Quantum probability and randomness

Editorial Quantum Probability and Randomness Andrei Khrennikov 1,∗and Karl Svozil 2 Linnaeus University, 351 95 Växjö, Sweden 1040 Vienna, Austria; svozil@tuwien.ac.at Received: 2 Januar

Quantum

Probability and

Randomness

Andrei Khrennikov and Karl Svozil Edited by

Printed Edition of the Special Issue Published in Entropy

Special Issue Editors

Andrei Khrennikov

Karl Svozil

MDPI• Basel • Beijing • Wuhan • Barcelona • Belgrade

Andrei Khrennikov

Linnaeus University

Sweden

Karl SvozilInstitute for Theoretical Physics of theVienna Technical University

This is a reprint of articles from the Special Issue published online in the open access journal Entropy

(ISSN 1099-4300) from 2018 to 2019 (available at: https://www.mdpi.com/journal/entropy/specialissues/Probability Randomness)

For citation purposes, cite each article independently as indicated on the article page online and asindicated below:

LastName, A.A.; LastName, B.B.; LastName, C.C Article Title Journal Name Year, Article Number,

 2019 by the authors Articles in this book are Open Access and distributed under the Creative

Commons Attribution (CC BY) license, which allows users to download, copy and build uponpublished articles, as long as the author and publisher are properly credited, which ensures maximumdissemination and a wider impact of our publications

The book as a whole is distributed by MDPI under the terms and conditions of the Creative Commonslicense CC BY-NC-ND

About the Special Issue Editors vii Andrei Khrennikov and Karl Svozil

Quantum Probability and Randomness

Reprinted from: Entropy 2019, 21, 35, doi:10.3390/e21010035 1 Mladen Paviˇci´c and Norman D Megill

Vector Generation of Quantum Contextual Sets in Even Dimensional Hilbert Spaces

Reprinted from: Entropy 2018, 20, 928, doi:10.3390/e20120928 6 Aldo C Mart´ınez, Aldo Sol´ıs, Rafael D´ıaz Hern´andez Rojas, Alfred B U’Ren, Jorge G Hirsch and Isaac P´erez Castillo

Advanced Statistical Testing of Quantum Random Number Generators

Reprinted from: Entropy 2018, 20, 886, doi:10.3390/e20110886 18

Maria Luisa Dalla Chiara, Hector Freytes, Roberto Giuntini, Roberto Leporini and Giuseppe Sergioli

Probabilities and Epistemic Operations in the Logics of Quantum Computation

Reprinted from: Entropy 2018, 20, 837, doi:10.3390/e20110837 31

Xiaomin Guo, Ripeng Liu, Pu Li, Chen Cheng, Mingchuan Wu and Yanqiang Guo

Enhancing Extractable Quantum Entropy in Vacuum-Based Quantum Random NumberGenerato

Reprinted from: Entropy 2018, 20, 819, doi:10.3390/e20110819 53

Marco Enr´ıquez, Francisco Delgado and Karol ˙ Zyczkowski

Entanglement of Three-Qubit Random Pure States

Reprinted from: Entropy 2018, 20, 745, doi:10.3390/e20100745 66

Margarita A Man’ko and Vladimir I Man’ko

New Entropic Inequalities and Hidden Correlations in Quantum Suprematism Picture of QuditStates

Reprinted from: Entropy 2018, 20, 692, doi:10.3390/e20090692 85

Arkady Plotnitsky

“The Heisenberg Method”: Geometry, Algebra, and Probability in Quantum Theory

Reprinted from: Entropy 2018, 20, 656, doi:10.3390/e20090656 102

Francisco Delgado

SU (2) Decomposition for the Quantum Information Dynamics in 2d-Partite Two-Level

Quantum Systems

Reprinted from: Entropy 2018, 20, 610, doi:10.3390/e20080610 148

Marius Nagy and Naya Nagy

An Information-Theoretic Perspective on the Quantum Bit Commitment Impossibility Theorem

Reprinted from: Entropy 2018, 20, 193, doi:10.3390/e20030193 189

Gregg Jaeger

Developments in Quantum Probability and the Copenhagen Approach

Reprinted from: Entropy 2018, 20, 420, doi:10.3390/e20060420 205

Reprinted from: Entropy 2018, 20, 284, doi:10.3390/e20040284 224

Andrei Khrennikov, Alexander Alodjants, Anastasiia Trofimova and Dmitry Tsarev

On Interpretational Questions for Quantum-Like Modeling of Social Lasing

Reprinted from: Entropy 2018, 20, 921, doi:10.3390/e20120921 229

Paul Ballonoff

Paths of Cultural Systems

Reprinted from: Entropy 2018, 20, 8, doi:10.3390/e20010008 253

Andrei Khrennikov was born in 1958 in Volgorad and spent his childhood in the town of Bratsk,

in Siberia, north from the lake Baikal In the period between 1975–1980, he studied at MoscowState University, department of Mechanics and Mathematics, and in 1983, he received his PhD

in mathematical physics (quantum field theory) at the same department In 1990, he becamefull professor at Moscow University for Electronic Engineering Since 1997, he has been aprofessor of applied mathematics at Linnaueus University, South-East Sweden, and since 2002,the director of the multidisciplinary research center at this university, as well as the InternationalCenter for Mathematical Modeling in Physics, Engineering, Economics, and Cognitive Science.His research interests are multidisciplinary, e.g., foundations of quantum physics, information, andprobability, cognitive modeling, ultrametric (non-Archimedean) mathematics, dynamical systems,infinite-dimensional analysis, quantum-like models in psychology, and economics and finances

He is the author of approcimately 500 papers and 20 monographs in mathematics, physics, biology,psychology, cognitive science, economics, and finances

Karl Svozil is a professor of theoretical physics at Vienna’s University of Technology He earned a

Dr Phil while studying philosophy and sciences in the old, “Humboldtian” tradition in Heidelbergand Vienna, emphasizing the unity of knowledge After attending the Lawrence Berkeley Laboratoryand UC Berkeley, he worked as a physicist in Vienna, with many shorter stays abroad—among them,the Lomonosov Moscow State University, Lebedev Physical Institute and ICPT Trieste He recentlyheld an honorary position at the University of Auckland and served as president of the InternationalQuantum Structures Association Svozil’s main interests include quantum logic, issues related to(in)determininsism in physics, and “relativizing” relativity theory in the spirit of Alexandrov’stheorem of incidence geometry

Editorial

Quantum Probability and Randomness

Andrei Khrennikov 1,∗and Karl Svozil 2

Linnaeus University, 351 95 Växjö, Sweden

1040 Vienna, Austria; svozil@tuwien.ac.at

Received: 2 January 2019; Accepted: 3 January 2019; Published: 7 January 2019

Keywords: quantum foundations; probability; irreducible randomness; random number generators;

quantum technology; entanglement; quantum-like models for social stochasticity; contextuality

The recent quantum information revolution has stimulated interest in the quantum foundations

by perceiving and re-evaluating the theory from a novel information-theoretical viewpoint [1 5].Quantum probability and randomness play the crucial role in foundations of quantum mechanics

It might not be totally unreasonable to claim that, already starting from some of the earliest(in hindsight) indications of quanta in the 1902 Rutherford–Soddy exponential decay law and thesmall aberrations predicted by Schweidler [6], the tide of indeterminism [7,8] was rolling againstchartered territories of fin de siécle mechanistic determinism Riding the waves were researchers like

Exner, who already in his 1908 inaugural lecture as rector magnificus [9] postulated that irreduciblerandomness is, and probability theory therefore needs to be, at the heart of all sciences; natural as well

as social Exner [10] was forgotten but cited in Schrödinger’s alike “Zürcher Antrittsvorlesung” of

1922 [11] Not much later Born expressed his inclinations to give up determinism in the world of theatoms [12], thereby denying the existence of some inner properties of the quanta which condition adefinite outcome for, say, the scattering after collisions

Von Neumann [13] was among the first who emphasized this new feature which was very differentfrom the “in principle knowable unknowns” grounded in epistemology alone Quantum randomness

was treated as individual randomness; that is, as if single electrons or photons are sometimes capable of

behaving acausally and irreducibly randomly Such randomness cannot be reduced to a variability

of properties of systems in some ensemble Therefore, quantum randomness is often considered as

irreducible randomness.

Von Neumann understood well that it is difficult, if not outright impossible in general, to checkempirically the randomness for individual systems, say for electrons or photons In particular, he

proceeded with the statistical interpretation of probability based on the mathematical model of von

Mises [14,15] based upon relative frequencies after admissible place selections.

At the same time, it is just and fair to note that the aforementioned tendencies to groundphysics, and by reductionism, all of science, in ontological indeterminism, have been stronglycontested and fiercely denied by eminent physicists; most prominently by Einstein Planck [16](p 539) (see also Earman [17] (p 1372)) believed that causality could be neither generally proved norgenerally disproved He suggested to postulate causality as a working hypothesis, a heuristic principle,

a sign-post (and for Planck the most valuable sign-post we possess) “to guide us in the motley confusion

of events”.

