Annals of the International Society of Dynamic Games Volume Series Editor Tamer Basar ¸ Editorial Board Tamer Basar, University of Illinois, Urbana ¸ Pierre Bernhard, I3S-CNRS and University of Nice-Sophia Antipolis Maurizio Falcone, University of Roma “La Sapieza” Jerzy Filar, University of South Australia, Adelaide Alain Haurie, HEC-University of Geneva Arik A Melikyan, Russian Academy of Sciences, Moscow Andrzej S Nowak, Wroclaw Univeristy of Technology and University of Zielona G´ o Leo Petrosjan, St Petersburg State University Alain Rapaport, INRIA, Montpelier Josef Shina, Technion, Haifa Annals of the International Society of Dynamical Games Advances in Dynamic Games Applications to Economics, Finance, Optimization, and Stochastic Control Andrzej S Nowak Krzysztof Szajowski Editors Birkhă user a Boston ã Basel • Berlin Andrzej S Nowak Wroclaw University of Technology Institute of Mathematics Wybrze˙ e Wypia´ skiego 27 z n 50-370 Wroclaw Poland and Faculty of Mathematics, Computer Science, and Econometrics University of Zielona G´ o Podg´ rna 50 o 65-246 Zielona G´ o Poland Krzysztof Szajowski Wroclaw University of Technology Institute of Mathematics Wybrze˙ e Wypia´ skiego 27 z n 50-370 Wroclaw Poland AMS Subject Classifications: 91A-xx, 91A05, 91A06, 91A10, 91A12, 91A13, 91A15, 91A18, 91A20, 91A22, 91A23, 91A25, 91A28, 91A30, 91A35, 91A40, 91A43, 91A46, 91A50, 91A60, 91A65, 91A70, 91A80, 91A99 Library of Congress Cataloging-in-Publication Data International Symposium of Dynamic Games and Applications (9th : 2000 : Adelaide, S Aust.) Advances in dynamic games : applications to economics, finance, optimization, and stochastic control / Andrzej S Nowak, Krzysztof Szajowski, editors p cm – (Annals of the International Society of Dynamic Games ; [v 7]) Papers based on presentations at the 9th International Symposium on Dynamic Games and Applications held in Adelaide, South Australia in Dec 2000 ISBN 0-8176-4362-1 (alk paper) Game theory–Congresses I Nowak, Andrzej S II Szajowski, Krzysztof III Title IV Series HB144.I583 2000 330’.01’5193–dc22 ISBN 0-8176-4362-1 2004048826 Printed on acid-free paper c 2005 Birkhă user Boston a All rights reserved This work may not be translated or copied in whole or in part without the written permission of the publisher (Birkhă user Boston, c/o Springer Science+Business Media, Inc., Rights a and Permissions, 233 Spring Street, New York, NY, 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden The use in this publication of trade names, trademarks, service marks and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights Printed in the United States of America 987654321 www.birkhauser.com SPIN 10988183 (KeS/SB) Contents Preface Contributors Part I ix xi Repeated and Stochastic Games Information and the Existence of Stationary Markovian Equilibrium Ioannis Karatzas, Martin Shubik and William D Sudderth Markov Games under a Geometric Drift Condition Heinz-Uwe Kă enle u 21 A Simple Two-Person Stochastic Game with Money Piercesare Secchi and William D Sudderth 39 New Approaches and Recent Advances in Two-Person Zero-Sum Repeated Games Sylvain Sorin 67 Notes on Risk-Sensitive Nash Equilibria Andrzej S Nowak 95 Continuous Convex Stochastic Games of Capital Accumulation Piotr Wiecek 111 Part II Differential Dynamic Games Dynamic Core of Fuzzy Dynamical Cooperative Games Jean-Pierre Aubin 129 Normalized Overtaking Nash Equilibrium for a Class of Distributed Parameter Dynamic Games Dean A Carlson 163 183 Cooperative Differential Games Leon A Petrosjan vi Contents Part III Stopping Games Selection by Committee Thomas S Ferguson 203 Stopping Game Problem for Dynamic Fuzzy Systems Yuji Yoshida, Masami Yasuda, Masami Kurano and Jun-ichi Nakagami 211 On Randomized Stopping Games El˙bieta Z Ferenstein z 223 235 Dynkin’s Games with Randomized Optimal Stopping Rules Victor Domansky 247 Stopping Games – Recent Results Eilon Solan and Nicolas Vieille Modified Strategies in a Competitive Best Choice Problem with Random Priority Zdzisław Porosi´ ski n 263 271 Optimal Stopping Games where Players have Weighted Privilege Minoru Sakaguchi 