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 1781–1880 of 3949. Source text is displayed literally.

1781stochastic matrix by itself supplies the reduction theorem but does1782not supply the symmetry used in the derivative of $P$.1783 1784\section{Migration and exponential synchronization}1785 1786Let $V=R_F$ be a common local replicator field with a Lipschitz constant1787$C$ on $\Delta(S)$ in Euclidean norm. Consider the additional coupling1788\[1789  \dot x^r=V(x^r)+\kappa\left(\sum_s W_{rs}x^s-x^r\right),1790  \qquad \kappa\geq 0.1791\]1792This term diffuses relative frequencies between equal-normalization1793populations. It mixes a particular type across populations even when1794its local frequency is initially zero. If the replicator has the1795ambient form above, the total mass vector $m$ satisfies1796$\dot m_r=(1-m_r)\phi_r+\kappa((Wm)_r-m_r)$ along any solution.1797The constant unit vector solves this linear inhomogeneous equation,1798with the coefficients taken along that solution. Uniqueness preserves1799each mass. The incoming linear terms have nonnegative off-diagonal1800coefficients for every type. Variation of constants, or local Picard1801iteration of the resulting positive integral equation for each type,1802preserves nonnegativity. Thus the model has a global flow on1803$\mathcal D_I$.1804 1805\begin{proposition}1806Let $|I|\geq 2$ and suppose $W$ is irreducible and reversible with a1807positive stationary distribution $\mu$. Let $\lambda>0$ be the1808smallest positive eigenvalue of $I-W$ in the $\mu$-weighted inner1809product. If $\kappa\lambda>C$, the squared distance to the synchronous1810subspace, in the product norm with weights $\mu$, decays at least at1811rate $2(\kappa\lambda-C)$.1812\end{proposition}1813 1814\begin{proof}1815For the mean $\bar x=\sum_r\mu_rx^r$ set $\delta^r=x^r-\bar x$ and1816$E=\tfrac12\sum_r\mu_r\lVert\delta^r\rVert^2$. Since the deviations1817have weighted mean zero, the term from $\dot{\bar x}$ in $\dot E$1818vanishes. The local contribution is1819$\sum_r\mu_r\langle\delta^r,V(x^r)-V(\bar x)\rangle\leq 2CE$.1820The migration contribution equals $-\kappa$ times the Dirichlet form1821\[1822  \left\langle\delta,(I-W)\delta\right\rangle_\mu1823  =\frac12\sum_{r,s}\mu_rW_{rs}1824               \lVert\delta^r-\delta^s\rVert^2.1825\]1826Reversibility makes $I-W$ self-adjoint. Irreducibility forces every1827zero of the displayed form to be constant across populations. Applying1828the spectral gap on the mean-zero subspace bounds the form below by1829$2\lambda E$. The claim follows by integrating1830$\dot E\leq -2(\kappa\lambda-C)E$.1831\end{proof}1832 1833Migration cancels from the mean equation by stationarity of $\mu$.1834The mean therefore satisfies1835$\lVert\dot{\bar x}-V(\bar x)\rVert\leq C\sqrt{2E}$. The estimate1836controls the deviation from the local law along the mean path, and the1837stability of any particular synchronous orbit is a further dynamical1838question. The partition theorem gives an exact invariant subspace. The1839migration estimate gives an attraction criterion for one such subspace.1840Both properties specify what a spatial reduction preserves and under1841which hypotheses it does so.1842\par1843 1844\bookpart{III}{Stochastic populations and aggregation}1845\chapter{Finite populations and Markov dynamics}1846\label{ch:markov}1847 1848Alexander starts the finite-population analysis from individual replacement and then asks what changes when interaction is local \cite[Sec.