This is a good place to remark that random features of an individual system can be discussed in

the framework of subjective probability theory The individual (irreducible) interpretation of quantum

randomness due to von Neumann matches well with the subjective probability interpretation ofquantum mechanics (QBism, see, e.g., [18,19])

The main reason for keeping the statistical interpretation was that the aforementioned individualrandomness of quantum systems was considered by von Neumann as one of the basic features ofnature (and not of the human mind!) Von Neumann was sure that such a natural phenomenon must

be treated statistically (by the same reason Bohr also treated quantum randomness statistically, see [20]for details)

In particular, von Neumann remarked [13] (pp 301–302), that, for measurement of some quantity

R for an ensemble of systems (of any origin),

It is not surprising that R does not have a sharp value , and that a positive dispersion exists However, two different reasons for this behavior a priori conceivable:

1 The individual systems S1, , SN of our ensemble can be in different states, so that the ensemble

[S1, , SN] is defined by their relative frequencies The fact that we do not obtain sharp values

for the physical quantities in this case is caused by our lack of information: we do not know in which state we are measuring, and therefore we cannot predict the results.

2 All individual systems S1, , SN are in the same state, but the laws of nature are not causal Then, the cause of the dispersion is not our lack of information, but nature itself, which has disregarded the principle of sufficient cause.

These are characterizations of epistemic and ontic indeterminism, respectively Von Neumannfavored the second, ontic, case which he considered “important and new” (and which he believed

to be able to corroborate [21]) Therefore, for von Neumann, quantum randomness is essentially astatistical exhibition of violation of causality, a violation of the principle of sufficient cause

We compare this kind of randomness with classical interpretations of randomness, see, e.g.,Chapter 2 [22]:

1 unpredictability (von Mises),

2 complexity-incompressibility (Kolmogorov, Solomonof, Chaitin),

3 typicality (Martin-Löf)

It seems that the interpretation of randomness as unpredictability (von Mises) is very close to theinterpretation of quantum randomness as an exhibition of acausality

The article by Pavicic and Megill [23], Vector Generation of Quantum Contextual Sets in Even

Dimensional Hilbert Spaces, is a novel contribution to quantum contextuality theory As is well

known, the most elaborated contextual sets, which offer blueprints for contextual experiments andcomputational gates, are the Kochen–Specker sets In this paper, a method of vector generation thatsupersedes previous methods is presented It is implemented by means of algorithms and programsthat generate hypergraphs embodying the Kochen-Specker property and that are designed to becarried out on supercomputers

Recent years were characterized by the tremendous development of quantum technology.Quantum random generators are among the most important outputs of this development As ispointed out in the review by Martínez et al [24], Advanced Statistical Testing of Quantum Random

Number Generators, the natural laws of the microscopic realm provide a fairly simple method to

generate non-deterministic sequences of random numbers, based on measurements of quantum states

In practice, however, the experimental devices on which quantum random number generators arebased are often unable to pass some tests of randomness In this review, two such tests are brieflydiscussed, the challenges that have to be encountered in experimental implementations are pointedout Finally, the authors present a fairly simple method that successfully generates non-deterministicmaximally random sequences

The connection between quantum logic and quantum probability is highlighted byDalla Chiara et al [25] in the paper entitled Probabilities and Epistemic Operations in the Logics of Quantum

Computation The authors stress that quantum computation theory has inspired new forms of quantum

logic, called quantum computational logics In this article, they investigate the epistemic operation(which is informally used in a number of interesting quantum situations): the operation “beingprobabilistically informed”.

In the paper entitled Enhancing Extractable Quantum Entropy in Vacuum-Based Quantum Random

Number Generator, Guo et al [26] commit to enhancing quantum entropy content in the vacuumnoise based quantum RNG They have taken into account main factors in this proposal to establishthe theoretical model of quantum entropy content, including the effects of classical noise, theoptimum dynamical analog-digital convertor (ADC) range, the local gain and the electronic gain

of the homodyne system

The work by Enríquez et al [27], Entanglement of Three-Qubit Random Pure States, is devoted

to studying entanglement properties of generic three-qubit pure states There are obtained thedistributions of both the coefficients and the only phase in the five-term decomposition of Acín et al for

an ensemble of random pure states generated by the Haar measure on U(8) Furthermore, the authors

analyze the probability distributions of two sets of polynomial invariants One of these sets allows

us to classify three-qubit pure states into four classes Entanglement in each class is characterizedusing the minimal Renyi–Ingarden–Urbanik entropy The numerical findings suggest some conjecturesrelating some of those invariants with entanglement properties to be ground in future analytical work

In the article New Entropic Inequalities and Hidden Correlations in Quantum Suprematism Picture

of Qubit States, Margarita A Man’ko and Vladimir I Man’ko [28] considered an analog of Bayes’formula and the nonnegativity property of mutual information for systems with one random variable.For single-qubit states, they presented new entropic inequalities in the form of the subadditivity andcondition corresponding to hidden correlations in quantum systems Qubit states are represented inthe quantum suprematism picture, where these states are identified with three probability distributions,describing the states of three classical coins, and illustrating the states by Triada of Malevich’s squareswith areas satisfying the quantum constraints

In the article by Plotnitsky [29], “The Heisenberg Method”: Geometry, Algebra, and Probability in

Quantum Theory, quantum theory is reconsidered in terms of the following principle, which can be

symbolically represented as QUANTUMNESS→PROBABILITY→ALGEBRA The principle states

that the quantumness of physical phenomena, that is, the specific character of physical phenomenaknown as quantum, implies that our predictions concerning them are irreducibly probabilistic, even indealing with quantum phenomena resulting from the elementary individual quantum behavior (such

as that of elementary particles), which in turn implies that our theories concerning these phenomenaare fundamentally algebraic, in contrast to more geometrical classical or relativistic theories, althoughthese theories, too, have an algebraic component to them

The work by Delgado [30], SU(2) Decomposition for the Quantum Information Dynamics in 2d-Partite

Two-Level Quantum Systems, presents a formalism to decompose the quantum information dynamics in

SU(22d ) for 2d-partite two-level systems into 2 d−1 SU(2) quantum subsystems It generates an easierand more direct physical implementation of quantum processing developments for qubits

The paper by Marius Nagy and Naya Nagy [31], An Information-Theoretic Perspective on the

Quantum Bit Commitment Impossibility Theorem, proposes a different approach to pinpoint the causes

for which an unconditionally secure quantum bit commitment protocol cannot be realized, beyond thetechnical details on which the proof of Mayers’ no-go theorem is constructed

In the Copenhagen approach to quantum mechanics as characterized by Heisenberg, probabilitiesrelate to the statistics of measurement outcomes on ensembles of systems and to individualmeasurement events via the actualization of quantum potentiality In the review by Jaeger [32],

Developments in Quantum Probability and the Copenhagen Approach, brief summaries are given of a series

of key results of different sorts that have been obtained since the final elements of the Copenhageninterpretation were offered and it was explicitly named so by Heisenberg—in particular, resultsfrom the investigation of the behavior of quantum probability since that time, the mid-1950s Thisreview shows that these developments have increased the value to physics of notions characterizing

the approach which were previously either less precise or mainly symbolic in character, includingcomplementarity, indeterminism, and unsharpness.

A new way of orthogonalizing ensembles of vectors by “lifting” them to higher dimensions isintroduced by Havlicek and Svozil [33] entitled Dimensional Lifting through the Generalized Gram-Schmidt

Process This method can potentially be utilized for solving quantum decision and computing problems.

Recently the mathematical formalism and methodology of quantum theory started to be widelyapplied outside of physics, especially in psychology, decision making, social and political science(see, e.g., [34]) This special issue contains one paper belonging to this area of research, the article ofKhrennikov et al [35], On Interpretational Questions for Quantum-Like Modeling of Social Lasing The

formalisms of quantum field theory and theory of open quantum systems are applied to modeling

socio-political processes on the basis of the social laser model describing stimulated amplification of

social actions The main aim of this paper is establishing the socio-psychological interpretations of the

quantum notions playing the basic role in lasing modeling

The article by Paul Ballonoff [36], Paths of Cultural Systems, is also devoted to applications outside

physics, namely to anthropology A theory of cultural structures predicts the objects observed byanthropologists A viable history (defined using pdqs) states how an individual in a populationfollowing such history may perform culturally allowed associations, which allows a viable history tocontinue to survive The vector states on sets of viable histories identify demographic observables ondescent sequences

We hope that the reader will enjoy the present issue, which will be useful to experts working inall domains of quantum physics and quantum information theory, ranging from experimenters, totheoreticians and philosophers

The cover of this electronic book was created by Renate Quehenberg and the editors would like tothank her for the graphical contribution to this special issue

Acknowledgments: We express our thanks to the authors of the above contributions, and to the journal Entropy

and MDPI for their support during this work.

Conflicts of Interest: The authors declare no conflict of interest.