285 Equilibrium in an Arbitration Procedure Vladimir V Mazalov and Anatoliy A Zabelin 295 Bilateral Approach to the Secretary Problem David Ramsey and Krzysztof Szajowski Part IV Applications of Dynamic Games to Economics, Finance and Queuing Theory Applications of Dynamic Games in Queues Eitan Altman 309 Equilibria for Multiclass Routing Problems in Multi-Agent Networks Eitan Altman and Hisao Kameda 343 Endogenous Shocks and Evolutionary Strategy: Application to a Three-Players Game Ekkehard C Ernst, Bruno Amable and Stefano Palombarini 369 vii Contents Robust Control Approach to Option Pricing, Including Transaction Costs Pierre Bernhard 391 S-Adapted Equilibria in Games Played over Event Trees: An Overview Alain Haurie and Georges Zaccour 417 Existence of Nash Equilibria in Endogenous Rent-Seeking Games Koji Okuguchi 445 A Dynamic Game with Continuum of Players and its Counterpart with Finitely Many Players Agnieszka Wiszniewska-Matyszkiel Part V 455 Numerical Methods and Algorithms for Solving Dynamic Games Distributed Algorithms for Nash Equilibria of Flow Control Games Tansu Alpcan and Tamer Basar ¸ A Taylor Series Expansion for H ∞ Control of Perturbed Markov Jump Linear Systems Rachid El Azouzi, Eitan Altman and Mohammed Abbad Advances in Parallel Algorithms for the Isaacs Equation Maurizio Falcone and Paolo Stefani 473 499 515 Numerical Algorithm for Solving Cross-Coupled Algebraic Riccati Equations of Singularly Perturbed Systems Hiroaki Mukaidani, Hua Xu and Koichi Mizukami 545 Equilibrium Selection via Adaptation: Using Genetic Programming to Model Learning in a Coordination Game Shu-Heng Chen, John Duffy and Chia-Hsuan Yeh 571 Two Issues Surrounding Parrondo’s Paradox Andre Costa, Mark Fackrell and Peter G Taylor Part VI 599 Parrondo’s Games and Related Topics State-Space Visualization and Fractal Properties of Parrondo’s Games Andrew Allison, Derek Abbott and Charles Pearce 613 viii Contents Parrondo’s Capital and History-Dependent Games Gregory P Harmer, Derek Abbott and Juan M R Parrondo 635 Introduction to Quantum Games and a Quantum Parrondo Game Joseph Ng and Derek Abbott 649 A Semi-quantum Version of the Game of Life Adrian P Flitney and Derek Abbott 667 Preface Modern game theory has evolved enormously since its inception in the 1920s in the works of Borel and von Neumann The branch of game theory known as dynamic games descended from the pioneering work on differential games by R Isaacs, L S Pontryagin and his school, and from seminal papers on extensive form games by Kuhn and on stochastic games by Shapley Since those early developmental decades, dynamic game theory has had a significant impact on such diverse disciplines as applied mathematics, economics, systems theory, engineering, operations research, biology, ecology, and the environmental sciences On the other hand, a large variety of mathematical methods from differential equations to stochastic processes has been applied to formulate and solve many different problems This new edited book focuses on various aspects of dynamic game theory, providing authoritative, state-of-the-art information and serving as a guide to the vitality of the field and its applications Most of the selected, peer-reviewed papers are based on presentations at the 9th International Symposium on Dynamic Games and Applications held in Adelaide, South Australia in December 2000 This conference took place under the auspices of the International Society of Dynamic Games (ISDG), established in 1990 The conference has been cosponsored by Centre for Industrial and Applicable Mathematics (CIAM), University of South Australia, IEEE Control Systems Society, Institute of Mathematics, Wrocław University of Technology (Poland), Faculty of Mathematics, Computer Science and Econometrics, University of Zielona G´ (Poland), ISDG Organizing Society, o and the University of South Australia Every paper that appears in this volume has passed through a stringent reviewing process, as is the case with publications for archival journals A variety of topics of current interest are presented They are divided in to six