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 2861–2960 of 3949. Source text is displayed literally.

2861A complete forward flow determines a functor2862\(B(\mathbb R_{\geq 0},+)\to\mathbf{Set}\). Here \(B\) makes2863the monoid of nonnegative times into a one-object category, and2864the functor sends addition of times to composition of evolution2865maps. The square says precisely that \(q\) is the component2866of a natural transformation. Smooth semiconjugacies form a2867category with ordinary identities and composition. The naturality2868condition uses a common clock. A state-dependent time change such2869as the positive rescaling in Chapter~\ref{ch:compositional-games}2870requires a separate description.2871 2872Given \(F\) and a surjective submersion \(q\), a suitable \(G\)2873exists exactly when \(Dq_xF(x)\) is constant on every fiber2874\(q^{-1}(y)\), as a vector in \(T_yN\). Necessity is immediate2875from \eqref{eq:exact-semiconjugacy}. For sufficiency, define the2876candidate at \(y\) to be this common value. A local smooth section2877of the submersion expresses it as a smooth field on a neighborhood2878of \(y\), and the expressions from different sections agree.2879Surjectivity also gives uniqueness. For an affine quotient of2880simplices by a partition, a global affine section is available.2881Choose a probability vector supported in each block and lift2882coarse masses to these fixed proportions. Evaluating \(Dq F\)2883on that section gives the smooth candidate whenever the2884fiber condition holds. The candidate preserves the coarse2885simplex whenever the original flow preserves the fine simplex,2886by applying the chain rule to a source solution through2887a lift of each coarse initial state.2888 2889\section{Partition quotients of selection}2890 2891Let \(b:S\twoheadrightarrow T\) be a partition map and let2892\(B_{\alpha i}\) be \(1\) if \(b(i)=\alpha\), and \(0\)2893otherwise. Write \(z=Bp\) for the image of the population2894\(p\), so \(z_\alpha=\sum_{b(i)=\alpha}p_i\). Let2895\(F_i(p)=p_i(f_i(p)-\bar f(p))\) with2896\(\bar f(p)=\sum_i p_i f_i(p)\). For every block with2897\(z_\alpha>0\) define2898\[2899 \widehat f_\alpha(p)2900 =\frac{1}{z_\alpha}\sum_{b(i)=\alpha}p_i f_i(p).2901\]2902Summing the microscopic equations and regrouping the mean gives2903\[2904 (BF(p))_\alpha2905 = z_\alpha\left(\widehat f_\alpha(p)2906        -\sum_{\beta:z_\beta>0}z_\beta\widehat f_\beta(p)\right)2907\]2908on the active blocks. The other block velocities vanish, since2909all of their microscopic masses are zero.2910 2911\begin{proposition}2912For a proposed coarse payoff field \(g(z)\), the partition2913sends the replicator field for \(f\) to the replicator field2914for \(g\) if and only if2915\(\widehat f_\alpha(p)-g_\alpha(Bp)\) is independent of \(\alpha\)2916among the active blocks for every \(p\).2917\end{proposition}2918\begin{proof}2919Subtract the two velocities on each active block and divide2920by its positive mass. The resulting equality requires the2921difference between the two payoffs to equal the weighted mean2922of those differences on the active blocks. Hence all of2923the differences are equal there. A common difference cancels2924in the centered payoffs and yields equality of the velocities.2925Inactive blocks contribute neither mass nor velocity to either2926calculation.2927\end{proof}2928 2929A sufficient condition is2930\(f_i(p)=g_{b(i)}(Bp)+c(p)\) with the same scalar \(c(p)\)2931for all strategies. For matrix payoffs, the corresponding2932form is \(A=B^{\mathsf T}\bar A B+\mathbf 1 c^{\mathsf T}\),2933where \(c\) is a constant vector and \(\bar A\) is the coarse2934payoff matrix. The proposition is weaker than pointwise2935payoff agreement within each block. Microscopic deviations2936from the proposed block payoff can cancel in the population2937average. This cancellation is tested at every microscopic2938population in the domain.2939 2940\section{Comparison with linear lumpability}2941 2942For a Markov equation \(\dot p=Qp\) with a column-conservative2943generator, the same partition is exact for another generator2944\(\overline Q\) if \(BQ=\overline Q B\). The column of2945\(BQ\) at a microscopic state \(i\) records the total signed2946rate into each coarse block. A matrix \(\overline Q\) exists2947precisely when these columns agree for all \(i\) in each2948source block. To see this, compare the matrices on each2949basis vector. If the equalities of columns hold, define the2950coarse column using any microscopic representative of the2951block. The sums of its entries vanish because the columns2952of \(Q\) sum to zero. Its off-diagonal entries are nonnegative2953because the corresponding target blocks contain no source state.2954Thus the resulting matrix is a Markov generator. Equivalently,2955the intertwining holds for the matrix exponentials at every2956nonnegative time, by applying it to each power and summing2957the exponential series. The implication in the reverse2958direction follows by differentiating the exponentials at2959zero. For open Markov processes, preservation of interfaces2960and fluxes adds constraints to the condition on the generators