References

Phys 2012, 42, 721–724 [CrossRef ]

Advances and problems Phys Scr 2014, T163, 010301.

limits of quantum mechanics: Theory and experiment, Volume 1 Found Phys 2015, 45, 707–710;

quantum mechanics: Theory and experiment, Volume 2 Found Phys 2015, doi:10.1007/s10701-015-9950-1.

A Hölder: Wien, Austria, 2016.

Wien, Germany, 1922.

12. Born, M Zur Quantenmechanik der Stoßvorgänge Z Phys 1926, 37, 863–867 [CrossRef ]

USA, 1955.

Philosophy of Science; Butterfield, J., Earman, J., Eds.; North-Holland: Amsterdam, The Netherlands, 2007;

pp 1369–1434.

Reconsideration of Foundations; Växjö University Press: Växjö, Schweden, 2002; pp 463–543.

Mod Phys 2017, 60, 136–148 [CrossRef ]

UK, 2016.

Entropy 2018, 20, 928 [CrossRef ]

20, 745 [CrossRef ]

[ CrossRef ]

Germany; New York, NY, USA, 2010.

c

 2019 by the authors Licensee MDPI, Basel, Switzerland This article is an open access

article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

Article

Vector Generation of Quantum Contextual Sets in Even Dimensional Hilbert Spaces

Mladen Paviˇci´c 1,2, * ,† and Norman D Megill 3,†

Institute Ruder Boškovi´c, 10000 Zagreb, Croatia

Received: 29 October 2018; Accepted: 24 November 2018; Published: 5 December 2018

Abstract: Recently, quantum contextuality has been proved to be the source of quantum

computation’s power That, together with multiple recent contextual experiments, prompts improvingthe methods of generation of contextual sets and finding their features The most elaboratedcontextual sets, which offer blueprints for contextual experiments and computational gates, arethe Kochen–Specker (KS) sets In this paper, we show a method of vector generation that supersedesprevious methods It is implemented by means of algorithms and programs that generate hypergraphsembodying the Kochen–Specker property and that are designed to be carried out on supercomputers

We show that vector component generation of KS hypergraphs exhausts all possible vectors that can

be constructed from chosen vector components, in contrast to previous studies that used incompletelists of vectors and therefore missed a majority of hypergraphs Consequently, this unified method

is far more efficient for generations of KS sets and their implementation in quantum computationand quantum communication Several new KS classes and their features have been found and areelaborated on in the paper Greechie diagrams are discussed

Keywords: quantum contextuality; Kochen–Specker sets; MMP hypergraphs; Greechie diagrams

1 Introduction

Recently, it has been discovered that quantum contextuality might have a significant place in

a development quantum communication [1,2], quantum computation [3,4], and lattice theory [5,6].This has prompted experimental implementation of 4-, 6-, and 8-dimensional contextual experimentswith photons [7 13], neutrons [14–16], trapped ions [17], solid state molecular nuclear spins [18],and paths [19,20]

Experimental contextual tests involve subtle issues, such as the possibility of noncontextual hiddenvariable models that can reproduce quantum mechanical predictions up to arbitrary precision [21].These models are important because they show how assignments of predetermined values to densesets of projection operators are precluded by any quantum model Thus, Spekkens [22] introducesgeneralised noncontextuality in an attempt to make precise the distinction between classical andquantum theories, distinguishing the notions of preparation, transformation, and measurement ofnoncontextuality and by doing so demonstrates that even the 2D Hilbert space is not inherentlynoncontextual Kunjwal and Spekkens [23] derive an inequality that does not assume that the valueassignments are deterministic, showing that noncontextuality cannot be salvaged by abandoningdeterminism Kunjwal [24] shows how to compute a noncontextuality inequality from an invariantderived from a contextual set/configuration representing an experimental Kochen-Specker (KS) setup.This opens up the possibility of finding contextual sets that provide the best noise robustness in

demonstrating contextuality The large number of such sets that we show in the present work canprovide a rich source for such an effort.

Quantum contextual configurations that have been elaborated on the most in the literature are the

KS sets, and, in this paper, we consider just them In order to obtain KS sets, so far, various methods ofexploiting correlations, symmetries, geometry, qubit states, Pauli states, Lie algebras, etc., have beenfound and used for generating master sets i.e., big sets which contain all smaller contextual sets [25–37]

All of these methods boil down either to finding a list of vectors and their n-tuples of

orthogonalities from which a master set can be read off or finding a structure, e.g., a polytope,from which again a list of vectors and orthogonalities can be read off as well as a master set they build

In the present paper, we take the simplest possible vector components within an n-dimensional Hilbert

space, e.g.,{0, ±1}, and via our algorithms and programs exhaustively build all possible vectors and

their orthogonal n-tuples and then filter out KS sets from the sets in which the vectors are organized.

For a particular choice of components, the chances of getting KS sets are very high We generate KS setsfor even-dimensional spaces, up to 32, that properly contain all previously obtained and known KSsets, present their features and distributions, give examples of previously unknown sets, and present ablueprint for implementation of a simple set with a complex coordinatization

2 Results

The main results presented in this paper concern generation of contextual sets from several basicvector components Previous contextual sets from the literature made use of often complicatedsets of vectors that the authors arrived at, following particular symmetries, or geometries,

or polytope correlations, or Pauli operators, or qubit states, etc In contrast, our approach considersMcKay–Megill–Paviˇci´c (MMP) hypergraphs (defined in Section2.1) from n-dimensional (nD) Hilbert

space (H n , n ≥ 3) originally consisting of n-tuples (in our approach represented by MMP hypergraph

edges) of orthogonal vectors (MMP hypergraph vertices) which exhaust themselves in formingconfigurations/sets of vectors (MMP hypergraphs) Already in [38], we realised that hypergraphsmassively generated by their non-isomorphic upward construction might satisfy the Kochen–Speckertheorem even when there were no vectors by means of which they might be represented (seeTheorem1), and finding coordinatizations for those hypergraphs which might have them, via standardmethods of solving systems of non-linear equations, is an exponentially complex task solvable only forthe smallest hypergraphs [38] It was, therefore, rather surprising to us to discover that the hypergraphsformed by very simple vector components often satisfied the Kochen–Specker theorem In this paper,

we present a method of generation of KS MMP hypergraphs, also called KS hypergraphs, via suchsimple sets of vector components

Theorem 1 (MMP hypergraph reformulation of the Kochen–Specker theorem).

There are nD MMP hypergraphs, i.e., those whose each edge contains n vertices, called KS MMP hypergraphs,

to which it is impossible to assign 1s and 0s in such a way that

(α) No two vertices within any of its edges are both assigned the value 1;

(β) In any of its edges, not all of the vertices are assigned the value 0.

In Figure1, we show the smallest possible 4D KS MMP hypergraph with six vertices and threeedges We can easily verify that it is impossible to assign 1 and 0 to its vertices so as to satisfy theconditions (α) and (β) from Theorem1 For instance, if we assign 1 to the top green-blue vertex, then,according to the condition (α), all of the other vertices contained in the blue and green edges must be

assigned value 0, but, herewith, all four vertices in the red edge are assigned 0s in violation of thecondition (β), or, if we assign 1 to the top red-blue vertex, then, according to the condition (α), all the

other vertices contained in the blue and red edges must be assigned value 0, but, herewith, all fourvertices in the green edge are assigned 0s in violation of the condition (β) Analogous verifications go

through for the remaining four vertices We verified that there is neither a real nor complex vector

solution of a corresponding system of nonlinear equations [38] We have not tried quaternions as

of yet

Figure 1 The smallest 4D KS MMP hypergraph without a coordinatization.

When a coordinatization of a KS MMP hypergraph exists, its vertices denote n-dimensional

vectors inH n , n ≥ 3, and edges designate orthogonal n-tuples of vectors containing the corresponding

vertices In our present approach, a coordinatization is automatically assigned to each hypergraph bythe very procedure of its generation from the basic vector components A KS MMP hypergraph with a

given coordinatization of whatever origin we often simply call a KS set.

