033362X, 15375331. https://doi.org/10.1086/338631. https://academic.
oup.com/poq/article-lookup/doi/10.1086/338631
Morier, D., & Bordiga, E. (1984). The conjunction fallacy: A task specific phenomenon? Personal-
ity and Social Psychology Bulletin, 10(2). https://doi.org/10.1177/0146167284102010. https://
journals.sagepub.com/doi/abs/10.1177/0146167284102010
Nielsen, M. A., & Chuang, I. L. (2010). Quantum computation and quantum information: 10th
Anniversary. Cambridge University Press.
Nosofsky, R. M. (1986). Attention, similarity, and the identification–categorization relationship.
Journal of Experimental Psychology: General, 115(1), 39–57. ISSN 1939-2222 (Electronic),
0096-3445 (Print). https://doi.org/10.1037/0096-3445.115.1.39
Penrose, R. (1990, December). The nonalgorithmic mind. Behavioral and Brain Sciences, 13(4),
692–705. ISSN 0140-525X, 1469-1825. https://doi.org/10.1017/S0140525X0008105X. https://
www.cambridge.org/core/product/identifier/S0140525X0008105X/type/journal_article
Piedeleu, R., Kartsaklis, D., Coecke, B., & Sadrzadeh, M. (2015). Open system categorical quantum
semantics in natural language processing. In CALCO (p. 14).
Plate, T. A. (1995, May). Holographic reduced representations. IEEE Transactions on Neural
Networks, 6(3), 623–641. ISSN 10459227. https://doi.org/10.1109/72.377968. http://ieeexplore.
ieee.org/document/377968/
Pothos, E. M., & Busemeyer, J. R. (2009, June). A quantum probability explanation for violations
of ‘rational’ decision theory. Proceedings. Biological Sciences, 276(1665), 2171–2178. ISSN
0962-8452. https://doi.org/10.1098/rspb.2009.0121
Pothos, E. M., & Busemeyer, J. R. (2013, June). Can quantum probability provide a new direction for
cognitive modeling? Behavioral and Brain Sciences, 36(3), 255–274. ISSN 0140-525X, 1469-
1825. https://doi.org/10.1017/S0140525X12001525. https://www.cambridge.org/core/product/
identifier/S0140525X12001525/type/journal_article
Pothos, E. M., & Trueblood, J. S. (2015, February). Structured representations in a quantum
probability model of similarity. Journal of Mathematical Psychology, 64–65, 35–43. ISSN
00222496. https://doi.org/10.1016/j.jmp.2014.12.001. https://linkinghub.elsevier.com/retrieve/
pii/S0022249614000832
Pothos, E. M., Busemeyer, J. R., & Trueblood, J. S. (2013). A quantum geometric model of similarity.
Psychological Review, 120(3), 679–696. ISSN 1939-1471, 0033-295X. https://doi.org/10.1037/
a0033142. http://doi.apa.org/getdoi.cfm?doi=10.1037/a0033142
Rosch, E. (1975). Cognitive representations of semantic categories. Journal of Experimental Psy-
chology: General, 104(3), 192–233. ISSN 1939-2222 (Electronic), 0096-3445 (Print). https://
doi.org/10.1037/0096-3445.104.3.192
Sadrzadeh, M., Kartsaklis, D., & Balkır, E. (2018, April). Sentence entailment in compositional
distributional semantics. Annals of Mathematics and Artificial Intelligence, 82(4), 189–218. ISSN
1573-7470. https://doi.org/10.1007/s10472-017-9570-x
Savage, L. J. (1954). The foundations of statistics. Wiley Publications in Statistics.
Smith, E. E., & Osherson, D. N. (1984, October). Conceptual combination with prototype
concepts. Cognitive Science, 8(4), 337–361. ISSN 0364-0213. https://doi.org/10.1016/S0364-
0213(84)80006-3. http://www.sciencedirect.com/science/article/pii/S0364021384800063
Quantum Computing and Cognitive Simulation 105

Smolensky, P. (1990, November). Tensor product variable binding and the representation
of symbolic structures in connectionist systems. Artificial Intelligence, 46(1–2), 159–216.
ISSN 00043702. https://doi.org/10.1016/0004-3702(90)90007-M. https://linkinghub.elsevier.
com/retrieve/pii/000437029090007M
Smolensky, P., & Legendre, G. (2006, July). The harmonic mind. The MIT Press. https://mitpress.
mit.edu/books/harmonic-mind-2-vol-set
Sozzo, S. (2014, June). A quantum probability explanation in Fock space for borderline contradic-
tions. arXiv:1311.6050 [physics, physics:quant-ph]. http://arxiv.org/abs/1311.6050
Sozzo, S. (2015, June). Conjunction and negation of natural concepts: A quantum-theoretic model-
ing. Journal of Mathematical Psychology, 66, 83–102. ISSN 0022-2496. https://doi.org/10.1016/
j.jmp.2015.01.005. https://www.sciencedirect.com/science/article/pii/S0022249615000073
Tversky, A. (1977). Features of similarity. Psychological Review, 84(4), 327–352. ISSN 1939-1471
(Electronic), 0033-295X (Print). https://doi.org/10.1037/0033-295X.84.4.327
Tversky, A., & Kahneman, D. (1983). Extensional versus intuitive reasoning: The conjunction
fallacy in probability judgment. Psychological Review, 90(4), 293–315. ISSN 1939-1471 (Elec-
tronic), 0033-295X (Print). https://doi.org/10.1037/0033-295X.90.4.293
Tversky, A., & Shafir, E. (1992, September). The disjunction effect in choice under uncertainty.
Psychological Science, 3(5), 305–310. ISSN 0956-7976. https://doi.org/10.1111/j.1467-9280.
1992.tb00678.x
Wang, Z., & Busemeyer, J. R. (2013, September). A quantum question order model supported by
empirical tests of an a priori and precise prediction. Topics in Cognitive Science, n/a–n/a. ISSN
17568757. https://doi.org/10.1111/tops.12040
Wang, Z., Busemeyer, J. R., Atmanspacher, H., & Pothos, E. M. (2013, September). The potential
of using quantum theory to build models of cognition. Topics in Cognitive Science, n/a–n/a. ISSN
17568757. https://doi.org/10.1111/tops.12043
Widdows, D., & Peters, S. (2003). Word vectors and quantum logic experiments with negation and
disjunction (p. 14). Stanford University.
Wiebe, N., Bocharov, A., Smolensky, P., Troyer, M., & Svore, K. M. (2019, February). Quantum
language processing. arXiv:1902.05162 [quant-ph]. http://arxiv.org/abs/1902.05162
Wittgenstein, L. (1953). Philosophical investigations. Philosophische Untersuchungen. Macmillan.
The Philosophy of Quantum Computing

Michael E. Cuffaro

Abstract From the philosopher’s perspective, the interest in quantum computation

stems primarily from the way that it combines fundamental concepts from two distinct
sciences: Physics, in particular Quantum Mechanics, and Computer Science, each
long a subject of philosophical speculation and analysis in its own right. Quantum
computing combines both of these more traditional areas of inquiry into one wholly
new, if not quite independent, science. Over the course of this chapter we will be
discussing some of the most important philosophical questions that arise from this
merger and philosophical lessons to be learned.

Keywords Quantum computing · Quantum humanities · Philosophy of science ·

Many-worlds interpretation · Church-Turing thesis

My work on this chapter benefited significantly from my interactions with students and other
audience members during and after a series of lectures I gave at the University of Urbino’s twenty-
third international summer school in philosophy of physics, held online in June 2020, in the midst
of the first wave of the COVID-19 pandemic, as well as a further lecture I gave for Michel Janssen’s
“The Age of Entanglement” honors seminar at the University of Minnesota in December 2020, as
the second wave of the pandemic began in earnest. Thanks to Ari Duwell, Eduardo Reck Miranda,
Philip Papayannopoulos, and Lev Vaidman for comments on a previous draft of this chapter. I
am also grateful for informal discussions, over the years, with Guido Bacciagaluppi, Jim Baggot,
Michel Janssen, Christoph Lehner, Lev Vaidman, and David Wallace; my presentation of the Everett
interpretation in Sect. 3, in particular, is significantly informed by what I take myself to have learned
from these discussions, though I hold only myself responsible for any mistakes or misunderstandings
in my presentation of the Everettian view. Section 2, on “Fundamental concepts” is heavily informed
by my recent work on related topics with Stephan Hartmann, Michael Janas, Michel Janssen,
and Markus Müller, as well as by informal correspondence with Jeffrey Bub and (the late) Bill
Demopoulos; though here again, I take sole responsibility for any mistakes. Finally, I gratefully
acknowledge the generous financial support of the Alexander von Humboldt Stiftung.

M. E. Cuffaro (B)
Munich Center for Mathematical Philosophy, Ludwig-Maximilians-Universität München,
Munich, Germany
e-mail: [email protected]

© Springer Nature Switzerland AG 2022 107

E. R. Miranda (ed.), Quantum Computing in the Arts and Humanities,
https://doi.org/10.1007/978-3-030-95538-0_3
108 M. E. Cuffaro

1 Introduction

From the philosopher’s perspective, the interest in quantum computation stems pri-
marily from the way that it combines fundamental concepts from two distinct sci-
ences: physics (especially quantum mechanics) and computer science, each long a
subject of philosophical speculation and analysis in its own right. Quantum comput-
ing combines both of these more traditional areas of inquiry into one wholly new
(if not quite independent) science. There are philosophical questions that arise from
this merger, and philosophical lessons to be learned. Over the course of this chapter,
we will be discussing what I take to be some of the most important.1
We begin, in Sect. 2, by introducing the fundamental concepts from physics and
computation that will be essential for framing the further philosophical discussion
that will follow. Section 2.1 and Sect. 2.2 introduce concepts from classical mechanics
and the classical theory of computation, respectively. In Sect. 2.1, we discuss the
concept of the state of a physical system as it is given in classical mechanics. We
emphasize in particular the way that we are invited, in classical mechanics, to think of
the state of a physical system as a compact description of the properties possessed by
it. These properties determine, in advance, the answers to the experimental questions
that we can pose of a system. And considering these questions and their answers leads
one to an alternative representation of a system’s state that is useful for representing
computation, a subject we then take up in more detail in Sect. 2.2.
In Sect. 2.2, we introduce the concept of a model of computation (or computational
architecture), the concept of the cost of carrying out a computation under a given
computational model, and explain what it means to solve a problem just as easily
under one computational model as under another. After discussing some of the more
important computational complexity classes, we then introduce two theses that can
be used to ground the model-independence of computational cost: the so-called
universality of Turing efficiency thesis (sometimes also referred to as the “strong”,
or “physical”, or “extended” Church-Turing thesis), and the invariance thesis. These
theses are both called into question by the existence of quantum computers. And
since some have taken the absolute model-independence guaranteed by them to
be foundational for the science of computational complexity theory, the question
arises of what to make of those foundations in light of the existence of quantum
computation. We will discuss this question, in an abstract way, in Sect. 2.3, where
I will argue that although the universality thesis must indeed be given up, this is
not a great loss. The invariance thesis, by contrast, remains, but in a different form,
transformed from a metaphysical to a methodological principle.

1Space does not permit me to exhaustively survey all of the philosophical issues brought up by
quantum computing. The interested reader can find a summary of other important issues in Hagar
and Cuffaro (2019).
The Philosophy of Quantum Computing 109

In Sect. 2.4, we turn to the physics behind quantum computation and begin by
introducing some of the more important concepts of quantum theory. Most impor-
tantly, we introduce the concept of a quantum state. We emphasize that the way that
a quantum state determines what the answers to experimental questions will be is
fundamentally different than the way that they are determined by a classical state.
We then turn to quantum computation proper in Sect. 3, where we review the basic
concepts of quantum computing and consider what we can say regarding the physical
explanation of the power of quantum computers. The many-worlds explanation of
quantum computing—the idea that quantum computers outperform classical com-
puters by running their computations in exponentially many physical universes—is
then introduced. We note two major problems that arise for this interpretation. The
first arises from the so-called preferred basis problem. This problem is a challenge
(that is arguably surmountable) for the many-worlds view in the more general context
of quantum mechanics. But we will see that it is especially problematic in the context
of quantum computers. The second major problem arises from the fact that there are
many different models of quantum computation, but the many-worlds explanation
of quantum computing only seems motivated by one of them.
In Sect. 4, we consider the role of quantum entanglement in enabling quantum
computers to outperform classical computers. We summarize an unsuccessful argu-
ment to the conclusion that quantum entanglement is insufficient to enable this quan-
tum “speedup” in Sect. 5, noting in Sect. 6 that reflecting on what it means to provide
a classical computer simulation of a quantum phenomenon should convince us to
reach the opposite conclusion. We continue the discussion of classically simulating
quantum computers in Sect. 7, and reflect on general aspects of the computational
perspective that the study of quantum computing provides on physical theory. We
note that reflecting on quantum computing emphasizes that there are important dif-
ferences in the methodological presuppositions that lie at the basis of physics and
computer science, respectively, and that conflating these can lead to confusion. We
also note the emphasis that studying quantum computation places on the fact that
quantum mechanics and classical mechanics are each alternative universal languages
for describing physical systems, and that the difference between quantum mechan-
ics and classical mechanics lies, fundamentally, in the differences in the expressive
power of these languages.2 We reflect on this in Sect. 8.

2By “quantum mechanics” I mean the fundamental theoretical framework shared in common by
every specific quantum-mechanical theory (quantum field theories, for instance) of a particular class
of systems; see Aaronson (2013b, pp. 110–111), Janas et al. (2022, Chap. 1 and §6.3), Nielsen and
Chuang (2000, p. 2), and Wallace (2019).
110 M. E. Cuffaro

2 Fundamental Concepts

2.1 Classical States

Classical mechanics (in its various forms3 ) was, prior to quantum mechanics, our
fundamental theoretical framework for describing the dynamics of physical systems
(i.e., how physical systems change over time). What is meant by a physical system
here is just one of the concrete objects that a particular physical theory describes. All
such systems are described as evolving through time in accordance with the dynam-
ical constraints that a theoretical framework applies universally to every physical
system. In classical mechanics (and the same is true, as we will see later, in quantum
mechanics), a physical system can be anything from a single particle to (in princi-
ple) the entire physical universe. Mathematically, though, what the dynamical laws
of a theory actually govern are the relations between the possible state descriptions
of systems. By the state description (or state specification) of a system is meant a
description of the particular physical situation that it happens to be in at a given
moment in time. For instance, at a given moment in time, a classical-mechanical sys-
tem will have a particular kinetic energy, it will be accelerating (or not) in a particular
direction, and so on.
It turns out that we do not have to explicitly specify each and every one of a sys-
tem’s dynamical properties to exactly and exhaustively specify the system’s dynam-
ics. In classical mechanics, given the dynamical laws, it is enough to specify values
for a system’s position and momentum. The value of any other dynamical prop-
erty can then be calculated by relating these with each other and with the values of
non-dynamical properties of the system such as, for instance, its charge or its mass.
Figure 1 illustrates the two different strategies described in the previous paragraph
for specifying the state of a physical system, for instance, a system composed of a
particle attached to the end of a spring constrained to move in one direction (see
Hughes, 1989, p. 73). On the left, we explicitly specify values (v1 , v2 , . . . ) for each
and every one of the physical system’s dynamical parameters. On the right, the values
of momentum, p, and position, q, are specified, and all other dynamical quantities are
calculated on their basis. In particular, the total energy, H , of the system is defined as
the sum of the kinetic energy, T , of the particle, and the potential energy, V , stored in
the spring, which are in turn defined in terms of p and q, respectively. Note that m, the
system’s mass, is a non-dynamical parameter that does not change over the history
of the system, and k is the force (the first derivative of momentum) per unit distance
required for the spring to be displaced from its equilibrium point (i.e., the point at
which it is neither stretched nor compressed). Other forces (e.g., gravitational forces)
are neglected but in principle can be included (for further discussion, see Hughes,
1989, §3.2.5).

3 These include Newtonian, Lagrangian, Hamiltonian, relativistic, and classical statistical mechan-
ics. For a recent comparison and philosophical discussion of Lagrangian and Hamiltonian mechan-
ics, see Curiel (2014).
The Philosophy of Quantum Computing 111

Fig. 1 Two different strategies for specifying the state of a physical system, for instance, a system
composed of a particle attached to the end of a spring constrained to move in one direction (see
Hughes, 1989, p. 73). On the left, we explicitly specify values (v1 , v2 , . . . ) for each and every one of
the physical system’s dynamical properties. On the right, the values of momentum, p, and position,
q, are specified, and all other dynamical quantities are calculated on their basis

The upshot is that specifying the position and momentum of a system provides
us with enough information, in classical mechanics, to completely characterize all
of the other dynamical properties of that system. Accordingly, the dynamical state,
ωt , of a system at some particular time t is given in classical mechanics by

ωt = (qt , pt ), (2.1)

where qt and pt are vectors (in three dimensions) representing the system’s position
and momentum, respectively, at t. Further, we can infer from the state of a system at
t, and the classical-mechanical laws of motion, exactly what the state of the system
will be in the next moment and at every other time both in the system’s past and in
its future (Hughes, 1989, §2.6).
Classical-mechanical states have another feature. Imagine all of the possible
(experimental) yes-or-no questions one might want to ask about the dynamical
properties of a particular system at a particular time, questions like: Is the value
of the dynamical property A within the range ? Completely specifying a sys-
tem’s dynamical state, i.e., specifying its momentum, p, and its position, q, yields a
simultaneous answer to all such questions irrespective of whether any question has
actually been asked. Indeed, this is what actually justifies our having considered the
values of p, q, and the other quantities derived from them to be (observable) prop-
erties of the system in the first place, properties possessed by the system whether or
not, and however, we enquire concerning them (Janas et al., 2022, §6.3). The same
goes for a system composed of many parts; for, after all, any observable property
A1 of a subsystem S1 of some larger system S is also an observable property of S.
Thus, given the state specification for a particular system we can construct a sort of
“truth table” which could be used, in principle, to answer any conceivable yes-or-no
question concerning the system and its subsystems (see Fig. 2).
In Fig. 2, I wrote Y and N (for “yes” and “no”), but I could have equally well used T
(for true) and F (for false), or alternately the binary digits 0 and 1. Using binary digits,
i.e., “bits”, is convenient because they are useful for representing numbers, which can
112 M. E. Cuffaro

Fig. 2 A “truth table” for a classical system composed of two subsystems whose states, respectively,
are (p1 , q1 ) and (p2 , q2 ). Various combinations of values for the state parameters pi and qi are given
on the left of the double-vertical line, with the superscript j in v j indicating the jth combination.
Relative to a given combination of values, the answers to experimental questions concerning the
values of derived observable quantities A, B, . . . on the right of the double-vertical line can be
determined

Fig. 3 A physical system consisting of four light switches. For each switch, we can ask: “Is switch
S turned on?” Using 0 to represent yes and 1 to represent no, the state of the overall system relative
to this question can be represented using the bit-string 0110 (which, in base-ten, is the number 6).
This is just one of the 24 = 16 possible states that a system like this one can be in. More generally,
for a system made up of n two-dimensional subsystems, the number of possible states for the system
as a whole is 2n

be manipulated abstractly using logico-mathematical operations.4 In particular, if we

build physical systems whose states can reliably be used to represent binary numbers
(in the sense that yes-or-no questions concerning their observable properties A, B,
. . . can be laid out as in Fig. 2, replacing 0 for Y and 1 for N ), and reliably evolve
them in ways that mirror a small basic set of logico-mathematical operations, then we
can (by combining these operations) use physical systems to carry out computations
that can, in principle, be arbitrarily complex (see Figs. 3 and 4).5

4 I say “logico-mathematical” because logical operations on bits can be thought of as modulo-2

arithmetical operations (see Boole, 1847).
5 For more general accounts of how physical systems can be used to represent computations, see

Fletcher (2018), Horsman et al. (2018), Maroney and Timpson (2018).

The Philosophy of Quantum Computing 113

Fig. 4 Some of the logico-mathematical operations, or “logic gates”, that can be used to manipulate
bits. OR and NOT together constitute a universal set of gates, in the sense that any more complicated
logico-mathematical operation, such as the “if-then” gate at the right, can be constructed using
combinations of OR and NOT. There are other universal sets besides {OR, NOT}, for instance:
{AND, NOT}, {NOR}, and {NAND}, where NAND is the “not-and” gate and NOR is the “not-or”
gate

2.2 Classical Computers

When the word “computer” is used in popular discourse today, generally what is
meant is a physical system like the one depicted in Fig. 3—not, of course, literally
a collection of light switches, but a physical system composed of subsystems that,
like the switches, can reliably be interpreted as being either on or off, and that can be
organized into registers, random access memory, hard drives, and other peripheral
systems such as keyboards, “mice”, touchpads, and so on, that can be manipulated
to reliably realize various logical operations like the ones depicted in Fig. 4. The
laptop I am writing this chapter with is an example of such a device. We turn such
devices on when we need them, and turn them off or “put them to sleep” when we
do not. But the word “computer” did not always signify a kind of machine. Prior
to the 1930s, “computation” generally meant an activity carried out by a human
being. Even after Alan Turing’s introduction of what later came to be known as
the “Turing machine” in 1936—an abstract device that is itself modeled on human
computation—“computers” were generally understood to be human beings until well
into the 1960s.6
As for the Turing machine: this, as mentioned, is also not a concrete physical
device (although it could, in principle, be instantiated by one) but an abstract math-
ematical model. Turing’s primary reason for introducing it had been to address the
so-called Entscheidungsproblem (which is the German word for ‘decision problem’),
an important but unsettled question that had arisen in the foundations of number
theory. The Entscheidungsproblem concerned the existence of an effective proce-

6The role of human computers in the United States of America’s space program, for instance, has
been documented in Shetterly (2016).
114 M. E. Cuffaro

dure for deciding whether an arbitrarily given expression in first-order logic can be
proven from the axioms of first-order logic. The answer, as Turing and, indepen-
dently, Alonzo Church were able to demonstrate, is that there is no such procedure.
Turing’s own proof relied on a penetrating philosophical analysis of the notion of
effective (human) computation, and a corresponding argument that one could design
an automatic machine (now called a Turing machine; see Fig. 5) to carry out the
essential activities of a human computer.
Turing (1936–7, pp. 249–51) argued that, for a computer to carry out a computa-
tion, it is essential that she have access to a notebook from which she can read and
onto which she can write various symbols in the course of her work. These symbols
need to be distinguishable by her from one another, on the basis of which Turing
argued that the alphabet from which she chooses her symbols must be a finite alpha-
bet. At any given moment during a computation, a computer may find herself in one
of a finite number (again, because she must be able to distinguish them) of states of
mind relevant to the task at hand which summarize her memory of the actions she
has performed up until that point along with her awareness of what she must now
do (pp. 253–4). The actions that are available to her are characterized by a finite set
of elementary operations, such as “read the next symbol” from the notebook, “write
symbol a” to the notebook, and so on. In an automatic machine, a control unit is
constructed to encode the machine’s “state of mind” (i.e., a logical representation
of its current state, its current input, and its transition function), which in general
changes after every operation of the read-write head. The latter moves back and
forth along a one-dimensional tape (the machine’s “notebook”), from which it reads
and onto which it writes various symbols from a finite alphabet, in conformity with
a particular finite set of exactly specifiable rules. Turing was able to show that no
automatic machine of this kind can be used to solve the Entscheidungsproblem.
Through the work of Turing, Church, and others, the modern science of computa-
tion, and in particular computability theory—the science of which problems can and
which cannot be solved by a computer, i.e., by anything or any person that can carry
out (but is restricted to) the essential activities associated with human computation—
was born.7
Irrespective of whether it is a machine or a human being that is doing the comput-
ing, the question of how much it actually costs to compute a solution to a computable
problem is also a very important one. In particular, we generally would like to know
how much time and how much space are required (though time is usually regarded
as the more important measure). The question of the cost of a given computation had
become especially important as more and more problems came to be carried out by
machines in the 1950s and 1960s. Even the earliest digital computers of the 1930s
and 1940s performed their computations far faster than human beings could possibly
perform them, and with improvements in design and in technology, machines could
be made to run ever faster. Through the need to use these machines’ resources effi-

7 For more on the Entscheidungsproblem and the early history of computer science, see Copeland
(2020), Davis (2000), Dawson Jr. (2007), Lupacchini (2018).
The Philosophy of Quantum Computing 115

Fig. 5 A version of what is now called a ‘Turing machine’, with a single bi-directional tape, a
control unit encoding a particular set of instructions or program, and a read-write head. Such a
machine is an idealized representation of the components involved in human computation. In a
universal Turing machine, the computer’s program is read from the tape at the beginning of the
computation. For a good introduction to the Turing machine and other related topics, see Martin
(1997)

ciently, a sub-field of computer science began to develop whose concern was with
how efficiently a given problem can be solved and on which types of machine.
In the context of the Turing machine model, time and space are quantified in
terms of the number of computational ‘steps’ (i.e., the number of basic operations)
and the number of squares of tape, respectively, that are needed to carry out a given
computation. Although it is in principle possible (assuming we have enough tape)
to instantiate a Turing machine physically,8 generally it makes more sense, from
a practical perspective, to design a mechanical computer in accordance with some
other model of computation. Particularly important is the von Neumann architecture,
a type of stored-program computer,9 that involves a central processing unit (CPU),
an instruction register and program counter, internal memory used to store data
(including the program being run), external memory used to store less frequently
needed data, and finally an input and an output mechanism. A von Neumann machine
can in principle compute all of the same problems as a universal Turing machine (see
Fig. 5), but a time step for a von Neumann machine is not concretely interpreted in the

8 A Turing machine’s tape does not need to be actually infinite in length. What is required is only that
the tape be indefinitely long, so that, for a given (finite) computation, the machine can be supplied
with enough tape to carry it out. To put it another way: What constitutes ‘enough tape’ to carry out
a computation is not part of the general definition of a Turing machine. It is, rather, assumed, for
the purposes of that definition, that enough tape to carry out a given computation can be supplied.
That said, certain variations on the Turing machine model restrict the ways in which tape can be
read by the control unit in various ways. For instance, some variants employ separate tape(s) for
the machine to write “rough work” on in addition to an output tape, some variants only allow the
read-write head to move in one direction along the tape, and so on.
9 Although the von Neumann architecture, or ‘von Neumann machine’ is only one of a number of

various types of stored-program computer, the terms have today (inappropriately, from a historical
point of view) come to be understood synonymously (Copeland, 2017).
116 M. E. Cuffaro

same way as it is for a Turing machine, just as ‘space’ for a von Neumann machine
is not measured in squares of tape.
There are many other universal models of computation besides von Neumann
machines and Turing machines. Nevertheless, at least when it comes to reasonable
classical computational models, it is possible to talk about the computational cost
of solving a particular problem (like, for instance, the problem of finding the prime
factors of a given integer), in a model-independent way; i.e., without having to
specify which machine model that cost is defined with respect to. There are two
things involved in explicating what it means for the computational cost of solving a
given problem to be model-independent in this sense. The first, mentioned just above
is the idea of a reasonable machine architecture.
Roughly, a reasonable (universal) machine architecture or model is one that it
would be possible to physically instantiate. Turing machines like the one described
in Fig. 5 and von Neumann machines are examples of reasonable machine models. A
computational model that employs an unbounded amount of parallel processing, by
contrast, is an example of an unreasonable model, or what van Emde Boas (1990, p. 5)
calls a model of the “second machine class”. There are legitimate theoretical reasons
for studying the complexity-theoretic properties of unreasonable models. But no
finite physical system could possibly instantiate an unreasonable architecture. There
are thus good reasons to want to distinguish the reasonable from the unreasonable
ones (see also, Dean, 2016).
The second thing we need to do, in order to explicate what it means for the compu-
tational cost of solving a given problem to be model-independent, is to make sense
of what it means to solve a given problem just as easily with one computational
architecture as with some other one. But before we can make sense of solving some-
thing just as easily, we need to first try and make sense of what it means to solve
something easily under one computational model. Consider, to begin with, a human
computer working alone. For a human computer, the difference between a problem
that could in principle require up to a thousand steps to solve, and one that could in
principle require up to a million steps, is very great indeed. With some justification,
this person might call the former problem easy to solve, and the latter problem hard.
Any criterion based on a fixed number of computational steps, however, would not
be very useful in the context of machines, at any rate certainly not any criterion based
on numbers this low. Even by the 1950s, a mechanical computer could carry out a
computation far faster than a human being. As of 2020, a typical laptop computer
can perform hundreds of thousands of Millions of Instructions per Second (MIPS)
(Wikipedia contributors, 2020), and computers are becoming more and more per-
formant every year. In the context of machine computation, it is far more useful to
ask how a given solution to a computational problem scales; i.e., how the number
of resources required to carry out a given solution to that problem grows with the
size of the input to the problem. Take, for instance, the problem to sort a given set of
numbers. All else equal, the more numbers there are to sort, the longer it will take.
How much longer? The fastest sorting algorithms (one of which is MergeSort)
will take on the order of n log n steps to sort n numbers in the worst case (Mehlhorn
& Sanders, 2008).
The Philosophy of Quantum Computing 117

In modern computational complexity theory, we say that a problem is easy to

solve, or alternately that it is tractable, if there exists an efficient algorithm to solve
it. If no efficient algorithm exists, then we say that the problem is hard, or intractable.
This definition, of course, merely trades in one informal notion for another, so let
us state more precisely what we mean by ‘efficient’. By an efficient algorithm for
solving a given problem, we mean an algorithm that will solve it in polynomial time;
i.e., in a number of time steps that are no more than a polynomial function of the
size of its input. Stated in mathematical terms, this means that for input size n, such
an algorithm will take on the order of n k steps, where k is some constant.10 As for
an algorithm that takes more than a polynomial number of steps, we consider it to
be inefficient, even if it might be the most efficient algorithm known for a particular
problem. An example of an inefficient algorithm is one that requires exponential
time; i.e., on the order of k n steps for input size n and some constant k.11 Even harder
problems take factorial time, i.e., on the order of n! steps, and so on.
The idea of using polynomial time as our criterion for distinguishing between
efficient and inefficient algorithms—what we will henceforth call the polynomial
principle—was introduced independently by Alan Cobham and Jack Edmonds in
1965.12 There are a number of good reasons for adopting the polynomial principle.
One reason is simply that the set of problems picked out by it has so far tended
to correspond with those that we would (so to speak) pre-theoretically regard as
efficiently solvable (Cuffaro, 2018b, §11.6).13 Perhaps the most useful feature of the
polynomial principle stems from the fact that polynomial functions compose (see
Fig. 6). That is, given an algorithm that runs in polynomial time, if we add to it
any number of calls to polynomial-time subroutines, the total running time of the
algorithm will still be polynomial (Arora & Barak, 2009, p. 27).14
Now that we have a handle on what it means for a problem to be solvable easily,
we can state what it means for a problem to be solvable just as easily on one type
of computational architecture as on another. We say that a problem is solvable just
as easily on machine model M1 as on machine model M2 if and only if there is an
algorithm for solving the problem on M1 that requires no more than a polynomial

10 “On the order of” is a technical term, usually symbolized in “big-oh notation” as O(T (n)). An
algorithm is O(T (n)) for some function T (n) if for every sufficiently large n, its actual running
time t (n) ≤ c · T (n) for some constant c. For instance, an algorithm that never takes more than 5n 3
steps is O(n 3 ).
11 A famous example of a problem for which only exponential-time solutions are known is the

Traveling Salesman Problem (Cook, 2012).

12 In some literature it is referred to as the Cobham-Edmonds thesis. Kurt Gödel anticipated the

principle, to some extent, in a private letter he wrote to John von Neumann in 1956. For further
discussion, see Cuffaro (2018b).
13 This correspondence is not perfect, but the usefulness of the polynomial principle is such that we

appeal to it despite this (Cuffaro, 2018b, §11.6).

14 It is easy to see this: Consider a program that consists of n k calls of a subroutine that takes n l

steps, where n is the number of bits used to represent the input, and k and l are finite constants. The
total number of steps needed to carry out this program will be n k+l . If k and l are finite constants
then so is k + l. In other words, n k+l is still a polynomial.
118 M. E. Cuffaro

Fig. 6 Inserting any number of calls to polynomial-time subroutines within the body of a
polynomial-time algorithm results in a polynomial-time algorithm

number of extra time steps as compared to M2 .15 Another way of putting this is that
anything that M2 can do is efficiently simulable by M1 .
Any problem that is solvable in polynomial time by a Turing machine is said to
belong to the class PTIME (for “polynomial time”), though it is usually referred
to simply as P. P is, of course, not the only complexity class. If a given problem is
such that, for a given solution to it, there is a Turing machine that will verify that the
solution is indeed a solution, then the problem is said to belong to the class NP, for
“nondeterministic polynomial time”. The reason for the name is that, equivalently,
this class of problems can be defined as those that can be solved in polynomial
time on a nondeterministic Turing machine. A nondeterministic Turing machine, or
choice machine, is such that at a given step of a given computation, an imagined
external operator of the machine can choose to have it transition in one way rather
than another. This is unlike a standard (deterministic) Turing machine for which
every computational step is completely determined in advance given a particular
input. Choice machines are interesting from a theoretical point of view, but given
that they involve an external operator they are not really automatic machines in the
sense that is relevant to our discussion here.
A variant of the choice machine that is automatic is a probabilistic Turing machine.
Such a machine’s choices are not made by an external operator but by the machine
itself, which we can imagine as equipped with the equivalent of a fair coin that it can
flip a number of times to decide whether to transition one way or another whenever
such a choice is open to it.16 The class of problems solvable in polynomial time
on a probabilistic Turing machine is called BPP (for “bounded error probabilistic
polynomial”). Note that what it means to be able to solve a given problem is not the
same for a probabilistic Turing machine as it is for a deterministic Turing machine.
In particular, a given “solution” output by a probabilistic Turing machine is allowed
to be wrong. We only demand that it be right with high enough probability so that,

15 Note that it makes sense to talk about solving a given problem just as easily on M1 as on M2
even when the problem under consideration is actually intractable for both. For instance, if some
problem requires 2n steps to solve on M1 and 2n + n 3 steps to solve on M2 then it is no harder,
from the point of view of the polynomial principle, to solve it on M2 than on M1 .
16 For more on probabilistic and nondeterministic Turing machines and how they compare to their

deterministic counterparts, see Cuffaro (2018b, §11.3).

The Philosophy of Quantum Computing 119

if we re-run the machine for the given input on the order of a polynomial number of
further times, our confidence in the majority answer will approach certainty.
There are very many more complexity classes besides these ones (see Aaronson,
2016). But we will stop here as we now have the basic ingredients with which to
state a couple of related theses that can be used to explicate what it means for the
cost of solving a given problem to be model-independent. As we will see a little
later, there are good reasons to prefer the second thesis, but for now, we simply
introduce both. First, the universality of Turing efficiency thesis (sometimes also
called the “strong”, “physical”, or “extended” Church-Turing thesis17 ) asserts that
any problem that can be efficiently solved on any reasonable machine model M (von
Neumann architecture, Harvard architecture, or whatever) can be efficiently solved
on a probabilistic Turing machine, or more formally

PolyM = BPP. (2.2)

In other words, the thesis implies that the set of problems solvable in polynomial
time does not grow beyond BPP if we allow ourselves to vary the underlying model.
Assuming the thesis is true, we do not need to worry about what model an algorithm is
implemented on when discussing the computational complexity of various problems;
we can simply focus on the abstract probabilistic Turing machine model and carry
out our analyses in relation to it.18
The second thesis we will introduce is called the invariance thesis (van Emde
Boas, 1990, p. 5),19 which asserts that given any two reasonable machine models
Mi and M j , Mi can efficiently simulate M j ; i.e., Mi can solve any problem just as
easily (in the sense explained above) as M j can. More formally

poly
∀i, j Mi ∼ M j . (2.3)

Note that the invariance thesis implies the universality thesis, but not vice versa.

17 See Timpson (2013, Chap. 6) for discussion of a different, more general thesis, that is only
indirectly relevant to computational complexity theory. For a discussion of how these theses relate,
see Cuffaro (2018b, §11.4).
18 We could have also expressed the thesis in terms of P rather than BPP. Although it was thought,

for many years, that there are more problems efficiently solvable on a probabilistic Turing machine
than on a standard Turing machine, a number of recent results have pointed in the opposite direction
(e.g., Agrawal et al., 2004), and it is now generally believed that classical probabilistic computation
does not offer any performance advantage over classical deterministic computation (Arora Barak,
2009, Chap. 20). In other words, it is now widely believed that P = BPP, or that it is just as easy
to solve a given problem on a deterministic Turing machine as it is on a probabilistic one. We have
nevertheless stated the universality thesis in terms of BPP because this will prove convenient when
it comes time to discuss the differences between classical and quantum computation. A quantum
computer is, from one point of view, just another kind of probabilistic computer (that calculates
probabilities differently), and it has the same success criterion as a classical probabilistic computer,
i.e., we only demand that a given “solution” be correct with “high enough” probability.
19 See also: Goldreich (2008, p. 33), who names it differently.
120 M. E. Cuffaro

2.3 Physical Perspectives on Computer Science

Neither the universality of Turing efficiency thesis, nor the invariance thesis, is a
mathematical theorem. These statements can be true or false, and for a long time
they were thought to be true, for none of the reasonable (classical) universal models
of computation that had been developed since the time of Turing were found to
be more efficient than the Turing model by more than a polynomial factor, and all
of them had been shown to be able to simulate one another efficiently (van Emde
Boas, 1990). Over the last three decades, however, evidence has been mounting
against universality and invariance, primarily as a result of the advent of quantum
computing (Aaronson, 2013b, Chaps. 10 and 15).
Quantum mechanics, as we will see in the next section, is an irreducibly probabilis-
tic theory. Thus, a quantum computer is a type of probabilistic machine. Analogously
to the way we defined BPP as the class of problems solvable (in the probabilistic
sense) in polynomial time on a probabilistic Turing machine, we can also define the
complexity class BQP (for “bounded error quantum polynomial”) as the class of
problems solvable (in the same sense) in polynomial time on a quantum computer.
It is easy for a quantum computer to simulate a probabilistic Turing machine. Thus

BPP ⊆ BQP, (2.4)

i.e., the class of problems efficiently solvable on a quantum computer is at least as

large as the class of problems efficiently solvable on a probabilistic Turing machine.
What is still unknown, though it would be surprising if it were not the case, is whether

BPP  BQP, (2.5)

i.e., whether the class of problems efficiently solvable on a quantum computer (a

physically realizable computational architecture) is strictly larger than the class of
problems efficiently solvable on a probabilistic Turing machine. Note that if Eq. (2.5)
is true, then both the universality and invariance theses are false.
There are some authors who view the consequences of the falsification of the
universality thesis, in particular, to be profound. Bernstein and Vazirani (1997), for
example, take it that computational complexity theory “rests upon” this thesis (p.
1411), and that the advent of quantum computers forces us to “re-examine the foun-
dations” (p. 1412) of the theory. The sense in which complexity theory rests upon
universality is expressed by Nielsen and Chuang (2000), who write that the falsity of
the thesis implies that complexity theory cannot achieve an “elegant, model indepen-
dent form” (p. 140). For Hagar (2007), the failure of model-independence shakes,
not only the foundations of complexity theory, but certain views in the philosophy of
mind that depend on the model-independence of computational kinds (pp. 244–245).
If we actually examine what the universality thesis is saying, however, then it is
not really clear, at least not prima facie, how it can ground the model-independence
of complexity-theoretic concepts. The statement of the thesis is that any efficiently
The Philosophy of Quantum Computing 121

solvable problem is solvable efficiently by a probabilistic Turing machine. A prob-

abilistic Turing machine is a particular model of computation, though. How can a
thesis whose very definition makes reference to a particular computational model
ground the model-independence of computational complexity theory? In fact, there
is a weak notion of model-independence being alluded to here. The point (see Nielsen
& Chuang, 2000, p. 140) is that, for any assertion of the form: “problem P is effi-
ciently solvable under computational model M”, the qualification “under computa-
tional model M” can always (given the truth of the universality thesis) be substituted
with “by a probabilistic Turing machine” without changing the truth value of that
sentence. Further, to show that such a sentence is true in general, it suffices to show
that it is true for a probabilistic Turing machine. Finally, because “by a probabilistic
Turing machine” qualifies every such sentence we can leave it off and still expect to
be understood. “Problem P is efficiently solvable by a probabilistic Turing machine”
is thus abbreviated to “problem P is efficiently solvable”. That we can do this is not
insignificant, but this is arguably not a particularly deep sense of model-independence
(see Cuffaro, 2018b, §11.6).
Far more interesting in relation to the question of model-independence is the
invariance thesis. Unlike the universality thesis, model-independence is built right
into the very statement of invariance. For after all, it amounts to the quite direct
claim that the details of any particular reasonable machine model, since these can be
efficiently simulated by any other reasonable model, are irrelevant for the purposes
of characterizing the complexity of a given problem.
Invariance, if taken to be true without qualification, clearly brings with it an
absolute notion of model-independence (at least with respect to physically reasonable
models). And if taken to be false (as it seems we should, given the existence of
quantum computation) it clearly precludes such a notion. Arguably, however, what
is (and always was) most valuable about the idea of invariance is not the absolute
model-independence that is implied when it is taken to hold without qualification.
As we will see later, the term “quantum computer” does not refer to some single
particular model of computation but is, rather, an umbrella term for a number of
distinct computational models,20 all of which have been shown to be computationally
equivalent in the sense that they are all efficiently simulable by one another.21 In
other words, what we have learned from the study of quantum computing is that, in
addition to the existence of an equivalence class of reasonable classical models that
satisfy the invariance thesis with respect to one another, there is a further class of
reasonable computational models, based on the principles of quantum mechanics,
that satisfy the invariance thesis with respect to one another. Thus, there are two
distinct equivalence classes of physically reasonable computational models from the
viewpoint of computational complexity theory. This is a discovery.

20 For instance, the quantum Turing model (Deutsch, 1985), the quantum circuit model (Deutsch,
1989), the cluster-state model (Briegel et al., 2009), the adiabatic model (Farhi et al., 2000), and so
on.
21 See, for instance, Aharonov et al. (2007), Nishimura and Ozawa (2009), Raussendorf and Briegel

(2002).
122 M. E. Cuffaro

Invariance, thought of as a guiding rule or methodological principle, rather than as

an absolute thesis, can be understood as grounding these investigations, and arguably
this was the point all along (Cuffaro, 2018b, §11.6). Through the search for equiv-
alence classes, we carve out the structure of the space of computational models,
yielding a notion of relative model-independence among the machine models com-
prising a particular equivalence class. And the fact that relative model-independence
exists within the space of computational models at all arguably tells us something
deep about how computer science connects up with the world, for the differences in
computational power between these two reasonable classes of computational model
are best understood by considering the physics needed to describe them. We dis-
cussed the physics of classical computers in Sect. 2.1. In the next section, we turn to
the physics of quantum computers.

2.4 Quantum States and Operations

We saw earlier that in classical mechanics, assigning a particular state, ω = (q, p), to
a system fixes the answer to every conceivable yes-or-no experimental question that
we can ask about it in advance, irrespective of whether we actually ask any questions
or not. And we saw how to arrange the possible questions we can ask about a system,
and their possible answers, into a truth-table-like structure like the one given in Fig.
2. Note that specifying the values of the answers to the questions on the right-hand
side of the double-vertical line in Fig. 2 is another way of representing the state of
a system like the one depicted in Fig. 3. We saw how one can manipulate such a
system in ways that can be represented abstractly as logico-mathematical operations
on binary digits, where the binary digits are abstract representations of the properties
of the system’s (two-level) subsystems.
The subject matter of quantum mechanics, just as it is for classical mechanics,
is physical systems, where these can be anything from a single particle to the entire
physical universe. In other words, quantum mechanics is not just a theory of the
small. It is, rather, a new universal language for describing physical systems,22 and it
describes our experience with those systems better and more accurately than classical
mechanics describes it. In quantum mechanics, just as in classical mechanics, a
system can be, at a given moment in time, in any one of a number of possible
physical states. These states are represented by state vectors (or “wave functions”).
The state vector for a two-dimensional quantum system or “qubit”, for instance, is
given in general by

|ψ = α|0 + β|1. (2.6)

When the complex numbers α and β are both non-zero, a qubit is said to be in
a superposition of the basis vectors |0 and |1. Just as an ordinary vector in the

22 See note 2 above.

The Philosophy of Quantum Computing 123

Fig. 7 An ordinary vector in the Cartesian plane. It can be decomposed into x and y coordinates,
as on the left, or into the alternative coordinates x  and y  as on the right. Similarly, a state vector,
in quantum mechanics, can be decomposed into more than one Hilbert space basis. Two common
bases (but not the only two) with which to decompose the state vector for a qubit are the {|0, |1}
basis (known as the computational basis) and the {|+, |−} basis

Cartesian plane can be decomposed, mathematically, into x and y coordinates, a state

vector for a two-dimensional quantum system can be decomposed, mathematically,
into two basis vectors for the (Hilbert) space that it is situated in (see Fig. 7). Two
common bases (but not the only two) with which to decompose the state vector for
a qubit are the {|0, |1} basis (known as the computational basis) and the {|+, |−}
basis.
Associated with a given basis is a particular experimental question; for instance,
“Is the qubit in the state |0 (as opposed to the state |1)?” in the case of the com-
putational basis, and in the case of the {|+, |−} basis, “Is the qubit in the state |+
(as opposed to the state |−)?” When a system in the state α|0 + β|1 is ‘asked’
whether it is in state |0 as opposed to the state |1—i.e., when it is measured in the
computational basis—-then with probability |α|2 the answer will come back as |0
and with probability |β|2 it will come back as |1; i.e.

  Pr = |α|2
M α|0 + β|1 −−−−→ |0 (2.7)
  Pr = |β|2
M α|0 + β|1 −−−−→ |1 (2.8)

where M is a measurement operator used to represent a computational basis mea-

surement. Note that there is also a basis corresponding to the question: “Is the qubit
in the state |ψ (as opposed to the state |φ)?” Why not just ask this question? One
reason is that measuring a quantum-mechanical system involves using a concrete
physical device, and it turns out that it is technologically more feasible to construct
measurement devices corresponding to some experimental questions than to others
(Janas et al., 2022, §6.5).
124 M. E. Cuffaro

In terms of their dynamics, states of quantum systems evolve linearly

 
U α|0 + β|1 = αU |0 + βU |1 (2.9)

and unitarily:

UU † |ψ = U † U |ψ = |ψ (2.10)

over time, where U is a unitary operator and U † is its adjoint.

Classical and quantum mechanics differ in the way that answers to experimental
questions are determined. In contrast to classical mechanics, specifying a quantum
system’s state at a given moment in time does not fix in advance the answers to the
experimental questions that can be asked of that system, or as Bub and Pitowsky
(2010) put it, the quantum state is not a truthmaker in relation to those questions.
It fails to be a truthmaker in two senses: First, a given state specification yields, in
general, only the probability that the answer to a given experimental question will
take on one or another value when asked. This in itself is not as much of a departure
from classical mechanics as one might think, however, because conditional upon the
selection of an experimental question, one can, in quantum mechanics, describe the
observed probabilities as stemming from a prior classical probability distribution
over the dynamical properties of the system that is determined in advance by the
quantum state.23
This brings us to the second, more important, sense in which the state of a quantum-
mechanical system fails to be a truthmaker in relation to questions that can be asked
about that system’s dynamical properties: The probability distributions associated
with the answers to individual experimental questions cannot be embedded into a
global prior probability distribution over all of the answers, as they can be for a
classically describable system (see Fig. 2). In quantum mechanics, one can only
say that conditional upon our inquiring about the observable A, there will be a
particular probability for the answer to that question to take on a particular value.
Thus, in quantum mechanics, unlike in classical mechanics, we actually have to ask
the system a question in order to get an answer from it.24
When quantum-mechanical systems are combined, they can sometimes become
entangled with one another. Consider, by way of contrast, the simple illustration in
Fig. 3 of a classically describable system composed of many subsystems. This system
is in a product state, which means that the state of the overall system can be expressed
as a product of the states of each individual subsystem, i.e., 0 A 1 B 1C 0 D , where the
individual states of the subsystems are 0 A , 1 B , 1C , and 0 D . In classical mechanics,
there is no other way to describe the combined state of a number of subsystems. Even

23This is analogous to the way we interpret probabilities in classical statistical mechanics.

24See also Janas et al. (2022, Chaps. 1 and 6), who call this (ironically) the “small” measure-
ment problem in contrast to the “big” (this label is also ironic) problem described in the previous
paragraph. These labels are originally due to Bub and Pitowsky (2010), who used them ironically
too.
The Philosophy of Quantum Computing 125

when these subsystems are correlated with one another (as the state of one side of a
coin is correlated with the state of the other side, for instance), as long as we include
facts about those correlations in our description of the overall system, then its overall
state can be factored into the individual states of its subsystems (Bell, 2004 [1981]).
This is true even if the subsystems happen to be far apart, for instance, if I print out
two copies of a document and mail each of them to distant parts of the globe.
In quantum mechanics, in contrast, a system composed of multiple subsystems
can sometimes be in a state like the following:

1  
√ |0 A |1 B − |1 A |0 B . (2.11)
2

This state, describing a system composed of two subsystems A and B, is not a product
of the individual states of its subsystems, for there is no way to factorize it into a state
of the form: |ψ A |φ B (the reader is invited to try). It is sometimes said that in an
entangled quantum system the whole is somehow prior to the parts (Howard, 1989),
or that the systems are nonlocally correlated (Maudlin, 2011), for if Alice measures
her qubit in the computational basis and receives the result |0 A , then a measurement
by Bob on his qubit (which in principle he may have taken to the opposite side of
the universe) is thereby instantaneously determined to be |1 B , although there is no
way for Bob to take advantage of this instantaneous determination for the purpose
of predicting the outcome of a measurement on his subsystem, short of waiting for
Alice to send him a classical signal (a text message or a phone call, for instance).25
Note that the correlation between the results of computational (i.e., {|0, |1}) basis
measurements on an entangled system in this state do not, in themselves, represent
as much of a departure as one might think from what one can describe using classical
language. For with respect to this specific correlation, Bell (2004 [1964]) showed that
it can be seen as arising from a classically describable system that has been prepared
in a particular way. It is only when we consider other experiments on such a system
that we see a departure from the possibilities inherent in classical description. In
particular, because

|0 + |1 |0 − |1

|+ = √ , |− = √ , (2.12)
2 2

Equation (2.11) can be rewritten, in the {|+, |−} basis, as

1  
√ |+ A |− B − |− A |+ B . (2.13)
2

Indeed, we obtain the same form for the state of the system regardless of which basis
we choose to express it in:

25This restriction on the use that can be made of nonlocal correlations in quantum mechanics is
called “no signaling”. For discussion, see Bub (2016), Cuffaro (2020).
126 M. E. Cuffaro

1  
√ |b1  A |b2  B − |b2  A |b1  B . (2.14)
2

Equations (2.11), (2.13), and (2.14) all yield the same probabilities for the results of
experiments.
If we collect a number of pairs of subsystems prepared in this entangled state, and
then ask of each pair: “Is this pair in the state |0|1 (as opposed to |1|0)?,” then
we will find that the answers “yes” and “no” will be obtained with equal probability.
Conditional upon this question, we can imagine these answers as arising from a prior
classical probability distribution over the properties of the subsystem pairs, half of
which we imagine to be in the state |0|1, and half of which we imagine to be in
the state |1|0. This prior probability distribution is incompatible, however, with
the prior classical probability distribution that we might imagine to be responsible
for the answers yielded from repeatedly asking the question: “Is this pair in the
state |+|− (as opposed to |−|+)?” In the context of the latter question, we can
imagine the answers as arising from a prior classical probability distribution over
the properties of the subsystem pairs, half of which we imagine to be in the state
|+|−, and half of which we imagine to be in the state |−|+. Asking a question
of an imagined ensemble of systems, half of which are in the state |0|1, and half of
which are in the state |1|0, will yield different statistics, however, than the ones that
would be yielded by asking the same question of an imagined ensemble of systems,
half of which are in the state |+|−, and half of which are in the state |−|+.
These probability distributions are, in this sense, incompatible. And yet all of these
statistics are predicted by one and the same quantum state, the one expressed in the
unitarily equivalent forms given in Eqs. (2.11–2.14). In quantum mechanics, unlike
in classical mechanics, the questions we ask or do not ask actually matter for the
ways that we can conceive of the underlying properties of a physical system, even
when the state of that system is fully specified (Janas et al., 2022, §6.5).

3 Quantum Computation and Parallel Universes

The are problems that are hard for a classical computer, but that a quantum computer
can solve easily (where “easy” and “hard” are meant in the computational sense
defined above in Sect. 2.2).26 The most famous example of such a problem is that of
factoring large integers into primes. The best-known classical algorithm for factoring
is the number field sieve (Lenstra et al., 1990), which takes on the order of 2(log N )
1/3

steps to factor a given integer N .27 No one has yet proven that this is the best that a

26 There are also problems for which a quantum computer, despite being unable to solve them easily,
can nevertheless solve them significantly more easily than a classical computer can. An example is
the problem to search an unstructured database, for which a quantum (“Grover’s”) algorithm can
reduce the number of steps required by a quadratic factor over any known classical algorithm. See:
Bennett et al. (1997), Grover (1996), and for further discussion see Cuffaro (2018b, p. 269).
27 For the meaning of ‘on the order of’ see fn. 10.
The Philosophy of Quantum Computing 127

classical computer can do. All the same, encryption algorithms such as RSA (named
for the paper by Rivest, Shamir, and Adleman in which it was introduced), that are
today widely used for secure transactions on the internet, rely on the assumption that
it is; i.e., that one cannot factor in polynomial time. In 1994, Peter Shor discovered
a quantum algorithm—now known as Shor’s algorithm—that can factor integers in
on the order of log N steps for a given integer N —an exponential “speedup” over
the number field sieve.
What explains this phenomenon? What is it, exactly, that allows quantum com-
puters to compute certain problems more efficiently than classical computers can?
Surprising as it may seem, there is still no consensus regarding the answer, neither
among the researchers working in the field, nor among the philosophers comment-
ing on this work. By contrast, in the popular literature on the topic, one answer
has tended to dominate all the others. This is the idea that quantum computers are
actually parallel devices that perform exponentially many classical computations
simultaneously, with each computation taking place in a parallel physical universe
(or “world”), different in certain ways from our own but just as real as it is.
Perhaps the strongest proponent of this view, which we will call the many-worlds
explanation of the power of quantum computers, is one of the fathers of the field,
David Deutsch, discoverer of the first quantum algorithm and of the universal quan-
tum Turing machine (Deutsch, 1985). The many-worlds explanation of quantum
computing is, for Deutsch, not just a speculative conjecture. For Deutsch, many-
worlds are the only plausible explanation for how quantum computers work. As he
puts it in his 1997 book, The Fabric of Reality: “[t]o those who still cling to a single-
universe world-view, I issue this challenge: Explain how Shor’s algorithm works”
(1997, p. 217).
Before we discuss what those who, like Deutsch, defend this view of quantum
computation are really saying, let us review briefly what the view amounts to in
more general terms in the context of quantum mechanics, for what the many-worlds
explanation of quantum computing amounts to is an application of this more gen-
eral philosophical interpretation of quantum mechanics to the case of quantum-
mechanical computers.28 Thus far, our examples of quantum-mechanical systems
have been of qubits, either considered singly or in combination with other qubits.
But quantum mechanics is not restricted, of course, to describing qubits. Quantum
mechanics is, rather, a universal language for describing any physical system, be it a
single particle or (in principle) the entire physical universe. From the point of view
of quantum mechanics, you and I, the laptop I am writing this chapter with, the table

28 The interpretation of quantum mechanics that we will be discussing in this section is one of
a number of related interpretations of quantum mechanics that are collectively referred to as the
“Everett interpretation”. These include, but are not limited to Hugh Everett III’s original formu-
lation (Everett III, 1956), the “Berlin Everettianism” of Lehner (1997), Lev Vaidman’s version of
Everett (Vaidman, 1998), so-called “many minds” variants (Albert & Loewer, 1988), and finally the
“many-worlds” variants that are the direct inspiration for the many-worlds explanation of quantum
computing. Belonging to the last group are DeWitt’s (1973 [1971]) view, as well as the “Oxford
Everett” interpretation (Deutsch, 1997; Saunders, 1995; Wallace, 2003, 2012) with which we will
be mostly concerned here.
128 M. E. Cuffaro

it is sitting on, and everything else in the world is a quantum-mechanical system.

And just as a qubit can be in a superposition of its basis states (see Eq. (2.6) and Fig.
7), so too can any physical system. As Schrödinger famously remarked, according
to quantum mechanics there are even ways of preparing a cat in a superposition of
“dead” and “alive” states (Schrödinger, 1935a).
When we actually come to pose a question to the system; i.e., to measure the qubit,
the cat, or what have you, we never actually find that it is in a superposition state.
Quantum mechanics predicts that the cat will be found to be either dead or alive,
the qubit to be either |0 or |1, and so on, with a certain probability. Thus, unlike
the physical state of a system as described by classical mechanics, we cannot simply
read off from the quantum state description of a system what the outcome of an
experiment on that system will be (except in special cases). This raises the question
of how to interpret the quantum state. Are we to take superpositions literally, despite
the fact that we never observe them? What would it mean if we did? The proponent
of the many-worlds interpretation of quantum mechanics answers yes to the first
question. As for the second question, the view of the many-worlds advocate is that
each branch of a superposition represents a distinct, classically describable (from the
point of view of that branch), physical universe.
Just as it is with a cat, or with any physical system, so it is with a quantum
computer. What is different, arguably, about a quantum computer (according to the
defender of the many-worlds explanation of quantum computing) is that in a quan-
tum computer the resources made available by these multiple physical universes are
harnessed to perform computations. This is a striking claim. But it can be said, in
favor of the many-worlds explanation, that some quantum algorithms certainly do
give us this impression. The following evolution is representative of a typical step
in many quantum algorithms (note that normalization factors have been omitted for
simplicity):

2
n
−1 2
n
−1
|xn |0 → |xn | f (x), (3.1)
x=0 x=0

What is being depicted here is the state of a quantum computer, composed of n + 1

qubits, where the first n qubits are called the input qubits of the computer, and the last
qubit is called the computer’s output qubit. The computer begins in the state described
on the left-hand side of the equation, and then transitions to the state described on
the right-hand side.
This notation is very compact, so let us unpack it. Considering, first, the left-hand
side: If we take the n input qubits as comprising, together, one subsystem of the
system, and the output qubit as comprising another subsystem, then we can say that
the input and output subsystems of the computer begin in a product state. In other
words, we can write the left-hand side of the equation in product form as
The Philosophy of Quantum Computing 129
2n −1 

LHS = |xn
|0. (3.2)
x=0

Note that, in Eq. (3.2), the n input qubits are together in a superposition of all of
the possible computational basis states for a system of n qubits. This is obscured
somewhat, both by the summation notation as well as by the shorthand, |xn , being
used to represent each superposition term. To clarify this, recall that for a two-
dimensional system, i.e., a single qubit, the computational basis states are, as we
pointed out earlier (see Eq. (2.6) and Fig. 7), the states |0 and |1. As for a system
composed of two qubits, it has four computational basis states, namely, |0|0, |0|1,
|1|0, and |1|1. In general, for a system of n qubits there are 2n basis states.

|0|0 . . . |0|0,
|0|0 . . . |0|1,
|0|0 . . . |1|0,
...
|1|1 . . . |1|1, (3.3)

each of which can be thought of as representing a binary number. For short, we can
represent these in decimal as {|0, |1, |2, |3, . . . |2n−1 }. A superposition of all
of these basis states is then given by

2
n
−1
|0 + |1 + |2 + |3 + . . . + |2 n−1
 = |xn , (3.4)
x=0

exactly as we see in Eq. (3.2).

On the right-hand side of Eq. (3.1), the n + 1 qubits of the quantum computer
are no longer in a product state. The state of the computer has now transitioned to
an entangled state: a superposition in which the state obtained for the output qubit,
conditional upon measuring it, is correlated with the state that will be obtained for
the input qubits, conditional upon measuring them. For instance, if we measure the
input qubits and get the state |1, then the state of the output qubit will be | f (1), for
the given function f , and similarly for every other value of x. What is the function
f ? That will depend on the particular problem that the algorithm is designed to
solve. More importantly, from the point of view of our current discussion, notice that
the right-hand side of the equation encodes exponentially many evaluations of that
function in the state of the computer! We can, or so it seems, appeal to the results of
all of these evaluations to help us with whatever problem we are trying to solve.
In reality, things are not so easy, for when we actually come to read off the
result of the computation, we will only ever find the computer’s output qubit to be
in one of its exponentially many superposition terms; i.e., in some state | f (a) for
some one particular a in the domain of f . If we are tempted to read the presence
of these exponentially many function evaluations in the description of the state of
130 M. E. Cuffaro

the computer literally, then the fact that only one of them can ever be accessed in
a particular run should give us some pause (cf. Mermin, 2007, p. 38). That said, in
any given run, there will be an, in general, non-zero probability of obtaining any one
of them, and this, perhaps, speaks in favor of viewing them all as somehow literally
there. As for the goal (which we should not lose sight of) of actually solving the
problem at hand, what is just as important as achieving the form of the state of the
computer on the right-hand side of Eq. (3.1) is the next step in a computation like
this one, which requires that we manipulate the system cleverly enough so that one
of the desired solutions is found with higher probability than the other, undesirable,
solutions (Pitowsky, 2002, §3).
The many-worlds explanation of a quantum computational process enjoins us to
take an evolution like the one given in Eq. (3.1) ontologically seriously. It affirms,
in other words, that things are indeed as they seem: The computer most definitely
is performing many simultaneous function evaluations in parallel when it is in a
state like the one on the right-hand side of Eq. (3.1). Moreover, the many-worlds
explanation directly answers the question of where this parallel computation occurs,
namely, in distinct physical universes. For this reason, it is also, arguably, one of the
more intuitive of the physical explanations of quantum speedup. And it is certainly
thought-provoking. It is no wonder, then, that it is the one most often mentioned in
the popular literature on quantum computation. And it has advocates in the more
serious literature as well.29
So far we have highlighted one potential problem for the many-worlds explana-
tion: Despite the appearance of parallel processing in an equation like Eq. (3.1), only
one of these function evaluations is ever accessible on a given run of the computer.
This is not to say that no story can be told from the point of view of the many-worlds
explanation about why this is so, but at the very least this is enough to cast doubt on
the claim that an evolution like the one given in Eq. (3.1) constitutes, all by itself,
evidence for the many-worlds explanation of the power of quantum computers. We
will return to this issue later. But for now, we need to consider a somewhat deeper
problem faced by the advocate of the many-worlds explanation. Recall that decom-
posing a qubit in the state |ψ into the computational basis states |0 and |1 (see
Eq. (2.6) and Fig. 7) is only one way to express this state. The same state can also be
expressed in other bases besides the computational basis. This means that a system
that is in the state
1 1
√ |0 + √ |1, (3.5)
2 2

from the point of view of the computational basis, is simply in the state:

|+ (3.6)

29In addition to Deutsch’s 1997 book, see Deutsch (2010), and see also Vaidman (2018 [2002],
§7) and Wallace (2012, Chap. 10). The strongest and most in-depth defence of the many-worlds
explanation of quantum computing that I am aware of is the one given by Hewitt-Horsman (2009).
The Philosophy of Quantum Computing 131

from the point of view of the {|+, |−} basis (see Eq. (2.12)). But now it is no longer
clear just how many universes we should take this qubit to exist in. If we decompose
its state in the computational basis, then it would seem that it exists in two worlds,
and if we decompose it into the {|+, |−} basis then it would seem that it only exists
in one. Which is it? If we decide that one of these decompositions is to be preferred
to the other, then the challenge for the advocate of the many-worlds explanation is to
give a compelling reason why, for prima facie there does not appear to be any reason
for preferring one of these decompositions over the other.
This is known as the preferred basis problem, and it is a problem for the many-
worlds view more generally, i.e., not just in the context of quantum computing. In
this more general context, advocates of the many-worlds picture (see, for instance,
Wallace, 2003) attempt to solve the preferred basis problem by appealing to the
dynamical process of decoherence. In quantum mechanics, as we saw (see Eq. (2.6)
and Fig. 7), the state of a system at a particular moment in time is described by
a state vector that evolves linearly and unitarily. However, the evolution thereby
described is the evolution of a system that is completely isolated from its external
environment, which is an idealization; in reality, it is actually never possible to
completely isolate a system from its environment,30 unless, perhaps, we take our
system of interest to be the universe in its entirety.31 But barring the universe as a
whole, when a system interacts with an external environment—for example, with
our measurement apparatus as we ask the system an experimental question—then
the terms in the superposition describing its state begin to decohere (Zurek, 2003
[1991]) and come to achieve a kind of independence from one another (although they
never decohere completely). On the many-worlds picture, we are to think of such
(approximately) decoherent terms as existing in independently evolving physical
universes. And in each of these independently evolving physical universes, there is,
in addition, a different version of ourselves, all of whom receive a different answer
to the experimental question that was asked. And with every additional question, the
universe as we know it is branching, spawning exponentially more and more versions
of the system, and more and more versions of ourselves along with it, and so on and
on and on.
Fundamentally, however, decoherence is an approximate phenomenon, and some
small amount of residual interference between worlds always remains despite it.
Nevertheless, decoherence tells us that, when the environment and system happen
to be correlated in a particular way, then a particular basis will emerge with respect
to which we can effectively describe the superposition terms expressed in the state
of the system as evolving independently of one another. As Wallace (2003, p. 90)
puts it: “the basic idea is that dynamical processes cause a preferred basis to emerge
rather than having to be specified a priori”. In this way, superposition terms that for
all practical purposes evolve stably and independently over time with respect to the
decoherence basis can be identified with different copies of measurement pointers,

30 At the very least, the gravitational effects of other distant systems will not be able to be neglected.
31 Some philosophers have questioned whether we should think of even the universe as a whole as
a closed system (see, for instance, Cuffaro & Hartmann, 2021).
132 M. E. Cuffaro

cats, experimenters, and whatever else is theoretically useful for us to include in our
ontology: “the existence of a pattern as a real thing depends on the usefulness—in
particular, the explanatory power and predictive reliability—of theories which admit
that pattern in their ontology” (Wallace, 2003, p. 93). Whatever else one may think
of the many-worlds view, decoherence, at least, does provide a principled way to
identify worlds in the wave-function.
For the advocate of the many-worlds explanation of quantum computation, how-
ever, there is still a problem. Appealing to decoherence may solve the preferred basis
problem for the purposes of describing the world of our everyday experience,32 but
the inner workings of a quantum computer are not part of that everyday experience.
The problem is not just that qubits are too small to see. The problem is that the super-
positions characteristic of quantum algorithms are coherent superpositions (Nielsen
Chuang, 2000, p. 278; see also Aaronson, 2013c; Cuffaro, 2012). Thus, the terms
in the wave-function of a quantum computer do not seem to meet the criterion for
world-identification advocated for by the many-worlds view.
Now, to be fair, a similar thing can be said with regard to our everyday experience.
So-called decoherent superpositions are (as I mentioned) not really decoherent, after
all. But they are decoherent enough, and for long enough, that it is useful (or so says
the many-worlds advocate) to think of the terms in such a superposition as inde-
pendent, and the worlds that they describe as ontologically real. Likewise, although
the superposition on the right-hand side of Eq. (3.1) is actually a coherent superpo-
sition, it may nevertheless be useful to think of the terms in that superposition as
independent, at least for the short time that the quantum computer is in that state
(Hewitt-Horsman, 2009, p. 876). The problem, however, is that it is the very fact
that they are coherent that allows us to “cleverly” extract desirable solutions from
these superpositions with higher probability than undesirable solutions (Bub, 2010;
Duwell, 2007).
Even if we grant that there is some heuristic value in describing a quantum com-
puter in a superposition state as evaluating functions in exponentially many parallel
worlds (I do not doubt that this was of some heuristic use to Deutsch, for instance,
even if I question whether it is necessary to think of such superpositions in this way),
it does not follow that this is enough to licence granting ontological status to those
worlds. Wallace (2003, p. 93) mentions (as we saw) explanatory power and predictive
reliability, for instance, and discusses the way that these and other ideas are applied
in contemporary physics to support the many-worlds view outside of the context of
quantum computing. It is not at all clear that these criteria are met in the context of
quantum computing, however, and even Wallace admits that they, for the most part,
are not: “There is no particular reason to assume that all or even most interesting
quantum algorithms operate by any sort of ‘quantum parallelism”’ (Wallace, 2010,
p. 70, n. 17). Wallace goes on: “Shor’s algorithm, at least, does seem to operate in

32 The preferred basis problem is not the only challenge that needs to be met by an advocate of the
Everett interpretation of quantum mechanics. Another issue that has been much discussed in recent
literature is the problem of how to account for probabilities on the Everettian view. For more on
this issue, see Adlam (2014), Dawid and Thébault (2015), Greaves and Myrvold (2010), Vaidman
(1998, 2012), and Wallace (2007).
The Philosophy of Quantum Computing 133

this way” (ibid.), but he does not describe how. Yet there are very plausible accounts
of how Shor’s algorithm works that do not appeal to massive parallelism at all (see
Bub, 2010). Far from it, on Jeffrey Bub’s account of Shor’s algorithm, the quan-
tum algorithm is more efficient than known classical algorithms because it performs
fewer, not more, computations (see also Bub, 2008).
The final reason that I will mention for being skeptical of the many-worlds expla-
nation of quantum computing is that it only really seems to be useful in the context
of one particular model of quantum computing. This is the so-called quantum circuit
model, the model for which many of the first quantum algorithms were designed
(Deutsch, 1989). This model is useful for abstract theoretical purposes, as well as
for pedagogical purposes, as it borrows many familiar ideas from the classical cir-
cuit model of computation (see Fig. 4). In the quantum circuit model, similarly to
the classical circuit model, logical circuits are constructed out of various “quantum
logic gates”. These instantiate unitary transformations of one or more qubits that are
prepared beforehand in computational basis states (typically qubits begin a compu-
tation in the state |0). The unitary gates transform the qubits’ states into various
superpositions of computational basis states, and at the end of the computation a
measurement is performed, again in the computational basis, on (some of) the qubits
and the results are read out.
Figure 8a depicts a number of important one- and two-qubit quantum gates. The
X gate implements a qubit-flip operation; i.e., it takes |0 → |1 and vice versa.
The Y gate takes |0 → i|1 and |1 → −i|0. The Z gate takes |0 → |0 and
|1 → −|1. The R gate takes |0 → |0 and |1 → i|1. The H (or Hadamard)
√ √
gate takes |0 → (|0 + |1)/ 2 and |1 → (|0 − |1)/ 2. The S gate takes |0 → |0 and
|1 → eiπ/4 |1. At the extreme right is the two-qubit CNOT (or controlled-not) gate. It
leaves the topmost qubit unchanged. The bottom qubit is then assigned the output of
taking the exclusive-or of both, i.e., this gate takes |0|0 → |0|0, |0|1 → |0|1,
|1|0 → |1|1, and |1|1 → |1|0. The X , Y , Z , R, H , and CNOT gates together
form the Clifford group of gates, which we will have more to say about later. If we
add the S gate to the Clifford group, they together form a universal set of gates,
i.e., any quantum circuit implementing any series of unitary transformations can be
simulated to arbitrary accuracy using combinations of these seven gates.
Figure 8b depicts a quantum circuit diagram for Deutsch’s Algorithm, which
determines whether a given function f on one bit is constant ( f (0) = f (1)) or
balanced ( f (0)
= f (1)): Two qubits are prepared in the product state |0|0 and are
each sent through an X -gate and a Hadamard gate, after which they are together input
to the two-qubit entangling unitary gate U f . The first qubit is then sent through a
further Hadamard gate and finally measured to yield the answer (see Deutsch, 1989).
Just as there are various models of universal classical computation (for instance,
the various versions of the Turing machine, as well as the von Neumann architecture,
and so on, that I mentioned above), there are various models of universal quantum
computation. One computational model that presents a particularly difficult problem
for those who would advocate for the many-worlds explanation is the cluster-state
model of quantum computing, also known as measurement-based and one-way quan-
tum computing (Raussendorf & Briegel, 2002; Raussendorf et al., 2003; Nielsen,
134 M. E. Cuffaro

Fig. 8 a A number of important one- and two-qubit quantum gates. The X , Y , Z , R, H , and CNOT
gates together form the Clifford group. If we add the S gate to the Clifford group, they together form
a universal set. b A quantum circuit diagram for Deutsch’s Algorithm, which determines whether
a given function f on one bit is constant ( f (0) = f (1)) or balanced ( f (0)
= f (1))

2006).33 In the cluster-state model, the computer begins, not in a product state but in
a highly entangled state, and measurements are performed not only at the end of the
computation but throughout it. These measurements are adaptive, in the sense that
each measurement is performed in a different basis, which depends on the random
outcomes of whatever previous measurements have been performed.
Why is this a problem for the advocate of the many-worlds explanation of quantum
computing? Because the fact that measurements are adaptive means, on the one hand,
that there is no principled way to select a preferred basis a priori in the context of a
given computation (Cuffaro, 2012, §4). Whichever basis we choose, few qubits will
actually be measured in that basis in the course of the computation. On the other
hand, there is no sense in which we can say that a preferred basis “emerges” from
the computational process. There is, therefore, no way to identify the worlds that the
computation as a whole is supposed to be happening in.
As I alluded to above, both the cluster-state model and the circuit model are
universal models of quantum computing. Thus, anything that the circuit model can
do can be done (and, as it happens, just as efficiently) in the cluster-state model and
vice versa. Perhaps, then, it could be argued that understanding the “true nature” of
algorithms in the cluster-state model requires that we first translate them into the
language of the circuit model, though I can think of no reason why one should think
so other than a desire to hold on to the many-worlds idea come what may. The proper
response to anyone who would put forward such an argument is that a large part
of what motivates those who adhere to the many-worlds explanation of quantum
computing in the first place is that it is useful for algorithm analysis and design to

33For introductions to cluster-state quantum computing aimed at philosophers, see Cuffaro (2012,
§4) and Duwell (2018, §4.5).
The Philosophy of Quantum Computing 135

believe that a quantum computer is carrying out its computations in parallel worlds.
This does not seem to be so for the cluster-state model. On the contrary, dogmatically
holding on to the view that many-worlds are, at root, physically responsible for
the speedup evinced in the cluster-state model is at best useless, for it is of no
help in designing algorithms for the cluster-state model. At worst, dogmatically
holding on to the many-worlds idea could prove positively detrimental if it prevents
us from exploiting the power of the cluster-state model or discovering other quantum
computational models in the future.

