5D Library › Evolutionary Game Theory, Direct and Indirect Reciprocity and the Emergence of Cooperation · 中文
Preface: Why Cooperation Has Not Been Selected Away
Note
Please read this first. This book introduces mathematical models from evolutionary game theory on how cooperation arises and is maintained, together with the experimental evidence, and uses short programs to make its arguments. Unless a source is given, the populations, strategies, teams, communities and data in the book are fictional or synthesised by the programs, and the parameters are illustrative. The book discusses mathematical models; it is not moral instruction and not business, management, investment or legal advice, and it makes no forecasts about any company, community or market.
“Fitness”, “replication” and “selection” in evolutionary games originally describe changes in the frequency of genes or strategies in a population; applying them to human learning, imitation and institutions is a matter of models and analogies. “Altruism” in biology is a definition in terms of fitness and says nothing about motives or moral worth; the Confucian and Mohist ethics read in Chapter 7 are moral philosophy, which is a different thing from calculating gene frequencies.
In 1964 W. D. Hamilton posed a question: natural selection favours individuals that leave more offspring, so how can behaviour that costs the actor and benefits others spread through a population? His answer was that when the beneficiary is a close enough relative of the actor, helping a relative is also a way of copying one’s own genes (Hamilton 1964). Seven years later Robert Trivers argued that unrelated individuals can also sustain cooperation through exchanges of “you help me and I help you” (Trivers 1971). Around 1980 Robert Axelrod carried the question into political science and economics with two computer tournaments (Axelrod 1980a, 1980b; Axelrod and Hamilton 1981).
This book asks four questions. First, in one-shot anonymous interactions with no additional mechanism, why are cooperators eliminated? Second, which mechanisms allow cooperation to survive evolution, and on what assumptions does each depend? Third, how do these mechanisms fare in real experiments with people: which predictions have been confirmed and which have not? Fourth, can these ideas be tested in a pre-registered controlled experiment?
What this book covers
Chapter 1 derives the replicator equation, compares the phase portraits and basins of attraction of the prisoner’s dilemma, the hawk–dove game and the stag hunt, and discusses fixation probabilities in finite populations; Chapter 2 discusses Hamilton’s rule, the Price equation, the five mechanisms summarised by Nowak and the disputes surrounding them; Chapter 3 discusses Axelrod’s tournaments, execution errors, generous tit for tat, win-stay, lose-shift and zero-determinant strategies; Chapter 4 discusses image scoring and second-order reputation norms in indirect reciprocity; Chapter 5 discusses spatial reciprocity on networks and two large experiments with human networks; Chapter 6 discusses altruistic punishment, antisocial punishment and Ostrom’s design principles in public goods games. Chapter 7 reads these studies beside passages from the Mencius, the Mozi and the Analects on “graded love”, “impartial care and mutual benefit” and “repaying resentment with uprightness”, and states where classical ethics ends and modern evolutionary theory begins. Chapter 8 offers a “four-step checklist for cooperation and protection against extortion” and a pre-registered cluster-randomised controlled experiment.
The book borders on several other volumes of 5D Library, which it only recalls briefly without repeating their programs. Chapter 3 of Volume 16, Cross-Cultural Contracts, Real Options and Game Theory of Southeast Asian Market Entry, discussed relational contracts, in which “the value of future cooperation outweighs the temptation to break a promise once”, the same idea as the probability of future interaction in Chapters 2 and 3 here; its Chapter 2 also discussed the reputation model of Kreps and Wilson. Chapter 1 of Volume 27, Mechanism Design, Incentive Compatibility and the Game Theory of Negotiation, discussed the experimental finding that voluntary contributions to public goods decline over rounds, and the punishment experiments of Chapter 6 continue from there. Volume 25, Social Network Topology, Structural Holes and Complex Contagion, discussed network structure and spreading; Chapter 5 here deals with a different process on networks, the imitation and selection of strategies. Volume 32, Algorithmic Game Theory, Braess’s Paradox and the Price of Anarchy, discussed Nash equilibria and potential games. Volume 14, Dynamical Systems of Neurotransmitters and Endocrinology, overlaps with this book only in one study of testosterone and the ultimatum game cited in its Chapter 3, which this book does not take up.