2.1 Formalism

MMP hypergraphs are those whose edges (of size n) intersect each other in at most n − 2

vertices [26,37] They are encoded by means of printable ASCII characters Vertices are denoted by one

of the following characters:1 2 9 A B Z a b z ! " # $ % & ’ ( ) * - / : ; < = > ? @ [\ ] ˆ _ ‘

{ | } ~ [26] When all of them are exhausted, one reuses them prefixed by ‘+’, then again by ‘++’, and so

forth An n-dimensional KS set with k vectors and m n-tuples is represented by an MMP hypergraph with k vertices and m edges which we denote as a k-m set In its graphical representation, vertices are depicted as dots and edges as straight or curved lines connecting m orthogonal vertices We handle

MMP hypergraphs by means of algorithms in the programs SHORTD, MMPSTRIP, MMPSUBGRAPH,VECFIND, STATES01, and others [5,30,38–41] In its numerical representation (used for computer

processing), each MMP hypergraph is encoded in a single line in which all m edges are successively given, separated by commas, and followed by assignments of coordinatization to k vertices (see 18-9

in Section2.2)

2.2 KS Vector Lists vs Vector Component MMP Hypergraphs

In Table1, we give an overview of most of the k-m KS sets (KS hypergraphs with m vertices and k edges) as defined via lists and tables of vectors used to build the KS master sets that one can

find in the literature These master sets serve us to obtain billions of non-isomorphic smaller KS sets

(KS subsets, subhypergraphs) which define k-m classes In doing so (via the aforementioned algorithms and programs), we keep to minimal, critical, KS subhypergraphs in the sense that a removal of any of

their edges turns them into non-KS sets Critical KS hypergraphs are all we need for an experimentalimplementation: additional orthogonalities that bigger KS sets (containing critical ones) might possess

do not add any new property to the ones that the minimal critical core already has The smallesthypergraphs we give in the table are therefore the smallest criticals Many more of them, as well as theirdistributions, the reader can find in the cited references Some coordinatizations are over-complicated

in the original literature For example (as shown in [37]), for the 4D 148-265 master, components

{0, ±i, ±1, ±ω, ±ω2}, where ω = e2πi/3, suffice for building the coordinatization, and for the 6D 21-7components{0, 1, ω} suffice In addition, {0, ±1} suffice for building the 6D 236-1216.

Table 1 Vector lists from the literature; we call their masters list-masters We shall make use of their

component-masters ω is a cubic root of unity: ω=e2πi/3.

F 8 3

B G

F 8 3

B

G 18−9 {0,±1,± i}

regular polytope

D

I 9 A

S C

5 O

b

H c

E

Q J

d e 8 X

"

r w v L

% G 8 5 2

9 7 J

f g j k Q

S O q n l

* )

YZ

W U T e V E

A IC

s u K

m R p

{0,±1}

u t P

+O +m

1

A C

% i g /XY b

c # <= l m s[

o : _‘

~

?

>

] h +n

| +7 +K +k

+2 +C {

+3

+I +5 +p +L +i v

’ +l

* +T +e r +R ) +P

x +Q

"

q +Z +W ( 9 +N +h G

W

$ \

&

} +A

+4

+G z +S +V +a O +d Q 8

B D

JL+H U+fM

@ +M +6

+1

+J +U +c

4 I N

+B E

+D

5

{0,±1}

Some of the smallest KS hypergraphs in the table have ASCII characters assigned and some donot This is to stress that we can assign them in an arbitrary and random way to any hypergraphand then the program VECFIND will provide them with a coordinatization in a fraction of a second.For instance,

18-9:1234,4567,789A,ABCD,DEFG,GHI1,I29B,35CE,68FH

{1={0,0,0,1},2={0,0,1,0},3={1,1,0,0},4={1,-1,0,0},5={0,0,1,1},6={1,1,1,-1},

7={1,1,-1,1}, 8={1,-1,1,1},9={1,0,0,-1},A={0,1,1,0},B={1,0,0,1},C={1,-1,1,-1},D={1,1,-1,-1},E={1,-1,-1,1},F={0,1,0,1},G={1,0,1,0},H={1,0,-1,0},I={0,1,0,0}}.(To simplify parsing, this notation delineates vectors with braces instead of traditional parentheses inorder to reserve parentheses for component expressions.)

However, a real finding is that we can go the other way round and determine the KS sets fromnothing but vector components{0, ±1}.

2.3 Vector-Component-Generated Hypergraph Masters

We put simplest possible vector components, which might build vectors and therefore provide

a coordinatization to MMP hypergraphs, into our program VECFIND Via its option -master,the program builds an internal list of all possible non-zero vectors containing these components.From this list, it finds all possible edges of the hypergraph, which it then generates MMPSTRIP viaits option-U separates unconnected MMP subgraphs We pipe the obtained hypergraphs throughthe program STATES01 to keep those that possess the KS property We can use other programs ofours, MMPSTRIP, MMPSHUFFLE, SHORTD, STATES01, LOOP, etc., to obtain smaller KS subsets andanalyze their features

The likelihood that chosen components will give us a KS master hypergraph and the speedwith which it does so depends on particular features they possess Here, we will elaborate onsome of them and give a few examples Features are based on statistics obtained in the process ofgenerating hypergraphs:

(i) the input set of components for generating two-qubit KS hypergraphs (4D) should contain numberpairs of opposite signs, e.g.,±1, and zero (0); we conjecture that the same holds for 3, 4, qubits;

with 6D it does not hold literally; e.g.,{0, 1, ω} generate a KS master; however, the following

combination ofω’s gives the opposite sign to 1: ω + ω2= −1;

(ii) mixing real and complex components gives a denser distribution of smaller KS hypergraphs;

(iii) reducing the number of components shortens the time needed to generate smaller hypergraphs

and apparently does not affect their distribution

Feature (i) means that, no matter how many different numbers we use as our input components,

we will not get a KS master if at least to one of the numbers, the same number with the oppositesign is not added Thus, e.g.,{0, 1, −i, 2, −3, 4, 5} or a similar string does not give any, while {0, ±1},

or{0, ±i}, or {0, ±( √5− 1)/2} do Of course, the latter strings all give mutually isomorphic KS

masters, i.e., one and the same KS master, if used alone More specifically, they yield a 40-32 masterwith 40 vertices and 32 edges as shown in Table2 When any of them are used together with othercomponents, although they would generate different component-masters, all the latter masters of aparticular dimension would have a common smallest hypergraph as also shown in Table2

Table 2 Component-masters we obtained List-masters are given in Table1 In the last two rows of

the number of them we successfully generated although there are many more of them except in the 40-32 class.

Contains List-Masters

1 4

D A 6 H I

F 8 3

F 8 3

F 8 3

F 8 3

• As for the features (ii) and (iii) above, components{0, ±1, ω} generate the master 180-203 which

has the following smallest criticals 18-9, 20 22-11, 22 26-13, 24 30-15, 30 31-16, 28 35-17,

33 37-18, etc This distribution is much denser than that of, e.g., the list-master 24-24 withreal vectors which in the same span of edges consists only of 18-9, 20-11, 22-13, and 24-15criticals or of the list-master 60-75 which starts with the 26-13 critical In AppendixA, we give adetailed description of a 21-11 critical with a complex coordinatization and give a blueprint for itsexperimental implementation;

• In [19], the reader is challenged to find a master set which would contain the "seven context star"21-7 KS critical (shown in Tables1and2) We find that{0, 1, ω} generate the 216-153 6D master

which contains just three criticals 21-7, 27-9, and 33-11,{0, 1, ω, ω2} generate 834-1609 master

from which we obtained 2.5× 107criticals, and{0, ±1, ω, ω2} generate 11808-314446 master from

which we obtained 3× 107criticals, all of them containing the seven context star Some of theobtained criticals are given in AppendixB;

• The 60-75 list-master contains criticals with up to 41 edges and 60 vertices, while the 2316-3052component-master generated from the same vector components contains criticals with up to close

to 200 edges and 300 vertices;

• The 60-105 list-master contains criticals with up to 40 edges and 60 vertices, while the 156-249component-master generated from the same vector components contains criticals with up to atleast 58 edges and 88 vertices;

• Components{0, ±1} generate 332-1408 6D master which contains the 236-1216 list-master while

originally components{0, ±1/2, ±1/ √3,±1/ √2, 1} were used;

• In [37], we generated 6D criticals with up to 177 vertices and 87 edges from the list-master 236-1216,while, now, from the component-master 11808-314446, we obtain criticals with up to 201 verticesand 107 edges;

• We did not generate 16D and 32D masters because that would take too many CPU days and wealready generated a huge number of criticals from submasters which are also defined by means ofthe same vector components in [37] See also Section3

3 Methods

Our methods for obtaining quantum contextual sets boil down to algorithms and programswithin the MMP language we developed to generate and handle KS MMP hypergraphs as themost elaborated and implemented kind of these sets The programs we make use of, VECFIND,STATES01, MMPSTRIP, MMPSHUFFLE, SUBGRAPH, LOOP, SHORTD, etc., are freely available fromour repository http://goo.gl/xbx8U2 They are developed in [5,29,30,38–40,47,48] and extended forthe present elaboration Each MMP hypergraph can be represented as a figure for a visualisationbut more importantly as a string of ASCII characters with one line per hypergraph, enabling us toprocess millions of them simultaneously by inputting them into supercomputers and clusters For thelatter elaboration, we developed other dynamical programs specifically for a supercomputer or cluster,which enable piping of our files through our programs in order to parallelize jobs The programs havethe flexibility of handling practically unlimited number of MMP hypergraph vertices and edges as wecan see from Table2 The fact that we did not let our supercomputer run to generate 16D and 36Dmasters and our remark that it would be "computationally too demanding" do not mean that suchruns are not feasible with the current computers, but that they would require too many CPU days onthe supercomputer and that we decided not to burden it with such a task at the present stage of ourresearch; see the explanation in Section2.3