parts: the first (five papers) treat repeated games and stochastic games, and the second (three papers) covers differential dynamic games The third part of the volume (nine papers) is devoted to the various extensions of stopping games, which are also known as Dynkin’s games In the fourth part there are seven papers on applications of dynamic games to economics, finance, and queuing theory The final two parts contain five papers which are devoted to algorithms and numerical solution approaches for dynamic games, and the section on Parrondo’s games (five papers) We wish to thank all the associate editors and the referees for their valuable contributions that made this volume possible Wrocław and Zielona G´ o Wrocław Andrzej S Nowak Krzysztof Szajowski Contributors Mohammed Abbad, D´ partement de Mathematiques et Informatique, Facult´ des e e Sciences B.P 1014, Universit´ Mohammed V, 10000 Rabat, Morocco e Derek Abbott, Centre for Biomedical Engineering (CBME) and Department of Electrical and Electronic Engineering, The University of Adelaide, Adelaide, SA 5005, Australia Andrew Allison, Centre for Biomedical Engineering (CBME) and Department of Electrical and Electronic Engineering, University of Adelaide, Adelaide, SA 5005, Australia Tansu Alpcan, Coordinated Science Laboratory, University of Illinois, 1308 West Main Street, Urbana, IL 61801, USA Eitan Altman, INRIA, B.P 93, 2004 route des Lucioles, 06902 Sophia-Antipolis Cedex, France Bruno Amable, Facult´ des Sciences Economies, Universit´ Paris X-Nanterre, e e 200 av de la R´ publique, 92000 Nanterre, France e Jean-Pierre Aubin, Centre de Recherche Viabilit´ , Jeux, Contrˆ le, Universit´ e o e Paris-Dauphine, 75775 Paris cx (16), France Rachid El Azouzi, University of Avignon, LIA, 339, chemin des Meinajaries, Agroparc B.P 1228, 84911 Avignon Cedex 9, France Tamer Basar, Coordinated Science Laboratory, University of Illinois, 1308 West ¸ Main Street, Urbana, IL 61801, USA Pierre Bernhard, Laboratoire I3S, UNSA and CNRS, 2000 route des Lucioles, Les Algorithmes – bˆ t Euclide 8, BP.121, 106903 Sophia Antipolis-Cedex, France a Dean A Carlson, Mathematical Reviews 416 Fourth Street, P.O.Box 8604, Ann Arbor, MI 48107-8604, USA Shu-Heng Chen, AI-ECON Research Center, Department of Economics, National Chengchi University, 64 Chi-Nan Rd., Sec.2, Taipei 11623, Taiwan Andre Costa, School of Applied Mathematics, University of Adelaide, Adelaide, SA 5005, Australia 656 J Ng and D Abbott ⎡ ⎤ sin(θA ) cos(θA ) ⎢ cos(θA ) sin(θA ) ⎥ ⎥, =⎢ ⎣− sin(θA ) cos(θA ) ⎦ cos(θA ) − sin(θA ) and for the second qubit, ˆ ˆ (θ) = I ⊗ ˆ (θ ) cos(θA ) ⊗ − sin(θA ) ⎡ cos(θA ) sin(θA ) ⎢− sin(θA ) cos(θA ) =⎢ ⎣ 0 0 = sin(θA ) cos(θA ) ⎤ 0 0 ⎥ ⎥ cos(θA ) sin(θA ) ⎦ − sin(θA ) cos(θA ) So the total system, |ψ , becomes |ψ = ˆ ˆ |ψ1 ψ2 , which is a superposition of the base kets i.e a|T T + b|T H + c|H T + d|H H As before, |a|2 , |b|2 , |c|2 , |d|2 represent the classical probability of obtaining the respective states if we measure the system and |a|2 + |b|2 + |c|2 + |d|2 = So what does it all mean if we find that the system is in the state, say, |T H ? Now the T represents the 1st qubit and the H , the 2nd qubit As we have defined qubit as the result of the 1st toss, and qubit as the 2nd, |T H means that we have a tail at the 1st toss, followed by a head It gives us the toss history of the set of games So, if a head is considered a win, and tails, a loss, then |b|2 is the probability of losing the first game, and then winning the second game5 Simulating Game B For Game B, we employ a very similar approach to Game A However, the difference is that we will now use a Controlled-Controlled-Rotation (CCRot) matrix A CCRot gate is a 3-qubit gate, where the 3rd bit is rotated when the first qubits are (Fig 6) The truth table of a CCRot gate is given in Table 2: Table 2: The truth table of a CCRot gate The rotation is given by θ |q0 |0 |0 |1 |1 |q1 |0 |1 |0 |1 |q2 |0 |0 |0 cos(θ )|0 + sin(θ )|1 ) From now on, we shall represent Heads as 1, and Tails as 0, i.e |00 ≡ |T T , |11 ≡ |H H etc Introduction to Quantum Games and a Quantum Parrondo