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 2401–2500 of 3949. Source text is displayed literally.

2401For instance, an adapter can expose the population state as an input2402and compute a payoff vector as its output. A downstream selection2403component then evaluates each strategy against that vector.2404Compositional optimization tests whether a strategy passes the2405resulting response relation. A dynamical model adds a state transition2406or vector field on populations, whose state change is an additional2407map. These two observations about the system can then be checked for2408agreement under the hypotheses of Chapter~\ref{ch:lyapunov}.2409 2410\section{Deterministic adapters}2411 2412The response formula gives a useful rule for changing the language in2413which an interface is described. Suppose2414\(L:(X',S')\to(X,S)\) and \(M:(Y,R)\to(Y',R')\) are lenses.2415Regard each as a singleton-strategy game with its sole response.2416For a context \((x',k')\) of \(M\circ\mathcal G\circ L\), the2417response test of the middle game has context2418\[2419 \left(v_L(x'),\ y\longmapsto u_M(y,k'(v_M(y)))\right).2420\]2421This follows by first applying the lens substitution to the return2422from \(M\), then carrying the input forward through \(L\). The2423singleton comparisons impose no further response condition, and the2424same formula therefore computes equilibria of the adapted game.2425If a downstream adapter translates every return by the same amount2426for the relevant decision context, a real maximizing response test2427is unchanged. More generally a strictly increasing transformation2428of every return in that test preserves the ordering of alternatives.2429The key hypothesis here is a common transformation of the returns2430being compared. A translation depending on the output may alter2431their order and must be included in the continuation itself.2432\par2433\chapter{Parallel games and higher comparisons}2434\label{ch:compositional-games}2435 2436Sequence describes a change of context through an intermediate2437interface. Parallel composition retains two interfaces and asks how2438the environment evaluates their joint output. A joint continuation2439can couple the decisions even when their play maps factor.2440 2441\section{Parallel composition}2442 2443For \(\mathcal G:(X,S)\to(Y,R)\) and2444\(\mathcal H:(U,T)\to(V,Q)\), their product has object types2445\((X\times U,S\times T)\) and \((Y\times V,R\times Q)\), and2446strategy set \(\Sigma_{\mathcal G}\times\Sigma_{\mathcal H}\).2447Its play and coplay are the products of the fixed-strategy lenses.2448Given a context with forward state \((x,u)\) and continuation2449\(k:Y\times V\to R\times Q\), an incumbent \((\sigma,\tau)\)2450produces \(y_* = P_{\mathcal G}(\sigma,x)\) and2451\(v_* = P_{\mathcal H}(\tau,u)\). Define the two continuations2452seen by the components as2453\begin{align*}2454 k_1(y)&=\pi_R k(y,v_*),&2455 k_2(v)&=\pi_Q k(y_*,v).2456\end{align*}2457The parallel best-response relation requires2458\((\sigma,\sigma')\in B_{\mathcal G}(x,k_1)\) and2459\((\tau,\tau')\in B_{\mathcal H}(u,k_2)\).2460A trial change in one component thus sees the same incumbent2461output from the other. In particular, even a product of games2462whose input states are independent can have coupled equilibrium2463conditions when the continuation is joint. If2464\(k(y,v)=(r(y),q(v))\), the continuations are \(r\) and \(q\)2465independently of \(\sigma,\tau\), so the equilibrium sets multiply.2466 2467\begin{proposition}[Interchange]2468Parallel composition commutes with sequential composition up to2469the canonical exchange of strategy factors. Its associativity,2470unit and symmetry are given by the corresponding product2471bijections.2472\end{proposition}2473 2474\begin{proof}2475Consider two chains, \(\mathcal G\) followed by \(\mathcal H\)2476and \(\widetilde{\mathcal G}\) followed by2477\(\widetilde{\mathcal H}\), with incumbent strategies2478\(\sigma,\tau\) and \(\widetilde\sigma,\widetilde\tau\),2479respectively. Let \(y_*,\widetilde y_*\) be the intermediate2480outputs, \(z_*,\widetilde z_*\) the final outputs, and \(k\) the2481joint outer continuation. In either order of assembly the test at2482\(\mathcal H\) uses \(y_*\) and \(z\mapsto\pi_1 k(z,\widetilde z_*)\).2483The test at \(\mathcal G\) uses the continuation2484\[2485 y\longmapsto C_{\mathcal H}2486       \bigl(\tau,y,\pi_1 k(P_{\mathcal H}(\tau,y),\widetilde z_*)\bigr).2487\]2488Indeed, composing within the first chain propagates the first2489outer continuation through \(\mathcal H\). Composing across the2490parallel pair first propagates \(k\) through the product lens,2491and then freezes the second intermediate value at2492\(\widetilde y_*\). Its downstream output is precisely2493\(\widetilde z_*\). The formulas for the other chain follow by2494using \(\pi_2\) and holding \(z_*\) fixed. Thus all four response2495tests agree. The play and coplay maps agree because composition of2496product lenses is componentwise.2497 2498For any number of parallel factors the continuation for one2499factor fixes all other incumbent outputs and projects to the2500return for that factor. This gives the same formula for either