4 Conclusions

The main result we obtain is that our vector component generation of KS hypergraphs (sets)exhaustively use all possible vectors that can be constructed from chosen vector components This is

in contrast to previous studies, which made use of serendipitously obtained lists of vectors curtailed

in number due to various methods applied to obtain them Hence, we obtain a thorough andmaximally dense distribution of KS classes in all dimensions whose critical sets can therefore bemuch more effectively used for possible implementation in quantum computation and communication

A comparison of Tables1and2vividly illustrates the difference

In AppendixA, we present a possible experimental implementation of a KS critical with complexcoordinatization generated from{0, ±1, ω} What we immediately notice about the 21-11 critical from

FigureA1is that the edges are interwoven in more intricate way than in the 18-9 (which has beenimplemented already in several experiments), exhibiting the so-calledδ-feature of the edges forming

the biggest loop within a KS hypergraph Theδ-feature refers to two neighbouring edges which share

two vertices, i.e., intersect each other at two vertices [37] It stems directly from the representation

of KS configuration with MMP hypergraphs Notice that theδ-feature precludes interpretation of

practically any KS hypergraph in an even dimensional Hilbert space by means of so-called Greechiediagrams, which by definition require that two blocks (similar to hypergraph edges) do not share more

than one atom (similar to a vertex) [6], on the one hand, and that the loops made by the blocks must

be of order five or higher (which is hardly ever realised in even dimensional KS hypergraphs—seeexamples in [37]), on the other

Our future engagement would be to tackle odd dimensional KS hypergraphs Notice that, in a 3DHilbert space, it is possible to explore similarities between Greechie diagrams and MMP hypergraphsbecause then neither of them can have edges/blocks which share more than one vertex/atom (via theirrespective definitions) and loops in both of them are of the order five or higher [26,39]

Author Contributions: Conceptualization, M.P.; Data Curation, M.P.; Formal Analysis, M.P and N.D.M.; Funding

Acquisition, M.P.; Investigation, M.P and N.D.M.; Methodology, M.P and N.D.M.; Project Administration, M.P.; Resources, M.P.; Software, M.P and N.D.M.; Supervision, M.P.; Validation, M.P and N.D.M.; Visualization, M.P.; Writing—Original Draft, M.P.; Writing—Review and Editing, M.P and N.D.M.

and Education (MSE) of Croatia through the Center of Excellence for Advanced Materials and Sensing Devices (CEMS) funding, and by grants Nos KK.01.1.1.01.0001 and 533-19-15-0022 This project was also supported by the Alexander or Humboldt Foundation Computational support was provided by the cluster Isabella of the Zagreb University Computing Centre, by the Croatian National Grid Infrastructure (CRO-NGI), and by the Center for Advanced Computing and Modelling (CNRM) for providing computing resources of the supercomputer Bura at the University of Rijeka in Rijeka, Croatia The supercomputer Bura and other information and communication

technology (ICT) research infrastructure were acquired through the project Development of research infrastructure for

laboratories of the University of Rijeka Campus, which is co-funded by the European regional development fund.

Miroslav Puškari´c from CNRM are gratefully acknowledged.

Conflicts of Interest: The authors declare no conflict of interest.

Abbreviations

The following abbreviations are used in this manuscript:

Appendix A 21-11 KS Critical with Complex States fromH2⊗ H2

Below, we present a possible implementation of a KS critical 21-11 with complex coordinatizationshown in FigureA1

The vector components of the first qubit on a photon correspond to a linear (horizontal, H, vertical,

V, diagonal, D, antidiagonal A) and circular (right, R, left L) polarization, and those of the second qubit

to an angular momentum of the photon(+2, −2) and (h, v) One-to-one correspondence between

them is given below

1

7 F

J

B D

E

I H G

K 9 A

8

6 5

21−11−a

B=(1,1,0,0) D=(1,1,−1,−1) E=(1,1,1,1) F=(1,−1,1,−1) G=(0,1,0,−1) H=(1,0,−1,0) I=(0,1,0,1) J=(1,−1,1,1) K=(0,0,1,0)

2=(1,−1,−1,−1) 1=(1,1,1,−1) 3=(1,0,0,1) 4=(0,1,−1,0) 5=(0,1,1,0) 6=(0,0,0,1) 7=(1,0,0,0) 8=(0,1,0,0) 9=(0,0,1,−1) A=(0,0,1,1)

L=(1,−1,−i,i) 3

C=(1,−1,i,−i) L

C

Figure A1 21-11 KS set with complex coordinatization.

An example of a tensor product of two vectors/states fromH2⊗ H2is:

|01 = |0, 1 = |01⊗ |12=

10



1

⊗

01



0

01



2

; | − 2 =

01

−1



2

.Now, one can read off our vertex states as follows:

We will now skip real states and go directly to those with imaginary components, C and L, to

illustrate how they can be implemented via circular polarization:

−1



i

1

−1



−i

1

Appendix B 6D Criticals from the Masters Containing the Seven Context Star.

The 216-153 KS master generated from{0, 1, ω} contains 21-7 and 27-9, which can be viewed as

21-7 with a pair ofδ-triplets interwoven with 21-7, as shown in FigureA2 The 834-1609 KS mastergenerated from{0, 1, ω, ω2}, which were used for a construction of 21-7 in [19], contains 39-13 as well.Equally so, the 11808-314446 master generated from{0, ±1, ω, ω2}.

39−13

(together with 21-11 and 27-9); see the text.

References

51, 102103 [CrossRef ]

Ann Henri Poincare 2011, 12, 1417–1429 [CrossRef ]

Implementation of an Eight-Dimensional Kochen-Specker Set and Observation of Its Connection with the

16 Bartosik, H.; Klepp, J.; Schmitzer, C.; Sponar, S.; Cabello, A.; Rauch, H.; Hasegawa, Y Experimental Test of

the Simplest Kochen-Specker Set for Quantum Information Processing Phys Rev Lett 2014, 113, 090404.

[ CrossRef ]

Philos Mod Phys 2004, 35, 151–176 [CrossRef ]

the Kochen-Specker Theorem arXiv 2018, arXiv:1805.02083.

Phys Lett A 1996, 212, 183–187 [CrossRef ]

J Phys A 2010, 43, 105304 [CrossRef ]

Phys Rev A 2013, 88, 012102 [CrossRef ]

J Phys A 2015, 48, 225301 [CrossRef ]

Phys Lett A 2017, 381, 1853–1857 [CrossRef ]

Contextual Sets arXiv 2011, arXiv:1105.1840.

43. Kernaghan, M Bell-Kochen-Specker Theorem for 20 Vectors J Phys A 1994, 27, L829–L830 [CrossRef ]

wpi.edu/Pubs/E-project/Available/E-project-042108-171725/unrestricted/MQPReport.pdf (accessed on

26 November 2018).

J Plus 2012, 127, 86 [CrossRef ]

Quantum Structures; Engesser, K., Gabbay, D., Lehmann, D., Eds.; Elsevier: Amsterdam, The Netherlands,

2007; pp 751–787.

of the 2017 40th International Convention on Information and Communication Technology, Electronics and Microelectronics (MIPRO 2017), Opatija, Croatia, 22–26 May 2017; Biljanovi´c, P., Ed.; Institute of Electrical and Electronics Engineers (IEEE), Curran Associates, Inc.: Red Hook, NY, USA, 2017; pp 195–200, ISBN 9781509049691.

c

 2018 by the authors Licensee MDPI, Basel, Switzerland This article is an open access

article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

Article

Advanced Statistical Testing of Quantum Random Number Generators

Aldo C Martínez 1 , Aldo Solis 2 , Rafael Díaz Hernández Rojas 3 , Alfred B U’Ren 2 ,

Jorge G Hirsch 2 and Isaac Pérez Castillo 4,5, *

ON K1N 6N5, Canada; mbac@ciencias.unam.mx

C.P 04510 Mexico, Mexico; aldo.solis@correo.nucleares.unam.mx (A.S.);

alfred.uren@correo.nucleares.unam.mx (A.B.U.); hirsch@nucleares.unam.mx (J.G.H.)

rafael.diaz.hr@gmail.com

México, Apdo Postal 20-364, Cd Mx., C.P 04510 Mexico, Mexico

Received: 20 October 2018; Accepted: 14 November 2018; Published: 17 November 2018

Abstract: Pseudo-random number generators are widely used in many branches of science, mainly in

applications related to Monte Carlo methods, although they are deterministic in design and, therefore,unsuitable for tackling fundamental problems in security and cryptography The natural laws ofthe microscopic realm provide a fairly simple method to generate non-deterministic sequences ofrandom numbers, based on measurements of quantum states In practice, however, the experimentaldevices on which quantum random number generators are based are often unable to pass some tests

of randomness In this review, we briefly discuss two such tests, point out the challenges that wehave encountered in experimental implementations and finally present a fairly simple method thatsuccessfully generates non-deterministic maximally random sequences