Game 657 Figure 6: A Controlled-Controlled-Rotation gate/matrix If qubit 2, |q2 , starts off in the |0 state, it will be rotated if both |q0 and |q1 are in the |1 state In matrix form, this is |000 |001 |010 |011 |100 |101 |110 |111 ⎡ ⎢0 ⎢ ⎢0 ⎢ ⎢0 ⎢ ⎢0 ⎢ ⎢0 ⎢ ⎣0 0 0 0 0 0 0 0 0 0 0 0 ⎤ 0 0 0 ⎥ ⎥ 0 ⎥ ⎥ 0 ⎥ ⎥ 0 ⎥ ⎥ 0 ⎥ ⎥ cos(θB ) sin(θB ) ⎦ − sin(θB ) cos(θB ) 0 0 0 The CCRot matrix is perfect for what we need to because in Game B of the HD game, the choosing of the coin for the 3rd toss (qubit) is dependent on the previous results (qubits) But Game B is a little more than the above As mentioned earlier, the above matrix will rotate the 3rd bit if the first qubits are In our context, this means that the state of the system is only changed if we won the previous games, i.e this simulates choosing and tossing coin in Game B What we need is to obtain other variations of the CCRot gate to simulate the other coins for the different possible histories So for coins 2, and 4, their respective matrices are: ⎡ ˆ win,win ⎢0 ⎢ ⎢0 ⎢ ⎢0 =⎢ ⎢0 ⎢ ⎢0 ⎢ ⎣0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ⎤ 0 0 0 ⎥ ⎥ 0 ⎥ ⎥ 0 ⎥ ⎥, 0 ⎥ ⎥ 0 ⎥ ⎥ cos(θB4 ) sin(θB4 ) ⎦ − sin(θB4 ) cos(θB4 ) 658 J Ng and D Abbott ⎡ ˆ win,lose ˆ lose,win ˆ lose,lose 0 0 ⎢0 0 0 ⎢ ⎢0 0 ⎢ ⎢0 0 0 =⎢ ⎢0 0 cos(θB ) sin(θB ) 3 ⎢ ⎢0 0 − sin(θB ) cos(θB ) 3 ⎢ ⎣0 0 0 0 0 0 ⎡ 0 0 ⎢0 0 0 ⎢ ⎢0 cos(θB ) sin(θB ) 0 3 ⎢ ⎢0 − sin(θB ) cos(θB ) 0 3 =⎢ ⎢0 0 ⎢ ⎢0 0 0 ⎢ ⎣0 0 0 0 0 0 ⎡ cos(θB2 ) sin(θB2 ) 0 0 ⎢− sin(θB ) cos(θB ) 0 0 2 ⎢ ⎢ 0 0 ⎢ ⎢ 0 0 =⎢ ⎢ 0 0 ⎢ ⎢ 0 0 ⎢ ⎣ 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ⎤ 0⎥ ⎥ 0⎥ ⎥ 0⎥ ⎥, 0⎥ ⎥ 0⎥ ⎥ 0⎦ ⎤ 0⎥ ⎥ 0⎥ ⎥ 0⎥ ⎥, 0⎥ ⎥ 0⎥ ⎥ 0⎦ ⎤ 0⎥ ⎥ 0⎥ ⎥ 0⎥ ⎥ 0⎥ ⎥ 0⎥ ⎥ 0⎦ Now Game B is obtained by putting these CCRot matrices one after another (Fig 7) i.e ˆB = ˆ0ˆ1ˆ2ˆ3 ⎡ ⎤ B2 ⎢ ⎥ ⎢ B3 ⎥ ⎥ =⎢ ⎢ ⎥, B 0⎦ ⎣ B4 Figure 7: Combining the different rotation matrices to form the Game B quantum gate A solid dot means that the control qubit must be in the |1 state for the gate to rotate the target qubit An open circle means that the control qubit must be in the |0 state Introduction to Quantum Games and a Quantum Parrondo Game 659 where B2 = cos(θB2 ) sin(θB2 ) , − sin(θB2 ) cos(θB2 ) B3 = cos(θB3 ) sin(θB3 ) , − sin(θB3 ) cos(θB3 ) B4 = cos(θB4 ) sin(θB4 ) − sin(θB4 ) cos(θB4 ) As the ˆ i matrices commute with each other, the order is not important All we need to is to vary the amount of rotation for each ˆ i , this gives us the required matrix for simulating Game B which we will denote by ˆ B 5.1 Combining Games A and B To combine the two games, all we need to is to decide on the number of games, create the correct rotation matrices for these games, and then apply these matrices to an initial state ket, |ψ0 = |00 For example, to play two games of Game A and a game of Game B, the final state of the system is |ψf = ˆ ˆ ˆ |000 (see Fig 8), which will be a superposition of all possible outcomes, so |ψf = a|000 + b|001 + c|010 + + h|111 This means that we can now plot a graph of probability vs outcome, and thus work out the most likely histories if we play the game infinite times (Fig 9) Fig 10 shows the results for playing two games of A’s followed by two games of B’s followed by two games of A’s etc for 10 games Figure 8: A 3-toss game, where we play two games of A followed by one game of B Discussion What we have done so far is essentially the same as calculating each of the possible outcomes by multiplying the respective probabilities So what we have here is a quantum game that produces classical results, where the two losing games combine to create a winning game (Fig 11) It is a quantum simulation of a classical system However, on a classical computer, to calculate every possible history for n games 660 J Ng and D Abbott Game Sequence = AAB 0.25 0.2 Probability 0.15 0.1 0.05 0 History Figure 9: Classical probability vs history plot of playing two games of A’s followed by one game of B The indices are binary strings converted into decimals For example, the bar for (010) is the probability of losing the first game, winning the second, and losing the third Game Sequence = ABABABABAB 