4 Quantum Computation and Entanglement

The many-worlds explanation of the power of quantum computers is not the only
explanation that has been proposed by philosophers. I have already mentioned the
upshot of Bub’s analysis of Shor’s algorithm in the previous section. Bub’s view,
more generally (2006, 2010), is that the key to understanding the power of quan-
tum computers lies in understanding the way that they exploit the novel logico-
probabilistic structure of quantum mechanics (Pitowsky, 1989). We saw above that
in quantum mechanics, the (classical) probability distribution over the answers to a
particular experimental question—the one that we infer, conditional upon our asking
that question—cannot be embedded into a global prior probability distribution over
all of the answers to all of the questions we might want to ask. But although these
individual probability distributions do not logically fit together as they do for a classi-
cal system, there are logical relations between them nonetheless, that are compactly
described by the quantum state.
For instance, as a general rule (which admits exceptions) one cannot use a quan-
tum system to compute a logical disjunction by computing the individual values
of each disjunct. This is simply because in the general case both disjuncts will not
be globally defined. However, quantum mechanics’ logical structure provides other
ways to compute a logical disjunction, and these other ways (which are not available
to a classical computer) are exploited by a quantum computer to compute certain
problems more efficiently than a classical computer can compute them.
Another view is Duwell’s (2018, 2021), who in contrast, agrees with the proponent
of the many-worlds explanation in identifying quantum parallelism as at least part of
the source of the power of quantum computers. Duwell, however, resists the tempta-
tion to make the a metaphysical inference from parallelism to many computational
worlds. For Duwell, the source of the power of a quantum computer lies in the way
that it can efficiently correlate multiple values of a function and use these correlations
to efficiently extract global information about the function. The disagreement over
whether a quantum computer performs more, or fewer, computations than a classical
computer is one that Duwell views as arising from conflicting intuitions about how
to appropriately describe the quantum systems that perform computational tasks.
In these and other candidate explanations of the power of quantum computers that
one encounters in the philosophical literature on the topic, the fact that quantum-
136 M. E. Cuffaro

mechanical systems exhibit entanglement (see Sect. 2.4) invariably plays an impor-
tant role. For Bub entanglement is absolutely central, as entanglement is a (direct)
manifestation of the fact that the logical structures of quantum and classical mechan-
ics differ. As for Duwell, he takes his quantum parallelism thesis to be completely
compatible with the idea that entanglement plays a central role in quantum comput-
ing, even if his explanation emphasizes the correlations between the values of the
function being evaluated by a system rather than the logical structure of its under-
lying state space (Duwell, 2018, p. 101). The many-worlds advocate, as well, views
quantum entanglement to be indispensable in the analysis of the power of quantum
computing (Hewitt-Horsman, 2009, p. 889), even if for the many-worlds advocate it
does not, by itself, suffice as a philosophical explanation for it.
The debate over the interpretation of the phenomenon of quantum entanglement
has historically been one of the central debates in the controversy over quantum the-
ory’s conceptual foundations. First emphasized by Albert Einstein, Boris Podolsky,
and Nathan Rosen in their 1935 criticism of the orthodox interpretation of quan-
tum mechanics, Erwin Schrödinger called it “the characteristic trait of quantum
mechanics, the one that enforces its entire departure from classical lines of thought”
(Schrödinger, 1935b, p. 555, emphasis in the original). It was only with the work of
Bell (2004 [1964], 2004 [1966]), however, that its significance for our understanding
of the break that quantum mechanics makes with classical physics was first made
fully clear.
Consider Louise and Robert, two friends who live in the 15th and 11th arrondisse-
ments, on the left bank and the right bank of the Seine, respectively, in the city of
Paris. Every morning, Louise and Robert each receive a letter in their respective
mailboxes that consists of a single sheet of paper inscribed with either a large bass
clef:

or a large treble clef

After awhile, Louise determines that it is equally likely, on any given day, that she
will receive a bass clef as it is that she will receive a treble clef. After awhile, Robert
determines the same. But when they compare their notes, they find that whenever
Robert receives a bass clef, so does Louise. Similarly, every time he receives a treble
clef, she does too. In other words, there is a one-in-two chance, on any given day, that
they both receive a bass clef, and a one-in-two chance that they both receive a treble
clef. No other combinations are ever observed. Thus, their outcomes are correlated,
The Philosophy of Quantum Computing 137

such that the probability distribution over the possible combinations of outcomes is
given by34 :

(4.1)

What explains this correlation? Well in this case it turns out that Louise and Robert
both play in a jazz ensemble. The band’s leader, Carsten, lives in the center of the city
(on the Île Saint-Louis). Every afternoon he flips a fair coin. Depending on whether
the coin lands heads or tails, he either writes a large bass clef or a large treble clef
on a sheet of paper, photocopies it, and sends one copy each to Robert and Louise. If
Robert receives a treble clef from Carsten, then he will know to take his tenor horn
to the jazz club that night. Otherwise, he will take his trombone. As for Linda, if she
receives a treble clef, she will know to bring her soprano clarinet. Otherwise, she
will bring her bassoon.
The result of Carsten’s coin flip is called the common cause of Louise’s and
Robert’s outcomes, and the story we tell about how Carsten’s coin flip determines
Louise’s and Robert’s outcomes is called a causal model for the correlations that they
see (see Pearl, 2009). If Louise and Robert do not know how Carsten is determining
what to send them, then they will wonder about the hidden-variable that explains
their correlation. In this case, the result of Carsten’s coin flip is actually a local
hidden-variable, since the process by which the outcome of the coin flip determines
what is written on the letters is confined to the small localized region in the vicinity
of the desk in Carsten’s central office. He flips the coin, takes note of the outcome
(the local hidden-variable), and writes the corresponding symbol.
Instead of simply flipping a coin while sitting at his desk, we can imagine a
more complicated, spatially distributed, process by which Carsten determines what to
write. For instance, Carsten might begin by flipping his coin, and then, corresponding
to heads or tails he might telephone Tilde or Bjarne, who live in the northern and
southern suburbs of the city, respectively, and ask whichever of them he calls to roll a
six-sided die at exactly seven-o’clock in the evening, and to afterward call him back
at either 7:13 PM, if it is Bjarne, or 7:18 PM, if it is Tilde, to tell him the outcome.
Then, if the result of the die roll is one, three, or six, Carsten will write a treble clef on
the sheet of paper before photocopying it and sending copies to Robert and Louise,
while if it is two, four, or five, he will write a bass clef. All of these actions together
constitute, just like the coin flip in our simpler scenario, a locally causal model to
explain Louise’s and Robert’s correlation. Why do we call it local even though it is
spatially distributed? Because the physical processes (the coin flips, die rolls, and

34 If the outcomes were completely uncorrelated, the probability distribution would be

138 M. E. Cuffaro

telephone calls) by which Carsten determines what to write propagate locally in the
physical sense, i.e., at a finite speed that is less than the speed of light.
Bell (2004 [1964]) showed, first, that there are certain probabilistic constraints—
what are now called Bell inequalities—on the statistics arising from measurements on
any locally correlated system. He then showed that these constraints are violated by
statistics that are predicted to arise from certain experiments on a system composed
of two qubits in the quantum-mechanical state35

1  
√ |0 A |1 B − |1 A |0 B , (4.2)
2

where the A and B subsystems can be as far apart as one likes. The proof of this
violation is known as Bell’s theorem and it, and its variants, have since been experi-
mentally confirmed many times over (Genovese, 2016).
The predicted violation only occurs for certain measurements. If we measure
both the A and B qubits in the computational basis (see Eq. (2.6) and Fig. 7), then
the predicted statistics will actually be compatible with the constraints imposed by
local hidden-variable theories, as Bell himself showed (2004 [1964], p. 16). But
as we rotate the measurement basis (see Fig. 7) that we use for B away from the
measurement basis that we use for A, quantum mechanics predicts that we will see
a violation, that it will reach a maximum at a certain point, and then decrease again
as the measurement basis for B begins to line up again with the measurement basis
for A.
Another name for the computational basis is the Z -basis. We call it the Z -basis
because the two computational basis vectors |0 and |1 correspond to the two possible
outcomes of a Z -basis measurement on a qubit, where “measuring a qubit in the Z -
basis” means sending it through a Z -gate (see Fig. 8) and then recording its state
(Janas et al., 2022, §6.5). Similarly, the basis vectors |+ and |− are the two possible
outcomes of an X -basis measurement on a qubit, and |y +  and |y −  are the two
possible outcomes of a Y -basis measurement. The X , Y , and Z gates, together with
the trivial I gate that leaves a qubit’s state unchanged, are known as the Pauli gates.
I will have more to say about this family of gates later. For now, I want to point out
that, for a system of qubits in the state given by Eq. (4.2), as long as both the A and
B qubit are measured in one of the Pauli bases, the predicted statistics arising from
those measurements will not violate the constraints that Bell’s inequality imposes
on local hidden-variable theories. In other words, if all that we have access to are
measurement devices that can measure a qubit in one of the Pauli bases, then there
will be no way to disprove some local hidden-variable story a skeptic might cook up to
explain the observed statistics. To experimentally disprove such a story, we will need
to have measurement devices that can measure in other measurement bases besides
these ones, the ones for which Bell showed that a violation of the Bell inequalities
will occur.

35 This state is identical to the one given in Eq. (2.11) but we repeat it here for convenience.
The Philosophy of Quantum Computing 139

Bell’s theorem should convince us that no local hidden-variable theory can repro-
duce the correlations arising from such measurements (i.e., from measurements in
bases other than X , Y , and Z ), but this does not bar a skeptic from positing a nonlocal
hidden-variable theory to recover them instead. An example of such a theory is one
in which the outcome of a measurement on A depends on the measurement basis
used to measure B. But since A and B could in principle be far apart in space, the
causal influence from B to A in such a theory will have to be propagated faster than
the speed of light. This is a hard pill to swallow, but one might be inclined to swallow
it anyway in order to avoid the other alternatives.36
Coming back to our discussion of quantum computers, recall that when we dis-
cussed the many-worlds explanation of quantum computing, we noted that the com-
puter’s state as given on the right-hand side of Eq. (3.1) is entangled. The question
arises, then, as to what role is played by entanglement more generally in enabling
quantum computers to outperform their classical competitors. There are two ways to
think of this question. We might ask, on the one hand, whether realizing an entangled
state is necessary to enable a quantum speedup. On the other hand, we might ask
whether it is sufficient to enable it.
There has been some debate surrounding the first question. On the one hand, Jozsa
and Linden (2003) have proven that realizing an entangled state is necessary to enable
speedup if a quantum computer can be assumed to be in a so-called “pure state”, i.e.,
in a state that represents a maximally specific description of the system from the point
of view of the quantum-mechanical formalism. This is the case when we represent
a system with a state vector as we have been doing up until now (and will continue
to do). On the other hand, it has been argued (Biham et al., 2004) that quantum
computers that are in “mixed states”—i.e., states which describe the computer in
a less than maximally specific way, either because we are ignorant of the actual
state of the computer, or because the computer is coupled to its environment—are in
some cases capable of achieving a quantum speedup over their classical competitors
without ever being entangled. There is insufficient space to discuss this (largely
technical) debate here, but in Cuffaro (2013), I argue that the purported counter-
examples to what I there call the “necessity of entanglement thesis” do not actually
demonstrate what they purport to show, but instead clarify the necessary role that
entanglement does play in quantum computation.
From the philosopher’s point of view, the more interesting question in relation to
the role of entanglement in quantum computing is the question regarding sufficiency,
for as we will see, reflecting on this question sheds light on the tension between
physical and computational ways of thinking that is at the heart of the science of
quantum computing, the tension that is the primary source of the insight this new
science brings into both physics and computation.

36 For further discussion, see Myrvold et al. (2020).

140 M. E. Cuffaro

5 The Gottesman–Knill Theorem

The main reason for being skeptical of the idea that entanglement suffices to enable
quantum speedup is the Gottesman–Knill theorem (Gottesman, 1999). This theorem
states that any quantum algorithm that employs (exclusively) some combination of
the following operations (which together form what we will call the Gottesman–
Knill set) is efficiently simulable by a classical computer: (i) The Clifford group of
gates (see Fig. 8), i.e., the X , Y , Z , R, H , and CNOT gates; (ii) Clifford group gates
conditioned on the values of classical bits (indicating, e.g., the results of previous
measurements); (iii) state preparation of a qubit in the computational basis (as one
typically does for each qubit prior to the beginning of a computation); (iv) measure-
ments in one of the Pauli bases (as one does at the end of a computation) (Nielsen &
Chuang, 2000, §10.5.4).
The reason the theorem represents a challenge, to the view that entanglement
suffices to enable a quantum computer to outperform a classical computer, is that by
using only the operations given in (i)–(iv), which according to the theorem are all
efficiently classically simulable, it is possible to generate an entangled state. Figure 9
depicts a quantum circuit to realize an entangled state using only operations from the
Gottesman–Knill set. It begins with the preparation of two qubits in the computational
basis state |0, then subjects them each to an X -gate, as a result of which they will
both be in the state |1. The first qubit is then run through a Hadamard gate which
√
transforms it into the superposition state (|0 − |1)/ 2, and following that both qubits
are run through a CNOT gate. The resulting state of the computer is an entangled
state.
It would be wrong to conclude (cf. Jozsa & Linden, 2003, pp. 2029–30), on the
basis of this, that entanglement is not sufficient to enable a quantum computer to out-
perform a classical computer. Why? Well, let us reflect on what the Gottesman–Knill
theorem is saying. The Gottesman–Knill theorem shows that there is a certain set of
quantum operations that can be efficiently simulated on a classical computer. Let us
then consider what we mean when we say that we have “efficiently classically sim-
ulated” something. We can say, to start with, that in a classical computer simulation
(efficient or otherwise), whatever computer is doing the simulating will be such that

Fig. 9 A quantum circuit (above), written out explicitly in terms of the unitary transformations
required to realize it (below), involving only operations from the Gottesman–Knill set. At the end
of this sequence of operations, the computer will be in an entangled state
The Philosophy of Quantum Computing 141

it can be given a classical description. As we saw in Sect. 2.1, a complete description

of a system, in classical mechanics, is always factorisable into complete descriptions
of all of its subsystems (see Fig. 3). Besides this, classically describable systems and
processes are locally causal, just like the convoluted procedure that Carsten uses to
determine what to stuff in the envelopes he sends to Louise and Robert. The classical
language of classical mechanics constrains all the descriptions of systems that can
be given in that language to be this way. The behavior of a classically describable
system, in other words, is always expressible in terms of local causes and effects, and
correlations between effects manifested by such a system’s subsystems can always
be explained by appealing either to common or direct causes, whose influences prop-
agate no faster than the speed of light. This is true no matter how large a system is
conceived to be. For classical mechanics aims to be a universal language for describ-
ing any physical system, be it a single particle or the entire physical universe. The
local causes of an effect will not always be known to us, but classical mechanics tells
us that they are there, and that we can find them in principle.
Let us return to Carsten, our band leader who lives on the Île Saint-Louis in the
center of Paris. Carsten has lately become interested in quantum computers as a way
of helping with composition,37 and has acquired a shiny silver box, now sitting on
his desk, which he claims is one. Quadeisha, a friend and fellow musician who has
happened by, is skeptical, and asks him to show her how it works. Obligingly, Carsten
turns on his computer and has it run what he tells her is a calculation whose result
depends on the prior generation of an entangled state. Quadeisha, still not satisfied,
asks Carsten to describe the algorithm by which the computer arrives at its result.
Upon discovering (after looking through the owner’s manual for the device) that the
algorithm employs no operations outside of the Gottesman–Knill set, she refuses to
believe that Carsten’s computer is quantum.
Quadeisha is right to be skeptical, even if the box on Carsten’s desk (which,
let us assume, he does not know how to open) actually is a quantum computer.38
Moreover, she need not have heard of the Gottesman–Knill theorem to be justified
in her skepticism. She need only be familiar with the work of John Bell. For it can be
shown that the combined effect of any sequence of Gottesman–Knill operations, for
any subsystem of the system to which they are applied, is equivalent to a measurement
in one of the Pauli bases, X , Y , Z , on a system whose state is given by one of the
basis vectors of a Pauli basis (Cuffaro, 2017, p. 115). Bell already showed us how to
provide a locally causal model to reproduce the statistics of such measurements on
a system in the state given in Eq. (4.2). Moreover, his technique is straightforwardly
extendable to other similarly entangled two-party systems (Cuffaro, 2017 §A.1),
and further techniques have been devised for constructing efficient locally causal
models to recover the statistics arising from Pauli-basis measurements on entangled

37For a review of some of the uses envisioned for quantum computers in music, see Miranda (2021).
38This is actually the attitude (and for similar reasons) that many theorists take toward certain claims
by private corporations to have built a working quantum computer (see, e.g., Aaronson (2013a),
Shin et al. (2014)).
142 M. E. Cuffaro

systems composed of three or more parties as well (see Cuffaro, 2017, §A.2, which
summarizes a technique first described in Tessier, 2004).39
The upshot of the Gottesman–Knill theorem is that a certain number of quantum
operations are efficiently classically simulable despite being capable of generating
entangled states. And we have just seen that to say that some phenomenon is “effi-
ciently classically simulable” is to say that it can be efficiently redescribed in a
locally causal way. Since any sequence of Gottesman–Knill operations is equivalent
to a measurement in one of the Pauli bases, X , Y , Z , on a system whose state is
given by one of the basis vectors of a Pauli basis, the Gottesman–Knill theorem is
essentially telling us that there are some statistics associated with measurements on
systems in entangled states that admit of a locally causal description.
We do not really need the Gottesman–Knill theorem to tell us this, for Bell’s
and related inequalities amount to general constraints that any locally causal model
of combined measurement statistics must satisfy, and we already know that in
some cases, i.e., for measurements in one of the Pauli bases, the predictions of
a locally causal model are compatible with quantum-mechanical predictions. It is
therefore misleading to conclude, on the basis of the Gottesman–Knill theorem, that
entanglement is not sufficient to enable quantum computational speedup. What the
Gottesman–Knill theorem shows us is that one also needs to use this entanglement
to its full potential, in particular, by not restricting a quantum computer to use only
a small proportion of the operations of which it is capable (i.e., by not restricting it
to the Gottesman–Knill set).
In the context of more general discussions of quantum mechanics, there are those
who claim that entanglement is the one and only distinguishing feature of the theory
(Schrödinger, 1935b). This is a controversial claim, but it is not proved false by the
fact that Bell (2004 [1964]) showed that some operations on some entangled states
can be redescribed in a locally causal way. Likewise, pointing out essentially the
same thing in the context of quantum computation is not an objection to the claim
that quantum entanglement constitutes a sufficient physical resource for realizing
speedup over classical computation. Indeed, the Gottesman–Knill theorem serves to
highlight the role that is actually played by entanglement in a quantum computer and
to clarify why and in what sense it is sufficient to preclude the computer’s evolution
from plausibly being classically simulated (for more on this, see Cuffaro, 2017).

6 Computer Simulations and Locally Causal Models

of Quantum Phenomena

Above we mentioned that techniques have been developed for providing classical
computer simulations, i.e., locally causal descriptions, to recover the statistics aris-
ing from Pauli-basis measurements on entangled systems whose subsystems are in
superpositions of eigenstates of Pauli observables, for any number of parties. The

39 The n-party case, for n ≥ 3, introduces subtleties which we will discuss in the next section.
The Philosophy of Quantum Computing 143

classical computer simulations of three or more party systems are especially inter-
esting in that they necessarily involve a small amount of communication. For three
parties, A, B, and C, one bit must be communicated between B and A in order to
make the protocol work. For n parties, n − 2 bits are required, i.e., the number of
bits required scales linearly with n (Tessier, 2004). Note that this counts as being
“easy” in the complexity-theoretic sense we discussed in Sect. 2.2.
From a certain point of view, however, it may seem quite odd to call a classi-
cal computer simulation of quantum correlations, that involves a certain amount of
communication between parties, a locally causal model of those correlations. Recall
that, earlier, when we introduced Bell’s theorem, we claimed that communication
between distant parties in the context of a Bell experiment constitutes a nonlocal
influence, i.e., one that would have to be communicated at a speed exceeding that of
light.
We can start to resolve the tension here if we consider, first, just what a locally
causal redescription of quantum measurement statistics is required to do. If a pair
of qubits, A and B, are prepared in an entangled state and then spatially separated,
quantum mechanics tells us that when, for instance, a Z -basis measurement is per-
formed on B, A’s state changes “instantaneously”, so that a Z -basis measurement on
A will yield an outcome that is correlated with the result of the experiment on B in a
particular way. Our challenge, however, is not to reproduce these apparent dynamics.
Our challenge is to come up with an alternative description of the statistics that we
actually observe in our experiments.
In order to assert that one has actually observed the results of a combined exper-
iment on two or more spatially separated subsystems, it must be assumed that the
results associated with the individual experiments on those subsystems have actually
been combined in some way. That is, we must somehow gather together the results
registered locally at the sites of the experiments on A and B. If Alice (a bassist)
is assigned to measure A, for instance, and Bob (who plays the piano) to measure
B, then once they record their individual results, they need to communicate them,
either to one another, or perhaps to Carsten who is sitting in his office on the Île
Saint-Louis, where he is in the process of showing Quadeisha how his quantum
computer is but one part of a network of spatially distributed quantum computers
that are together performing a computation that depends on the prior generation of
a system in an entangled state. While Alice and Bob are making their way toward
Carsten and Quadeisha with their results, however, there is time for Alice and Bob to,
surreptitiously, exchange further information with one another by local means (Bob
might call Alice on his cell phone, for instance, as they are making their way toward
Carsten’s office) in order to coordinate the outcomes of their individual measure-
ments and “correct” them if necessary, unbeknownst to Quadeisha. And since this
further information is propagated locally, i.e., at a speed no greater than that of light,
it can be considered as part of the common cause of the actually observed (combined)
measurement event, and thus as part of a classical, locally causal, description of that
measurement event (see Cuffaro, 2017, §4, Fig. 1).
Now, given a particular combined calculational outcome at a time t, one can
try to explain it in one of two ways: (i) Carsten’s way, i.e., as the result of a bona
144 M. E. Cuffaro

fide quantum-mechanical process manifesting nonlocal behavior, or (ii) Quadeisha’s

more skeptical way, i.e., as the result of a locally causal mechanism whose different
computational components are communicating with one another at less than the
speed of light. Without further evidence—if Carsten cannot or will not open his silver
box—it is perfectly legitimate to side with Quadeisha here, and be a skeptic with
respect to Carsten’s quantum-mechanical description of the box’s observed behavior.
Quadeisha’s alternative account of the observed correlations is a perfectly good one
in such a scenario, for her proposed classical alternative can solve the problem just as
easily, in the complexity-theoretic sense, as Carsten’s purported quantum computer.
Bell’s and related inequalities specify constraints on what a locally causal descrip-
tion that accounts for a combined probability distribution must be like, but in and of
themselves they are little more than formal statements. Bell’s (2004 [1964]) origi-
nal inequality, for instance, is essentially just a theorem of probability (Bell, 2004
[1981]; Janas et al., 2022, §3.4, Pitowsky, 1989, 1994). This is no fault. But if we are
to make a meaningful distinction between what is and is not ruled out by them, then
we must consider the context within which they are being applied. Normally this
does not need to be made explicit, for normally the context is what I have elsewhere
called the theoretical context (Cuffaro, 2017, p. 107), wherein we are considering
alternative theories of the natural world. In this context, Bell’s inequalities help us
to answer the question of whether there may be some deeper hidden-variable theory
of the natural world underlying quantum mechanics, and what such a theory needs
to be like.
They do not answer that question all by themselves, however. Any alternative
hidden-variable theory of the natural world must do more than merely satisfy the
constraints imposed by the Bell inequalities. It must also be plausible. Thus, besides
reproducing the well-confirmed statistical predictions of quantum mechanics, any
deeper candidate theory should also, for instance, be consistent with our other well-
confirmed theories of physics such as special and general relativity (and if not, then
a compelling reason will need to be given for why we should not worry about such
contradictions). It is on the basis of such plausibility constraints, and not on the basis
of the Bell inequalities, that we rule out many of the so-called “loopholes” to Bell’s
theorem.
Quadeisha’s alternative account of the observed correlations in Carsten’s office
counts as a particularly implausible example of such a loophole in the theoretical
context.40 But the theoretical context is not the only context that the Bell inequalities
are relevant to. Nor is it the context appropriate to a discussion of the respective
capabilities of quantum and classical computers. The context that is appropriate here
is what I have elsewhere called the practical context (Cuffaro, 2017, p. 107). In the
practical context, what concerns us is not alternative theories of the natural world.
What concerns us is what we are capable of building with the aim of reproducing

40Quadeisha’s loophole is actually conceptually similar to the “collapse locality” loophole at the
heart of Adrian Kent’s causal quantum theory (Kent, 2005). For discussion see Cuffaro (2017, pp.
104–106). For a more general discussion of the methodology of no-go theorems, see Dardashti
(2021).
The Philosophy of Quantum Computing 145

the statistical predictions of quantum mechanics. The plausibility constraints that

restrict us in the practical context are not the same as the ones that restrict us in
the theoretical context. But there are plausibility constraints in the practical context
nevertheless.
In particular, in the practical context, we should rule out, as implausible, alternative
locally causal descriptions of systems that would be too hard (in the complexity-
theoretic sense) for us to build. And we should rule in alternative locally causal
descriptions of systems that can be built easily (again, in the complexity-theoretic
sense). The complexity involved in the specification of such a system, in other
words, needs to be tractable. If, for instance, Quadeisha’s basis for being skeptical of
Carsten’s claim to own a quantum computer is that she can specify a locally causal
model for his computer’s calculation that requires a number of additional resources
that scales exponentially, then Carsten will be justified in dismissing her skepticism,
for he can with justice say that it is wildly implausible that anyone can have built
a machine according to Beatrice’s specifications, because simulating the quantum
calculation would amount, for such a machine, to an intractable task (Cuffaro, 2017,
§5).41

7 Computational Perspectives on Physics

In the previous section we discussed classical computer simulations of quantum-

mechanical phenomena, and we made the point that such simulations can be thought
of as locally causal alternative descriptions of those phenomena. This is not the
way that they are normally framed in the literature on quantum information and
computation. The way that it is normally put is that such simulations quantify the
extent to which quantum mechanics differs from classical mechanics, by enumerating
the number of resources required to account for quantum-mechanical phenomena
using classical means (see, for instance, Brassard, 1999; Rosset et al., 2013; Tessier,
2004; Tessier et al., 2005; Toner & Bacon, 2003).
Framing their significance in this way is certainly not incorrect, and it can be very
theoretically useful.42 But from the philosopher’s perspective there is value in, so to
speak, calling a spade a spade. A description of a plausible classical computational
model for efficiently simulating some of the operations that can be carried out by a
quantum computer is the same kind of thing as a plausible alternative local hidden-
variable account of some particular class of quantum phenomena, in the simple sense
that in both cases these constitute plausible locally causal models of quantum phe-
nomena. The difference between them lies in the way that one interprets the word

41 Note that I am taking ‘tractable’ here in a relative sense. That is, the resources required by a
classical computer to reproduce a particular effect should differ tractably from those required by a
quantum computer. Or in other words: it should not be essentially harder for the classical system
to produce the same effect as the quantum system.
42 The general subject of classical simulations of quantum systems is an important and burgeoning

area of modern physics (see, for example, Borges et al., 2010; Lee & Thomas, 2002).
146 M. E. Cuffaro

‘plausible’ in each case. In the practical context we assume that a system has been
built by a rational agent to achieve a particular purpose. We do not presuppose this
in the theoretical context.43 Clearly, certain assumptions that would be regarded as
plausible in the former context will not be in the latter and it is important to be on
guard against conflating them. As we saw in the previous section, practical investi-
gators attempting to isolate and/or quantify the computational resources provided by
physical systems may be in danger of conceptual confusion if they are not cognisant
of the differences between the practical and theoretical contexts. More positively, it
must not be assumed by practical investigators that every ‘no-go result’, formulated
in the theoretical context, will constrain what they can accomplish in the practical
context (Cuffaro, 2018a). More generally it is important to think seriously about
what one means by a ‘plausible locally causal model’ regardless of the context of
investigation.
Calling a spade a spade gives us, moreover, insight into how the practical context
can illuminate the investigations of the traditional philosopher of physics in the con-
text of quantum mechanics. The silly story we told in the previous section about how
Carsten, Alice, and Bob manage to reproduce, by communicating just a few classical
bits, all of the correlational phenomena associated with Pauli-basis measurements on
an n-party quantum system in an entangled state, besides representing just another
silly classical story, emphasizes perhaps the most fundamental difference that we can
point to between classical and quantum mechanics. Classical and quantum mechanics
are, fundamentally, alternative universal languages for describing physical systems
(Janas et al., 2022, §6.3). And by investigating the computational power inherent
in those languages we gain insight into the respective logical structures of classical
and quantum theory, into the logical structure of quantum mechanics that enables
the efficient representation of correlational phenomena by quantum-mechanical sys-
tems, and the logical structure of classical mechanics which precludes this (Janas et
al., 2022, §4.2).

8 Concluding Remarks

Reflecting on the lessons of quantum mechanics, Niels Bohr wrote that:

In representing a generalization of classical mechanics suited to allow for the existence of
the quantum of action, Quantum mechanics offers a frame sufficiently wide to account for
empirical regularities which cannot be comprised in the classical way of description (Bohr,
1948, p. 316, my emphasis).

This is a lesson with which the artist will be familiar.44 From time to time, in
literature, music, and in other forms of art, new methods of writing and composing
emerge that allow one to express the subtleties and nuances of experience more

43 This statement is not meant to express any sort of theological opinion. It is merely a statement
about how science operates, at least in this century.
44 A similar point is made in Janas et al. (2022, Sect. 6.3, note 22).
The Philosophy of Quantum Computing 147

easily than before. E. M. Forster (1942, p. 28) once said, about Virginia Woolf, that
“She pushed the light of the English language a little further against the darkness.”
In music, the (re-)introduction, into the Western musical tradition, of modality and
chromaticism through Jazz, Blues, and other forms of popular music, and also through
art music (see Vincent, 1951), has enabled modern composers in the Western tradition
to explore musical landscapes and express aspects of our emotional and intellectual
experience that composers limited to the major and minor scales, so dominant during
Western music’s classical period, cannot easily capture. In his musings about quantum
mechanics, Bohr himself was wont to appeal to an analogy with poetry. Werner
Heisenberg recalls, in his memoirs, a conversation he had with Bohr in which the
latter stated that:
We must be clear that, when it comes to atoms, language can be used only as poetry. The poet,
too, is not nearly so concerned with describing facts as with creating images and establishing
mental connections” (Heisenberg, 1971, p. 41).

These and like analogies were not lost on actually practising poets and other artists
of the same period (see Mairhofer, 2021). The subject of quantum mechanics has
even been found to lend itself very naturally to the format of the comic book (see Bub
& Bub, 2018, and the review by Cuffaro & Doyle, 2021, especially §4). As for the
science of quantum computing: The science of quantum computing takes advantage,
as we have seen throughout the course of this chapter, of the increased expressive
power of the quantum-mechanical language. It shows us that using the new language
of quantum mechanics, we can interact with what we take to be quantum-mechanical
systems in ways that cannot be easily replicated classically.
In the course of this chapter we have reflected on the fundamentals of classical and
quantum physics and computation, on the resources used and on the explanation of the
power of quantum computers, and finally on the broader insights that can be gleaned
from the science of quantum computing both for physics and for computation. The
philosophical issues addressed in this chapter are not the only issues brought up by
the science of quantum computing. But I hope to have convinced the reader, with
this sample of some of the more central ones, of quantum computing’s potential for
illuminating the world that we are all participants in.

References

Aaronson, S. (2013a). D-Wave: Truth finally starts to emerge. Posted: June 05, 2013. Retrieved
August 11, 2014, from
www.scottaaronson.com/blog/?p=1400
Aaronson, S. (2013b). Quantum computing since Democritus. Cambridge University Press.
Aaronson, S. (2013c). Why philosophers should care about computational complexity. In B. J.
Copeland, C. J. Posy, & O. Shagrir (Eds.), Computability: Turing, Gödel, Church, and beyond
(pp. 261–327). MIT Press.
Aaronson, S. (2016). Complexity zoo. complexityzoo.uwaterloo.ca/Complexity_Zoo.
Adlam, E. (2014). The problem of confirmation in the Everett interpretation. Studies in History and
Philosophy of Modern Physics, 47, 21–32.
148 M. E. Cuffaro

Agrawal, M., Kayal, N., & Saxena, N. (2004). PRIMES is in P. Annals of Mathematics, 160, 781–
793.
Aharonov, D., van Dam, W., Kempe, J., Landau, Z., Lloyd, S., & Regev, O. (2007). Adiabatic quan-
tum computation is equivalent to standard quantum computation. SIAM Journal on Computing,
37, 166–194.
Albert, D., & Loewer, B. (1988). Interpreting the many worlds interpretation. Synthese, 77, 195–213.
Arora, S., & Barak, B. (2009). Computational complexity: A modern approach. Cambridge Uni-
versity Press.
Bell, J. S. (2004 [1964]). On the Einstein-Podolsky-Rosen paradox. In Speakable and unspeakable
in quantum mechanics (pp. 14–21). Cambridge University Press.
Bell, J. S. (2004 [1966]). On the problem of hidden variables in quantum mechanics. In Speakable
and unspeakable in quantum mechanics (pp. 1–13). Cambridge University Press.
Bell, J. S. (2004 [1981]). Bertlmann’s socks and the nature of reality. In Speakable and unspeakable
in quantum mechanics (pp. 139–158). Cambridge University Press.
Bennett, C. H., Bernstein, E., Brassard, G., & Vazirani, U. (1997). Strengths and weaknesses of
quantum computing. SIAM Journal on Computing, 26, 1510–1523.
Bernstein, E., & Vazirani, U. (1997). Quantum complexity theory. SIAM Journal on Computing,
26, 1411–1473.
Biham, E., Brassard, G., Kenigsberg, D., & Mor, T. (2004). Quantum computing without entangle-
ment. Theoretical Computer Science, 320, 15–33.
Bohr, N. (1948). On the notions of causality and complementarity. Dialectica, 2, 312–319.
Boole, G. (1847). The mathematical analysis of logic. Philosophical Library. Reprinted: Thoemmes
Press (1998).
Borges, C. V. S., Hor-Meyll, M., Huguenin, J. A. O., & Khoury, A. Z. (2010). Bell-like inequality
for the spin-orbit separability of a laser beam. Physical Review A, 82, 033833.
Brassard, G., Cleve, R., & Tapp, A. (1999). Cost of exactly simulating quantum entanglement with
classical communication. Physical Review Letters, 83, 1874–1877.
Briegel, H. J., Browne, D. E., Dür, W., Raussendorf, R., & den Nest, M. V. (2009). Measurement-
based quantum computation. Nature Physics, 5, 19–26.
Bub, J. (2006). Quantum computation from a quantum logical perspective.
arXiv:quant-ph/0605243v2
Bub, J. (2008). Quantum computation and pseudotelepathic games. Philosophy of Science, 75,
458–472.
Bub, J. (2010). Quantum computation: Where does the speed-up come from? In A. Bokulich & G.
Jaeger (Eds.), Philosophy of quantum information and entanglement (pp. 231–246). Cambridge
University Press.
Bub, J. (2016). Bananaworld, quantum mechanics for primates. Oxford University Press.
Bub, J., & Pitowsky, I. (2010). Two dogmas about quantum mechanics. In S. Saunders, J. Barrett,
A. Kent, & D. Wallace (Eds.), Many worlds? Everett, quantum theory, and reality (pp. 433–459).
Oxford University Press.
Bub, T., & Bub, J. (2018). Totally random: Why nobody understands quantum mechanics. Princeton
University Press.
Church, A. (1936). An unsolvable problem of elementary number theory. American Journal of
Mathematics, 58, 345–363.
Cobham, A. (1965). The intrinsic computational difficulty of functions. In Y. Bar-Hillel (Ed.),
Logic, methodology and philosophy of science: Proceedings of the 1964 international congress
(pp. 24–30). North-Holland.
Cook, W. J. (2012). In pursuit of the traveling salesman: Mathematics at the limits of computation.
Princeton University Press.
Copeland, J. B. (2017). The modern history of computing. In E. N. Zalta (Ed.), The Stanford
encyclopedia of philosophy (Winter 2017 ed.). Metaphysics Research Lab, Stanford University.
Copeland, J. B. (2020). The Church-Turing thesis. In E. N. Zalta (Ed.), The Stanford encyclopedia
of philosophy (Summer 2020 ed.). Metaphysics Research Lab, Stanford University.
The Philosophy of Quantum Computing 149

Cuffaro, M. E. (2012). Many worlds, the cluster-state quantum computer, and the problem of the
preferred basis. Studies in History and Philosophy of Modern Physics, 43, 35–42.
Cuffaro, M. E. (2013). On the necessity of entanglement for the explanation of quantum speedup.
arXiv:1112.1347v5
Cuffaro, M. E. (2017). On the significance of the Gottesman-Knill theorem. The British Journal for
the Philosophy of Science, 68, 91–121.
Cuffaro, M. E. (2018). Reconsidering no-go-theorems from a practical perspective. The British
Journal for the Philosophy of Science, 69, 633–655.
Cuffaro, M. E. (2018). Universality, invariance, and the foundations of computational complexity
in the light of the quantum computer. In S. O. Hansson (Ed.), Technology and mathematics:
Philosophical and historical investigations. Springer.
Cuffaro, M. E. (2020). Information causality, the Tsirelson bound, and the ‘being-thus’ of things.
Studies in History and Philosophy of Modern Physics, 72, 266–277.
Cuffaro, M. E., & Doyle, E. P. (2021). Essay review of Tanya and Jeffrey Bub’s Totally random:
Why nobody understands quantum mechanics: A serious comic on entanglement. Foundations of
Physics, 51, 28:1–28:16.
Cuffaro, M. E., & Hartmann, S. (2021). The open systems view. arXiv:2112.11095v1
Curiel, E. (2014). Classical mechanics is Lagrangian; it is not Hamiltonian. The British Journal for
Philosophy of Science, 65, 269–321.
Dardashti, R. (2021). No-go theorems: What are they good for? Studies in History and Philosophy
of Science, 86, 47–55.
Davis, M. (2000). The universal computer: The road from Leibniz to Turing. W. W. Norton and
Company.
Dawid, R., & Thébault, K. P. Y. (2015). Many worlds: Decoherent or incoherent? Synthese, 192,
1559–1580.
Dawson, J. W., Jr. (2007). Classical logic’s coming of age. In D. Jacquette (Ed.), Philosophy of
logic (pp. 497–522). Elsevier.
Dean, W. (2016). Squeezing feasibility. In A. Beckmann, L. Bienvenu, & N. Jonoska (Eds.) Pursuit
of the Universal: Proceedings of the 12th Conference on Computability in Europe (pp. 78–88).
Springer International Publishing.
Deutsch, D. (1985). Quantum theory, the Church-Turing principle and the universal quantum com-
puter. Proceedings of the Royal Society of London. Series A. Mathematical and Physical Sciences,
400, 97–117.
Deutsch, D. (1989). Quantum computational networks. Proceedings of the Royal Society of London.
Series A. Mathematical and Physical Sciences, 425, 73–90.
Deutsch, D. (1997). The fabric of reality. Penguin.
Deutsch, D. (2010). Apart from universes. In S. Saunders, J. Barrett, A. Kent, & D. Wallace (Eds.),
Many worlds? Everett, quantum theory, and reality (pp. 542–552). Oxford University Press.
DeWitt, B. S. (1973 [1971]). The many-universes interpretation of quantum mechanics. In DeWitt
& Graham (1973, pp. 167–218).
Duwell, A. (2007). The many-worlds interpretation and quantum computation. Philosophy of Sci-
ence, 74, 1007–1018.
Duwell, A. (2018). How to make orthogonal positions parallel: Revisiting the quantum parallelism
thesis. In M. E. Cuffaro & S. C. Fletcher (Eds.), Physical perspectives on computation, compu-
tational perspectives on physics (pp. 83–102). Cambridge University Press.
Duwell, A. (2021). Computation and physics. Cambridge University Press.
Edmonds, J. (1965). Paths, trees, and flowers. Canadian Journal of Mathematics, 17, 449–467.
Einstein, A., Podolsky, B., & Rosen, N. (1935). Can quantum-mechanical description of physical
reality be considered complete? Physical Review, 47, 777–780.
Everett III, H. (1956). The theory of the universal wave function. In DeWitt & Graham (1973, pp.
3–140).
Farhi, E., Goldstone, J., Gutmann, S., & Sipser, M. (2000). Quantum computation by adiabatic
evolution. Tech. Rep. MIT-CTP-2936, MIT. arXiv:quant-ph/0001106
150 M. E. Cuffaro

Fletcher, S. C. (2018). Computers in abstraction/representation theory. Minds & Machines, 28,

445–463.
Forster, E. M. (1942). Virginia Woolf. The Rede Lecture 1941. Cambridge University Press.
Genovese, M. (2016). Experimental tests of Bell’s inequalities. In Bell & Gao (2016, pp. 124–140).
Gödel, K. (1956). Private letter to John von Neumann, 20 March 1956. Translated by Wensinger
in: Sipser (1992).
Goldreich, O. (2008). Computational complexity: A conceptual perspective. Cambridge University
Press.
Gottesman, D. (1999). The Heisenberg representation of quantum computers. In S. P. Corney, R. Del-
bourgo, & P. D. Jarvis (Eds.), Group22: Proceedings of the XXII International Colloquium on
Group Theoretical Methods in Physics (pp. 32–43). International Press. Longer version available
at: arXiv:quant-ph/9807006v1
Greaves, H., & Myrvold, W. (2010). Everett and evidence. In S. Saunders, J. Barrett, A. Kent, &
D. Wallace (Eds.), Many worlds? Everett, quantum theory, and reality (pp. 181–205). Oxford
University Press.
Grover, L. K. (1996). A fast quantum mechanical algorithm for database search. In Proceedings of
the Twenty-eighth Annual ACM Symposium on Theory of Computing, STOC ’96 (pp. 212–219).
Association for Computing Machinery.
Hagar, A. (2007). Quantum algorithms: Philosophical lessons. Minds & Machines, 17, 233–247.
Hagar, A., & Cuffaro, M. (2019). Quantum computing. In E. N. Zalta (Ed.), The Stanford encyclo-
pedia of philosophy (Winter 2019 ed.). Metaphysics Research Lab, Stanford University.
Heisenberg, W. (1971). Physics and beyond. Harper & Row.
Hewitt-Horsman, C. (2009). An introduction to many worlds in quantum computation. Foundations
of Physics, 39, 869–902.
Horsman, D., Kendon, V., & Stepney, S. (2018). Abstraction/representation theory and the natural
science of computation. In M. E. Cuffaro & S. C. Fletcher (Eds.), Physical perspectives on
computation, computational perspectives on physics (pp. 127–152). Cambridge University Press.
Howard, D. (1989). Holism, separability, and the metaphysical implications of the Bell experiments.
In J. T. Cushing & E. McMullin (Eds.), Philosophical consequences of quantum theory (pp. 224–
253). University of Notre Dame Press.
Hughes, R. I. G. (1989). The structure and interpretation of quantum mechanics. Harvard University
Press.
Janas, M., Cuffaro, M. E., & Janssen, M. (2022). Understanding quantum Raffles: Quantum mechan-
ics on an informational approach: Structure and interpretation. Springer.
Jozsa, R., & Linden, N. (2003). On the role of entanglement in quantum-computational speed-up.
Proceedings of the Royal Society of London. Series A. Mathematical, Physical and Engineering
Sciences, 459, 2011–2032.
Kent, A. (2005). Causal quantum theory and the collapse locality loophole. Physical Review A, 72,
012107.
Lee, K. F., & Thomas, J. E. (2002). Experimental simulation of two-particle quantum entanglement
using classical fields. Physical Review Letters, 88, 097902.
Lehner, C. (1997). What it feels like to be in a superposition. And why. Synthese, 110, 191–216.
Lenstra, A. K., Lenstra Jr., H. W., Manasse, M. S., & Pollard, J. M. (1990). The number field sieve.
In Proceedings of the Twenty-second Annual ACM Symposium on Theory of Computing, STOC
’90 (pp. 564–572). Association for Computing Machinery.
Lupacchini, R. (2018). Church’s thesis, Turing’s limits, and Deutsch’s principle. In M. E. Cuffaro
& S. C. Fletcher (Eds.), Physical perspectives on computation, computational perspectives on
physics (pp. 60–82). Cambridge University Press.
Mairhofer, L. (2021). Atom und Individuum: Bertolt Brechts Interferenz mit der Quantenphysik. De
Gruyter.
Maroney, O. J. E., & Timpson, C. G. (2018). How is there a physics of information? On characterising
physical evolution as information processing. In M. E. Cuffaro & S. C. Fletcher (Eds.), Physical
The Philosophy of Quantum Computing 151

perspectives on computation, computational perspectives on physics (pp. 103–126). Cambridge

University Press.
Martin, J. C. (1997). Introduction to languages and the theory of computation (2nd ed.). McGraw-
Hill.
Maudlin, T. (2011). Quantum non-locality and relativity (3rd ed.). Wiley-Blackwell.
Mehlhorn, K., & Sanders, P. (2008). Algorithms and data structures. Springer.
Mermin, D. N. (2007). Quantum computer science: An introduction. Cambridge University Press.
Miranda, E. R. (2021). Quantum computer: Hello, music! In E. R. Miranda (Ed.) Handbook of arti-
ficial intelligence for music: Foundations, advanced approaches, and developments for creativity.
Springer.
Myrvold, W., Genovese, M., & Shimony, A. (2020). Bell’s theorem. In E. N. Zalta (Ed.) The Stanford
encyclopedia of philosophy (Fall 2020 ed.). Metaphysics Research Lab, Stanford University.
Nielsen, M. A. (2006). Cluster-state quantum computation. Reports on Mathematical Physics, 57,
147–161.
Nielsen, M. A., & Chuang, I. L. (2000). Quantum computation and quantum information. Cambridge
University Press.
Nishimura, H., & Ozawa, M. (2009). Perfect computational equivalence between quantum Tur-
ing machines and finitely generated uniform quantum circuit families. Quantum Information
Processing, 8, 13–24.
Pearl, J. (2009). Causality: Models, reasoning, and inference (2nd ed.). Cambridge University Press.
Pitowsky, I. (1989). Quantum probability—Quantum logic. Springer.
Pitowsky, I. (1994). George Boole’s ‘conditions of possible experience’ and the quantum puzzle.
British Journal for the Philosophy of Science, 45, 99–125.
Pitowsky, I. (2002). Quantum speed-up of computations. Philosophy of Science, 69, S168–S177.
Raussendorf, R., & Briegel, H. J. (2002). Computational model underlying the one-way quantum
computer. Quantum Information and Computation, 2, 443–486.
Raussendorf, R., Browne, D. E., & Briegel, H. J. (2003). Measurement-based quantum computation
on cluster states. Physical Review A, 68, 022312.
Rivest, R. L., Shamir, A., & Adleman, L. (1978). A method for obtaining digital signatures and
public-key cryptosystems. Communications of the ACM, 21, 120–126.
Rosset, D., Branciard, C., Gisin, N., & Liang, Y.-C. (2013). Entangled states cannot be classically
simulated in generalized Bell experiments with quantum inputs. New Journal of Physics, 15,
053025.
Saunders, S. (1995). Time, quantum mechanics, and decoherence. Synthese, 102, 235–266.
Schrödinger, E. (1935a). Die gegenwärtige situation in der quantenmechanik. Naturwissenschaften,
23, 807–812; 823–828; 844–849. Translated in: Trimmer, J. D. (1980) Proceedings of the Amer-
ican Philosophical Society, 124, 323–338.
Schrödinger, E. (1935b). Discussion of probability relations between separated systems. Mathe-
matical Proceedings of the Cambridge Philosophical Society, 31, 555–563.
Shetterly, M. L. (2016). Hidden figures. HarperCollins.
Shin, S. W., Smith, G., Smolin, J. A., & Vazirani, U. (2014). How “quantum” is the D-Wave
machine? arXiv:1401.7087v2
Shor, P. W. (1994). Algorithms for quantum computation: Discrete logarithms and factoring. In
35th Annual Symposium on Foundations of Computer Science, 1994 Proceedings (pp. 124–134).
Tessier, T. E. (2004). Complementarity and entanglement in quantum information theory. Ph.D.
thesis, The University of New Mexico, Albuquerque, New Mexico.
Tessier, T. E., Caves, C. M., Deutsch, I. H., & Eastin, B. (2005). Optimal classical-communication-
assisted local model of n-qubit Greenberger-Horne-Zeilinger correlations. Physical Review A,
72, 032305.
Timpson, C. G. (2013). Quantum information theory & the foundations of quantum mechanics.
Oxford University Press.
Toner, B. F., & Bacon, D. (2003). Communication cost of simulating Bell correlations. Physical
Review Letters, 91, 187904.
152 M. E. Cuffaro

Turing, A. M. (1936–7). On computable numbers, with an application to the Entscheidungsproblem.

Proceedings of the London Mathematical Society. Second Series, s2-42, 230–265.
Vaidman, L. (1998). On schizophrenic experiences of the neutron or why we should believe in
the many-worlds interpretation of quantum theory. International Studies in the Philosophy of
Science, 12, 245–261.
Vaidman, L. (2012). Probability in the many-worlds interpretation of quantum mechanics. In Y.
Ben-Menahem & M. Hemmo (Eds.), Probability in physics (pp. 299–311). Springer.
Vaidman, L. (2018 [2002]). Many-worlds interpretation of quantum mechanics. In E. N. Zalta (Ed.)
The Stanford encyclopedia of philosophy (Fall 2018 ed.). Metaphysics Research Lab, Stanford
University. First published: 2002.
van Emde Boas, P. (1990). Machine models and simulations. Algorithms and Complexity. In J. van
Leeuwen (Ed.), Handbook of theoretical computer science (Vol. A, pp. 1–66). MIT Press/Elsevier.
Vincent, J. (1951). The diatonic modes in modern music. Mills Music. Page references to second
edition (1974) published by Curlew Music Publishers.
Wallace, D. (2003). Everett and structure. Studies in History and Philosophy of Modern Physics,
34, 87–105.
Wallace, D. (2007). Quantum probability from subjective likelihood: Improving on Deutsch’s proof
of the probability rule. Studies in History and Philosophy of Modern Physics, 38, 311–332.
Wallace, D. (2010). Decoherence and ontology. In S. Saunders, J. Barrett, A. Kent, & D. Wallace
(Eds.), Many worlds? Everett, quantum theory, and reality (pp. 53–72). Oxford University Press.
Wallace, D. (2012). The emergent multiverse. Oxford University Press.
Wallace, D. (2019). On the plurality of quantum theories: Quantum theory as a framework, and its
implications for the quantum measurement problem. In S. French & J. Saatsi (Eds.), Realism and
the quantum (pp. 78–102). Oxford University Press.
Wikipedia contributors. (2020). Instructions per second. In Wikipedia, The Free Encyclopedia.
Posted on 15 September, 2020, 11:15.
Zurek, W. H. (2003 [1991]). Decoherence and the transition from quantum to classical—revisited.
arXiv:quant-ph/0306072v1
Quantum-Like Cognition and
Rationality: Biological and Artificial
Intelligence Systems

Andrei Khrennikov

Abstract This is a short introductory review on quantum-like modeling of cogni-

tion with applications to decision-making and rationality. The aim of the review is
twofold: (a) to present briefly the apparatus of quantum information and probability
theory useful for such modeling; (b) to motivate applications of this apparatus in
cognitive studies and artificial intelligence, psychology, decision-making, social and
political sciences. We define quantum rationality as decision-making that is based on
quantum information processing. Quantumly and classically rational agents behaves
differently. A quantum-like agent can violate the Savage Sure Thing Principle, the
Aumann theorem on impossibility of agreeing to disagree. Such an agent violates the
basic laws of classical probability, e.g., the law of total probability and the Bayesian
probability inference. In some contexts, “irrational behavior” (from the viewpoint of
classical theory of rationality) can be profitable, especially for agents who are over-
loaded by a variety of information flows. Quantumly rational agents can save a lot
of information processing resources. At the same time, this sort of rationality is the
basis for quantum-like socio-political engineering, e.g., social laser. This rationality
plays the important role in the process of decision-making not only by biosystems, but
even by AI-systems. The latter equipped with quantum(-like) information processors
would behave irrationally, from the classical viewpoint. As for biosystems, quantum
rational behavior of AI-systems has its advantages and disadvantages. Finally, we
point out that quantum-like information processing in AI-systems can be based on
classical physical devices, e.g., classical digital or analog computers.

Keywords Mathematical formalism · Quantum mechanics · Open quantum

systems · Quantum instruments · Quantum Markov dynamics · Psychological
effects · Cognition · Decision-making · Classical rationality · Quantum
rationality · Information overload

A. Khrennikov (B)
International Center for Mathematical Modeling in Physics and Cognitive Sciences, Linnaeus
University, 351 95 Växjö, Sweden
e-mail: [email protected]

© Springer Nature Switzerland AG 2022 153

E. R. Miranda (ed.), Quantum Computing in the Arts and Humanities,
https://doi.org/10.1007/978-3-030-95538-0_4
154 A. Khrennikov

1 Introduction

‘Information’ is the new paradigm reflecting the main character of modern human
society as an information society. The tremendous development of information tech-
nologies of the last 20 years has dramatically changed our lifestyle through the World
Wide Web of the Internet and the mobile-connectivity web that have led to the cre-
ation of virtual social networks. Recent information technologies caused the digital
transformation of human communications and, consequently, of the whole society.
Unavoidably, challenges of the new information culture become the focus of inten-
sive studies and a meeting point for researchers not only from physics, information
engineering, and artificial intelligence, but also from biology, psychology, decision-
making, cognitive, social and political sciences, economics and finance. The aim of
this note to review briefly the recent applications of the quantum information and
probability theories outside of physics. This line of research is known as quantum-like
modeling (Asano et al., 2011, 2012, 2014, 2015a, 2015b, 2017a, 2017b; Basieva &
Khrennikov, 2015; Bagarello & Oliveri, 2013; Busemeyer & Bruza, 2012; Bagarello
et al., 2019; 2018; Basieva et al., 2017; Busemeyer et al., 2006; Boyer-Kassem et
al., 2015;Busemeyer et al., 2014; Dzhafarov & Kujala, 2012; Dzhafarov et al., 2015;
Haven, 2005; Haven & Khrennikov, 2013; Haven et al., 2017; Khrennikov, 1999,
2003, 2004a, 2004b, 2006, 2010, 2014, 2015a, 2015b, 2016; Khrennikov et al., 2014;
Khrennikova, 2016, 2017; Khrennikov & Basieva, 2014b,2014a; Khrennikov et al.,
2018; Ozawa & Khrennikov, 2020a, 2020b; Pothos & Busemeyer, 2009; Surov et al.,
2019; Wang & Busemeyer, 2013, 2014; White et al., 2014). This is not consideration
of genuine quantum physical processes in cognitive systems, but modeling behavior
of generally macroscopic cognitive or AI agents processing information on the basis
of quantum laws.
We want to present in more detail consequences of such information processing
for rationality. In classical decision-making, rational agents are mathematically mod-
eled as probabilistic information processors using Bayesian update of probabilities:
rational = Bayesian. Quantum state update is generally non-Bayesian (Ozawa &
Khrennikov, 2020b). We define quantum rationality as decision-making that is based
on quantum state update. Quantum and classical rational agents behave differently.
For instance, a quantum(-like) agent can violate the Savage Sure Thing Principle
(Savage, 1954) (see Busemeyer & Bruza, 2012; Haven & Khrennikov, 2013; Haven
et al., 2017) and the Aumann theorem (Aumann, 1976) on impossibility of agree-
ing to disagree (see Haven et al., 2017; Khrennikov & Basieva, 2015b; Khrennikov,
2014b).
In quantum-like modeling, the brain is treated as a black box in which informa-
tion processing cannot be described by classical probability (Kolmogorov, 1933)
(CP) (cf. with genuine quantum physical models of brain’s functioning (Arndt et
al., 2009; Bernroider, 2017; Bernroider & Summhammer, 2012; Hameroff, 1994;
Penrose, 1989; Umezawa, 1993; Vitiello, 1995, 2001)). And there is a plenty of
such nonclassical statistical data—in cognitive psychology, game theory, decision-
making, social science, economics, finances, and politics. In decision theory, such
Quantum-Like Cognition and Rationality: Biological and Artificial Intelligence Systems 155

data was coupled to probability fallacies and irrational behavior of agents. We pro-
pose to apply the most well-developed nonclassical theory of probability, namely,
based on the mathematical formalism of quantum theory.
One may think that the appeal to quantum probability (Khrennikov, 2016) (QP)
and information to model decision-making by humans is too exotic. However, we
recall that as early as the 1970s, Tversky (one of the most cited psychologists of all
time) and Kahneman (Nobel prize in economics in 2002, for prospect theory, which
he co-developed with Tversky) have been demonstrating cases where CP-prescription
and human behavior persistently diverge (Kahneman, 2003; Kahneman & Tversky,
1972, 1979, 1984; Kahneman & Thaler, 2006). Today, we are at the theoretical
cross-roads, with huge divisions across conflicting, entrenched theoretical positions.
Should scientists continue use CP as the basis for descriptive and normative
predictions in decision making and accept continuous generation of paradoxes?
Should we abandon probability theory completely and instead pursue explana-
tions based on heuristics, as Tversky and Kahneman proposed?
The use of the probabilistic and statistical methods is really the cornerstone of the
modern scientific methodology. Thus, although the heuristic approach to decision-
making cannot be discarded completely, it seems more natural to search novel proba-
bilistic models for decision-making. Our suggestion is to use QP and more generally
quantum information, instead of heuristics of Tversky and Kahneman.
We stress that quantum-like modeling does not appeal to genuine quantum physi-
cal processes in biosystems. Quantum-like information processing can be performed
by macroscopic systems, as cells, animals, or humans. Even artificial intelligence
need not be based on quantum physical processors, as, e.g., quantum computers.
Quantum-like modeling opens the door to creation of AI-systems processing infor-
mation by respecting the laws of quantum theory, but equipped with classical phys-
ical processors. Some step in this direction was done within the recent studies on
quantum(-like) information retrieval (see, e.g., Aerts et al., 2019; Melucci, 2015;
van Rijsbergen, 2004): algorithms based on the complex Hilbert state representation
of information, but driven on classical computers, demonstrate superiority compar-
ing with the traditional (“classical”) algorithms. Of course, creation of successfully
working genuine quantum computers and simulators would give the possibility for
creation of the genuine quantum AI-systems. However, since behavior of quantum-
like and genuine quantum AI-systems is based on the same formalism and method-
ology, the theory of quantum-like cognition, rationality, and artificial intelligence
would be useful even for quantum physical AI-systems.
For reader’s convenience, this paper contains basics of CP and QP. We present
them in two steps. Section 2 is the informal comparative introduction to CP versus
QP; then Sects. 7, 9 present briefly, but mathematically rigorously the CP and QP
formalisms (the later is presented as a part of the quantum formalism: quantum states
and observables, the Born rule for probability of outcome of a measurement, the
projection postulate and quantum state update). However, these formal presentations
of CP and QP are used only in the last part of this work devoted to comparison of
classical and quantum versions of the Aumann theorem. The preceding part devoted
to classical vs. quantum rationality is written at the heuristic level.
156 A. Khrennikov

2 Brief Comparison: Classical Versus Quantum Probability

CP was mathematically formalized by Kolmogorov (1933) (see Sect. 7 for details).

This is the calculus of probability measures, where a nonnegative weight p(A) is
assigned to any event A. The main property of CP is its additivity: if two events
O1 , O2 are disjoint, then the probability of disjunction of these events equals the
sum of probabilities:
P(O1 ∨ O2 ) = P(O1 ) + P(O2 ).

In fact, powerful integration theory that is needed for calculation of averages demands
σ-additivity: 
P(∪ j O j ) = P(O j ), (1)
j

where O j ∩ Oi = ∅, i = j.
QP is the calculus of complex amplitudes or in the abstract formalism complex
vectors. Thus, instead of operations on probability measures one operates with vec-
tors. We can say that QP is a vector model of probabilistic reasoning. Each complex
amplitude ψ gives the probability by Born’s rule: Probability is obtained as the
square of the absolute value of the complex amplitude.

p = |ψ|2 (2)

(for the Hilbert space formalization, see Sect. 9, Formula (17). By operating with
complex probability amplitudes, instead of the direct operation with probabilities,
one can violate the basic laws of CP.
In CP, the law of total probability (LTP) is derived by using additivity of probability
and the Bayes formula, the definition of conditional probability,

P(O2 ∩ O1 )
P(O2 |O1 ) = , P(O1 ) > 0. (3)
P(O1 )

Consider the pair, A and B, of discrete classical random variables. Then


P(B = β) = P(A = α)P(B = β|A = α).
α

Thus, in CP the B-probability distribution can be calculated from the A-probability

and the conditional probabilities P(B = β|A = α).
In QP (Khrennikov, 2016), classical LTP is perturbed by the interference term
(Khrennikov, 2010, 2016); for dichotomous quantum observables A and B of the
von Neumann-type, i.e., given by Hermitian operators  and B̂, the quantum version
of LTP has the form:
Quantum-Like Cognition and Rationality: Biological and Artificial Intelligence Systems 157

P(B = β) = P(A = α)P(B = β|a = α) (4)
α

 
+2 cos θα1 α2 P(A = α1 )P(B = β|A = α1 )P(A = α2 )P(B = β|a = α2 )
α1 <α2
(5)
If the interference term is positive, then the QP-calculus would generate a probability
that is larger than its CP-counterpart given by the classical LTP (2). In particular, this
probability amplification is the basis of the quantum computing supremacy.

3 Classical (Bayesian) Versus Quantum (Generally

Non-Bayesian) Rationality

In classical theory of decision-making, rational behavior of agents is formalized with

the Savage Sure Thing Principle (STP) (Savage, 1954):
If you prefer prospect b+ to prospect b− if a possible future event A happens (a =
+1); and you prefer prospect b+ still if future event A does not happen (a = −1);
then you should prefer prospect b+ , despite having no knowledge of whether or not
event A will happen.
Savage’s illustration refers to a person deciding whether or not to buy a certain
property shortly before a presidential election, the outcome of which could radically
affect the property market:
“Seeing that he would buy in either event, he decides that he should buy, even
though he does not know which event will obtain”.
STP is considered as the axiom of rationality of decision makers (Savage, 1954).
It plays the important role in decision-making and economics in the framework
of Savage’s subjective utility theory. In the latter, probability is formalized in the
classical probabilistic framework (Kolmogorov, 1933) and it is endowed with the
subjective interpretation.
We remark that STP is a simple consequence of the law of total probability—LTP
(see (5)). Violation of LTP implies violation of STP. Thus, the degree of satisfaction
of LTP can be used as a statistical test of classical (STP-type) rationality.
In cognitive psychology, violation of STP is known as the disjunction effect. A
plenty of statistical data was collected in cognitive psychology in experiments demon-
strating disjunction effect. For example, in experiments of the Prisoners’ Dilemma
type (Kahneman, 2003; Kahneman & Thaler, 2006; Kahneman & Tversky, 1972,
1979, 1984). Such data violate LTP. The latter implies irrationality (from classical
viewpoint) of agents participating in experiments (mainly students).
We recall that LTP is derived from two assumptions that are firmly incorporated
into the Kolmogorov axiomatics:
1. Additive law for probability.
2. Baeys formula for conditional probability.
158 A. Khrennikov

Therefore, violation of LTP and, hence, of STP (and classical rationality) is generated
either by violation of additivity of probability or the Bayes law for conditional proba-
bility or by the combination of these factors. Generally, this leads to the impossibility
to use in decision-making Bayesian inference. Quantum(-like) agents proceed with
more general inference machinery based on the quantum state update.
Hence, classical rationality is Bayesian inference rationality and quantum ratio-
nality is non-Bayesian inference rationality.1 In the light of above considerations,
one can ask:
Are quantum agents irrational?
As was discussed, by using QP it is possible to violate LTP and hence STP.
Therefore, generally quantum-like agents are (classically) irrational. However, we
can question the classical probabilistic approach to mathematical formalization of
decision making and, consequently, the corresponding notion of rationality. We define
quantum(-like) rationality as respecting the quantum calculus of probabilities and
the quantum formula for interference of probabilities, LTP with the interference term
(5).

4 Advantages, Disadvantages, and Roots of Quantum

Rationality

4.1 Liberalization of Decision-Making

Quantum rationality means more freedom for decision-making, liberalization of

this process. Generally, liberalization has its advantages and disadvantages. The
main advantage of quantum rationality is that such agents can come to essentially
larger spectrum of possible decisions than classically rational agents. Some quan-
tum decisions are classically unapproachable. One of the main disadvantage is that
decisions—events of decisions—can belong to incompatible decision algebras (σ-
algebras of events in Kolmogorov axiomatics of CP, see Sect. 7). In such a case, it is
impossible to come to consensus. We recall that in CP all possible decision events
are unified in one common event-algebra. They can always be joined consistently
with operations of conjunction, disjunction, and negation.
We remark that the CP-QP interplay is closely connected to the interplay of
classical and quantum logic. Classical logic operates in a single Boolean algebra.
Quantum logic operates in the lattice of orthogonal projectors in complex Hilbert
space. This lattice can be represented as union of partially overlapping Boolean
algebras representing compatible events given by commuting projectors. We recall
that measure-theoretic realization of Boolean algebras are precisely σ-algebras of
CP (Sect. 7). However, quantum logic is not just a collection of Boolean algebras.

1Of course, non-Bayesian probability updates are not reduced to quantum, given by state transfor-
mations in the complex Hilbert space. One may expect that human decision-making violates not
only classical, but even quantum rationality.
Quantum-Like Cognition and Rationality: Biological and Artificial Intelligence Systems 159

It is based on the consistent transition from one algebra to another—through unitary

transformations. The latter can be treated as change of logic coordinates.
By solving a problem, a cognitive system selects one of Boolean subalgebras of
quantum logic; typically as simple as possible. Why? We shall discuss this question
below.