Keywords: Bell inequalities; algorithmic complexity; Borel normality; Bayesian inference; model

selection; random numbers

1 Introduction

Monte Carlo methods are one of the essential staples of the basic sciences in the modern age.Although these gained prominence during the early 1940s, thanks to secret research projects carriedout in Los Alamos Scientific Laboratory by Ulam and von Neumann [1,2], their origins may be tracedback to the famous Buffon’s needle problem, posed by Georges-Louis Leclerc, Comte de Buffon, in the18th century In the present day, Monte Carlo “experiments” are seen as a broad class of computationalalgorithms that use repeated random sampling to obtain numerical estimates of a given natural ormathematical process In order to use these methods efficiently, fully random sequences of numbersare needed Back in the 1940s, this was a tall order, and various methods to generate random sequenceswere used (some of them literally using roulettes), until von Neumann pioneered the concept ofcomputer-based random number generators During the following years, these became the standardtool in Monte Carlo methods and are still generally well-suited for many applications However, thesecomputer-based methods generate pseudo random numbers [3], which means that the generatedsequence can be determined given an algorithmic program and an initial seed, two ingredients whichare hardly random Thus, in order to achieve a truly unpredictable source of random numbers, we must

eliminate these two deterministic aspects The former is easy to overcome using, for example, a pattern

of keystrokes typed on a computer keyboard as a random seed On the other hand, the algorithmicprogram could be replaced, for instance, by a classical chaotic system [4] Examples of the latter abound

in the area of weather prediction and climate sciences

In recent years, however, the community has been moving towards using the fundamental lawsdictating the behaviour of the quantum realm for the generation of sequences of truly random numbers.This seems, at a first glance, to be at odds with the following rather nạve thought: if the natural laws ofthe microscopic world are considered to be a computer program under which a system evolves from aninitial state (a seed), should not its corresponding generated sequence also be predictable? It turns outthat Quantum Mechanics, in its current standard view, related to the Copenhagen interpretation, has aspecial ingredient that makes the random sequence inherently unpredictable for both the generatorand the observer Such a strange behaviour has been eloquently recast over the years in variousforms, famously by the quote “spooky action at a distance” due to Einstein, or mathematically by thecelebrated work of Bell [5,6] The application of quantum randomness in cryptography has given rise

to the concept of device independent randomness certification, which, in a nutshell, corresponds tothose processes that violate Bell’s inequalities [7,8] However, there seems to be some confusion in theliterature regarding two different properties of a given sequence of random numbers The first one,rather important as we have argued above, is whether the sequence is truly random, meaning that it

is unpredictable In contrast, the second one is related to the issue of assessing whether or not it isbiased It is crucial to keep in mind that these two properties are independent, as evidenced by therandom number generator Quantis [9], which is based on a quantum system and is able to pass thestandard tests of randomness (NIST (National Institute of Standards and Technology) suite) [10] buthas difficulties with other tests [11]

Due to recent advances in quantum technologies, and since the NIST suite has been examined

in other works [12], together with a critical view on the use of p-values on which the NIST suite

relies [13], it becomes necessary to consider other criteria for measuring the performance of quantumrandom number generators Thus, we focus solely on two recently introduced approaches: the firstone is based on algorithmic complexity theory evaluating incompressibility and bias at the same time,since an incompressible sequence is necessarily an unbiased one [14], while the second one relies onBayesian model selection Both methods are based on solid structures which lead to a definition ofrandomness that is very intuitive and which arises independently of the development of randomnumber generators We apply them to analyze sequences of random bits generated in our laboratoryusing quantum systems We also address the issue about the origin of the biases observed whenutilizing these types of devices

“algorithmically” random

We now introduce a remarkable result from algorithmic information theory: the Borel-normalitycriterion due to Calude [14], which allows us to asymptotically check whether a sequence is not

“algorithmically” random Assuming we are given a string = {1001010110110 · · · } of || = n bits

(We will only consider binary sequences, but our results are easily generalizable to other alphabets),the idea of the Borel-normality criterion consists primarily of dividing the original sequence into

consecutive substrings of length i and then computing the frequencies of occurrence of each of them.

For brevity and later use, let us defineΩ(i)as the set of 2i substrings that can be formed with i characters,

let ibe the sequence obtained after dividing it into substrings, and|| i ≡ [||/i] Additionally, let N j

i ()

be the number of times the j-th substring of length i appears in  For example, when considering

substrings of length i= 1, we are looking at the frequencies of the symbols Ω(1)= {0, 1} that conform

the original string, while for i = 2, we have to consider the frequencies of four substrings, namely

Ω(2)= {00, 01, 10, 11} According to Calude, a necessary condition for a sequence to be maximally

random is that the deviations of these frequencies with respect to the expected values in the idealrandom case should be bounded as follows [14,15]:

N i j ()

|| i − 1

2i

log2(n)

values of i, the Borel-normality condition cannot guarantee that it is indeed random.

Recently, a Bayesian criterion has been introduced [16,17] by some of the authors of the presentarticle to test, from a purely probabilistic point of view, whether a sequence is maximally random

as understood within information theory [18] The method works by exploiting the Borel-normalitycompression scheme and then recasting the problem of finding possible biases in the sequence as an

inferential one in which Bayesian model selection can be applied Specifically, for a fixed value of i,

we need to consider all the possible probabilistic models, henceforth denoted as{M (i) α } α, that couldhave generated the sequence Each such model determines a unique probability assignation to the

elements ofΩ(i), which depends on a set of prior parametersθ For these parameters, the Jeffreys’

prior, PJeff(θ), turns out to be a convenient choice of prior parameter distribution, as it entails the

“Occam Razor principle” in which more complex models are penalized, as well as being mathematicallyconvenient for the case at hand; some other advantages are pointed out in [16,17]

Next, the question of finding all the generative models that can produce a sequence is ultimately

solved by noticing that all the possible probabilities assignations are in a one-to-one correspondencewith all possible partitions ofΩ(i) Since obtaining the partitions of any set is a straightforwardcombinatorial task [19], we are able to determine all the relevant models when searching for possiblebiases in the generation of For instance, when i = 1, there are two possible models: one in which

the two elements ofΩ(1)are equiprobable, corresponding to the partition{{0, 1}} of Ω(1)into one

subset—i.e., the same set—and another model with probabilities p0= θ, p1= 1 − θ corresponding to

the partition{{0}, {1}} of Ω(1)into two subsets Even though it might seem that the first model isjust a particular case of the second one (by lettingθ= 1/2), we should keep in mind that the priordistributions are different in both cases,δ (θ − 1/2) and 1

π √

θ (1−θ), respectively, thus yielding two

different models Analogously, for i= 2, there is a single unbiased model, which corresponds to thepartition ofΩ(2)into one subset (with probabilities p j = 1/4, for j = 00, 01, 10, 11), and 14 additional

models associated with the different ways of dividingΩ(2)into subsets The latter are related to thenumber of ways of distinguishing among the elements ofΩ(2)during the assignation of probabilities,and thus any of these models would entail some bias when generating a sequence Note that, in general,

for any value of i, we will face a similar situation in which a single model can produce an unbiased,

and hence maximally random, sequence by means of a uniform distribution, while the rest of themwill be some type of categorical distribution

Once all the models have been identified, the remaining part is the computation of the posterior

distribution P

M (i) α , which from an inferential point of view is the most relevant distribution as

it gives the probability that the modelM (i) α has indeed produced the given sequence Note that,

since a generative approach was adopted, we have direct access to the distribution P

 (i) α ,

which can be combined with the parameters’ prior PJeff(θ) to obtaining the distribution P  (i) α =



dθ P  (i) α ,θ PJeff(θ) One of the most important results of [16] is that this marginalization can

be accomplished exactly for all the models and any value of i Therefore, we can obtain the posterior

distribution by a simple application of Bayes’ rule:

M (i) γ being the prior distribution in the space of models that can generate sequences of

strings of length i Therefore, the best model α that describes the dataset is quite simply given by

α = arg maxα P

If the best modelM (i) α  turns out to be the unbiased one for all possible lengths i of the substrings,

then we can say that the process that generated that dataset was maximally random However,

it remains to discuss how large the length i of substrings can be for a given dataset of n bits To answer this, we first note that, for any set containing N elements, the possible number of partitions is given by the N-th Bell number B N[19] Thus, for a given i, all possible partitions of the setΩ(i) will result in B2i

models to be tested, and, therefore, it is expected for them to be sampled at least once when observing

 This means that B2imax ∼ n, which, for sufficiently large n, yields imax= log2(log2(n)), precisely as

in the Borel-normality criterion

Randomness characterization through Bayesian model selection has some clear and naturaladvantages, as already pointed out in [16], but, unfortunately, it has an important drawback:

the number of all possible models for a given length i, given by B2i, grows supra-exponentially

with i: indeed, for i = 1, we have two possible models, for i = 2, we have 15 possible models, for i = 3,

we have instead 4140 possible models, while, for i= 4, we have 10,480,142,147 models Thus, even

if we are able to acquire data for the evaluation of these many models, it becomes computationallyimpractical to estimate the posterior for all of them using Equation (2) There is an elegant strategy

to overcome this difficulty: one can derive bounds similar to those provided by the Borel-normalitycriterion, by comparing the log-likelihood ratio between the maximally random model and themaximally biased one This yields the following bound for the frequencies of occurrence [17]:

3 Ideal Random Number Generation

While intuition dictates that quantum random number generators (QRNG) should be superior

to their classical counterparts, such a comparison was carried out in [11] and very recently in [20],with rather disappointing results For the classical case, the authors used three Pseudo-RandomNumber Generators (PRNG): the generators included in the software packages Mathematica andMaple, and the digits ofπ expressed in base 2 For QRNG, they used two devices: (i) Quantis,

developed by IDQ [9], a quantum random number generator interfaced with a common computer, and(ii) an experiment from a quantum optics group in Vienna The latter experiment consists of a very weaklight source, attenuated to the single photon level, a beam splitter, and two single photon detectors.Leaving aside the question of which QRNG performs better, the real surprise was that the PRNGs comeout in this test with a superior performance, by far, as compared to their quantum counterparts Thisresult appears to be at odds with the natural randomness associated with quantum phenomena Why

is it that the inherent quantum randomness does not translate into better performance with respect toclassical systems? Does randomness, as discussed in this paper, have no impact on the performance ofthe generators? Is this a fundamental or a technical problem?

