Indexed text attachment
evolutionary-game-theory-in-higher-categories.tex
Source for Evolutionary Game Theory in Higher Categories
View paper · Download original file
Lines 3709–3808 of 3949. Source text is displayed literally.
3709 3710A program strategy can be represented by a fixed artifact,3711with an evaluation map from its identity and the current3712context to a score. The test distribution, budget,3713counterfactual comparison class, and timing determine3714whether the score estimates the intended fitness.3715An inner process can generate or train candidate3716artifacts. An outer process can then keep a distribution3717over verified candidates and update that distribution using3718scores. Exponential weights keeps the distribution in the3719simplex and is insensitive to score shifts common to3720all candidates in a round. These facts are properties of3721the update. They alone say nothing about the behavior3722or validity of any candidate.3723 3724There is also a compositional concern. If an evaluator3725reports the same payoff for every strategy in a coarse3726fiber, deterministic aggregation can commute with3727selection. If it reports only a randomized label, the3728calculation of Chapter~\ref{ch:stochastic-compression}3729identifies a covariance defect even when the fine3730payoffs are conditional expectations of coarse3731scores. This gives a test of a proposed reporting3732interface before using it as a summary of population3733behavior. Delayed reports, context changes, or3734evaluation mismatch introduce separate errors. A3735learning guarantee is conditional on specifying3736those mechanisms and bounding their effect.3737 3738The constructions in the book put equilibrium,3739dynamics, stochastic modeling, and presentation3740changes into compatible types. A model of interacting3741agents adds an interpretation of strategies and a3742mechanism for producing scores at the chosen interface.3743The resulting learning statements remain relative3744to that mechanism.3745\par3746 3747\appendix3748\chapter{Notation}3749\label{app:notation}3750\small\raggedright3751\setlength{\parskip}{0pt}3752\begin{description}[leftmargin=2em,style=nextline,itemsep=.31\baselineskip]3753\item[$I$, $\Delta(I)$, $x_i$] Type set, its probability simplex,3754 and mass assigned to type $i$. See Chapter~\ref{ch:population}.3755\item[$f(x)$, $\langle x,f(x)\rangle$] The current fitness vector3756 and its mean under $x$.3757\item[$R_f(x)$, $M(x)$] The replicator field and the matrix3758 $\operatorname{diag}(x)-xx^\top$ that converts a payoff3759 covector into a selection direction. See Chapters~\ref{ch:replicator}3760 and~\ref{ch:stochastic-compression}.3761\item[$D(p\Vert x)$] Relative entropy of a comparator $p$ from3762 the state $x$, finite when the support of $p$ is contained in3763 that of $x$. See Chapters~\ref{ch:lyapunov} and~\ref{ch:agent-learning}.3764\item[$\operatorname{supp}(x)$] The types with positive mass in $x$.3765 The associated support face contains trajectories from $x$ for3766 the replicator field without mutation.3767\item[$P$, $Q$] In stochastic chapters, transition kernels or3768 generator matrices with conventions specified at their use.3769 In parametrized games, $P$ can also denote a parameter object.3770\item[$A^\top$, $C_A(x)$] The pullback of payoff vectors along3771 a column-stochastic channel $A$, and its conditional covariance3772 $\operatorname{diag}(Ax)-A\operatorname{diag}(x)A^\top$.3773 See Chapter~\ref{ch:stochastic-compression}.3774\item[$(X,S)$, $(Y,R)$] Forward and backward types of an open-game3775 interface. See Chapter~\ref{ch:open-games}.3776\item[$\mathcal E_{/B}$, $P\times_B X$]3777 The slice over a context object $B$ and its fiber product.3778 See Chapter~\ref{ch:contextual-games}.3779\item[$BG$, $X//G$] The delooping of an internal group object and3780 the homotopy quotient of a $G$-object. See Chapter~\ref{ch:higher-equivariance}.3781\end{description}3782\chapter{Machine-checked selection identities}3783\label{app:formalization}3784 3785The accompanying Lean project formalizes the algebra that makes3786deterministic aggregation compatible with selection. It uses Lean~4.19.0,3787pinned mathlib, and 23 audited theorems. The index types are arbitrary3788finite types. The statements apply uniformly to every cardinality,3789including reindexing along a map that leaves some target coordinates3790empty. See \cite{lean4} for the proof assistant.3791 3792\section{Drift and the normalized update}3793 3794Let $x,u:I\to R$, where $I$ has a finite type structure. The3795formal definitions are3796\[3797 m(x)=\sum_i x_i,\qquad3798 \mu_x(u)=\sum_i x_i u_i,\qquad3799 F(x,u)_i=x_i\big(u_i-\mu_x(u)\big).3800\]3801Over a commutative ring the kernel checks3802\[3803 \sum_i F(x,u)_i=(1-m(x))\mu_x(u).3804\]3805It follows that the drift is tangent to the mass-one hyperplane.3806The zero-coordinate property, invariance under adding a common payoff3807on that hyperplane, and linearity under uniform payoff scaling are3808also checked. These are identities of algebraic expressions used in