~4]{alexander}. To make this distinction precise, separate the population state, the law of an update, and the statistic that describes the next state. A stochastic update assigns a probability to each possible successor. Its support determines what can happen, and its probability weights determine the relative likelihoods. Both are needed to describe extinction, fixation, and stationary behavior.1849 1850\section{Count states and fitness}1851 1852Let \(S\) be a nonempty finite strategy set. Fix a population size \(N\geq 2\). A labeled configuration is a function \(z:\{1,\ldots,N\}\to S\). Its count vector is \(n(z)\), with \(n_i(z)\) the number of individuals using strategy \(i\). The set of count states and the associated frequencies are1853\[1854 C_N(S)=\left\{n\in\mathbb N^S:\sum_{i\in S}n_i=N\right\},1855 \qquad x_i=\frac{n_i}{N}.1856\]1857Count states are orbits of the action of label permutations on configurations. The induced action on an update law can be used to test whether the orbit statistic itself evolves by a Markov chain. Under exchangeable sampling, the law of the updated counts depends only on the current counts. On a graph, two configurations in the same orbit can place the same strategy at vertices with different neighborhoods. The Markov property for counts then requires a separate argument.1858 1859Suppose \(A_{ij}\) is the payoff to strategy \(i\) against \(j\), and that an individual's opponent is sampled uniformly from the other \(N-1\) individuals. The expected payoff of a present type \(i\) is1860\[1861 u_i^N(n)=\sum_{j\in S}\frac{n_j-\mathbf 1_{\{i=j\}}}{N-1}A_{ij}.1862\]1863If \(n_i=0\), there is no individual of type \(i\) whose payoff is needed. Give each currently present type a strictly positive reproductive weight \(f_i(n)\). For instance a positive response \(f_i(n)=\phi(u_i^N(n))\) implements a prescribed mapping from payoffs to reproductive probabilities. The choice is part of the model. Adding the same constant to all payoffs preserves many static comparisons. Adding the same constant to positive reproductive weights changes ratios \(f_i/f_j\) in general and hence can change the stochastic dynamics. A common positive multiplication of all weights at a state leaves their normalized ratios unchanged. The distinction parallels the several meanings of payoff equivalence in Chapter~\ref{ch:payoff-equivalence}.1864 1865\section{Birth, mutation, and replacement}1866 1867The birth--death process updates one position at a time. It belongs to the overlapping-generation tradition of Moran's model \cite{moran-random}. Here the parent is chosen with probability proportional to its reproductive weight, the offspring type is sampled from a mutation kernel, and a uniformly chosen member of the current population is replaced. The parent is eligible for replacement in this convention. Let \(M_{ik}\) be the probability that a type \(i\) parent produces type \(k\), with each row of \(M\) summing to one. Define1868\[1869 b_i(n)=\frac{n_i f_i(n)}{\sum_j n_j f_j(n)},\qquad1870 c_k(n)=\sum_i b_i(n)M_{ik},\qquad1871 d_j(n)=\frac{n_j}{N}.1872\]1873For an absent type we set \(n_i f_i(n)=0\) by convention. The offspring distribution is \(c(n)\). Let \(e_i\) denote the count vector with a unit at \(i\) and zero elsewhere. If \(k\neq j\) and \(n_j>0\), the transition from \(n\) to \(n+e_k-e_j\) has probability \(c_k(n)d_j(n)\). The probability of remaining at \(n\) is \(\sum_j c_j(n)d_j(n)\). These probabilities sum to one because all ordered pairs of offspring and removed types have been accounted for. A death rule depending on the parent or on its location replaces the product \(c_kd_j\) by a joint kernel. In that case the joint law is part of the dynamics.1874 1875The update can also be written as a composition of kernels, retaining only the information required by the next stage. With \(D_n=\{(k,j)\in S^2:n_j>0\}\), each arrow below is well defined on its displayed states:1876\[1877\begin{tikzcd}[column sep=large]1878 \{n\} \arrow[r,"c\otimes d"{above}] &1879 D_n \arrow[r,"{\text{replacement}}"{above}] &1880 C_N(S).