In order to explore this apparent paradox, we will discuss the different technical and designdifficulties associated with quantum random number generation using light These days, it isstraightforward to detect single photons using avalanche photo-diodes (APD), devices capable ofdetecting up to a few million single photons per second with> 60% detection efficiencies employing

relatively simple electronics With this simple design in mind, we only need a single photon source, abeam splitter (BS), and a couple of single-photon detection devices in order to set up a QRNG device.This minimalistic design is sketched in Figure1

Figure 1 Ideal experimental setup for a nạve QRNG (quantum random number generators) based

on an individual photon source and a beam splitter (BS) Neglecting the possible losses, a photon will activate only one of the two detectors, therefore producing a random bit per photon.

4 Experimental Challenges in Random Number Generation

Suppose now that we have a single-photon source and we want to generate a sequence of bits

to be tested against the bounds given by Equation (1) Let us start by first focusing on the so-called

Borel level, the word length i, which can take a maximum value of imax= log2(log2(n)) In Table1, we

report how imaxgrows with n, up to a value of imax= 6 In order to achieve a Borel level imax= 6, wewill require a dataset of length 1018events Assuming that our single-photon generator and detectorscan cope with around a million of events per second, we would then require on the order of 600 years

to generate a sequence of that length! It turns out that imax= 5 is a more realistic value, since it leads

to a required dataset of size 4.3 × 109events, which can be realistically produced in a couple of hours

Table 1 Necessary data lengths for maximum Borel level imax=log2(log2(n)) The double exponential relation grows so quickly that it is not possible get to level 6.

Calude criterion The first component that we must be wary about is the BS A regular BS usually has

an error figure in the region of 1%, which is very high with respect to the stringent tolerance would

need in order fulfill Borel normality Is it plausible to correct this using a Polarizing Beam Splitter(PBS), instead of the BS, with an active control through feedback of the state of polarization so as tocompensate for any bias in the PBS? In what follows, we investigate this question through a simpleexperiment The state of polarization of a single photon entering the PBS can be written as

where|V and |H refer to the vertical and horizontal polarization components, respectively We can

approach the state in Equation (5) by transmitting the laser beam trough a half wave plate (HWP) so

as to achieve arbitrary rotation of the linear polarization Assuming a perfect, unbiased BS, we would

need an incoming polarization state with a = b = 1/ √2 so that the resulting sequence of bits isunbiased If, on the other hand, the PBS exhibits biases (e.g., due to manufacturing error), we canadjust the orientation of the above-mentioned half wave plate so as to adjust precisely the value of our

coefficients a and b to compensate for the PBS bias.

Our experimental setup, shown in Figure2, can be regarded as the minimal realistic devicefor the implementation of a QRNG The main questions which we wish to address are: (i) howgood are the sequences of bits generated by such a device? In addition, (ii) do they pass theBorel-normality criterion?

AL1

APD2 APD1

Motor Controller

Figure 2 Experimental setup The relative angle of the half wave plate is controlled in order to

reduce bias.

The input state is prepared using the beam from a laser diode (LD) The beam is transmittedthrough a set of neutral density filters (NDF) with a combined optical density 7.3 for attenuation

to a level compatible with the maximum recommended count rate of our single photon detectors.The beam is then transmitted through a half wave plate (HWP) mounted on a motorized rotation stage

so as to control its orientation angle relative to the PBS axes The PBS splits the beam into two spatialmodes according to the H and V polarizations, each of which is coupled with the help of an asphericlens (AL1 and AL2) into a multimode fiber leading to an avalanche photodiode (APD1 and APD2)

We include a polariser (P1 and P2), with an extinction ratio, defined as the ratio of the maximum to theminimum transmission of a linearly polarized input, of 100,000:1 prior to each of the aspheric lenses(AL1 and AL2) for a reduction of the non-polarized intensity reaching the detectors

Suppose now that we prepare our system so that the average relative power P i/(P0+ P1) of each

APD detector i= 0, 1 starts ideally at 1/2 In Figure3, we show how this average relative powerevolves with time (see curve labelled “without feedback”) Note that, even though the system starts

in a perfectly balanced state, it rapidly deviates from this condition The slow change of this curvecan be attributed to thermal drift while the oscillatory component with a period of approximatelyhalf an hour is related to the air conditioning system in the laboratory These effects can be effectivelycompensated by rotating the HWP After some study of the response function of our experimentalsetup, a correction every minute with a proportional controller was sufficient to correct for all theseeffects leading to a steady response (see curve labelled “with feedback”) [21]

Figure 3 Evolution in time of the normalized power The device starts in a perfect balanced state but

quickly deviates from this condition By using a feedback mechanism, we can obtain stability of the normalized power within an error of 0.004.

5 First Battery of Results

We have used the experimental setup described in the previous section to generate a sequence of

4,294,967,296 bits, allowing us to test the Borel–Normality criterion up to level imax= 5 The results

of this analysis are depicted in Figure4 In the plot, bars represent the deviations from the idealvalue for all the strings at each Borel level For instance, in the first part of the analysis (purple bars),there are only two bars corresponding to the frequency of occurrences of substrings “1” and “0”

As our initial setup is very fine-tuned and stable, the bars have practically zero height, with value

5× 10 −6 The green bars represent the second part of the analysis, or Borel level two, corresponding to

frequencies of occurrences of symbols{00, 01, 10, 11}, and so on In the same figure, the horizontal

lines represent the bound given by the right-hand-side of Equation (1) Our first battery of results are aclear disappointment: only the first set for substrings of length one clearly passes the test, while, forhigher lengths, our QRNG fails miserably to pass Calude’s criterion

Figure 4 Results from Borel analysis The first two boxes correspond to the deviations from the mean

at the first Borel level; these exhibit the same height but opposite signs The next four bars (green) represent the deviations for level two, i.e., “00”, ”01”, ”10”, “11” The blue and yellow boxes represent the deviations for level three and four, respectively The red lines correspond to Borel’s bound, which turns out to be much smaller than the deviations Only the first level passes the test.

Furthermore, a closer look at the green bars shows that events 00 and 11 appear more frequentlythan expected, by about 0.005%, (while events 01 and 10 appear less frequently than expected bythe same margin) This effect reveals a correlation between “equal events”, that is, the same digitappearing twice In terms of our experiment, this means that it is more probable to observe an event

in a detector once a previous event has already been recorded Other parts of our test validate this:for Borel level three, the yellow bars indicate that events with alternate zeroes and ones (010 and 101)appear less frequently than expected, also by about 0.005% At Borel levels four and five, the largerdeviations appear for events 0101 and 1010, clearly in accordance with the previous results Thisindicates that certain parts of our experimental setup are introducing unwanted correlations betweenbits, which results in the magnitude of some of the deviations to be 50 times larger than expected Ourexperimental effort clearly does not suffice for our sequences to pass the Borel–Normality criterion.How is this possible?

6 APD Effects on Introducing Correlations

The two main effects in the behavior of our APDs which can introduce undesired correlations

in the resulting sequences of bits are called after-pulsing and dead time [22,23] The first effect,roughly speaking, corresponds to a false detection event due to the residual effects of an avalanchetriggered by a previous event, while the dead time is the time period after each event during whichthe system is not able to record a subsequent incoming optical signal In this case, we have a typicaldead time of 22 ns, a maximum after-pulsing probability of 1%, and a dark counts rate of 100 counts/s

While the device exhibits a linear behaviour up to 5× 106counts/s, the detection rates used in ourexperiments are an order of magnitude lower.