Claims corrected after checking
Several statements in the author’s original framework needed correction or qualification after checking. First, the replicator equation is given as in the framework; the dynamics was introduced by Taylor and Jonker in 1978, and the name “replicator equation” comes from a 1983 paper by Schuster and Sigmund; the concept of an evolutionarily stable strategy (ESS) comes from Maynard Smith and Price’s paper of 1973, but that paper used computer simulation to compare five strategies of “limited war”, and the two-strategy hawk–dove game is a later simplification. “In one-shot anonymous games defectors eliminate cooperators” holds only for the prisoner’s dilemma; in the hawk–dove game the two strategies coexist, and in the stag hunt the outcome depends on the initial proportions. Second, the five thresholds summarised by Nowak in 2006 (, , , , ) are correctly stated, but they come from five different models with different assumptions: is the condition for tit for tat to be an ESS against always defect, and the thresholds for “risk dominance” and for the one-third law in finite populations rise to and ; holds only under death–birth updating, weak selection and , the exact threshold for a finite population is , and under birth–death updating a network never favours cooperation; requires weak selection and rare group splitting. The “five mechanisms” are not an agreed classification: kin selection and group selection are often regarded as two mathematical descriptions of the same process, and the criticism of inclusive fitness theory by Nowak, Tarnita and Wilson in 2010 drew a reply signed by 137 authors. Price’s 1970 paper gives the covariance formula; the multilevel form that splits change into a “between-group” and a “within-group” part comes from his 1972 paper. Third, the paper by Axelrod and Hamilton in Science in 1981 is a synthesis of the tournaments with biology; the results of the two tournaments were published in Axelrod’s two papers of 1980 (14 programs in the first, 62 in the second). Tit for tat can never score more than its opponent in any single match, and its victory depends on the field of entries. Under 1 % execution errors two tit-for-tat players do fall into a cycle of retaliation; their long-run average payoff drops to the average of the four outcomes whatever the size of the error rate, and fewer errors only mean that it takes longer to get there. The forgiveness probability of generous tit for tat is the upper limit that still resists always defect (Molander’s “optimal level of generosity” is this boundary), not a target to adopt as it stands: at exactly that value always defect ties with it, and with noise it gains slightly, so in practice the value should stay a little below it; the general formula is , which is 1/3 for Axelrod’s payoffs (5, 3, 1, 0). Win-stay, lose-shift (“Pavlov”) resists always defect only if , that is in the donation game; with Axelrod’s payoffs the two sides are exactly equal. Zero-determinant extortion strategies never lose a single match but do not prevail in evolution in large populations. Fourth, Nowak and Sigmund’s image score of 1998 rises with each act of giving and falls with each refusal, so refusing to help someone with a bad image also costs points; this flaw was pointed out by Leimar and Hammerstein (2001) and by Panchanathan and Boyd (2003). The “leading eight” norms were found by Ohtsuki and Iwasa in 2004 and analysed in 2006; what they share is that helping a good individual is good, refusing a good individual is bad, refusing a bad individual is good (justified punishment), and a bad individual who helps a good one can become good again (forgiveness). “Refusing to help a bad individual costs nothing” holds for all eight norms; “helping a bad individual is judged bad” holds only for some of them, such as stern judging, while under simple standing helping a bad individual is still good. All these conclusions assume that everyone shares the same public assessments; when assessments are private and noisy, cooperation can collapse. Fifth, Nowak and May’s 1992 model is the deterministic rule “imitate the highest scorer in the neighbourhood”; for the rule of Ohtsuki and colleagues (2006) see the second point. The two human network experiments are reported as in the framework: Gracia-Lázaro and colleagues (2012) had 1,229 secondary-school students play the prisoner’s dilemma on a lattice of 625 and a heterogeneous network of 604, and cooperation did not differ between the two networks; Rand and colleagues (2014) found that cooperation in fixed networks remained stable when . Rand and colleagues considered that the parameters of the first experiment also satisfied ; the disagreement is unresolved, and the book reports both side by side. Sixth, Fehr and Gächter’s experiment of 2002 had 240 participants in groups of four, regrouped every period; the experiment of Herrmann, Thöni and Gächter (2008) was run in 16 cities in 15 countries (Switzerland contributed two, St. Gallen and Zurich), with 1,120 university students in