0.014 0.012 Probability 0.01 0.008 0.006 0.004 0.002 0 100 200 300 400 500 History 600 700 800 900 1000 Figure 10: Probability vs history plot of playing 10 games These games are played by playing two games of A’s, followed by two games of B’s and so on 661 Introduction to Quantum Games and a Quantum Parrondo Game Game Sequence = ABABABABABABABA 0.3 Periodic 0.2 0.1 Capital 0.1 Game A 0.2 0.3 0.4 0.5 0.6 Game B Games Played 10 12 14 Figure 11: When we combine the two losing games (Game A and Game B) in a periodic fashion (alternate game A and B), a winning game results This result is identical to that of classical simulations [7] require 2n bits On a quantum computer, on the other hand, only n qubits are required: an improvement of log(n) The rotation matrices are 2n x2n because it is a classical way of representing, calculating and simulating quantum processes Looking back at Fig 2, it is natural to ask what happens if we extend the axes to allow the coefficients of |0 and |1 , a and b, to be negative and/or complex? These complex amplitudes are taken into account by a phase factor eiφ which is inserted into the rotation matrices So the basic rotation matrix, ˆ (θ ), becomes iφ sin(θ) ˆ (θ, φ) = e cos(θ) −iφ sin(θ) e cos(θ) Since the actual probabilities are the lengths of the state kets, these complex amplitudes will still produce classical results under normal circumstance However, if made to interact with each other, they can produce radically different results Two probability amplitudes of the same magnitude and different signs can cancel each other out, resulting in destructive interference This will not happen in classical game theory because classical probabilities are always real and positive So how we cause these quantum probability amplitudes to interact? Eisert’s approach [2] was to employ an entangling gate, J , which calculates the payoff of the two parties By varying the entangling parameter (which is essentially a phase parameter) in J , Eisert’s result showed that the classical problem of Prisoner’s dilemma is a subset of the quantum game, and there is no longer a dilemma when the game is fully explored in the quantum regime 662 J Ng and D Abbott Parrondo’s game can be seen as a competition between two players: Casino (C) and Parrondo (P) Both Parrondo’s and Casino’s aim is to maximize the winnings or minimize the losses As mentioned above, despite the Casino’s Game A and Game B being originally unfairly biased against Parrondo, he can construct a combined winning game by playing the games in certain sequences However, one of the Casino’s business strategists has read Parrondo’s published paper, and brought the issue up at the next Casino board meeting At that meeting, it was decided that the Casino should employ quantum mechanics to help them turn the tables back in their favor This was done by implementing a Casino Gate (Fig 12), C(φ), cos(γ ) sin(γ ) For and a “de-Casino” gate C † (φ), where C(φ) = C † (φ) = − sin(γ ) cos(γ ) n games, the resultant state is |ψf = C † (γ )G(n)C(γ )|0 ⊗n where G(n) is the collection of quantum gates that describes the sequence of games played This can be thought of as the player walking through the doors from the classical world into the quantum Casino, and later, from the Casino back into the classical world By setting γ = π/2, Game A remains the same but Game B becomes a winning Figure 12: The Casino adopts a quantum strategy by employing entangling gates, J (γ ) The degree of entanglement is determined by γ , with γ = representing no entanglement, and γ = π/2 representing maximum entanglement Figure 13: The player also adopts a quantum strategy by employing phase gates, P (φ) Introduction to Quantum Games and a Quantum Parrondo Game 663 game, yet through Parrondo’s strategy, the combined game is now a losing game (Fig 14) Interestingly, although Game B wins faster than Game A, the combined game is still losing In fact, it loses faster than just playing Game A on its own This is a different paradox to the original! AAAAAAAA Capital 10 C.Casino,C.Player C.Casino,Q.Player Q.Casino,C.Player Q.Casino,Q.Player 