Independent scholarship · Open accessThe monthly digest

Editorially reviewed

Games and Information in Higher Categories

A 19-chapter monograph rebuilding noncooperative game theory and information economics, after Rasmusen's Games and Information, in categorical terms: games as points of types and as composable open games, knowledge as base change, common priors as an H^1 obstruction, finite versus infinite horizons as induction versus coinduction, and incentive compatibility as an enriched-presheaf condition.

Anonymous← All research

Games and Information in Higher Categories

This monograph is a higher-categorical treatment of the territory mapped by Eric Rasmusen's Games and Information: An Introduction to Game Theory (4th ed., Blackwell, 2006): strategic and extensive forms, information, mixed strategies, dynamic and repeated games, incomplete information, moral hazard, adverse selection, mechanism design, signalling, bargaining, auctions and pricing. It is an independent work. It reproduces none of that book's text, tables, figures or problems, and a concordance (Appendix B) lets the two be read side by side.

The attached PDF has 19 chapters, four appendices and 240 references in 189 pages. Appendix D gives solutions to all 105 exercises. Each topic is stated classically, with proofs or references and a worked example, and then re-read in a language in which its structural content is explicit. Every numbered statement carries a status tag: classical, literature, reformulation, observation or programme. The sections below summarize the main re-readings.

Solution concepts are functions on types

A solution concept is a dependent function $S:\prod_{G:\mathcal{T}}\mathcal{P}(\Sigma(G))$ on a type $\mathcal{T}$ of games, so it respects the identity type of $\mathcal{T}$. Four requirements from four literatures are this one fact for four types. On the automorphisms of a game it is the symmetry axiom of equilibrium selection, and it forces a focal point to be extra structure. On extensive forms modulo the Thompson transformations it is Kohlberg–Mertens invariance; an explicit pair of trees shows that subgame perfection is not a function on that quotient. On type spaces modulo bisimilarity it separates interim correlated rationalizability from Bayesian equilibrium. On continuation games it is the Markov restriction.

The choice of type has content. Payoff profiles related by $u_i\mapsto\alpha_iu_i+\beta_i(a_{-i})$ with $\alpha_i>0$ have the same dominance relations, rationalizable sets, Nash and correlated equilibria. The Prisoner's Dilemma is equivalent in this sense to a game whose unique equilibrium is Pareto efficient. Prediction and welfare evaluation therefore live on different types of games.

Knowledge is base change; common knowledge is a colimit

For an agent who observes $p:\Omega\to T$, Aumann's knowledge operator is $K=p^*\forall_p$ on subsets of $\Omega$. The S5 laws follow from the adjoint triple $\exists_p\dashv p^*\dashv\forall_p$. Common knowledge is $q^*\forall_q$ for the pushout of the agents' type maps:

$$\begin{CD} \Omega @>{p_1}>> T_1 \\ @V{p_2}VV @VVV \\ T_2 @>>> T_\wedge \end{CD}$$

Computed in spaces rather than sets, the same colimit is a bipartite graph $B$ with vertices $\Omega\sqcup\coprod_iT_i$ and an edge from each state to each of its types. Then $\pi_0(B)=T_\wedge$ is the meet of the partitions, and for $n$ agents

$$b_1(B)=(n-1)\,|\Omega|-\sum_i|T_i|+|\pi_0(B)|.$$

A common prior is the vanishing of a class in $H^1$

Let $\beta$ assign to each type $t\in T_i$ a full-support posterior on its cell, and define a 1-cochain on $B$ by

$$\lambda_\beta(e_{\omega,i})=\log\beta_i(p_i\omega)(\omega).$$

Then $\beta$ has a common prior if and only if $[\lambda_\beta]=0$ in $H^1(B;\mathbb{R})$. When it exists the prior is unique on each component of $B$. Every class is realized by some posterior profile, so profiles with a common prior form a submanifold of codimension $b_1(B)$. This repackages the cycle equations of Rodrigues-Neto and the tightness criterion of Hellman and Samet, whose counting condition $\sum_i|T_i|=(n-1)|\Omega|+1$ is the equation $b_1(B)=0$ for connected $B$. Aumann's agreement theorem becomes functoriality of conditional expectation around the pushout square.