The mechanism by which dead time introduces correlations in our data, particularly inexperimental arrangements with two or more APDs as in our case, is as follows: suppose that wehave an event in one of our detectors During its dead time, it will have zero probability of recordinganother incoming event in that detector, while the other detector still exhibits a non-zero probability ofrecording an event On the other hand, the way after-pulsing introduces correlations in our data is byincreasing the probability of observing consecutively the same event, resulting in the observation of anexcess of the events{00} and {11} in Figure4

These two effects can somewhat be corrected either by re-designing our experimental setup and/ormodifying the software Instead of following this route to generate maximally random sequences ofbits, let us pursue a rather simple solution as discussed below

7 Random Number Generation Using Time Measurements

We now follow a method introduced in [3] Suppose thatρ (x) is the probability density function

of a continuous random random variable X on an interval x ∈ (a, b) Let us further assume that its

real value x is represented up to a given precision so that we assign a parity to x according to the

parity of its least significant digit Next, we divide the interval(a, b) into an even number 2L of bins

This method can be very easily implemented in the lab as follows Suppose that the random

variable X is the time difference between two consecutive photon arrivals to the detector In our case,

these times are of the order of 500 ns to 10 μs A typical sequence of these time differences look like:

ordered using its length i = 1 (purple bins), i = 2 (green bins), i = 3 (blue bins), i = 4 (orange bins), and imax= 5 (yellow bins) The solid red line corresponds to the Borel bound, 8.6 × 10 −5 In the samegraph, the green line is the Bayesian bound given the right-hand-side of Equation (4) that depends

on i and therefore it is not a constant, as is the case for the Borel bound Finally, the height of the

various background colored boxes correspond to the values given by the left-hand-side expression ofthe Bayesian bound

Figure 5 Results for generation using the least significant bits of time tags In this case, the deviations

are very small, so this generation scheme is excellent The solid red lines represent the maximum deviations allowed by the Calude test, while the solid green lines correspond to the bounds given the Bayesian approach.

As we can see, this extremely simple QRNG passes the Borel-normality criterion up to i= 5,

and nearly passes the Bayesian criterion (passes it for i ≤ 4 and slightly exceeds the bound for i = 5).

Notice that, while the previous experimental setup required an accurate balance between zeroes and

ones, in the present case we already have very small deviations at Borel level i= 1, less than 10−5,

showing the convenience of this method For i= 2 (green bins), the deviations are much larger, almost

three times the value for i= 1, but nevertheless they pass the test again by a considerable margin.These results show the lack of correlations between consecutive events, which is the main drawback

of the previous approach It is important to note that, in the results at Borel level i= 4, there is a

substantial increase in the deviations compared with i= 1, 2, 3 While this increase may indicate thepresence of some as yet unidentified correlations, these are of an insufficient magnitude to reach thebounds On the other hand, this experimental setup fails to pass some of the requirements of theBayesian scheme The deviations derived from the Bayesian criterion are shown in Table2, and also

in Figure5 As we can see, all Borel levels pass the Bayesian test, except for the last one, albeit by asmall margin

Table 2 Comparison between the left-hand-side and the right-hand-side of the Borel-type bounds given

bounds are satisfied for the first four Borel levels, but not the last one by a slight margin.

to one, indicating that, given the dataset, this is the most likely model to have generated such data

For Borel level i= 4, we are only able to analyse those models which are in the vicinity, in parameterspace, to the maximally random model These models correspond to partitioningΩ(5)into two subsets,resulting in 32,767 models, giving a total of 32,768, including the maximally random one In this case,

it turns out that the most likely model is not the maximally random one Actually, using the value ofthe posterior probability, this model is ranked in the position 9240 out of all the explored models, and

therefore the sequence of bits fails to pass the Bayesian criterion already at Borel level i= 4 Note that

i= 5 was not included in Table3because we lack the computational power to address this Borel level

Table 3 Value of the posterior distribution P

random model Note that the prior distribution for each Borel model is a flat distribution along all the

random model is the most plausible.

as Bell’s theorem, that simply tells us that, given some initial conditions, it is impossible to predict

the outcome of a single measurement However, in the present review, we have shown that QRNGsactually perform rather poorly in tests of randomness as compared to classical PRNGs The reason isfairly simple: unpredictability has nothing to do with bias, and while experimental devices based onQuantum Mechanics may produce a truly unpredictable random signal, they also tend, more oftenthan not, to introduce correlations In particular, for QNRGs based on optical devices, we have beenable to account for two, perhaps amongst the many, effects that introduce bias in our data While inour own experimental work involving a QNRG we have failed to obtain sequences which obey theBorel and Bayesian criteria, we were able to show that extracting sequences from the least significantdigits of times of arrival represents a promising strategy.

Author Contributions: J.G.H and A.B.U conceived and designed the experiments; A.C.M and A.S performed

the experiments; I.P.C and R.D.H.R developed the method based on Bayesian Inference; A.C.M, A.S and R.D.H.R analysed the data All authors contributed to writing the paper.

Funding: Financial support of UNAM-DGAPA-PAPIIT IA103417 and IN109417 is acknowledged.

Conflicts of Interest: The authors declare no conflict of interest.

References

Math Ser 1951, 12, 36–38.

IEEE Trans Circuits Syst I Fundam Theory Appl 2001, 48, 281–288 [CrossRef ]

University Press: Cambridge, UK, 2004.

Luo, L.; Manning, T.A.; et al Random numbers certified by Bell’s theorem Nature 2010, 464, 1021–1024.

[ CrossRef ] [ PubMed ]

https://www.idquantique.com/random-number-generation/products/quantis-random-number-generator (accessed on 9 October 2018).

Number Generators for Cryptographic Applications; Technical Report; Booz-Allen and Hamilton Inc.: Mclean,

VA, USA, 2001.

In Proceedings of the IEEE International Symposium on Circuits and Systems (ISCAS 2007), New Orleans,

LA, USA, 27–30 May 2007; pp 1437–1440.

New York, NY, USA, 2010.

Pérez Castillo, I Improving randomness characterization through Bayesian model selection Sci Rep 2017,

7, 3096 [CrossRef ] [ PubMed ]

Modelos Master’s Thesis, Posgrado en Ciencias Físicas UNAM, Mexico, Mexico, 2017 (In Spanish)

19. Pemmaraju, S.; Skiena, S.S Computational Discrete Mathematics: Combinatorics and Graph Theory with Mathematica;

Cambridge University Press: Cambridge, UK, 2003.

Quantum Randomness arXiv 2018, arXiv:1806.08762.

Bachelor’s Thesis, Facultad de Ciencias, UNAM, Mexico, Mexico, 2017 (In Spanish)

© 2018 by the authors Licensee MDPI, Basel, Switzerland This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

Italy; hfreytes@gmail.com (H.F.); giuntini@unica.it (R.G.); giuseppe.sergioli@gmail.com (G.S.)

viale Marconi 5, I-24044 Dalmine (BG), Italy; roberto.leporini@unibg.it

Received: 28 August 2018; Accepted: 28 October 2018; Published: 31 October 2018

Abstract: Quantum computation theory has inspired new forms of quantum logic, called quantum

computational logics, where formulas are supposed to denote pieces of quantum information, while

logical connectives are interpreted as special examples of quantum logical gates The most natural

semantics for these logics is a form of holistic semantics, where meanings behave in a contextual way.

In this framework, the concept of quantum probability can assume different forms We distinguish

an absolute concept of probability, based on the idea of quantum truth, from a relative concept of probability (a form of transition-probability, connected with the notion of fidelity between quantum

states) Quantum information has brought about some intriguing epistemic situations A typicalexample is represented by teleportation-experiments In some previous works we have studied aquantum version of the epistemic operations “to know”, “to believe”, “to understand” In this article,

we investigate another epistemic operation (which is informally used in a number of interestingquantum situations): the operation “being probabilistically informed”

Keywords: quantum logics; quantum probability; holistic semantics; epistemic operations

1 Introduction

Quantum information and quantum computation have inspired new developments of some basicconcepts of the quantum theoretic formalism, which for a long time had been regarded as mysterious

and potentially paradoxical In this framework the concept of quantum probability has been investigated

according to new perspectives, giving rise to possible applications to fields that are far apart frommicrophysics (cognitive and social sciences, semantics of natural languages and of the languages ofart, see, for instance, [1 3])

As is well known, the basic idea of quantum computation theory is that information can be storedand transmitted by quantum physical objects Accordingly, pieces of quantum information can beidentified with states of some special quantum systems that are storing the information in question

In the simplest case a piece of quantum information corresponds to a pure state of a single particle:

a qubit (or qubit-state), the quantum counterpart of the classical concept of bit Mathematically aqubit can be represented as a quantum superposition (living in the two-dimensional Hilbert spaceC2),whose form is

|ψ = c0|0 + c1|1,

is it that the inherent quantum randomness does not... concept of quantum probability can assume different forms We distinguish

an absolute concept of probability, based on the idea of quantum truth, from a relative concept of probability. .. with alternate zeroes and ones (010 and 101)appear less frequently than expected, also by about 0.005% At Borel levels four and five, the largerdeviations appear for events 0101 and 1010, clearly