5 BBBBBBBB Capital C.Casino,C.Player C.Casino,Q.Player Q.Casino,C.Player Q.Casino,Q.Player 2 AAABBAAA Capital 10 C.Casino,C.Player C.Casino,Q.Player Q.Casino,C.Player Q.Casino,Q.Player 5 ABABABAB Capital 10 C.Casino,C.Player C.Casino,Q.Player Q.Casino,C.Player Q.Casino,Q.Player 5 Games Played Figure 14: Plotting Capital vs Games for varying γ and φ and varying game sequences (as labeled) Classical Casino uses γ = 0, Quantum Casino uses γ = π/2 Classical Parrondo uses φ = 0, Quantum Parrondo uses φ = π/2 As can be seen, for a classical casino, the results are the same regardless of whether Parrondo uses a quantum coin or not However, it did not take long for Parrondo to realise that this sudden change of fortune is not simply a statistical abnormality, but rather, due to the Casino’s quantum strategy So he decides to beat the Casino at their own game again, and adopts a quantum strategy as well This is done with a phase-shift gate (Fig 13), P (φ) = eiφ Now, the resultant state is |ψf = C † (γ )G(n)P (φ)C(γ )|0 ⊗n This e−iφ causes both Game A and Game B to become winning games, and combine to create a winning overall game So when both the Casino and Parrondo adopt quantum strategies, the original Parrondo game is no longer a paradox So as can be seen, Parrondo’s best strategy lies in employing a quantum strategy He is guaranteed to win regardless of the strategy used by the Casino (Fig 15) 664 J Ng and D Abbott Unfortunately (or fortunately, depending on how one prefers to see the situation), the same cannot be said for the Casino however If the casino adopts a quantum strategy, Parrondo can choose a classical strategy and play only the winning Game B or a quantum strategy and still win AAAAAAAA BBBBBBBB 10 10 5 0 5 2 Player (φ) 0 1 Player (φ) Casino (γ) AABBAABB 0 Casino (γ) ABABABAB 10 10 5 0 5 2 Player (φ) 0 Casino (γ) 1 Player (φ) 0 Casino (γ) Figure 15: Plotting the capital after playing games as labeled The axes are plotted from to 2π γ is the casino’s parameter, while φ is Parrondo’s parameter At γ = φ = 0, we have an entirely classical game In fact, for all γ = 0, the results are the same as classical results, so in the plots, we have a straight line at γ = 0, regardless of what φ is Conclusion Parrondo’s games are of general interest as they illustrate how two losing coin tossing games can win when combined either in deterministic or non-deterministic sequences For this phenomenon to occur, there must be coupling between the games In Section we saw that the CD games couple via capital-based statedependence and the HD games couple via history-based state-dependence The open question is, can a quantum Parrondo game be designed such that the coupling is via quantum entanglement? For the case of non-deterministic sequences of games A and B, Game A can be thought of as “noise” that breaks up the state-dependent rules that are biasing Introduction to Quantum Games and a Quantum Parrondo Game 665 Game B to lose – and this is why the combination of A and B wins (“the Boston Interpretation”) So another open question for quantum Parrondo games is, can the effect of Game A be in fact replaced by some form of decoherence such as a measurement? Acknowledgements We would like to thank Gerard Milburn, Bill Munro, Ben Travaglione and Michael Nielsen of the SRC for Quantum Computer Technology, University of Queensland for all the inspirational and educational discussions over the course of this work Thanks are also due to Wanli Li, Dept of Physics, Princeton, for a number of manuscript suggestions Funding from GTECH and the Sir Ross and Sir Keith Smith Fund is gratefully acknowledged REFERENCES [1] Benjamin S C and Hayden P M., Multi-player quantum games Phys Rev A 64:030301(R), 2001 See also LANL Preprint quant-ph/0007038 [2] Eisert J., Wilkens M and Lewenstein M., Quantum games and quantum strategies Phys Rev Lett., 83:3077, 1999 See also LANL Preprint quantph/9806088 [3] Harmer G P and Abbott D., Losing strategies can win by Parrondo’s paradox Nature (London), 402:864, 1999 [4] Li C-F., Zhang Y-S., Huang Y-F and Guo G-C., Quantum strategies of quantum