4.2 Dysfunctional Disagreement and Information Overload

We point out that recently humans started to use widely the quantum decision-making
technique, especially in social and political decision-making. In particular, this can
explain the recent trend in increasing dysfunctional disagreement in, e.g., political
debates and generally in social life of the modern society. This disagreement cannot
be explained by insufficient information supply. There is a lot of information. The
problem is in opposite—in information overload (Khrennikov, 2020). The informa-
tion flows generated by mass-media and internet are so powerful, that by making a
decision on the concrete complex problem humans are not able to construct the joint
algebra for all possible events related to the problem under consideration. Different
issues involved in a problem are treated in different algebras. Moreover, these alge-
bras can differ essentially depending on agent’s social network and his information
environment (or it would better to say “a part of the information environment that is
selected by this agent”).
In the situation when an agent is overloaded with variety of information and
he or she should make rapidly the decision on a complex problem, the quantum
information processing (without attempting to unify all received information within
a single Boolean algebra) demonstrates its superiority over classical one, at least
from the viewpoint of minimization of computational resources and speed up of
decision-making. Thus, in the modern society quantum rationality beats classical
rationality, with “just one” casualty—inconsistency of some decisions. However,
this is inconsistency w.r.t. classical Boolean logic. If all agents behave quantumly,
then such inconsistencies become invisible.
The above discussion is applicable not only to biological systems, as say humans,
but also to AI-systems. The latter when operating with powerful information flows
would also prefer to use the quantum logic and QP. For each problem, a quantum(-
like) AI-system, say robot, selects a proper Boolean algebra, but at the same time
it keeps the possibility to use other Boolean algebras, corresponding to selection
of different orthonormal bases in the Hilbert state space. However, transition from
one Boolean algebra to another realized with a unitary operator U in the state space
demands a lot of computational resources, because state space has big dimension;
for n-qubit state space, it is 2n . In situation of information overload and temporal
constraints, an agent (biological or AI) has no possibility to perform such transition.
Moreover, to get more or less complete image of the situation, an agent has to make
transitions to a variety of Boolean algebras corresponding to incompatible variables
which are represented by noncommuting Hermitian operators.
160 A. Khrennikov

4.3 Social Laser

One of the consequences of information overload is that information loses its con-
tent. A human has no possibility to analyze deeply the content of communications
delivered by mass-media and social networks. People process information without
even attempting to construct an extended Boolean algebra of events. They oper-
ate with labels such as say COVID-19, vaccination, pandemy without trying to go
deeper beyond this labels. Contentless information behaves as a bosonic quantum
field which is similar to the quantum electromagnetic field. Interaction of humans
with such quantum information field can generate a variety of quantum-like behav-
ioral effects. One of them is social lasing, stimulated amplification of social actions
(SASA) (Khrennikov, 2015a, 2016, 2018; Khrennikov et al., 2018, 2019; Tsarev
et al., 2019). In social laser theory, humans play the role of atoms, social atoms
(s-atoms). Interaction of the information field composed of indistinguishable (up
to some parameters, as say social energy) excitations with gain medium composed
of s-atoms generate the cascade type process of emission of social actions. SASA
describes well, e.g., color revolutions and other types of mass protests (see Khren-
nikov, 2020 for detailed presentation).

5 Classical Versus Quantum Approach to the Problem

of Agreement on Disagree

Aumann’s approach (Aumann, 1976) to common knowledge and his theorem that
rational agents cannot agree on disagree play the crucial role in theory of rationality.
This theorem implies that rational agents acting under very natural conditions would
never agreeing on disagree. Hence, it couples rationality with consistency of actions
of decision makers. The main puzzle raised by Aumann’s theorem is that people
often “agree on disagree”.
How can we explain this contradiction between the statement which was mathe-
matically proved in the well-established framework of decision theory and the real
behavior of humans?
The typical solution is that one of the two basic assumptions of the Aumann
theorem is violated in the real processes of decision-making. We recall the Aumann’s
assumptions:
1. Common knowledge.
2. Common priors.
However, in modern information society these assumptions are very natural. Society
is homogeneous and the majority of people have common priors. Information is
openly distributed via mass-media and internet. Of course, there are attempts to
use insider information. But, such attempts, e.g., at the financial market, are subject
Quantum-Like Cognition and Rationality: Biological and Artificial Intelligence Systems 161

for punishment. Violations of the Aumann theorem cannot be reduced to insiders’

activity.
In my paper (Khrennikov & Basieva, 2014b) (see also Haven et al., 2017; Khren-
nikov, 2015b), it was pointed out that the Aumann theorem (Aumann, 1976) is also
based on the third, so to say hidden, assumption, namely, rationality of agents. As
was discussed in Sect. 3, classical notion of rationality is based on Bayesian infer-
ence in decision-making which is CP-formalized in scientific theory. If real agents
are not classically, but quantumly rational, they can violate the Aumann theorem,
even under the assumptions of common knowledge and common priors.
The basic QP-departure from the classical Aumann’s model is the existence of
incompatible information representations of the world by different agents. To model
this situation, instead of the set-theoretical (Boolean) partitions of the space of the
states of the world , we have to consider partitions of the unit operator in com-
plex Hilbert space H (space of the quantum states of the world) consisting of the
mutually orthogonal projectors. In general these partitions can be incompatible, i.e.,
the corresponding question-operators of different agents need not commute. Here,
we proceed with the simplest mathematical model of quantum measurements based
on the projection operators, measurements of the von Neumann-Lüders type (Von
Neumann-Luders, 1955). Generalization to measurements represented as quantum
instruments (Davies, 1976; Davies & Lewis, 1970; Okamura & Ozawa, 2016; Ozawa,
1984, 1997, 2004, Yuen, 1987) seems to be possible, but technically nontrivial.
In short, the main reason for this is that the basic quantum element introducing
violations into Aumann’s theorem is usage of a more general rule of updating of
probabilities, the quantum analog of the Bayesian updating, see Sect. 3 for the gen-
eral discussion. A few different sources of incompatibility are combined in this rule.
Besides the most evident source, namely, possible incompatibility of information
representations of agents (decision makers), two other sources also play important
roles. These are possible incompatibilities of information representations with quan-
tum events and common prior states. They both can contribute non-trivially into the
interference term perturbing the matching of posterior probabilities; even in contexts
with common prior states and nontrivial common knowledge (see Haven et al., 2017;
Khrennikov, 2015b for details).

6 Common Knowledge: Illustrative Examples

The notion of common knowledge plays the crucial role in various problems of coordi-
nation of actions and approaching conventions—in philosophy, economics (including
accounting and capital market research), game theory, statistics, computer science,
artificial intelligence. This notion is not simple and we shall devote this section to
the informal discussion, irrelevant to the use of CP or QP. For illustration, we shall
present the examples of the crucial impact of common knowledge in the problems
of social coordination. The first one is based on author’s personal experience and the
second one is commonly used in literature on common knowledge.
162 A. Khrennikov

We start with the remark that common knowledge is generalization of a simpler

notion mutual knowledge: everybody in a group of people is aware about some fact
or event. Now we present the first illustrative example for the notion of common
knowledge:
Example 1 (Earthquake, Japan, 11 March 2011 (Haven et al., 2017)) At that time
I was participating in Quantum Bio-information workshop at the Noda-city campus
of Tokyo University of Science, the fourth floor of the conference building. Around
15.00 the building started to shake, strongly shake.2 This shaking of the building and
that this is a sign of a strong earthquake was the mutual knowledge for the workshop
participants, everybody seen this. However, people did not try to escape from the
building until somebody loudly said: “this is a very strong earthquake.” Immediately
we sprung outside the building. This announcement of the fact known to everybody
in the conference room made the mutual knowledge common and this changed our
behavior crucially. After this announcement each workshop’s participant knew: “each
participant knows that each participant knows that earthquake is very strong.” and
so on, ad infinitum.
However, this was not the end of this story. After a half-hour staying at some
distance from the conference building and seeing that there were no more signs of
earthquake, we decided to return to the conference room and continue workshop. In
the middle of the session the building started to shake again and sufficiently strongly.
However, nobody said publicly that shaking is strong. Session continued. (May be
participants expected such an announcement from session’s chairman who may be
expected it from conference’s organizers.) My collaborator who was the last speaker
of this session told that she was really scared during her talk and very angry when
people started to ask questions after she finished.
Thus common knowledge is an essentially stronger assumption than simply
mutual knowledge. To have common knowledge means not only that everybody knows
some information E, but even that everybody knows that everybody knows E and
that everybody knows that everybody knows that everybody knows E and so on, ad
infinitum.
Remark. (Ad infinitum) The definition of common knowledge is based on the
infinite hierarchy of levels of knowing. This presence of infinity might make the
impression that this notion is not useful for concrete applications in which the infinite
level of commonality is inapproachable. However, this is not the case. Of course,
common knowledge is an ideal notion, but its role in science is similar to the role of
other ideal notions, such as, e.g., a point, straight line, irrational number. We cannot
proceed mathematically without such ideal notions. This is the exhibition of the
transcendental structure of human reasoning.
Example 2 (Blue eyes paradox (Friedell, 1969) There is an island populated by
people with blue and green eyes, say k people have blue eyes, others have green

2 Comment: earthquakes nearby Tokyo happen often; in some periods things in room shake practi-
cally everyday. Amplitudes vary day to day; of course each time I estimated their strength. But this
is difficult to do subjectively....
Quantum-Like Cognition and Rationality: Biological and Artificial Intelligence Systems 163

eyes. For island’s inhabitants, the number k is unknown. At the beginning nobody
knows the color of her/his eyes. There is the very strict rule:
If a person finds that she/he has blue eyes, that person must move from the island
before sunrise.
At this island everybody knows eye colors of others. But, it is forbidden to discuss
eye colors and there are no mirrors or similar devices. Once a stranger comes to the
island and announced to all the people:
“At least one of you has blue eyes”.
Thus it became common knowledge. The problem: what is the eventual outcome
of this public announcement?
Consider first the simplest case, k = 1. Some person will recognize that she/he
alone has blue eyes (by seeing only green eyes in the others) and leave at the first
sunrise. Let now k = 2. At the first sunrise nobody leaves the island. Then two people
having blue eyes by seeing only one person with blue eyes, and that no one left on
the first sunrise understand that k > 1. They leave at the second sunrise. And so on
by using the inductive argument. The paradox is that if k > 1 then the stranger told
to people at this island what they already have known: there are blue-eyed people.
However, without stranger’s announcement this fact was not common knowledge.
Its becoming common knowledge had dramatic consequences for inhabitants of the
island.
For k = 2, it is first-order knowledge. Each person having blue eyes knows that
there is someone with blue eyes, but she/he does not know that the other blue-eyed
person has this same knowledge. For k = 3, it is second order knowledge. After 2
days, each person having blue eyes knows that a second blue-eyed person knows that
a third person has blue eyes, but no one knows that there is a third blue-eyed person
with that knowledge, until the third day arrives. And so on...

7 Kolmogorov’s Model of Classical Probability Theory

Since CP-formalization of common knowledge is performed in the measure-theoretic

framework, it is useful to recall the Kolmogorov’s model (Kolmogorov, 1933) of CP.
The Kolmogorov probability space (Kolmogorov, 1933) is any triple

(, F, P),

where  is a set of any origin and F is a σ-algebra of its subsets, P is a probability

measure on F. The set  represents random parameters of the model. Kolmogorov
called elements of  elementary events. This terminology is standard in mathematical
literature. Sets of elementary events are regarded as events.
The essence of Kolmogorov’s approach is that not any subset of  can be treated
as an event. For any stochastic model, the system of events F is selected from the
very beginning. The main mathematical point is that F has to be a σ-algebra.
164 A. Khrennikov

We remind that a σ-algebra is a system of sets which contains  and empty set,
it is closed with respect to the operations of countable union and intersection and to
the operation of taking the complement of a set. For example, the collection of all
subsets of  is a σ-algebra. This σ-algebra is used in the case of finite or countable
sets:
 = {ω1 , ..., ωn , ...}. (6)

However, for “continuous sets”, e.g.,  = [a, b] ⊂ R, the collection of all possi-
ble subsets is too large to have applications. Typically it is impossible to describe a
σ-algebra in the direct terms. To define a σ-algebra, one starts with a simple system
of subsets of  and then consider the σ-algebra which is generated from this sim-
ple system with the aid of aforementioned operations. In particular, one of the most
important for applications σ-algebras, the so-called σBorel-algebra, is constructed
in this way by staring with the system consisting of all open and closed intervals of
the real line. In a metric space (in particular, in a Hilbert space), the Borel σ-algebra
is constructed by starting with the system of all open and closed balls.
Finally, we remark that in American literature the term σ-field is typically used,
instead of σ-algebra.
The probability is defined as a measure, i.e., a map from F to nonnegative real
numbers which is σ-additive:

P(∪ j A j ) = P(A j ), (7)
j

where A j ∈ F and Ai ∩ A j = ∅, i = j. The probability measure is always normal-

ized by one:
P() = 1. (8)

In the case of a discrete probability space, see (6), the probability measures have
the form 
P(A) = p j , p j = P({ω j }).
ω j ∈A

In fact, any finite measure μ, i.e, μ() < ∞, can be transformed into the probability
measure by normalization:

μ(A)
P(A) = , A ∈ F. (9)
μ()

A (real) random variable is a map ξ :  → R which is measurable with respect

to the Borel σ-algebra B of R and the σ-algebra F of . The latter means that,
for any set B ∈ B, its preimage ξ −1 (B) = {ω ∈  : ξ(ω) ∈ B} belongs to F. This
condition provides the possibility to assign the probability to the events of the type
“values of ξ belong to a (Borel) subset of the real line.” The probability distribution
of ξ is defined as
Quantum-Like Cognition and Rationality: Biological and Artificial Intelligence Systems 165

Pξ (B) = P(ξ −1 (B)). (10)

In the same way we define the real (and complex) vector-valued random variables,
ξ :  → Rn and ξ :  → Cn .
Let ξ1 , ..., ξk be real-valued random variables. Their join probability distribution
Pξ1 ,...,ξk is defined as the probability distribution of the vector-valued random vari-
able ξ = (ξ1 , ..., ξk ). To determine this probability measure, it is sufficient to define
probabilities

Pξ1 ,...,ξk (1 × ... × k ) = P(ω ∈  : ξ1 (ω) ∈ 1 , ...., ξk (ω) ∈ k )

where  j , j = 1, ..., k, are intervals (open, closed, half-open) of the real line.
We remark once again that LTP (5) is a theorem within Kolmogorov probability
theory (Kolmogorov, 1933). We also recall that LTP plays the basic role in Bayesian
probability inference.

8 Classical Formalization of Common Knowledge and

Aumann Theorem

8.1 States of the World

Aumann’s considerations are applicable to a finite number of agents, call them i =

1, 2, ..., N . These individuals are about to learn the answers to various multi-choice
questions, to make observations.
Mathematically the situation is represented with the aid of classical probability
space (based on the Kolmogorov axiomatics, 1933). Typically it is assumed that the
state space  representing all possible states of the world is finite.

8.2 Agents’ Information Representations

Each agent creates its information representation for possible states of the world
based on its own possibilities to perform measurements, “to ask questions to the
world.” Mathematically these representations are given by partitions of  : P (i) =
(P j(i) ), where, for each agent i,

∪ j P j(i) =  and P j(i) ∩ Pk(i) = ∅, j = k.

Thus an agent cannot get to know the state of the world ω precisely; she/he can
only get to know to which element of its information partition P j(i) ≡ P (i) (ω) this ω
belongs. In this set-theoretic model of knowledge, by definition the agent i knows
166 A. Khrennikov

an event E in the state of the world ω if the element of his information partition
containing this ω is contained in E :

P (i) (ω) ⊂ E. (11)

In logical terms, this can be written as P (i) (ω) ⇒ E, the event P (i) (ω) implies the
event E; we also remark that {ω} ⇒ P (i) (ω).

8.3 Common Prior

It is assumed that on  there is defined a probability p, the common prior of all

agents. In the accordance with the measure-theoretic model of probability theory
(Kolmogorov, 1933) there is given a σ-algebra, say F, of subsets of , its elements
represent events (“propositions” in some interpretations), and there is given a proba-
bility measure p defined on F. In the knowledge models it is typically assumed that
F is generated by agents’ partitions, i.e., this is the minimal σ-algebra containing
all systems of set P (i) , i = 1, ..., N . It is important to point out that, in particular,
such a σ-algebra contains all subsets of the form P j(1) 1
∩ ... ∩ P j(N
N
)
. Hence, in the
classical knowledge model the prior probability is assigned not only to the individ-
ual elements of agents’ information representations, i.e., P j(i) → p(P j(i) ), but even
to more complex events

P j(1)
1
∩ ... ∩ P j(N
N
)
→ p j1 ... jN ≡ p(P j(1)
1
∩ ... ∩ P j(N
N
)
). (12)

Thus by agreeing on the prior the agents have to agree on numerous conjunctive
probabilities.

8.4 CP-Formalization of the Notion of Common Knowledge

We consider the systems of sets P̃ (i) = {∪m P j(i) m

} consisting of finite unions of the
elements of the systems P and the system P̃ = ∩i P̃ (i) . A set O belongs to the
(i)

system P̃ if it belongs to any P̃ (i) . Thus, for each i, it can be represented as

O = ∪m P j(i)
m
, (13)

for some finite set of indexes (depending on i).

We now repeat the definition of common knowledge for two agents (and we
continue to proceed with two agents):
ACN An event E is common knowledge at the state of the world ω if 1 knows E,
2 knows E, 1 knows 2 knows E, 2 knows 1 knows E, and so on.
Quantum-Like Cognition and Rationality: Biological and Artificial Intelligence Systems 167

In theory of common knowledge, the basic role is played by the set of all states
of the world for which E is common knowledge; it is denoted by the symbol κE.
As was shown by Aumann (1976), this set of states of the world belongs to P̃ and,
hence, for each i, it can be represented (in the case κE = ∅) in the form (see (13)):

κE = ∪m P j(i)
m
. (14)

Let E be an event. For a state of the world ω, each agent i updates the common prior
p(E) on the basis of the observation the element P (i) (ω) of its information partition.
(For this agent, it means that the state of the world ω is contained in P (i) (ω).) This
update is given by the conditional probability

qi (ω) = p(E ∩ P (i) (ω))/ p(P (i) (ω)).

We remark that the conditional probability qi (ω) is defined to be the same for all
states of the world ω in a given element of partition. Thus, in fact,

qi (ω) ≡ qik ,

where ω ∈ Pk(i) = P (i) (ω).

8.5 Aumann Theorem

Now, Aumann’s theorem states that if both

q1 (ω) = q1 and q2 (ω) = q2 (15)

are common knowledge and prior probabilities are the same, then necessarily q1 =
q2 —simply because

qi = p(E|κCq1 q2 ) = p(E ∩ κCq1 q2 )/ p(κCq1 q2 ), (16)

where Cq1 q2 is the event (15): “the first agent by updating the prior probability of the
event E assigns the value q1 and the second agent the value q2 .”
168 A. Khrennikov

9 Basics of Quantum Formalism

9.1 States

Denote by H a complex Hilbert space. For simplicity, we assume that it is finite

dimensional. Pure states of a system S are given by normalized vectors of H and
mixed states by density operators (positive semi-definite operators with unit trace).

9.2 Observables, Born Rule for Probability, and Projection

Postulate

In the original quantum formalism (Von Neumann, 1955), physical observable A is

represented bya Hermitian operator Â. We consider only operators with discrete
spectra: Â = x x Ê A (x), where Ê A (x) is the projector onto the subspace of H
corresponding to the eigenvalue x. Suppose that system’s state is mathematically
represented by a density operator ρ. Then the probability to get the answer x is given
by the Born rule:

Pr{A = x ρ} = T r [ Ê A (x)ρ] = T r [ Ê A (x)ρ Ê A (x)] (17)

and according to the projection postulate the post-measurement state is obtained via
the state transformation:

Ê A (x)ρ Ê A (x)
ρ → ρx = . (18)
T r Ê A (x)ρ Ê A (x)

For reader’s convenience, we present these formulas for a pure initial state ψ ∈ H.
The Born’s rule has the form:

Pr{A = x ρ} = Ê A (x)ψ 2 = ψ| Ê A (x)ψ. (19)

The state transformation is given by the projection postulate:

ψ → ψx = Ê A (x)ψ/ Ê A (x)ψ . (20)

Here the observable-operator  (its spectral decomposition) uniquely determines the

feedback state transformations I A (x) for outcomes x

ρ → I A (x)ρ = Ê A (x)ρ Ê A (x). (21)

Quantum-Like Cognition and Rationality: Biological and Artificial Intelligence Systems 169

The map x → I A (x) given by (21) is the simplest (but very important) example
of quantum instrument (Davies, 1976; Davies & Lewis, 1970; Ozawa, 1984, 1997;
Yuen, 1987).

9.3 Quantum Logic

Following von Neumann (1955) and Birkhoff and von Neumann (1936) we represent
events, propositions, as orthogonal projectors in complex Hilbert space H.
For an orthogonal projector P, we set H P = P(H ), its image, and vice versa, for
subspace L of H, the corresponding orthogonal projector is denoted by the symbol
PL .
The set of orthogonal projectors is a lattice with the order structure: P ≤ Q
iff H P ⊂ HQ or equivalently, for any  ∈ H, |P ≤ |Q. This lattice is
known as quantum logic. Thus in classical Boolean logic events are represented by
sets and in quantum logic events are represented by orthogonal projectors.
We recall that the lattice of projectors is endowed with operations “and” (∧),
conjunction, and “or” (∨), disjunction. For two projectors P1 , P2 , the projector
R = P1 ∧ P2 is defined as the projector onto the subspace H R = H P1 ∩ H P2 and
the projector S = P1 ∨ P2 is defined as the projector onto the subspace H R defined
as the minimal linear subspace containing the set-theoretic union H P1 ∪ H P2 of sub-
spaces H P1 , H P2 : this is the space of all linear combinations of vectors belonging
these subspaces. The operation of negation is defined as the orthogonal complement:

P ⊥ = {y ∈ H : y|x = 0 : for all x ∈ H P }.

In the language of subspaces the operation “and”, conjunction, coincides with the
usual set-theoretic intersection, but the operations “or”, disjunction, and “not”, nega-
tion, are nontrivial deformations of the corresponding set-theoretic operations. It is
natural to expect that such deformations can induce deviations from classical Boolean
logic.
Consider the following simple example. Let H be two-dimensional
√ Hilbert space
with the orthonormal basis (e1 , e2 ) and let v = (e1 + e2 )/ 2. Then Pv ∧ Pe1 = 0 and
Pv ∧ Pe2 = 0, but Pv ∧ (Pe1 ∨ Pe2 ) = Pv . Hence, for quantum events, in general the
distributivity law is violated:

P ∧ (P1 ∨ P2 ) = (P ∧ P1 ) ∨ (P ∧ P2 ) (22)

As can be seen from our example, even mutual orthogonality of the events P1 and
P2 does not help to save the Boolean laws.
Thus quantum logic relaxes some constraints set by classical Boolean logic, in
particular, the distributivity between the operations of conjunction and disjunction.
170 A. Khrennikov

10 Quantum Formalization of Common Knowledge and

Aumann Theorem with Interference Term

10.1 Quantum States of the World

In our quantum-like model, the “states of the world” are given by pure states. Thus
the unit sphere S1 (H ) in a complex Hilbert space H represents (up to phase factors)
all possible states of the world.

10.2 Agents’ Quantum Information Representations

Questions posed by agents are mathematically described by Hermitian operators, say

A(i) . We state again that events (propositions) are identified with orthogonal projec-
tors. For the state of the world , an event P occurs (takes place with probability 1)
if  belongs to H P .
To simplify considerations, we proceed in the case of the finite dimensional state
space of the world, m = dim H < ∞. Each Hermitian operator can be represented
as a linear combination of orthogonal projectors to its eigen-subspaces; the questions
of agents can be expressed as

A(i) = a (i) (i)
j Pj , (23)
j

where (a (i) (i) (i)

j ) are real numbers, all different eigenvalues of A , and (P j ) are the
orthogonal projectors onto the corresponding eigen-subspaces. Here (a j ) encode
possible answers to the question of the ith agent.3 The system of projectors P (i) =
(P j(i) ) is the spectral family of A(i) . Hence, for any agent i, it is a “disjoint partition
of unity”:  (i)
Pk = I, Pk(i) ∧ Pm(i) = 0, k = m. (24)
k

We remark that (24) is simply the lattice-theoretical expression of the following

operator equalities:  (i)
Pk = I, Pk(i) Pm(i) = 0, k = m. (25)
k

3 Although in quantum physics the magnitudes of these numbers play an important role, in quantum
information theory the eigenvalues are merely formal labels encoding information which can be
extracted from a state with the aid of an observable. In the case of dichotomous answers, we can
simply use zero to encode “no” and one to encode “yes”.
Quantum-Like Cognition and Rationality: Biological and Artificial Intelligence Systems 171

This spectral family can be considered as information representation of the world

by the ith agent. In particular, “getting the answer a (i)
j ” is the event which is mathe-
(i)
matically described by the projector P j .
If the state of the world is represented by  and, for some k0 , P ≤ Pk(i)0
, then

p (Pk(i)
0
) = TrP Pk(i)0 = 1 and, for k = k0 , p (Pk(i) ) = TrP Pk(i) = 0.

Thus, in this case, the event Pk(i)

0
happens with the probability one and other events
from information representation of the world by the ith agent have zero probability.
However, opposite to the classical case, in general  need not belong to any
concrete subspace4 H P (i) . Nevertheless, for any pure state , there exists the mini-
k

mal projector Q (i)
 of the form m P j(i)
m
such that P ≤ Q (i)
 . This projector can be
constructed in the following way. Each state  determines the set of indexes

O(i) = { j : P j(i)  = 0}. (26)

Then the minimal projector majorating the one-dimensional projector corresponding

to the state  has the form:  (i)
Q (i)
 = Pj . (27)
j∈O(i)

The projector Q (i)

 represents the ith agent’s knowledge about the -world. We
remark that p (Q (i)
 ) = 1.
Consider the system of projectors P̃ (i) consisting of sums of the projectors from
P (i) :  (i)
P̃ (i) = {P = P jm }. (28)
m

Then
Q (i)
 = min{P ∈ P̃
(i)
: P ≤ P}, (29)

see (26), (27) for the constructive definition.

Thus in general, for the ith agent, the picture of the world is not only fuzzy (i.e.,
based on her partition of unity), but also probabilistic corresponding to the set of
probabilities, ( p (Pk(i) ), k = 1, 2, ...).

4We state again that in the classical probability model the states of the world are encoded by points
of . Take one fixed state ω. Since information representation of each agent is a partition of , for
each i there exists an element of partition, say P j(i) , containing this ω. For this state of the world,
the ith agent should definitely get the answer a (i) (i)
j corresponding the element P j . Thus any agent
is able to resolve uncertainty at least for her/his information representation (although she/he is not
able to completely resolve uncertainty about the state of the world). In the quantum case, an agent
is not able to resolve uncertainty even at the level of her/his information representation. And the
prior probability is updated in this uncertainty context.
172 A. Khrennikov

The ith agent picture of the world can be represented by the density operator
(mixed quantum state)
 
ρ(i) = Pk(i) P Pk(i) = p (Pk(i) )ρ(i)
k ,
k k

Pk(i) P Pk(i)
where ρ(i)
k = . Since each ρ(i)
k is a pure state, the ith picture of the world
TrPk(i) P Pk(i)
is given by the mixture of pure states, corresponding to “cells" Pk(i) .

10.3 Quantum Way of Common Knowledge Formalization

Definition 1 For the -state of the world and the event E, the ith agent knows E if

Q (i)
 ≤ E. (30)

It is evident that if, for the state of the world , the ith agent knows E, then
 ∈ HE . In general the latter does not imply that E is known (for the state ).5
However, if  ∈ E = P j(i) , then this event is known for i. The same is valid for any
event of the form E = P j(i)1
∨ ... ∨ P j(i)
k
(= P j(i)
1
+ ... + P j(i)
k
); if  ∈ HE , then such
E is known for i.
We remark that the straightforward analog of the classical definition, see (11),
would be based on condition P j(i) ≤ E for

P ≤ P j(i) , (31)

instead of more general condition (30). However, it would trivialize the class of
possible states of the world, because condition (31) is very restrictive.
We shall use the standard definition of common knowledge, see ACN, but based
on the quantum representation of knowing an event, see Definition 1. As in the
classical case, we have that “Where something is common knowledge, everybody
knows it.”
We recall that in the classical case, for each event E, there is considered the set
of all states of the world for which E is common knowledge. It is denoted by the
symbol κE.

5For example, the state space H is four dimensional with the orthonormal basis (e1 , e2 , e3 , e4 ), the
projectors P1 and P2 project H onto the subspaces with the bases (e1 , e2 ) and (e3 , e4 ), respectively.
Here (P1 , P2 ) is information representation of√an agent. Let E be the projector onto the subspace
with the basis (e1 , e4 ) and let  = (e1 + e4 )/ 2. Then Q  = I, the unit operator. Hence, E is not
known for this agent, although it belongs to H E .
Quantum-Like Cognition and Rationality: Biological and Artificial Intelligence Systems 173

This definition is naturally generalized to the quantum case. Here κE is defined

as the projector on the subspace consisting of all states of the world for which E is
common knowledge.
Similar to the set-theoretic framework, we introduce the system of projectors

P̃ = ∩i P̃ (i) .

We remark that (by definition) a projector P ∈ P̃ if and only if, for each i = 1, ..., N ,
it can be represented in the form

P= P j(i)
m
. (32)
m

Examples illustrating how this common knowledge structure works can be found
in Haven et al. (2017).
Now, before to formulate the quantum version of Aumann’s theorem, we recall that
while in the classical Aumann scheme the update of the prior probability distribution
on the basis of information representations of agents plays the crucial role. The
quantum analog of the Aumann scheme is based on the quantum procedure of the
state update as the result of measurement, in the simplest case this is the projection
postulate based update.

10.4 Quantum Version of Aumann’s Theorem

The quantum common prior assumption is formulated naturally: both agents assign
to possible states of the world the same quantum probability distribution given by
the density operator ρ, a priori state. The agents do not know exactly the real state
of the world which is always a pure state and in general a possible state of the world
appears for them as a mixed quantum state ρ.

Theorem 1 (Khrennikov & Basieva, 2014b) Under assumptions of common prior

and common knowledge, the following interference version of Aumann’s theorem
holds:  
1 
qi − q s = TrPk(i)j æPk(i)m E − TrPk(s)j æPk(s)m E . (33)
TræˇCq1 ...qN j=m j=m

If the amplitude of right-hand side of (33), the interference term between updates
of the probability of the event E by two different agents, is small, we can say that the
agents named i and s practically agree. The interference term can be considered as
a measure of “agreement on disagree” between the agents.
174 A. Khrennikov

11 Concluding Discussion

As was confirmed by plenty of experimental statistical data from cognitive psychol-

ogy, decision-making, social and political sciences, generally human agents make
their decisions by violating CP-laws (see introduction). It is natural to look for a
more general formalism to model cognition and decision-making. The QP-formalism
serves as the basis for such extension of CP-based approach. In this chapter, we dis-
cussed the consequences of the quantum-like model for rationality. Quantum ratio-
nality, i.e., based on quantum probability inference, differs crucially from classical
Bayesian rationality. It has its advantages and disadvantages; some of them were
discussed in this paper. The main reason for the use of quantum(-like) informa-
tion processing is information overload, especially in the situation demanding quick
decision-making under temporal and computation resource restrictions. Since the
modern information society is characterized by intensification of information flows
generated by mass-media, social networks, and working process, we can expect
increasing use of quantum information processing by human agents and, hence, the
phenomenon of quantum rationality will become more common. From the classical
rationality viewpoint, this kind of rationality can be viewed as irrationality. However,
it seems the only choice in the modern society.
The main danger of quantum rational behavior is that such agents become a very
good medium for social engineering; in particular, a good active medium for social
lasing. The latter can be used to generate instability throughout the world, in the form
of mass protests and color revolutions.
The quantum-like approach to modeling of cognition, decision-making, and ratio-
nality will definitely find applications in artificial intelligence, for two types of AI-
systems:
• Systems equipped with genuine quantum information processing devices, say
quantum computers or simulators.
• Systems equipped with classical information processing devices, say classical
digital or analog computers, realizing quantum(-like) information processing.
Personally I do not share the generally high expectation for successful realization
of genuine quantum physical computing project, especially hopes that such quantum
devices can be useful for AI-systems, say robots. I think that quantum information
processing based on classical computational devices has better perspectives. But,
since in science it is always difficult to make prognoses for future development,
both types of AI-systems, genuine quantum and quantum-like, have to be studied.
In future, the output of this paper may become useful for modeling rationality in
collectives composed of quantum and quantum-like robots and other AI-systems.
Quantum-Like Cognition and Rationality: Biological and Artificial Intelligence Systems 175

References

Aerts, D., Khrennikov, A., Melucci, M., & Toni, B. (Eds.). (2019). Quantum-like models for infor-
mation retrieval and decision-making. STEAM-H: Science, technology, engineering, agriculture,
mathematics & health. Springer.
Arndt, M., Juffmann, T., & Vedral, V. (2009). Quantum physics meets biology. HFSP Journal,3(6),
386–400. https://doi.org/10.2976/1.3244985
Asano, M., Basieva, I., Khrennikov, A., & Yamato, I. (2017a). A model of differentiation in quantum
bioinformatics. Progress in Biophysics and Molecular Biology,130, Part A, 88–98.
Asano, M., Basieva, I., Khrennikov, A., Ohya, M., Tanaka, Y., & Yamato, I. (2012). Quantum-
like model for the adaptive dynamics of the genetic regulation of E. coli’s metabolism of glu-
cose/lactose. Systems and Synthetic Biology, 6, 1–7.
Asano, M., Basieva, I., Khrennikov, A., Ohya, M., Tanaka, Y., & Yamato, I. (2015b). Quantum
information biology: From information interpretation of quantum mechanics to applications in
molecular biology and cognitive psychology. Foundations of Physics, 45(10), 1362–1378.
Asano, M., Khrennikov, A., Ohya, M., Tanaka, Y., & Yamato, I. (2014). Violation of contextual
generalization of the Leggett-Garg inequality for recognition of ambiguous figures. Physica
Scripta,T163, 014006.
Asano, M., Khrennikov, A., Ohya, M., Tanaka, Y., & Yamato, I. (2015a). Quantum adaptivity in
biology: From genetics to cognition. Springer.
Asano, M., Basieva, I., Khrennikov, A., Ohya, M., & Tanaka, Y. (2017b). A quantum-like model of
selection behavior. Journal of Mathematical Psychology, 78, 2–12.
Asano, M., Ohya, M., Tanaka, Y., Basieva, I., & Khrennikov, A. (2011). Quantum-like model of
brain’s functioning: Decision making from decoherence. Journal of Theoretical Biology, 281(1),
56–64.
Aumann, R. J. (1976). Agreeing on disagree. The Annals of Statistics,4, 1236–1239.
Bagarello, F. (2019). Quantum concepts in the social, ecological and biological sciences. Cambridge
University Press.
Bagarello, F., Basieva, I., Pothos, E. M., & Khrennikov, A. (2018). Quantum like modeling of deci-
sion making: Quantifying uncertainty with the aid of Heisenberg-Robertson inequality. Journal
of Mathematical Psychology, 84, 49–56.
Bagarello, F., & Oliveri, F. (2013). A phenomenological operator description of interactions between
populations with applications to migration. Mathematical Models and Methods in Applied Sci-
ences, 23(03), 471–492.
Basieva, I., & Khrennikov, A. (2015). On the possibility to combine the order effect with sequential
reproducibility for quantum measurements. Foundations of Physics, 45(10), 1379–1393.
Basieva, I., Pothos, E., Trueblood, J., Khrennikov, A., & Busemeyer, J. (2017). Quantum probability
updating from zero prior (by-passing Cromwell’s rule). Journal of Mathematical Psychology, 77,
58–69.
Bernroider, G. (2017). Neuroecology: Modeling neural systems and environments, from the quan-
tum to the classical level and the question of consciousness. Journal of Advanced Neuroscience
Research, 4, 1–9.
Bernroider, G., & Summhammer, J. (2012). Can quantum entanglement between ion transition
states effect action potential initiation? Cognitive Computation, 4, 29–37.
Birkhoff, J., & von Neumann, J. (1936). The logic of quantum mechanics. Annals of Mathematics,
37(4), 823–843.
Boyer-Kassem, T., Duchene, S., & Guerci, E. (2015). Quantum-like models cannot account for the
conjunction fallacy. Theory and Decision, 10, 1–32.
Busemeyer, J., & Bruza, P. (2012). Quantum models of cognition and decision. Cambridge Univer-
sity Press.
Busemeyer, J. R., Wang, Z., Khrennikov, A., & Basieva, I. (2014). Applying quantum principles to
psychology. Physica Scripta,T163, 014007.
176 A. Khrennikov

Busemeyer, J. R., Wang, Z., & Townsend, J. T. (2006). Quantum dynamics of human decision
making. Journal of Mathematical Psychology, 50, 220–241.
Davies, E. B. (1976). Quantum theory of open systems. Academic Press.
Davies, E. B., & Lewis, J. T. (1970). An operational approach to quantum probability. Communi-
cations in Mathematical Physics, 17, 239–260.
Dzhafarov, E. N., & Kujala, J. V. (2012). Selectivity in probabilistic causality: Where psychology
runs into quantum physics. Journal of Mathematical Psychology, 56, 54–63.
Dzhafarov, E. N., Zhang, R., & Kujala, J. V. (2015). Is there contextuality in behavioral and social
systems? Philosophical Transactions of the Royal Society A, 374, 20150099.
Friedell, M. (1969). On the structure of shared awareness. Behavioral Science, 14, 28–39.
Hameroff, S. (1994). Quantum coherence in microtubules. A neural basis for emergent conscious-
ness? Journal of Consciousness Studies, 1, 91–118.
Haven, E., & Khrennikov, A. (2013). Quantum social science. Cambridge University Press.
Haven, E., Khrennikov, A., & Robinson, T. R. (2017). Quantum Methods in Social Science: A First
Course. WSP.
Haven, E. (2005). Pilot-wave theory and financial option pricing. International Journal of Theoret-
ical Physics, 44(11), 1957–1962.
Kahneman, D., & Thaler., R. (2006). Utility maximization and experienced utility. Journal of
Economic Perspectives,20, 221–232.
Kahneman, D. (2003). Maps of bounded rationality: Psychology for behavioral economics. Amer-
ican Economic Review, 93(5), 1449–1475.
Kahneman, D., & Tversky, A. (1972). Subjective probability: A judgment of representativeness.
Cognitive Psychology, 3(3), 430–454.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econo-
metrica, 47, 263–291.
Kahneman, D., & Tversky, A. (1984). Choices, values and frames. American Psychologist, 39(4),
341–350.
Khrennikov, A. (2004b). Information dynamics in cognitive, psychological, social, and anomalous
phenomena. Ser.: Fundamental Theories of Physics.
Khrennikov, A. (2010). Ubiquitous quantum structure: From psychology to finances. Springer.
Khrennikov, A. (2015a). Quantum-like model of unconscious-conscious dynamics. Frontiers in
Psychology,6, Art. N 997.
Khrennikov, A. (2016). Probability and randomness: Quantum versus classical. Imperial College
Press.
Khrennikov, A. (2020). Social laser. Jenny Stanford Publ.
Khrennikov, A., & Basieva, I. (2014a). Quantum model for psychological measurements: From the
projection postulate to interference of mental observables represented as positive operator valued
measures. NeuroQuantology,12, 324–336.
Khrennikov, A., Alodjants, A. Trofimova., A., & Tsarev, D. (2018). On interpretational questions
for quantum-like modeling of social lasing. Entropy,20(12), 921.
Khrennikov, A., Basieva, I., Dzhafarov, E.N., Busemeyer, J. R. (2014). Quantum models for psy-
chological measurements: An unsolved problem. PLOS ONE, 9, Art. e110909.
Khrennikov, A., Basieva, I., Pothos, E. M., & Yamato, I. (2018). Quantum probability in decision
making from quantum information representation of neuronal states. Scientific Reports, 8. Article
number: 16225.
Khrennikov, A. (1999). Classical and quantum mechanics on information spaces with applications
to cognitive, psychological, social and anomalous phenomena. Foundations of Physics, 29, 1065–
1098.
Khrennikov, A. (2003). Quantum-like formalism for cognitive measurements. Biosystems, 70, 211–
233.
Khrennikov, A. (2004a). On quantum-like probabilistic structure of mental information. Open Sys-
tems and Information Dynamics, 11(3), 267–275.
Khrennikov, A. (2006). Quantum-like brain: Interference of minds. BioSystems, 84, 225–241.
Quantum-Like Cognition and Rationality: Biological and Artificial Intelligence Systems 177

Khrennikov, A. (2015a). Towards information lasers. Entropy, 17(10), 6969–6994.

Khrennikov, A. (2015b). Quantum version of Aumann’s approach to common knowledge: Sufficient
conditions of impossibility to agree on disagree. Journal of Mathematical Economics, 60, 89–104.
Khrennikov, A. (2016). Quantum Bayesianism as the basis of general theory of decision-making.
Philosophical Transactions of the Royal Society A, 374, 20150245.
Khrennikov, A. (2016). Social laser: Action amplification by stimulated emission of social energy.
Philosophical Transactions of the Royal Society, 374(2054), 20150094.
Khrennikov, A. (2018). Social laser model: From color revolutions to Brexit and election of Donald
Trump. Kybernetes, 47(2), 273–278.
Khrennikova, P. (2014). A quantum framework for ‘Sour Grapes’ in cognitive dissonance. In H.
Atmanspacher, E. Haven, K. Kitto, & D. Raine (Eds.), Quantum interaction. QI 2013. Lecture
Notes in Computer Science (Vol. 8369). Springer.
Khrennikova, P. (2016). Quantum dynamical modeling of competition and cooperation between
political parties: The coalition and non-coalition equilibrium model. Journal of Mathematical
Psychology, to be published.
Khrennikova, P. (2017). Modeling behavior of decision makers with the aid of algebra of qubit
creation-annihilation operators. Journal of Mathematical Psychology, 78, 76–85.
Khrennikov, A., & Basieva, I. (2014b). Possibility to agree on disagree from quantum information
and decision making. Journal of Mathematical Psychology, 62(3), 1–5.
Khrennikov, A., Toffano, Z., & Dubois, F. (2019). Concept of information laser: From quantum
theory to behavioural dynamics. The European Physical Journal Special Topics, 227(15–16),
2133–2153.
Kolmogorov, A. N. (1933). Grundbegriffe der Wahrscheinlichkeitsrechnung. Springer.
Melucci, M. (2015). Introduction to information retrieval and quantum mechanics. Springer.
Okamura, K., & Ozawa, M. (2016). Measurement theory in local quantum physics. Journal of
Mathematical Physics,57, 015209.
Ozawa, M. & Khrennikov, A. (2020a). Application of theory of quantum instruments to psychology:
Combination of question order effect with response replicability effect. Entropy,22(1), 37. 1-9436.
Ozawa, M. & Khrennikov, A. (2020b). Modeling combination of question order effect, response
replicability effect, and QQ-equality with quantum instruments. arXiv:2010.10444
Ozawa, M. (1984). Quantum measuring processes for continuous observables. Journal of Mathe-
matical Physics, 25, 79–87.
Ozawa, M. (1997). An operational approach to quantum state reduction. Annals of Physics (N.Y.),
259, 121–137.
Ozawa, M. (2004). Uncertainty relations for noise and disturbance in generalized quantum mea-
surements. Annals of Physics (N.Y.), 311, 350–416.
Penrose, R. (1989). The Emperor’s new mind. Oxford University Press.
Pothos, E., & Busemeyer, J. R. (2009). A quantum probability explanation for violations of ‘rational’
decision theory. Proceedings of Royal Society B, 276, 2171–2178.
Savage, L. J. (1954). The foundations of statistics. Wiley.
Surov, I. A., Pilkevich, S. V., Alodjants, A. P., & Khmelevsky, S. V. (2019). Quantum phase stability
in human cognition. Frontiers in Psychology, 10, 929.
Tsarev, D., Trofimova, A., Alodjants, A., et al. (2019). Phase transitions, collective emotions and
decision-making problem in heterogeneous social systems. Scientific Reports, 9, 18039.
Umezawa, H. (1993). Advanced field theory: Micro, macro and thermal concepts. AIP.
van Rijsbergen, C. J. (2004). The geometry of information retrieval. Cambridge University Press.
Vitiello, G. (2001). My double unveiled: The dissipative quantum model of brain. Advances in
Consciousness Research, John Benjamins Publishing Company.
Vitiello, G. (1995). Dissipation and memory capacity in the quantum brain model. International
Journal of Modern Physics B, 9, 973.
Von Neumann, J. (1955). Mathematical foundations of quantum mechanics. Princeton University
Press.
178 A. Khrennikov

Wang, Z., & Busemeyer, J. R. (2013). A quantum question order model supported by empirical
tests of an a priori and precise prediction. Topics in Cognitive Science, 5, 689–710.
Wang, Z., Solloway, T., Shiffrin, R. M., & Busemeyer, J. R. (2014). Context effects produced by
question orders reveal quantum nature of human judgments. PNAS, 111, 9431–9436.
White, L. C., Pothos, E. M., & Busemeyer, J. R. (2014). Sometimes it does hurt to ask: The
constructive role of articulating impressions. Cognition, 133(1), 48–64.
Yuen, H. P. (1987). Characterization and realization of general quantum measurements. In M.
Namiki et al. (Ed.), Proceedings of the 2nd International Symposium of Foundations of Quantum
Mechanics (pp. 360–363).
Quantum Music, Quantum Arts
and Their Perception

Volkmar Putz and Karl Svozil

Abstract The expression of human art, and supposedly sentient art in general, is
modulated by the available rendition, receiving and communication techniques. The
components or instruments of these techniques ultimately exhibit a physical, in par-
ticular, quantum layer, which in turn translates into physical and technological capac-
ities to comprehend and utilize what is possible in our universe. In this sense, we can
apply a sort of Church-Turing thesis to art, or at least to its rendition.

Keywords Quantum computer music · Quantum music theory · Music

perception · Church–Turing thesis · Bose and Fermi modelling

1 Realm of Quantum Expressibility

A short glance at historic practices of music and artistic expression, in general,

suggests that there has been, and still is, a fruitful exchange of ideas between crafts-
manship, technology, and (material) sciences on the one hand, and entertainment,
artistry, and creativity on the other hand. Impulses and ideas flow back and forth, very
much like in the accompanying fields of mathematics and natural sciences. This is
even true, in particular, for culinary subjects such as molecular gastronomy, where it
has been argued that “food processing dominates cuisines”: because even if all of the
French recipes would have been erased from people and other memories, most if not
all of these revered dishes could be “recovered” by merely following “reasonable”
rules of food processing This (2005)—which strongly are linked to technology, such
as the “domestication” of fire.

V. Putz (B)
Pädagogische Hochschule Wien, Grenzackerstraße 18, 1100 Vienna, Austria
e-mail: [email protected]
K. Svozil
Institute for Theoretical Physics, Vienna University of Technology, Wiedner Hauptstraße
8-10/136, 1040 Vienna, Austria
e-mail: [email protected]

© Springer Nature Switzerland AG 2022 179

E. R. Miranda (ed.), Quantum Computing in the Arts and Humanities,
https://doi.org/10.1007/978-3-030-95538-0_5
180 V. Putz and K. Svozil

It thus comes of no surprise that the evolution of quantum physics brought about
the quest for the quantum arts; and in particular, for quantum music Putz and Svozil
(2017) and quantum fine arts, especially quantum visual art. Indeed, every aspect of
human life can be re-evaluated and reframed in terms of the quantum paradigm.
In our (not so humble) opinion there are two immediate issues: One issue is the
tendency to re-introduce irrational “magic”, a sort of “quantum hocus pocus” Svozil
(2016) that brings it close to the esoteric, and fosters a kind of pseudo-religion
allegedly justified by the most advanced contemporary physics.
Another, converse, issue is the temptation to argue that, just like in quantum
computing (Mermin, 2007, Section 1.1), “any art is quantum” as the “underlying
physical layer” of any (classical) artistic expression is governed by the laws of quan-
tum mechanics. However, we emphasize upfront that we have to resist this temptation
towards a premature flattening and folding of the quantum phenomena into classi-
cal molds. Rather we consider quantum arts, and, in particular, quantum music, as
operations exploiting certain very special transformations of physical internal states,
subject to very carefully controlled conditions.
So what exactly are these very special transformations that characterize quantum
art? In this regard, we can proceed almost in parallel to the development of quantum
computation Fortnow (2003); Nielsen and Chuang (2010); Mermin (2007), and point
out some central assets or capacities:
(i) parallelization through coherent superposition (aka simultaneous linear com-
bination) of classically mutually exclusive tones or signals that are acoustic,
optic, touch, taste, or otherwise sensory;
(ii) entanglement not merely by classical correlation Peres (1978) but by relational
encoding Schrödinger (1935); Brukner and Zeilinger (1999); Zeilinger (1999);
Brukner et al. (2002) of multi-partite states such that any classical information is
“scrambled” into relational, joint multi-partite properties while at the same time
losing value definiteness about the single constituents of such multi-partite states
—this can be seen as a sort of zero-sum game, a tradeoff between individual
and collective properties;
(iii) complementarity associated with value (in)definiteness of certain tones or sig-
nals that is acoustic, optic, touch, taste, or otherwise: if one such observable is
definite, another is not, and vice versa;
(iv) contextuality is an “enhanced” form of complementarity and value indefinite-
ness that can be defined in various ways Dzhafarov et al. (2017); Abram-
sky (2018); Grangier (2002); Aufféves and Grangier (2018, 2020); Grang-
ier (2020); Budroni et al. (2021), in particular, emphasizing homomorphic,
structure-preserving nonembeddability into classical schemes Specker (1960);
Kochen and Specker (1967); Svozil (2021)
Those criteria or observables constitute significant signatures of quantum behav-
ior. The transformations and processing of classical-to-quantum states or quantum
states exhibiting these features can be considered musical, optical, or other instru-
ments or “transmitters” for the creation of quantum art. Similarly, assorted trans-
Quantum Music, Quantum Arts and Their Perception 181

Fig. 1 (Color online) Temporal succession of quantum tones |c , |d , . . ., |b  in the C major
scale forming the octave basis B

formations process quantum art. Finally, the process of information transmission

requires instruments of perception or “receivers” (Shannon, 1948, Fig. 1).
Let us mention typical components and theoretical entities as example trans-
formations. For instance, Hadamard transformations produce perfect “mixtures” of
classically mutually exclusive signals. Quantum Fourier transforms produce gen-
eralized mixtures. All of them have to be uniformly unitary—that is, in terms of
the various equivalent formal definitions, they have to transform orthonormal basis
into orthonormal ones, they have to preserve scalar products or norms, and their
inverse is the adjoint. One of the physical realizations is in terms of generalized
beam splitters Reck et al. (1994); Zukowski et al. (1997).
Depending on whether we are willing to contemplate genuine quantum receivers
or merely classical ones we end up with either a quantum cognition or with merely a
classical cognition of this quantum art; and, in particular, of quantum music. In the
first, radical deviation from classical music, we would have to accept the possibility
of human or sentient consciousness or audience to perceive quantum impressions.
This is ultimately a neurophysiologic question. It might well be that the pro-
cessing of signals exterior to the receiving and perceiving “somewhere along those
channels” requires a breakdown to classicality; most likely through the introduction
of stochasticity Glauber (1986). This is very much in the spirit of Schrödinger’s
cat Schrödinger (1935) and (later) quantum jellyfish Schrödinger (1995) metaphors
based on the assumption that, ultimately, even if decoherence by environmental
intake can be controlled, there cannot be any simultaneous co-experience of being
both dead and alive, just as there might not be any co-experience of passing into a
room by two separate doors simultaneously.
On the other hand, nesting of the Wigner’s friend type von Neumann (1996);
Everett (1957); Wigner (1995); Everett (2012), suggests that there might be substance
to a sort of mindful co-experience of two classical distinct experiences. Whether such
experiences remain on the subconscious primordial level of perception, or whether
this can be levied to a full cognitive level is a fascinating question on its own that
exceeds the limited scope of this article.
182 V. Putz and K. Svozil

2 Quantum Musical Tones

In what follows we closely follow our nomenclature and presentation of quantum

music Putz and Svozil (2017). Those formal choices are neither unique nor compre-
hensive. Alternatives are mentioned.
We consider a quantum octave in the C major scale, which classically consists
of the tones c, d, e, f , g, a, and b, represented by eight consecutive white keys
on a piano. (Other scales are straightforward.) At least three ways to quantize this
situation can be given:
(i) bundling octaves by coherent their superposition (aka simultaneous linear com-
bination), as well as
(ii) considering pseudo-field theoretic models treating notes as field modes that are
either bosonic or fermionic.
The seven tones c, d, e, f , g, a, and b of the octave can be considered as belong-
ing to disjoint events (maybe together with the null event 0) whose probabilities
should add up to unity. This essentially suggests a formalization by a seven (or eight)
dimensional Hilbert space C7 or C8 ) with the standard Euclidean scalar product. The
respective Hilbert space represents a full octave.
We shall study the seven-dimensional case C7 . The seven tones forming one octave
can then be represented as an orthonormal basis B of C7 by forming the set theoretical
union of the mutually orthogonal unit vectors; that is, B = {|c , |d , . . . |b },
where the basis elements are the Cartesian basis tuples
 
|c  = 0, 0, 0, 0, 0, 0, 1 ,
 
|d  = 0, 0, 0, 0, 0, 1, 0 ,
...
 
|b  = 1, 0, 0, 0, 0, 0, 0

of C7 . Figure 1 depicts the basis B by its elements, drawn in different colors.

2.1 Bundling Octaves into Single Tones

Pure quantum musical states could be represented as unit vectors |ψ ∈ C7 which
are linear combinations of the basis B; that is,

|ψ = αc |c  + αd |d  + · · · + αb |b , (1)

with coefficients αi satisfying

|αc |2 + |αd |2 + · · · + |αb |2 = 1. (2)

Quantum Music, Quantum Arts and Their Perception 183

Equivalent representations of |ψ are in terms of the one-dimensional subspace {|φ |

|φ = α|ψ, α ∈ C} spanned by |ψ, or by the projector Eψ = |ψψ|.
A musical “composition”—indeed, and any succession of quantized tones form-
ing a “melody”—would be obtained by successive unitary permutations of the state
|ψ. The realm of such compositions would be spanned by the succession of all uni-
tary transformations U : B → B mapping some orthonormal
 basis B into another
orthonormal basis B ; that is Schwinger (1960), U = i |  i i |.

2.2 Coherent Superposition of Tones as a New Form of

Musical Parallelism

One of the mind-boggling quantum field theoretic features of a “bundling” within sin-
gle modes is the possibility of the simultaneous “co-existence” of classically exclud-
ing musical states, such as a 50:50 quantum g in the
 C major scale obtained by sending
1 1  
|0g  through the Hadamard gate H = √2 1
, resulting in √12 |0g  − |1g  , and
1 −1
depicted in Fig. 2 by a 50 white 50 black; that is, gray, tone (though without the
relative “−” phase).
This novel form of musical expression might contribute to novel musical experi-
ences; in particular, if any such coherent superposition can be perceived by the audi-
ence in full quantum uniformity. This would require the cognition of the recipient
to experience quantum coherent superpositions—a capacity that is highly specula-
tive. It has been mentioned earlier that any such capacity is related to Schrödinger’s
cat Schrödinger (1935) and quantum jellyfish Schrödinger (1995) metaphors, as
well as to nestings of the Wigner’s friend type von Neumann (1996); Everett (1957);
Wigner (1995); Everett (2012).

2.3 Classical Perception of Quantum Musical Parallelism

In the following, we shall assume that quantum music is “reduced” to the continuous
infinity of its classical forms. Then, if a classical auditorium listens to the quan-
tum musical state |ψ in Eq. 1, the individual classical listeners may perceive |ψ

 
Fig. 2 Representation of a 50:50 quantum tone |g  = √1 |0g  − |1g  in gray (without indicating
2
phase factors)
184 V. Putz and K. Svozil

very differently; that is, they will hear only a single one of the different tones with
probabilities of |αc |2 , |αd |2 , . . ., and |αb |2 , respectively.
Indeed, suppose that classical recipients (aka “listeners”) modeled by classical
measurement devices acting as information-theoretic receivers are assumed. Then
any perception (aka “listening” or reception) of a quantum musical state that is
in a coherent superposition—with some coefficients 0 < |αi | < 1—because of the
supposedly irreducably stochastic Zeilinger (2005) quantum-to-classical transla-
tion Svozil (2004) represents an “irreducible” Peres (1980); Scully and Drühl (1982);
Greenberger and YaSin (1989); Scully et al. (1991); Zajonc et al. (1991); Kwiat et al.
(1992); Pfau et al. (1994); Chapman et al. (1995); Herzog et al. (1995) stochastic
measurement. This can never render a unique classical listening experience, as the
probability to hear the tone i is |αi |2 . Therefore, partitions of the audience will hear
different manifestations of the quantum musical composition made up of all varieties
of successions of tones. These experiences multiply and diverge as more tones are
perceived.
For the sake of a demonstration, let us try a two-note quantum composition. We
start with a pure quantum mechanical state in the two-dimensional subspace spanned
by |c  and |g , specified by
 
4 3 1 4
|ψ1  = |c  + |g  = . (3)
5 5 5 3

|ψ1  would be detected by the listener as c in 64% of all measurements (listenings),

 
01
and as g in 36% of all listenings. Using the unitary transformation X = , the
10
next quantum tone would be
 
3 4 1 3
|ψ2  = X|ψ1  = |c  + |g  = . (4)
5 5 5 4

This means for the quantum melody of both quantum tones |ψ1  and |ψ2  in
succession—for the score, see Fig. 3—that in repeated measurements, in 0.642 =
40.96% of all cases c − g is heard, in 0.362 = 12.96% of all cases g − c, in
0.64 · 0.36 = 23.04% of all cases c − c or g − g, respectively.

Fig. 3 (Color online) A two-note quantum musical composition—a natural fifth

Quantum Music, Quantum Arts and Their Perception 185

3 Quantum Musical Entanglement

Quantum entanglement Schrödinger (1935) is the property of multipartite quantum

systems to code information “across quanta” in such a way that the state of any
individual quantum remains irreducibly indeterminate; that is, not determined by
the entangled multipartite state Schrödinger (1935); Brukner and Zeilinger (1999);
Zeilinger (1999); Brukner et al. (2002). Thus the entangled whole should not be
thought of as composed of its proper parts. Formally, the composite state cannot be
expressed as a product of separate states of the individual quanta.
A typical example of an entangled state is the Bell state, | −  or, more generally,
states in the Bell basis spanned by the quantized notes e and a; that is

1
| ±  = √ (|0e |1a  ± |1e |0a ) ,
2
(5)
1
|±  = √ (|0e |0a  ± |1e |1a ) ,
2

A necessary and sufficient condition for entanglement among the quantized notes
e and a is that the coefficients α1 , α2 , α3 , α4 of their general composite state
|ga  = α1 |0e |0a  + α2 |0e |1a  + α3 |1e |0a  + α4 |1e |1a  obey α1 α4 = α2 α3
(Mermin, 2007, Sec. 1.5). This is clearly satisfied by Eqs. (5). Figure 4 depicts the
entangled musical Bell states.

−  and | +  in the first bar, and |−  and |+  in

Fig. 4 Quantum musical entangled states |ea ea ea ea
the second bar (without relative phases)

− +
Fig. 5 (Color online) Quantum musical entangled states for bundled octaves |ea   and |ea   in

the first bar, and |−

ea   and |+
ea   in the second bar (without relative phases)
186 V. Putz and K. Svozil

Entanglement between different octaves can be constructed similarly. Figure 5

depicts this configuration for an entanglement between e and a  .

4 Quantum Musical Complementarity and Contextuality

Although complementarity Pauli (1933) is mainly discussed in the context of observ-

ables, we can present it in the state formalism by observing that, as mentioned earlier,
any pure state |ψ corresponds to the projector Eψ = |ψψ|. In this way, any two non-
vanishing nonorthogonal and noncollinear states |ψ and |φ with 0 < |φ|ψ| < 1
are complementary. For the dichotomic field approach, Fig. 6 represents a configu-
ration of mutually complementary quantum tones for the note a in the C major scale
(a), and mutually complementary linear combinations as introduced in Sect. 2 (b).
Complementarity can be extended to more advanced configurations of contexts.
These quantum configurations and their associated quantum probability distributions,
if interpreted classically, either exhibit violations of classical probability theory,
classical predictions, or nonisomorphic embeddability of observables into classical
propositional structures Dzhafarov et al. (2017); Abramsky (2018); Grangier (2002);
Aufféves and Grangier (2018, 2020); Grangier (2020); Budroni et al. (2021); Specker
(1960); Kochen and Specker (1967); Svozil (2021).

5 Bose and Fermi Model of Tones

An alternative quantization to the one discussed earlier is in analogy to some

fermionic or bosonic—such as the electromagnetic—field. Just as the latter one
in quantum optics Glauber (1970, 2007) and quantum field theory Weinberg (1977)
is quantized by interpreting every single mode (determined, for the electromagnetic

Fig. 6 Temporal succession of complementary tones (a) for binary occupancy |φa  = αa |0a  +
βa |1a , with |αa |2 + |βa |2 = 1 with increasing |αa | (decreasing occupancy), (b) in the bundled
octave model, separated by bars
Quantum Music, Quantum Arts and Their Perception 187

Fig. 7 Temporal succession of tones |c , |d , . . ., |b  in an octave in the C major scale with
dicreasing mean occupancy

field for instance by a particular frequency and polarization) as a sort of “container”—

that is, by allowing the occupancy of that mode to be either empty or any positive
integer (and a coherent superposition thereof)—we obtain a vast realm of new musi-
cal expressions which cannot be understood in classical terms.
Whereas in a “bosonic field model” occupancy of field modes is easy to be corre-
lated with the classical volume of the corresponding tone, in what follows we shall
restrict ourselves to a sort of “fermionic field model” of music which is characterized
by a binary, dichotomic situation, in which every tone has either null or one occu-
pancy, represented by |0 = (0, 1) or |1 = (1, 0), respectively. Thus every state of
such a tone can thus be formally represented by entities of a two-dimensional Hilbert
space, C2 , with the Cartesian standard basis B = {|0, |1}.
Any note |i  of the octave consisting of |c , |d , . . ., |b , in the C major scale
can be represented by the coherent superposition of its null and one occupancies;
that is,
|i  = αi |0i  + βi |1i , (6)

with |αi |2 + |βi |2 = 1, αi .βi ∈ C.

Every tone is characterized by the two coefficients α and β, which in turn can
be represented (like all quantized two-dimensional systems) by a Bloch sphere, with
two angular parameters. If we restrict our attention (somewhat superficially) to real
Hilbert space R2 , then the unit circle, and thus a single angle ϕ, suffices for a char-
acterization of the coefficients α and β. Furthermore, we may very compactly notate
the mean occupancy of the notes by gray levels. Now, in this “fermionic setting”,
with the mean occupation number of any tone between 0 and 1 the gray level does
not indicate the volume of the corresponding tone but the mere chance of it being
present or not, see also Sect. 2. Figure 7 depicts a sequence of tones in an octave in
the C major scale with decreasing occupancy, indicated as gray levels.
In this case, any nonmonotonous unitary quantum musical evolution would have
to involve the interaction of different tones; that is, in a piano setting, across several
keys of the keyboard.
188 V. Putz and K. Svozil

6 Quantum Visual Arts

Just as for the performing arts such as music one could contemplate the quantum
options and varieties for the visual arts. Suffice it to say that the notion of “color”
experience can be extended to the full quantum optical varieties that result from
the electromagnetic field quantization, as already mentioned earlier. Incidentally,
Schrödinger published a series of papers on classical color perception Schrödinger
(1924); Schrödinger and Niall (2017) until around 1925. Yet to our best knowledge he
never considered the particular quantum aspects of human color and light perception.
Human rod cells respond to individual photons Hecht et al. (1942); Westheimer
(2016). Moreover, recent reports suggest that humans might be capable of “being
aware” of the detection of a single-photon incident on the cornea with a probability
significantly above chance Tinsley et al. (2016). It thus may be suspected that this
area of perception presents the most promising pathway into truly quantum percep-
tion. Speculations how this issue may be transferred to the perception of sound are
compelling.
Let us state up front that quantum visual art, and, in particular, quantum paral-
lelism, is not about additive color mixing, but it is about the simultaneous existence
of different, classically mutually exclusive “colors”, or visual impressions in general.
Quantum visual arts use the same central assets or capacities (i)–(iv) mentioned ear-
lier in Sect. 1. It can be developed very much in parallel to quantum music but requires
the creation of an entirely new nomenclature. The perception of quantum visual art is
subject to the same assumptions about the cognitive capacities to comprehend these
artifacts fully quantum mechanically or classically. This will be shortly discussed in
the following section.

7 Can Quantum Art Render Cognitions and Perceptions

Beyond Classical Art?

Suppose for a moment that humans are capable to sense, receive and perceive quan-
tum signals not only classically but in a fully quantum mechanical way. Thereby, they
would, for instance, be capable of simultaneously “holding” different classically dis-
tinct tones at once—not just by interference but by parallel co-existence. This would
result in a transgression of classical art forms, and in entirely novel forms of art.
The existence of such possibilities depends on the neurophysiology of the human,
or, more generally, sentient, perception apparatus. Presently the question as to
whether or not this is feasible is open; the answer to it is unknown.
In the case that merely classical perceptions are feasible, we would end up with
a sort of Church-Turing thesis for music. In particular, quantum music would not be
able to “go beyond” classical music for a single observer, as only classical renditions
could be perceived. Of course, as we mentioned earlier, quantum music might “sound
differently for different observers”. To this end, we might conceptualize a kind of
Quantum Music, Quantum Arts and Their Perception 189

universal musical instrument that is capable of rendering all possible classical notes.
Pianos and organs might be “for all practical purposes good” approximations to such
a universal device.
Quantum music and quantum arts, just like quantum computing Deutsch (1985),
or computations starting and ending in real numbers but using imaginary numbers
as intermediaries Musil (1906), might be a sort of bridge crossing elegantly a gap
between two classical domains of perception. And yet they could be so much more
if only the quantum could be “heard” or “sensed”.

8 Summary

We have contemplated the various extensions of music, and arts in general, to the
quantum domain. Thereby we have located particular capacities which are genuine
properties. These involve parallelization through coherent superposition (aka simul-
taneous linear combination), entanglement, complementarity and contextuality. We
have reviewed the nomenclature introduced previously Putz and Svozil (2017) and
considered particular instances of quantum music. Then we have briefly discussed
quantum visual arts.
The perception of quantum arts depends on the capacity of the audience to either
perceive quantum physical states as such, or reduce them to classical signals. In the
first case, this might give rise to entirely novel artistic experiences. We believe that
these are important issues that deserve further attention, also for sentient perception
in general and human neurophysiology, in particular.

Acknowledgements This research was funded in whole, or in part, by the Austrian Science Fund
(FWF), Project No. I 4579-N. For the purpose of open access, the author has applied a CC BY
public copyright licence to any Author Accepted Manuscript version arising from this submission.

