Independent scholarship · Open accessThe monthly digest

Indexed text attachment

evolutionary-game-theory-in-higher-categories.tex

Source for Evolutionary Game Theory in Higher Categories

View paper · Download original file

Lines 3149–3248 of 3949. Source text is displayed literally.

3149rule for that motion. A connection or transition process would3150be additional data.3151 3152\section{Parametrization and effective descent}3153 3154In an $\infty$-topos $\mathcal E$, let $X,Y,P$ be objects over a3155base $B$. A family of evaluations from $X$ to $Y$ with parameter3156$P$ is a morphism $P\times_B X\to Y$. By the adjunction3157for the exponential in $\mathcal E_{/B}$, such a family is3158classified by $P\to[Y^X]_B$, where $[Y^X]_B$ denotes the3159fiberwise exponential. An inverse-image or presentation change3160can preserve evaluation while changing parameter coordinates.3161The pullback formula follows by applying the functor3162composing structure maps to the exponential adjunction:3163for $q:C\to B$ and $T\to C$,3164\[3165 q_!(T\times_C q^*X)\simeq q_!T\times_B X .3166\]3167Thus $q^*[Y^X]_B\simeq[(q^*Y)^{q^*X}]_C$ by Yoneda.3168If $U\to B$ is an effective epimorphism with \v{C}ech nerve3169$U_\bullet$, then3170\[3171 \mathcal E_{/B}\simeq3172 \lim_{[n]\in\boldsymbol\Delta}\mathcal E_{/U_n}.3173\]3174This is descent for slices in an $\infty$-topos3175\cite[\S6.1.3]{lurie-htt}. Objects and morphisms over $B$3176can thus be specified locally along with a homotopy-coherent3177chain of overlap comparisons. The comparison extends to3178evaluation families since their parameter maps land in the3179compatible exponential object.3180 3181This gives a semantics for local games. A family can be observed3182by changing the base, and the result of interpreting that family3183is independent of a selected cover. For the concrete smooth3184model the relevant gluing equations reduce to equalities; for3185spaces with groupoid-valued states a commuting triangle may3186instead carry an invertible comparison. One must still specify3187which points are probability states and which maps define3188fitness. The categorical descent theorem glues these structures3189once they are defined in the chosen semantics. It does not3190supply a probability law from the base topos.3191 3192\section{Monodromy and restricted parameter choices}3193 3194A loop in $B$ transports its strategy fiber by a permutation.3195The image of the resulting representation of the fundamental3196group is a monodromy group $M$. In a connected component,3197a flat state is a locally constant probability vector in the3198trivializations. It is specified by a vector $x$ at one base3199point that is fixed by every permutation in $M$. To prove it,3200transport a choice from that point along paths. This is independent3201of the path exactly when every based loop fixes the vector.3202Flat states therefore have a common value within each orbit of3203$M$ on types. If $O$ is such an orbit and $w_O$ is its mass,3204each of its coordinates has value $w_O/|O|$. The flat3205parameter region is consequently a simplex on monodromy orbits.3206A general smooth section of the bundle can vary within a3207context and needs only to obey transition laws.3208 3209A flat payoff matrix, by the same argument, is fixed under3210conjugation by the monodromy permutations. Its entry depends3211on an orbit of the diagonal action on ordered pairs of types.3212These conditions can force parameter tying, even when local3213populations have several distinguishable types. Conversely,3214the pair orbits may retain interaction patterns that are3215lost if one replaces all types by their orbit labels before3216giving the fitness. Context determines which changes of3217presentation are legitimate. The next chapter measures a3218different failure, when a coarse description is a randomized3219channel rather than a permutation.3220\par3221\chapter{Selection under stochastic compression}3222\label{ch:stochastic-compression}3223 3224An observer can group strategies by a classification rule.3225A deterministic grouping sends each strategy to one class,3226whereas a randomized classifier sends it to a distribution of3227classes. These two changes of description have different effects3228on replicator dynamics. When fitness is pulled back along the3229classifier, the difference is given by a conditional covariance.3230 3231\section{Columns as conditional distributions}3232 3233Let $I,J$ be nonempty finite sets, and let $A$ be a3234column-stochastic matrix from $I$ to $J$. Its column $a_i$3235is the conditional distribution of the label $j\in J$ given3236a strategy $i\in I$. Write $x\in\Delta(I)$ and $z=Ax$.3237The pushforward of a distribution is $Ax$. The corresponding3238pullback of a payoff vector $\gamma\in\mathbb R^J$ is3239$A^\top\gamma$, since3240\[3241 \langle x,A^\top\gamma\rangle=\langle Ax,\gamma\rangle.3242\]3243This is expectation along a Markov kernel, expressed in3244dual coordinates \cite{fritz-markov}. If $g:\Delta(J)\to\mathbb3245R^J$ is the observed fitness, consider the fine fitness3246$f(x)=A^\top g(Ax)+c(x)\mathbf 1_I$, where $c$ is a3247common payoff offset. This assumption controls the dependence3248of a fine type's payoff on all its possible labels. It must