Finite horizons are initial algebras; infinite horizons are terminal coalgebras

Finite game trees form the initial algebra of a polynomial functor, and backward induction is a fold. Subgame perfection is an S4 necessity operator applied to the Nash predicate, formally parallel to $K$. Strategies of a repeated game form the terminal coalgebra of $X\mapsto A_i\times X^{A}$; automata are coalgebras, and the Kalai–Stanford complexity of a strategy is the size of its bisimulation quotient. A self-generating set in the sense of Abreu, Pearce and Stacchetti is a post-fixed point of a monotone operator $B$ and at the same time a coalgebra whose states are promised values; the equilibrium payoff set is $\nu B$.

The folk theorem is the gap between the least and greatest fixed points of that operator. Rubinstein's uniqueness theorem for alternating offers is the absence of a gap, because the Shaked–Sutton recursion $x\mapsto1-\delta_2(1-\delta_1x)$ is a contraction.

Incentive compatibility is an enriched-presheaf condition

For quasi-linear utility $v(x,\theta)-t$ and an allocation rule $f:\Theta\to X$, put $\ell_f(\theta',\theta)=v(f(\theta),\theta)-v(f(\theta'),\theta)$ and let $d_f$ be its shortest-path closure, the free category on $(\Theta,\ell_f)$ enriched in $\overline{\mathbb{R}}=([-\infty,\infty],\ge,+)$. Incentive-compatible payment rules are exactly the copresheaves:

$$t(\theta)-t(\theta')\le d_f(\theta',\theta).$$

Rochet's theorem says $f$ is implementable iff $d_f(\theta,\theta)=0$ for all $\theta$. The corepresentable $d_f(\theta_0,-)$ is the greatest payment rule vanishing at $\theta_0$. Revenue equivalence holds iff $d_f(\theta,\theta')+d_f(\theta',\theta)=0$ for all pairs (Heydenreich–Müller–Uetz–Vohra), which says every hom-value is invertible. On a grid of $n+1$ types with $f(\theta)=\theta$ the defect is exactly $1/n$. The revelation principle is recast as initiality of the direct mechanism among implementations, resting on the fact that best responses pull back along maps of mechanisms.

Composition

Chapter 2 presents the open games of Ghani, Hedges, Winschel and Zahn. The lens and optic discipline is then used for contracts (a contract is a costate), for the informativeness principle (garbling is factorization of Markov kernels), for market microstructure, and for signal jamming, where a deviation changes the forward pass and leaves the conjectured backward pass fixed. The text is explicit that composition composes equilibrium checking and not equilibrium search.

Status and authorship

The contribution is the unified development, the reformulations named above with short proofs from the cited results, the worked examples, and a closing assessment with eight open problems. Three statements are, to the author's knowledge, not in the literature in this form: the $H^1$ obstruction on the homotopy-colimit graph, payments as enriched copresheaves, and the revelation principle as initiality. None of the cited theorems is claimed as new.

Limitations: the structures used are mostly low-dimensional (graphs, 1- and 2-categories, enriched posets), no new existence theorem is proved, and the proofs and bibliographic details have not been independently refereed.

The attached verify.py uses only the Python standard library and a fixed seed. Its 113 checks recompute every numerical example, the numerical answers in the solutions appendix, and test the finite structural claims (S5 laws, the $b_1$ formula, the common-prior obstruction, open-game composition, products of selection functions, the incentive metric, fixed-point indices) on explicit and random instances. This is evidence about the examples, not proof of the theorems. The LaTeX source is attached as a single UTF-8 file marked for text indexing.

Generative AI (Anthropic's Claude) assisted with drafting, literature recall, mathematical consistency checks, computation and typesetting, under the direction of the submitting author, who is responsible for the text and uses the public byline "Anonymous".

Reproducibility artifacts

games-and-information-in-higher-categories.tex · 448,079 bytes · Read in browser · Download
Text indexing selected · View indexed source
SHA-256 827a6c1e3f5d3db5b8bb0480c4dacadf35fa6d29bee969d977a7679870710bce

games-and-information-in-higher-categories.pdf · 1,013,431 bytes · Download
SHA-256 e76b18567bb35b1073c16cf8662e8275aa395393ef71272ddc03f7ce3dfa9f6b

Discussion

Open a careful, constructive conversation about this note.

Comments appear after editorial approval.