References

This, H. (2005). Modelling dishes and exploring culinary ‘precisions’: the two issues of molecular
gastronomy. British Journal of Nutrition, 93, S139.
Putz, V., & Svozil, K. (2017). Quantum music. Soft Computing, 21, 1467. arXiv:1503.09045.
Svozil, K. (2016). Quantum hocus-pocus, Ethics in Science and Environmental Politics (ESEP).
16, 25. arXiv:1605.08569.
Mermin, D. N. (2007). Quantum Computer Science. Cambridge: Cambridge University Press.
Fortnow, L. (2003). One complexity theorist’s view of quantum computing. Theoretical Computer
Science, 292, 597.
Nielsen, M. A., & Chuang, I. L. (2010). Quantum Computation and Quantum Information. Cam-
bridge: Cambridge University Press. 10th Anniversary Edition.
Peres, A. (1978). Unperformed experiments have no results. American Journal of Physics, 46, 745.
Schrödinger, E. (1935). Die gegenwärtige Situation in der Quantenmechanik. Naturwissenschaften,
23, 807.
190 V. Putz and K. Svozil

Brukner, Č, & Zeilinger, A. (1999). Operationally invariant information in quantum measurements.
Physical Review Letters, 83, 3354. quant-ph/0005084.
Zeilinger, A. (1999). A foundational principle for quantum mechanics. Foundations of Physics, 29,
631.
Brukner, Č., Zukowski, M., & Zeilinger, A. (2002). The essence of entanglement.
arXiv:quant-ph/0106119, translated to Chinese by Qiang Zhang and Yond-de Zhang, New
Advances in Physics (Journal of the Chinese Physical Society).
Dzhafarov, E. N., Cervantes, V. H., & Kujala, J. V. (2017). Contextuality in canonical systems of
random variables. Philosophical Transactions of the Royal Society A: Mathematical, Physical
and Engineering Sciences, 375, 20160389. arXiv:1703.01252 .
Abramsky, S. (2018). Contextuality: At the borders of paradox. In E. Landry (Ed.) Categories for the
Working Philosopher (pp. 262–285). Oxford University Press, Oxford, UK. arXiv:2011.04899.
Grangier, P. (2002). Contextual objectivity: a realistic interpretation of quantum mechanics. Euro-
pean Journal of Physics, 23, 331. arXiv:quant-ph/0012122.
Aufféves, A., Grangier, P. (2018). Extracontextuality and extravalence in quantum mechanics. Philo-
sophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences,
376, 20170311. arXiv:1801.01398.
Auffèves, A., & Grangier, P. (2020). Deriving born’s rule from an inference to the best explanation.
Foundations of Physics, 50, 1781. arXiv:1910.13738.
Grangier, P. (2020). Completing the quantum formalism in a contextually objective framework.
arXiv:2003.03121.
Budroni, C., Cabello, A., Gühne, O., & Kleinmann, M. J. (2021). Quantum contextuality: Åke
Larsson. arXiv:2102.13036 [quant-ph].
Specker, E. (1960). Die Logik nicht gleichzeitig entscheidbarer Aussagen. Dialectica, 14, 239.
arXiv:1103.4537 .
Kochen, S., Specker, E. P. (1967). The problem of hidden variables in quantum mechanics. Journal
of Mathematics and Mechanics (now Indiana University Mathematics Journal), 17, 59.
Svozil, K. (2021). Varieties of contextuality emphasizing (non)embeddability. arXiv:2103.06110.
Shannon, C. E. (1949). Bell System Technical Journal 27, 379 (1948), reprinted in C. E. Shannon and
W. Weaver: The Mathematical Theory of Communication, University of Illinois Press, Urbana,
Illinois.
Reck, M., Zeilinger, A., Bernstein, H. J., & Bertani, P. (1994). Experimental realization of any
discrete unitary operator. Physical Review Letters, 73, 58.
Zukowski, M., Zeilinger, A., & Horne, M. A. (1997). Realizable higher-dimensional two-particle
entanglements via multiport beam splitters. Physical Review A, 55, 2564.
Glauber, R. J. (1986). Amplifiers, attenuators, and schrödinger’s cat. Annals of the New York
Academy of Sciences, 480, 336.
Schrödinger, E. (1995). The Interpretation of Quantum Mechanics. Dublin Seminars (1949-1955)
and Other Unpublished Essays. Woodbridge, Connecticut: Ox Bow Press.
von Neumann, J. (1932, 1996). Mathematische Grundlagen der Quantenmechanik, (2nd ed.). Berlin,
Heidelberg: Springer, English translation in [54].
Everett, H., III. (1957). Relative State formulation of quantum mechanics. Reviews of Modern
Physics, 29, 454.
Wigner, E. P. (1961, 1962, 1995) Remarks on the mind-body question. In I. J. Good (Ed.), The
Scientist Speculates (pp. 284–302). London, New York, and Berlin: Heinemann, Basic Books,
and Springer-Verlag.
Everett, H., III. (2012). The Everett Interpretation of Quantum Mechanics: Collected Works 1955–
1980 with Commentary, edited by J. A. Barrett & P. Byrne. Princeton, NJ: Princeton University
Press.
Schwinger, J. (1960). Unitary operators bases. Proceedings of the National Academy of Sciences
(PNAS), 46, 570.
Zeilinger, A. (2005). The message of the quantum. Nature, 438, 743.
Quantum Music, Quantum Arts and Their Perception 191

Svozil, K. (2004). Quantum information via state partitions and the context translation principle.
Journal of Modern Optics, 51, 811. arXiv:quant-ph/0308110.
Peres, A. (1980). Can we undo quantum measurements? Physical Review D, 22, 879.
Scully, M. O., & Drühl, K. (1982). Quantum eraser: A proposed photon correlation experiment
concerning observation and “delayed choice” in quantum mechanics. Physical Review A, 25,
2208.
Greenberger, D. M., & YaSin, A. (1989). “Haunted” measurements in quantum theory. Foundation
of Physics, 19, 679.
Scully, M. O., Englert, B.-G., & Walther, H. (1991). Quantum optical tests of complementarity.
Nature, 351, 111.
Zajonc, A. G., Wang, L. J., Zou, X. Y., & Mandel, L. (1991). Quantum eraser. Nature, 353, 507.
Kwiat, P. G., Steinberg, A. M., & Chiao, R. Y. (1992). Observation of a “quantum eraser”: a revival
of coherence in a two-photon interference experiment. Physical Review A, 45, 7729.
Pfau, T., Spälter, S., Kurtsiefer, C., Ekstrom, C. R., & Mlynek, J. (1994). Loss of spatial coherence
by a single spontaneous emission. Physical Review Letters, 73, 1223.
Chapman, M. S., Hammond, T. D., Lenef, A., Schmiedmayer, J., Rubenstein, R. A., Smith, E., &
Pritchard, D. E. (1995). Photon scattering from atoms in an atom interferometer: Coherence lost
and regained. Physical Review Letters, 75, 3783.
Herzog, T. J., Kwiat, P. G., Weinfurter, H., & Zeilinger, A. (1995). Complementarity and the quantum
eraser. Physical Review Letters, 75, 3034.
Pauli, W. (1933). Die allgemeinen Prinzipien der Wellenmechanik. In H. Geiger & K. Scheel (Ed.),
Handbuch der Physik (Vol. 24, p. 126). Berlin: Springer.
Glauber, R. J. (1969). Quantum theory of coherence. In S. M. Kay & A. Maitland (Ed.), Quan-
tum Optics: Proceedings of the Scottish Universities’ Summer School in Physics 1969. London:
Academic Press.
Glauber, R. J. (2007). Amplifiers, attenuators and Schrödingers cat. In Quantum Theory of Optical
Coherence (pp. 537–576). Wiley-VCH Verlag GmbH & Co. KGaA.
Weinberg, S. (1977). The search for unity: Notes for a history of quantum field theory. Daedalus,
106, 17.
Schrödinger, E. (1924). Über den Ursprung der Empfindlichkeitskurven des Auges. Die Naturwis-
senschaften, 12, 925.
Schrödinger, E., Niall, K. K. (2017). Erwin Schrödinger’s Color Theory. Springer International
Publishing.
Hecht, S., Shlaer, S., & Pirenne, M. H. (1942). Energy, quanta, and vision. Journal of General
Physiology, 25, 819.
Westheimer, G. (2016). History of physiological optics in the twentieth century, in Handbook of
Visual Optics: Fundamentals and Eye Optics, Volume One (Chap. 1, pp. 1–10). CRC Press, Taylor
& Francis Group.
Tinsley, J. N., Molodtsov, M. I., Prevedel, R., Wartmann, D., Espigulé-Pons, J., Lauwers, M., &
Vaziri, A. (2016). Direct detection of a single photon by humans. Nature Communications, 7.
https://doi.org/10.1038/ncomms12172.
Deutsch, D. (1985). Quantum theory, the Church-Turing principle and the universal quantum com-
puter. Proceedings of the Royal Society of London. Series A, Mathematical and Physical Sciences
(1934–1990), 400, 97.
Musil, R. (1906). Die Verwirrungen des Zöglings Törleß. Wien und Leipzig: Wiener Verlag, project
Gutenberg ebook # 3471.
von Neumann, J. (1955). Mathematical Foundations of Quantum Mechanics. Princeton, NJ: Prince-
ton University Press, German original in [27].
Quanta in Sound, the Sound of Quanta:
A Voice-Informed Quantum Theoretical
Perspective on Sound

Maria Mannone and Davide Rocchesso

Abstract Humans have a privileged, embodied way to explore the world of sounds,
through vocal imitation. The Quantum Vocal Theory of Sounds (QVTS) starts from
the assumption that any sound can be expressed and described as the evolution of a
superposition of vocal states, i.e., phonation, turbulence, and supraglottal myoelastic
vibrations. The postulates of quantum mechanics, with the notions of observable,
measurement, and time evolution of state, provide a model that can be used for
sound processing, in both directions of analysis and synthesis. QVTS can give a
quantum-theoretic explanation to some auditory streaming phenomena, eventually
leading to practical solutions of relevant sound-processing problems, or it can be
creatively exploited to manipulate superpositions of sonic elements. Perhaps more
importantly, QVTS may be a fertile ground to host a dialogue between physicists,
computer scientists, musicians, and sound designers, possibly giving us unheard
manifestations of human creativity.

voi ce
1 Sound ↔ Quanta

Sometimes, when kids imitate sounds around them, they are blamed for producing
weird noises. However, they are unknowingly using their own voice as a probe to
investigate the world of sounds, and thus they are probably performing some exper-
iments. Some of the these kids will become sound designers, other ones composers,
other sound engineers and physicists; some other ones, will blame future kids, and
the cycle repeats.

M. Mannone
European Centre for Living Technology (ECLT), Ca’ Foscari University of Venice, Venice, Italy
M. Mannone (B) · D. Rocchesso
Department of Mathematics and Computer Sciences, University of Palermo, Palermo, Italy
e-mail: [email protected]; [email protected]
D. Rocchesso
e-mail: [email protected]

© Springer Nature Switzerland AG 2022 193

E. R. Miranda (ed.), Quantum Computing in the Arts and Humanities,
https://doi.org/10.1007/978-3-030-95538-0_6
194 M. Mannone and D. Rocchesso

What is a vocal imitation? It is the attempt to reproduce some essential features

of a sound, thought of or actually heard, with the human voice. The imitation can
refer to characteristics of the sound, or to its hypothetical sources (Gaver 1993).
The human brain catches some salient sound features, and the voice attempts to
reproduce them. Sometimes, even poets (some of the kids above became poets as
well) coined new words as to include auditory dimensions in their poetry, producing
examples of onomatopoeia. This happened for example at the beginning of 20th
Century, with the poems by the futurist Filippo Tommaso Marinetti, where the words
Zang Tumb Tumb imitate motor and war noises (Marinetti 1914), and with La fontana
malata (The sick fountain) by Aldo Palazzeschi, where the words Clof, clop, cloch
(Palazzeschi 1909) mimic the intermittent flow of water and the noise of falling
drops.
Onomatopoeia gives poetical dignity to vocal imitations. Vocal imitations raised
the interest of science as well. In the framework of a recent European project, the
voice has been shown to be a powerful means to produce sound sketches, which can
be transformed into refined sound designs through interactive voice-driven sound
manipulations. In this sense, the machine extracts sound from the embodied imagi-
nation of the sound designer (Delle Monache et al. 2018). Tools of this kind, taking
the form of an augmented microphone, have been prototyped (Rocchesso et al. 2016)
and, with the purpose of turning the microphone into a music controller, even devel-
oped into products (Vochlea microphone 2022).
A few scholars may argue that there are precise tools to investigate sounds, such
as the Fourier formalism, which uses decompositions based on sinusoidal functions,
and all formalisms inspired by Fourier’s approach (von Helmholtz 1870; Koenig
and Delwin 2015). However, this formalism is not immediately understandable in
everyday communications, and it is less directly manipulable than vocal imitations.
While the Fourier formalism is powerful at the level of persons with some education
in sound and music, it is not the way laypersons communicate and reason about sonic
realities.
A powerful support to both qualitative and quantitative communication and rea-
soning on sound is given by sound visualizations. Spectrograms display the spectrum
of frequencies through time of a sound. With spectrograms, we can easily compare
sounds and investigate how their properties change through time. However, the vocal
imitations of a natural or an artificial sound, which appears as completely intuitive
to humans (again, think about the kid giving voice to a toy car), might be hard to find
by comparison of the spectrogram of the vocal imitation with the spectrogram of the
original sound. It can be possible to investigate some emerging properties of both
sounds, but it can be really hard while dealing, for example, with a vocal imitation of
a motor, or some other mechanical noise, that has a really different spectral profile
than human voice. Thus, Fourier-driven sound visualization has some limitations in
revealing the embodied perceptual features of sounds.
Which are the characteristics of human voice? The utterances of humans and many
mammals can be decomposed into overlapping chunks that fall within three primitive
classes: phonation, turbulence, and supraglottal myoelastic vibrations (Friberg et al.
2018). In phonation, the source is in the vocal folds. In turbulence, the source is in
Quanta in Sound, the Sound of Quanta 195

chaotic motion of inhaled or exhaled air. Supraglottal myoelastic vibrations include

several kinds of low-frequency oscillations or pulse trains generated with different
parts of the vocal apparatus, such as lips or tongue. We can build up a new formalism
to describe the sound based on these components.
And what are the characteristics of sound as it is produced out of our body?
Sound is made of waves of rarefaction and compression, produced by vibrating
strings, air-filled pipes, vibrating membranes or plates, and so on. Consider the
simplest of these systems, which is probably the flexible string fastened to a rigid
support at both ends. This is one of the most important models in physics, which has
been used to demonstrate fundamental phenomena, in acoustics as well as in other
areas of physical sciences. In fact, whilst vibrating strings have often been used as a
paradigm for quantum mechanics, the vice versa, that is, using quantum mechanics
as a paradigm to understand sound, was proposed in the nineteen-forties by Gabor
(1947), who imagined how sound analysis and synthesis could be based on acoustical
quanta, or wavelets.1 His seminal work has been extensively carried on and expanded
both by scientists and musicians, and is certainly at the root of granular approaches
to sound and music (Roads 2001).
A variety of ideas and methods of quantum mechanics have been applied to
describe forms and phenomena pertaining to that form of art whose medium is
sound, that is music. For example, tonal attractions have been modeled as metaphor-
ical forces (Blutner and beim Graben 2020), quantum parallelism has been proposed
to model music cognition (Dalla Chiara et al. 2015), the quantum formalism has
been proposed as a notational tool for music-visual “atomic” elements (Mannone
et al. 2020), the non-Markovianity of open quantum systems has been proposed
as a measure of musical memory within a score (Mannone and Compagno 2014).
Quantum computing, that is computation based on actual physical quantum pro-
cesses (Nielsen 2010), starts being used to control sound synthesizers and computer-
generated music (Miranda 2021a, b). The opposite practice, that is using sonification
as a means to make the actions of quantum algorithms perceivable as musical lines
and recognizable as patterns, has appeared with the flourishing of quantum comput-
ing as an area of theoretical computer science (Weimer 2010).
This chapter is part of a book on quantum-theoretical and -computational
approaches to art and humanites, and its first chapter provides an excellent intro-
duction to quantum theory and quantum computing. Nevertheless, we give a few
basic notions, essentially the postulates of quantum mechanics, in Sect. 1.1. For now,
we can say that quantum mechanics is a branch of physics, where, in a nutshell:
• Matter and energy, seen at the level of subatomic particles, are described as discrete;
• We describe particles as points or as probability waves to find them in some places;
• If we know the momentum of a particle, we don’t know its position, and vice versa;
• The measurement influences the state of what is measured: the observer (subject)
influences the observed (object).

1 A wavelet is a wave-like oscillation under a finite temporal envelope.

196 M. Mannone and D. Rocchesso

In 1935, Albert Einstein tried to resist to quantum mechanics, postulating hidden

variables to justify such a bizarre behavior (Einstein et al. 1935). Some further studies
showed that, if Einstein was right, some inequalities should be satisfied (Bell 1964),
but quantum-mechanical systems can be conceived and implemented that actually
violate such inequalities (Aspect et al. 1982). This means that a local realistic view
of the world does not apply to quantum phenomena. According to another Nobel
prize, Richard Feynman, nobody really understood quantum mechanics (Feynman
1995).
One might ask: if quantum mechanics is so difficult to be interpreted and under-
stood, why is it so often invoked to explain mundane affairs that have nothing to
do with particle physics? As a possible explanation, the formalism is based on a
few postulates, it assumes linearity and unitary (energy-preserving) time evolution,
and it gives a probabilistic framework capable to explain concurrent and interfering
phenomena.
Let us go back to sound and voice. If quantum mechanics can be joined with the
sound, and the sound with the voice, thus quantum mechanics can be joined with
the voice, and this is our idea: a Quantum Vocal Theory of Sound (Rocchesso and
Mannone 2020).
This approach is not opposed to the richness and complexity of Fourier formalism,
spectrograms, and so on. It presents a different paradigm, a different starting point,
using the primitive components of human voice. The novelty is that these components
appear as useful not only to investigate the voice itself, but also to face the complexity
of the world of general sound. It is a strong statement, but it actually follows the
intuition: each kid knows well how to imitate the vroom vroom of a car, a long time
before learning how to read an equation and how to interpret a graph.
The QVTS approach can be exploited to investigate sound, decomposing it into its
essential features through the analysis step. But QVTS can also help do the opposite,
that is, create new sounds, in the synthesis step. Sound synthesis can lead to creative
applications; some possible applications are described later on in this chapter.
The structure of the chapter is the following. In Sect. 1.1, we remind of some
basics of quantum mechanics. In Sect. 2, we present the fundamental ideas of QVTS.
In Sect. 3, examples of sound processing based on the QVTS, with audible and
interpretable outcomes, are given. In Sect. 4, we describe our vision on the future of
QVTS keeping an eye on interdisciplinary collaborations and creative applications.
As an example of possible creative applications, we sketch the structure of a piece
based on vocal states.

1.1 Some Postulates to Live by

An observable is a physical quantity, that can be described as a mathematical (Hermi-

tian, linear) operator. Each operator acts on a complex vector space, the state space.
The space where quantum observables live is the separable Hilbert space. It’s sepa-
Quanta in Sound, the Sound of Quanta 197

rable, because we can distinguish the components along different axes. In the case
of QVTS, the space is related with the vocal primitives, and it is separable as well,
as we will see in the next section.
A quantum state, that is a unit-length vector in the state space, can be seen as
a superposition of values with some probabilities. An eigeinstate is a characteristic
state of some operator. After the measurement process, the probability wave collapses
to a certain value: it is the eigenvalue of a certain operator, and the system is in an
eigenstate of that operator.
Probability is a key concept in quantum mechanics. According to the principle of
uncertainty, we cannot know, let’s say, the position and the momentum of a particle
with the same precision. The more is the information we have on position, the less
we know about momentum, and vice versa.
Let us say more on the idea of quantum measurement. Consider a Cartesian frame-
work with axes x, y, and z—a tridimensional space with three mutually orthogonal
axes. We can perform measurements along each of the axes. Quantum measure
implies a change in the measured entity. If the measurement along the direction x
can give a positive result, in all subsequent measurements along the same direction
we will have a positive value. A measure along x would zero out the y and z compo-
nents, while leaving only the x component with value 1. If, before measurement, there
is a given probability to get a specific outcome, after the effective measurement of
that outcome, the probability to get the same value in each subsequent measurement
along the same direction is 100%. In fact, in quantum mechanics, the measurement of
a state implies the destruction of part of the initial information, and thus the process is
called destructive measure. A quantum state is a superposition of eigenstates, which
are reduced to a single state after the measurement. Such state collapse happens in the
context including both the system and the measuring entity, through the interaction
of the two (Rovelli 2021). Intuitively, it’s like observing and taking a picture of a
person, and blocking him or her as the represented image along that specific shooting
direction. (Be careful the next time you’ll take pictures).
Dennis Gabor first exploited the paradigm of quantum theory to investigate
sound (Gabor 1947), instead of doing the usual vice versa, with sound and strings used
as metaphors to understand quantum waves. Gabor proposed the concept of quantum
of sound, as a unit-area cell in the time-frequency plane, which could be called phon,
from the Greek fwn». On the other hand, we start from a vocal description of sound,
to define the phon as the set of vocal primitive operators.

2 The Quantum Vocal Theory of Sound

In a recent article (Rocchesso and Mannone 2020), we have proposed the basics for
a Quantum Vocal Theory of Sound (QVTS). Here, we summarize its main ideas, and
then we propose some hints for future developments.
First of all, let us define the phon formalism, where the word phon indicates the
quantum of sound, expressed in the state space of vocal primitives. With the phon
198 M. Mannone and D. Rocchesso

Fig. 1 The Bloch sphere

adapted to QVTS, to
represent the phon space.
Hand-drawing by
D. Rocchesso

formalism, we can define vocal states, and extend the quantum bit (qubit) language
to the human voice. Some quantum-mechanical concepts, such as state preparation
and measurement, can be extended to the domain of voice as well.
Consider a space with three independent directions: x, y, and z. In the QVTS, the
three axes of this “phonetic space” have a vocal meaning:
• z is the phonation, giving waveforms of different pitches;
• x is the turbulence, giving noises of different brightnesses;
• y is the myoelasticity, giving pulsations at different tempos (thought of as slow
pulse trains).
Such three-dimensional space is sketched in Fig. 1 where, at the intersections between
each of the axes and the unit (called Bloch) sphere, we can find two mutually orthog-
onal vectors, each depicted as a tiny sketchy spectrogram.
Given a sound recording of human voice, if we measure phonation using a specific
computational tool (such as SMS-tools Bonada et al. 2011), it is possible to separate
such component from the rest, and all subsequent measurements of phonation will be
giving the same result. If we measure a primitive component first, and then another
one, the result is generally dependent on the order of the two operations: A fact that
is known as non-commutativity. Figure 5 shows a couple of example spectrograms
illustrating the difference.
A vocal state can be described as a superposition of phonation, turbulence, and
myoelasticity with certain probabilities. We can thus define a phon operator σ as a
3-vector operator, providing information on the x, y, and z components through its
specific directions in the 3d phonetic space. Each component of σ is represented by
a linear operator, so we have σx , σ y , and σz .
Quanta in Sound, the Sound of Quanta 199

2.1 Preparation and Measurement Along Axes

According to the postulates of quantum mechanics, it is possible to perform mea-

surements along one arbitrary axis of the 3d phonetic space and, as a result, we will
have prepared the phon along that specific axis.
A quantum measurement is represented by an operator, called a projector, acting
on the state, and provoking its collapse onto one of its eigenvectors. If the system is
in a state |ψ and then we make a measurement, the probability to get the result j is
given by:
p j := pm ( j|ψ) = ψ| M j |ψ = ψ| M †j M j |ψ , (1)

where the set {M j } is a projector system in the Hilbert space of states. {M j } is

a complete set of Hermitian and idempotent matrices. An Hermitian matrix is a
complex matrix, that is equal to its transposed conjugate (indicated by the † symbol
in Eq. 1). It has real eigenvalues. Idempotent means that, if we apply multiple times
an operator, the result is the same as if we applied the operator just once. With
an orthonormal basis of measurement vectors |a j , the elementary projectors are
M j = |a j a j |, and the system collapses into |a j .

2.1.1 Measurement Along Z

A measurement along the z axis is performed through the operator σz . The eigenvec-
tors (or eigenstates) of σz are |u and |d, corresponding to pitch-up phonation and
pitch-down phonation, with eigenvalues λu = +1 and λd = −1, respectively:

σz |u = |u , σz |d = − |d .

The eigenstates |u and |d are orthogonal, i.e., u|d = 0, and they can be represented
as column vectors    
1 0
|u = , |d = . (2)
0 1

The operator σz can also be represented in matrix form as

 
1 0
σz = . (3)
0 −1

Applying a measurement along the z direction to a generic phon state |ψ corre-
sponds to pre-multiply it by one of the measurement operators (or projectors)
 
10
Mu = |u u| =
00
200 M. Mannone and D. Rocchesso

or  
00
Md = |d d| = ,
01

and to normalize the resulting vector to have length one. Such operators satisfy the
completeness relation Mu + Md = I , summing up to the unit operator.
A generic phon state |ψ can be expressed as

|ψ = αu |u + αd |d , (4)

where the coefficients are complex numbers, αu = u|ψ, and αd = d|ψ. Being
the system in state |ψ, the probability to measure pitch-up is

pu = ψ| Mu† Mu |ψ = ψ|u u|u u|ψ = ψ|u u|ψ = αu∗ αu (5)

and, similarly, the probability to measure pitch-down is pd = ψ|d d|ψ = αd∗ αd ,

where ∗ denotes complex conjugation. The completeness relation ensures that pu
and pd sum up to one.
If we repeatedly prepare a state ψ and measure it along the z direction, we get the
average value
⎛ ⎞
 
σz  := λm pm = ψ| ⎝ λm Mm† Mm ⎠ |ψ = ψ|σz |ψ , (6)
m={u,d} m={u,d}

where the sum within brackets is called the observable of the measurement.
In quantum computing terminology, the vectors2 (2) give the computational basis
of a qubit vector space. The operator (3) corresponds to a Z gate, which acts as a
phase flip on the second state of the computational basis.

2.1.2 Measurement Along X

The eigenstates of the operator σx are |r  and |l, corresponding to turbulent primitive
sounds having different spectral distributions, one with the rightmost (or highest-
frequency) centroid and the other with the lowest-frequency centroid. Their respec-
tive eigenvalues are λr = +1 and λl = −1, such that

σx |r  = |r  , σx |l = − |l .

2 In quantum computing, the vectors of the computational basis are normally called |0 and |1.
Quanta in Sound, the Sound of Quanta 201

If the phon is prepared |r  (turbulent) and then the measurement apparatus is set
to measure σz , there will be equal probabilities of getting pitch-up or pitch-down
phonation as an outcome. This measurement property is satisfied if |r  is defined as

1 1
|r  = √ |u + √ |d . (7)
2 2

A similar definition is given for |l, such that the two eigenstates of turbulence are
orthogonal (r |l = 0):
1 1
|l = √ |u − √ |d . (8)
2 2

In matrix form, the turbulence operator is expressed as

 
01
σx = , (9)
10

and its eigenvectors are

√1 √1
|r  = 2 , |l = 2 . (10)
√1
2
− √12

Applying a measurement along the x direction to a generic phon state |ψ corre-
sponds to pre-multiply it by one of the measurement operators
 
1 11
Mr = |r  r | =
2 11
or  
1 1 −1
Ml = |l l| = ,
2 −1 1

and to normalize the resulting vector to have length one. Such operators satisfy the
completeness relation Mr + Ml = I .
In quantum computing, the operator (9) corresponds to a X gate, which is the
equivalent of the NOT gate in classical logic circuits, as it flips the states of the
computational basis. The vectors (7) and (8) form the Hadamard basis, often denoted
with the symbols {|+ , |−}.
Preparation in one of the states of the Hadamard basis {|r  , |l}, followed by
measurement along the z axis, results in an operation that is equivalent to coin
flipping, +1 or −1 being obtained with equal probability.
202 M. Mannone and D. Rocchesso

2.1.3 Measurement Along Y

The eigenstates of the operator σ y are | f  and |s, corresponding to slow myoelastic
pulsations, one faster and one slower,3 with eigenvalues λu = +1 and λd = −1, such
that
σ y | f  = | f  σ y |s = − |s .

If the phon is prepared | f  (pulsating) and then the measurement apparatus is set to
measure σz , there will be equal probabilities for |u or |d phonation as an outcome.
This measurement property is satisfied if

1 i
| f  = √ |u + √ |d , (11)
2 2

where i is the imaginary unit.

Likewise, the |s state can be defined in such a way that the two eigenstates of
pulsation are orthogonal ( f |s = 0):

1 i
|s = √ |u − √ |d . (12)
2 2

In matrix form, the pulsation operator is expressed as

 
0 −i
σy = , (13)
i 0

and its eigenvectors are

√1 √1
|f = 2 , |s = 2 .
√i
2
− √i 2

Applying a measurement along the y direction to a generic phon state |ψ corre-
sponds to pre-multiply it by one of the measurement operators
 
1 1 −i
Mf = | f  f | =
2 i 1
or  
1 1 i
Ms = |s s| = ,
2 −i 1

and to normalize the resulting vector to have length one. Such operators satisfy the
completeness relation M f + Ms = I .

3In describing the spin eigenstates, the symbols |i and |o are often used, to denote the in–out
direction.
Quanta in Sound, the Sound of Quanta 203

The matrices (3), (9), and (13) are called the Pauli matrices. In quantum computing,
these are all useful one-qubit gates.

2.2 Measurement Along an Arbitrary Direction

Orientingthe measurement

 apparatus in the phonetic space along an arbitrary direc-
tion n = n x , n y , n z means taking a weighted mixture of Pauli matrices:
 
nz n x − in y
σn = σ · n = σx n x + σ y n y + σz n z = . (14)
n x + in y −n z

2.2.1 Sines + Models and the Phon Space

The Harmonic plus Noise model (Bonada et al. 2011) is well suited to describe
measurement and preparation in the phonation-turbulance planar section of the 3d
phonetic space. An arbitrary direction in such plane is described by the operator
 
cos θ sin θ
σn = , (15)
sin θ − cos θ

where θ is the angular direction, pointing to a superposition of phonation and turbu-

lence (see Fig. 1). The eigenstate for eigenvalue +1 is

|λ1  = [cos θ/2, sin θ/2] , (16)

the eigenstate for eigenvalue −1 is

|λ−1  = [− sin θ/2, cos θ/2] , (17)

and the two are orthogonal. Suppose we prepare the phon to pitch-up |u. If we rotate
the measurement system along n, the probability to measure +1 is

p(+1) = u|λ1  λ1 |u = |u|λ1 |2 = cos2 θ/2, (18)

and the probability to measure −1 is

p(−1) = |u|λ−1 |2 = sin2 θ/2. (19)

The expectation (or average) value of measurement is therefore


σn  = λ j p(λ j ) = (+1) cos2 θ/2 + (−1) sin2 θ/2 = cos θ. (20)
j
204 M. Mannone and D. Rocchesso

Fig. 2 Spectrogram of a vocal sound which is a superposition of phonation and supgraglottal

myoelastic vibration. A salient pitch (horizontal dotted line) as well as quasi-regular train of pulses
(vertical dotted lines) are automatically extracted

Symmetrically, if we prepare the phon in state |λ1  and we measure along the z
axis, we get a pitch-up with probability cos2 θ/2 and a pitch-down with probability
sin2 θ/2.
More generally, the Sines plus Noise plus Transients model (Verma et al. 1997)
may be suitable to describe measurement and preparation in the whole 3d phonetic
space, where supraglottal myoelastic vibrations are made to correspond to transient
pulse trains. For example, consider the vocal fragment4 whose spectrogram is repre-
sented in Fig. 2. An extractor of pitch salience and an extractor of onsets5 have been
applied to highlight respectively the phonation (horizontal dotted line) and myoelas-
tic (vertical dotted lines) components in the spectrogram. In the z − y plane, there
would be a measurement orientation and a measurement operator that admit such
sound as an eigenstate.

2.3 Purity and Mixing

In quantum mechanics, the density operator is a mathematical object that describes

the statistical (pure or mixed) state of a quantum system, and it is usually represented
as a matrix. A pure state is not referred to a moral condition, but to a separability
of states. A mixed state indicates an inseparability of states from the viewpoint of

4 It is one of the example vocal sounds considered in Rocchesso et al. (2016), and taken from New-
man (2004).
5 The feature extractors are found in the Essentia library (Bogdanov et al. 2013).
Quanta in Sound, the Sound of Quanta 205

the observer, who has some degree of epistemic uncertainty. Thus, the concept of
density matrix generalizes the concept of state superposition. The density operator
is defined as 
ρ= p j |ψ j  ψ j | , (21)
j

where p j indicates the probability for the j-state. The density operator for a pure
state is ρ = |ψψ|, and the trace of the associated density matrix is tr[ρ 2 ] = 1. For a
mixed state, tr[ρ 2 ] < 1. It can be shown that the density matrix (21) can be expressed
as a composition of Pauli matrices as in (14), with the addition of the identity matrix.
From such representation, pure states can be shown to lay on the surface of the Bloch
sphere, while mixed states stay inside the sphere, with the completely chaotic state
being found at the origin (Cariolaro 2015). A pure state can contain a superposition,
but such a composition is defined with certainty. A mixed state is a probabilistic
mixing. The mixed state is inseparable. The generalization introduced by mixed
states can represent the audio concept of mixing, thus coming useful in composition
of auditory scenes.

2.4 Not Too Sure? Uncertainty Can Be Measured

In the wonderland of quantum mechanics, it can happen that, the better we know
something, the lesser we know something else. In QVTS, the more precise our
knowledge of phonation, the less precise our measurement of turbulence. In quantum
mechanics, if we measure two observables L and M simultaneously in a single
experiment, the system is left in a simultaneous eigenvector of the observables only
if L and M commute, i.e., if their commutator [L, M] = LM − ML vanishes. When
the commutator is different from zero, we say that the two operators do not commute.
This
 happens
 with measurement operators along different axes. It is the case of
σx , σ y = 2iσz . As a consequence for QVTS, phonation and turbulence cannot be
simultaneously measured with certainty.
The uncertainty principle is one of the key ideas of quantum mechanics. It is
based on Cauchy–Schwarz inequality in complex vector spaces. According to the
uncertainty principle, the product of the two uncertainties is at least as large as half
the magnitude of the commutator:

1
LM ≥ |ψ| [L, M] |ψ| (22)
2

Let L = T = t be the time operator, and M = W = −i dtd be the frequency oper-

ator. Applying them to the complex oscillator Aeiωt , we get a time-frequency uncer-
tainty, where the uncertainty is minimized by the Gabor function, a sinusoid win-
dowed by a Gaussian (Irino and Patterson 1997).
206 M. Mannone and D. Rocchesso

2.4.1 The Order Matters

Kids learn that multiplying a times b, with a, b natural numbers, is the same as
multiplying b times a—and, since early age, they think that commutativity is always
verified. Reading a book and then going for a walk might be the same stuff as going
for a walk and then reading a book (maybe). Quantum mechanics does not work that
way, and the same for QVTS. If we record a singer, we take away vowels, and then we
take again away vowels, the result is the same—the recording is in an autostate of no-
vowels. If we take away vowels, and then we take away the noise, the result is different
from what we could hear if we do the opposite, that is, taking away the noise and
then the vowels. More precisely, the measurement operators oriented along different
axes do not commute. For example, let A be an audio segment. The measurement (by
extraction) of turbulence by the measurement operator turbulence-right Mr = |r r |
leads to Mr (A) = A . A successive measurement of phonation by the measurement
operator pitch-up Mu = |uu| gives Mu (A ) = A , thus Mu (A ) = Mu Mr (A) =
A . If we perform the measurements in the opposite order, with phonation first and
turbulence later, we obtain Mr Mu (A) = Mr (A∗ ) = A∗∗ . We expect that [Mr , Mu ] =
0, and thus, that A∗∗ = A . The diagram in Fig. 3 shows non-commutativity in the
style of category theory.
In bra-ket notation, this fact can be expressed as

Mr Mu |A = |r  r |u u|A = r |u |r  u|A =

(23)
Mu Mr |A = |u u|r  r |A = u|r  |u r |A .

Given that r |u is a scalar and u|r  is its complex conjugate, and that |u r | is
generally non-Hermitian, we get

[Mr , Mu ] = |r  r |u u| − |u u|r  r | =

(24)
= r |u |r  u| − u|r  |u r | = 0,

or, in terms of matrices

Fig. 3 A non-commutative Mu Mr
diagram representing the
non-commutativity of
measurements of phonation
A
Mr Mu A
(Mu ) and turbulence (Mr ) on A =
audio A Mu Mr A∗∗
A∗
Mr M u
Quanta in Sound, the Sound of Quanta 207

audio

sines+noise sines+noise

Harmonic Stochastic

sines+noise sines+noise

Stochastic Harmonic

Fig. 4 On the left, an audio segment is analyzed via the sines+noise model. Then, the noise part
is submitted to a new analysis. In this way, a measurement of phonation follows a measurement of
turbulence. On the right, the measurement of turbulence follows a measurement of phonation. This
can be described via projectors through equation (23), and diagrammatically in Fig. 3

   
1 11 10 1 10 11
[Mr , Mu ] = − =
2 11 00 2 00 11
   (25)
1 10 1 11 1 0 −1 i
= − = = σ y = 0.
2 10 2 00 2 1 0 2

On audio signals, measurements of phonation and turbulence can be performed

using the sines + noise model (Bonada et al. 2011), as described in Fig. 4. The
measurement of phonation is performed through the extraction of the sinusoidal
component, while the measurement of turbulence is performed through the extraction
of the noise component with the same model. The spectrograms for A and A∗∗ in
Fig. 5 show the results of such two sequences of analyses on a segment of female
speech, confirming that the commutator [Mr , Mu ] is non-zero.
Consider again Fig. 1, which shows a representation of the phon space using the
Bloch sphere. There are small spectrograms at the extremities, in correspondence
of |s, | f , |u, |d, |r , and |l. Applying σz σx to a state, we get the flipped state
we would obtain if we had applied σx σz . If we apply a pitch operator and then a
turbulence operator (or vice versa) to a slow impulse train (|s or | f ), we get another
impulse train.

2.5 Time Flies

The variation of quantum states in time can be obtained through the application of
time evolution operators on them. Similarly, suitable time operators can make the
density matrix vary in time as well. Given a density operator ρ(t0 ) at time t0 , its time
variation is obtained applying a unitary operator U (t0 , t):
208 M. Mannone and D. Rocchesso

Fig. 5 Top: spectrogram corresponding to a measurement of phonation Mu following a measure-

ment of turbulence Mr , leading to Mu Mr (A) = A ; bottom: spectrogram corresponding to a mea-
surement of turbulence Mr following a measurement of phonation Mu , leading to Mr Mu (A) = A∗∗

ρ(t) = U† (t0 , t)ρ(t0 )U(t0 , t). (26)

This is the most general definition: There are no assumptions on states (mixed or
pure), and the only assumptions on the operator U are that it is unitary, i.e., U† U = I,
with I the identity matrix, and that it depends only on t and t0 .
But actually there is more. The unitary operator U, evaluated at a tiny time incre-
ment
, is related to the Hamiltonian H, describing the energy of the system:

U(
) = I − i
H. (27)

For a closed and isolated system, H is time-independent, and the unitary operator
becomes U(t) = eiH(t−t0 ) . However, nature is more complex, things are not isolated,
and usually H is time-dependent, and the time evolution is given by an integral.
To complicate things even more, with a non-commutative Hamiltonian, an explicit
solution cannot be found. The problem can be circumvented by considering local
time segments where the Hamiltonian is locally commutative.
Quanta in Sound, the Sound of Quanta 209

An evolving state can, at a certain time, be subject to measurement. The quantum

measurement operator (or projector) acts on the state and make it collapse onto one
of its eigenvectors. If we have a mixed state, the system collapses into an ensemble
of states.
In the QVTS, the phon state evolution is subject to restoring forces, and the
Hamiltonian depends on the state orientation in the phon space. Such evolution is
alike that of a spin in a magnetic field. The Hamiltonian can be expressed as
 
ω ω nz n x − in y
H = σ ·n = , (28)
2 2 n x + in y −n z

whose energy eigenvalues are E j = ± ω2 , with energy eigenvectors |E j . An initial

phon |ψ(0) can be expanded in the energy eigenvectors as

|ψ(0) = α j (0) |E j  , (29)
j

where α j (0) = E j |ψ(0), and the time evolution of the state is

 
|ψ(t) = α j (t) |E j  = α j (0)e−i E j t |E j  . (30)
j j

Where do the restoring forces come from, in the sound domain? Broadly speaking,
they come from the local temporal sound production context. Similarly to the con-
cept of coarticulation in phonetics, the locally-defined Hamiltonian is determined by
neighboring sounds, extending their effects in the short-term future or past (Daniloff
and Hammarberg 1973). In practice, we can rely on an audio analysis system, such as
the Short-Time Fourier Transform (STFT), to extract and manipulate slowly-varying
features such as pitch salience or spectral energy to determine the components of the
Hamiltonian (28). Considered a slice of time and an audio signal, the initial phon
state can be made to evolve subject to a time-dependent yet commutative Hamiltonian
expressed as
H (t) = e−kt S, (31)

where S is a time-independent Hermitian matrix and k governs the spreading of

coarticulating features. Such Hamiltonian evolution has been inspired by a quantum
approach to image segmentation (Youssry et al. 2015), or figure-ground segregation.
For evolution in the phon space, the matrix S can be set to assume the structure (28),
where the components of potential energy can be extracted as audio features through
time-frequency analysis. For example, the n z component can be made to correspond
to the extracted pitch salience, and the n x component can be made to correspond to
the extracted noisiness. In the time slice under examination, an initial |u state will
evolve to a final state
t
|ψ(t) = e−i 0 H(τ )dτ
|u = U(0, t) |u , (32)
210 M. Mannone and D. Rocchesso

which in general will be a superposition (4) in the phon space. A measurement in the
computational (phonation) basis will make it collapse to |u or |d according to the
probabilities αu∗ αu or αd∗ αd , respectively. If there are two competing and concurrent
pitch lines, the Hamiltonian evolution followed by measurement may thus make
a pitch following process stay on one line or jump to the other one. In this way,
auditory streaming processes (Bregman 1994) and figure-ground segregation can be
mimicked.

3 Quantum Vocal Sound Processing

In this section, we present some examples that show how the quantum formalism,
as assimilated by the QVTS, can be used together with classical signal processing,
for creative yet controllable analysis/synthesis tasks. Given the time-frequency
representation of an audio signal, as provided by the STFT, the elements of the S
matrix of the Hamiltonian (31) can be computed from decimated audio features.
For example, pitch salience can be extracted from time-frequency analysis (Salamon
and Gomez 2012), and used as the n z component. The exponential factor can be set
to g(m) = e−km , where m is the frame number within a segment of M frames. The
time evolution (32) can be computed by approximating the integral with a cumulative
sum. Starting from an initial state (e.g., |u), the phon goes through repeated cycles of
Hamiltonian evolution, measurement, and collapse. The decision to measure phona-
tion or turbulence can be based on the degree of pitchiness that the evolution within
a certain audio segment has reached. Since the observable σz has eigenvalues ±1 for
eigenvectors |u and |d, a measure of the degree of pitchiness can be given by the
distance of σz |ψ from |ψ . The degrees of noisiness and transientness can be
similarly determined using the observables σx and σ y , respectively.
When doing actual audio signal processing based on the QVTS, several degrees
of freedom are available to experiment with: The decimation factor or number M of
frames in a segment; The exponential damping factor k; The thresholds for switching
to a certain measurement direction in phon space; The decision to collapse or not—
this is a freedom we have if we are using a classical computer!

3.1 Playing with Pure States

3.1.1 Fugue Following

Consider the beginning of the Fugue from the Toccata and Fugue in D Minor, BWV
565, by Johann Sebastian Bach (Fig. 6). In this fragment, there is only one voice,
played by the left hand. However, the design of this sequence actually creates the
illusion of two voices: an upper line with an ostinato A, and a lower line with notes
G, F, E, D, C, D, E, F, and so on.
Quanta in Sound, the Sound of Quanta 211

Fig. 6 Toccata and Fugue in D minor BWV 565 by J. S. Bach: beginning of the Fugue

Fig. 7 The synthetic recording of the excerpt in Fig. 6 through Hamiltonian evolution. The upper
line with the repeated A is evident in the first part of the graph, while the second part contains a
fragment of the melody of the lower line

The score fragment of Fig. 6 was automatically rendered with piano samples at 100
bpm and analyzed via the STFT,6 with pitch salience and noise energy extracted via
the SMS-tools (Bonada et al. 2011). Setting the parameters frame decimation M =
10, exponential damping k = 0.1, threshold of pitchiness 0.9, collapse decimation
5, we obtain a phon evolution from pitch-up phonation represented by the green dots
of Fig. 7. The red and yellow lines represent the two most salient pitches, as features
extracted from the STFT. In the first part of the evolution, the green dots mainly
correspond to the ostinato of note A, while, in the second part, they mainly follow
the ascending scale fragment (A B C D...). Even without any noise added, transient
and noise components are inherent in piano samples (e.g., key noise) and, therefore,
the phon is subject to non-negligible forces and actually moves from the initial
|u state. Different runs will give different evolutions, as the collapse is governed
by probability amplitudes, but in general we observe that the Hamiltonian listener
follows the upper or the lower melodic line for some fractions of time. Interestingly,
the melodic line following is stable, or even more stable, if we add a strong white
noise to the signal, with an amplitude that is about one tenth of the signal. An example
evolution is depicted in Fig. 8, where the effect of the added noise is only apparent
after second 5, when effectively there is no signal. Figure 9 shows the phon evolution
when the fugue is drowned into noise. In this case the melodic contour following is
more easily disrupted. The zero-pitch green dots represent points where measurement
and collapse have been oriented to the x direction, for the effect of thresholding in

6 Sample rate 44100Hz, window size 2048, transform size 4096, hop size 1024.
212 M. Mannone and D. Rocchesso

Fig. 8 The synthetic recording of the excerpt in Fig. 6 through Hamiltonian evolution, with the
added noise (amplitude 0.1). The upper line with the repeated A is evident in the first part of the
graph, while the second part contains a fragment of the melody of the lower line

Fig. 9 The same fragment of Fig. 6 completely drowned into noise (amplitude 1.0)

pitchiness. In a resynthesis, these dots correspond to noise bursts, while the other
dots come from z-oriented measurements and produce pitched notes.
In repeated runs of Hamiltonian fugue following, we can see multiple melodic
lines emerging as the time evolution of some initial sound/vocal state. The collapse
of the phon to a state or another can be interpreted as the attention shifts from figure
to ground, or vice versa (Bregman 1994; Bigand et al. 2000).
The proposed example can be the starting point for a wider investigation in the
field of auditory bistability. Bistability is an intriguing topic in cognition. As a refer-
ence for quantum effects in cognition, especially regarding superposition and non-
classical probability, we may refer to Yearsley and Busemeyer (2016) for a theoretical
quantum-based approach to explain cognitive acts. The idea of bistability is also faced
in an article on mathematics and image/music (Mannone et al. 2020). It exploits the
Dirac formalism used in quantum mechanics to represent images as superpositions
of essential visual forms. There is a minimum number of forms which allows the
Quanta in Sound, the Sound of Quanta 213

recognizability of the form. With a little abuse of terminology, we can consider this
limiting, minimum value of simple forms as the limit of a Gestalt neighborhood, as
the discrete version of a topological neighborhood, having in the center the initial,
complete, not-approximated visual form. When an image is bistable, we can imagine
to have two neighborhoods as the two faces of a thin cylinder. We can see one face
or the other one; but we cannot see the two faces together. This is the core idea of
bistability. While classic examples of bistability are visual, also auditory illusions
can be constructed, e.g., with different auditory streamings (Byrne et al. 2019). These
cases might be analyzed with the help of QVTS, as we did with the beginning of the
Fugue from the Toccata and Fugue BWV 565.

3.1.2 Glides Tunneling

Continuity effects have been very important to derive a perceptual organization of

sound for auditory scene analysis (Bregman 1994). Gestalt principles such as prox-
imity or good continuation are often used to describe how listeners follow concurrent
pitch lines and extract temporal patterns from a scene. A simple yet significant case
is that of two gliding and crossing tones interrupted, at the crossing point, by a short
burst of noise (Ciocca and Bregman 1987).
Figure 10 (top) shows the spectrogram of two gliding and crossing tones, inter-
rupted by a 200ms-band of white noise, intervening at time 1.5s. The red and yellow
lines are the traces of the two most salient pitches, as extracted using the Essentia
library (Bogdanov et al. 2013). With stimuli such as this, listeners most often report
perceiving a single frequency-varying auditory object tunneling (Vicario 1960) the
interruption. Depending on the temporal extension and intensity of the noise burst,
a perceived V-shaped trajectory may be predominant over a rectilinear continuation,
thus making proximity prevail over good continuation.
It is interesting to use the case of crossing glides interrupted by noise as a test for
Hamiltonian evolution, with the matrix S of the Hamiltonian (31) computed from
decimated audio features such as pitch salience and noise energy. As a result of such
feature extraction from a time-frequency representation, we obtain two potentials,
for phonation and turbulence, which drive the Hamiltonian evolution. Figure 10 also
displays (middle) the computed salience for the two most salient pitches and (bottom)
the energy traces for two bands of noise (1kHz–2kHz and 2kHz–6kHz). It is clear
how the pitch extractor becomes uncertain when the two tones get close in frequency
and start beating, and it wiggles around during the noisy interruption.
Figure 11 shows an example evolution of the phon state, starting from |u. In this
specific run of the evolution, the phon sticks to phonation (one of the two pitches)
until well inside the noise band, even though pitch has very little salience within
the interruption (see Fig. 10, middle), with only occasional switches to turbulence
(the zero-pitch green dots in Fig. 11). Right after the noise interruption, the phon
evolution is still uncertain, until it steadily takes a |u state, thus giving an overall
V-shaped bouncing trajectory. In this instance, proximity is shown to prevail over
good continuation. Due to the statistical nature of quantum measurement, another run
214 M. Mannone and D. Rocchesso

Fig. 10 Tracing the two most salient pitches and noise energy for two crossing glides interrupted
by noise

Fig. 11 Tracking the phon state under Hamiltonian evolution from pitch-up
Quanta in Sound, the Sound of Quanta 215

of the evolution may well produce a downward-crossing trajectory, and turbulence

bursts may be found at different times. Such uncertainty on noise location is consistent
with the known perceptual fact that bursts of noise overlapped to a noise transition are
not precisely located, with errors that can be up a few hundred milliseconds (Vicario
1963).
With this example, we have given a demonstration of how quantum evolution of
the phon state can be set to reproduce relevant phenomena in auditory perception,
with possible applications in computational auditory scene analysis.

3.2 Playing with Mixed States

The quantum concept of mixing, briefly described in Sect. 2.3, can be related to the
familiar audio concept of mixing. At the start of a Hamiltonian evolution, the initial
state may be mixed, i.e., known only as a probabilistic mixture. For example, at time
zero we may start from a mixture having 13 probability of |u and 23 probability of
|d. The density matrix would evolve in time according to equation (26).
When a pitch measurement is taken, the outcome is up or down according to

P[m = i|ρ] = T r [ρ Mi ], (33)

and the density matrix that results from collapsing upon measurement is given by

Mi ρ Mi
ρ (i) = . (34)
T r [ρ Mi ]

The density matrix can be made audible in various ways, thus sonifying the Hamil-
tonian evolution. For example, the completely chaotic mixed state, corresponding to
the half-identity matrix ρ = 21 I, can be made to sound as noise, and the pure states
can be made to represent distinct components of an audio mix.

3.2.1 Glides Tunneling

Given the same audio scene of the two crossing glides interrupted by noise (Fig. 10),
we may follow the Hamiltonian evolution from an initial mixed state. We can choose
to make the pure states to sound like the upper or the lower of the most salient
pitches, and the completely mixed state to sound like noise. These three components
can be mixed for states with intermediate degrees of purity. If pu and pd are the
respective probabilities of |u and |d as encoded in the mixed state, the resulting
mixed sound can be composed by a noise having amplitude min ( pu , pd ), by the
upper pitch weighted by pu − min ( pu , pd ), and by the lower pitch weighted by
pd − min ( pu , pd ). Figure 12 shows an example of evolution from the mixed state
having probabilities 13 and 23 , with periodic measurements and collapses ruled by
equation (34).
216 M. Mannone and D. Rocchesso

Fig. 12 Amplitudes of components |u, |d, and noise resulting from a Hamiltonian evolution from
a mixed state

Fig. 13 Sound synthesis obtained from the density matrix evolution from a mixed state, using the
component amplitudes depicted in Fig. 12

The analyzed audio scene and the model parameters, including the computed
Hamiltonian, are the same as used in the evolution of pure states described in
Sect. 3.1.2. The amplitudes of the three components can be used as automated knobs
to control two oscillators and a noise generator, producing the sound of spectro-
gram Fig. 13, characterized by a prevailing upward tone with a downward bifurcation
and a noisy tail.
Quanta in Sound, the Sound of Quanta 217

3.2.2 Vocal Superposition

As another example of mixed state evolution, we consider again the vocal sound
whose spectrogram is depicted in Fig. 2. It was chosen as an example of actual
superposition of phonation and slow myoelastic vibration. Despite the presence of
only one definite pitch, we can prepare the phon in an initial mixed state, having
1
3
probability of |u and 23 probability of |d, and compute a Hamiltonian evolution
based on potentials deduced from the time-frequency analysis, namely pitch salience,
noise component, and detected onsets. As in the example of Sect. 3.2.1, we chose
to assign phonation amplitudes equal to pu − min ( pu , pd ) and pd − min ( pu , pd )
to the components |u and |d, respectively, and turbulence amplitude min ( pu , pd )
to the noise component. In addition, here we extract a pulsating component as well,
corresponding to slow myoelastic vibration, whose amplitude is derived from the
probabilities p f and ps of fast or slow pulsation. For example, p f is derived from
T r [ρ | f   f |], which is similar to equation (33).
Figure 14 shows the amplitude profiles that are extracted from the Hamiltonian
evolution, where we chose to measure phonation when min ( pu , pd ) > 0.5, otherwise
measuring along the slow myoelastic vibration axis. With non-physical freedom, we
collapsed the mixed state, along |u, |d, | f , or |s, using an equation similar to (34),
once every five measurements. The resulting sound, which can be considered as a
quantum-inspired audio effect, has the spectrogram depicted in Fig. 15, where the
most salient pitch and the onsets have been extracted again and superimposed.

Fig. 14 Amplitudes of components |u, |d, turbulence, and slow myoelastic vibration, resulting
from a Hamiltonian evolution from a mixed state, run on the vocalization of Fig. 2
218 M. Mannone and D. Rocchesso

Fig. 15 Sound synthesis obtained from the density matrix evolution from a mixed state, using the
component amplitudes depicted in Fig. 14, and all three components of phonation, turbulence, and
slow myoelastic vibration. The most salient pitch and onsets, as extracted from the synthetic sound,
are displayed as red dashed lines

3.3 From Signal Processing to Quantum Computing,

and Back

In digital audio, talking about quantization does not mean referring to quantum theory.
Instead, quantization is meant to be the reduction of a continuous range of signal
amplitude values to a finite set of discrete values, with a cardinality that depends on the
number of bits dedicated to represent each discrete-time sample. Signal quantization
introduces a kind of noise, which tends to have a spectrotemporal structure that
somehow follows the signal, thus becoming audible as a distortion for low-amplitude
signals. A cure for quantization noise is dithering, i.e., adding some tiny broadband
noise to the audio signal itself, before quantization, thus making quantization noise
more spectrally uniform and perceptually tolerable (Pohlmann 1995). That injecting
dither noise to signals and systems can make human and machine processing more
robust is a fact that has been known for a long time, and widely applied in a variety of
fields, including audio and image processing. In quantum-inspired sound processing,
as illustrated in the example of Sect. 3.1, dithering can be used to control how erratic
leading-pitch attribution can be, in auditory scenes of competing sources.
As opposed to low-amplitude noise, that may actually make the pitch evolution
of a phon more stable, when a high-amplitude noise burst is encountered, it actually
acts as a bounce on the phon state, making it rotate by an angle θ . A sequence of
bursts, such as that of the example in Sect. 3.2, is much like a sequence of bounces
between billiard balls of highly different weights. Recently, the classically mechanic
behavior of balls, whose weights are in ratios of powers of 100, has been shown to be
Quanta in Sound, the Sound of Quanta 219

perfectly analogous to the kernel of the Grover algorithm for quantum search (Brown
2020), which is based on unitary reflections in the state space.
In the examples of Sects. 3.1 and 3.2, the quantum evolution is driven by potentials
that are derived from the same audio that is being processed. To turn these evolu-
tions into quantum algorithms we should freeze a reference audio segment, extract
the feature-based potentials from the time-frequency representation, and convert
the elementary unitary transformations into quantum gates, arranged along left-to-
right wires. Each stage of the quantum algorithm would represent a bounce or a
measurement in the phon space, as long as the operators are consistent with the
postulates of quantum mechanics. It should be noted that we have only been consid-
ering single-qubit (or single-phon) operators. The universe of multiple and entangled
phons remains to be explored.
In both quantum mechanics and sound signal processing, unitary operators and
unitary transformations have a central role. In fact, in physically-inspired sound syn-
thesis and digital audio effects, unitary matrix transformations are often found, as
scattering elements in feedback delay networks for artificial reverberation (Pulkki
et al. 2011). In these structures, if the feedback matrix A is chosen to be unitary,
an initial pulse bounces indefinitely, at each bounce scattering into a multiplicity of
other pulses. In the ball-within-the-box (BaBo) model (Rocchesso 1995), the matrix
A can be interpreted as a scattering ball that redistributes the energy from incoming
wavefronts into different directions, each corresponding to a planar wave loop, which
produces a harmonic series. Indeed, the matrix A does not have to be unitary for the
feedback structure to be lossless (Schlecht and Habets 2017). However, even staying
within the class of unitary matrices, A can be chosen for its scattering properties,
ranging from the identity matrix (no scattering at all) to maximally-diffusive struc-
tures (Rocchesso 1997; Schlecht 2020). A promising perspective for future quantum
sound processing, is to find realizable quantum operators for such matrices. In par-
ticular, the Hadamard operator and the Householder reflection are extensively used
in quantum algorithms, and these were proposed as reference matrices for feedback
delay networks with maximally-diffusive properties (Jot 1997). In the context of
QVTS, a Hadamard gate H converts a phon from |r  to |u, and from |l to |d. If
followed by a measurement in the computational basis, it can be used to discriminate
between the two turbulent states. If inserted in a phon manipulation sequence, it
determines a switch in the vocal state of sound. Loops are not allowed in quantum
computing (Nielsen 2010), but by spatially unfolding feedback, a reverberator based
on feedback delay networks may be converted to a quantum algorithm with several
stages of unitary operators, acting as scattering elements on a multiplicity of phons.
As a non-negligible detail, banks of delay lines of different lengths, in the order of
tens of milliseconds, should be interposed between consecutive scattering sections.
Shor’s algorithm for factorization of a large integer N relies on an efficient way
of finding the periodicity (modulo N ) of the function a x , constructed from a ran-
domly chosen integer a that is smaller than N and coprime with it. To compute
the periodicity of a function, the Quantum Fourier Transform (QFT) operator is
used, which transforms a superposition
 of N = 2n computational basis states on
n qubits, with coefficients x = x0 , x1 , . . . , x N −1 , into another superposition with
220 M. Mannone and D. Rocchesso
 
coefficients X = X 0 , X 1 , . . . , X N −1 = D F T (x), that is the Discrete Fourier
Transform of x. Using quantum parallelism, such DFT is implemented with O(n 2 )
quantum gates, while classically that would take O(n2n ) steps. Recently, a direct
transposition of the Fast Fourier Transform (FFT) into the form of a quantum circuit
has been proposed and called the Quantum FFT (QFFT) (Asaka et al. 2020). Instead
of the amplitude encoding used for the QFT, a basis encoding is −1
used, where a data
sequence x is expressed as a tensor product of vector spaces Nj=0 |x j . A potential
impact on audio signal processing would be that quantum parallelism would allow
to perform all frames of a N -bins STFT simultaneously, with O(N log2 N ) gates.
The aggregate features of an audio segment would then be encoded in the resulting
vector of qubits.

4 Quantum Evolution of the State (of the Art)

A respectable scientific theory helps find new results, confirms expectations, extends
the validity of known laws bringing them toward the realm of the unknown and
(formerly) inexplicable, and so on.
An exciting scientific theory leaves room for imagination and artistic creativity.
New ideas can arise from the interdisciplinary dialogue between people of differ-
ent fields. QVTS is intrinsically interdisciplinary, and we think it can enhance the
dialogue between worlds.
Interchanges between music and quantum mechanics constitute a relatively new
and flourishing research area. Our contribution to this field is the addition of the
human voice, and the use of vocal primitives as a probe to more generally investigate
the world of sounds. In Sect. 3 we have proposed some examples of a creative use of
QVTS, where the Hamiltonian evolution is the starting point for sound synthesis. In
this section, we suggest some further creative applications.
The density matrix can be exploited to improve source separation techniques. In
fact, the operation of partial trace on density matrix allows us to separate a system
from the environment, the reservoir. Choosing on which part of the whole system
we are making the operation of partial trace, we can interchangeably choose which
part we are neglecting. For example, given a polyphonic vocal recording, we can
establish that singer 1 is the system and singers 2, 3, and 4 are the environment (thus,
we can perform the partial trace on singers 2-3-4), or that singer 2 is the system and
singers 1, 3, and 4 are the environment, and so on. In fact, as a practical interest
in the domain of QVTS, we can think of a general recording, with multiple voices,
and interpret it as a statistic mixture of states. Voices might be organized as a solo
singing voice against a a background of several other voices of a choir—a quantum
choir. Therefore, QVTS may help analyze choral music. In addition, it can give us
hints also on how to create music. Creativity can precisely take off from mixtures of
states and vocal polyphony.
Because QVTS constitutes a bridge between sounds and quantum formalism, we
can play with the symmetries of particle processes and transform them to musical
Quanta in Sound, the Sound of Quanta 221

symmetries, thus giving voice to quantum processes. We can create correspondences

between certain quantum properties of particles and the sounds, their transformations,
and musical transformations. For example, an inversion of the spin could be musically
rendered with an inversion of the pitch interval; a quantum superposition can be
rendered with the simultaneous playing of different orchestral sections to create a
“cloud of sound.” A quantum measurement, with the subsequent collapse of the sound
wave, could be rendered with the sudden silence of other orchestral sections, and
with the remaining sound of a section, or even only one instrument sound. Musical
structures can be thought of as transformations over time of “states” (short musical
sequences or essential musical ideas for example). According to this metaphor, we
might describe the time evolution of quantum states, including density matrices
describing inseparable state superpositions through generated musical structures.
These hints should be compared with perceptual criteria, to create an idea of the
processes in the mind of the listener.
Finally, we may imagine an interface where the user can modify states on the
Bloch sphere, modifying the synthesis in real time. Such an interface might allow
a “Quantum Synthesis,” maybe the Gabor’s dream. A quantum synthesizer with
potential for development has indeed been recently proposed by Costa Hamido et al.
(2020), where the quantum circuits such as the one for Grover’s search can be run on
a simulator or a quantum computer, and probability distributions and computation
steps can be heard, with auditory exploitation of quantum noise.
We end this section with a fun, original musical fragment, or, better, a set of
instructions to make music directly out of the QVTS-Bloch’s sphere. A suitable
synthesizer as the one hypothesized above could make this attempt a concrete tool
for creative purposes. Let us imagine a short musical composition with two vocal
(not instrumental) lines, created out of moving states on the Bloch’s sphere. As
another homage to Bach, we can be inspired by the structure of the Two-Part Inven-
tions, with the parts imitating each other, as in a simple counterpoint. Thus, we can
provocatively call our attempt Two-Part Quantum Invention No. 1. Figure 16 shows
a tentative notation, with a schematic sphere derived from Fig. 1, and the sequence
of state variations. Generalization of the proposed idea to more voices and intricate
counterpoints is up the reader. This structure could be used as a set of instructions
for vocal improvisation, similarly to the “quantum improvisations” Pauline Oliveros
used to conduct (Oliveros 2002). Conceptually, the two voices can be instances of
the same evolving phon, from which we can, in principle, extract infinite counter-
point lines. If the parallel motion of parts causes troubles in classical counterpoint
because of the feeling of sameness, intrinsic parallelism is the real advantage of
quantum computation, eventually leading to quantum supremacy for some compu-
tational problems. Music counterpoint may actually give voice to quantum parallel
computations.
222 M. Mannone and D. Rocchesso

Fig. 16 Sketch for a music composition based on Bloch’s sphere for QVTS (by M.M.).

4.1 Concluding Remarks

Starting from kids’ playing and moving toward futuristic scenarios of quantum choirs
and Quantum Inventions, in this chapter we presented the fundamental ideas of
the Quantum Vocal Theory of Sound (QVTS), along with some proposal of future
developments.
We aimed to discuss a supplemental formalism to describe sound (with the vocal
probe), rather than proving any “wrongness” or obsolescence of the classical formal-
ism, such as Fourier analysis, for sound. Our supplemental formalism is an alternative
one, it gives a new perspective, and it has the advantage of providing more infor-
mation, especially regarding Gestalt-related phenomena, as in the case of bistable
states.
QVTS is interdisciplinary in nature, as it provides a bridge between sound sources,
sound production, human perception, and the intuitive identification of sounds and
sound sources. In addition, the uncertainty (or fuzziness) that is proper to quantum
Quanta in Sound, the Sound of Quanta 223

thinking might be compared with the approximation of human intuitive assessments

about sounds, sound sources, and sound identification.
Beyond the theoretical foundation, our aim is to foster the creation of machines that
measure sounds in terms of vocal primitives and express sonic processes as evolving
superpositions of vocal primitives. We hope that our presentation of QVTS may lead
to further questions, research, developments, as well as to artistic contributions.
May it be a simple metaphor, or a quantitative tool, the core of quantum mechanics
is more and more inspiring for musicians, performers, and scientists, shedding light
on new and unexplored collaborations and insights.
All of that, can lead to the sound of quanta.

References

Asaka, R., Sakai, K., & Yahagi, R. (2020). Quantum circuit for the fast Fourier transform. Quantum
Information Processing, 19, 277.
Aspect, A., Dalibard, J., & Roger, G. (1982). Experimental test of Bell’s inequalities using time-
varying analyzers. Physical Review Letters, 49(25), 1804.
Bell, J. S. (1964). On the Einstein Podolsky Rosen Paradox. Physics Physique Fizika, 1(3), 195–200.
https://doi.org/10.1103/PhysicsPhysiqueFizika.1.195.
Bigand, E., McAdams, S., & Forêt, S. (2000). Divided attention in music. International Journal of
Psychology, 35(6), 270–278.
Blutner, R., & Beim Graben, P. (2020). Gauge models of musical forces. Journal of Mathematics
and Music. https://doi.org/10.1080/17459737.2020.1716404.
Bogdanov, D., Wack, N., Gómez Gutiérrez, E., Gulati, S., Herrera Boyer, P., Mayor, O., Roma
Trepat, G., Salamon, J., Zapata González, J. R., & Serra, X. (2013). Essentia: An audio analysis
library for music information retrieval. In Proceedings of the 14th Conference of the International
Society for Music Information Retrieval (ISMIR) (pp. 493–498), Curitiba, Brazil
Bonada, J., Serra, X., Amatriain, X., & Loscos, A. (2011). Spectral processing. In U. Zölzer (Ed.),
DAFX: Digital Audio Effects (pp. 393–445). John Wiley & Sons Ltd.
Bregman, A. S. (1994). Auditory scene analysis: The perceptual organization of sound. Cambridge,
MA: MIT Press.
Brown, A. R. (2020) Playing Pool with |ψ: from Bouncing Billiards to Quantum Search. Quantum
4, 357, Verein zur Förderung des Open Access Publizierens in den Quantenwissenschaften.
Byrne, Á., Rinzel, J., & Rankin, J. (2019). Auditory streaming and bistability paradigm extended
to a dynamic environment. Hearing Research,383, 107807.
Cádiz, R., & Ramos, J. (2014). Sound synthesis of a gaussian quantum particle in an infinite square
well. Computer Music Journal, 38(4), 53–67.
Cariolaro, G. (2015). Quantum Communications. Cham: Springer.
Ciocca, V., & Bregman, A. S. (1987). Perceived continuity of gliding and steady-state tones through
interrupting noise. Perception & Psychophysics, 42(5), 476–484, Springer .
Costa Hamido, O., Cirillo, G. A., & Giusto, E. (2020). Quantum synth: a quantum-computing-based
synthesizer. In Proceedings of the 15th International Conference on Audio Mostly (pp. 265–268)
(2020)
Dalla Chiara, M. L., Giuntini, R., Leporini, R., Negri, E., & Sergioli, G. (2015). Quantum informa-
tion, cognition, and music. Frontiers in Psychology, 6, 1583.
224 M. Mannone and D. Rocchesso

Daniloff, R. G., Hammarberg, R. E. (1973). On defining coarticulation. Journal of Phonetics, 1(3),

239–248. http://www.sciencedirect.com/science/article/pii/S0095447019313889
Delle Monache, S., Rocchesso, R., Bevilacqua, F., Lemaitre, G., Baldan, S., & Cera, A. (2018).
Embodied sound design. International Journal of Human-Computer Studies, 118, 47–59.
Einstein, A., Podolsky, B., & Rosen, N. (1935). Can quantum-mechanical description of physical
reality be considered complete? Physical Review, 41, 777. http://www.drchinese.com/David/
EPR.pdf
Feynman, R. (1995). The Character of Physical Law. Cambridge (MA): MIT Press.
Friberg, A., Lindeberg, T., Hellwagner, M., Helgason, P., Salomão, G. L., Elowsson, A., Lemaitre,
G., & Ternström, S. (2018). Prediction of three articulatory categories in vocal sound imitations
using models for auditory receptive fields. The Journal of the Acoustical Society of America,
144(3), 1467–1483
Gabor, D.: Acoustical quanta and the theory of hearing. Nature, 159, 591–594 (1947)
Gaver, W. W. (1993). What in the world do we hear? An ecological approach o auditory event
perception. Ecological Psychology, 5(4), 285.
Ghirardi, G. (1999). Quantum superpositions and definite perceptions: envisaging new feasible
experimental tests. Physics Letters A, 262(1), 1–14.
Irino, T., & Patterson, R. D. (1997). A time-domain, level-dependent auditory filter: The gam-
machirp. The Journal of the Acoustical Society of America, 101(1), 412–419.
Jot, J. -M. (1997) Efficient models for reverberation and distance rendering in computer music and
virtual audio reality. In Proceedings: International Computer Music Conference 1997, Thessa-
loniki, Hellas, 25–30 September 1997, 236–243. The International Computer Music Association
(1997)
Juett, J. C. (2010). Pauline Oliveros and quantum sound. Liminalities: A Journal of Performance
Studies, 6(2), 1–10. http://liminalities.net/6-2/oliveros.pdf
Koenig, D. M., & Delwin, D. F. (2015). Spectral Analysis of Musical Sounds with Emphasis on the
Piano. New York: Oxford University Press.
Kulpa, J. (2019). QuBits, an interactive virtual reality project and compositional space for sound
and image. Advisor: E. Campion. Ph.D. dissertation, Berkeley, California. https://escholarship.
org/uc/item/5tq9x2hb
Kulpa, J., Campion, E., & Cella, C. (2020) QuBits, a system for interactive sonic virtual reality.
In Proceedings of ICMC 2020, Santiago, Chile. http://www.carminecella.com/research/2020_
ICMC_QuBits.pdf
Lindeberg, T., & Friberg, A. Idealized computational models for auditory receptive fields. PLoS
One, 10(3), e0119032. https://doi.org/10.1371/journal.pone.0119032
Mannone, M. (2018). Introduction to gestural similarity. An application of category theory to the
orchestra. Journal of Mathematics and Music, 12(2), 63–87.
Mannone, M., & Compagno, G. (2014). Characterization of the degree of Musical non-markovianity.
https://arxiv.org/abs/1306.0229
Mannone, M., Favali, F., Di Donato, B., & Turchet, L. (2020). Quantum GestART: identifying
and applying correlations between mathematics, art, and perceptual organization. Journal of
Mathematics and Music. https://doi.org/10.1080/17459737.2020.1726691.
Marinetti, F. T. (1914) Zang tumb tumb. Milano, Edizioni futuriste di poesia. http://parliamoitaliano.
altervista.org/zang-tumb-tumb/.
Miranda, E. R. (2021) Creative quantum computing: Inverse FFT sound synthesis, adaptive sequenc-
ing and musical composition. In: A. Adamatzky (Ed.), Alternative Computing, World Scientific
(2021)
Miranda, E. R. (2021). Quantum computer: hello, music! In E. Miranda (Ed.), Handbook of Artificial
Intelligence for Music: Foundations, Advanced Approaches, and Developments for Creativity.
Springer Nature
Newman, F. (2004). MouthSounds: How to Whistle Pop: Boing, and Honk... for All Occasions and
Then Some. Workman Publishing.
Quanta in Sound, the Sound of Quanta 225

Nielsen, M. A. & Chuang. I. L. (2010). Quantum Computation and Quantum Information. Cam-
bridge University Press.
Oliveros, P. (2002) Quantum listening: From practice to theory (to practise practice). In K. Siu Tong
& C. Sin-wai (Eds.), Culture and Humanity in the New Millennium: The Future of Human Values
(pp. 27–44). The Chinese University Press.
Palazzeschi, A. (1909). La fontana malata, from Poemi. Firenze, Cesare Blanc. https://www.
pensieriparole.it/poesie/poesie-d-autore/poesia-38384
Pohlmann, K. C. (1995). Principles of Digital Audio (3rd ed.). USA: McGraw-Hill Inc.
Pulkki, V., Lokki, T., & Rocchesso, D. (2011) Spatial Effects. DAFX: Digital Audio Effects (pp.
139–183). Wiley.
Roads, C. (2001). Microsound. Cambridge, MA: MIT Press.
Rocchesso, D. (1995). The ball within the box: A sound-processing metaphor. Computer Music
Journal, 4(19), 47–57. The MIT Press. http://www.jstor.org/stable/3680990
Rocchesso, D., Mauro, D. A., & Delle Monache, S. (2016). MiMic: The Microphone as a Pencil. In
Proceedings of the TEI ’16: Tenth International Conference on Tangible, Embedded, and Embod-
ied Interaction, in Eindhoven, Netherlands (pp. 357–364). New York, NY, USA: Association for
Computing Machinery, TEI ’16. https://doi.org/10.1145/2839462.2839467
Rocchesso, D., Mauro, D., & Drioli, C. (2016). Organizing a sonic space through vocal imitations.
Journal of the Audio Engineering Society, 64(7/8), 474–483. http://www.aes.org/e-lib/browse.
cfm?elib=18333
Rocchesso, D. (1997). Maximally diffusive yet efficient feedback delay networks for artificial rever-
beration. IEEE Signal Processing Letters, 4(9), 252–255. https://doi.org/10.1109/97.623041.
Rocchesso, D., & Mannone, M. (2020). A quantum vocal theory of sound. Quantum Information
Processing, 19, 292. https://doi.org/10.1007/s11128-020-02772-9.
Rovelli, C. (2021). Helgoland. Penguin Books.
Salamon, J., & Gomez, E. (2012). Melody extraction from polyphonic music signals using pitch
contour characteristics. IEEE Transactions on Audio, Speech, and Language Processing, 20(6),
1759–1770.
Schlecht, S. J. (2020). FDNTB: The feedback delay network toolbox. In Proceedings of the 23rd
International Conference on Digital Audio Effects DAFx-20.
Schlecht, S. J., & Habets, E. A. P. (2017). On lossless feedback delay networks. IEEE Transactions
on Signal Processing, 65(6), 1554–1564. https://doi.org/10.1109/TSP.2016.2637323.
Shamma S. (2013). Spectro-temporal receptive fields. In D. Jaeger , R. Jung (Eds.), Encyclopedia
of Computational Neuroscience. New York, NY: Springer. https://doi.org/10.1007/978-1-4614-
7320-6_437-1.
Spence, C. (2011) Crossmodal correspondences: A tutorial review. Attention, Perception, & Psy-
chophysics, 73, 971–995. https://doi.org/10.3758/s13414-010-0073-7
Spence, C., Senkowski, D., & Röder, B. (2009). Crossmodal processing. Experimental Brain
Research, 198(107), 107–111 (2009). https://link.springer.com/article/10.1007/s00221-009-
1973-4.
Verma, T. S., Levine, S. N., & Meng, T. H. (1997). Transient modeling synthesis: A flexible anal-
ysis/synthesis tool for transient signals. In Proceedings of the International Computer Music
Conference (pp. 48–51)
Vicario, G. B. (1963). La “dislocazione temporale” nella percezione di successioni di stimoli dis-
creti (The “time displacement” in the perception of sequences of discrete stimuli). Rivista di
Psicologia,57(1), 17–87.
Vicario, G. (1960). L’effetto tunnel acustico (The acoustic tunnel effect). Rivista di Psicologia, 54,
41–52.
Vochlea microphone. Retrieved from https://vochlea.com/.
von Helmholtz, H. (1870). Die Lehre von den Tonempfindungen als physiologische Grundlagefürdie
Theorie der Musik. Braunschweig: F. Vieweg und sohn.
Weimer, H. (2010). Listen to quantum computer music. Retrieved Jan. 20, 2021, from http://www.
quantenblog.net/physics/quantum-computer-music.
226 M. Mannone and D. Rocchesso

Yearsley, J. M., & Busemeyer, J. R. (2016). Foundations of probability theory in psychology and
beyond (special issue). Journal of Mathematical Psychology, 74, 99.
Youssry, A., El-Rafei, A., & Elramly, S. (2015). A quantum mechanics-based framework for image
processing and its application to image segmentation. Quantum Information Processing, 14(10),
3613–3638.
Quantum Cinema and Quantum
Computing

Renate C.-Z. Quehenberger

It cannot help striking me that this extension of Algebra ought to

lead to a further extension similar in nature to the Geometry of
three dimensions; & that again perhaps to a further extension
into some unknown region & so on ad infinitum possibly! (Ada
Lovelace in a letter to her tutor Augustus De Morgan, 1832,
quoted in Padua, 2015, 60).