measurement Phys Lett A 280:257, 2001 See also LANL Preprint quant-ph/0007120 [5] Marinatto L and Weber T., A quantum approach to static games of complete information Phys Lett A 272:291, 2000 See also LANL Preprint quantph/0004081 [6] Meyer D.A., Quantum strategies Phys Rev Lett., 82:1052, 1999 See also LANL Preprint quant-ph/9804010 [7] Parrondo J M R., Harmer G P and Abbott D., New paradoxical games based on brownian ratchets Phys Rev Lett., 85(24):5226–5229, December 2000 See also LANL Preprint cond-mat/0003386 A Semi-quantum Version of the Game of Life Adrian P Flitney Centre for Biomedical Eng (CBME) Department of Electrical and Electronic Engineering The University of Adelaide Australia aflitney@physics.adelaide.edu.au Derek Abbott Centre for Biomedical Engineering (CBME) Department of Electrical and Electronic Engineering The University of Adelaide Australia dabbott@eleceng.adelaide.edu.au Abstract A version of John Conway’s game of Life is presented where the normal binary values of the cells are replaced by oscillators which can represent a superposition of states The original game of Life is reproduced in the classical limit, but in general additional properties not seen in the original game are present that display some of the effects of a quantum mechanical Life In particular, interference effects are seen Key words Cellular automata, quantum games, quantum cellular automata AMS Subject Classifications Primary 68Q80; Secondary 37B15 Introduction John Conway’s game of Life [10] is a well-known two-dimensional cellular automaton where cells are arranged in a square grid and have binary values generally known as dead or alive The status of the cells change in a discrete fashion, each “generation” depending upon the number of neighboring cells that are alive, the general idea being that a cell dies if there is either overcrowding or isolation There are many different rules that can be applied for birth or survival of a cell and a number of these give rise to interesting properties such as still lives (stable patterns), oscillators (patterns that periodically repeat), spaceships or gliders (fixed shapes that move across the Life universe), glider guns, and so on [3,12,11] Conway’s original rules are some of the few that are balanced between survival and extinction of the Life “organisms.” In this version a dead (or empty) cell 668 A P Flitney and D Abbott d   = alive p = empty or dead d   d d   p   d   (i) block (a) dd    dd    (ii) tub d   p   p d d   (b) initial p ddd     p d   p   p d d   1st gen 2nd gen p p    dd p p    dd p p    dd p p dd    p p d   d p p p d   p p p   d p p    dd blinker (c) d   d p p p p p d   p p p   d p p    dd initial 1st gen 2nd gen beacon Figure 1: A small sample of the simplest structures in Conway’s Life: (a) the simplest still-lives (stable patterns), the block and the tub, and the simplest oscillators (periodic patterns), (b) the blinker and (c) the beacon, both of period two A number of blocks and blinkers will normally evolve from any moderate-sized random collection of alive and dead cells becomes alive if it has exactly three living neighbors, while an alive cell survives if and only if it has two or three living neighbors Much literature on the game of Life and its implications exists For a recent discussion on the possibilities of this and other cellular automata the interested reader is referred to reference [24] The simplest still lives and oscillators are given in figure 1, while figure shows a glider, the simplest and most common moving form A large enough random collection of alive and dead cells will, after a period of time, usually decay into a 669 A Semi-quantum Version of the Game of Life p   p d p   p d d   p p dd    p p p p p initial p p p d   p p    dd p    p dd p p p p 1st gen p p p p p d d   p   p p    dd p p p d   3rd gen p p   p d p p p   d p ddd     p p p p 2nd gen p p p p p p p d   p p d d     p p dd    4th gen Figure 2: In Conway’s Life, the simplest spaceship (a pattern that moves continuously through the Life universe), the glider The figure shows how the glider moves one cell diagonally