fixed groups of four over 10 periods; with punishment, average contributions ranged from 5.7 in Athens to 18.0 in Boston (out of 20), and the cooperation-enhancing effect of punishment was at least weakly significant in 11 of the 16 participant pools (not significant in the other five). The experiment of Gächter, Renner and Sefton (2008) showed that the net benefit of punishment depends on the length of interaction: with 10 periods groups with punishment earned less net, with 50 periods they earned more. Among Ostrom’s eight design principles, “monitoring” and “graduated sanctions” are two separate principles; the principles were induced from long-enduring common-pool resource institutions, and a 2010 review of 91 studies found them reasonably well supported. Seventh, the classical texts: the three quotations from “Jin Xin I”, “Universal Love II” and “Xian Wen” and their chapter titles are correct; the phrase “repay resentment with kindness” also appears in Chapter 63 of the Laozi, and the “someone” in the Analects may have had this view in mind; Zhu Xi explained “uprightness” as “in what one loves and hates, takes and rejects, being entirely impartial and without selfishness”, not “an eye for an eye”. Eighth, the Chapter 8 trial distinguishes “not contributing towards cooperators” from “justifiably not contributing towards defectors” in a public goods game, but a contribution to a public good goes to the whole group and cannot be aimed at one person; the protocol therefore splits each community at random into groups of three every round, and withholding counts as justified only when both other group members are labelled “bad” at the start of the round. The book uses this single rule throughout: the simple-standing rule of Chapter 4 with the strictest form of justification, under which withholding from someone in bad standing is good, withholding from someone in good standing is bad, and contributing to someone in bad standing costs nothing. First-order image scoring does not distinguish justified from unjustified withholding, so this rule is not image scoring. Each chapter states these points in place, and they are collected in the editorial notes published with the book.
How to read the boxes
As in other volumes of 5D Library, different kinds of statement are marked separately.
Evidence
Evidence boxes report published research findings, with sources.
Calculation
Calculation boxes give formulas and short Python programs. Every program runs unchanged in a standard Python 3 environment using only the standard library, and its actual output is printed beneath it. The “populations”, “strategies”, “communities” and data in the book are fictional or synthesised by the programs; parameters are illustrative values, not estimates for any real system.
Hypothesis · not established
Hypothesis boxes contain the author’s ideas, are marked “not yet established”, and state what result would show them to be wrong.
Source text
Source boxes quote classical texts, with the author’s explanation.
Note
Note boxes give reminders and explain choices.
About this edition
This is version 1.0.0, the first complete edition: a preface and eight chapters, in Chinese and English. The book contains 15 programs and 4 vector figures. The text, boxes and code were drafted with an AI writing tool from the author’s title, framework and notes; the author is responsible for reviewing and revising them and for checking every source. Corrections, and references the author has missed, are welcome through the 5D Library website.
Disclosure
The author uses the works of 5D Library, including this book, as the methodological basis of his community courses, the 5D Ascension Cards (Dimensional Shift Card Decks) and the Yafu AI knowledge base, and therefore has a commercial interest in the subject of this book. The author also runs communities and founded and runs Asia Influencer X and Yafu Global, whose business includes community operations and influencer and brand partnerships, and therefore has a direct commercial interest in the subjects of cooperation, reputation and community governance discussed in this book. The book discusses mathematical models of evolutionary games; it is not moral instruction, promotes no course, card deck or service, and is not investment or business advice; it uses no client or community data from these communities and businesses; the Chapter 8 trial is designed to be run and analysed by researchers independent of the author’s business and does not recruit from the author’s community, students or customers, including the clients of these businesses; all the code and the study protocol are published openly for anyone to check.
References
Fan, Y. (2026). Evolutionary Game Theory, Direct and Indirect Reciprocity and the Emergence of Cooperation: From Hamilton's Rule and Axelrod's Tournaments to Reputation and Costly Punishment, Read Beside Classical Confucian and Mohist Ethics [演化博弈、直接与间接互惠及合作涌现的数学模型] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.23788551 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).