Abstract The author discusses philosophical, ontological and phenomenological

questions of quantum phenomena and their possible visualization by means of
quantum geometry as well as links to the complex number space, prime numbers
and lattices used for quantum computation. The chapter engages with intuitively
developed images realized as digital dynamic 3D geometry. Some of these images,
which could be relevant in the context of Quantum Computing, are introduced. The
author discusses discoveries that became the basics of Quantum Computing through
a journey into the history of Quantum Physics, which lead to the notion of quantum
information and the conception of quantum computers.

Keywords Quantum humanities · Higherdimensional geometry of Quantum

states · 3D Penrose patterns · Heisenberg’s uncertainty principle · Poincaré
conjecture

1 Introduction

This chapter on Quantum Cinema and Quantum computing engages with intuitively
developed images realized as digital dynamic 3D geometry. Some of these images,
which could be relevant in the context of quantum computing, are presented here.

R. C.-Z. Quehenberger (B)

Independent SciArt Researcher, QC:L, Vienna, Austria
Luxor, Egypt
e-mail: [email protected]; [email protected]
URL: https://quantumcinema.uni-ak.ac.at/

© Springer Nature Switzerland AG 2022 227

E. R. Miranda (ed.), Quantum Computing in the Arts and Humanities,
https://doi.org/10.1007/978-3-030-95538-0_7
228 R. C.-Z. Quehenberger

The exciting development of quantum computation is rooted in the fundamentally

new interpretation of quantum physics which became the “it from bit” doctrine by
John Archibald Wheeler, who linked information, physics and quantum phenomena
by suggesting all things physical are information-theoretic in origin (Wheeler,
1989). According to Claude Shannon’s communication protocol, information is neg-
entropy (Shannon, 1948). His model of the communication process defines Wheeler’s
information paradigm. Shannon’s original communication model limits the interest
in signalling systems that include transmitter, receiver, channel and noise (Abel
and Trevor, 2005). It gave rise to the new discipline of quantum information and
quantum information science, the study of how a quantum computer works.
Quantum computers rely on qubits as their basic unit of information. Whereas
traditional computers use bits, which is a unit of information based on binary and
can either be in an ‘on’ state or an ‘off’ state, quantum computers use qubits instead.
A qubit introduces a new state: both ‘off’ and ‘on’, at the same time. This creates
three states instead of two, allowing for much higher quantities of data to be trans-
ferred. The current issue (which has to be overcome) is that quantum machines
can generate control signals for multiple qubits at cryogenic temperatures—100
millikelvin, or—273.05 °C, which is almost absolute zero (Pauka et al., 2021).
The world’s biggest quantum computers currently operate with just 50 or so qubits,
whereas the control platform with the cryogenic chip to control thousands of qubits
is the next big step to scale up quantum computing. While Google is predicting the
first error-free quantum computer by the end of the decade.
Here we will discuss some philosophical ontological and phenomenological ques-
tions of quantum phenomena and their possible visualization by means of quantum
geometry as well as links to the complex number space, prime numbers and lattices
used for quantum computation.
We will tackle discoveries that became the basics of quantum computing in a
parcour through the history of quantum physics, which lead to quantum information
and the conception of quantum computers.

2 Quantum Cinema—A Digital Vision

Linear algebra is the mathematical language of quantum computation. It is repre-

sented by high-dimensional vector spaces as developed by Hermann Grassmann,
who was not only inspired by his father, a crystallographer, but also by ancient
Indian philosophy that speaks of the infinite dimensional Net of Indra.— How
can we imagine the complex network of 1-dimensional vectors embedded in a
higher-dimensional spatial continuum?
This question about visualization of higher mathematics and the imagination of
quantum phenomena was the task of the interdisciplinary SciArt research project
Quantum Cinema and Quantum Computing 229

Quantum Cinema—a digital vision (2010–2013) 1 mentored by the then most eminent
Austrian physicist Helmut Rauch, who first proved the 4π-value of the spin of a
neutron (Rauch et al., 1975). As a pioneer in the field of neutron interferometry, he
said,
There are strong indications, that we may understand the physical world as interference
pattern (Rauch, in a private conversation, 2012).

The Quantum Cinema project aimed to develop visual tools for the perception
of quantum phenomena. After all, computer animation is the perfect medium for
overcoming projective geometry, which is foundational for many scientific concepts.
We hold that only the invention of digital media makes the appropriate outlook of
higher mathematics possible, and animated 3D graphics are the new medium for a
holographic depiction model which was realized with the help of a group of artists
with trained imagination skills.
Quantum Cinema is based on the idea that geometry must not remain restricted
to the plane, 2-dimensional page—merely a shadow of rigid shapes projected onto a
plane as it is the case with projective geometry—but may create an abstract movie of
previously unimaginable quantum phenomena in order to gain cognitive and visual
access to the quantum realms.—Hence, the title Quantum Cinema.
The term Quantum Cinema was originally coined by the Austrian media theorist
Peter Weibel—who was so kind as to head this project—for his (then) futuristic
vision of a cinema that can be perceived without further external device directly in
the brain (Weibel, 2005). Nowadays, human enhancement projects like Neuralink,
brain–machine interfaces (BMIs) for full-bandwidth data streaming from a device
are under development (Musk, 2019).
Questions of imagination and visualization of quantum phenomena relevant in
the context of quantum computation shall be discussed here:
– How can we understand quantum phenomena such as the wave-particle duality
and the superposition of particles and the entanglement of two or more particles?
These are important features for the application of quantum algorithms since they
are produced by a discrete Fourier transform over the amplitudes of a particle’s wave
function. Fourier transform allows for any wave to be uniquely decomposable into
an infinite superposition of trigonometric waves of different frequencies (Fourier,
1822).
– How can we imagine phenomena on the foundations of nature which serve as the
most effective ingredients for quantum computing technology?
– How can we imagine the tiniest processes in nature, known as wave-particle
duality, that serve as bits in quantum computers?
– How can we imagine quantum fluctuations, the wavelike state of a particle that
produces a numeric output by means of a quantum computer?

1I am grateful to Quantum Cinema artists Nikola Tasic, Christian Magnes, Rudi Friemel, Kathrin
Stumreich and the mathematician and civil engineer Dr. Hans Katzgraber.
230 R. C.-Z. Quehenberger

We hope that 3D animated geometry as performed in the Quantum Cinema project

could be interesting in regard to topological quantum systems, especially our new
visualization of a hypersphere, a geometrical object that is used in Quantum geometry
even though only insufficient construction methods exist.
Finally, features used in quantum computation such as Fourier transform, Bern-
hard Riemann’s zeta function and Fuchsian functions will be discussed in relation
to the famous Poincaré’s conjecture about the fundamental polyhedron and the
homology sphere which turns out to be homeomorphic to a 3-sphere, as we will
show below.—Is this an alternative answer to the Clay math millennium’s problem?
Different lines of thought from brilliant minds have been creating challenges as
well as obstacles that have nevertheless led to the current state of development of
quantum computers.

3 Lady Lovelace—Poetical Science and the Role

of Imagination

Lady Lovelace is meanwhile widely recognized for the conception of the first
computer algorithm, published as a footnote to her translation of Italian military
engineer Luigi Menabrea article into English (Lovelace, 1842).
Although the further development of algebras remained widely abstract, in the
1830s, Augusta Ada King, the Countess of Lovelace, developed her imagination of
algebraic operations in higher dimensions.
Her interest in mathematics was fostered by her mother, Lady Annabella Byron,
dubbed “the Princess of Parallelograms” for her own fascination with mathematics.
She tried to keep her, as Lord Byron’s daughter, away from poetry to prevent her
from following in his footsteps. So Lady Lovelace was tutored by the mathematician
Augustus De Morgan, who was, with George Boole, one of the founders of formal
logic and developers of modern algebra. In Boolean algebra, True and False are
replaced by 0 and 1, which became the cornerstones of a binary decision-making
system to which the third indeterminate value, the state of superposition, was added
for quantum computation. In quantum logic, De Morgan’s Laws play an important
role in the projection-valued truth value assignment of observational propositions in
quantum mechanics.
By 1841, Ada Lovelace was developing a concept of “Poetical Science” in which
scientific logic would be driven by imagination. “Poetical Science” can also be under-
stood as a precursor of SciArt, a collaboration between the arts and sciences as it
is accomplished partly nowadays. For Lovelace, imagination is “the discovering
faculty, pre-eminently—science and religion were the same and imagination itself
was divine.
Imagination penetrates into the unseen worlds around us, the worlds of Science.
She saw mathematics metaphysically, as “the language of the unseen relations
between things” (Holmes, 2015).
Quantum Cinema and Quantum Computing 231

The development of an appropriate geometrical imagination for these (still)

unknown regions of most abstract higher mathematics was also the topic of the
SciArt project called Quantum Cinema—a digital vision.
It is a gift to the art world that Ada Lovelace’s prediction, that one day we will
be able to make music with the mathematical engine, was realized in a wide range
of electronic devices and music computer programs (Lovelace in Toole, 2010, 366–
67). While the vision of her comrade, the engineer of the analytical engine, Charles
Babbage, was about a machine with divine abilities which apparently became reality.
He spoke of an engine that like god—whom he called the 1st programmer—will
detect and record each whisper and every movement of a leaf (Babbage, 1838).
Nowadays, we may well recognize in his words a metaphor for surveillance programs
currently harnessed by quantum computing technologies.

3.1 Art and Science

The physicist David Bohm spoke of an art form of science. The scientist is an artisan
who has to create sensitive instruments in order to discover oneness and totality—
like quantum computation technology, one might add—which aid perception and so
make possible both the testing of new ideas for their truth or falsity, and the disclosure
of new and unexpected kinds of facts (Bohm, 2004, 105).
On the other hand, the artist may create visual access to subscendent structures,
to what Plato called the world of ideas, that express the harmony and beauty in
nature—categories which remained more valid in the arts than in science.—The
beautiful formula as a guarantee for truth became obsolete.
Computer-generated art is based on mathematical algorithms which were previ-
ously considered as an art form.
Algorism is the art by which at present we use those Indian figures, which number two times
five. (Alexandre de Villedieu, poem Carmen de Algorismo, 1240)

This phrase alludes to the Indian origin of the numerals (Indorum ars numerandi)
or Hindu numerals. The term algorithm goes back to the name of Persian mathe-
matician Muhammad ibn Musa al-Khwārizmı̄ (780–850), who is arguably the most
important mathematician of the Middle Ages. Besides calculating the most exact
calendar of his time and his contributions to architecture and poetry, he developed
two distinct branches of mathematics, both of which owe their name to him: algebra
and algorithms (Brezina, 2006).
For the French, thirteenth century poet Alexander de Villa Dei mathematics was
part of the Liberal Arts and algorithms an art form. A century later, Leonardo da
Vinci pursued painting as a scientific method. For example, his advice on the mode
of studying anatomy was,
Study the science first, and then follow the practice which results from that science. (Leonardo
da Vinci [1490] 1877, 13)
232 R. C.-Z. Quehenberger

The Austrian quantum physicist Wolfgang Pauli confirmed that at the time of the
“re-opening” of the Platonic Academy in Florence (1438–39) sponsored by Cosimo
de’ Medici, Leonardo was the scientist while Marsilio Ficino, the “first academic”,
and Pico della Mirandola founder of esotericism, did not practice sober science, but
religious mysticism, which originated in a new cosmic feeling and expressed itself
in particular in the deification of space (Pauli, 1955, 8). In the fifteenth century, the
German philosopher and theologian Nicolas of Cusa proclaimed that space is infinite
and boundless, like the n-dimensional space in mathematics later on.
The invention of perspective and projective geometry centuries before by Renais-
sance artists, such as Paolo Uccello and Albrecht Dürer, only gained importance in
mathematics in the nineteenth century. Projective geometry was the basis for the
invention of cinema in the nineteenth century. The projection of an image onto a
screen by a beam of light led to new worlds’ arising at the point at infinity. Whereas
according to Gilles Deleuze narrative cinema produces “time crystals” by freezing
space and time into moving images taken from the real world (Deleuze, 1991, 132).
3D digital quantum cinema renders higher-dimensional virtual space configurations
composed of 3-dimensional moving spaces in time.
It is less known that Dürer also explored shapes of higher-dimensional spaces. He
designed Penrose type irregular five-fold patterns, one of them exactly matching the
Ammann-Beenker Tiling (Dürer, folio 142, in Ruppich, 1969, 73). The astronomer
Johannes Kepler also designed 4-dimensional polyhedra during his search for the
shape of the space with the potential for the formation of matter and the emergence
of motion (Lück, 2000). A closer look tells us that Kepler’s Pythagorean approach
towards the discovery of the three laws of planetary motion lead him in fact into
higher-dimensional space configurations (Quehenberger, 2022).
The history of arts and literature is full of comments on scientific shortcomings.
For instance, in the second half of the nineteenth century, during the hype about the
4th dimension, the poet Edwin Abbott Abbott made fun of limited mathematical
imaginations in his satirical novel about “Flatlanders” who cannot imagine a sphere
(Abbott, 1884). H. G. Wells invented a “Time Machine” (1895) for travelling on
the time axis that inspired relativity theory significantly, while during the cubist art
movement in the 1910s, Marcel Duchamp and Pablo Picasso interpreted aspects of
the 4th dimension by depicting the simultaneity of different perspectives in their
paintings. Inspired by conversations with Henri Poincaré, Duchamp also criticized
the “free choice of metrics” as applied in metric tensors with wooden rulers carved
in the arbitrary shape of dropped threads in his artwork (3 Standard Stoppages,
1913–14).
The famous “melting clocks” (The Persistence of Memory, 1931) paintings by
the surrealist Salvatore Dalí are a reaction to General relativity, wherein “time and
space lose the last remnant of physical reality.” (Einstein, 1916, 117). It appears
that the pseudo 4-dimensional Minkowski space led to a mental loss of a proper
imagination of 4-dimensional systems. Quantum Cinema models are trying to over-
come this deficit. In 2012, we produced the Quantum Fluxus Dada Duett as Art’s
Birthday present, a satirical dadaist sound poem on 100s of different currently
Quantum Cinema and Quantum Computing 233

used space concepts with Erwin Schrödinger’s sentence, “this is absurd” as refrain
(Quehenberger & Weibel, 2012).
We like to recall pataphysique as proclaimed by the French artist and writer Alfred
Jarry at the beginning of the twentieth century. He called for an ideology-free space
in science and announced a science beyond physics and metaphysics, a science of
imagination that underlies everything (Jarry, [1911] 1996). Only recently, SciArt has
emerged as the 3rd culture, opening up new collaborations between art and science.

3.2 The Role of Imagination

According to the nineteenth century German philosopher, Arthur Schopenhauer,

we perceive the world and its representations, we create it as a whole using the
imagination (1819). Schopenhauer, as well as a century later the quantum physicist
Erwin Schrödinger, was inspired by the ancient Indian philosophy of the Vedanta
School with its most important underlying proposition that matter has no essence
independent of mental perception; that existence and perceptibility are convertible
terms (Jones, 1794).
Schopenhauer criticized the abolition of Euclidean geometry by comparing it to
somebody’s cutting off their legs in order to walk with crutches (Schopenhauer, 1892,
vol. I, 17). Since the emergence of many different non-Euclidean geometries during
the nineteenth century, geometry completely lost its authority for proofing algebraic
theories as it? had been the case in antiquity with Euclidean geometry. The result
is that sophisticated algebraic geometry concepts remain abstract. From the artist’s
perspective, it is astonishing that mathematics and physics has no reliable image of
the 4th dimension, not to speak of higher dimensions.
By developing a higher-dimensional Euclidean geometry by means of 3D anima-
tion we have tried to overcome this deficit, which was addressed by the German
mathematician Felix Klein, famous for his Erlangen program, which aimed to unite
all geometries (Klein, 1872)—150 years ago:
The developments of recent decades have meant that in Germany in many cases the focus
has been on abstract, logical inquiries of geometry, whereas the training of the appropriate
outlook has been neglected. (Klein, 1892, 32)

It seems that the training of the imagination for higher-dimensional mathematical

objects has still not been accomplished. Therefore, we claim that it is a task for artists
to employ their visual expertise to engage with this problem of the visualization of
higher mathematics, and computer animation is the appropriate medium here.
The most eminent mathematician, engineer, and physicist around 1900, the poly-
math Henri Poincaré spoke about the usefulness of geometric figures in the theory
of imaginary functions and integrals caught between imaginary limits. He strived for
visual access to higher mathematics that has not even been achieved today.
N-dimensional geometry has a real object; nobody doubts it today. The beings of hyperspace
are susceptible to definitions precisely like those in ordinary space, and if we cannot represent
234 R. C.-Z. Quehenberger

them, we can design them and study them. If therefore, by example, Mechanics with more
than three dimensions must be condemned as devoid of any object, it is not the same with
the hypergeometry (Poincaré, 1895, 1).

What Poincaré called “hypergeometry” in his “Analysis situs” articles later

became the field of topology. The most well-known figures in topology are tori and
bezels, many holed tori, trousers, among other figures obviously unable to provoke
mental pictures of a dynamic discrete higher-dimensional space-time continuum
which we need in order to understand what is going on the Planck scale.
That’s why we engaged in inventing a hypergeometry based on Poincaré’s original
ideas of interconnected spaces with digital artistic tools that allowed the building of
a dynamic spatial continuum by means of a newly (re-)discovered heptahedron, the
3-dimensional representation of the Penrose kites and darts tiling named epita-
hedron (E±) (Quehenberger, 2013; 2014a, 2014b, 2014c; 2016, 2019). Penrose
patterns are irregular patterns of the golden ratio filling the plane with arbitrarily
large regions, with fivefold and tenfold rotational symmetry first designed by Roger
Penrose (Penrose, 1974).
We’ve investigated quantum phenomena by means of geometrical experiments
using these unit cells of 5-dimensional space along with the topology of the (hyper-)
complex number space (Weibel-Katzgraber-Quehenberger, 2012).
Quantum Cinema visualizations are inspired by David Bohm so as to visually and
geometrically test the assumption of the Penrose pattern as an answer to ontological
questions, concerning the implicate order as David Bohm called it:
In short, the implicate order, as I would state it, is the unitary deep structure (holo-archy) of
level-1, which subscends or underlies the explicate surface structures of elementary particles
and waves. (Bohm, 1982, 168)

3.3 How Can We Imagine a Simple Quantum System?

Quantum Cinema research worked on the assumption that the aperiodic Penrose
tiling—a 2-dimensional slice of the 5- or 10-dimensional space (depending on its
fivefold or tenfold symmetries) which is used as models for quasicrystals (Shechtman
et al., 1984) can be adopted to quantum physics (Fig. 1).
The dual of the Penrose patterns is a cubic grid formed by parallels into five distinct
directions of space, named Ammann-bars after their explorer, Robert Ammann
(Grünbaum & Shephard, 1987, 571ff.). This grid is associated with the icosahedral
group, isomorphous to the group A5 (de Brujin, 1981). Penrose tilings are defined
in the 6-dimensional space through a strip and projection method. This gives rise
to the idea that is appropriate for the visualization for the quantum space which
is differing from the usual position and momentum-space. Accordingly, our first
intuitively designed quantum space looks like as shown in Fig. 2.
In quaternionic quantum mechanics, the spin–orbit interaction demands that the
(Euclidean) space–time dimension be 6-dimensional. Pauli correspondence is char-
acterized by the fact that the state space HP1 n S4 of a spin 1/2 particle system admits
Quantum Cinema and Quantum Computing 235

Fig. 1 The convex epitahedron E+ whose edges are partitions of the pentagrid; a polyhedron in
4-dimensional space (image: Quantum Cinema/Renate Quehenberger 2012© Bildrecht, Vienna
2021)

a natural embedding in R5 (Brody and Gräfe, 2011). Also, Louis de Broglie was
convinced of a 5-dimensional wave mechanics (de Broglie, 1927).
These facts seem to support our visualization method based on the Penrose tiling,
the 2-dimensional slice of the 5-dimensional space.
The British physicist Alain Mackay showed that a set of six vectors lies in parallel
to the fivefold rotation axis of the icosahedron (Mackay, 1981). This could possibly
be useful for the hexagonal grids used in quantum computation.

4 The Heisenberg Uncertainty Principle

and the Copenhagen Interpretation

Since the beginnings of modern quantum theory (around 1926) there has been
a dispute about whether or which “images” of quantum events are, in principle,
possible.
We are tempted to regard it as the biggest irony in the history of quantum
mechanics, that the aim for “Anschaulichkeit” (German for “visual accessibility”)
as expressed by Werner Heisenberg in his first published article on the uncertainty
principle (Heisenberg, 1927), turned into the very opposite when the ramifications
of the principle were codified in the Copenhagen Interpretation, which states that the
236 R. C.-Z. Quehenberger

Fig. 2 The complex space model of a particle with straight lines, parallel into 5 directions
constructed from two interlocked icosahedra (image: Quantum Cinema/Renate Quehenberger,
2011© Bildrecht, Vienna 2021)

wave equation is not a real phenomenon but allows only statistical predictions.—How
would a quantum computer work then?
The Copenhagen Interpretation, which appears to be “orthodox” quantum theory
today (cf. Bohr, Heisenberg, and others), called for a ban on images: In principle, the
Heisenberg Uncertainty Principle prohibits thinking about processes and quantum
mechanical formalism in general. This was compared with the Byzantine iconoclasm
of the eight and the ninth centuries, when the “magical power” of images was opposed
and the divine principle was promoted as an obscured power of a non-imaginative
entity. In orthodox quantum theory, formalism itself has ascribed a “magic” (Feynman
et al.), whereby “images” generated in computer models (of causal, so-called “hid-
den” parameters) the result of their materialistic approach, threatening that magic by
demystification (Grössing, 2003).
Various “no-go theorems” were cited, which supposedly proved that such
“images” (which could, via “hidden parameters”, to a certain extent consti-
tute a “deeper” explanatory authority “behind” the observed quantum mechan-
ical processes) are impossible, although the “causal” interpretation after Louis de
Broglie—who first proposed matter waves (de Broglie, 1925), David Bohm and
others—refuted exactly this as viable counterexample.
Quantum Cinema and Quantum Computing 237

Since then a “realism” on the bottom of reality is disputed although proved in a

gedankenexperiment by Schrödinger’s grandson and friends, known as the Barrett-
Rudolph (PBR) theorem (Pusey et al., 2012). Matthew F. Pusey, Jonathan Barrett and
Terry Rudolph have shown that the distributions for |0 and |+ cannot overlap. If the
same can be shown for any pair of quantum states |ψ0  and |ψ1  then the quantum
state can be inferred uniquely from λ. In this case, the quantum state is a physical
property of the system. This results in the introduction of a new no-go theorem for
hidden variables models of quantum theory. They find it surprising that although
quantum states are the key mathematical objects in quantum theory, physicists have
been unable to agree on what a quantum state represents.
Terry Rudolph is currently working on the application of his grandfather’s entan-
glement phenomenon to enable photonic quantum computing. Instead of entangling
individual photons to perform calculations, this technology involves the entangle-
ment of entire light beams to create a “continuous-variable cluster state” over which
computations can take place (Pool, 2020).

4.1 Heisenberg’s Movie Frame Analogy

Perhaps one has to conceive the transition from one stationary state to the next in a similar
way to the transition from one image to another in films. The transition does not take place
suddenly. Rather, the one image becomes weaker and the other slowly emerges and becomes
stronger. For a while, both images permeate each other and one does not know which is
actually meant. Perhaps there is thus a state in which one does not know whether an atom is
in the upper or the lower state. (Heisenberg, 1989, 31)

The young Heisenberg illustrated his ideas about quantum indeterminacy to Albert
Einstein with a movie frame analogy: Maybe we have to imagine the transition from
one stationary state to the next in a similar way like the transition from one picture
to the next in some films. There are not enough discrete frames available in order to
achieve a clear image and the same happens with measurement. Only an ensemble
of particles allows the appropriate measurement outcome, just as you need a series
of images to make pin-sharp movie pictures (Heisenberg, 2011 [1979], 39).
Heisenberg’s analogy argument that we will never be able to see the still picture
between one and the next film frame in terms of mathematical “Anschaulichkeit”,
was the topic of our geometrical experiment applied to a geometrical model.
In the Quantum Cinema project, Heisenberg’s movie frame analogy was applied
to moving straight lines, partitions of the 6-dimensional space grid, which must be
considered to be in constant movement. The still picture shows all possible positions
of the points on the two circles merged in the white lines. Our geometrical digital
experiment comprised of possible positions of two points connected to two circles
in blue and red. The straight lines rotating at different speeds generated traces of
white lines that can be interpreted as probability distribution for the localization of
a particle, as shown in the still images (Fig. 3).
238 R. C.-Z. Quehenberger

Fig. 3 3D animated digital experiment: slowly moving centre points of circles create distorted
white lines (b) the traces of fast movement points creating a distorted bundle of white lines (image:
Quantum Cinema/R. Quehenberger 2012© Bildrecht, Vienna 2021)

We take two circles (a, b) with centre points A and B on a triangle. The two circles
perform rotations over these 3 points. The distance AB can be regarded as minimum
distance h. Thereby the points on the circle outline run into turbulence and the line is
no longer a straight line but generates different forms of loops on the circle line. Any
point and position of the point on this loop will be in a fast wavelike motion. One
cannot “measure” both the position and momentum of the point if the simulation is
played at full speed.
Any point and position of the point on this loop will be in fast motion so that
one cannot measure the position and momentum simultaneously. But Heisenberg’s
idea that the still image between one and the next picture cannot be seen in terms of
mathematical visual access turns into something different: in 3D digital geometry,
one can see all images in-between if the movie is run in very slow motion. The still
image shows all possible positions of the points on the two circles merged in the
white lines.
The Quantum Cinema 3D digital geometry examples on the uncertainty principle
were actually inspired by Johannes Kepler’s concept of quantum and fluxus with
the idea of “exire”—points in motion that create lines. He was relying on Proclus’
dynamic interpretation of Euclid’s “Elements” which was also the formation principle
of the hyper-Euclidean geometry we have supposedly only (re-)invented.
Alan Mackay also mentioned that unfortunately, Kepler lived too early to learn
that the five regular polyhedra are in fact (as we now believe!) eigenfunctions of
Schrödinger’s equation and thus do describe the symmetries of the planetary electrons
surrounding an atomic nucleus (Mackay, 1990, 3).
Based on our higher-dimensional geometry, we reached the conclusion that the
cause for the Heisenberg indeterminacy lies in the fact that higher-dimensional space
itself has to be regarded as being in constant motion. On a moving background of a
6-dimensional space grid, it is impossible to fix a moving particle and measure the
position and momentum of an object.
Quantum Cinema and Quantum Computing 239

5 Up, Up to Higher Dimensions

The discussion on higher dimensions goes back to Ancient Greece, while evidence
for higher-dimensional space models can be found in ancient Egypt. Meanwhile,
Plato’s 5th element could be identified as unit cells of the 5-dimensional space
(Quehenberger, 2016).
Plato’s description of the composition of triangles that form the well-known
Platonic solids starts with the tetrahedron, the octahedron and the icosahedron, and
ends with the cube. He does not mention the dodecahedron, which does not fit into his
concept because it does not consist of triangles and was ascribed only to his scholar
Theaetetus, who also proved that there are only five regular convex polyhedra, the
Platonic solids, before Euclid. Plato himself added instead this obscure remark:
[...] since there is another fifth configuration the god used it for the delineation of the
universe.” (Plato, Timaios, 55c)

Only the possible compositions of dodecahedra from the 3D representation of the

Penrose Kites and Darts make it clear that he kept the secret, which he and Pythagoras
probably found in ancient Egyptian mystery schools. Only by not telling the name
of the 5th regular solid, he left it open that the shape of the 5th element leads to a
range of dodecahedral spaces in three and more dimensions which may well be used
as models for the Universe, as Poincaré suggested in 1904.
We should not forget that the modern Standard Theory describes particles in terms
of Group theory, which relies on the symmetries of Platonic solids conforming Plato’s
idea that the formation of matter is caused by traces of shapes in motion.
We also have to consider that Group theory requires a 5-dimensional space, as
Poincaré pointed out:
An immediate induction shows us that the classification of algebraic surfaces and the theory
of their birational transformations are intimately connected with the classification of real
closed hypersurfaces in the space of five dimensions from the point of view of Analysis
situs. (Poincaré, 1900, 19)

Echoing Pythagorean harmonies, Alan L. Mackay, who coined the term “general-
ized crystallography”, anticipated the existence of aperiodic fivefold crystals before
the discovery of quasicrystals, revealed:
What the Pythagoreans really discovered was group theory and of course this can be applied
to a vast range of circumstances. (Mackay, 1990, 3)

At this point, we have to consider that Plato was visualizing Pythagoras, and
both are evidently relying on ancient Egyptian mystery schools and mythology that
explains the origami of 5-dimensional space conveyed as sex and crime stories about
gods and goddesses (Quehenberger, 2013, 2019). Pythagoras understood the signif-
icance of making the connection between the sensory aspects of our world and the
realm of numbers with the goal of realizing the pre-established harmony of the
cosmos (Miller, 1996, 181).
240 R. C.-Z. Quehenberger

Pythagoras’ theorem also plays a key role in quantum mechanics. It is used to

find the length of the vector when complex numbers are used in quantum mechanics.
It is fundamentally impossible to predict with certainty the outcome of a future
measurement on a system prepared in a pure state, unless the state is an eigenstate
of the observable to be measured. The best prediction is probabilistic, given by the
Born rule, which states that the probability density of finding a particle at a given
point, when measured, is proportional to the square of the magnitude of the particle’s
wave function at that point. Hence, the Born rule is a consequence of the complex
Pythagorean theorem: the measure of a subset of the complex line is the sum of the
measures of its projections on all eigenspaces (Mandolesi, 2019).
In Euclidean geometry, the Pythagorean theorem is represented among the three
sides of a right triangle. It states that the area of the square whose side is the
hypotenuse (the side opposite the right angle) is equal to the sum of the areas of
the squares on the other two sides. The probability in quantum mechanics must then
take the form Pc = Pa + Pb , since the two possibilities are mutually exclusive due
to orthogonality of measuring either the value associated with the subspace A or the
value associated with the subspace B, is given by: |c|2 = |a|2 + |b|2 . The quaternions,
discovered by Hamilton, are a 4-dimensional algebra, H = {a + bi + cj + dk: a, b,
c, d √∈ R}. It is a normed division algebra, meaning that the norm |a + bi + cj +
dk|= (a2 + b2 + c2 + d2 ) obeys |qq | = |q||q | for all q, q ∈ H. The unit sphere in H,
a 3-sphere, is, therefore, a group (Baez, 2017).
The dynamics of quaternions remind us of the Greek term dynamis. In his dialogue,
Theaitetos Plato speaks of “dynamis”, sometimes translated as square roots, to indi-
cate the transformation from a quadrangle (tetragonos) to a square, of an imaginary
meaning and also that these numbers possess a potential and dynamics for returning
to the real numbers (Tydecks, 2020).
Even though the particulars of Plato’s ‘theory of everything’ as outlined in Timaeus, strike the
modern sensibility as absurd (if not borderline psychotic), there are many parallels between
his picture of the universe and that embodied in string theory. (Yau & Nadis, 2010, 326)

The mathematicians Shing-Tung Yau and Steve Nadis mention that geometriza-
tion, the idea that the physics we observe springs directly from geometry, stands at
the core of both approaches. Would they be surprised to read that Plato’s 5th element
could be composed as 26-dimensional space, as required by String Theory for one
particle to inhabit in order to achieve the unification?
On digesting this higher-dimensional information from Plato’s world of ideas,
we may well convince ourselves that the Pythagoreans also worked with this
higher-dimensional space concept that was lost in religious doctrines in the ensuing
centuries.
Higher dimensions were again considered at the turn of the seventeenth century
by Renée Descartes, who was the first to introduce the notion of higher dimensions
in relation to equations of higher order (Descartes [1637] 1902, 13). Descartes is
well-known for the x, y coordinate system, however, as early as 1610, he also stated
that a combination of systems of coordinate axes is a prerequisite for the depiction of
higher-dimensional spaces. He worked with the German weaver Johann Faulhaber,
Quantum Cinema and Quantum Computing 241

who also produced calculations for Johannes Kepler—on higher order simplices.
Faulhaber worked on cubic equations based on Girolamo Cardano, who had first
published Tartaglia’s solution a century earlier (1545) and defined formulas for sums
up to the 17th power, which is known as the Faulhaber Formula (Faulhaber, 1631).
Faulhaber’s work is still relevant for computation because his works influenced
Jakob Bernoulli, and Bernoulli numbers are the core ingredients of the first so-called
software program of Ada Lovelace that she developed for the Analytical Engine
designed by Charles Babbage.
The Analytical Engine weaves algebraic patterns like the Jaquard loom weaves flowers and
leaves. (Lovelace, in Padua, 2015, 88)

It is notable that punched cards were invented for weaving on a loom in 1804
by Jean Marie Jacquard and were once commonly used in data storing and data
processing applications or to directly control automated machinery even after the
development of electronic computers in the 1950s and until recently.

5.1 Higher-Dimensional “Quantum Stuff” and the Curse

of Dimensionality

Until now, dimensionality in computational software programs like mathematica is

solved by assigning different degrees of freedom from 1-dimensional lines but not
from a spatial compound. In contrast to this, we took inspiration from Henri Poincaré,
who was primarily interested in the physical and philosophical implications of the
meaning of the concept of dimension, states:
The fundamental proposition of Analysis Situs is, that space is a continuum of three
dimensions. (Poincaré, 1912)

He inspired the Dutch mathematician Luitzen E. J. Brouwer, who founded intu-

itionism to a dimension concept based on “lawless sequences” (German: Wahlfolgen).
We assume that we can apply this concept to our hyper-Euclidean space compounds
of interconnected spaces that can be ascribed to a certain dimensionality (Brouwer,
1913).
Alas, there is currently no appropriate concept of dimensionality that can be easily
applied to digital computation. For this lack of functionality, the term "curse of
dimensionality” was coined by Richard E. Bellman. The problems arose in dynamic
programming when analyzing and organizing data in high-dimensional spaces was
not possible, but do not occur in low-dimensional settings, such as the three-
dimensional physical space of everyday experience (Bellman, 1957). Consequently,
in computer science, higher dimensions are avoided by the process of dimension
reduction. It is the transformation of data from a high-dimensional space into a
low-dimensional space so that the low-dimensional representation retains some
meaningful properties of the original data. While this method seems viable for
signal processing, its application in natural sciences—e.g., in bioinformatics—seems
242 R. C.-Z. Quehenberger

rather problematic, because important information concerning complex structure

components gets lost.
Although applying our continuous higher-dimensional space concept to computer
programs remains a challenge.
The founding fathers of quantum physics, Erwin Schrödinger and Louis de
Broglie, had good imaginations for quantum phenomena taking place in a higher-
dimensional configuration space:
One actually no longer has to assign a psi function to this totality of particles, which represents
a wave propagating in ordinary 3-dimensional physical space but a psi function that represents
a wave propagating in abstract space, the configuration space, which has three dimensions
as many times as there are parts in the system. (de Broglie, 1943, 145f. Transl. RCZQ)

Erwin Schrödinger, who first proposed entanglement imagined the wave func-
tion—which he formulated for the first time—as the description of a phenomenon
located in higher-dimensional space. He regarded the purely statistical Copen-
hagen interpretation, which takes the wave function merely as probability, intro-
duced by Max Born after Heisenberg discovered quantum indeterminacy, as a
terrible misinterpretation of his theory. This development of quantum theory went
in a completely different direction than Schrödinger had wanted. Abstract statistics
doesn’t demand any kind of imagination. Even the Nobel laureate Gerard’t Hooft
admitted, speaking about higher dimensions, “I leave them out whenever I can” (‘t
Hooft in QC-Interview, Quehenberger, 2011).
Contributing to its complexity, the American physicist Richard Feynman was the
first to imagine a quantum computer that would be able to calculate the simulation
of complex quantum systems:
But the full description of quantum mechanics for a large system with R particles is given
by a function ψ(x1 , x2 , . . ., xR , t) which we call the amplitude to find the particles x1 , . . .,
xR , and therefore, because it has too many variables, it cannot be simulated with a normal
computer with a number of elements proportional to R or proportional to N. (Feynman,
1982)

Only recently, the discussion on higher dimensions gained new momentum. In the
operational approach to general probabilistic theories, one distinguishes two spaces,
the state space of the “elementary systems” in the quantum world and the physical
space in which “laboratory devices” are embedded. Each of those spaces has its own
dimensionality. This means there is a different minimal number of real parameters
(coordinates) needed to specify the state of the system and for a point within the
physical space (Dakic and Brukner, 2016).

6 Schrödinger’s “Entanglement”

Entanglement is the most important quantum phenomenon applied to quantum

encryption since entangled particle pairs serve as the primary technical component
in quantum computers.
Quantum Cinema and Quantum Computing 243

The rather romantic term entanglement was coined by the Austrian physicist—
and notorious womanizer who lived with two wives—Erwin Schrödinger, whose
name is tied to the famous Schrödinger wave equation.
Schrödinger tried to understand the possible unification of matrix mechanics by
Heisenberg, Jordan algebra and his wave mechanics, since partial differential equa-
tions are equivalent to quadratic forms of infinitely many variables (Schrödinger in
a letter to Born, April 14, 1926, in Meyenn, 2010, 201).
Finally, Paul Dirac’s attempt to prove the equivalence of matrix mechanics and
wave mechanics by essential use of the delta function and his bra-ket formalism
based on hypercomplex numbers similar to William Rowan Hamilton’s quaternions
led to a deeper understanding of the foundations (Dirac, 1930). As artists, we are
trying to understand mathematical abstract formalism in a way that can be visualized
geometrically. We are less focussed on measurement outcomes than on ontological
questions—what is really happening on the subscendent level-1?
In the information interpretation of quantum mechanics, information is the most
fundamental basic entity (Svozil, 2000). Every quantized system associated with a
definite discrete amount of information can be traced back to Erwin Schrödinger’s
“catalogue of expectation values” mentioned in his seminal papers “On the Current
Situation of Quantum Mechanics” (Schrödinger, 1935). Here, he also introduced the
famous simultaneously dead and alive “Schrödinger’s cat” metaphor.
In this drastically cruel thought experiment about a quantum cat in a closed
box with cyanide, while hidden from the observer, the cat is dead and alive at the
same time. It is intended to serve a better understanding of the phenomenon of the
superposition of a quantum state.
Well, ’state’ is a term that everybody uses today, even the holy PAM, but it doesn’t make it
more rich in content (Schrödinger in a letter to Pauli, Dublin, beginning of July 1935, in
Meyenn, 1985, 406)

The “holy PAM” is Paul Adrien Maurice Dirac, the British physicist who first
defined the spin of a particle. Wolfgang Pauli famously said of him: “There is no
God and Dirac is his prophet.” (Pauli, in Meyenn, 2010, 406).
Schrödinger’s (1935) articles were immediate answers to Albert Einstein’s
question “Can quantum–mechanical description of physical reality be considered
complete?” wherein he assumed a local deterministic realism of quantum states
(Einstein et al., 1935).
Schrödinger also reacted to the forbidden determinism by introducing the term
“entanglement” for the undetermined state depending on an earlier random event
instead of Einstein’s “spooky action at a distance” of which Isaac Newton had already
stated:
I believe no man who has in philosophical matters a competent faculty of thinking, can ever
fall into it. (Newton in a letter to Richard Bentley in 1693, quoted by Price, 2013)

Entanglement is a property of correlations between two or more quantum systems

that defy classical description.
244 R. C.-Z. Quehenberger

Quantum non-locality does not prove that “signals” travel “faster than light”.
Rather, it shows that at a deep level of reality the speed of light as a limiting factor is
irrelevant because phenomena are instantaneously connected regardless of distance.
Because the spin of a particle does not exist until a measurement is made, the act
of making the measurement and determining the axis of spin of particle 1 will also
determine the spin of particle 2 no matter how far apart it is from particle 1. Particle
2 will instantly respond to the state of particle 1, even if it is on the other side of the
universe.
At any instant measurement of particle 1 is performed, particle 2, which may be
thousands of miles away, will acquire a definite spin—“up” or “down” if a vertical
axis, “left” or “right” and a horizontal axis is chosen. How does particle 2 know
which axis was chosen? There is no time for it to receive that information by any
conventional signal (Capra, 1982, 85).
Wolfgang Pauli approached the entangled quantum system as follows:
One must differentiate between different layers of reality, an R that contains all the informa-
tion that can be obtained from measurements on 1 and one (deducible from R) r which only
contains the information which can be obtained by measurements on 1 alone.
A complete description would have to assign characteristics to the state of the particle
1 which must already contain all those properties of 1 which — according to possible
measurements on 2, which do not interfere with 1 — can be predicted with certainty. (Pauli
in a letter to Heisenberg, 1935, in Meyenn, 1985, 404)

6.1 A Proposal for a Space that Enables Schrödinger’s

“Entanglement”

The effect of the environment through decoherence becomes more accessible in the
artistic interpretation of a higher-dimensional framework based on the Penrose tiling
shown above:
Let us assume that for any interconnected system of two or more measurable
quantum objects of higher order, say qubits (no matter how far they are separated),
the “miracle” of a “minimal distance effect” occurs in a spatial framework that
exhibits principles of the aperiodic Penrose tiling. For this, Conway’s theorem holds
that for any circular region of diameter d, the distance from the perimeter of the
“home town to the perimeter” of the duplicate town is never more than d times half
of the cube of the golden ratio, or 2.11 times d.
If you walk in the right direction, you need not go more than that distance to find yourself
inside an exact copy of your home town. The theorem also applies to the universe in which
you live. Every large circular pattern (there is an infinity of different ones) can be reached
by walking a distance in some direction that is certainly less than about twice the diameter
of the pattern and more likely about the same distance as the diameter. On a Penrose pattern
you are always very close to a duplicate of home. (Gardner, 1989, 10)

Following this assumption, we derive the 3D representation of the entanglement

picture as interlocked spaces of two unit cells of the 5-dimensional space, supposedly
Quantum Cinema and Quantum Computing 245

what Plato called a “figure of marriage”. In his symposium, Plato let Socrates quote
his teacher Diotima, “Love is the medium”.

6.2 Entangled States in Quantum Computing

The strikingly nonclassical properties of entangled states lead to more and more
novel demonstrations of fundamental properties of quantum mechanical systems
(Greenberger-Horne-Zeilinger, 1993). The young field of quantum information
exploits such entangled quantum states for new types of information transmission
and information processing (Bennett, 1995).
In 1995, the Austrian physicist Anton Zeilinger and co-workers reported
the first experimental realization of quantum communication, verifying the
increased capacity of a quantum information channel by “quantum dense coding”
(Mattle, Zeilinger et al., 1996). The applied scheme was theoretically proposed
by Bennett and Wiesner and utilizes two entangled two-state systems (Bennett &
Wiesner, 1992).
The output of a quantum information processor is, by convention, measured in
the basis states with Boolean values |0 and |1 denoted as A and B, often referred
to as Alice and Bob. Any controlled unitary gate can be realized using at most two
CNOT gates and four single-qubit gates. This acts on the state of two qubits, known
as the control qubit, C, and the target qubit, T (Barnett, 2009, 146). √
Typically, four specific two-qubit states with the maximal value of 2/ 2 are desig-
nated as Bell states. These Bell pairs represent all the ways that can be chosen to have
correlations or anticorrelations between the |0 and |1 states of the computational
basis and the |+ and |− states of the X basis.
These four states form a complete basis for two qubits. So any state, entangled or
not, can be expressed as a superposition of Bell states. They are known as the four
maximally entangled two-qubit Bell states and form a maximally entangled basis
known as the Bell basis of the 4-dimensional Hilbert space for two qubits.
From the perspective of quantum information theory, any two-state quantum
system (represented by a0 |0 + a1 |1) can be used to provide a physical imple-
mentation of a qubit like the bit in classical information theory. Examples that have
been realized in the laboratory include the two orthogonal polarization states of a
photon, the orientation of a spin-half particle, a pair of electronic energy levels in an
atom, ion, or quantum dot, and a pair of paths in an interferometer. However, it is
convenient to treat a qubit as though it were a spin-half particle and to describe it by
Pauli operators (σ_x, σ_y, σ_z) (Barnett, 2009, 46).
246 R. C.-Z. Quehenberger

6.3 A Space Composed of Cells

In the 1950s, the physicist Léon Brillouin, who applied information theory to physics,
introduced space cells on the fundamental level of reality. Before he made major
contributions to crystallography by finding the elementary, irreducible zone, the
first Brillouin zone reduced by all symmetries in the point group of the lattice.
He sketched a “miscellaneous” cell on the foundations of the quantum realm with
information content per element of data (Brioullin, 1952, 72). The term “miscel-
laneous” stands for a compound of elements in movement which also applies to
the 6-dimensional space lattice and inherent symplectic manifolds in movement.
We envisage a compound of dynamic elements (points and straight lines) forming
5-simplices, 3D representation of the Penrose Kites and Darts tiling (E±). All parts
of these unit cells of the 5-dimensional space, its points, edges and faces have to be
thought of as being in constant movement, forming different angles and composi-
tions. We imagine iterations of the size of the lattice forming simplices with parti-
tions of circle decoration (cf. Conway decoration of the Penrose kites and darts
tiling) forming wavelike structures in movement. Their traces would, according to
Plato, lead to the formation of matter. We imagine that a particle as expressed by
Schrödinger’s wave equation is the result of permutations of space cells according
to symmetry operations in accordance with group theory.
Our Platonic approach seems to meet symplectic Clifford algebra as interpreted by
Basil Hiley: The basis is not matter or fields in interaction in space–time, but a notion
of “structure process” from which the geometry of space–time and its relationship to
matter emerge together, providing a way that could underpin general relativity and
quantum theory (Hiley, 2016). All the quantum properties of Schrödinger and Pauli
particles can be described entirely from within a Clifford algebra taken over the reals,
there is no need to resort to any “wave function” (Hiley & Callaghan, 2010).
Such a space in motion would not permit the measurement of the momentum
and position of a particle, and provides a logical geometrical framework for the
Heisenberg uncertainty principle: Higher-dimensional space, as we know it from
visualizations of hypercubes, is in motion.
In this respect, the dynamic structure of the infinite 5-dimensional space may be
a visualization of the “pattern that connects” as coined in cybernetics by Gregory
Bateson. Only by means of 3D animated geometry can we visualize the idea of
dynamic building blocks for systems like “Jumbled Boxes”, as Gordon Pask drew
them (Clarke, 2009, 50).
Various Unified Theories can be visualized by means of this 3D representation and
its composite configurations, such as of the 26-dimensional space continuum arising
from inside the 4-dimensional dodecahedron, identified as Poincaré’s dodecahedral
space (Quehenberger, 2018).
However, we do not expect that this model will be accepted anytime soon since the
physics community remains agnostic about an underlying space. In Quantum Gravity
emergent space–time scenarios, space and time are denied at the fundamental level
of reality (Oriti, 2006). In our model “time” is represented in the 4th dimension of
Quantum Cinema and Quantum Computing 247

reality on the globe, while if we apply the same model to quaternionic quantum
mechanics we get quantum space–time crystals, as proposed by Nobel Laureate
Frank Wilczek (2012). This system, which is periodic in time and also in space, was
realized experimentally by confining ions in a ring-shaped trapping potential with a
static magnetic field soon thereafter.

7 Hidden Parameters: Never Neglect a Woman (Grete

Hermann)

Progress in quantum mechanics could have been achieved much faster if John von
Neumann’s Mathematical Foundations of Quantum Mechanics (1932) would have
been less determining. His proof that no hidden variables can exist held for decades
until it was finally refuted by John Bell’s with his non-locality theory on hidden
variables (Bell, 1966). Everybody continued to cite the von Neumann proof until
John Bell (Bell, 1964) rediscovered that von Neumann’s no hidden variables proof
was based on an assumption that “can only be described as silly […] so silly […]”
(Mermin, 1993).
But back in 1935, the German philosopher and mathematician Grete Herman
already disproved John von Neumann’s proof that no hidden variables can exist.
She pointed out a glaring deficiency in his argument, however, she seems to have
been entirely ignored. When the Einstein–Podolsky–Rosen paper was published,
Wolfgang Pauli asked his German colleague Werner Heisenberg to write a “pedagogic
response” concerning the determinism in quantum mechanics to the “old boys” (Pauli
in a letter to Heisenberg, Zurich, 15 June, 1935 in Meyenn, 1985, 412).
Heisenberg replied that Grete Hermann already had already written a treatise
which he found “overall reasonable (despite a philosophical rational tendency)”
(letter from Heisenberg to Pauli, Leipzig, 2 July 1935, Meyenn, 1985, 408).
Grete Hermann was the first physicist who asked for the ontologic grounds of
observables. With her philosophical request to find other characteristics which deter-
mine the path of a particle, we should not stop searching for underlying causes.
She called for an improvement of the quantum mechanical formalism (Herzenberg,
2008):
We need to find the causations of the measurement results and predictions — there is only
one sufficient reason to stop looking for it — that you know already the source. (Hermann,
1935)

Grete Hermann distinguished between causality and predictability, and empha-

sized the fact that they are not identical:
The fact that quantum mechanics assumes and pursues a causal account also for unpredictable
occurrences proves that an identification of these two concepts is based on a confusion.
(Hermann, 1935)

Grete Hermann showed that causality was retained in the sense that after an
interaction, causes could be assigned for a particular effect. This allows for the
248 R. C.-Z. Quehenberger

possibility that physical processes may be strictly determined even though the exact
prediction is not possible (Lenzen, 1969).
Her work with Werner Heisenberg, Carl Friedrich von Weizsäcker and other
prominent physicists during the early 1930s led Hermann to the idea of a rela-
tional interpretation of quantum mechanics. This term refers to an interpretation of
quantum theory that eliminates the concepts of the absolute state of a system, which
was only recently developed more fully by Carlo Rovelli. Niels Bohr arrived consid-
erably later at the relational concept of quantum mechanical states by considering
the Einstein–Podolsky–Rosen thought experiment, and he became known for this
approach, while Grete Hermann’s work in introducing it has been all but forgotten
(Jammer, 1974, 208).
Hermann has studied in Göttingen under Emmy Noether, one of the iconic figures
of twentieth century mathematics, who revealed the very important general connec-
tion between symmetries and conservation laws in physics, known as Noether’s
theorem.
In her thesis, Grete Hermann demonstrated that Noether’s proof of the Lasker–
Noether theorem could be turned into polynomial ideals in Noetherian rings. This was
the first efficient algorithm for computing primary decompositions for polynomial
rings over a field of characteristic 0 (Hermann, 1926).
Hermann completed her work before the availability of computers, and even before
the idea of an effective procedure had been formalized. Algorithms that compute
primary decompositions are very useful nowadays for quantum algorithms.

7.1 More Ignorance and Paradoxes

Quantum computing is ironically built upon a paradox: Although “hidden param-

eters” were “forbidden” for decades, they became the foundations of Quantum
Information.
Two decades after Grete Hermann, David Bohm found that in principle, these
“hidden” variables determine the precise results of each individual measurement
process. In practice, however, in measurements that we now know how to carry
out, the observing apparatus disturbs the observed system in an unpredictable and
uncontrollable way, so that the uncertainty principle is obtained as a practical limi-
tation on the possible precision of measurements. Although widely applied, John
von Neumann’s description of measurement is insufficiently general for the simple
reason that most performed observations are not of this type. The real world is noisy,
and this ensures that our observations will include errors (Barnett, 2009, 92).
David Bohm pointed out that von Neumann’s proof that quantum theory is not
consistent with hidden variables and does not apply to his interpretation, because the
hidden variables he speaks of depend both, on the state of the measuring apparatus and
the observed system and so his theory would go beyond von Neumann’s assumptions
(Bohm, 1952).
Quantum Cinema and Quantum Computing 249

Ironically, Bohm’s hidden variable theory and Bohmian mechanics were falsely
declared as deterministic and was even “verboten” (forbidden) to be taught in recent
decades (Basil Hiley in QC-interview, Quehenberger, 2012a). In its original form, the
mathematical formalism of the de Broglie-Bohm theory refers to hypothetical ontic
elements, such as the quantum potential comparable with John Bell’s "beables”. John
Bell was inspired by his teacher David Bohm, and his idea of an underlying reality
that we can imagine like a pattern in movement (Bohm, 2004, 92).
“Bell’s inequality” and separately the Kochen–Specker theorem by Simon Kochen
and Ernst Specker (1967) first demanded contextuality to be a feature of quantum
phenomenology was a correction of the history of quantum mechanics that opened
up a new field of research, namely quantum information—and Bell’s non-locality
theorem became the basic feature of quantum computation.
Also, the Kochen–Specker theorem demonstrates that it is not possible to repro-
duce the predictions of quantum theory in terms of a hidden variable model
where the hidden variables assign a value to every projector deterministically and
non-contextually (Kunjwal & Spekkens, 2015).
When John Bell started to work on the foundations of quantum mechanics, there
was hardly any interest in such topics. Even worse, working on foundations was
not considered to be a proper topic for a physicist, so the CERN employée wrote his
seminal article during his sabbatical leave, as his then-colleague Reinhold Bertlmann
tells: John Clauser, the first who had the courage to carry out an experiment on
Bell inequalities in the 1970s had to struggle enormously to get the resources for
doing the experiment. Only after the experiments of Alain Aspect in the 1980s the
community slowly began to realize that there was something essential to Bell’s s
theorem. In the 1990s, the field of applications of entangled states and of Bell’s
theorem opened up, followed by experiments on quantum teleportation, quantum
cryptography, long-distance quantum communication and the realization of some
of the basic entanglement-based concepts in quantum computation (Bertlmann &
Zeilinger, 2017, vi).
Aspect was following Bell’s example by using entangled photons instead of elec-
trons, which proves physical reality must be un-local and all results show the impossi-
bility of a Local Realistic interpretation of quantum mechanics (Aspect et al., 1981).
After that, the perfect anti-correlation of entangled pairs has been repeated many
times since the easiest method of producing entangled states was achieved by Anton
Zeilinger’s group after their entangled photon experiment (Kwiat et al., 1995). The
pairs of polarization-entangled photons were produced by type-II down-conversion
in a nonlinear beta-barium-borate crystal that was incidentally found in the desk of
their colleague, Ralph Höpfel. The angle of 0.72° in their experimental set-up seems
to indicate the 72° of the 5-dimensional space grid.
The second obstacle seems to be the lack of realism attributed to the quantum
world. How could we build technology on something the existence of which is
questioned?
The realist quantum approaches often distinguish between ψ-ontic, ontological
quantum theories and ψ-epistemic theories (Fraassen, 1972). “Ontic states” are
250 R. C.-Z. Quehenberger

what Terry Rudolph calls real states versus “the mist (quantum state) is just knowl-
edge/information about the real state” referred to as “psi-ontic” and “psi-epistemic”,
respectively (Rudolph, 2017). In other words, type-I ψ-ontic interpretations are
defined as those in which the probabilities of measurement outcomes are determined
by intrinsic properties of the observed system. The other, ψ-epistemic, theories view
the quantum state as merely representing knowledge of an underlying objective
reality in a sense somewhat analogous to that in which a state in classical statistical
mechanics assigns a probability distribution to points in phase space (Cabello, 2017).
Schrödinger was considering to assign a physical realism not to the wave function
ψ itself but to the quadratic function of ψ (Schrödinger in a letter to H. A. Lorentz,
June 6, 1926, in Meyenn, 2010, 254ff.).
Twenty-five years ago, David Bohm and Basil Hiley described the close resem-
blance between key properties of deterministic, hidden variable, pilot-wave theory
and emergence theory in the book The Undivided Universe (1993). Their theory
describes the emergent formation of ordered (i.e., negentropic) states in nonlinear,
self-organizing systems, such as deterministic chaos (Walleczek et al., 2019).
David Bohm’s philosophy of the undividable wholeness where there are “no
separate whirlpools interacting” was inspired by the Fourier Universe where the
subscendent space is a carpet-like pattern of frequencies. This enfolded [or implicate]
order contains the general totality of what is called the “explicate” or “unfolded” order
(Bohm, 1980, xv).
We believe that the fractal nature of the Penrose tilings and their 3D representations
where each triangle is decorated with a fraction of a circle, as suggested by John
H. Conway for the Kites and Darts tiling, is appropriate to visualize this pattern
of frequencies. Similarly, like the triangles with curves, the underlying reality as
conceptualized by String theorists consists of fractions of flat space with a string
attached, named Dirichlet branes (Polchinski, 1995). This idea seems to meet our
concept with the triangular faces of epitahedra with the 1-dimensional “string-like”
circular decoration. Joseph Polchinski also found that the product of the electric and
magnetic charges is a single Dirac unit, and that the quantum of charge takes the
value required by string duality.
If we assign some realism to the straight lines and attribute electric force to them,
(like Maxwell and Faraday’s lines of force) and magnetic force to circles, we derive
the image of the wavelike nature of reality suggested by Joseph Fourier and Bohm.

8 The Geometry of Quantum States

In their seminal book on Quantum Geometry, physicists Ingemar Bengtsson and

Karol Zyczkowski pointed out that Quantum mechanics is a significant generalization
of classical probability theory and show how quantum states can be represented by
geometrical elements (Bengtsson & Zyczkowski, 2006).
Quantum Cinema and Quantum Computing 251

Usually, the observables are pictured as points on the surface of a sphere, the
Bloch sphere—it is easy to visualize, but in some respects, it is quite misleadingly
simple (Bengtsson & Zyczkowski, 2006, 215).
It is a sphere of unit radius, with each point on its surface corresponding to a
different pure state. A given observable consists of two antipodal points on the surface
of the Bloch ball representing a pair of mutually orthogonal states two orthogonal
pure states. The north and south poles correspond to the quantum states so that the
four possible results 00, 01, 10 and 11 correspond to measurements of the two qubits
as the Bell basis.
During the Quantum Cinema research project, we were trying to find a more
detailed 4-dimensional representation of such a quantum system.
There are examples of quantum states depicted as compositions of nested Platonic
solids: Partially transposed states occupy the remaining four corners of the unit cube,
which is filled with two tetrahedra in an antipodal position. Two states are regarded
as “equally entangled” if they differ only by a choice of basis in Hilbert spaces H1
and H2 . The separable subset in 2 × 2 dimensions is the intersection of two tetrahedra
and can be represented by an octahedron (Vollbrecht and Werner, 2000).
Another representation of probabilities of maximally entangled states by using
an octahedron inside a tetrahedron is quite similar to our intuitive “entanglement”
image above (Fig. 4), where two epitahedra form the dodecahedron as Boolean
intersection. We may assume that this emerging dodecahedral space can be compared
to the maximal Kús-Zyczkowski ball of absolutely separable states that lies within
the double pyramid. It touches the faces of the pyramids at the four points which
mark the closest separable states to the four Bell states. The entangled Weyl states are
located in the four corners of the tetrahedron outside the double pyramid, extending
from the separable states to the maximally entangled Bell states at the tip (Bertlmann-
Friis-Bulusu, 2016).

8.1 A New Object in Quantum Geometry

There are several indicators for fivefold symmetries in the quantum system. Most
famously Klyachko and co-workers assigned a 5-vector an orthogonality graph in the
form of a pentagram, and in this way derive a Kochen–Specker inequality for spin
1 systems where the set of all functions forms a Vorob’ev-Keller cone (Klyachko
et al., 2008). This 5-vector can be visualized by the 5 edges of the pyramidal cone of
the 3D representation of the Penrose Kites and Darts tiling E±. We assume that E±
is a golden 5-simplex that corresponds to the unique optimal configuration of C5 , of
which it is said that it looks like an umbrella (Bharti, Cabello et al., 2019).
We maintain that Mermin’s magic pentagram as well as several other contextu-
ality proofs can be drawn back to higher-dimensional spatial structures, namely the
two antipodal entangled simplices (Quehenberger, 2017). The here presented entan-
gled simplex representation (2E±) (Fig. 5) also seems to conform to the Penrose
Dodecahedron as contextuality proof, an image that goes back to Penrose’s idea of 5
252 R. C.-Z. Quehenberger

Fig. 4 “Entangled state” of two epitahedra with a dodecahedron in the centre; 3D animation of
rotating faces result in full circles of the Conway circle decoration of the Penrose tiles, (image:
Quantum Cinema/Renate Quehenberger, 2011© Bildrecht, Vienna 2021)

Fig. 5 3D Representation of
a three-partite quantum
system indicating spin states
in the framework of the
Poincaré sphere. The sphere
is drawn with three great
circles and labels for six
basic polarization H (linear
horizontal), V (linear
vertical), D (linear diagonal),
A (linear antidiagonal), R
(right-hand circular) and L
(left-hand circular) and
images of the polarization
vectors for each where the (i,
0) (image: Quantum
Cinema/Renate
Quehenberger, 2017©
Bildrecht, Vienna 2021)
Quantum Cinema and Quantum Computing 253

interlocked cubes (Zimba & Penrose, 1993). In our “entanglement figure”, the apexes
of these two irregular pyramids (E±) with dynamical edges from partitions of the
pentagrid may be assigned to the spin up ↑/right and the spin down ↓/left, the other
two measurements would then show 01, 10 (see Fig. 11).
We assume that it was possible to translate the magic Mermin pentagram from
a finite projective geometry description by Michel Planat et al. (2006) into a hyper-
Euclidean representation by means of the double epitahedron (2 E+). The deviated
Pauli operators are in projective geometry represented by three points of PG(1, 2),
their three “Jacobson” counterparts, and the four points whose both coordinates are
zero-divisors.
These points on Mermin’s magic pentagram correspond with the intersecting
points of the entangled simplex figure.
Our 3D representation of the magic pentagram seems to correspond with the three
different interlocked spaces associated with Stokes relations, summing up to zero.
The coordinate system of Stokes vectors with components S1 , S2 and S3 (usually
depicted as three vectors) appear in their spatial form. It appears that from these two
entangled simplices (2E + , or better an indeterminate E± flipping from E+ to E-?)
we derive the picture for a three-partite quantum system. Inside the Poincaré sphere,
it illustrates the three spaces of possible polarizations of electromagnetic waves and
their corpuscular behaviour. In this model, we may assign one direction of the spin in
one epitahedron (E + beige), right (R), and another direction to the other (E + grey),
left (L) (see Fig. 5), where the states |0 and |1 are the eigenstates (Quehenberger,
2017).
Entangled “state” of spin 1/2 particles can be assigned to the Pauli operators, a
set of three 2 × 2 complex matrices (usually denoted as σ_x, σ_y, and σ_z … by the
Greek letter sigma (σ).
Here they are represented by the node points of the edges of the three intersecting
space cells while partitions of the edges surrounding the dodecahedral centre perform
the Clifford algebra of Pauli spin matrices as permuting elements. This construc-
tion, applicable for any number of degrees of freedom, reveals an interesting cyclic
geometric structure that reminds us of a “quantum- merry-go-round” (cf. Uchida
et al., 2015).
It looks as if this 8-dimensional representation of qubits (if we count two inter-
locked 4-dimensional systems) inside the Poincaré sphere could lead to a more
detailed vision of quantum circuits. The advantage of the entangled E± figure, the
two antipodal, spinning golden simplices, for the “conventional” Platonic structure
is that it can be considered as a “real” partition of higher-dimensional space and the
rational principles of algebraic structures may be applied in a dynamic way for a
simulation of movement. Yet, it seems that until there is more discussion, abstract
formalism will continue to prevail:
I’ve always regarded the arrangement into a 5-pointed star of the ten operators in the eight-
dimensional proof of the BKS theorem as nothing more than a useful mnemonic for what
they are. All you have to remember is that the four going along one line of the star are XXX,
XYY, YXY, and YYX.
(N. David Mermin in an email to the author, 05/21/2015)
254 R. C.-Z. Quehenberger

8.2 Quantum Gates

Charles Babbage first realized the idea of gates in his design of the Analytical Engine
with barrels that might act on 50, 100 or more vertical systems of control levers with
cranks, pivots and sliders (Bromley, 1982). These gates now became quantum gates—
for example, the Hadamard gate, which is the one-qubit version of the quantum
Fourier transform. Although Babbage knew Charles Fourier socially, Babbage did
not seem to grasp the significance of his 1811 work on heat propagation, nor did he
seem to know of Joule’s endeavours with heat and mechanical energy. Both could
not have had any idea that their works would merge in a quantum computer driven
by Fourier amplitudes.
In quantum computation, it is sometimes useful to represent a quantum operation
or protocol as a quantum circuit that is reminiscent of the equivalent circuits in
electrical and electronic engineering.
Here, De Morgan’s Laws are applied as powerful rules of Boolean algebra and set
theory that relates the three basic set operations (union, intersection and complement)
to each other. They denote a pair of transformation rules that are both valid rules of
inference, providing an indispensable tool when simplifying digital circuits involving
and/or, and not gates in quantum computing (Mano, 1988).
If A and B are subsets of a universal set U, de Morgan’s laws state that.
(A ∪ B)’ = A’ ∩ B’.
(A ∩ B)’ = A’ ∪ B’.
where ∪ denotes the union (OR), ∩ denotes the intersection (AND) and A’ denotes
the set complement (NOT) of A in U, i.e., A’ = U\A (Bernal, 2005).
In order to create a quantum algorithm, the CNOT and Hadamard gates followed
by measurement in the computational basis constitute a Bell-state measurement. It
is carried out on the first two qubits and leaves the last in one of the states |ψ-, σz|ψ|,
σx|ψ| and σy|ψ. These four possible measurements of Bell states are associated with
the binary numbers 00, 01, 10 and 11 (Barnett, 2009, 152).
The unitary operators on a single qubit, X, Y, Z correspond to the measurement
of the spin along the x, y, and z axes, respectively. The X matrix is often known as
the quantum gate, by analogy to the classical gate, while the Hadamard gate, which
is also key to quantum computing, is the one-qubit version of the quantum Fourier
transform.
The X and Z Pauli matrices are also sometimes referred to as the bit flip and phase
flip matrices: the X matrix takes |0 to |1 and |1 to |0, earning the name bit flip;
and the Z matrix leaves |0 invariant, and takes |1 to –|1, with the extra factor of
−1 added known as a phase factor, so justifying the term phase flip. The Pauli gates
X, Y and Z can be obtained by appropriately setting the phase shifts α = π/2, β2 =
π, or α = π/4 and β4 = –π/2 (Hao et al., 2019).
The computation is performed step by step as follows:
Alice performs a Pauli X and Z gate to her half of the entangled pair, dependent
on the message she would like to send. She then sends her half of the entangled pair
Quantum Cinema and Quantum Computing 255

in a string of N possible symbols through a CNOT gate to Bob, who then measures
the two qubits in the Bell basis to infer the intended 2-bit classical message.
Given Alice’s measurements, Bob performs one of four operations on his half of
the EPR pair and recovers the original quantum state. If Alice performs a measure-
ment and obtains the result |00 then Bob’s system will be in the state |ψ (Nielsen and
Chuang, 2010, 27).
The secret key determines a second transformation, this time on the encrypted
message. This transformation is known as decryption, the inverse of encryption,
allowing Alice to recover the original message (Nielsen and Chuang, 2010, 641).

