Indexed text attachment
evolutionary-game-theory-in-higher-categories.tex
Source for Evolutionary Game Theory in Higher Categories
View paper · Download original file
Lines 1549–1648 of 3949. Source text is displayed literally.
1549since several copied entries contribute several terms to1550that sum.1551 1552In detail, let \(q\colon S'\twoheadrightarrow S\) be a1553surjection and pull a payoff vector \(f\) back to \(f\circ q\).1554The mass assigned to the fiber of \(i\) is proportional to1555\(\lvert q^{-1}(i)\rvert\exp(f_i/\eta)\). The effective1556coarse payoff includes a shift of \(\eta\log\lvert1557q^{-1}(i)\rvert\), which cancels from the response when1558the fiber sizes agree. More generally, using positive1559reference weights \(w_j\) on \(S'\) makes the mass of a fiber1560proportional to1561\(\exp(f_i/\eta)\sum_{j:q(j)=i} w_j\).1562The weighted response therefore commutes with lumping for1563pulled-back fitness vectors when one pushes the reference1564weights forward. To extend this calculation to a1565semiconjugacy of the revision equations, the fine fitness1566must be constant within each fiber and must be computed1567from the coarse state by the same rule at every instant.1568 1569One can derive the logit rule by regularizing the response1570optimization. Let \(H(y)=-\sum_i y_i\log y_i\), using the1571continuous extension \(0\log0=0\). For fixed \(x\), write1572\(Z=\sum_j\exp(F_j(x)/\eta)\). Substituting the definition1573of \(B^\eta\) gives1574\[1575 y\cdot F(x)+\eta H(y)1576 =\eta\log Z-\eta D(y\Vert B^\eta(x)).1577\]1578The identity is valid also for \(y\) on a proper face because1579the second argument of entropy has full support. By the1580nonnegativity established in Chapter~\ref{ch:lyapunov}, the1581unique regularized maximizer is \(B^\eta(x)\). Compare its1582objective value with that of a pure maximizing strategy,1583which has entropy zero. Since \(H(y)\le\log |S|\),1584\[1585 0\le \max_i F_i(x)-B^\eta(x)\cdot F(x)1586 \le\eta\log |S|.1587\]1588The entropy bound follows by computing relative entropy1589against the uniform distribution. In particular, each1590logit rest point is an \(\eta\log|S|\)-approximate Nash1591state, in the sense that this is a bound on the gain1592from any unilateral population-level deviation.1593 1594The dynamics \(\dot x=B^\eta(x)-x\) have a unique forward1595solution for locally Lipschitz \(F\). The same integrating1596factor used for best response shows forward invariance.1597For a fixed \(\eta\), continuity on the compact simplex1598gives a positive lower bound for every response coordinate.1599Thus the solution is interior for all \(t>0\), with each1600coordinate bounded uniformly away from zero after any1601positive time. Brouwer's fixed point theorem ensures a1602rest point for continuous \(F\). Every rest point is1603interior and satisfies \(x=B^\eta(x)\).1604 1605If \(\eta_k\to0\) through positive parameters and1606\(x^{(k)}=B^{\eta_k}(x^{(k)})\) converges, its limit is Nash.1607For a strategy strictly suboptimal at the limit, continuity1608gives a payoff gap of at least some \(\delta>0\) compared1609with a maximizing strategy there, for all large \(k\).1610Its probability in \(x^{(k)}\) is at most1611\(\exp(-\delta/\eta_k)\), hence is zero in the limit.1612Only maximizers can retain mass. This argument does not1613choose which Nash states arise along branches of fixed points.1614 1615The potential case gives a convergence result at a fixed1616noise level. Let \(V\) be a potential in the sense of1617Chapter~\ref{ch:lyapunov}, continuously differentiable on1618the simplex in its affine hull. On the interior the logit1619dynamics satisfy1620\begin{align*}1621 \frac{d}{dt}(V+\eta H)(x)1622 &=\eta\sum_i (B^\eta_i(x)-x_i)1623 \bigl(\log B^\eta_i(x)-\log x_i\bigr)\\1624 &=\eta\bigl[D(B^\eta(x)\Vert x)+D(x\Vert B^\eta(x))\bigr].1625\end{align*}1626The first equality uses1627\(F_i(x)=\eta\log B^\eta_i(x)+\eta\log\sum_j e^{F_j(x)/\eta}\)1628and the fact that the velocity sums to zero. Thus \(V+\eta H\)1629strictly increases except at rest points. The interior bounds1630after positive time make the compactness argument of1631Chapter~\ref{ch:lyapunov} applicable. If \(V\) is concave,1632then \(V+\eta H\) is strictly concave on the simplex. Each1633interior logit rest point satisfies its constrained1634first-order condition, so there is only one. All trajectories1635converge to it. The existence, limit, and convergence1636statements use different hypotheses on fitness and on the1637response rule. More general perturbation and revision1638theories are developed by Sandholm \cite{sandholm-population}.1639\par1640\chapter{Interacting populations and spatial structure}1641\label{ch:structured}1642 1643The comparison neighborhood in Chapter~\ref{ch:revision} consisted of1644individuals drawn from the same population. An individual can instead1645interact frequently with nearby populations while drawing its behavioral1646replacement from its own population. An individual who moves changes the1647composition of both the source and destination. These mechanisms give1648different equations and lead to different forms of reduction. We study