over a period of four generations collection of still lives and oscillators like those shown here while firing a number of gliders off towards the outer fringes of the Life universe The recent interest in quantum games [1,2,4,6,7,9,16–20] suggests the possibility of applying the idea of superposition of states in quantum mechanics to the game of Life Unfortunately Conway’s Life is irreversible while, in the absence of a measurement, quantum mechanics is reversible In particular, operators that represent measurable quantities must be unitary A full quantum Life would be problematic given the known difficulties of quantum cellular automata [21] Recently, in an attempt to generalize von Neumann’s universal constructor [22] to quantum mechanics, it was found that a quantum universal constructor capable of selfreproduction cannot exist with finite resource in a deterministic universe [23] This could have important bearing in understanding life from a quantum theoretic viewpoint Interesting behavior can still be obtained in a semi-quantum mechanical Life by representing the cells by classical sine-wave oscillators with a period equal to one generation, an amplitude between zero and one, and a variable phase The amplitude of the oscillation represents the coefficient of the alive state so that the square of the amplitude gives the probability of finding the cell in the alive state when a measurement of the “health” of the cell is taken If the initial state of the system contains at least one cell that is in a superposition of eigenstates the neighboring cells will be influenced according to the coefficients of the respective eigenstates, propagating the superposition to the surrounding region If the coefficients of the superpositions are restricted to positive real numbers we not expect to see qualitatively new phenomena By allowing the coefficients to be complex, that is, by allowing phase differences between the oscillators, qualitatively new phenomena, for example interference effects, may arise The inter- 670 A P Flitney and D Abbott ference effects we see are those due to an array of classical oscillators with phase shifts and are not fully quantum mechanical Our cellular automaton should be distinguished from quantum cellular automata discussed in references [5,8,13–15] A First Model To represent the state of a cell we introduce the following notation: |ψ = a|alive + b|dead , (1) subject to the normalization condition |a|2 + |b|2 = (2) |a|2 and |b|2 represent the probabilities of measuring the cell as alive or dead respectively If the values of a and b are restricted to non-negative real numbers we cannot get destructive interference The model still differs from a classical probabilistic mixture since it is the amplitudes that are added and not the probabilities In our model |a| is the amplitude of the oscillator Restricting a to non-negative real numbers corresponds to the oscillators all being in phase The birth, death and survival operators have the following effects B|ψ = |alive , (3) D|ψ = |dead , S|ψ = |ψ A cell can be represented by the vector a b The B and D operators are not unitary Indeed they can be represented in matrix form by B∝ 1 , 0 D∝ 0 , 1 (4) where the proportionality constant is not relevant for our purposes After applying B or D (or some mixture) the new state will require (re-) normalization so that the probabilities of being dead or alive still sum to unity | is the standard quantum mechanical notation to be read as “the state of ” ... to check There are two existence results in [2], Theorems 7. 1 and 7. 2, that not rely on such an assumption Here we present the analogue of the second of them α Theorem 6.2 Suppose that the variables... nothing, whereas the other half receive an income of units of money, all of which they pay back to the bank since the interest rate is r1 = As there are no lenders, the books balance Theorem 6.1 now... and are owned by the individual agents However, the agents are required to offer the goods in the market, and not receive the proceeds until the start of the subsequent period The assumption that