8.3 Quantum Lattices—Why Not Use a 5-Dimensional Cubic

Grid?

Quantum computing and lattices have had a close relationship ever since the work of
Regev in 2004. He showed that the unique shortest vector problem is equivalent to
bounded distance decoding the hidden subgroup problem on dihedral groups (Regev,
2004). The ordered lattices of all classical crystals are derived from three Platonic
solids: the tetrahedron, the cube and the octahedron. Most lattice theories in quantum
electrodynamics (QED) also rely on cubic lattices up to the 25-dimensional Leech
lattice, which was discussed in the rise of higher-dimensional String Theories.
The grid of the 5-dimensional space formed by the edges of 12 E± (named
epita-dodecahedron) is also actually a cubic lattice forming 24 subspaces and two
counter-movements, totaling up to 26 dimensions if we count all subspaces and
movements (Quehenberger, 2013, 2014a, 2014b, 2014c). The issue of its dimension-
ality was discussed with Gerard’t Hooft, who first counted the 12 polyhedra forming
the epita-dodecahedron and considered the epita-dodecahedron as 12-dimensional.
On realizing that they form another 12 partitions from their intersection in the centre,
he also saw 24 dimensions (in a personal conversation during the Time and Matter
conference, Venice, 2012).
The essential ideas in quantum computation require operations upon quantum stuff
as a user-defined circuit mapped onto space–time. In the space–time computer frame-
work, interactions are taking place in discrete time intervals defined by a discrete set of
space–time points, each labelling a setting-outcome pair. Any given quantum compu-
tation may be expressed as “tomographically complete” sets of quantum circuits.
The structure of setting-outcome pairs at discrete events is the computational lattice
(Hardy & Lewis, 2019, 1).
The problem with this model is, that “space–time points” are not sufficient to
describe a continuum.
Here we suggest using a 6-dimensional space–time framework for one particle
and the same six dimensions for the real world—if we are prepared to restructure the
epistemic philosophical weltbild. The latter includes time in the 4th dimension and
light as electromagnetic turbulence in the 5th dimension (Quehenberger, 2012b).
256 R. C.-Z. Quehenberger

For quantum computing, a procedure to construct new lattices also enables the
labelling of the signals belonging to particular signal constellations (Vieira et al.,
2006). Signal constellations are considered to consist of the barycentres of the regular
hyperbolic polygons with 4 g edges, denoted by F4g (Carvalho, 2001).
Hyperbolic polygons are partitions of some hyperbolic Poincaré half-plane models
imagined as the signal space, where the noise is modelled as a hyperbolic Gaussian
random variable.
Here we depict the hyperbolic space as hemispheres, defining the hypersphere
forming the 4-dimensional space in imaginary movement. These hemispheres may
also represent Boolean sets A and B as shown in Fig. 6.
The transformation matrices associated with the generators of 4g, with m =
1,..., 2 g and.
 iπ

e 4g 0
C= −iπ (1)
0 e 4g

is the matrix corresponding to the elliptic transformation with order 4 g.

Our symplectic representation is composed of golden 5-cells named epitahedra
(E±). Here we identify the edges of the trapezoid basis of E± as polygons with 4 g.

Fig. 6 Cubic Lattice of the five-dimensional space and the 6-dimensional space–time model with
Venn diagram of subsets AB visualizing de Morgans’s law (image: Quantum Cinema/Renate
Quehenberger 2021© Bildrecht, Vienna 2021)
Quantum Cinema and Quantum Computing 257

Formula (1) above poses a challenge to our imagination—how can we bring together
the term e iπ as an infinite logarithmic wavelike phenomenon with the space–time
grid?
Within the context of the design of digital communications systems in homo-
geneous spaces, in particular in hyperbolic spaces, it is necessary to establish a
systematic procedure for the construction of lattices (order of a quaternion algebra)
as the basic entity to the design of signal constellations (quotient of an order by a
proper ideal) to be employed in the transmission of digital information.
The computational lattice x = (lx, xt), composed of “wired” events is a tessel-
lation of octagons and squares. Octagons are convenient because they adjoin to form
“causal-diamond-like structures”. The two-qubit gates situated on the octagons form
a regular lattice, the gate lattice.
The choice of “causal-diamond-like structures” is the simplest construction of a
rhomb by drawing two circles with the same radius, where the centrepoint of one
circle lies on the circumference of the second. The same rhomb is forming the pattern
of the golden mean, known as rhombic Penrose tiling and the lattice for quasicrystals,
which we use for visualization of quantum phenomena.
By proposing the pentagrid as a model for a quantum lattice we also think of the
infinite 5-dimensional space with angles in 72° (Fig. 7). Its inverse number, as well
as its value behind the comma, seem to indicate a relationship to the experimental
value for the Bell parameter function that was experimentally determined to be Sexp
= 2.73± = 0.02 (Weihs et al., 1998)—named “awkward Irish angles” by John Bell
(Bertmann and Friis, 2008).

9 The Hypersphere S3

The hypersphere S3 serves as a model for the Universe as well as for many features in
quantum physics. Despite its importance, usually in literature, for the understanding
of the topology of space, the depiction of the 3-sphere (S3 ) remains to date more
or less a question mark. There is no model for the 4-dimensional space and the
hypersphere (S3 ) is declared as “unimaginable”.
For the hyper-Euclidean model of the hypersphere, which may also serve as a
representation of the earth’s complex system, we take an ordinary 3-dimensional
sphere (in mathematics denoted as S2 : a 2-sphere, because it takes only the surface
into account) and attach hemispheres on each point of its surface (Quehenberger,
Rubin et al., 2022).
So we arrive at a complex 3-dimensional representation of a sphere with a
4-dimensional “real” Euclidean “tangent space” consisting of interconnected, hemi-
spheres in imaginary movement. The model satisfies William Rowan Hamilton’s
mathematical foundations as played out in his Algebra as the Science of Pure Time
(Hamilton, 1837), which appears to have been widely neglected. In his Lectures
on Quaternions (1853), Hamilton noted that complex numbers (quaternions) √ are a
composite of real and imaginary numbers (z = x + iy ∈ C where i = −1) living
258 R. C.-Z. Quehenberger

Fig. 7 Epita-dodecahedron, composed of 12 E± forming a 4-dimensional dodecahedron, repre-

senting the infinite 5-dimensional space, b centre of the epita-dodecahedron, with Platonic and
Archimedean solids; (see also: Epita-dodecahedron (2014), QC-film presented at the ICM Seoul
(KR) 2014 (image: Quantum Cinema/Renate Quehenberger 2012© Bildrecht, Vienna 2021)

in 4-dimensional space—the piece of the curved space surrounding a sphere. In this

way, the 4-dimensional space becomes visually accessible. General relativity avoids
Hamiltonian formulations but recognizes that time is part of the physical system
which quantum gravity is trying to quantize (Wüthrich, 2014). The confusion can
be avoided by assigning time to the 4th dimension of space in imaginary movement
instead of dealing with arbitrary points of space–time (Fig. 8).
In the 3-dimensional representation, we can visualize the embedding space R4 —
which is also the complex vector space C2 (Bengtsson & Zyczkowski, 2006, 82).
The visualization of a Clifford translation of a G + element, particularly a state, is
usually assigned to appear along the big circle of the S3 sphere. The circle is an
intersection of the sphere with the plane of the Hamiltonian lift (Soiguine, 2017).
Quantum Cinema and Quantum Computing 259

Fig. 8 Construction of the hypersphere: Attaching hemispheres on the ball started with icosahedral
symmetry. More and more hemispheres are added until the centre of one is in the radius of its
neighbour; the hypersphere fully packed with hemispheres exhibits hexagonal hyperbolic cells on
the surface of S3 (image: Quantum Cinema/Renate Quehenberger 2012© Bildrecht, Vienna 2021)

The currently still most quoted construction of a hypersphere is called Hopf

fibration, which is a bijection between the points on a sphere resulting in sets of
intertwining great circles on the surface of a sphere (Hopf, 1931).
The idea of fibered spaces goes back to the nineteenth century pedagogue Johann
Friedrich Herbart who perceived the world as continua of fibre bundles. Herbart
inspired the mathematician Bernhard Riemann to conceptualize 1-dimensional mani-
folds. However, these spaghetti-like spaces are counter-intuitive for an artist’s mind.
Nevertheless, we find these 1-dimensional lines as circles below (Fig. 9) and as
wavelike logarithmic fibre spirals (Fig. 10).
Also, the Heegaard splitting is a decomposition of a compact oriented 3 manifold
that results from dividing it into two handle bodies and does not lead to a repre-
sentation of S3 . In topology S1 is a circle, S2 is a hyperbolic circle, two of which
form the sphere S2. However, it is not possible to glue two spheres together in order
to create a hypersphere S3 . Consequently, a full picture of the hypersphere remains
absent (Bengtsson & Zyczkowski, 2006, 425). Regardless of the lack of a proper
imagination, the 3-sphere is an example of the 3-manifolds which play an essential
role in topological quantum computing. Topological methods are used to assist in
the performance of quantum error correction (Nielson and Chuang, 2010, 499).
260 R. C.-Z. Quehenberger

Fig. 9 Construction Epita-dodecahedron, 3D Kites and Darts Penrose tilings with circle decoration
in motion exhibit the spherical shape of a 3-sphere. (image: Quantum Cinema/Renate Quehenberger
2012© Bildrecht, Vienna 2021)

Topology speaks of handlebody-knots embedded in the 3-sphere S3 . Please

note, that we get a torus if we cut the 4-dimensional space around the compact
3-dimensional space of the sphere in the centre.
In the context of 3-manifolds, some authors use Hausdorff topological spaces
with a group-theoretical substitute for 3-manifolds, the fundamental group of a knot
complement also known as knot group (Asselmeyer-Maluga, 2021). In our construc-
tion of the hypersphere, the knot is the fundamental space-generating element: It
starts not with a knot in 1-dimensional string but with the knot of a 2-dimensional
plane forming a pentagon. The triangular partitions of this pentagon undergo another
knot-like enfolding that results in a double 3D pentagon, which we have named the
epitahedron (E±). Twelve of these two 5-cells, form the epita-dodecahedron (Fig. 7),
whose circles in movement form spherical shapes homeomorphic to the 3-sphere, as
shown below in Fig. 9.
Quantum Cinema and Quantum Computing 261

Fig. 10 The 4π evolution of e (vimeo.com/50142342) evolution of the imaginary number e (image:

Quantum Cinema/Renate Quehenberger 2012© Bildrecht, Vienna 2021)

The problem is that 3-manifolds require a 5-dimensional space like R5 as an

embedding space for which they do not have any visual representation since the
visualizations of the Quantum Cinema project are not yet widely discussed.
Today the main source for 3-manifolds remains mathematician William Thurston,
who explains that the 3-sphere S3 could be easily visualized as R3 , together with one
point at infinity (Thurston[, 1980], 2002), 1). However, Thurston’s description of a
horosphere, as mentioned by Carl Friedrich Gauss and Nikolai I. Lobachevsky in the
early 1800s, seems more suitable for the hyper-Euclidean visualization method. It is
a Euclidean sphere tangent to the unit sphere where the point X of tangency is the
centre of the horosphere (Thurston, [1980] 2002, 38). We interpret the horosphere
as attached to the sphere S2 .
When working on the foundations of topology, Henri Poincare investigated topo-
logical properties that characterized a sphere. He was able to show that the Poincaré
sphere had a fundamental group of order 120, while the 3-sphere had a trivial
fundamental group.
Our intuitively applied construction method for the hypersphere S3 —the full
covering of the sphere S2 with hemispheres where each centre of one lies in the
radius of its neighbour—requires starting with icosahedral symmetry, as we see above
(Fig. 8). We assume that Poincaré must have had a similar picture in mind when he
suggested the icosahedral symmetries for the dodecahedral space. Surprisingly, we
find hyperbolic hexagonal shapes resulting from the intersecting hemispheres on the
2-sphere as the outer surface of S3 .
We animated the epita-dodecahedron according to the description of the Poincaré
dodecahedron space by William Threlfall and Herbert Seifert in their first joint paper
(1931). It is characterized by the use of a dodecahedron, the opposite faces of which
262 R. C.-Z. Quehenberger

are identified after a turn by π/5 in order to construct a “spherical dodecahedron

space”. Analogous to their description, we use the 3-dimensional pentagons E±, a
golden pentachoron (aka 5-cell or 4-simplex), to visualize the principle of the counter-
movements of the opposite polyhedra E + instead of the pentagons and a spherical
dodecahedron only appears inside from the red and green circle decorations. The
same geometry of the dodecahedral space and its dual hypersphere was suggested to
be responsible for the Earth’ intrinsic dynamics (dC Rubin et al., 2021).
Hence, we get a 3-manifold that is composed according to Henri Poincaré’s call
for a hypergéométrie in the visualization of the space of 5 dimensions.
— Please note the relationship to Poincaré’s famous question, “Is it possible that a
manifold with a vanishing fundamental group is not homeomorphic to the 3-sphere?”
(Poincaré, 1904, 46).
Poincaré assumed all these manifolds are simply connected, i.e., homeomorphic
to hyper-spheres (Poincaré, 1895, 277). William Thurston established the following
weak version of the Poincaré conjecture: If a simply connected 3-manifold M is
a cyclic branched covering space of S3 , then M is in fact homeomorphic to S3 .
We may well recognize this “cyclic branched covering space” as full circles of the
epita-dodecahedron (Fig. 9).
With its circles in motion, this infinite 5-dimensional space in the shape of a dodec-
ahedron reveals the same shape as the hypersphere. This picture of a 4-dimensional
dodecahedron, the epita-dodecahedron, composed of the 3D representation of the
Penrose Kites and Darts and animated according to the description of the Poincaré
homology sphere gives rise to the claim that is homeomorphic to the hypersphere.
However, we claim that Ricci flow surgery and Ricci calculus is relying on a system
of point transformations and not considering Poincaré’s special group transforma-
tions. Moreover, any proof of the Poincaré conjecture cannot be void of any statement
about the shape of the hypersphere (Quehenberger, 2021).

10 Quantum Algorithms and Fourier Transform

Fourier transform, or better quantum Fourier transform, became the dominant tool
in quantum computing. Fourier transform may be used to solve the hidden subgroup
problem, a generalization of the phase estimation and order-finding problems that
has among its special cases an efficient quantum algorithm for the discrete logarithm
problem, another problem thought to be intractable on a classical computer (Nielson
and Chuang, 2011, 217).
In quantum computing the Fourier transform is the key to a general procedure
known as phase estimation—also described as black box operation—which is, in
turn, the key for many quantum algorithms.
In mathematics, the discrete Fourier transform over an arbitrary ring generalizes
the discrete Fourier transform of a function whose values are complex numbers, i.e.,
performing the transform in the integers mod N for some integer N (Schönhage &
Strassen, 1971). The algorithm uses recursive Fast Fourier transforms in rings with
Quantum Cinema and Quantum Computing 263

2n+1 elements, a specific type of number theoretic transform (Knuth, 1997, 305).
Quantum algorithms can be summarized as follows: First, reduce the question at hand
to something resembling a hidden subgroup problem, then extend the existing hidden
subgroup machinery to give a quantum algorithm for this particular variant. The key
step in solving a hidden subgroup problem is the ability to compute the quantum
Fourier transform over a group with exponentially many elements. So the relationship
between the Fourier transforms over different cyclic groups of an arbitrary fixed
superposition (Hales & Hallgren, 2000).
Suppose a unitary operator U has an eigenvector |u > with eigenvalue e2πiφ , where
the value of φ is unknown. The product representation allows the construction of an
efficient quantum circuit computing the Fourier transform and provides insight into
algorithms based upon the quantum Fourier transform (Nielson and Chuang, 2011,
217).
A classical computer requires exponentially more operations to compute the
Fourier transform than a quantum computer. The quantum circuit construction of
the quantum Fourier transform apparently requires logic gates of exponential preci-
sion in the number of qubits used. Fourier analysis involves expanding periodic
functions on the unit circle written in terms of a series of sines and cosines functions
which are referred to as harmonics. Considering higher-dimensional analogues of
the harmonics on the unit n-sphere, one arrives at the spherical harmonics—which
we imagine as logarithmic spirals around the unit sphere.
In our visualization of the 4π evolution of e (Fig. 10), the natural algorithm e is
depicted by means of logarithmic spirals emerging from the unit sphere and curling
around it. This construction is similar to the idea of fibered spaces; we coloured the
spiral on the left red, and spirals on the right blue. Uneven numbers of logarithmic
spirals mix, while even numbered spirals appear in bundles of the same colour. This
is certainly another topic that requires deeper investigation.
At least it provides a suggestion for the wave state of a quantum particle. It reminds
us of the measurement of the 4π value of a neutron by Helmut Rauch, which we
discussed with him in depth during the Quantum Cinema project.
This wave bundle figure is reminiscent of a charged-particle motion (factors of
4π/c) and Maxwell’s equations describing a curl in an electric field that implies a
change in the magnetic field equal to the current density, occurring at the speed of
light. The given distribution of electric current density J is derived from the magnetic
field B simply as (c/4πi)∇ × B (Vasyliunas, 2005). This reminds us of Schrödinger’s
dream of a visual interpretation of the matrix elements “as amplitudes of partial
oscillations of the electrical moment of the atom”, which he had to abandon (Meyenn,
2010, 204).
According to Heisenberg’s theory, the Schrödinger equation of a light quantum
has a value of the order of 1 (Heisenberg in a letter to Pauli, May 30, 1935, in Meyenn,
1985, 397).
However, we may imagine the quantum object, the ideal quantum Fourier trans-
form on n qubits, results if the controlled-Rk gates are performed to a precision
r = 1/p(n) for some polynomial p(n).
264 R. C.-Z. Quehenberger

The prime state made of n qubits that corresponds to the quantum superposition
of all prime numbers is less than 2n (we take n > 1 so that 2n is not a prime), where
each prime number can be expressed in binary form p = p0 20 + p1 21 + ... + pn−1 2
n−1
.
All the states in the sum are orthogonal and the normalization of the state is related
to the squared root of the number of primes less than 2n , namely π(2n ). In the case
√ n = 3,
of, for example,
|P3 = 1/ 4 (|2 + |3 + |5 + |7
= 1/2 (| ↑↓↑+ | ↑↓↓ + | ↓↑↓ + | ↓↓↓) where the qubits |0 and |1 are described
by the spin polarized states ↑aand ↓nof a spin 1/2 particle (Latorre & Sierra, 2013).
Applying the controlled-R2 gate produces the state of the qubit system.
At most n/2 swaps are required, and each swap can be accomplished using three
controlled-CNOT gates. Accordingly, this circuit provides a (n2 ) algorithm for
performing the quantum Fourier transform.
The ability to tune the control frequency of a qubit within the silicon chip by engi-
neering its atomic configuration is extremely important. The molecules can be created
with different resonance frequencies. This means that controlling the spin of one
qubit will not affect the spin of the neighbouring qubit, leading to fewer errors—an
essential requirement for the development of a full-scale quantum computer (Voisin,
2020).

10.1 Quantum Algorithms and Fuchsian Groups

Poincaré’s research on the fundamental polyhedron has special importance for

quantum computation since quantum gates are elements of the fundamental group
represented as SU(2) matrices. In quantum computation, the arithmetic Fuchsian
groups play an important role as a basis for code construction and signal constellation
design (Cavalcante, Palazzo et al., 2005).
The arithmetic Fuchsian group is a subgroup of PSL(2\R) and derived from a
quaternion algebra. The term “Fuchsian groups” for a class of groups that are impor-
tant in the theory of modular functions was coined by Henri Poincaré because he
was inspired by the work on inverse functions by Lazarus I. Fuchs (Fuchs, 1880).
Poincaré used them as a fundamental concept in the development of the
theory of automorphic functions and created a completely new area of mathe-
matics (Poincaré, 1882). Discrete and continuous transformations also play an impor-
tant part in the analysis of differential equations carried out by Felix Klein. Which
led to the bestowal of the name “Fuchsian” causing difficulties between Poincaré and
Felix Klein. When Klein complained about the name because he was also working
on the same algebraic structures, Poincaré countered in German by quoting Goethe’s
Faust, “Name ist Schall und Rauch” (a name is sound and smoke) in a letter to Klein,
4 April 1882 (Verhulst, 2012, 43).
Quantum Cinema and Quantum Computing 265

Poincaré was building his research on a fundamental polyhedron in Group theory

on the so-called Fuchsian groups, where each group is associated with a funda-
mental polygon with edges identified by certain motions and closed paths in a 3-
manifold defined by a polyhedral region with faces identified by certain geometric
transformations (Poincaré, 1892).
In this context, he was pondering certain shapes suitable as a fundamental region,
like a cube, a pyramid or a torus belonging to a family of 3-dimensional manifolds
and obtained as quotients of R3 by certain groups.
He assumed all these manifolds were simply connected, i.e., homeomorphic to
hyper-spheres (Poincaré, 1904, 277). The question as to whether his conjecture was
accurate or not became one of the famous millennium problems (Milnor, 2000, 2003).
Because of their 4-dimensional character, these polygons related to Fuchsian
groups are usually thought of as being part of the hyperbolic plane. Fuchsian groups
are associated with a regular tessellation and given edge-pairing transformations
tessellation whose regular hyperbolic polygon generates an oriented compact surface
with genus g > 2. This algorithm can be implemented in Maple or Mathematica
(Benedito et al., 2016).
Distinct from the usual projective geometry interpretation that the quaternion
space with Fuchsian groups can be treated as a tessellation of the sphere, we assume
the quaternionic space is spanned over orthogonally attached vectors as real tangent
space surrounding the sphere, forming the 4-dimensional space as shown in Fig. 6
above in the area of the yellow circles. The two intersecting circles may also represent
the space of real states, with disjoint regions (cf. Jevtic & Rudolph, 2015).
Accordingly, our construction of the hyperspace is composed of hyperplanes for
the upper-half plane H = {z ∈ C: Im(z) > 0} attached on each point of the sphere (S2 )
and constructs a hyperbolic space which is a real hemisphere on top of the hyperplane
instead of the usual projective models of the hyperbolic plane, known as Poincaré
disk or Beltrami–Klein disk model.
This gives us the images for the nodes as above, and the hyperbolic hexagon is only
the basis of the closed mirroring self-dual epitahedra (E±) that is the fundamental
polyhedron of the icosahedral/dodecahedral space we can compose of it. This image,
which naturally contains the lattice Z5 from the edges of the polyhedra (E±), enables
lattices to be imagined that are associated with signal constellations in our presumed
real hyperspace.

10.2 Factorization of Prime Numbers

A quantum computer can factor a number exponentially faster than the best known
classical algorithms. The most spectacular discovery in quantum computing to date
is that quantum computers can efficiently perform some tasks which are not feasible
on a classical computer.
For example, finding the prime factorization of an n-bit integer is thought to
require exp((n1/3 log2/3 n)) operations using the best classical algorithm known at
266 R. C.-Z. Quehenberger

the time of writing, the so-called number field sieve (Nielsen and Chuang, 2010,
216).
Prime numbers are central objects in Mathematics and Computer Science. They
appeared dramatically in Quantum Computation through Shor’s algorithm, which
converts the hard problem of factorization into a polynomial one using quantum
interference (Shor, 1997). The entanglement properties of the Prime state remain to
be explored in more detail.
The Prime state is highly entangled, and its entanglement measures encode number
theoretical functions, such as the distribution of twin primes.

10.3 Bernoulli Numbers and the Riemann Hypothesis

Bernoulli numbers are characterized by the three most useful features, namely, a
recursive equation, an explicit formula and a generating function. They were first
applied to computer science in the first conception for the first published complex
computer program by Ada Lovelace in note G on the Analytical Engine, where she
describes an algorithm for generating Bernoulli numbers with Babbage’s machine
(Lovelace, 1842). Bernoulli factory algorithms sample the probability f (λ). There is
an analogy to the Bernoulli factory problem called the quantum Bernoulli factory,
with the same goal of simulating functions of unknown probabilities. For every odd
n > 1, Bn = 0. For every even n > 0, Bn is negative if n is divisible by 4 and positive
otherwise.
Bernoulli numbers can also be expressed in terms of the famous Riemann zeta
(ξe) function, an extension of Euler’s zeta (ξ) function.
The Swiss mathematician Leonard Euler made pioneering and influential discov-
eries in many other branches of mathematics, such as analytic number theory,
complex analysis, and infinitesimal calculus. He was invited to Russia by Empress
Catherine II thanks to his good relations with the Bernoulli brothers and succeeded
Daniel Bernoulli as professor of mathematics at the Academy of St. Petersburg in
1733. Euler discovered a formula relating ξ(2 k) to the Bernoulli numbers, yielding
results such as ξ(2) = π2 /6 and ξ(4) = π4 /90. The values of the zeta function are
expressed as powers of π and depend on these Bernoulli numbers.
Bernhard Riemann noted that his zeta function had trivial zeros at −2, −4, −6, …
and that all non-trivial zeros were symmetric about the line Re(s) = 1/2. He calculated
the first six non-trivial zeros of the function and observed that they were all on the
same straight line. In a report published in 1859, Riemann stated that this might very
well be a general fact (Riemann, 1859).
In 1900, David Hilbert listed proving or disproving the Riemann hypothesis as
one of the most important unsolved problems confronting modern mathematics.
Since then numerous attempts to solve this conjecture have been published. It
is central to understanding the overall distribution of the primes. The Hilbert-Pólya
Conjecture proposes a spectral interpretation of the non-trivial zeros of the Riemann
zeta function. Hilbert called the set of eigenvalues of an operator the “spectrum”, by
Quantum Cinema and Quantum Computing 267

analogy with the spectral lines produced by the radiation frequencies of atoms. Prime
numbers are also related to the harmonic mean as formulated by the Pythagoreans.
For example, in 1901, von Koch showed that the Riemann hypothesis was equiv-
alent to the logarithmic expression similar to the one Euler found (von Koch,
1901).
or(x) = Li (x) + 0 (x1/2 ln x)
Jacques Hadamard and de la Vallee Poussin proved the prime number theorem,
by showing that. √
ad(x) = Li (x) + 0 (xe -a ln x )
for some positive constant a, and they did this by bounding the real part of the zeros
in the critical strip away from 0 and 1. The error term is directly dependent on what
was known about the zero-free region within the critical strip.
The function f satisfies the equation f (1 − s) = f(s) for the complex argument
s ≡ σ + iτ. Or, as the British mathematician Godfrey Harold Hardy put it, ξ (1/2 +
it) and proved that the Riemann zeta function has infinitely many real zeros. Hardy
also showed that infinite many non-trivial zeros lie on the critical line Re(s) = 1/2
(Hardy, 1914).
The Austro-Hungarian mathematician Marcel Riesz showed in 1916 that
the Riemann hypothesis was equivalent to the claim that it is true for any e larger
than 1/4 if we had, x tending to infinity on the positive real axis (Riesz, 1916). Also,
all points where f vanishes are located on the critical line σ = 1/2 and all lines of
constant phase of f corresponding to ±π, ±2π, ±3π… merge with the critical line
(Schleich et al., 2018). This π-periodicity is visualized in Fig. 11, where after every
π -turn the prime numbers appear on one line with the most beautiful pattern of
“entangled” prime numbers on the y-axis visualized by small blue circles.
It seems to be a visual confirmation of Godfrey H. Hardy and John E. Littlewood’s
Twin Prime Conjectures, (Hardy & Littlewood, 1921). An analytical explanation of
these periodical structures obtained is an open problem worthy of investigation by
the experts in the analytical number theory (Machadoa & Luchkob, 2021).
In our visualization, the numbers 1, 2, 3, 4 are lying inside the radius of the sphere,
while number 5 is already the first number on the spiral arm crossing the x-axis above
the north pole of the sphere, representing the imaginary part of the system just like
the red and blue bundles of lines in Fig. 10. By placing all numbers consecutively on
the ±x and ±y-axis, we arrive at all even numbers on the x-axis and all odd numbers
on the y-axis. Furthermore, we place the imaginary numbers i on a logarithmic spiral
originating on the surface of the unit sphere at point 1/2 of a Cartesian coordinate
system, which is the half diameter of the unit sphere that represents the real numbers.
This implies that in our visualization, the so-called “critical line” with non-trivial
zeros is a geodesic on the real number space unit sphere and the imaginary numbers
on the logarithmic spiral form the 4-dimensional space around a central unit sphere
which can be identified as Hamiltonian/quaternionic/Lorentzian space–time.
For quantum computation, Jose I. Latorre and German Sierra showed how to
construct the quantum prime state, a quantum sequence state based on the sequence
of prime numbers with an error below the bound that allows for verification of
268 R. C.-Z. Quehenberger

Fig. 11 The prime numbers in a logarithmic 4πi solution, all prime numbers on the y-axis, produce
a series of twin prime numbers which are depicted by small blue circles, seemingly “entangled”
(image: R. Quehenberger 2012© Bildrecht, Vienna 2021)

the Riemann hypothesis. This algorithm can be further combined with the quantum
Fourier transform to yield an estimate of the prime counting function more efficiently
than any classical algorithm (Latorre & Sierra, 2013). They assume it is likely that
the quantum correlations emerging from the Prime state are profoundly related to
theorems in Number Theory. Their claim is simple: All non-trivial zeros have a real
part of 1/2.
With the help of computers, more than ten billion zeros of the zeta function lying
on the critical, clearly highlighting Riemann’s extraordinary intuition, have been
calculated to date (Bayer, 2018). However, calculations ad infinitum do not provide
sufficient proof (van de Lune et al., 1986).
Finally, we call for the reestablishment of the authority of geometry for proving
algebraic operations so that we could claim the million dollars—at least for the
Poincaré conjecture since, despite many claims to have done so, Henri Poincaré’s
question has never really been answered in the way we have shown here.
Quantum Cinema and Quantum Computing 269

Although purely imaginative, geometry may provide a visual correspondence to

algebraic descriptions of quantum systems. Moreover, we theorize that the much
discussed “hidden parameters” are to be found in the above described dynamic 5-
dimensional space.
John Bell did not agree that: “In science, whatever is provable must not be believed
without proof.” (Dedekind, 1888), but in all cases, the process of confirmation—
and not the proof—remains the ultimate source from which knowledge derives its
authority; it is the “experience of truth” (Bell, 2000).
In this context, we may recall Plato’s theory of forms, which states: “…in any
case there are forms although not perceivable but only accessible through the mind”
(Plato, 97). In this tradition of Platonic thought, John Bell quoted Hermann Weyl, who
shared his lifelong conviction that intuition, or insight—rather than proof—furnishes
the ultimate foundation of mathematical knowledge (Bell, 2000).
Transparent penetrable space, the purest image of my knowing, cannot be inspected but must
be seen intuitively, and within it my inspecting itself is so seen. The light is not outside of
me, but rather in me. (Hermann Weyl, as quoted by Bell, 2000)

References

Abbott, E. A. (1884). Flatland: A romance of many dimensions. Seeley & Co.

Aspect, A., Grangier, P., & Roger, G. (1981). Experimental tests of realistic local theories via Bell’s
theorem. Physical Review Letters, 47, 460–463.
Babbage, C. (1838). The ninth Bridgewater treatise. A fragment. J. Murray.
Baez, J. C. (2017). From the icosahedron to E8. arXiv:1712.06436v2
Barnett, S. (2009) Quantum information (Oxford Master Series in Physics: Atomic, Optical, and
Laser Physics). Oxford University Press.
Bayer, P. (2018). The Riemann hypothesis: The great pending challenge. Mèthode Science Studies
Journal 8, 34-–41. https://doi.org/10.7203/metode.8.8903
Bell, J. S. (1964). On the Einstein Podolsky Rosen paradox. Physics Physique Fizika, 1, 195–200.
Bell, J. S. (1966). On the problem of hidden variables in quantum theory. Reviews of Modern
Physics, 38, 447–452.
Bell, J. S. (2000). Hemann Weyl on intuition and the continuum. Philosophia Mathematica, 8(3),
259–273.
Bellman, R. E. (1957). Rand corporation dynamic programming. Princeton University Press.
Benedito, C. W. D. O., Palazzo, R., & Interlando, J. C. (2016). An algorithm to construct arithmetic
Fuchsian groups derived from quaternion algebras and the corresponding hyperbolic lattices.
Journal of Pure and Applied Algebra, 220(5), 1902–1923.
Bengtsson, I., & Zyczkowski, K. (2006). Geometry of quantum states: An introduction to quantum
entanglement. Cambridge University Press.
Bennett, C. H. (1995). Quantum information and computation. Physics Today, 48(10), 24.
Bennett, C. H., & Wiesner, S. J. (1992). Communication via one- and two-particle operators on
Einstein-Podolsky-Rosen states. Physical Review Letters, 69, 2881.
Bharti, K., Ray, M., Varvitsiotis, A., Kwek, L.-C., & Cabello, A. (2019). Local certification of
programmable quantum devices of arbitrary high dimensionality. arXiv:1911.09448v1
Bohm, D. (1952). A suggested interpretation of the quantum theory in terms of “Hidden” variables.
I Physical Review, 85, 166 and Part. II Physical Review, 85, 180.
270 R. C.-Z. Quehenberger

Bohm, D. (1982b). The holographic paradigm and other paradoxes, exploring the leading edge of
science. In: K. Wilber (Ed.). Shambala.
Bohm, D. (2004). On creativity. In: L. Nichol (Ed.). Routledge.
Bohm, D., & Hiley, B. (1993). The undivided universe. Routledge.
Bernal, R. A. (2005). De Morgan’s laws revisited: To be AND/OR NOT to be, paper PO25,
PharmaSUG.
Bertlmann, R. A., & Friis, N. (2008). Theoretical physics T2 quantum mechanics course of lectures,
T2–script of Sommersemester 2008 by Reinhold A. Bertlmann.
Bertlmann, R., & Zeilinger, A. (Eds.). (2017). Quantum [Un]Speakables II. Springer International
Publishing Switzerland.
Brezina, C. (2006). Al-Khwarizmi: The inventor of algebra. The Rosen Publishing Group.
Brillouin, L. (1956). Science and information theory. Academic Press.
Brody, D. C., & Gräfe, E.-M. (2011). Six-dimensional space-time from quaternionic quantum
mechanics. Physical Review D, 84, 125016. arXiv:1105.3604
de Broglie, L. (1925). Recherches sur la Théorie des Quanta, Thesis 1924. Annales de Physique,
10e série, t. III.
de Broglie, L. (1927). L’univers à cinq dimensions et la mécanique ondulatoire. Journal De Physique
Et Le Radium, 8, 66–73.
de Broglie, L. (1943). Die Elementarteilchen, Individualität und Wechselwirkung, H Coverts; orig.
(1941). Continu et discontinu en physique moderne (Continuous and discontinuous in Modern
Physics), Paris: Albin Michel.
Bromley, A. G. (1982). Charles Babbage’s analytical engine, 1838. Annals, 4(3).
Brouwer, L. E. J. (1913). Über den natürlichen Dimensionsbegriff. Reine Und Angewandte
Mathematik, J, 42, 146–215.
Cabello, A. (2017). Interpretations of quantum theory: A map of madness. In: O. Lombardi, S.
Fortin, F. Holik, C. López (Eds.), What is quantum information? Cambridge University Press.
Cavalcante, R. G., Lazari, H., Lima, J. D., & Palazzo, R. Jr. (2005). A new approach to the design of
digital communication systems. In: A. Ashikhimin & A. Barg (Eds.), Discrete mathematics and
theoretical computer science (Vol. 68, pp. 145–177). DIMACS Series. American Mathematical
Society.
Carvalho, E. D. (2001). Construction and labeling of geometrically uniform signal constellations
in Euclidean and hyperbolic spaces. Doctoral Dissertation, FEEC-UNICAMP.
Capra, F. (1982). The turning point. Simon and Schuster.
Clarke, B. (2009). Heinz von Foerster’s Demons. In: B. Clarke, M. B. N. Hansen (Eds.), Emergence
and embodiment: New essays on second-order systems theory. Duke University Press.
Dakić, B., & Brukner, Č. (2016). The classical limit of a physical theory and the dimensionality of
space. Fundamental Theory Physics, 181, 249–282.
Deleuze, G. (1989). Part 2: Cinema 2, University of Minnesota Press. Here quoted in German from:
Transl. Klaus Englert, (1991) Das Zeit-Bild, Kino 2, Frankfurt/Main: Suhrkamp.
Descartes, R. ([1637] 1902). Discours de la Méthode. Adam et Tannery.
Dirac, P. A. M. (1930). The principles of quantum mechanics. Clarendon Press.
de Bruijn, N. G. (1981). Algebraic Theory of Penrose’s Non-periodic Tilings of the Plane I and II,
Proc Koninklijke Nederlandse Akademie van Wetenschapen, 84, 39–52 and 53–66.
Einstein, A. (1916). Die Grundlage der allgemeine Relativitätstheorie. Annalen Der Physik, 49,
769–822.
Einstein, A., Podolsky, B., & Rosen, N. (1935). Can quantum-mechanical description of physical
reality be considered complete? Physical Review, 47, 777–780.
Faulhaber, J. (1631). Academia Algebrae, darinnen di miraculo’sische, Inventiones zu den höchsten
Costen weiters continuirt u. profitiert werden, Augsburg: Johann Remelins.
Feynman, R. (1982). Simulating physics with computers. International Journal of Theoretical
Physics, 21, 6/7, 467–488.
Fourier, J. B. J. (1822). Theorie analytique de la chaleur. Firmin Didot.
Quantum Cinema and Quantum Computing 271

van Fraassen, B. (1972). A formal approach to the philosophy of science in paradigms and paradoxes:
The philosophical challenge of the quantum domain. In R. G. Colodny (Ed.), Paradigms and
paradoxes: The philosophical challenge of the quantum domain (pp. 303–366). University of
Pittsburgh Press.
Fuchs, L. (1880). Über die Verallgemeinerung des Kehrungsproblems. Journal De Crelle, 89, 151–
169.
Gardner, M. (1989). Penrose tiles to trapdoor ciphers. H. Freeman & Co.
Greenberger, D. M., Horne, M. A., & Zeilinger, A. (1993). Multiparticle interferometry and the
superposition. Physics Today, 46(8), 22.
Grünbaum, B., & Shephard, G. C. (1987). Tilings and patterns. W. H. Freeman and Company.
Grössing, G. (2003). Zum Bilderstreit in der Quantentheorie, http://phaidon.philo.at/~nulleins/arc
hiv/archiv.html
Hales, L., & Hallgren, S. (2000) An improved quantum Fourier transform algorithm and appli-
cations. In: Proceedings of the 41st Annual Symposium on Foundations of Computer Science
(pp. 45–46).
Hao, L., Wang, Z., Yang, J., et al. (2019). Anomalous magnetoresistance due to longitudinal spin
fluctuations in a Jeff = 1/2 Mott semiconductor. Nature Communications, 10, 5301. https://doi.
org/10.1038/s41467-019-13271-6
Hardy, G. H. (1914). Sur les zeros de la fonction zeta ζ(s). Comptes Rendus De L’académie Des
Sciences, 158, 1012–1014.
Hardy, G. H., & Littlewood, J. E. (1921). The zeros of Riemann’s Zeta-function on the critical line.
Mathematische Zeitschrift, 10(3–4), 283–317. https://doi.org/10.1007/bf01211614
Hardy, L., & Lewis, A. G. M. (2019) Quantum computation with machine-learning-controlled
quantum stuff. arXiv:1911.13282v1
Hermann, G. (1926). Die Frage der endlich vielen Schritte in der Theorie der Polynomideale.
Mathematische Annalen, 95, 736–788.
Hermann, G. (1935). Die naturphilosophischen Grundlagen der Quantenmechanik. Abhandlungen
Der Fries’schen Schule, 6(2), 69–152.
Heisenberg, W. (1927) Über den anschaulichen Inhalt der quantentheoretischen Kinematik und
Mechanik (On the Perceptual Content of Quantum Theoretical Kinematics and Mechanics),
Copenhagen, Institut for theoretical Physics at the Kopenhagen University, March 23 and
Zeitschrift für Physik 43, 172–198.
Heisenberg, W. (1989). Encounters with Einstein: And other essays on people, places, and particles.
Princeton University Press.
Heisenberg, W. (2011). Quantenphysik und Philosophie, Stuttgart: Philip Reclam jun. (reprint from
1979).
Herzenberg, C. L. (2008) Grete Hermann: An early contributor to quantum theory.
arxiv.org/pdf/0812.3986.
Hiley, B. J., & Callaghan, R. E. (2010). The Clifford algebra approach to quantum mechanics B:
The Dirac particle and its relation to the Bohm approach. arXiv:1011.4033v1
Hiley, B. J.(2016). Structure process, weak values and local momentum. Journal of Physics:
Conference Series, 701(1). Article id. 012010
Hopf, H. (1931). Über¯ die Abbildungen der dreidimensionalen Sphäre auf die Kugelfäche,
Mathematische, Annalen, 104(1), 637–665.
Holmes, R. (2015). Computer Science: Enchantress of abstraction. Nature, 525, 30–32.
Toole, B. A. (2010) Ada, the enchantress of numbers: Poetical science. Critical Connection.
Jammer, M. (1974). The philosophy of quantum mechanics: The interpretations of quantum
mechanics in historical perspective. Wiley.
Jevtic, S., & Rudolph, T. (2015). How Einstein and/or Schrödinger should have discovered Bell’s
theorem in 1936. Journal of the Optical Society of America B - Optical Physics, 32, A50-A55.
ISSN 0740-3224
Jarry, A. ([1911] 1996). Exploits and Opinions of Dr. Faustroll, Pataphysician, Boston: Exact
Change.
272 R. C.-Z. Quehenberger

Jones, W. (1794). On the philosophy of the Asiatics. Asiatic Researches, IV, 164.
Jones, W. (1794). On the philosophy of the Asiatics. Asiatic Researches, IV, 164.
Klein, F. (1872). Vergleichende Betrachtungen über neuere geometrische Forschungen, Programm
zum Eintritt in die philosophische Fakultät und den Senat der Friedrich-Alexander-Universität
zu Erlangen, Erlangen: Deichert.
Klein, F. (1892). Vorlesungen über Nicht-Euklidische Geometrie. In Klein, Felix (1928)
Vorlesungen über Nicht-Euklidische Geometrie, Rosemann, W. (Ed.) Berlin: Springer.
Klyachko, A., Can, M. A., Binicioğlu, S., & Shumovsky, A. S. (2008). Simple test for hidden
variables in spin-1 systems. Physical Review Letters, 101, 020403.
Knuth, D. E. (1997). The art of computer programming, volume 2: Seminumerical algorithms, 3rd
edn. Addison-Wesley Professional.
von Koch, H. (1901). Sur la distribution des nombres premiers. Acta Mathematica, 24, 159–182.
Kochen, S., & Specker, E. P. (1967). The problem of hidden variables in quantum mechanics.
Journal of Mathematics and Mechanics, 17(1967), 59–87.
Kunjwal, R., Spekkens, R. W. (2015). From the Kochen-Specker theorem to noncontextuality
iinequalities without assuming determinism, submitted to QPL. www.cs.ox.ac.uk/qpl2015/pre
proceedings/50.pdf
Kwiat, P. G., Mattle, K., Weinfurter, H., Zeilinger, A., Sergienko, A. V., & Shih, Y. H. (1995).
New high-intensity source of polarization-entangled photon pairs. Physical Review Letters, 75,
4337–4341.
Latorre, J. I., & Sierra, G. (2013) Quantum computation of prime number functions. Quantum
Information & Computation, 14(7). arXiv:1302.6245v3
Lenzen, V. F. (1969). Concepts and reality in quantum mechanics. Philosophy of Science, 16(4),
279–286.
Leonardo da Vinci. (1877). Treatise on painting (J. F. Rigaud, Trans.) George Bell & Sons.
Lovelace, A. (1842). Notes to “A Sketch of the Analytical Engine invented by Charles Babbage by LF
Menabrea”, translated with a detailed commentary Ada Augusta, Countess of Lovelace, Biblio-
thèque Universelle de Genève, No. 82, October 1842. In: R. Taylor, (Ed.), Scientific memoirs,
selections from transactions of foreign academies and learned societies and from foreign journals
(Vol. 3, Article XXIX). FSA (1843).
van de Lune, J., te Riele, H. J. J., & Winter, D. T. (1986). On the zeros of the Riemann zeta function
in the critical strip, iv. Mathematics of Computation, 46(174), 667–681.
Lück, R. (2000). Dürer-Kepler-Penrose, the development of pentagon tilings. Materials Science
and Engineering A, 294–296.
Machadoa, T., & Luchkob, Y. (2021). Multidimensional scaling and visualization of patterns in
distribution of nontrivial zeros of the zeta-function. Communications in Nonlinear Science and
Numerical Simulation, 102, 105924.
Mackay, A. L. (1990). Crystals and fivefold symmetry. In: I. Hargittai (Ed.), Quasicrystals, networks,
and molecules of fivefold symmetry. VCH.
Mackay, A. L. (1981). De nive Quinquangula — On the Pentagonal Snowflake, Kristallogafiya 26,
1981, 910–919. Soviet Physics Crystallography, 26(1981), 517–522.
Mandolesi, A. L. G. (2019). Quantum fractionalism: The Born rule as a consequence of the complex
Pythagorean theorem. 1905.08429. Physics Letters A, 384(28), 126725.
Mattle, K., Weinfurter, H., Kwiat, P. G., & Zeilinger, A. (1996). Dense Coding in Experimental
Quantum Communication, Physical Review Letters, 76, 4656.
Mano, M. M. (1988). Computer engineering: Hardware design. Prentice Hall.
Mermin, N. D. (1990). Simple unified form for the major no-hidden-variables theorems. Physics
Review Letters, 65(27), 3373–3376.
Mermin, N. D. (1993). Hidden variables and the two theorems of John Bell. Reviews of Modern
Physics, 65, 803–815.
von Meyenn, K. (1985) Wolfgang Pauli: Wissenschaftlicher Briefwechsel mit Bohr, Einstein,
Heisenberg u.a. Band II: 1930–1939. Springer.
Quantum Cinema and Quantum Computing 273

von Meyenn, K. (2010). Eine Entdeckung von ganz außerordentlicher Tragweite: Schrödingers
Briefwechsel zur Wellenmechanik und zum Katzenparadoxon. Springer.
Miller, A. I. (1996). Insights of genius, imagery and creativity in science and art. Springer.
Milnor, J. (2000). The Poincaré conjecture. https://www.claymath.org/sites/default/files/poincare.
pdf
Milnor, J. (2003). Towards the Poincaré conjecture and the classification of 3-manifolds. Notices
of the AMS, 50(50), 1226–1233.
Musk, E. (1990) Neuralink, An integrated brain-machine interface platform with thousands of
channels. bioRxiv: 703801.https://doi.org/10.1101/70380
von Neumann, J. (1932). Mathematische Grundlagen der Quantenmechanik. Julius Spinger.
Nielsen, M. A., & Chuang, I. L. (2010). Quantum computation and quantum information, 10th
Anniversary Edition, CUP. Cambridge University Press.
Oriti, D. (2006). Disappearance and emergence of space and time in quantum gravity. In: D. Rickles
(Ed.), Studies in history and philosophy of modern physics. Special issue “Quantum Gravity”.
arXiv:1302.2849.
Padua, S. (2015). The thrilling adventures of Lovelace and Babbage—The (Mostly) true story of
the first computer. Pantheon.
Pauli, W. (1955). Die Wissenschaft und das abendländische Denken, Wolfgang Pauli Nachlass
im CERN-ARCH-PMC-06–291, WB 4/IVA, letter [2707]. http://cds.cern.ch/record/96387/files/
CERN-ARCH-PMC-06-291.pdf?version=1.
Pauka, J., et al. (2021). A cryogenic CMOS chip for generating control signals for multiple qubits.
Nature Electronics, 4, 64–70. https://doi.org/10.1038/s41928-020-00528-y
Penrose, R. (1974). The role of aesthetics in pure and applied mathematical research, southend-on-
sea. Bulletin of Mathematical Analysis and Applications, 10, 266–271.
Plato. (2003). Timaios. Reclam.
Poincaré, H. (1882). Theorie des Groupes Fuchsiens. Acta Mathematica, 1, 1–63.
Poincaré, H. (1895). Analysis Situs, Journal de l’École Polytechnique, 11e Série—Premier Cahier,
Paris. (Transl. by J. Stillwell: Papers on Topology, Analysis Situs and Its Five Supplements, Henri
Poincaré, 2009).
Poincaré, H. (1892). Sur l’Analysis situs, note de Henri Poincare. Comptes Rendus Des Seances
De L’academie Des Sciences, 115, 633–636.
Poincaré, H. (1900). Second supplement to analysis situs. Proceedings of the London Mathematical
Society, 32, 277–308.
Poincaré, H. (1904). Cinquième complément à l’analysis situs, Rend. Circ. Mat.: Palermo, 18,
45–110.
Poincaré, H. (1912). Pourquoi l’espace á trois dimensions. Revue De Métaphysique Et De Morale,
4, 483–504.
Polchinski, J. (1995). Dirichlet-Branes and Ramond-Ramond charges. arxiv.org/abs/hep-
th/9510017.
Pool, R. (2020). A new kind of quantum. https://spie.org/news/photonics-focus/novdec-2020/a-
new-kind-of-quantum?SSO=1.
Price, H. (2013). Taming the quantum spooks. https://aeon.co/essays/can-retrocausality-solve-the-
puzzle-of-action-at-a-distance.
Pusey, M. F., Barrett, J., & Rudolph, T. (2012). On the reality of the quantum state. arXiv:1111.
3328
Quehenberger, R. (2011). QC expert talk with Gerard ‘t Hooft (Utrecht University) 12 November
2011, Vienna. http://vimeo.com/34081638
Quehenberger, R., Weibel, P., Rauch, H., Katzgraber, H., & Friemel, R. (2012). A new digital
3-D dynamic geometry for the visualization of complex number space. In: Proceedings of
International Conference on Computational Electronics and Nanotechnology (ICOCENT-2012)
Conference, March 2012. Jaipur (pp. 287–292).
Quehenberger, R. (2012a). Quantum cinema expert-talk with Basil Hiley and Helmut Rauch.
vimeo.com/50374353.
274 R. C.-Z. Quehenberger

Quehenberger, R., & Weibel, P. (2012). Recording of “Quantum Fluxus Dada Duett“ with voice
of Erwin Schrödinger (recording “Do electrons think?” by courtesy of Ruth Braunitzer) Live
stream with Zabelka, M. (violin) as ARTs Birthday present 2012, Ö1 Kunstradio. http://www.
kunstradio.at/2012A/22_01_12.html
Quehenberger, R. (2012b). A reflection on theories of light. In: A. Khrennikov, H. Atmanspacher,
A. Migdall, & S. Polyakov (Eds.), Quantum theory: Reconsideration of foundations 6. AIP
Conference Proceedings 1508 (pp. 459–463).
Quehenberger, R. (2013). Quantum information traced back to ancient Egyptian mysteries.
Technoetic Arts: A Journal of Speculative Research, 11(3), 319–334.
Quehenberger, R. (2014a). A newly found golden heptahedron named epitahedron. In: G. Darvas
(Ed.) Journal Symmetry: Culture and Science. Symmetrion no. 25, 177–192.
Quehenberger, R. (2014b). A new visualization of the homology sphere. ICM 2014. In: International
Congress of Mathematicians, Seoul (KR) Poster presentation, P5 #29.
Quehenberger, R. (2014c). Epitadodecahedron, film presented at IMAGINARY at ICM 2014 &
International Conference of Mathematicians, ICM Seoul 2014. www.imaginary.org/film/the-
epita-dodecahedron-visualizing-poincares-dodecahedral-space
Quehenberger, R. (2016). Description of the cover-images: The Epitahedron. In: G. Darvas (Ed.),
Symmetry: Culture and science (Vol. 27(1), pp. 56–58). Symmetrion.
Quehenberger, R. (2017) A Proposal for a Psi-ontological model based on 5-dimensional
geometry. In: QCQMB Workshop: Quantum Contextuality in Quantum Mechanics and
Beyond, Prague 2017, http://www.psych.purdue.edu/~ehtibar/workshop/files/Poster_prag_Q
CQMB%202017_RCZQ_A0_corr.png
Quehenberger, R. (2018). A quest for an epistemic reset in higher dimensional space, Entropy, Best
Poster Awards at Linnaeus Conference: Towards Ultimate Quantum Theory (UQT). https://res.
mdpi.com/data/1st-place-r.c.-z.-quehenberger.pdf
Quehenberger, R. (2019) Enfolding the 5th dimension: With origami to the 3D representation of the
Penrose kites & darts pattern, symmetry: art and science, 2019. In: 11th Congress and Exhibition
of SIS Special Theme: Tradition and Innovation in Symmetry, Katachi 形 Kanazawa, Japan,
November 25–30, 2019.
Quehenberger, R., Kern L., & Tirelli, D. dC. Rubin, S. (2022). GAIA 5.0—A five-dimensional
geometry for the 3D visualization of Earth’ climate complexity. In: T. Lanza (Ed.) S. Illing-
worth et al., (Eds.) Special issue: Five years of earth sciences and art at the EGU (2015–2019).
Geoscience Communication, Copernicus, 2020–2021.
Quehenberger, R. (2021). Interactive comment on GAIA 5.0—A five-dimensional geometry for the
3D visualization of Earth’ climate complexity. Geoscience Communication Discussion. https://
doi.org/10.5194/gc-2020-27-AC3
Quehenberger, R. (2022, forthcoming) Kepler & die 5. Dimension, (Transl.: Kepler & the 5th
Dimension, forthcoming) Aichbergiana, Handreiche zum Alltag, No. 27.
Riemann, G. F. B. (1859). Über die Anzahl der Primzahlen unter einer gegebenen Grösse.
Monatsberichte der Berliner Akademie, 671–680.
Rauch, H., Zeilinger, A., Badurek, G., Wilfing, A., Bauspiess, W., & Bonse, U. (1975). Verification
of coherent spinor rotation of fermions. Physics Letters A, 54(6), 425–427.
Regev, O. (2004). Quantum computation and lattice problems. SIAM Journal on Computing, 33(3),
738–760.
Riesz, M. (1916). Sur l’hypothèse de Riemann. Acta Mathematica, 40, 185–190.
Rudolph, T. (2017). Q is for quantum. Kindle Edition by Terry Rudolph.
Rupprich, H. (Ed.) (1969). Dürer. Schriftlicher Nachlass, Berlin: Deutscher Verlag für Kunst-
wissenschaft, 1956/1966/1969, 3rd vol. 1969.
Schleich, W. P., Bezděková, I., Kim, M. B., Abbott, P. C., Maier, H., Montgomery, H. L., &
Neuberger, J. W. (2018). Equivalent formulations of the Riemann hypothesis based on lines
of constant phase. Physica Scripta, 93(6), [065201]. https://doi.org/10.1088/1402-4896/aabca9
Quantum Cinema and Quantum Computing 275

Schopenhauer, A. (1819). Die Welt als Wille und Vorstellung (Engl.: The World as Will and Repre-
sentation), in: Arthur Schopenhauers sämtliche Werke in sechs Bänden (Ed. Eduard Grisebach),
Verlag von Philipp Reclam jun., 1892.
Schönhage, A., & Strassen, V. (1971). Schnelle Multiplikation grosser Zahlen. Computing, 7, 281–
292.
Schrödinger, E. (1935). Die gegenwärtige Situation in der Quantenmechanik, Naturwissenschaften,
23:48, 807–812, 823–828; 844–849.
Schrödinger in a letter to Born, 14 April 1926, in Meyenn, 2010, p. 201) “... sich irgendwie mit der
Atommechanik von Heisenberg, Jordan und mir vereinigen lassen. Ich kann mir natürlich unge-
fähr denken, wie das geht, denn partielle Differentialgleichungen sind ganz allgemein äquivalent
mit quadratischen Formen von unendlich vielen Variabeln; aber ich wüsste doch gern genauer,
wie es zusammenhängt und wäre Ihnen sehr dankbar, wenn...”
Shannon, C. E (1948). A Mathematical Theory of Communication, 27, 379–423 & 623–656.
Shechtman, D., Blech, I., Gratias, D., & Cahn, J. W. (1984). Metallic phase with long-range
orientaional order and no translational symmetry. Physical Review Letters, 53(20), 1951–1953.
Shor, P. W. (1997). Polynomial-time algorithms for prime factorization and discrete logarithms on
a quantum computer. SIAM Journal on Computing, 26(5), 1484–1509.
Soiguine, A. (2017). Quantum computing in geometric algebra formalism: Light Beam Guide
Implementation Future Technologies Conference (FTC) 29–30 November 2017, Vancouver,
Canada
Svozil, K. (2000). The information interpretation of quantum mechanics. arxiv.org/abs/quant-
ph/0006033v2.
Threlfall, W. & Seifert, H. (1931).Topologische Untersuchung der Diskontinuitätsbereiche endlicher
Bewegungsgruppen des dreidimensionalen sphärischen Raumes, Mathematische Annalen 104(1),
1–70.
Thurston, W. P. ([1980] 2002). The geometry and topology of three-manifolds, notes distributed
by Princeton University, Electronic version 1.1, March 2002. http://www.msri.org/publications/
books/gt3m/
Toole, B. A. (2010). Ada, the Enchantress of Numbers: Poetical Science, Sausalito CA: Critical
Connection.
Tydecks, W. (2020). Re-entry nach Spencer-Brown, presentation at formlabor, Berlin, 31 January
2020.
Uchida, G., Bertlmann, R. A., & Hiesmayr, B. C. (2015) Entangled entanglement: The geometry
of GHZ states. Physics Letters A, 379, 2698, arxiv.org/abs/1410.7145.
Vasyliunas, V. M. (2005). Relation between magnetic fields and electric currents in plasmas. Annales
Geophysicae, 23(7), 2589–2597.
Vieira, L. V., Palazzo, R., & Faria, M. B. (2006). On the arithmetic Fuchsian groups derived from
quaternion orders VI. In: International Telecommunications Symposium (ITS2006). https://doi.
org/10.1109/ITS.2006.4433342
Verhulst, F. (2012). Henri Poincaré: Impatient Genius, Springer US.
Voisin, B., et al. (2020). Valley interference and spin exchange at the atomic scale in silicon. Nature
Communications. https://doi.org/10.1038/s41467-020-19835-1
Walleczek, J., Grössing, G., Pylkkänen, P., & Hiley, B. (2019). Emergent quantum mechanics:
David Bohm centennial perspectives. Entropy, 21(2), 113. https://doi.org/10.3390/e21020113
Weibel, P. (2005). Time slot—Geschichte und Zukunft der apparativen Wahrnehmung vom
Phenakistiskop bis zum Quantenkino, Cologne: Walther König.
Weihs, G., Jennewein, T., Simon, C., Weinfurter, H., & Zeilinger, A. (1998). Violation of Bell’s
inequality under strict Einstein locality conditions. Physical Review Letters, 81, 5039–5043.
Wheeler, J. A. (1983) Law without law. In: J. A. Wheeler & W. H. Zurek (Eds.). Quantum theory
and measurement. Princeton University Press.
Wheeler, J. A. (1989). Information, physics, quantum: the search for links. In: Proceedings III
International Symposium on Foundations of Quantum Mechanics (pp. 354–368). Tokyo.
276 R. C.-Z. Quehenberger

Wilczek, F. (2012). Quantum time crystals. Physical Review Letters, 109(16): 160401. arXiv:1202.
2539
Wüthrich, C. (2014). Raiders of the lost spacetime. In: D. Lehmkuhl (Ed.), Towards a theory of
spacetime theories. Birkhäuser. arXiv:1405.5552v1
Yau, S.-T., & Nadis, S. (2010). The shape of inner space, string theory and the geometry of the
universe’s hidden dimensions. Basic Books.
Zeilinger, A. (1999). A foundational principle for quantum mechanics. Foundations of Physics, 29,
631–643.
Zeilinger, A., Bernstein, H. J., & Horne, M. A. (1994). Information transfer with two-state, two-
particle quantum systems. Journal of Modern Optics, 41, 2375–2381.
Zimba, J., & Penrose, R. (1993). On Bell non-locality without probabilities: More curious geometry.
Studies in History and Philosophy of Science. SHPSB524
How to Make Qubits Speak

Bob Coecke, Giovanni de Felice, Konstantinos Meichanetzidis,

and Alexis Toumi

Abstract This is a story about making quantum computers speak, and doing so
in a quantum-native, compositional and meaning-aware manner. Recently we did
question-answering with an actual quantum computer. We explain what we did, stress
that this was all done in terms of pictures, and provide many pointers to the related
literature. In fact, besides natural language, many other things can be implemented
in a quantum-native, compositional and meaning-aware manner, and we provide
the reader with some indications of that broader pictorial landscape, including our
account on the notion of compositionality. We also provide some guidance for the
actual execution, so that the reader can give it a go as well.

Keywords Quantum natural language processing · Compositionality ·

ZX-calculus · String diagrams · DisCoPy software

1 Networks of Words

What is a word?
In particular, what is the meaning of a word? This could be what you find in a
dictionary. Then again, before there were dictionaries people also used words, and
so did people that could not read. In fact, most words we know we didn’t look up
in a dictionary. We probably learned them by hearing them used a lot. This idea of
meaning is closer to how machines learn meanings of words today: they learn them

B. Coecke (B) · G. de Felice · K. Meichanetzidis · A. Toumi

Cambridge Quantum/Quantinuum, 17 Beaumont St, Oxford OX1 2NA, UK
e-mail: [email protected]
G. de Felice
e-mail: [email protected]
K. Meichanetzidis
e-mail: [email protected]
A. Toumi
e-mail: [email protected]

© Springer Nature Switzerland AG 2022 277

E. R. Miranda (ed.), Quantum Computing in the Arts and Humanities,
https://doi.org/10.1007/978-3-030-95538-0_8
278 B. Coecke et al.

from the context in which they are used, an idea coined by Wittgenstein as “meaning
is use”, and expressed by Firth’s derived dictum (Firth, 1957):
“You shall know a word by the company it keeps”.
What is a sentence?
In particular, what is the meaning of a sentence? A sentence is not just a “bag of
words”,1 but rather, a kind of network in which words interact in a particular fashion.
In fact, in a very particular fashion, given that when we hear a sentence that we never
heard before, provided we do understand the words that occur in it, then we surely
also understand that sentence. That’s exactly why we don’t have dictionaries for
sentences—besides the fact a ‘sentence dictionary’ would take up an entire library,
if not more. There is an important academic question here:
• How do we deduce the meaning of a sentence given the meanings of its words?
We can also ask the converse question:
• Can we infer the meanings of words in sentences, from the meanings of sentences?
Both turn out to also be essential practical questions.
Some 10 years ago, in (Coecke et al., 2010), BC Mehrnoosh Sadrzadeh and Steve
Clark started to draw networks in order to address the first of these questions. These
networks look like this:

that
Alice likes (the) flowers Bob gives Claire

(1)

The usual technical term for these networks is “string diagram” (see e.g. Baez & Stay
2011). These string diagrams provided a straight-forward manner for how meanings
of words combine in order to produce the meaning of a sentence, even resulting in a
cover-heading feature in New Scientist.2
In order to better see how these string diagrams do so, let’s consider a simpler
example:
Alice hates Bob
(2)

The idea here is that the boxes represent the meanings of words and that the wires
are channels through which these meanings can be transmitted. So, in the example
above, the subject Alice and the object Bob are both sent to the verb hates, and

1 Here, ‘bag’ is a widely used technical term, referring to the fact that in many ‘natural language
processing’ applications grammatical structure has been entirely ignored for many years (Harris,
1954).
2 https://www.newscientist.com/article/mg20827903-200-quantum-links-let-computers-

understand-language/.
How to Make Qubits Speak 279

together they then make up the meaning of the sentence. This idea scales to much
bigger sentences like the one depicted above, and even to large text made up of
multiple sentences (Coecke, 2019).
This flow of words in sentences can be traced back to work originally started in the
1930s by Adjukiewicz (1935) and later by Bar-Hillel (1953), Chomsky (1956, 1957)
and Lambek (1958, 2008), among others. What they did was unifying grammatical
structures across different languages as a single mathematical structure. An overview
of these developments from the perspective of string diagrams is in Coecke (2013).

2 Language Is Quantum-Native

One particularly interesting aspect of this graphical framework for linguistics was
that the string diagrams were inherited from previous work that provided quantum
theory with a network-like language (Coecke, 2005). This work was accumulated
eventually in a 900-page book (Coecke & Kissinger, 2017).
In summary, a direct correspondence was established between, on the one hand,
the meanings of words and quantum states, and on the other hand, grammatical
structures and quantum measurements. This is illustrated in Fig. 1.
Obviously this led to the question: Can one make quantum computers handle
natural language? This was first proposed in a paper by Will Zeng and BC in 2016
(Zeng & Coecke, 2016), creating a new paradigm for natural language processing
(NLP) in a quantum computing context.
The idea of making quantum computers process natural language is just not only
incredibly cool but also a very natural thing to do for the reasons indicated above.
More researchers started to take an interest, and at some point Intel supported a basic
attempt to develop some of the ideas contained in (Zeng & Coecke, 2016) on their
quantum simulator (Intel News Release, 2019; O’Riordan et al., 2020).

single-system state GHZ-state two-system state single-system state

who

Quantum states She hates Bob

Quantum measurement

Bell-tests

Fig. 1 Illustration of how word meanings can be interpreted as quantum states, and grammatical
structure in terms of quantum measurement
280 B. Coecke et al.

However, there were some significant challenges to the proposal. Most impor-
tantly, there weren’t any sufficiently capable quantum computers that could imple-
ment the NLP tasks proposed. Additionally, an assumption was made that one could
encode word meanings on the quantum computer using quantum random access
memory (QRAM) (Giovannetti et al., 2008), which to this day, and despite theoreti-
cal progress and experimental proposals, remains a distant possibility.

3 Our New Attempt

In the past year, we have been examining ways to use existing NISQ (= Noisy
Intermediate Scale Quantum) devices, in the first instance one of IBM’s quantum
devices, for NLP.
The string diagrams as depicted above can’t be interpreted directly by IBM’s
machine, which instead, needs something in the form of a ‘quantum circuit’. Natural
language when we map it to a ‘quantum circuit skeleton’ now looks like this (Coecke
et al., 2020):

Alice αA αp αB Bob
βA βB

α
β

hates
α + γ
β
γ

(3)

In this picture as well as in our experiments we used the ‘ZX-calculus’ (Coecke &
Duncan, 2011) for drawing quantum circuits, which is again part of the same string
diagram language of quantum theory—we give the translation of ZX-calculus circuit
components in standard circuit gates in Fig. 2.
How to Make Qubits Speak 281

X-preparation X-phase gate Z-phase gate CNOT gate X-measurement

α α

⎛ ⎞
      1 0 0 0
1 1 0 1 0 ⎜0 1 0 0⎟

H◦ ◦H ⎜ ⎟ 1 0
0 0 eiα 0 eiα ⎝0 0 0 1⎠
0 0 1 0

Fig. 2 Translation of ZX-calculus circuit components in standard circuit gates. For a short tutorial
see (Coecke & Duncan, 2012), or for a more extensive ones see (Coecke & Kissinger, 2017; van de
Wetering, 2020)

The building blocks of the ZX-language are these ‘spiders’:

... ...
α α
... ... (4)

and rules for computing and reasoning look as follows:

...
α ...
... = = =
... α+β
β
...
...
(5)
So we are genuinely dealing with an entirely pictorial language, and in fact, recently it
was shown in (Hadzihasanovic, 2018) that all the equations that can be derived using
the usual quantum mechanical formalism, can also be derived using only pictures!
The key part of the passage from a sentence diagram representing grammatical
structure like (2) to a quantum circuit uses the ZX-calculus in a fundamental way.
For example, the fact that the CNOT-gate arises by sticking together two spiders as
follows:

(6)

is essential. Using this decomposition, the circuit arises as follows from (2):
282 B. Coecke et al.

*hates* Alice *hates* Bob

Alice Bob
=
(7)

We could then also decide to reduce the number of qubits as follows:

Alice Bob
Alice *hates* Bob
= *hates*

(8)

From this (3) arises when parametrising the word meanings by gates. A detailed
discussion of the entire passage is in (Coecke et al., 2020).
In the form (3), natural language can be implemented on NISQ devices, and of
course, will still work well as these devices scale in terms of size and performance.
Crucially, our solution provides a way forward in the absence of QRAM. By employ-
ing quantum machine learning we do not directly encode the meanings of words, but
instead quantum gates—those in the circuit (3) carrying Greek letters—learn their
meanings directly from text (Meichanetzidis et al., 2020). By way of analogy with
classical machine learning, in quantum machine learning we can indeed use quantum
circuits instead of classical neural networks in order to learn patterns from data (Ma
et al., 2019; Benedetti et al., 2019). Interestingly, neural network architectures are the
state-of-the-art in classical NLP, but the majority of methods do not take advantage
of grammatical structures. In contrast, we saw that our approach to natural language
naturally accommodates both grammar and meaning.
Using our framework, once the meanings of words and phrases are encoded as
quantum gates, we are able to encode the meaning of grammatical sentences on
quantum hardware. Posing a question to the quantum computer, constructed by the
vocabulary and grammar the quantum computer has learned, it returns the answer—
as illustrated in Fig. 3.3
Naturally, we next turned our attention toward the design and execution of an
experiment that is non-trivial, not least of all since our design is predicated on the
program being scalable. This means that the dimension of the meaning space grows
significantly with the number of qubits available whilst the size of the circuits dictated
by the grammar does not grow too large with the size of the sentence.

3Earlier theoretical work by our team towards question-answering includes (Coecke et al., 2018;
Felice et al., 2019).
How to Make Qubits Speak 283

Fig. 3 A quantum computer putting an end to social media disputes

4 The Actual Experiment

Once a quantum circuit is created from a sentence, it needs to be evaluated in order

to compute the meaning of that sentence. We may choose to perform this evaluation
on a classical computer, where we employ state-of-the-art methods for performing
the costly task of multiplying exponentially big matrices, or, we may choose to
implement the circuit on a quantum computer. This is of course what we decided
to do. A schematic presentation of the experiment on IBM quantum hardware is in
Fig. 4.
As we see in the quantum circuit (3), each of the parts of speech (subject, object,
verb) is in the quantum circuit a function of some parameters. For example, there are
sets of parameter values α A , β A , α B , β B , α, β, . . . such that:

Fig. 4 A schematic presentation of the actual experiment on IBM quantum hardware, depicting
both the training and testing parts, with θ = {α A , β A , α B , β B , α, β, . . .}
284 B. Coecke et al.

subject(α A , β A ) = Alice
object(α B , β B ) = Bob
verb(α, β, . . .) = hates

The values are determined empirically by a text corpus and are then used to answer
questions about the corpus. In order to ensure that our experiment can be executed
effectively on near-term NISQ devices, but at the same time be complex enough to
be interesting, we chose a vocabulary of a few words, for example:

{Alice, Bob, loves, hates, rich, silly}

and generated not some, but all grammatical sentences from their combinations. From
these sentences, we created their corresponding parameterised circuits. Moreover,
we interpret the language diagrams such that the sentence space is one-dimensional,
i.e. just a number indicating the truth-value of the sentence:

0 ≤ sentence(α A , β A , α B , β B , α, β, . . .) ≤ 1

A value close to 1 represents “tr ue” and a value close to 0 indicates “ f alse”.
The labeled toy corpus would look like:

Cor pus = {(Alice loves Bob, f alse), (Bob is silly, tr ue), (Alice is rich, tr ue), ...}

Now that we have our corpus of sentences we split the corpus:

Cor pus = T rain ∪ T est

in a training set T rain and a test set T est. Sentences in the training set T rain are
used to do supervised quantum machine learning in order to learn the parameters
that result in the correct measurement of the truth-labels. In this way, the parame-
ters for the circuits that prepare the meaning states for nouns {Alice, Bob}, verbs
{is, loves, hates}, and adjectives {rich, silly}, are learned.
The scheme for learning the parameters for words in sentences in the training set
is as follows. The circuit of a sentence in the training set is evaluated for the current
set of parameters on the quantum computer. By sampling measurement outcomes
we estimate:
|sentence(α A , β A , α B , β B , α, β, . . .)|2

This number is read by a classical computer that checks how well this matches the
desired truth label of the sentence. If there is a mismatch, our system updates the
parameters so that the quantum computer may evaluate an updated circuit. Iterating
this procedure until the parameters converge and all truth labels are reproduced for
the sentences in the training set.
How to Make Qubits Speak 285

After training, sentences in T est are used to estimate how well the truth labels
of new sentences, i.e. not in T rain, are inferred. These new sentences share the
same vocabulary as the sentences used for training, but they are grammatically and
semantically different.
Note that finding the optimal sequence of updates is, in general, a so-called ‘hard
optimization problem’, so it is important that our quantum meaning space is well
designed and allows the learning to be as tractable as possible. This design feature
is critical.
With this learning framework, we can now ask questions of the quantum com-
puter, as long as the questions are grammatical sentences expressed in terms of the
vocabulary and grammar used during training. We are pleased to add that in our
experiment, questions can, in fact, be posed as compositions of already learned sen-
tences. For example, we can use the relative pronoun who (Sadrzadeh et al., 2013)
(which we model in terms of CNOT gates within a quantum circuit) and ask:

Does Bob who is silly love Alice who is rich?

This is the same as asking whether:

who who
Bob is silly loves Alice is rich

(9)
is tr ue. This amounts to evaluating a bigger quantum circuit than the one that the
model has been trained on. However, because the model was trained on the same
grammar and vocabulary as used to express the question, we get the expected truth
label, false in this case.
One critical enabling feature of our experiment is making computers understand
the pictures that we have been drawing. For this we used GdF and AT’s DisCoPy
Python library (Felice et al., 2020). Another critical feature is effective and efficient
compiling and optimization. In order to run a circuit on a quantum device, it needed
to be compiled. Compiling entails morphing the circuit such that quantum operations
are expressed in terms of device-native operations, as well as accommodating for
the quantum processor’s restricted connectivity. For this task, we used Cambridge
Quantum Computing’s quantum software development platform, t|ket (Sivarajah
et al., 2020) which again crucially makes use of ZX-calculus.
286 B. Coecke et al.

5 QNLP

We’ve done it, it exists, so let’s give the kid a name: ‘quantum natural language pro-
cessing’, or QNLP in short. A much more detailed account on QNLP as well as the
data for a larger-scale experiment can be found in (Coecke et al., 2020; Meichanet-
zidis et al., 2020; Lorenz et al., 2021). With the successful execution of QNLP on
quantum hardware, and the great promise for the future, Cambridge Quantum Com-
puting has now resourced an Oxford based team dedicated to QNLP, and structure-
driven quantum AI more generally, evidently including each of us. Let’s single out
a few special features of QNLP.
Firstly, as already indicated above, QNLP is quantum-native. What we mean by
that is that the model of natural language that we employ is a quantum model (Clark
et al., 2014). Not just because we think that having a quantum model is particularly
cool, but because it was effectively the case that the most economic manner for
bringing language meaning and grammatical structure together was provided by the
‘categorical quantum mechanics’ framework (Abramsky & Coecke, 2004; Coecke,
2005; Coecke & Kissinger, 2017).4 The immediate consequence of this is that when
we try to ‘simulate’ QNLP on a classical computer, it would be exponentially expen-
sive, just as it is the case for simulation quantum systems on a classical computer.
Hence, QNLP truly loves being on quantum hardware.
Secondly, it is fair to say that it is meaning-aware, at the very least, it is definitely
more meaning-aware than the current deep learning based NLP implementations.
Indeed, the language pictures clearly indicate how the meanings flow in a meaningful
way. We say a bit more about this below in Sect. 9.
Thirdly, as already indicated in (Zeng & Coecke, 2016), there are many more
advantages to using quantum computers, most notably, we get quantum speed-up for
a large variety of NLP tasks, like question-answering. This space of opportunities is
in fact still largely unexploited, and the mission of our QNLP team is to do just that,
in as many possible manners as we can.

6 Beyond Language

We are now well on the way to make qubits speak. But can we make them do other
things as well? We are pretty sure we can, a first thing could be to make qubits
represent all of the inputs that we get through our fleshy physical embodiment,
like taste, smell, vision, hearing. The technical term for the kinds of spaces that
represent these sensory modes is ‘conceptual spaces’. They were introduced by Peter
Gärdenfors (Gärdenfors, 2000, 2014). There already are existing frameworks that
represent the input of these senses in a manner that exactly matches our language

4Categorical quantum mechanics is really just a fancy name for doing quantum mechanics in terms
of pictures.
How to Make Qubits Speak 287

diagrams, and hence allow for these sensory inputs to interact with each other (Bolt
et al., 2018).
These conceptual spaces are convex, and fit well with some other work within
the framework of our natural language diagrams, namely on the use of the quantum
mechanical concept of density matrices, which also form convex spaces (Ludwig,
1985; Barrett, 2007). This additional convex structure allows one to bring many more
linguistic features into play, like word ambiguity. For example, the word queen is
clearly ambiguous, as it can be a royal, a rock band, a bee, a chess piece, a playing card,
etc. In order to represent that ambiguity, we simply add all the different meanings in
order to form a density matrix (Piedeleu et al., 2015):

ρqueen = |queen-royalqueen-royal|
+ |queen-bandqueen-band|
+ |queen-beequeen-bee|
+ |queen-chessqueen-chess|
+ |queen-cardqueen-card|
+ ...

Alternatively, these density matrices can also be used to express hierarchical

relationships between words (Bankova et al., 2019), for example, lion, tiger and
cheeta, are examples of big cat, which further generalises to mammal, which
generalises to vertebrate, etc. Using density matrices we encode this hierarchy
as follows:

ρlion = |00|
ρtiger = |++|
ρcheeta = |−−|
ρbig cat = |00| + |11|
ρmammal = |00| + |11| + |22|
ρvertebrate = |00| + |11| + |22| + |33|

In (Coecke & Meichanetzidis, 2020; Lewis, 2019a, b; De las Cuevas et al., 2020;
Meyer & Lewis, 2020) this use of density matrices in language is further elaborated
upon.

7 String Diagrams Are Everywhere

Sticking with the theme of big cats, what’s the difference between a tiger and a lion?
In particular, why does a tiger have stripes, while a lion doesn’t (Fig. 5).
A more traditional scientist will tell you that these days we understand all of that
perfectly well. Before, if we dissected these animals, we found exactly the same
288 B. Coecke et al.

Fig. 5 Two close relatives with very different coats

organs. If we do some smaller scale animal chopping, we end up with cells that
again have exactly the same structure. However, digging even further, and hitting the
molecular level, we encounter DNA, and then we truly understand the difference.
Really? Some obscure humongously large code that is impossible to grasp by any
human ‘explains’ the difference? It explains as much the difference between these
two animals as the codes of two computer programs written in terms of 0’s and 1’s
explain the difference between accounting software and the operating system of your
smart phone.
That said, most of the exact sciences have adopted this perspective of understand-
ing things by breaking them down to smaller things, like elementary particles in
particle physics, and the elements of a set in set-theory. In contrast, in the arts and
humanities things really get their meanings by putting them in a context. For example,
in the social sciences properties are mostly thought of in terms of behaviours, rather
than by trying to build a map of the inside of someone’s brain, and most modern art
has very little meaning without a context.
We think that we can all agree that for our tiger and our lion, what explains the
difference is really the hunting process which each of them needs to do in order to
survive, and their respective coats provide the best camouflage for doing so, given they
hunt in very different environments. While of course these differences are encoded in
the genes, it is the process of evolution that makes our tiger and our lion look as they
do. The role of DNA in explaining the stripes of tigers is similar to the role of our
parameters α A , β A , α B , . . . in computing the meaning of sentences. Indeed, learning
the optimal parameters is an evolution-like process that depends on the environment
of the words, i.e. on the training data that we feed the machine.
So the two views explained above are really about either focussing on components
vs. context, and the latter is very much in line with Wittgenstein’s “meaning is use”.
We are saying all of this because the language diagrams that we saw above are the
mathematical incarnation of this way of thinking, for example:
How to Make Qubits Speak 289

tiger hunts deer in mangrove swamp

(10)

Indeed, this way of thinking requires a completely different kind of mathematics,

be it in the form of our diagrams, or its symbolic counterpart, so-called monoidal
categories.
This is not the place to go into this new kind of mathematics, but the interested
reader may want to check out the book (Coecke & Kissinger, 2017), which is probably
the most comprehensive account on these language diagrams, and their communality
with quantum theory. Shorter more informal accounts can be found in (Coecke, 2017).
For the braver reader, the symbolic counterpart of all of this exists within the realm
of category theory, and tensor categories in particular (Coecke & Paquette, 2011;
Selinger, 2011). Meanwhile, string diagrams have started to pop up in a wide variety
of disciplines, for example, in computer science (Abramsky, 1996; Pavlovic, 2013),
where they have been for a while, in engineering (Bonchi et al., 2014; Baez & Fong,
2015), in economy and game theory (Ghani et al., 2018; Felice et al., 2020), and
even in cognition (Tull & Kleiner, 2020; Signorelli et al., 2020).
We believe that this is the kind of mathematics that would also serve the social
sciences very well. As it is quite new, most mathematics that one finds in the social
sciences is statistics, which, in all fairness, is what one uses in absence of anything
better.
With new mathematics also come new software tools. For this we are using the
new very flexible DisCoPy software (Felice et al., 2020) which we already mentioned
above, and continues to be further developed by AT, GdF and collaborators. We’re
inviting everyone to play around with this toolbox, and even better, help us out
further developing DisCoPy—we may even reward you with a pie. For ZX-calculus
specifically there is now also the PyZX software (Kissinger & van de Wetering,
2019), which communicates directly with DisCoPy. PyZX also helps us out with our
QNLP experiments, by making the circuits that we feed into the quantum computer
as simple as possible, a task for which ZX-calculus is now setting the state-of-the-art
(Duncan et al., 2019; Cowtan et al., 2019; Beaudrap et al., 2020).

8 Compositionality

Above we saw how we can turn language diagrams into circuits. A typical thing
about circuits is, of course, that you can compose different circuits together for
forming bigger circuits. This is the principle of “compositionality”, and in fact,
string diagrams themselves are a result of this way of thinking. We can now, once
we moved to circuits, compose several sentences together in order to produce larger
text:
290 B. Coecke et al.

Alice hates Bob likes beer

(11)

But why do we have to go to quantum circuits to do so? Indeed, it goes without

saying that in absence of quantum circuits we can also compose sentences in order
to form larger text, so there should be manner for doing so without having to dive in
the realm of the quantum world.
This is indeed the case, when stopping halfway in the middle when passing from
language diagrams to quantum circuits (Coecke, 2019). There we also find circuits
that can be composed, and interestingly, some of the features specific to different
languages (e.g. specific orderings of words like subject-verb-object in English) are
gone (Coecke & Wang, 2020). So in a manner we end up with something that is more
universal than language diagrams, and, for example, stands in a particularly close
relationship with visual representations (Coecke, 2019). Here’s the language circuit:

Harmonica Snaky Frank Claudio

brother

hangs

gang

shoots

shoots

(12)

and here’s the corresponding visual representation:

How to Make Qubits Speak 291

Harmonica Snaky Frank Claudio

(13)

Maybe there is something deeper here. Maybe the origin of language should be
looked for in the visual world, rather than in some symbolic abstract realm. The
passage from these visual circuits to the more complex grammatical structure may
actually boil down to forcing something that wants to live in two (or more) dimensions
on a line, and force additional bureaucracy (or, bureau-crazy?) like the ordering of
words, which ends up being different for different languages. Why forced into one-
dimension? Since it seems to be hard for us humans to communicate verbally in any
other way than in terms of one-dimensional strings of sounds.
In fact, the very success of string diagrams is that it allows things to live in two
(or more) dimensions. Here’s a demonstration of this fact. Consider the operations of
parallel composition ⊗ (a.k.a. ‘while’) and sequential composition ◦ (a.k.a. ‘after’).
Typically, in textbooks, this structure is represented by one-dimensional strings
of algebraic symbols, and then, equations are needed to express their interaction
(a.k.a. bifunctoriality (Mac Lane, 1998)):

(g1 ⊗ g2 ) ◦ ( f 1 ⊗ f 2 ) = (g1 ◦ f 1 ) ⊗ (g2 ◦ f 2 )

292 B. Coecke et al.

On the other hand, representing it in terms of two-dimensional diagrams we have:

       
f1 f2
g1 ⊗ g2 ◦ f1 ⊗ f2 = g1 g2 ◦ f1 f2 =
g1 g2

⎛ ⎞ ⎛ ⎞
   
⎜ f1 ⎟ ⎜ f2 ⎟ f1 f2
⎜ ⎟ ⎜ ⎟
g1 ◦ f1 ⊗ g2 ◦ f2 =⎜ ⎟⊗⎜ ⎟ =
⎝ g ⎠ ⎝ g ⎠ g1 g2
1 2

That is, by using the two-dimensional format of diagrams, the symbolic equation
is always satisfied, so it is not needed anymore. In other words, the equation was a
piece of bureaucracy due to the one dimensional symbolic representation.

9 Meaning-Aware Relational Wholism

Let’s end with a bit of philosophy, which puts our language diagrams within a broader
context of different approaches that one may take towards machine learning. We’ll
again do this by means of a metaphor. Assume one wants to design a home. One way
to do so would be to train a machine using many different images of homes, possibly
with some indications of appreciation. Then, let the machine be creative and design
a home based on all of that input.
And it comes up with an amazing design. But what dit it really do? It kind of
carved everything out of one big block of stuff, and there will be forks and plates on
the table, chairs at the table, pictures on the wall, since that’s were they were in all of
the training images. However, the machine is unaware of what the use is of each of
these objects, and in particular that there is a very special relationship between the
plate and the fork, and also with the chair at the table (and less so with the picture
on the wall), and even more so, that there is a person who happens to be not in the
picture, but who wants to exploit this special relationship with respect to the food
they are cooking in the kitchen.
On the other hand, our language diagrams, when applied to a broader context, can
capture these relationships, in contrast to the black-box approach that is taken within
machine learning:
How to Make Qubits Speak 293

chair table plate Bob food

is at

is on

is on

is in

is in

(14)

The kind of thinking encoded in our language diagrams, which is fundamentally

wholistic while still aiming to grasp relationships between the parts, is what we
mean by the title of this section.
What our language diagrams do is express how words interact with each other.
Once we have established that structure, one thing that we can do is compute the
meaning of a sentence from the meanings of words. The more interesting thing
however, is to derive the meanings of words from the meaning of sentences. This is
exactly what we did in our QNLP experiment: from the knowledge that the sentences
in T rain were true, we figured out the optimal parameters α A , β A , α B , β B , α, β, . . .
which determine the meanings of the words.
This is really what this paper was all about, instantiated in order to make qubits
speak. After this philosophy interlude...

10 ...Let’s Get Playing!

First, we need some data to play with. For this we have two options: (1) we generate
an artificial dataset by designing the grammar ourselves, this is the simplest option
and the one we have implemented in (Coecke et al., 2020) or (2) we take some real-
world text data (e.g. every sentence from Alice in Wonderland) and use a parser to
construct the diagrams automatically for us. In any case, we begin our experiment
with a collection of diagrams for sentences, with some annotation telling us whether
each sentence is true or false (or whether or not it comes from Alice in Wonderland).
Next, we need to choose an ansatz for each type of word, i.e. an appropriate circuit
shape. Instantaneous quantum polynomials (IQP) (Shepherd & Bremner, 2009) are
a reasonable choice, since they are shallow circuits but are still believed to be hard
to simulate for classical computers. Once we’ve chosen the circuit shape for each
type, this defines a landscape, the total number of parameters defining our model:
the sum of the parameters defining each word. For each point in this landscape, i.e.
for each combination of parameters, we can run the circuits corresponding to each
294 B. Coecke et al.

sentence in our dataset and check how far away we are from the correct label. The
sum of the distance between our model’s prediction and the correct label is called
the loss function, this is the objective that we want to minimise.
Once we have defined the landscape and the loss, we can feed them as input
to the optimisation algorithm of our choice. The easiest choice, and the one we
implemented in (Coecke et al., 2020), is to pick a black box optimiser such as
simultaneous perturbation stochastic approximation (SPSA) (Spall, 1992). A more
advanced solution is to go fully diagrammatic and compute the gradient of our loss
directly from the shape of the diagrams for sentences (Toumi et al., 2021). When
the process has converged to some (locally) optimal parameters, tadam! We have
successfully trained our quantum natural language model. We may now evaluate
it on some testing data, typically a subset of the sentences that we withhold from
the training set. Note that crucially, even though our model has never seen these
sentences during training, hopefully it has already learnt the meanings for the words,
and it knows how to combine these meanings in new ways.
Thus, a typical QNLP experiment may be summarised as follows: (1) we draw
the diagrams for each sentences, (2) we pick a circuit shape for each word type,
(3) we define a loss function, (4) we find the optimal parameters and (5) we test
the model’s prediction. These five steps fit in a few lines of Python using the tools
from the DisCoPy library, the result can then be presented in the form of a Jupyter
notebook, see the documentation for plenty of examples:
https://discopy.readthedocs.io/en/main/notebooks.qnlp.html
Also, here is a more detailed tutorial as well:
https://discopy.readthedocs.io/en/main/notebooks/qnlp-tutorial.html

11 Outlook

So what’s next? Well, as we speak the editor of this volume is himself getting to play,
and obviously, we will also embark on a musical adventure. Evidently, we will also
further develop QNLP bringing in more conceptual depth, for the specific case of
language as well as towards broader AI features. Experimentally, we keep pushing
along as hardware improves.
All together this is just the beginning of an adventure that, we hope, on the one
hand, will create the next generation of AI, and on the other hand, will further the
quest towards better understanding the human mind and its interaction with the world
around us.
How to Make Qubits Speak 295

References

Abramsky, S. (1996). Retracing some paths in process algebra. In CONCUR’96: concurrency theory,
vol. 1119 of Lecture notes in computer science (pp. 1–17). Springer.
Abramsky, S., & Coecke, B. (2004). A categorical semantics of quantum protocols. In Proceed-
ings of the 19th annual IEEE symposium on logic in computer science (LICS) (pp. 415–425).
arXiv:quant-ph/0402130.
Adjukiewicz, K. (1935) Die syntaktische Konnexität. Studia Philosophica.
Baez, J. C., & Fong, B. (2015). A compositional framework for passive linear networks.
arXiv:1504.05625
Baez, J. C., & Stay, M. (2011). Physics, topology, logic and computation: a Rosetta stone. In
B. Coecke (ed.), New structures for physics, lecture notes in physics (pp. 95–172). Springer.
Bankova, D., Coecke, B., Lewis, M., & Marsden, D. (2019). Graded hyponymy for compositional
distributional semantics. Journal of Language Modelling, 6(2), 225–260.
Bar-Hillel, Y. (1953). A quasiarithmetical notation for syntactic description. Language, 29, 47–58.
Barrett, J. (2007). Information processing in generalized probabilistic theories. Physical Review A,
75, 032304.
Benedetti, M., Lloyd, E., Sack, S., & Fiorentini, M. (2019). Parameterized quantum circuits as
machine learning models. Quantum Science and Technology, 4, 043001.
Bolt, J., Coecke, B., Genovese, F., Lewis, M., Marsden, D., & Piedeleu, R. (2018). Interacting
conceptual spaces I: Grammatical composition of concepts. In M. Kaipainen, A. Hautamäki,
P. Gärdenfors, & F. Zenker (eds.), Concepts and their applications, synthese library, studies in
epistemology, logic, methodology, and philosophy of science. Springer.
Bonchi, F., Sobocinski, P., & Zanasi, F. (2014). A categorical semantics of signal flow graphs. In
CONCUR’14: concurrency theory. Lecture notes in computer science (vol. 8704, pp. 435–450).
Springer.
Chomsky, N. (1956). Tree models for the description of language. I.R.E. Transactions on Information
Theory, IT, 2, 113–124.
Chomsky, N. (1957). Syntactic Structures. Mouton.
Clark, S., Coecke, B., Grefenstette, E., Pulman, S., & Sadrzadeh, M. (2014). A quantum teleportation
inspired algorithm produces sentence meaning from word meaning and grammatical structure.
Malaysian Journal of Mathematical Sciences, 8, 15–25. arXiv:1305.0556.
Coecke, B. (2005). Kindergarten quantum mechanics. In A. Khrennikov (ed.), Quantum theory:
Reconsiderations of the foundations III (pp. 81–98). AIP Press. arXiv:quant-ph/0510032
Coecke, B. (2013). An alternative Gospel of structure: Order, composition, processes. In C. Heunen,
M. Sadrzadeh, & E. Grefenstette (eds.), Quantum physics and linguistics. A compositional, dia-
grammatic discourse (pp. 1 – 22). Oxford University Press. arXiv:1307.4038
Coecke, B. (2017). From quantum foundations via natural language meaning to a theory of every-
thing. In S. B. Cooper & M. I. Soskova (eds.), The incomputable: Journeys beyond the turing
barrier, theory and applications of computability (pp. 63–80). Springer International Publishing.
arXiv:1602.07618
Coecke, B. (2019). The mathematics of text structure. arXiv:1904.03478
Coecke, B., De Felice, G., Marsden, D., & Toumi, A. (2018). Towards compositional distributional
discourse analysis. In M. Lewis, B. Coecke, J. Hedges, D. Kartsaklis, & D. Marsden (eds.),
Proceedings of the 2018 workshop on compositional approaches in physics, NLP, and social
sciences. Electronic proceedings in theoretical computer science (vol. 283, pp. 1–12)
Coecke, B., de Felice, G., Meichanetzidis, K., & Toumi, A. (2020). Foundations for near-term
quantum natural language processing. arXiv preprint arXiv:2012.03755
Coecke, B., & Duncan, R. (2011). Interacting quantum observables: Categorical algebra and dia-
grammatics. New Journal of Physics, 13, 043016. arXiv:quant-ph/09064725
Coecke, B., & Duncan, R. (2012). Tutorial: Graphical calculus for quantum circuits. In International
workshop on reversible computation (pp. 1–13). Springer
296 B. Coecke et al.

Coecke, B., & Kissinger, A. (2017). Picturing quantum processes. A first course in quantum theory
and diagrammatic reasoning. Cambridge University Press.
Coecke, B., & Meichanetzidis, K. (2020). Meaning updating of density matrices. arXiv:2001.00862
Coecke, B., & Paquette, É. O. (2011). Categories for the practicing physicist. In B. Coecke (ed.),
New structures for physics, lecture notes in physics (pp. 167–271.) Springer. arXiv:0905.3010
Coecke, B., Sadrzadeh, M., & Clark, S. (2010). Mathematical foundations for a compositional
distributional model of meaning. In J. van Benthem, M. Moortgat, & W. Buszkowski (eds.), A
festschrift for Jim Lambek. Linguistic analysis (vol. 36, pp. 345–384). arxiv:1003.4394
Coecke, B., & Wang, V. (2020). De-humanising grammar for improved machine use. Draft paper.
Cowtan, A., Dilkes, S., Duncan, R., Simmons, W., & Sivarajah, S. (2019). Phase gadget synthesis
for shallow circuits. arXiv preprint arXiv:1906.01734
de Beaudrap, N., Bian, X., & Wang, Q. (2020). Fast and effective techniques for t-count reduc-
tion via spider nest identities. In 15th conference on the theory of quantum computation, com-
munication and cryptography (TQC 2020). Schloss Dagstuhl-Leibniz-Zentrum für Informatik.
arXiv:2004.05164
de Felice, G., Di Lavore, E., Román, M., Toumi, A. (2020). Functorial language games for question
answering. arXiv preprint arXiv:2005.09439
de Felice, G., Meichanetzidis, K., & Toumi, A. (2019). Functorial question answering. arXiv preprint
arXiv:1905.07408
de Felice, G., Toumi, A., & Coecke, B. (2020). DisCoPy: Monoidal categories in python. Accepted
for ACT 2020. arXiv preprint arXiv:2005.02975
De las Cuevas, G., Klingler, A., Lewis, M., & Netzer, T. (2020). Cats climb entails mammals move:
preserving hyponymy in compositional distributional semantics. arXiv:2005.14134 [quant-ph]
Duncan, R., Kissinger, A., Pedrix, S., & van de Wetering, J. (2019). Graph-theoretic simplification
of quantum circuits with the zx-calculus. arXiv preprint arXiv:1902.03178
Firth, J. R. (1957). A synopsis of linguistic theory, 1930–1955. Studies in Linguistic Analysis, 1–32.
Gärdenfors, P. (2000). Conceptual spaces: The geometry of thought. MIT Press.
Gärdenfors, P. (2014). The geometry of meaning: Semantics based on conceptual spaces. MIT Press.
Ghani, N., Hedges, J., Winschel, V., & Zahn, P. (2018). Compositional game theory. In Proceedings
of the 33rd annual ACM/IEEE symposium on logic in computer science, (pp. 472–481). ACM.
Giovannetti, V., Lloyd, S., & Maccone, L. (2008). Quantum random access memory. Physical review
letters, 100(16), 160501.
Hadzihasanovic, A., Ng, K. F., & Wang, Q. (2018). Two complete axiomatisations of pure-state
qubit quantum computing. In Proceedings of the 33rd annual ACM/IEEE symposium on logic in
computer science (pp. 502–511). ACM.
Harris, Z. S. (1954). Distributional structure. Word, 10(2–3), 146–162.
Kissinger, A., & van de Wetering, J. (2019). Pyzx: Large scale automated diagrammatic reasoning.
arXiv preprint arXiv:1904.04735
Lambek, J. (1958). The mathematics of sentence structure. American Mathematics Monthly, 65.
Lambek, J. (2008). From word to sentence. Milan: Polimetrica.
Lewis, M. (2019a). Compositional hyponymy with positive operators (pp. 638–647). Varna, Bul-
garia: In Proceedings of Recent Advances in Natural Language Processing.
Lewis, M. (2019b). Towards negation in discocat. In: Proceedings of SemSpace.
Lorenz, R., Pearson, A., Meichanetzidis, K., Kartsalkis, D., & Coecke, B. (2021). Qnlp in
practice: Running compositional models of meaning on a quantum computer. arXiv preprint
arXiv:2102.12846.
Ludwig, G. (1985). An axiomatic basis of quantum mechanics (vol. 1). Berlin Heidelberg: Derivation
of Hilbert Space. Springer-Verlag.
Ma, Y., Tresp, V., Zhao, L., & Wang, Y. (2019). Variational quantum circuit model for knowledge
graph embedding. Advanced Quantum Technologies, 1800078.
Mac Lane, S. (1998). Categories for the working mathematician. Springer.
How to Make Qubits Speak 297

Meichanetzidis, K., Gogioso, S., de Felice, G., Toumi, A., Chiappori, N., & Coecke, B. (2020).
Quantum natural language processing on near-term quantum computers. Accepted for QPL 2020.
arXiv preprint arXiv:2005.04147
Meichanetzidis, K., Toumi, A., de Felice, G., & Coecke, B. (2020). Grammar-aware question-
answering on quantum computers. arXiv:2012.03756
Meyer, F., & Lewis, M. (2020). Modelling lexical ambiguity with density matrices. In Proceed-
ings of the 24th Conference on Computational Natural Language Learning (pp. 276–290).
arXiv:2010.05670 [cs.CL]
O’Riordan, L. J., Doyle, M., Baruffa, F., & Kannan, V. (2020). A hybrid classical-quantum workflow
for natural language processing. arXiv preprint arXiv:2004.06800
Pavlovic, D. (2013). Monoidal computer I: Basic computability by string diagrams. Information
and computation, 226, 94–116.
Piedeleu, R., Kartsaklis, D., Coecke, B., & Sadrzadeh, M. (2015). Open system categorical quantum
semantics in natural language processing. In 6th conference on algebra and coalgebra in computer
science (CALCO 2015). Schloss Dagstuhl-Leibniz-Zentrum fuer Informatik.
Intel News Release. (2019). Intel to support the Irish Centre for High End Computing on new col-
laborative quantum computing project, March 6, 2019. https://newsroom.intel.ie/news-releases/
intel-to-support-the-irish-centre-for-high-end-computing-on-new-collaborative-quantum-
computing-project/
Sadrzadeh, M., Clark, S., & Coecke, B. (2013). The Frobenius anatomy of word meanings I: Subject
and object relative pronouns. Journal of Logic and Computation, 23, 1293–1317. arXiv:1404.5278
Selinger, P. (2011). A survey of graphical languages for monoidal categories. In B. Coecke (ed.),
New structures for physics, Lecture notes in physics (pp. 275–337). Springer. arXiv:0908.3347
Shepherd, D., & Bremner, M. J. (2009). Temporally unstructured quantum computation. Proceed-
ings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 465(2105), 1413–
1439.
Signorelli, C. M., Wang, Q., & Khan, I. (2020). A compositional model of consciousness based on
consciousness-only. arXiv preprint arXiv:2007.16138
Sivarajah, S., Dilkes, S., Cowtan, A., Simmons, W., Edgington, A., & Duncan, R. (2020). t|ket>:
A retargetable compiler for NISQ devices. arXiv preprint arXiv:2003.10611
Spall, J. C., et al. (1992). Multivariate stochastic approximation using a simultaneous perturbation
gradient approximation. IEEE Transactions on Automatic Control, 37(3), 332–341.
Toumi, A., Yeung, R., & de Felice, G. (2021). Diagrammatic differentiation for quantum machine
learning. arXiv preprint arXiv:2103.07960
Tull, S., & Kleiner, J. (2020). Integrated information in process theories. arXiv preprint
arXiv:2002.07654
van de Wetering, J. (2020). ZX-calculus for the working quantum computer scientist.
arXiv:2012.13966
Zeng, W., & Coecke, B. (2016). Quantum algorithms for compositional natural language processing.
Electronic Proceedings in Theoretical Computer Science, 221. arXiv:1608.01406
A Quantum Computing Approach
to Harnessing the Logic of the Mind
for Brain–Computer Interfacing

Eduardo Reck Miranda

Abstract This chapter presents a new quantum computing-based method to study

and harness neural correlates of mental activity for the development of Brain–
Computer Interface (BCI) systems. It introduces the notion of logic of the mind,
whereby neurophysiological data are encoded as logical expressions representing
mental activity. These representations then become instructions for BCI control.
Long complex logical expressions are intractable for classical computers. Hence, a
quantum computing algorithm is proposed. Despite the fact that quantum hardware
for running the proposed algorithm is not yet available commercially, we provide a
simulation, which demonstrates it plausibility. An attempt at running the system on a
quantum computer is also detailed, accompanied with an explanation of current limi-
tations. In addition to building BCI systems, we argue that our quantum computing-
based approach to study brain functioning and neural correlates of the mind has the
potential to impact medical research, including diagnosing brain disorder.

1 Introduction

The human brain is allegedly the most complex object known to mankind: it has circa
one hundred billion neurones forming a network of quadrillions of connections. The
amount of information that circulates through this network is immense.
There is a school of thought, called Dualism, which considers the mind and the
brain as separate entities (Rozemond, 1988). What is more, it has been suggested
that minds would not even need brains to exist. Although the separation between
mind and brain enjoys some currency in philosophical circles, it is generally agreed
nowadays that the mind results from the functioning of the brain.
The scientific community does not have yet a clear understanding of how brain
activity gives rise to the mind. Even though the behaviour of individual neurones is
fairly well understood nowadays, the way in which they cooperate in ensembles of

E. R. Miranda (B)
Interdisciplinary Centre for Computer Music Research (ICCMR), University of Plymouth,
Plymouth, UK
e-mail: [email protected]

© Springer Nature Switzerland AG 2022 299

E. R. Miranda (ed.), Quantum Computing in the Arts and Humanities,
https://doi.org/10.1007/978-3-030-95538-0_9
300 E. R. Miranda

millions has been difficult to unpack. Yet, this is of paramount importance to fathom
how the brain creates the mind.
The advent of increasingly sophisticated brain scanning technology has been
enabling a plethora of research activity to comprehend the neuronal correlates of
mental activities (Kraft et al., 2008; Vartanian et al., 2013). This comprehension is
paramount for philosophy, psychology, medicine and engineering. Indeed, emerging
technology that enables users to control systems with their mind banks on such
developments.
A Brain–Computer Interface, or BCI, is a piece of equipment that enables users to
control systems with their mind. It reads signals from the brain and harnesses them
for communicating with a computer; for example, to control a mechanical device,
such as a robotic arm, a musical instrument or a wheelchair (Fig. 1).

Fig. 1 A BCI harness brain signals to communicate with devices

A Quantum Computing Approach to Harnessing the Logic … 301

This chapter proposes a novel approach to study and harness neuronal correlates
of mental activity for the development of BCI systems. It introduces the notion of a
logic of the mind, where neurophysiological data are encoded as logical expressions
representing mental activity. These representations then become instructions for BCI
control. Ultimately, the logic of the mind is tool for studying the neuronal correlates
of mental activities.
Effective logical expressions are likely to be extensive and complex, involving
dozens of variables. Large expressions require considerable computational power to
be processed. This is problematic for BCI applications because brain data need to be
processed rapidly in order to execute sequences of commands on the fly.
Quantum computers hold much promise in terms of processing speed for a number
of problems that are intractable by classical computers, including those involving
evaluation of logical expressions. Hence, the rational for using quantum computers
to process the logic of the mind.
Quantum computers are fundamentally different from the typical computer as
we know it. The speed-up prophecy depends, amongst other things, on algorithms
that cleverly exploit fundamental properties of quantum physics, such as superpo-
sition, entanglement and interference. As it will be demonstrated below, the proposed
quantum algorithm to process logic of the mind expressions takes advantage of those
properties.
As a proof-of-concept, the chapter presents a system that reads the electroen-
cephalogram of a person at given lapses of time and builds logic expressions encoding
brain activity. The system generates respective quantum circuits and submits them to
a quantum computer. The computer then checks the satisfiability of the expressions.
Those expressions that are satisfied trigger commands to control a robotic arm.
As an additional example, we introduce an application of a BCI to make music. In
this case, the system converts the satisfiability results into sounds. The BCI becomes
a musical instrument controlled by the mind of the player.
The paper begins with a brief introduction to BCI and the electroencephalogram
(EEG) which is the neurophysiological signal that is normally used in BCI systems,
including the one presented in this chapter. Then, it briefly discusses how the EEG
corresponds to mental states, and presents the logic of the mind concept. Next, there
is a short introduction to quantum computing. Here, we focus on the basics deemed
necessary to understand the proposed quantum algorithm that deals with logic of
the mind expressions. Then, the paper shows how those expressions are processed
and describes the proof-of-concept BCI systems. The limitations of current quantum
computing hardware technology are also discussed.

2 Brain–Computer Interfacing

By and large, BCI research is concerned with the development of assistive technology
for persons with severe motor impairment. However, BCI technology is also being
developed for other applications, such as computer games (Hasan & Gan, 2012),
302 E. R. Miranda

biometrics (Palaniappan, 2008) and cursor control (Wilson & Palaniappan, 2011).
And this author has developed BCI for making music (Miranda, 2006; Miranda et al.,
2011).
There are different types of brain signals and respective sensors to read them.
The type of signal that is most commonly used for BCI technology is the
electroencephalogram.

2.1 The Electroencephalogram (EEG)

Neurones communicate with one another through electrical impulses. Neuronal elec-
trical activity can be recorded with electrodes placed on the scalp (Fig. 2). This
recording is called the electroencephalogram, or EEG. The EEG conveys the overall
activity of millions of neurones in the brain in terms of electric current. This is
measured as the voltage difference between two or more electrodes, one of which is
taken as a reference. See (Marcuse et al., 2015) for more details.
It is also possible to record electrical brain activity with electrodes surgically
implanted under the skull, on the surface of the cortex or deep inside the brain. This
is often called electrocorticography (ECoG) or intracranial EEG (iEEG). Whereas
implanted electrodes provide a far better signal to work with than surface ones, brain
implants are not yet routinely used for BCI systems for health and safety reasons.
Other technologies for brain scanning include functional Magnetic Resonance

Fig. 2 Cap furnished with

electrodes relay EEG data to
a computer for further
processing
A Quantum Computing Approach to Harnessing the Logic … 303

Fig. 3 Convention for

placing electrodes on the
scalp

Imaging (fMRI), near-infrared spectroscopy (NIRS) and magnetoencephalography

(MEG). However, these technologies are prohibitively expensive, less portable and/or
offer inadequate time-resolution resolution for a BCI application (McFarland &
Wolpaw, 2017).
For this project, we used a device manufactured by g.tec1 to read the EEG. It
consists of a cap furnished with electrodes and a transmitter that relays the EEG
wirelessly to a computer for further processing.
Electrodes positioning on the head may vary depending on the purpose of the
system or experiment. A commonly adopted convention is shown in Fig. 3. The
terminology for referring to the positioning of electrodes uses letters to indicate a
brain region and a number: Fp (for pre-frontal), F (for frontal), C (for central), T (for
temporal), P (for parietal) and O (for occipital). Odd numbers are for electrodes on
the left side of the head and even numbers are for those on the right side; the letter
‘z’ stands for the central region.
The EEG from scalp electrodes is a difficult signal to handle. It is filtered by the
meninges (the membranes that separate the cortex from the skull), the skull and the
skin before it reaches the electrodes. Moreover, the signal is weak: it is measured in
terms of microvolts (μV ). It needs to be amplified considerably in order to be useful
for a BCI. But amplification invariably brings spurious signals. Thus, the EEG needs
to be harnessed with signal processing and analysis methods in order to render it
suitable for a BCI system.

1 https://www.gtec.at/.
304 E. R. Miranda

Power spectrum analysis is a popular method to harness the EEG. This method
breaks the EEG signal into different frequency bands and reveals the distribution
of power between them. Power spectrum analysis is useful because it can reveal
patterns of brain activity, and a computer can be programmed to recognize and
translate them into commands for a system. Although this chapter focuses on power
spectrum analysis, it is worth noting that there are other EEG analysis methods for
detecting mental activity as well; e.g. Hjorth analysis (Oh et al., 2014).
Typically, users must learn how to voluntarily produce specific patterns of EEG
signals in order to be able to control something with a BCI. A fair amount of research
is underway to develop lexicons of detectable EEG patterns, understand what they
mean and develop methods to train users to produce them.

2.2 The Semantics of the EEG

The EEG conveys information about mental activity (Anderson & Sijercic, 1996;
Petsche & Etlinger, 1998). In medicine, the EEG is an important tool for diagnosis
of brain disorders.
There has been an increasing amount of research aimed at the identification of
EEG correlates of all sorts of mental activities (Hinterberger et al., 2014; Jeunet
et al., 2020; Nicolas-Alonso & Gomez-Gil, 2012; So et al., 2017; Yelamanchili,
2018). For instance, Giannitrapani (1985) identified EEG patterns associated with
abstract intellectual tasks, such as doing arithmetic operations mentally. Guenter and
Brumberg (2011) discovered neural correlates for the production of speech, which
can be detected in the EEG. And more recently, Daly et al. (2018) detected EEG
signatures correlated with emotions elicited while subjects were listening to music.
It is generally known that the distribution of power in the spectrum of the EEG can
indicate different states of mind. For example, a spectrum with salient low-frequency
components is often associated with a state of drowsiness, whereas a spectrum with
salient high-frequencies is associated with a state of alertness. Giannitrapani (1985)
linked the prominence of low-frequency components with a passive state of mind,
as opposed to an active state, which is characterized by high-frequency spectral
components.
Research exploring mental correlates of the EEG normally considers spectral
components up to 40 Hz (Kropotov, 2008). There are four recognized spectral
frequency bands, also known as EEG rhythms, each of which is associated with
specific mental states (Table 1).
The exact boundaries of the bands listed in Table 1 and their respective mental
states vary from one author to another.
The EEG that gives raise to mental states is highly dynamic. Figure 4 shows a
snapshot of EEG mapped onto 2D and 3D topographic representations using the
OpenVibe software (Renard et al., 2010). The line cursor on the ‘Signal display’
pane shows the precise moment of the snapshot. At this moment, prominent beta
rhythms are detected at the lateral sides of the brain.
A Quantum Computing Approach to Harnessing the Logic … 305

Table 1 Typical EEG rhythms and associated mental states. Frequency bands are in Hertz (Hz)
Bands Rhythms Mental states
f <4 Delta Sleep; can indicate of cerebral anomaly
4≤ f <8 Theta Drowsiness; hypnotic state; can indicate cerebral anomaly
8 ≤ f < 15 Alpha Relaxed, meditative, unfocused, almost drowsy state of mind
15 ≤ f < 40 Beta High arousal, active thinking, consciously focusing state of mind

Fig. 4 Snapshot of EEG activity mapped onto 2D and 3D topographical representations (The
electrodes’ labelling arrangement here is slightly different from the convention introduced in Fig. 2.)

It is common knowledge that different regions of the brain have distinct roles;
e.g. the visual cortex processes images, the auditory cortex processes sound, the
motor cortex controls our limbs and so on (Squire et al., 2008). Assorted regions
of the brain cooperate to perform mental tasks. Moreover, spectral amplitudes of
scalp EEG are constantly changing over the skull surface. Therefore, states of mind
cannot be accurately inferred simply by looking at a snapshot of the averaged EEG
from the whole set of electrodes at once. Rather, one needs to look at the time-based
interrelationships between spectral components of the EEG recorded simultaneously
at different locations on the scalp.
306 E. R. Miranda

Fig. 5 Tracking the trajectories of two EEG rhythms in time

3 An EEG-Based Logic of the Mind

Petsche and Etlinger (1998) advocate that different bandwidths of the EEG spectrum
are distinct windows to a dynamic landscape of electrical brain activity. It is suggested
that mental states are correlated with interrelations between these windows. We
follow this up by proposing a method to represent those interrelationships using
logical expressions. Hence, the notion of an EEG-based logic of the mind.
Consider a system, which extracts information from the EEG of a number of
electrodes at given lapses of time; e.g. periods lasting for 500 ms. For the sake of
clarity, let us say that the system extracts beta and alpha rhythms from the EEG signals
(Table 1). With this information, the system tracks the behaviour of these rhythms
over the scalp as time progresses. For this example, it tracks the electrodes’ locations
where beta and alpha rhythms displayed most power. This is shown in Fig. 5: beta
rhythms (denoted by the red electrodes) were detected prominently by electrode Fp2
at time t 0 . Next, at time t 1 they were detected prominently by electrode Fz and then
by electrode T4 at time t 2 . And alpha rhythms (denoted by the blue electrodes) were
detected prominently by electrodes T5, T3 and C3, respectively.
Next, at each time step t n , the EEG information is encoded as a logic expression.
The variables represent the activity of the respective electrodes.
In order to keep the example simple, let us consider only the following subset of
electrodes: {Fp2, Fz, T3, C3, T4, T5}. In this case, the electrodes that detect the most
prominent EEG signals are represented in the expression as ‘True’. All the others are
‘False’. Thus, the expressions for the time lapses in Fig. 5 could be written follows2 :

2 See (Smith, 2020) for an introduction to formal logic. The symbol ¬ is the negation operator (i.e.
NOT), and ∧ stands for the logical conjunction operator (i.e. AND). Variables corresponding to a
True electrode are in bold for clarity. These expressions could be stated in reduced form, but we
leave them expanded here for the sake of intelligibility.
A Quantum Computing Approach to Harnessing the Logic … 307

t 0 : β(Fp2 ∧ ¬Fz ∧ ¬T3 ∧ ¬C3 ∧ ¬T4 ∧ ¬T5)

∧ α(¬Fp2 ∧ ¬Fz ∧ ¬T3 ∧ ¬C3 ∧ ¬T4 ∧ T5)

t 1 : β(¬Fp2 ∧ Fz ∧ ¬T3 ∧ ¬C3 ∧ ¬T4 ∧ ¬T5)

∧ α(¬Fp2 ∧ ¬Fz ∧ T3 ∧ ¬C3 ∧ ¬T4 ∧ ¬T5)

t 2 : β(¬Fp2 ∧ ¬Fz ∧ ¬T3 ∧ ¬C3 ∧ T4 ∧ ¬T5)

∧ α(¬Fp2 ∧ ¬Fz ∧ ¬T3 ∧ C3 ∧ ¬T4 ∧ ¬T5)

The first term, inside the parenthesis on the left side of the conjunction operator,
corresponds to the beta rhythms and the second term to the alpha ones.
As a matter of fact, the expressions above could have been written with other
logical operators. For instance, t 0 could have been written with the logical disjunction
operator3 between the beta and alpha terms:

t k : β(Fp2 ∧ ¬Fz ∧ ¬T3 ∧ ¬C3 ∧ ¬T4 ∧ ¬T5)

∨ α(¬Fp2 ∧ ¬Fz ∧ ¬T3 ∧ ¬C3 ∧ ¬T4 ∧ T5)

The forms of those logical expressions depend on what they are meant to repre-
sent. They might, for example, depend on the correlated mental states which they are
supposed to stand for. For instance, it is generally agreed that increased beta activity in
the pre-frontal cortex corresponds to ‘focused attention in problem-solving’ (Ligeza
et al., 2015). The pre-frontal cortex is covered by electrodes Fp1 and Fp2, and to some
extent F7, F3, Fz, F4 and F8. Thus, the expressions for t 0 and t 1 above would corre-
spond to (or ‘satisfy’ in logic parlance) this mental state. Conversely, the expression
for t 2 would not.
The problem of establishing which values for the variables of a logical expression
can render the whole expression True, or satisfiable, is known as the Boolean satis-
fiability problem. As an illustration, consider this simple expression, with variables
A and B: (¬A ∧ B). If A = False and B = True, then this expression is True. Hence,
these values for A and B render the expression satisfiable. But if, say, B = False then
the whole expression would be False.
The ability to represent EEG correlates in terms of logical expressions is useful
for implementing BCI systems because they can be programmed to activate actions
associated to specific expressions. As the EEG is acquired and analysed, a system
would check when the information satisfies given logical expressions. Then, it would
perform the respective actions for those expressions that return True. For example:
If at t n , β(a ∧ ¬(b ∧ c ∧ d ∧ e ∧ f )) ∨ α(¬(a ∧ b ∧ c ∧ d ∧ e) ∧ f )) = True.
then { move robot arm to the left by 180°};

3 The symbol for the logical disjunction operator (i.e. OR) is ∨.

308 E. R. Miranda

If at t n , β(b ∧ ¬(a ∧ c ∧ d ∧ e ∧ f ) ∧ α(¬(a ∧ b ∧ c ∧ d ∧ f ) ∧ e)) = True.

then { move robot arm to the left by 45°};
If at t n , β((a ∨ d) ∧ ¬(b ∧ c ∧ e ∧ f )) ∧ α(¬(a ∧ b ∧ c) ∧ (d ∨ e ∨ f )) = True.
then { move robot arm to the right by 90°};
and so on.
Satisfiable complex Boolean expressions may return True for different combina-
tions of logic values. And such expressions can have a great number of variables and
a variety of satisfactory combinations. These are properties that make the logic of
the mind concept rather powerful for BCI.
In addition to building BCI systems, the proposed EEG-based logic of the
mind is potentially useful for cataloguing mental states and EEG correlates. We
envisage systems to automatically generate logical expressions while subjects are
performing specific mental tasks. The hypothesis is that such systems are able to
evolve sophisticated time-based ontologies of mental activity in unprecedented ways.
The caveat of developing a catalogue of mental states represented as logical
expressions is that Boolean satisfiability problems are very demanding in terms
of computation. Such expressions could easily grow to dozens of logical variables.
Expressions with, say, 50 variables would require as many as 250 combinations
to be evaluated. A mid-range personal computer is capable of performing circa 2
billion operations per second. Thus, such a computer would take almost an entire
week to evaluate an expression. Emerging quantum computing technology promises
considerable speed-up for tasks such as these (Ball, 2014). Hence, the rationale for
researching the potential of quantum computing for BCI.

4 A Brief Introduction to Quantum Computing and Logic

Operations

4.1 The Basics of Quantum Computing

This section introduces the basics of quantum computing and logic operations. It
focuses on the basics deemed necessary to contextualize and understand how to
program quantum computers for solving logical expressions. For more detailed expla-
nations of quantum computing, the reader is referred to (Bernhardt, 2019; Grum-
bling & Horowitz, 2019; Johnston et al., 2019; Mermin, 2007; Rieffel & Polak,
2011).
Classical computers manipulate information represented in terms of binary digits,
each of which can value 1 or 0. They work with microprocessors made up of billions
of tiny switches activated by electric signals. Values equal to 1 and 0 reflect the on
and off states of the switches, respectively.
A Quantum Computing Approach to Harnessing the Logic … 309

In contrast, a quantum computer deals with information in terms of quantum bits,

or qubits. Qubits are subject to the laws of quantum mechanics because they operate
at the subatomic level. At the subatomic level, a quantum object does not necessarily
exist in a determined state. Its state is unknown until one observes it.
A qubit is a two-level quantum system where the two basis states are usually
written as |0 and |1. In fact, a qubit can be in state |0, |1 or (unlike a classical bit)
in a state of superposition, which is a linear combination of both. Superposition is the
first fundamental property of quantum computing mentioned at the Introduction. To
a greater extent, the art of programming a quantum computer involves manipulating
qubits to perform operations while they are in such indeterminate state. This makes
quantum computing fundamentally different from digital computing.
The difficulty with building quantum processors using superconducting tech-
nology is that they need to be well isolated from the environment in order to keep the
qubits coherent to perform computations. Quantum processors rely on qubits being
kept in superposition for as long as possible. However, keeping a qubit in superpo-
sition is like balancing a tiny thin coin upright on a floating surface: any movement,
even the tiniest vibration, will cause it to fall to head or tail. And this gets harder
with groups of entangled qubits (e.g. when the state of a qubit depends on the state
of another). This fall is referred to as decoherence.
Still, total isolation is impossible, because one needs to access the qubits in order
to read information. The very act of reading qubits can change their state, because it
is an intervention from the external world.
In order to picture a qubit, imagine a transparent sphere with opposite poles.
From its centre, a vector whose length is equal to the radius of the sphere can point to
anywhere on the surface. In quantum mechanics, this sphere is known as the Bloch
sphere. And the vector is referred to as a state vector. The opposite poles of the sphere
are denoted by |0 and |1, which is the notation used to represent quantum states
(Fig. 6).
A qubit’s state vector can point at anywhere on the Bloch sphere’s surface. Math-
ematically, this is described in terms of polar coordinates using two angles, θ and
ϕ. The angle θ is the angle between the state vector and the z-axis (latitude) and the
angle ϕ describes the vector’s position in relation to the x-axis (longitude).
When a qubit is in a state of superposition of |0 and |1, the state vector could
be pointing anywhere between the two. However, we cannot really know where
exactly a state vector is pointing to until we read the qubit. In quantum computing
terminology, the act of reading a qubit is called ‘measurement’. Measuring the qubit
will make the vector point to one of the poles and return either 0 or 1 as a result.
The state vector of a qubit in superposition is described as a linear combination
of two vectors, |0 and |1, as follows:

|  = α|0  + β|1 , where|α|2 + |β|2 = 1.

The state vector | is a superposition of vectors |0 and |1 in a two-dimensional
complex space, referred to as Hilbert space, with amplitudes α and β. Consider the
squared values of α and β as probability values representing the likelihood of the
310 E. R. Miranda

Fig. 6 Bloch sphere (Source

Smite-Meister, https://com
mons.wikimedia.org/w/
index.php?curid=5829358)

measurement return 0 or 1. For instance, let us assume the following:

√
1 3
|  = α|0  + β|1 , whereα = andβ =
2 2
 2  2
In this case, α  = 0.25 and β  = 0.75. Therefore, the measurement of the
qubit has a 25% chance of returning 0 and a 75% chance of returning 1.
Quantum computers are programmed using sequences of commands, or quantum
gates, that act on qubits. For instance, the ‘not gate’, performs a rotation of 180°
around the x-axis. Hence, this gate is often called the X gate (Fig. 7). A more generic
rotational Rx(ϑ) gate is available for quantum programming, where the angle for
the rotation is specified. Therefore, Rx(180) applied to |0 or |1 is equivalent to
applying X to |0 or |1. In essence, all quantum gates perform qubit rotations, which
change the amplitude distribution of the system.
An important gate is the Hadamard gate (referred to as the H gate). This gate puts
the qubit into a superposition state consisting of an equal-weighted combination of
two opposing states: | = α|0 + β|1, where |α|2 = 0.5 and |β|2 = 0.5 (Fig. 8).
For other gates, please consult the references given earlier.
Qubits in a program typically start in ground state |0, and then a sequence of
gates are applied. Then, the qubits are read and the results are stored in standard
digital memory, which are accessible for further handling. A quantum algorithm is
often depicted as a circuit diagram of quantum gates, showing sequences of gate
operations on the qubits (Fig. 9).
The vertical z-axis of the Bloch sphere forms the so-called standard computational
basis. The x-axis forms the conjugate computational basis and y-axis for the circular
computational basis. As its name suggest, the standard basis is the most commonly
used, and it is the one adopted for the work presented in this chapter. A detailed
A Quantum Computing Approach to Harnessing the Logic … 311

Fig. 7 The X gate rotates a qubit’s state vector (pointing upwards on the figure on the left side) by
180° around the x-axis (pointing downwards on the figure on the right side)

Fig. 8 The Hadamard gate puts the qubit into a superposition state halfway two opposing poles

explanation of these bases and their significance to computation can be found in

(Bernhardt, 2019). What is important to bear in mind here is that changing the basis
on which a quantum state is expressed, corresponds to changing the measurement
performed to read the outcomes of the computations. It is important to note that
changing the basis to express a state does not change anything physical per se.
What makes quantum computation interesting are gates that operate on multiple
qubits, such as the conditional (or controlled) X gate; referred to as CX gate. The CX
312 E. R. Miranda

Fig. 9 A quantum algorithm depicted as a circuit of quantum gates. The squares with dials represent
measurements, which are saved on classical registers represented at the bottom line

gate is an entangling gate. Depending on the input control state, it can put 2 qubits
in entanglement, which is the second fundamental property of quantum computing
mentioned at the introduction. Entangled qubits can no longer be thought of as
independent units. They become one quantum entity described by a state vector of
its own right.
The CX gate applies an X gate on a qubit only if the state of another qubit is |1.
Thus, the CX gate establishes a dependency of the state of 1 qubit with the value of
another. The schematic representation of the CX gate is shown in Fig. 10. In fact,
any quantum gate can be built in controlled form. And entanglement can take place
between more than 2 qubits.
The Bloch sphere is useful for visualizing what happens with a single qubit, but it
is not suitable for multiple qubits, in particular when they are entangled. Entangled
qubits can no longer be thought of as independent units. They become one quantum
entity described by a state vector of its own right. Hence, from now on, we will have
to use mathematics to represent quantum systems.
The notation used above to represent quantum states (|, |0, |1), is called Dirac
notation. It provides an abbreviated way to represent vectors. For instance, |0 and
|1 represent the following vectors, respectively:

Fig. 10 The CX gate creates

a dependency of the state of
1 qubit with the state of
another. In this case, q1 will
be flipped only if q0 is |1
A Quantum Computing Approach to Harnessing the Logic … 313
   
1 0
|0 = and|1 =
0 1

Mathematically, quantum gates are expressed as matrices. For instance, the X gate
is represented as
 
01
X=
10

Therefore, quantum gate operations are expressed as matrix operations. Thus, the
application of an X gate to |0 is the multiplication of a matrix (gate) by a vector
(qubit state), which looks like this:
     
01 1 0
X (|0) = × = = |1
10 0 1

Quantum processing with multiple qubits is represented by means of tensor

vectors. A tensor vector is the result of the tensor product, represented by the symbol
⊗, of 2 or more vectors. A system of 2 qubits looks like this |0 ⊗ |0, but it is
normally abbreviated to |00. It is useful to look at the expanded form of the tensor
product to trace what it represents:
⎡ ⎤ ⎡⎤
1×1 1
   
1 1 ⎢ 1×0 ⎥ ⎢0⎥
|00 = |0 ⊗ |0 = ⊗ =⎢ ⎥=⎢ ⎥
0 0 ⎣ 0×1 ⎦ ⎣0⎦
0×0 0

The CX gate, for instance, is defined by the matrix:

⎡ ⎤
1 0 0 0
⎢0 1 0 0⎥
CX = ⎢
⎣0
⎥
0 0 1⎦
0 0 1 0

Thus, application of CX to |10 is represented as

⎡ ⎤ ⎡ ⎤ ⎡⎤
1 0 0 0 0 0
⎢0 1 0 0⎥ ⎢0 ⎥ ⎢0⎥
C X (|10) = ⎢
⎣0
⎥×⎢ ⎥=⎢ ⎥
0 0 1⎦ ⎣ 1 ⎦ ⎣0⎦
0 0 1 0 0 1
314 E. R. Miranda

Table 2 CX gate table,

Input Result
where q1 is flipped only if q0
is |1 |00 |00
|01 |11
|10 |10
|11 |01

The resulting vector is then abbreviated to |11 as shown below:

⎡ ⎤
0
   
⎢0⎥
⎢ ⎥ = 0 ⊗ 0 = |1 ⊗ |1 = |11
⎣0⎦ 1 1
1

Table 2 shows the resulting quantum states of CX gate operations, where the
second qubit flips only if the first qubit is |1. Note that in quantum computing,
qubit strings are usually enumerated from the right end of the string to the left: e.g.
|q2  ⊗ |q1  ⊗ |q0 . This is the norm adopted in this chapter from now on. Thus, the
‘first qubit’ is the rightmost one.
Another useful controlled gate is the multiple controlled form of the X gate, also
known as the Toffoli gate (Aharonov, 2003). An example of a 3-qubit Toffoli gate
(also known as CCX or CCNOT) in shown in Fig. 11.

Fig. 11 The 3-qubit Toffoli

gate creates a dependency of
the state of 1 qubit with the
state of 2 qubits. In this case,
q2 flips only if both, q0 and
q1 are |1

Table 3 shows resulting quantum states of the 3-qubit Toffoli gate portrayed in
Fig. 11.

4.2 Quantum Logic Operators

In digital logic, one can build any logic operator and entire Boolean expressions with
just one basic NAND operator (Akerkar and Akerkar, 2004). Likewise, quantum
A Quantum Computing Approach to Harnessing the Logic … 315

Table 3 The 3-qubit Toffoli

Input Result
gate table where q2 flips only
if both, q0 and q1 are |1 |000 |000
|001 |001
|010 |010
|011 |111
|100 |100
|101 |101
|110 |110
|111 |011

logic operators and expressions can be built using only the X gate and its controlled
forms. A few examples of such quantum logic operators are shown in Fig. 12.
In addition to the X gate, another useful basic gate with which we can build
Boolean expressions is the Rz(ϑ) gate. This gate rotates the state vector of a qubit
around the z-axis of the Bloch sphere by a given angle ϑ. In cases where the angle
ϑ is equal to 180° then the Rz(ϑ) gate is often called as the Z gate (Fig. 13).
Rotations around the z-axis represent changes in the phase of a qubit. For a qubit
in uniform superposition with an equal probability to be measured |0 or |1, the state
vector will point to the equator line. The Z gate thus reverses the phase of the qubit,
while maintaining its superposition.
Boolean gates built using controlled Z gates encode the results of operations in
the phases of the qubits. This allows for the representation of multiple outcomes in
superposition, which is something that cannot be done using digital bits. Hence, the
beauty and processing power of quantum computing.

Fig. 12 Quantum logic operators built with X gates. It is assumed that qubit c is initialized to |0
316 E. R. Miranda

Fig. 13 The Z gate rotates a qubit’s state vector by 180° around the z-axis. The opposite ends for
this state vector are notated as |+ and |−

A controlled Z gate acts only when both, the control qubit(s) and the target qubit are
pointing to |1. Therefore, the logic circuit will change the phases only of those qubits
that satisfy the operations they represent. These changes are referred to as ‘spin-
marking’ the possible outcomes. Examples of logic operators built with controlled
Z gates are shown in Fig. 14.
Notice that one cannot simply link sequences logic statements using a combination
of controlled X and Z gates in a circuit. Their ‘inputs’ and ‘outputs’ are incompatible
because they operate on different computational basis.
Bear in mind that rotations on the z-axis do not affect the amplitudes of the state
vector on the standard computational basis (z-axis of the Bloch sphere). And if
one measures a phase logic operator on the standard basis, phase information is

Fig. 14 Quantum phase logic operators using Z gates

A Quantum Computing Approach to Harnessing the Logic … 317

Fig. 15 The amplitude amplifier circuit4

lost. Hence, in order to be extracted, phase information needs to be converted into

amplitude information. This is done by means of a technique known as amplitude
amplification.
An amplitude amplifier is a device that increases the probability of revealing which
qubits have been spin-marked by the Z gates in a logical operation. The combina-
tion of spin-marking and amplitude amplification illustrates the third fundamental
property of quantum computing that makes it different from classical computing:
interference. Figure 15 shows how the circuit for amplitude amplification looks like.
In this case, the circuit is for three qubits. If more qubits are needed, then identical
gate sequences are added for each additional one.

5 Building Logic Expressions from EEG Information

In order to explain how to build logical expressions from EEG information auto-
matically, let us begin by defining an example of a scheme for representing EEG
information as logic expressions with only three clauses and three logical variables
(A, B, and C) as follows:

(Clause1 ) ∧ (Clause2 ) ∧ (Clause3 )

Each clause has two terms of the form (T er m 1 ∨ T er m 2 ). Below are examples
of logic expressions in the proposed format:

(¬A ∨ C) ∧ (¬B ∨ ¬C) ∧ (A ∨ C)

(A ∨ B) ∧ (¬B ∨ C) ∧ (A ∨ ¬C)
(A ∨ B) ∧ (¬B ∨ ¬C) ∧ (A ∨ C)

4 In the case of controlled Z gates, the black dot and the boxed circuit representation have the same
effect; sometimes only a vertical line of black dots are used.
318 E. R. Miranda

Next, consider that the clauses represent EEG information captured by electrodes
positioned at specific places on the head (Fig. 3), as shown in Table 4 and illustrated
in Fig. 16.
In this example, we are dealing with a BCI system that banks on EEG beta rhythms
in order to activate commands to control a hypothetical system.
The terms of the clauses are stipulated as follows: for each clause, the system
selects the two electrodes with the two highest EEG amplitudes. As an example, let
us pretend that at a certain moment in time the electrodes Fp1 and O1 registered
the highest amplitudes. Then, the system analyses the spectrum of the EEG from
these two electrodes and extracts the most prominent frequency component in the
spectrum, for each electrode. Let us say that the strongest component for Fp1 was
33.18 Hz and for O1 was 23.61 Hz. These correspond to terms A and C of Clause1 ;
that is, A = 33.18 Hz and C = 23.61 Hz.
Next, the system checks if the frequencies conveyed by the respective terms are
beta rhythms. If a frequency is equal to, or higher than, 15 Hz, then the respective
term is True. Otherwise, it is False. In this case A = True and C = True. Therefore,
these two terms form the clause (A ∨ C). Had Fp1 been equal to, say, 10.36 Hz, then
this term would have been (¬A ∨ C) instead.
For this example, the EEG analysis for the nine electrodes resulted in the following
expression: (A∨B)∧(¬B∨¬C)∧(A∨C). The next step is to verify if this expression
is satisfiable. In other words, the system checks if those EEG values would render

Table 4 Specific electrodes

A B C
are allocated to distinct A, B
and C variables for each Clause1 Fp1 T3 O1
clause Clause2 Fz Cz Oz
Clause3 Fp2 T4 O2

Fig. 16 Electrodes and respective logic variables allocations

A Quantum Computing Approach to Harnessing the Logic … 319

this expression True. Let us examine how the system generates a quantum circuit to
check the satisfiability of logical expressions.

6 Generating Quantum Circuits for Logic Satisfiability

The system uses X and Z gates, and their multiple-controlled forms, to build logical
expressions. The one introduced above are formed by three clauses linked by conjunc-
tion operators. And each of the three clauses are formed by terms linked by disjunction
operators.
The three disjunction clauses are built with X gates. And then, they are linked
with conjunction operators implemented with a controlled Z gate. The circuit requires
only 6 qubits: three of them (q0, q1 and q2) represent the logical variables A, B and
C. The other three (q3, q4 and q5) are ancillary qubits, which are used to represent
the results of the disjunction clauses. The full circuit is shown in Fig. 17 and the
respective Quil5 code in Code 1.

To begin with, all qubits are initialized to |0 and the ones representing the logical
variables are put in uniform superposition with the H gate.
The disjunction clauses are specified, one after the other, as shown in dashed boxes
at the top of Fig. 17. The results from each clause are held in ancillary qubits: N1 in q3,
N2 in q4 and N3 in q5. Then, the tripartite conjunction operation is implemented with

5 Quil is a quantum instruction language developed by Rigetti.

320 E. R. Miranda

Fig. 17 Quantum circuit for the expression (A ∨ B) ∧ (¬B ∨ ¬C) ∧ (A ∨ C)

a 3-qubit controlled Z gate applied to the ancillary qubits: they will be spin-marked
by the phase-logic AND operation.
Next, the three disjunction clauses need to be uncomputed in order to return
the ancillary qubits to their initial states. Uncomputation is achieved by running the
respective logic sub-circuits back to front; the ‘Inverted clauses’ in the middle section
of Fig. 17. Then, amplitude amplification is applied to the logic variables to reveal
the spin-marked qubits in terms of amplitudes. Finally, the qubits representing the
logic values are measured to yield the result.
A Quantum Computing Approach to Harnessing the Logic … 321

Fig. 18 Result yielded by running the circuit example for 5,000 shots. Binary numbers were
converted to decimals for plotting on the horizontal axis. Vertical axis are the times each of the
results were observed

The ancillary qubits are not part of the result. Therefore, they are not measured
in the end. However, they can interfere with the measurements because they are
entangled with the qubits that represent the logical variables. Hence, the ancillary
qubits must be uncomputed to make them return to their unentangled state.6
In this example, the circuit spin-marks the quantum states |001, |011 and |101.
Thus, each time the algorithm runs, it will output either 001, 011 and 101 with much
higher probability than any of the other possible values. The plot in Fig. 18 shows the
times that each of the possible outcomes were observed after running and measuring
the circuit for 5,000 times, or 5,000 ‘shots’ in quantum computing terminology, on
a Rigetti’s Quantum Virtual Machine (QVM).7 See Appendix 1 for more examples.
What does this result mean? Bearing in mind that in quantum computing qubit
strings are enumerated from the right end of the string to the left (i.e. |CBA), the
expression (A ∨ B) ∧ (¬B ∨ ¬C) ∧ (A ∨ C) is satisfied when:
Case 1, |001: A = True, B = False and C = False.
Case 2, |011: A = True, B = True and C = False.
Case 3, |101: A = True, B = False and C = True.
In other words, a BCI command associated with this expression is triggered when
significant beta rhythms are detected as follows:
(a) by electrodes Fp1 and Fp2 (Case 1)

6 Refer to (Johnston et al., 2019) for a didactic discussion on uncomputing ancillary qubits and
implications to measurement.
7 The Rigetti Quantum Virtual Machine is an implementation of the Quantum Abstract Machine

described in (Smith et al., 2017). Noise simulation models to account for the effect of quantum
hardware decoherence were not used in the examples discussed in this chapter.
322 E. R. Miranda

(b) by electrodes Fp1, Fp2 and T3 (Case 2)

(c) by electrodes Fp1, Fp2 and O2 (Case 3)
In terms of mental states, this expression encodes a state of focused attention. It is
well known that beta rhythms in the frontal regions of the brain are associated with
this mental state (Berta et al., 2013). We programmed the proof-of-concept to move
the robotic arm vertically, upwards and downwards, in response to detecting a state
of focusing attention in the brain of the user. An additional example is the case of
expression (A ∨ B) ∧ (B ∨ A) ∧ (B ∨ C), which is shown in Appendix 1. With this
expression, the robotic arm moves horizontally.
Whereas an in-depth account of mental states associated to EEG is far beyond
the scope of this chapter, the example above offers a glimpse into the potential for
neurotechnology of the proposed logical of the mind with quantum computing.
The next section describes our method to render into sound the outcomes from
running a quantum circuit to check the satisfiability of a logic expression, over a
number of shots.

7 Technical and Practical Considerations

7.1 Reducing Outliers

In the example presented in Sect. 6, the system spin-marked the quantum states |001,
|011 and |101. As shown in Fig. 18, it outputted 001, 011 and 101 significantly
more times than any of the other possible values. But there are five outliers, whose
probability of being observed were nearly 12% of the spin-marked ones. These
outliers take place due to the very nature of the amplitude amplification algorithm
(Johnston et al., 2019).
The probabilities of the outliers appearing could have been squashed further by
repeating the circuit between the initial Hadamard gates and the measurement section
a few times. What happens here is that interference can amplify the amplitudes (or
‘probabilities’) of spin-marked qubits and decrease the amplitudes of all others.
For the example above, repeating the circuit three times squashes the outliers to
negligible levels, as shown in Fig. 19. However, one must exercise caution here.
The number of repetitions needs to be specified carefully. In this particular case, if
the circuit is repeated only twice, then the probabilities of the outliers would have
increased instead. Moreover, a more important caveat is that such repetitions increase
circuit’s depth (i.e. the overall length of the circuit) considerably. On quantum
hardware, this escalates significantly the likelihood of errors due to decoherence.
The problem of decoherence poses severe limitations to the quantity of successive
gates that can be used in a circuit for a real quantum processor. The higher the number
of gates sequenced one after the other, the higher the likelihood of decoherence to
occur.
A Quantum Computing Approach to Harnessing the Logic … 323

Fig. 19 Results yielded by running three copies of the circuit in Fig. 17 for 5,000 shots on a Rigetti’s
QVM. Binary numbers were converted to decimals for plotting on the horizontal axis. Vertical axis
are the times each of the results were observed

7.2 Running on Quantum Hardware

At the time of writing, quantum processors struggle to maintain coherence for more
than a dozen successive gates involving superposition and entanglement. Currently,
coherence is assessed in terms of a few microseconds rather than seconds. However,
research is progressing fast to improve this (Cho, 2020). In addition to improving
hardware technology, there is much research activity to develop efficient quantum
error correction methods.
In order to illustrate how challenging the problem of decoherence is, let us
examine what happened when we ran Code 1 on the state of the art Rigetti’s Aspen-8
quantum chip, with no added noise correction algorithm. Aspen-8 affords three-
fold connectivity (i.e. supports 3-qubit gates) and coherence times of approximately
20 µs.
Figure 20 plots the outcomes after running the code for 2,048 shots on the quantum
chip. The results do not match the ones obtained with the QVM simulator; they are
all over the place. This is because the circuit is indeed too deep for this chip.
In truth, even the circuit for the first logic clause (A ∨ B) is problematic. The Quil
code to examine the satisfiability of this simple clause is shown in Code 2. Table 5
shows the outcomes that satisfy this clause. Thus, one can predict that each time we
run Code 2, it would output either 000, 101, 110, 111. Indeed, Fig. 21 shows the times
that each of the possible outcomes were observed after running the circuit for 2,048
shots on Rigetti’s QVM. They match the prediction perfectly. Yet, the outcomes from
running exactly the same code on Aspen-8 are not accurate (Fig. 22). Why is this so?
324 E. R. Miranda

Fig. 20 Results from running the circuit shown in Fig. 17 for 2,048 shots on a Rigetti’s Aspen-
8 quantum chip. Binary numbers were converted to decimals for plotting on the horizontal axis.
Vertical axis are the times each of the results were observed

Table 5 The outcomes that satisfy the logical clause (A ∨ B)

(A ∨ B) = |q2  B = |q1  A = |q0  Outcome Boolean
0 0 0 000 = 0 True
0 0 1 001 = 1 False
0 1 0 010 = 2 False
0 1 1 011 = 3 False
1 0 0 100 = 4 False
1 0 1 101 = 5 True
1 1 0 110 = 6 True
1 1 1 111 = 7 True
A Quantum Computing Approach to Harnessing the Logic … 325

Fig. 21 Results from running the circuit shown in Code 2 for 2,048 shots on Rigetti’s Quantum
Virtual Machine (QVM). Binary numbers were converted to decimals for plotting on the horizontal
axis. Vertical axis are the times each of the results were observed

Fig. 22 Results from running the circuit shown in Code 2 for 2048 shots for 2048 trials on Rigetti’s
Aspen-8 quantum chip. Binary numbers were converted to decimals for plotting on the horizontal
axis. Vertical axis are the times each of the results were observed
326 E. R. Miranda

Fundamentally, quantum computing programming languages are built upon a

handful of universal gates that physically act on quantum chips. Rigetti’s Aspern-
8 is a superconducting quantum chip, which can enact the following basic gates:
Rx(ϑ), Rz(ϑ), CZ and XY(ϑ).8 These are referred to as Quil’s native gates. The
measurements are done natively in the computational basis (z-axis). All other stan-
dard Quil gates (H, X, Z, Y, CNOT and so on) are built using native gates. Thus,
before compilation, a standard Quil code needs to be transpiled9 to native Quil. Code
3 shows the transpilation from Code 2.
Even though the circuit shown in Code 2 is 12 lines long and uses only eight
standard gates, its transpilation results in 39 lines of code with as many as 35 native
Quil gates. Unfortunately, this is too deep for Aspen-8. The real culprit here is the
CCNOT, Toffoli gate. This gate is notoriously expensive in native gates.
Similar tests on an IBM’s quantum computing resources can be found in
Appendix 2.

7.3 Quantum Advantage?

The expressions in Sect. 6 are limited to three logic variables. One may not even need
a computer to check their satisfiability, even less so a quantum computer. Should the
clauses have entailed a unique variable for each electrode, then the task of checking
the expressions’ satisfiability would have been harder. But still, not hard enough to
justify the need for a quantum computer.
However, if the expressions encompassed the whole set of 20 electrodes shown in
Fig. 3, each of which corresponding to a unique logic variable, then the satisfiability
problem would become considerably harder. In this case a quantum computer could
well be advantageous.

8 XY(ϑ) produces a parameterized iSWAP gate, which is not priority for discussion on this chapter.
For more details, see (Abrams et al., 2019).
9 A transpiler is a program that translates a piece of code into another at the same level of abstraction.

It is different from a compiler whose output is in a lower level of abstraction than the input.
A Quantum Computing Approach to Harnessing the Logic … 327

The amplitude amplification technique is a core component of the so-called

Grover’s algorithm. Introduced by Grover (1996), this algorithm uses amplitude
amplification to search for an element in an unstructured set of N elements. A brute-
force classic algorithm would scan all elements in the set until it finds the one that is
sought after. In the worst-case scenario, the element in question could have been the
last one to be checked, which means that the algorithm would have made N queries
to find it. Provided that a given problem can be encoded √
efficiently in terms of qubits,
Grover’s algorithm would be able find a solution with N queries. Thus, Grover’s
algorithm provides a quadratic speedup. This benchmarking also applies for logic
satisfiability problems.
Ignoring for the moment the format of the logical expressions discussed in Sect. 5,
let us suppose that a system would need to verify if the values A = True, B = True
and C = False satisfy a given logical expression. In this case, the three Boolean
variables amount to eight possible combinations for A, B and C; that is, 23 = 8.
Therefore, a brute-force algorithm would need to make up to eight checks to get
the answer. That is, the system would have to run the√algorithm up to eight times.
Conversely, Grover’s algorithm could solve this with 8 = 2.8 runs. Indeed, recall
that the convincing results shown in Fig. 19 were obtained by running three copies
of the circuit.
The difference in performance escalates significantly as the number of logic vari-
ables increases. For instance, 20 variables would require up to√220 = 1, 048, 576
checks classically, whereas Grover’s algorithms would require 220 = 1,024 runs.
Brushing aside any thorough comparison between the processing clocks of clas-
sical and quantum hardware, a quantum computer running Grover’s algorithms would
certainly outperform a classical computer running brute-force search. However, the
slightly disappointing news is that the number of qubits needed to implement large
Grover’s-like circuits is prohibitive the present times.
Nevertheless, despite the limited capacity of current quantum hardware tech-
nology, the quantum computing research community remains optimistic. The
industry is aiming at producing quantum chips with over 1,000 qubits by 2023 (Cho,
2020).

8 A Quantum BCI Musical Instrument

This section presents a practical application of the logic of the mind: a BCI sound
synthesiser.
We developed a method to render sounds with the results from checking the
satisfiability of the kinds of logical expressions introduced in Sects. 5 and 6 (Fig. 23).
The system generates sounds using an additive synthesiser. The additive synthesis
technique is informed by the theory of Fast Fourier Transform, or FFT (Muller, 2015).
It is based on the notion that sounds can be characterized as a sum of sine waves.
Additive synthesis works by deploying a number of sine wave sound generators
(e.g. digital oscillators) to produce partials, which are added up to produce the final
328 E. R. Miranda

Fig. 23 The example system architecture

result. A sine wave is characterised by a frequency value (i.e. speed of the cycle) and
an amplitude value (i.e. strength of the signal).
Our synthesiser comprises eight digital oscillators, each of which requires a
frequency value in Hertz (Hz) and an amplitude, whose value varies between 0
(silence) and 1 (loudest). The outputs are summed and a Hanning function is applied
to the result to give a smooth fade in and fade out bell shape to the sound.
In Fig. 24, individual partials are represented on the left-hand side of the figure,
where a bar on the ‘freq’ axis (frequency domain) has a certain magnitude on the
‘amp’ axis (amplitude domain). The spectrum of the resulting sound is represented
on the right-hand side; it contains eight partials. A schematic representation of the
resulting sound, depicting its bell-like shape is also shown.
For every lapse of time the system checks the satisfiability of the respective logic
expression, as discussed earlier, and uses the results to activate the oscillators of the
synthesiser. Figure 25 delineates how the qubit measurements plotted in Fig. 18 are
associated with oscillators. Notice that there are as many oscillators as the number
of different quantum states that the respective circuit can return: in this case, 23 = 8.
That is, each possible output is associated with a different partial of the resulting
sound. In additive synthesis, changes to the frequencies and amplitudes of the oscil-
lators modify the timbre of the resulting sound. A schematic representation of this
is given in Fig. 26.
Each oscillator of the synthesizer is assigned a frequency; for instance: osc 1 =
55.0 Hz, osc 2 = 164.81, osc 3 = 329.63 Hz and so on. This is fully customisable, and
the system can be set to change these frequencies algorithmically. The amplitudes are
normalized proportionally to the number of times the respective quantum state was
observed after a number of pre-specified shots. Thus, in Fig. 25, the first oscillator,
whose amplitude is controlled by the number of times the state |000 was observed,
A Quantum Computing Approach to Harnessing the Logic … 329

Fig. 24 Additive sound synthesis works by adding up a number of sine waves

will produce a partial that is much quieter than the second oscillator, whose amplitude
is given by number of times that state |001 was observed.
The duration of the time lapses is also customisable. However, it should not be
shorter than the window of time that is required to read and analyse EEG data.
Our experiments suggest that at least 1 s is needed to capture sufficient EEG data
to associate with a mental state. But in order to perceive differences clearly in the
timbre of the sounds, it is suggested to let it play for no less than 5 s. Of course, time
can vary and can be defined algorithmically should one wish to do so.
As a sound is being played, the system processes the EEG for the next time lapse,
builds and runs the circuit, and synthesizes a new sound immediately after the current
one. And the cycle continues for as long as required.
330 E. R. Miranda

Fig. 25 From qubit measurements to sound synthesis

Fig. 26 Schematic representation of a sequence of five sounds and their respective partials shown
above the time lapse line
A Quantum Computing Approach to Harnessing the Logic … 331

9 Concluding Remarks

This chapter presented an approach to interfacing the brain and a quantum computer.
Central to this approach is a method to encode brain activity as logic expressions
representing states of mind. These expressions are associated with commands to
control systems and/or machines with brain signals.
In addition to building brain–computer interfaces, the chapter put forward the
notion of a logic of the mind as a means to study neural correlates of the mind
and brain functioning, with potential benefits to medicine; e.g. for diagnosing brain
disorder. The EEG is an important physiological signal for diagnostics (Tatum IV,
2008). It is used to detect traces of epilepsy, dementia, cancer, inflammation, sleep
disorders and more. For instance, Clarke et al. (2013) reported that persistent excess
of beta rhythms at the frontal regions of the brain can indicate hyperactivity disorder.
It would be straightforward to encode this as a logic of the mind expression.
The EEG is currently the best signal for BCI systems. However, it is not necessarily
the best for research into understanding brain functioning and mental correlates. For
instance, functional magnetic resonance imaging (fMRI) measures the minuscule
variations in blood flow that occurs with brain activity. It is used to determine which
parts of the brain are more active than others when handling particular tasks. It has
higher spatial resolution than EEG has, and it can detect information deep inside the
brain.
The logic of the mind requires the processing of large Boolean expressions, which
are computationally very demanding. Even more so if we were to use fMRI scanning.
Today’s average classical computer would take hours, if not days, to evaluate a
handful of expressions. Hence, the rationale for using quantum computers.
At the time of writing, quantum computing hardware is of limited capacity for
realistic applications. But once fully developed, quantum computers promise to be
considerably faster than their classical counterparts to run certain types of algorithms,
such as the one introduced in this chapter. In the meantime, simulations on virtual
quantum machines enable research and development of quantum algorithms, which
would eventually run optimally when robust quantum hardware is available.
By way of a practical demonstration, we presented a BCI system that renders the
logic of the mind into sounds. We introduced a method to sonify quantum measure-
ments obtained from evaluating logic expressions. In this case, the states of mind
do not activate control commands. Instead, they are mapped directly onto sound.
Effectively, the system can be thought of as a novel musical instrument.
Incidentally, the sound synthesis technique introduced above is a contribution to
the field quantum computing on its own right, in the sense that it provides a method
to represent the wavefunction of a quantum system auditorily.

Acknowledgements The author would like to thank Mathew Wilson and James Hefford of the
Department of Computer Science at the University of Oxford for reviewing the quantum computing
content of this chapter. Also, many thanks to Palaniappan Ramaswamy of the School of Computing at
the University of Kent for scrutinizing the BCI content. Their meticulous comments and suggestions
contributed significantly to the clarity and rigour of this work.
332 E. R. Miranda

Appendix 1: Examples from the BCI System

Below are four examples of results taken from a run of the BCI system presented in
this chapter, using Rigetti’s QVM.
In relation to the BCI musical instrument, the different results produced variations
in the timbre of the sounds. Recordings of the respective sounds, and other examples,
are available online at SoundClick: https://soundclick.com/LogicoftheMind.

Example 1

This is an example where the logic expression is unsatisfiable (Table 6 and Fig. 27).

Table 6 The EEG analysis and generated expression for Example 1

Expression: (¬C ∨ B) ∧ (C ∨ A) ∧ (¬C ∨ B)
Clause 1: (¬C ∨ B) Clause 2: (C ∨ A) Clause 3: (¬C ∨ B)
Term 1: Term 1: Term 1:
Selected electrode = O1 Selected electrode = Oz Selected electrode = O2
Logic variable = C Logic variable = C Logic variable = C
Frequency = 13.1649 Hz Frequency = 31.1541 Hz Frequency = 8.22338 Hz
Term 2: Term 2: Term 2:
Selected electrode = T3 Selected electrode = Fz Selected electrode = T4
Logic variable = B Logic variable = A Logic variable = B
Frequency = 24.7721 Hz Frequency = 31.5992 Hz Frequency = 27.0611 Hz
A Quantum Computing Approach to Harnessing the Logic … 333

Fig. 27 Results from running the circuit for Example 1 for 1,000 times. Binary numbers were
converted to decimals for plotting on the horizontal axis. Vertical axis are the times each of the
results were observed
334 E. R. Miranda

Example 2

See Table 7, Fig. 28.

Table 7 The EEG analysis and generated expression for Example 2

Expression: (B ∨ A) ∧ (C ∨ A) ∧ ((¬A ∨ ¬C)
Clause 1: (B ∨ A) Clause 2: (C ∨ A) Clause 3: (¬A ∨ ¬C)
Term 1: Term 1: Term 1:
Selected electrode = T3 Selected electrode = Oz Selected electrode = Fp2
Logic variable = B Logic variable = C Logic variable = A
Frequency = 20.6042 Hz Frequency = 18.7471 Hz Frequency = 8.0119 Hz
Term 2: Term 2: Term 2:
Selected electrode = Fp1 Selected electrode = Fz Selected electrode = O2
Logic variable = A Logic variable = A Logic variable = C
Frequency = 21.2267 Hz Frequency = 32.5744 Hz Frequency = 10.3202 Hz

Fig. 28 Results from running the circuit for Example 2 for 1,000 times. Binary numbers were
converted to decimals for plotting on the horizontal axis. Vertical axis are the times each of the
results were observed
A Quantum Computing Approach to Harnessing the Logic … 335

Example 3

See Table 8, Fig. 29.

Table 8 The EEG analysis and generated expression for Example 3

Expression: (¬C ∨ A) ∧ (C ∨ A) ∧ (¬B ∨ ¬A)
Clause 1: (¬C ∨ A) Clause 2: (C ∨ A) Clause 3: (¬B ∨ ¬A)
Term 1: Term 1: Term 1:
Selected electrode = O1 Selected electrode = Oz Selected electrode = T4
Logic variable = C Logic variable = C Logic variable = B
Frequency = 13.7849 Hz Frequency = 17.5519 Hz Frequency = 12.6194 Hz
Term 2: Term 2: Term 2:
Selected electrode = Fp1 Selected electrode = Fz Selected electrode = Fp2
Logic variable = A Logic variable = A Logic variable = A
Frequency = 23.9491 Hz Frequency = 18.6791 Hz Frequency = 13.1322 Hz
336 E. R. Miranda

Fig. 29 Results from running the circuit for Example 3 for 1,000 times. Binary numbers were
converted to decimals for plotting on the horizontal axis. Vertical axis are the times each of the
results were observed
A Quantum Computing Approach to Harnessing the Logic … 337

Table 9 The EEG analysis

Expression: (A ∨ B) ∧ (B ∨ A) ∧ (B ∨ C)
and generated expression for
Example 4 Clause 1: (A ∨ B) Clause 2: (B ∨ A) Clause 3: (B ∨ C)
Term 1: Term 1: Term 1:
Selected electrode Selected electrode Selected electrode
= Fp1 = Cz = T4
Logic variable = A Logic variable = B Logic variable = B
Frequency = Frequency = Frequency =
15.0409 Hz 30.1681 Hz 18.4086 Hz
Term 2: Term 2: Term 2:
Selected electrode Selected electrode Selected electrode
= T3 = Fz = O2
Logic variable = B Logic variable = A Logic variable = C
Frequency = Frequency = Frequency =
18.6357 Hz 32.1824 Hz 19.1024 Hz

Example 4

See Table 9, Fig. 30.

Fig. 30 Results from running the circuit for Example 4 for 1,000 times. Binary numbers were
converted to decimals for plotting on the horizontal axis. Vertical axis are the times each of the
results were observed
338 E. R. Miranda

Appendix 2: Study Using IBM Quantum Computing

Resources

Figure 31 shows the results from running the excerpt corresponding to the logic clause
(A ∨ B) on an IBM quantum computer simulator, for 2,048 shots. The Quil code
shown in Code 2, in Sect. 7.2, was translated into the IBM’s OpenQASM language,
as shown in Code 8.

Not surprisingly, the results shown in Fig. 31 are identical to those obtained with
Rigetti’s QMV, as shown in Fig. 21.
The outcomes from running the exact OpenQASM code on IBM’s Santiago
processor (ibmq_santiago v1.1.1) is shown Fig. 32. Although there were errors,
in overall the results are comparable to those produced by the simulator. IBM’s
Santiago processor proved to be more resilient than Rigetti’s Aspen-8 in this case.
A Quantum Computing Approach to Harnessing the Logic … 339

Fig. 31 Results from running the circuit for the logic clause (A ∨ B) for 2,048 trials on IBM’s
simulator

However, it should be noted the programming for both machines were kept simple,
with no error correction and no a priori hardware calibration. Aspen-8 might have
performed just as well with appropriate calibration and noise mitigation procedures.
A detailed technical discussion on the idiosyncrasies of these machines, calibration,
error mitigation strategies and so on, is beyond the scope of this chapter.

Fig. 32 Results from running the circuit for the logic clause (A ∨ B) for 2,048 trials on IBM’s
Santiago processor
340 E. R. Miranda

In order to assess how well IBM’s hardware would fare with a deeper circuit,
we added a third logic variable and a phase-logic AND operation to implement the
expression ((A ∨ B) ∧ C), as shown in Fig. 33.
Considering that qubit strings are enumerated from the right end of the string to
the left (i.e. |CBA), the expression ((A ∨ B) ∧ C is satisfied when:
Case 1, |101: A = True, B = False and C = True.
Case 2, |110: A = False, B = True and C = True.
Case 3, |111: A = True, B = True and C = True.
Figure 34 shows the outcomes from running the circuit on the IBM quantum
computer simulator, for 2,048 shots. Not surprisingly, the measurements are fairly
accurate. Yet, the outcomes from running exactly the same code on IBM’s Santiago
processor (ibmq_santiago v1.1.1) are not accurate (Fig. 35). The transpiled code
comprised 75 gates; the circuit is too deep for this processor.

Fig. 33 Quantum circuit for the expression ((A ∨ B) ∧ C)

Fig. 34 Results from running the circuit in Fig. 33 for 2,048 trials on IBM’s simulator
A Quantum Computing Approach to Harnessing the Logic … 341

Fig. 35 Results from running the circuit in Fig. 33 for 2048 trials on IBM’s Santiago processor

References

Abrams, D. M., Didier, N., Johnson, B. R., da Silva, M. P., & Ryan, C. A. (2019). Implementation
of the XY interaction family with calibration of a single pulse. Retrieved October 24, 2020, from
https://arxiv.org/abs/1912.04424v1
Aharonov, D. (2003). A simple proof that Toffoli and Hadamard are quantum universal. https://
arxiv.org/abs/quant-ph/0301040
Akerkar, R., & Akerkar, R. (2004). Discrete mathematics. Person Education India.
Anderson, C., & Sijercic, Z. (1996). Classification of EEG signals from four subjects during
five mental tasks. In Solving Engineering Problems with Neural Networks: Proceedings of the
Conference on Engineering Applications in Neural Networks (EANN’96), London, UK.
Ball, P. (2014). Questioning quantum speed. Physics World, January: 38–41.
Bernhardt, C. (2019). Quantum computing for everyone. The MIT Press.
Berta, R., Bellotti, F., De Gloria, A., Pranantha, D., & Schatten, C. (2013). Electroencephalo-
gram and physiological signal analysis for assessing flow in games. IEEE Transactions on
Computational Intelligence and AI in Games, 5(2), 164–175.
Cho, A. (2020). IBM promises 1000-qubit quantum computer—A milestone—By 2023.
ScienceMag.org 15 Sep 2020. Retrieved October 30, 2020, from https://www.sciencemag.org/
news/2020/09/ibm-promises-1000-qubit-quantum-computer-milestone-2023
Clarke, A. R., Barry, R. J., Dupuy, F. E., McCarthy, R., Selikowitz, M., & Johnstone, S. J. (2013).
Excess beta activity in the EEG of children with attention-deficit/hyperactivity disorder: A
disorder of arousal? International Journal of Psychophysiology, 89(3), 314–319.
Daly, I., Williams, D., Malik, A., Weaver, J., Kirke, A., Hwang, F., Miranda, E. R., & Nasuto, S.
J. (2018). Personalised, multi-modal, affective state detection for hybrid brain-computer music
interfacing. IEEE Transactions on Affective Computing, 11(1), 111–124.
Giannitrapani, D. (1985). The electrophysiology of intellectual functions. Karger.
Grover, L. K. (1996). A fast quantum mechanical algorithm for database search. In Proceedings of
28th Annual ACM Symposium on the Theory of Computing. https://arxiv.org/abs/quant-ph/960
5043
342 E. R. Miranda

Grumbling, E., & Horowitz, M. (Eds.). (2019). Quantum computing: Progress and prospects.
National Academies Press. https://doi.org/10.17226/25196
Guenter, F. H.„ & Brumberg, J. S. (2011). Brain-machine interfaces for real-time speech synthesis.
In 2011 Annual International Conference of the IEEE Engineering in Medicine and Biology
Society, Boston, MA, pp. 5360–5363.
Hasan, B. A. S., & Gan, J. Q. (2012). Hangman BCI: An unsupervised adaptive self-paced brain-
computer interface for playing games. Computers in Biology and Medicine, 42(5), 598–606.
Hinterberger, T., Zlabinger, M., & Blasser, K. (2014). Neurophysiological correlates of various
mental perspectives. Frontiers in Human Neuroscience, 8(637).
Jeunet, C., Tonin, L., Albert, L., Chavarriaga, R., Bideau, B., Argelaguet, F., Millan, J. del. R.,
Lecuyer, A., & Kulpa, R. (2020). Uncovering EEG correlates of covert attention in soccer
goalkeepers: Towards innovative sport training procedures. Scientific Reports, 10(1705).
Johnston, E. R., Harrigan, N., & Gimeno-Segovia, M. (2019). Programming quantum computers.
O’Reilly Media Inc.
Kraft, E., Gulyás, B., & Pöppel, E. (2008). Neural correlates of thinking. Spriger Science & Business
Media.
Kropotov, J. (2008). Quantitative EEG, event-related potentials and neurotherapy. Academic Press.
Ligeza, T., Wyczesany, M., Tymorek, A. D., & Kaminski, M. (2015). Interactions between the
prefrontal cortex and attentional systems during volitional affective regulation: An effective
connectivity reappraisal study. Brain Topography, 29, 253–261.
Marcuse, L. V., Fields, M. C., & Yoo, J. J. (2015). Rowan’s primer of EEG (2nd ed.). Elsevier.
McFarland, D. J., & Wolpaw, J. R. (2017). EEG-based brain-computer interfaces. Current Opinion
in Biomedical Engineering, 4, 194–200.
Mermin, N. D. (2007). Quantum computer science: An introduction. Cambridge University Press.
Miranda, E. R. (2006). Brain-computer music interface for composition and performance.
International Journal on Disability and Human Development, 5(2), 119–125.
Miranda, E. R., Magee, W. L., Wilson, J. J., & Palaniappan, R. (2011). Brain-computer music
interfacing (BCMI): From basic research to the real world of special needs. Music and Medicine,
3(3), 134–140.
Muller, M. (2015). Fundamentals of music processing: Audio, analysis, algorithms, applications.
Springer.
Nicolas-Alonso, L. F., & Gomez-Gil, J. (2012). Brain computer interfaces, a review. Sensors, 12(2),
1211–1279.
Oh, S.-H., Lee, Y.-R., & Kim, H.-N. (2014). A novel EEG feature extraction method using Hjorth
parameter. International Journal of Electronics and Electrical Engineering, 2(2), 106–110.
https://doi.org/10.12720/ijeee.2.2.106-110
Palaniappan, R. (2008). Two-stage biometric authentication method using thought activity brain
waves. International Journal of Neural Systems, 18(1), 59–66.
Petsche, H., & Etlinger, S. C. (1998). EEG and thinking: Power and coherence analysis of cognitive
processes. Austrian Academy of Sciences.
Renard, Y., Lotte, F., Gibert, G., Congedo, M., Maby, E., Delannoy, V., Bertrand, O., & Lecuyer,
A. (2010). OpenViBE: An open-source software platform to design, test, and use brain-computer
interfaces in real and virtual environments. Presence, 19(1), 35–53.
Rieffel, E., & Polak, W. (2011). Quantum computing: A gentle introduction. The MIT Press.
Rozemond, M. (1988). Descartes’s dualism (p. 1998). Harvard University Press.
Smith, P. (2020). An introduction to formal logic. Logic matters, 2nd edn. Retrieved October 10,
2020, from https://www.logicmatters.net/ifl/
Smith, R., Curtis, M. J., & Zeng, W. (2017). A practical quantum instruction set architecture.
Retrieved October 13, 2020, from https://arxiv.org/abs/1608.03355v2
So, W. K. Y., Wong, S. W. H., Mak, J. N., & Chan, R. H. M. (2017). An evaluation of mental
workload with frontal EEG. PLoS One, 12(4), e0174949.
Squire, L., Berg, D., Bloom, F., du Lac, S., Ghosh, A., & Spitzer, N. (Eds.). (2008). Fundamental
neuroscience (3rd ed.). Academic Press.
A Quantum Computing Approach to Harnessing the Logic … 343

Vartanian, O., Bristol, A. S., & Kaufman, J. C. (Eds.). (2013). Neuroscience of creativity. The MIT
Press.
Wilson, J. J., & Palaniappan, R. (2011). Analogue mouse pointer control via an online steady state
visual evoked potential (SSVEP) brain–computer interface. Journal of Neural Engineering, 8(2),
025026.
Yelamanchili, T. (2018). Neural correlates of flow, boredom, and anxiety in gaming: An elec-
troencephalogram study. Missouri University of Science and Technology, Masters Theses 7812.
Retrieved October 10, 2020, from https://scholarsmine.mst.edu/masters_theses/7812
The History of Games for (Quantum)
Computers

James R. Wootton

Abstract Computer games are not just one application of computers, they are a
multitude. A wide variety of computational tasks are combined, all running as fast
as possible, to deliver the best possible experience to the player. It is reasonable to
expect that, somewhere in this maelstrom of computation, we can find something
that quantum computers will excel at. If the quest to combine games and quantum
computers was a game itself, it would be one of open-world exploration. Though
there may be no well-defined, concrete goals to achieve, there is something that
will guide us: this game is a sequel. By looking at how we combined games with
computers the first time, we can get some idea of what we can expect in this quantum
successor.

Keywords Quantum computing · Computer games · Quantum speedup ·

Procedural generation · Terrain generation

1 Introduction

Computer games are not just one application of computers, they are a multitude. A
wide variety of computational tasks are combined, all running as fast as possible,
to deliver the best possible experience to the player. It is reasonable to expect that,
somewhere in this maelstrom of computation, we can find something that quantum
computers will excel at.
With this we begin the adventure! If the quest to combine games and quantum
computers was a game itself, it would be one of open-world exploration. Though
there may be no well-defined, concrete goals to achieve, there is something that
will guide us: this game is a sequel! By looking at how we combined games with
computers the first time, we can get some idea of what we can expect in this quantum
successor.

J. R. Wootton (B)
IBM Quantum, IBM Research – Zurich, Zurich, Switzerland
e-mail: [email protected]

© Springer Nature Switzerland AG 2022 345

E. R. Miranda (ed.), Quantum Computing in the Arts and Humanities,
https://doi.org/10.1007/978-3-030-95538-0_10
346 J. R. Wootton

2 The 1950s: What Can Games Do for Computers?

Since the early history of computing, people have been interested in how it relates to
games. This has come in multiple forms: Can computers implement games? Can they
play games, and perhaps even beat humans? Could they offer experiences that would
be impossible without them? We now know that the answer to all these questions is
‘yes’ since the last few decades have conclusively shown what computers can do for
games.
Back in the 1950s, however, the combination of games and computers got off to a
slow start. Early examples were less about what computers could do for games and
more about what games could do for computers.
One of the first examples was Bertie the Brain exhibited at the Canadian National
Exhibition in 1950. It used vacuum tubes and light bulbs and played Tic-tac-toe. Not
because it offered a better play experience than using pen and paper, but because
it helped to show off the vacuum tubes. The fact that we are writing about it now,
over 70 years later, shows that it certainly had some impact. It showed us the first
possible reason for making games with computers: to showcase new technology.
Unfortunately, in this case, however, the specific technology it was showcasing soon
became obsolete.
Then, there was Nimrod in 1951. Again, vacuum tubes and light bulbs, but this time
the game was Nim. It was shown at the Festival of Britain with a stated aim, among
other things, to ‘…illustrate the algorithm and programming principles involved’.
This gave us a second reason to make games with computers: for education. Program-
ming would have seemed a strange and arcane art in those days. How better to make
it relatable than to show it playing a game?
Next was a research project in 1952 at Cambridge University, now commonly
referred to as OXO. The idea was to study human–computer interaction, done using
another implementation of tic-tac-toe (or ‘noughts and crosses’ as it is known in
Cambridge). It again used vacuum tubes, but this time had cathode ray displays: the
first graphical upgrade of the games industry. Nevertheless, the driving principle was
research, giving us another reason to make games with computers.
Another project to note was that done at IBM Research in the 1950s: making
artificial intelligence (AI) to play Checkers and beat a human opponent (Samuel,
1959). This would be something that IBM would repeat decades later, using Deep
Blue to beat Gary Kasparov at chess, and then Watson to beat champions at Jeopardy.
DeepMind has also been prominent in this area, using AlphaGo to beat a champion
at Go. All of which are more examples of using games to showcase technology.
There were more games in the 1950s, but the ones here show what the decade was
all about. Whether it was to showcase the technology, to educate about the technology
or to research the technology, games were used to help computers rather than the
other way around. An exception was Tennis for Two, a game made in 1958 for fun;
and it signalled what was coming.
The History of Games for (Quantum) Computers 347

3 The 1960s: What Can Computers Do for Games?

In 1962, a team at MIT decided to put their new PDP-1 computer to good use. They
embarked on a project that would:
• Test out the new device (and later, new installations).
• Showcase its capabilities.
• Be fun!
They created a game. But it wasn’t just an existing game, with a computer-based
implementation tacked on as a gimmick. It was something new: space battles without
the need to build your own spaceship. It was called Spacewar.
This was the first clear example of computers doing something for games. They
were used to make something new, and something fun. Functions made for trigono-
metric calculations were repurposed to let the player experience flying around a
star.
It was not the only game this decade to be set in space. There was also the original
Lunar Lander as well as the unimaginatively named Space Travel. More down to
Earth was an economics simulator, The Sumerian Game, which was an early ancestor
of games like SimCity.
In each case, the computer was used to do exactly what computers were made
for: crunching numbers and performing simulations. But rather than being used for
payroll calculations or planning a real trip to the moon, the computers were used to
provide unique play experiences.

4 The 1970s: Commercial Success

Not many people know about Spacewar! or Space Travel. Any familiarity with Lunar
Lander is usually due to one of its later incarnations, rather than the 1960s text-based
original. This was not due to any secrecy on the part of the developers: they usually
freely shared the source code. It was more because few had the room for a room-sized
computer, never mind the money.
The 1970s is when it became possible to bring the hardware to the masses. The first
arcade game was Computer Space, conceived of as a coin-operated Spacewar! that
ran on hardware dedicated only to the game. The first generation of home consoles
soon followed. With these breakthroughs, computer games became something you
could sell to the masses, either in an arcade or at home. The games industry had truly
begun.
Something else also started in the 1970s, when Paul Benioff began to think
about computers that operate according to the principles of quantum mechanics
(Benioff, 1980). It has taken a long time since then to fully formulate the idea,
explore applications and develop the required hardware. But we are finally close to
fruition.
348 J. R. Wootton

5 The 2010s: What Can Games Do for Quantum

Computers?

By 2010 we already had many examples of algorithms that could run on quantum
computers, to solve problems that would be intractable for even the biggest supercom-
puter; see here for a few examples (Jordan, 2021). However, the hardware required
to actually run these algorithms was far beyond what was available. Instead of the
many thousands of high-quality qubits that are needed, just having two qubits was
pretty impressive in 2010. This hardware was also not easy to use, being hidden away
in labs around the world and programmed primarily through the manual labour of
research students.
The state of affairs did not seem much better in 2015 when this author, a
theoretical physicist working on quantum error correction, proposed a quantum
computing experiment for 5-qubits (Wootton, 2016). Proposing an experiment was
easy. Convincing a research group with the required hardware to actually run it would
be much harder. And running it would be harder still.
This all changed in 2016 when IBM released their Quantum Experience: A web
interface allowing direct access to a 5-qubit device (Quantum Composer and Lab).
On one fateful day, just before lunchtime, this author suddenly realized that this
provided exactly what was required for his proposed experiment. By the time he left
the work, the experiment had been run. A revolution had begun!
Since then, a great many others like this author have used direct and simple access
to quantum hardware as part of their scientific research. But that is beyond the scope
of this chapter. With access to new computational hardware, there are also other
things we can do. It took us until early 2017 to realize that this included making
games.
Most probably the first-ever game using a quantum computer is a simple version of
Battleships (Wootton, 2017). The display was just ASCII in a terminal. The gameplay
was no better than you could do with pen and paper. Rather than being made to be
played, it was made to be the basis of a blog post.
The idea was to give an example of quantum programming that wasn’t a complex
algorithm, but instead something more tangible. The states of a qubit were used to
encode whether a ship had been sunk or not. Quantum gates were used to imple-
ment attacks. Without realizing it at the time, the author repeated the history of
Nimrod, mentioned above. Just as normal programming seemed like an arcane art
then, quantum computing can seem so now. So the game was designed to ‘illustrate
the algorithm and programming principles involved’. It was a quantum game made
for the purpose of education.
The next major quantum game was Quantum Awesomeness (Wootton, 2018).
This was based on the fact that there are many aspects of a quantum processor that
determine how useful it is: number of qubits, which pairs of qubits we are allowed to
apply a two-qubit manipulation to, and what imperfections are present. Thus, a puzzle
game was designed to help people get an idea of how different devices compare.
The History of Games for (Quantum) Computers 349

Fig. 1 An example of Quantum Awesomeness

The game was designed so that the puzzles were based on the specifications of
the device used to run it. The more qubits you had, and the more complex the layout
of possible two-qubit gates, the better the puzzles would be. The more imperfections
there were, the less it would seem to make any sense. Therefore, to compare devices
and see which was best, one only needed to see which was the most fun. It is not as
informative as looking at benchmarking data, but it is a lot more accessible to the
public.
An example can be seen in Fig. 1. The coloured circles represent qubits on the
device. The connections between them show us which pairs of qubits may be directly
manipulated through two-qubit gates. And hence the device is itself the puzzle board.
The principle behind the puzzle is simple: a random pairing of the qubits is chosen,
and the player has to work out what the pairing is. This is implemented by running a
quantum program whose details are defined by the chosen pairing. From the results,
a number is calculated for each qubit. This number should be exactly equal for the
two qubits of each pair, and so the player’s job should be a simple one: they just need
to find the pairs of equal numbers.
For an example such as the one above, finding this pairing is indeed a very easy job;
it is the pairs labelled H, I, J, K, L, M G and V if the reader is wondering. However,
these results come from a simulation of a quantum computer, done without any of the
imperfections inherent in real quantum hardware. An example from real hardware
(with a different pairing) can be seen in Fig. 2.
In the case shown in Fig. 2, the numbers that should be equal are not as equal
as they should be. We see that some pairs of qubits are behaving fairly well. Others
qubits, like the one whose number is 14, seem to be giving us total nonsense. This,
therefore, allows us to get an idea of how well the qubits work on this device, and
also allows us to compare qubits between devices.
This game was designed so that the player didn’t need to have direct access
to the devices. Instead, puzzles could be made once and played later. This is also

Fig. 2 Running Quantum Awesomeness on quantum hardware

350 J. R. Wootton

demonstrated in the image above, where you can still play the game despite the
device it ran on having been decommissioned some time ago. This was an example
of making quantum games for the purpose of research, repeating the history of the
1950s.
The next game to use a real quantum device was Q|Cards, made in 2019 by
a team working at Quantum Wheel, the Fifth Game Jam held in Helsinki, Finland
(Cards). This is a card game in which each player has a qubit and each card represents
a quantum gate. The quantum computer is used at the very end to judge who played
their gates best.
It is a fun game to play, but it also helps to teach people about quantum gates.
Instead of getting them to sit through a lecture or read through a textbook, playing
the game helps us explain the basics of quantum computing in a friendly atmosphere.
Another quantum game made for education.
The motivation behind these games was exactly the same as in the 1950s: education
and research. They also served the other big purpose of 1950s games, to showcase
the new technology, since their very existence helped to get the word out in the form
of blog posts, tweets and more.
The short summaries of the above-mentioned games are intended simply to give
a flavour of the quantum games of the 2010s. We will look at one of these examples
in detail later in this chapter.

6 Quantum Games Without Quantum Computers

If the history of computer games began in the 1950s, then the 1940s was its prehistory.
One relic of this era was Turochamp, a chess AI developed by Alan Turing and David
Champernowne (Copeland, 2004). It had one big problem: it was too complex for
the computer hardware of the time. It was run in 1948, but with a human manually
simulating the processes. This too has a parallel in recent experiments with quantum
games. It is easier to simulate simple quantum programs on a normal computer than
to run them on real quantum hardware. This allows us greater flexibility to explore
what qubits can do, within the familiar and fun context of hacking together simple
games.
Many games have been made in this way; for instance, QPong (a quantum version
of Pong) (QPong), QiskitBlocks (building quantum programs in a Minecraft-like
world) (QiskitBlocks), to cite but two.
One example of our own was not so much a game, but a method for procedural
generation. It used just a single qubit (the smallest element of quantum computation)
to create a landscape to walk around. The generated terrain was far simpler than can
easily be created using non-quantum methods, but that was beside the point. The
project was intended to provide a hands-on means by which people could learn all
the intricacies of single-qubit manipulations, by creating their own terrain generator.
This project will be explained in detail in Sect. 9.
The History of Games for (Quantum) Computers 351

7 The 2020s: What Can Quantum Computers Do

for Games?

What can quantum computers actually do for games? Like everything, they bring
some benefits but also some design constraints. For one thing, the need to cool the
hardware to near absolute zero means that you’ll never get one inside a home console.
They’ll live on the cloud, and so probably aren’t something that you want to send jobs
to every frame. For any game that might use a quantum computer, the vast majority
of it will still run on a normal computer. We just need to find a small corner that
quantum computers can excel at.
Though speeding up computation is the big promise of quantum computers, the
aim is mostly to take problems that could never otherwise be tackled within a human
lifetime and instead solve them within a reasonable timescale. Tasks that can already
be done very quickly with conventional hardware, will continue to be done with
conventional hardware, with no hope of a quantum speedup.
The most reasonable points in a game to insert quantum results are those where
responses are not expected in less than a second. This includes tasks such as setting
up the levels, puzzles or worlds that the player will experience. These jobs can be
done during a loading screen, or even during the game design process, ensuring that
the quantum computer has time to do its thing. The future of quantum computation
in games, therefore, looks likely to be dominated by procedural generation.
As with any attempt to find applications for quantum computers, we also need to
think about the different eras that the technology will go through. Our main aim is
a future era, where all imperfections are removed by error correction, where you’ll
have more qubits than you’ll ever need, and where you’ll be able to do any two-qubit
gates you want. This is the era of fault-tolerant, scalable quantum computing. With
this hardware, we will be able to implement the many algorithms developed over the
last few decades. By looking at the kinds of problems that they can solve, we can
begin to speculate on how they will be used in procedural generation.
Certain possibilities immediately jump out. Quantum computation has been
shown to offer more efficient computational complexity for constraint satisfiability
problems, which could be useful in searching the space of different forms of content
for those that satisfy desired properties. Also, since many problems in procedural
generation can be expressed in the abstract language of nodes connected by a network,
quantum speed-ups for the analysis of such networks could offer additional tools.
These are motivating examples, but they are admittedly not defined very concretely.
This is because quantum computation is a field that has mostly concerned itself with
asymptotic properties of a computation’s runtime: When thinking about how this
time grows with the size of a problem, we know much about what shape the graph
would be but little about the exact values. Games, on the other hand, are an applica-
tion that depends greatly on the exact speed at which calculations can be performed.
A factor of 10 in speed can be the difference between something being useful or
useless in a game.
352 J. R. Wootton

The exact runtime of quantum algorithms cannot currently be accurately predicted,

nor can the exact size of problems for which conventional hardware must give way
to a quantum approach. These quantities will depend on exactly how fault-tolerant
quantum computers finally come to fruition, which is still an area of active research.
Forecasting exactly how these devices will be useful for procedural generation in
games is therefore not possible.
However, quantum computers do exist today in non-fault-tolerant form. They are
already pushing the boundary of what can be reproduced with conventional hardware.
Though the ‘textbook’ algorithms might not apply, we can seek to create tailor-made
methods for the device we are using and the kind of results we want to get.
Over the next few years, we can seek to discover our own algorithms, to make
interesting and useful content for procedural generation. This may not be able to rely
on solving complex constraint satisfiability problems and such like, but the devices
are already able to generate interesting content. This author has been using them
for purposes exactly like these over the past few years, both for games in ‘game
jams’ and papers for research conferences (Wootton 2020a; Wootton 2020b). As the
hardware gets ever more sophisticated, so too will the possibilities for what we can
do with it. In this way, we can hope to reach some kind of quantum Spacewar!, the
first example of a unique play experience made possible with the new technology.

8 Battleships with Partial NOT Gates

As discussed in the last section, one reason we might want to combine games with
quantum computers is for education. The most obvious method is to make a game
such that players will learn about quantum computing through play. But people also
enjoy making games, allowing us to also explore a very different possibility: learning
about quantum computing by using it to make a game.
This was the idea behind the pioneering Battleships with partial NOT gates
(Wootton, 2017). It was not made to offer a good player experience but instead
was made to form the basis of a blog post. It was used as a simple and relatable
example of quantum computers doing something, which would be less imposing
than the quantum algorithms of textbooks. It was the Nimrod of quantum games, as
mentioned earlier, built to ‘…illustrate the algorithm and programming principles
involved’.
The game itself is a variant of the abovementioned Battleships. The central idea is
that there are ships that get attacked, and after a given number of hits they will sink.
Everything else, such as inputs, outputs and the handling of turns, is just bookkeeping.
Like any quantum algorithm, any quantum game will require a hybrid approach.
Conventional digital computers will be used for some tasks, and quantum computers
will be used for others. In this case we will use quantum computation for the central
game mechanic. Thus, it will keep track of how damaged each ship is and implement
attacks. Before we do this, let us start with an explanation of what a bit is. Even
The History of Games for (Quantum) Computers 353

though they have become a ubiquitous concept in our increasingly digital world,
they are not something that people spend much time thinking about.
A bit is simply the smallest possible chunk of information. It can take one of two
possible values, which we typically call 0 and 1. We can use bits to store binary
values. For example, if we are recording whether a ship has been destroyed or not,
we could use 1 for it being destroyed and 0 for it being fully intact. To read out the
information, we simply look at the value that was recorded.
The qubits at the heart of quantum computing are the quantum version of a bit.
They too can only give outputs of 0 or 1. So we will use exactly the encoding described
above, but with a qubit instead of a bit: The qubit state is certain to output 0 to encode
the state of a ship that is perfectly undamaged, and that for 1 for a ship that has been
destroyed.
A program for qubits takes the form of a so-called ‘quantum circuit’. These are
like the Boolean circuits with which conventional digital computing can be expressed
when considering very low levels of abstraction. We will use the Qiskit framework
(Qiskit) created by IBM Quantum and built in Python, to write these circuits. In this
Battleships game, each player has five ships, so we begin by creating a circuit for
five qubits:
from qiskit import QuantumCircuit
qc = QuantumCircuit(5)

On its own, we cannot do much with this circuit. We cannot even extract an output
of 0 or 1 from the qubits. For that we would need a circuit which does not just have
a qubit but has a bit on which an output can be written. So, let us modify our circuit
to not only have 5 qubits but also 5 bits to serve as their outputs.
qc = QuantumCircuit(5,5)

To actually extract these outputs, we need to use a so-called measure gate, which
specifies exactly when the process of extracting the output should be performed. For
convenience we will keep this in a separate circuit, defined for 5 qubits (and bits) as
follows.
m_z = QuantumCircuit(5,5)
for j in range(5):
m_z.measure(j,j)

Note that qubits and bits are indexed from 0 in Qiskit. So, the five qubits and
bits are referred to as 0, 1, 2, 3 and 4. The measure(j,j) command is telling the
circuit to extract an output from the jth qubit and write the answer to the jth bit.
After creating a circuit, the final step is to run it. For complex circuits we have
no choice but to use actual quantum hardware for this. For 5 qubits, however, it is
possible to simulate using a conventional digital computer:
from qiskit import Aer
backend = Aer.get_backend(’aer_simulator’)
job = backend.run(qc.compose(m_z), shots=10, memory=True)
result = job.result()
354 J. R. Wootton

outputs = result.get_memory()

Let us unpack what is happening here. Firstly, the circuit we ran was specifically
the combination of qc and m_z: our original five qubits with all the measurements
placed at the end. This combined circuit is created in Qiskit with the command
qc.compose(m_z). The circuit was run on a so-called ‘backend’. In this case the
simulator known as the ‘Aer simulator’ was used. Using a real quantum device is as
easy as changing the backend object to one that refers to a real quantum device so
that the ‘run‘ function sends the job to that device over the cloud.
The process is repeated for ten shots, so we can see if there is any randomness in
the results. The output of each process will be a 5-bit string, given that our circuit
has five output bits. Since we used shots = 10 to ask for ten runs of the same
process, the results are returned as a list of the ten outputs. Specifically, we will get
the following list:
[‘00000’, ‘00000’, ‘00000’, ‘00000’, ‘00000’, ‘00000’, ‘00000’,
‘00000’, ‘00000’, ‘00000’]

Each output is ’00,000’. This means that each of the five qubits output a 0 in
each case. This is due to the way qubits are initialized: they always start out in the
state that is certain to output a 0. Our five ships, therefore, start off fully undamaged.
To implement an attack on a ship we need to apply an operation to its qubits, known
as a gate. The simplest example would be a completely fatal attack, flipping the ship
directly from undamaged to destroyed. The corresponding qubit must therefore be
flipped from the state certain to output 0 to the one certain to output 1. This effect can
be achieved by two quantum gates, known as X and Y gates. Both can be regarded
as quantum equivalents of the NOT gate for standard bits, which flips between the 0
and 1 states. In the following, we will focus on the X gate to implement the attack.
So, let us get on with it, and launch our most destructive torpedo at the first ship,
which is the one corresponding to the qubit labelled 0:
qc.x(0)

This adds the new operation to the circuit qc: an X gate on qubit 0. If we run the
combined circuit now, we will get the result ’00,001’. Here the result for the first
ship appears at the first bit on the right. It is a 1, denoting that the first ship has been
destroyed. Therefore, our X gate had the desired effect.
Now let us load up a new torpedo and take aim for the second ship. But let us
suppose that this ship is made of sturdier stuff, so one torpedo will not be enough to
sink it. Instead, it will only half destroy its target. In order to implement this attack,
we would like to do something like ‘half of an X gate’. This means we need to find
a gate which, when applied twice, leads to an end result that is exactly the same as
an X gate.
If we were using a conventional digital computer, with a conventional bit repre-
senting each ship and a NOT gate to implement attacks, this would not be possible.
Either you flip the bit value, or you do nothing. There is no in-between. More bits
The History of Games for (Quantum) Computers 355

and different types of gate would be required to implement the possibility of a half-
damaged ship. Of course, using more conventional bits is exactly the best course of
action in this case, but it is not what our game is about. We will doggedly stick with
just the one bit for each ship, and happily pay the price of making it quantum.
With quantum bits we can access so-called ‘superposition states’, which are not
quite 0 and not quite 1. That is, when we measure them, the results will appear random,
making it easy to think that they are simply the result of an overly engineered coin
flip. However, that is not the case. These states are just as definite and well-defined
as the ones whose outputs are certain. We could define measurements for which they
would give certain outputs instead of random ones, as we will see in the next section.
But when we ask the question of whether they are 0 or 1, they do not fit neatly into
either box. They cannot use a third option to declare that they are neither: then they
would no longer be bits since bits are only allowed two options by definition. Thus,
they simply answer randomly. With that in mind, let us implement the following
attack on the second ship (the one whose qubit is labelled 1):
from math import pi
qc.rx(pi/2, 1)

Instead of just an X gate, this is an RX gate. This gate requires two arguments: as
before, we specify which qubit it acts on, but we also supply the value of pi/2. If we
run the combined circuit now, we will find that the result is randomly either ’00001’
or ’00011’. In both cases, the ships that we have not attacked remain undamaged
(represented by their 0 s) and the first ship remains destroyed (represented by its 1).
The second ship, however, which is represented by the second qubit on the right, is
randomly either 0 or 1. By taking multiple samples we can determine that these two
possibilities occur with equal probability.
In order to see that this is not just a fancy coin flip, let us implement the attack
again with another qc.rx(pi/2, 1). The entire quantum circuit qc is now:
qc = QuantumCircuit(5) # attack on the first ship
qc.x(0) # attack on the second ship
qc.rx(pi/2, 1)
qc.rx(pi/2, 1)

Running the combined circuit now yields the result ’00011’ with certainty.
The first RX gate took its qubit from the state certain to output 0 to a superposition
state between 0 and 1. The second RX gate completed the journey, taking the qubit
to the state certain to output 1. Together they had exactly the same effect as a single
X gate.
Now let us attack the next ship (the one whose qubit is labelled 2). This is even
sturdier than the last and will require three torpedoes to sink. The required effect can
be implemented with rx(pi/3, 2). Applying this gate once will give random
results, but with a bias towards 0. For two applications the bias will instead favour
1. For three we get 1 with certainty, and the ship is destroyed.
Similarly, we can use pi/4 for a gate that must be repeated four times to form an X
gate, and so on. This argument can be thought of as representing an angle, as we will
356 J. R. Wootton

see in the next section. The angle pi is required for an X gate (and so rx(pi, 0)
would be another way to express the X gate with which we attacked the first ship).
For multiple sequential applications of RX gates on the same qubit, these angles
effectively add together.
This is basically all there is to the quantum machinery behind Battleships with
partial NOT gates. It does not show us how or why one might want to build qubits, or
why the ability to do partial forms of the gates used in conventional digital computers
will confer any computational advantage. The aim that it does achieve, however, is
to show them in action. And to show that quantum computers can do a relatable job
without the need for an abstract algorithmic approach.

9 Single-Qubit Procedural Generation

Now that the qubit is no longer completely unfamiliar, let us get to know it better.
This can again be done in the context of using it to implement something that can be
used in a game.

9.1 The Basic Properties of Qubits

One fundamental property of quantum objects is that they must be described by

multiple incompatible attributes. This means that there are attributes of the object
that cannot simultaneously be well-defined: If one is defined with absolute certainty,
the others must be completely ambiguous.
This is not the easiest concept to grasp, since it is not a type of behaviour that we
recognize from the large, non-quantum objects of our everyday life. If we know the
size of an apple, for example, its colour does not suddenly become undefined. Never-
theless, such things are true of quantum objects. Qubits have been specifically created
to be the easiest quantum object for us to access and manipulate. So, by playing with
them, we can start to build up some intuition about this quantum behaviour.
We now know enough to write down some rules that any quantum bit must obey.
1. As a type of bit, it can only give us values of 0 and 1.
2. As a quantum object, it has multiple incompatible attributes.
3. As a type of bit, it can store no more than a single binary value.
Put 1 and 2 together, and we find that a qubit must have multiple attributes that
can be measured, but all must give values of only 0 or 1. Put 2 and 3 together, and
we find that only one of these attributes can be useful to store a binary value at any
one time. This is because assigning a definite value for one of the attributes forces
the others to be completely random.
The History of Games for (Quantum) Computers 357

9.2 Properties of an Initialized Qubit

As in the last section, we will be considering the concept of quantum circuits. We will
also repeat the trick of having a circuit called qc, which contains all the manipulations
we want to do before extracting an output:
qc = QuantumCircuit(1,1)

Next, we will use an additional circuit to extract the output, via the process known
as ‘measurement’. The standard form of measurement, as we used in the previous
section, is implemented simply using the measure gate in Qiskit:
m_z = QuantumCircuit(1, 1)
m_z = m_z.measure(0, 0)

Whenever one simply refers to ‘measurement’ in quantum circuits, this is usually

what is meant. However, to be more specific, it is known as the ‘z measurement’.
The circuit qc defined above is an empty circuit. Since nothing is done to it, it is
in the standard state for a freshly initialized qubit. In order to finding out how this
behaves, we need to extract an output by combining qc with the measurement circuit
and running it:
backend = Aer.get_backend(’aer_simulator’)
job = backend.run(qc.compose(m_z), shots=10, memory=True)
result = job.result()
outputs = result.get_memory()

As we might expect from the previous section, the result is 0 with certainty:
[‘0’, ‘0’, ‘0’, ‘0’, ‘0’, ‘0’, ‘0’, ‘0’, ‘0’, ‘0’]

This is the behaviour we can expect from a freshly initialized qubit whenever it is
measured using m_z. However, this is not the only way to measure. Another method
is the so-called ‘x measurement’. This is done with the following circuit, m_x:
m_x = QuantumCircuit(1, 1)
m_x.h(0)
m_x.measure(0, 0)

As you might have noticed, this requires an extra element compared to m_z. This
is because, at the time of writing, Qiskit doesn’t allow us to ask for the x measurement
directly, and so we need to use a little hack.
By running the combination of qc and m_x we will get a random mixture of 0s
and 1s as our results. To see the probability for each possible output, we need to
extract many results. For this it is more convenient to get the results in a less verbose
form, which simply lists each type of result along with the number of samples for
which it occurred. The syntax for this is as follows, for 1000 samples:
backend.run(qc.compose(m_x), shots=1000).result().get_counts()
358 J. R. Wootton

This gets the so-called ‘counts’ dictionary. Here’s an example of the kind of thing
you’ll see:
{‘1’: 499, ‘0’: 501}

Here the two possible results, 0 and 1, both occur roughly 500 times out of the 1000
total samples. From this we can conclude that the results are completely random,
giving 0 or 1 with 50/50 probability.
There are actually an infinite number of ways to extract an output from a qubit. But
we can create a complete description by using just three. The x and z measurements
are two of them. The final one is known as the ‘y measurement’:
m_y = QuantumCircuit(1, 1)
m_y.rx(pi/2, 0)
m_y.measure(0, 0)

By running the combination of qc and m_y we will again get a random mixture
of 0 s and 1 s as our results. With more samples, we can confirm that it is again
50/50. With statistics on the results of these three types of measurement, we can
fully characterize a single qubit state.
Here we characterized the state that a qubit is in when initialized at the beginning
of a circuit. The result was that such a qubit is certain to give an output of 0 for m_z,
but gives random results for the other two.
Now we can investigate the properties of other possible qubit states. We will do
this by applying all of the possible single-qubit operations to a qubit.

9.3 The X Gate

Let us examine the effects of the quantum operator known as the X gate:
qc = QuantumCircuit(1)
qc.x(0)

Since we are now well acquainted with all the important ways to get an output,
we can run them all for this new qc and see what happens. We will find that the x
and y measurements still give random results, but the z measurement is now certain
to output a 1. The X gate simply flips the bit value that is output for m_z. It serves
as the quantum form of a standard NOT logic operator, as discussed above.
If we add another X gate to the circuit, we will see the flip effect again: the output
for m_z will go back to being 0.

9.4 The H and RX Gates

Next, let us examine the behaviour of the H gate:

The History of Games for (Quantum) Computers 359

qc = QuantumCircuit(1)
qc.h(0)

If we run this with our three types of measurement, we will find that it is now
m_y and m_z that give random results, and m_x for which the outcome is certain
to be 0. This is an effect of what we might call the ‘conservation of certainty’ in
quantum systems. The H gate has unlocked the ability to store a definite bit value in
the results of m_x. However, it also takes away the ability to do so with m_z. This
prevents us from storing 2-bit values in the same qubit, and so preserves its identity
as truly being a type of bits.
A similar effect can be done using the RX gate that we saw earlier. This requires
an additional parameter: an angle expressed in radians. For example, with the angle
–pi/2:
qc = QuantumCircuit(1)
qc.rx(–pi/2, q)

The effect of this is to make it m_y that gives the output 0 with certainty, and the
other two that are random.
For both H and RX gates with this specific angle, the qubit was certain to output
0 for one of the measurements. With other gates we can make these certain to be 1
instead. The simplest way is to add an x gate at the beginning of each circuit.
With the gates we have so far, we are able to set the output of a qubit to be 0
or 1, and also to change which of the measurements see randomness and which see
certainty. These represent the most basic kinds of operations, known as the Clifford
gates.

9.5 The Bloch Sphere

The simplest way to move beyond the Clifford gates is to revisit the RX gate, and
simply use an angle that is not a multiple of pi/2. Specifically, let us use the following
code:

qc = QuantumCircuit(1).
qc.rx(–pi/4,q).

The results for this will not be as clear cut as we’ve seen before. Here is an example
of what you might see for 1000 samples for each type of measurement:

Results for an x measurement: {’1’: 488, ’0’: 512}.

Results for a y measurement: {’1’: 851, ’0’: 149}.
Results for a z measurement: {’1’: 152, ’0’: 848}.
Here we find that the results from m_x are completely random, but the other two
are not. The results from m_z have some randomness, but with a bias towards 0.
360 J. R. Wootton

The results from m_y are similarly random but with a bias towards 1. The limited
certainty of the qubit has therefore been shared between these two possible types of
output. Both are mostly, but not completely certain of what output to give.
To fully understand what is happening here, we need a way of visualizing the
results. Specifically, we will plot the probability for the outcomes 0 and 1 for each of
the three types of measurement on a 3D plot. Since this is well known to be a useful
thing to do, it has a name: the Bloch sphere.
Let us first use this visualization on an empty circuit: one that outputs a 0 with
certainty for a z measurement, but random results for the others:
qc = QuantumCircuit(1)

The state of the qubit is represented as a point in 3D space. The x, y and z axes are
used to show the probabilities for x, y and z measurements, respectively. If the output
is certain to be 0 for each measurement, the qubit state is depicted on one extreme
along the corresponding axis. If it is certain to be 1, it is depicted on the other side.
For a completely random result, it is in the middle. In this case the certainty of a 0 for
the z measurement puts the point at the very top of the image. For the x and y axes,
the point lies in the middle. The state corresponds to the point labelled |0 (Fig. 3).
Next, let us plot the state after applying the X gate to the qubit, for which the z
measurement is certain to output 1:
qc = QuantumCircuit(1)
qc.x(0)

Here the point is at the bottom of the Bloch sphere (Fig. 4). The positions of
these outcomes are completely opposed to each other. The state in this case has been
labelled |1.

Fig. 3 Bloch sphere

representing a qubit in state
|0
The History of Games for (Quantum) Computers 361

Fig. 4 Bloch sphere

representing a qubit in state
|1

We find similarities for the states whose output is certain for the x measurement.
Let us examine again the state where the H gate is used to ensure that the qubit is
certain to output 0 for the x measurement:
qc = QuantumCircuit(1)
qc.h(0)

Although the visualization does not include a specific label for this state, it is
usually referred to as the |+ state (Fig. 5).
Let us now construct a circuit that is certain to output 1 for an x measurement:
qc = QuantumCircuit(1)
qc.x(0)

Fig. 5 Bloch sphere

representing a qubit in state
|+
362 J. R. Wootton

Fig. 6 Bloch sphere

representing a qubit in state
|–

qc.h(0)

As we should expect, the corresponding point lies on the opposite side to the one
above. It is a state known as |– (Fig. 6).
The states corresponding to the certainty of a 0 or 1 outcome for y measurement
similarly correspond to the opposite points on the y-axis. This then gives us six
extreme points on the surface of a sphere. Each represents an outcome that is certain
for one of the three types of measurement, but completely random for the other two.
With other gates, we can explore the entire surface of the sphere. The RX gate for
a given angle corresponds to the rotation of the state around the x-axis by that angle.
There are similarly ry and rz gates that rotate around their respective axes.
For example, here is the –pi/4 rotation with RX gate that we considered earlier,
with the point corresponding to the state highlighted as a blue point. From its place-
ment, it is easy to see why it is biased towards 0 for a z measurement and 1 for a y
measurement: it lies exactly between those points (Fig. 7).
The fact that the qubit state appears bound to the surface of a sphere is a conse-
quence of the conservation of certainty. It is an expression of exactly how the trade-off
for certainty between the different measurements works.
Now we have seen all that needs to be known about a single qubit. All single-qubit
manipulations can be constructed from the gates discussed above. Everything else,
like special names for specific choices of angle and axis, is just administration.
The History of Games for (Quantum) Computers 363

Fig. 7 Bloch sphere

representing a qubit after the
application of the RX gate
with angle –pi/4

9.6 Single-Qubit Terrain Generation

Using the properties of the single qubit, we will now devise an algorithm for terrain
generation. This is based upon a proof-of-principle game created for the Ludum Dare
45 game jam (Wootton, 2019).
This algorithm for procedural generation is based around a single function:
get_height(i, j).
The function above will simply tell us the height of the terrain at any given point,
specified by the coordinates i and j. Note that here we use i and j as coordinates,
rather than the usual x and y because this section has already been using X and Y as
gates, and x and y as types of measurements.
The height in this function is calculated using a single-qubit quantum circuit.
Specifically, gates are applied that depend upon the coordinates of the point. The
choice of gates presented here was chosen quite arbitrarily. The reader is encouraged
to play around with alternatives.
One aspect of the choice of gates that was done very deliberately was to ensure that
the circuit created for any given point (i, j) is quite similar to that for its neighbours,
to ensure smooth terrain. For a specific example of this, let us consider a circuit that
consists of an rx and a ry rotation. For these, we shall slowly change the angle for
the RX gate as we move from point to point along the horizontal axis of the map,
and similarly change the angle for ry as we move along the vertical axis of the map.
Note that this is an arbitrary choice, and does not reflect any deep meaning. There
is no connection between the axes on the map and those of the Bloch sphere. It was
chosen simply because it seemed like a nice choice. Again, the reader is encouraged
to play around with alternatives.
After these gates, a z measurement is performed and the probability of getting the
output 1 is calculated. This probability could itself be used as the height. However,
it was found that more appealing terrain results when using p as the height. Again,
364 J. R. Wootton

this is a rather arbitrary choice and the reader can try different approaches. Below is
the resulting function:
def get_height(i, j):
qc = QuantumCircuit(1, 1)
# perform rotations, whose angles depend on i and j
qc.x(0)
# low frequency rotations to create island shape
qc.rx((1/32)*i*pi, 0)
qc.ry((1/32)*j*pi, 0)
# perform a z measurement
qc.measure(0, 0)
# determine the probability of a 1
counts = backend.run(
qc,shots=1000
).result().get_counts()
try:
p = counts[’1’]/1000
except:
p=0
# return p2 as the height
return p**2

In order to see what kind of terrain this generates, we can run it for a set of points
and plot the output. Figure 8 plots the results for a 30 × 30 set of points around
the origin. A terrain colour map is used, which colours low values as blue, and then
green, up to high values as brown and then white.
Though this is unmistakably some terrain, it clearly isn’t very interesting or real-
istic. In order to make it nicer, we can add some more gates. As one example, we
could break up the single peak by adding gates for which the angles of rotation make
larger changes as we move from one point to its neighbours, as follows (Fig. 9):

Fig. 8 Example of a simple

terrain generation
The History of Games for (Quantum) Computers 365

Fig. 9 Slightly more

sophisticated example of
terrain generation

def get_height(i, j):

qc = QuantumCircuit(1, 1)
# perform rotations, whose angles depend on i and j
qc.x(0)
# low frequency rotations to create island shape
qc.rx((1/32)*i*pi, 0)
qc.ry((1/32)*j*pi, 0)
qc.rx((1/16)*i*pi, 0)
qc.ry((1/16)*j*pi, 0)
qc.rx((1/8)*i*pi, 0)
qc.ry((1/8)*j*pi, 0)
# perform a z measurement
qc.measure(0,0)
# determine the probability of a 1
counts = backend.run(
qc,shots=1000
).result().get_counts()
try:
p = counts[’1’]/1000
except:
p=0
# return p2 as the height
return p**2

The example shown in Fig. 9 still is not very realistic. In order to create something
better, one should experiment with various new combinations of gates. We could also
try adding in a randomly generated seed. Note that this would not mean using the
randomness of the quantum computer: our calculation and use of a probability for the
height is an effort to avoid this randomness and make something deterministic from
it instead. Rather it is a random choice of parameters in the circuits we run in order to
366 J. R. Wootton

make each map unique. There are many ways that the process could be seeded, so we
again need to make an arbitrary choice. Since this method for procedural generation
was developed for a game jam, the method for seed generation was inspired by the
theme of the jam. In this case, the theme of Ludum Dare 45 was ‘Start with nothing’.
For this reason, the game starts with no seed, and hence no terrain. As the player
explores, their (presumably random) path was used to generate the world as they
explore it.
Another way of improving this method is to not just run the circuit with a z
measurement, but with x and y measurements too. This additional information could
then also be used to determine the characteristics of the terrain at any given point.
There are many further ideas that can be explored, but the limitations of this
method must also be remembered. Manipulating a single qubit is, as we saw from
the Bloch sphere, much like rotating a ball. As such, we won’t get anything uniquely
quantum that could not be done using standard tools for 3D rotations.

10 Conclusion

The computer games industry is, by definition, dependent on computational tech-

nology to develop ever more engaging experiences. As such, new tools enabled
by fault-tolerant quantum computers will certainly be seized upon as soon as they
become available.
But there is more to computer games than the industry. It is also an area beloved
by hobbyists, used for artistic expression, and a basis for education. These are the
approaches that will see the first benefits from quantum computers, as people begin
to explore the new technology through the medium of making games.
As with games for conventional computers in the 1950s and 1960s, it is likely that
we will see many examples of games being useful for quantum computers before
we see quantum computers become truly useful for games. The years ahead will
be ones of experimentation, with quantum computing rising from the grassroots of
game jams projects and indie developers. It’s hard to tell when quantum approaches
will reach the big AAA game studios, but we will have a lot of fun on the journey!

References

Benioff, P. (1980). The computer as a physical system: A microscopic quantum mechanical Hamil-
tonian model of computers as represented by Turing machines. Journal of Statistical Physics.,
22(5), 563–591.
Q Cards. https://zhamul.itch.io/qcards
Copeland, B. J. (2004). The essential Turing (pp. 563–564). Oxford University Press
Jordan, S. (2021) Quantum algorithm zoo. https://quantumalgorithmzoo.org
Qiskit. https://qiskit.org
QiskitBlocks. https://github.com/JavaFXpert/QiskitBlocks
The History of Games for (Quantum) Computers 367

QPong. https://github.com/HuangJunye/QPong
IBM Quantum Composer and Lab. https://quantum-computing.ibm.com
Samuel, A. L. (1959). Some studies in machine learning using the game of checkers. IBM Journal
of Research and Development, 3(3), 210–229.
Wootton, J. R. (2016). Demonstrating non-Abelian braiding of surface code defects in a five qubit
experiment. Quantum Science and Technology, 2(1).
Wootton, J. R. (2017). How to program a quantum computer. Retrieved December 03, 2020, from
https://medium.com/qiskit/how-to-program-a-quantum-computer-982a9329ed02
Wootton, J. R. (2018). Benchmarking of quantum processors with random circuits. https://arxiv.
org/abs/1806.02736
Wootton, J. R. (2019). Genesis. https://ldjam.com/events/ludum-dare/45/genesis-1
Wootton, J. R. (2020a). A quantum procedure for map generation, 2020 IEEE Conference on Games
(CoG), 73-80.
Wootton, J. R. (2020b). Procedural generation using quantum computation, International Confer-
ence on the Foundations of Digital Games, 98.