Indexed text attachment
evolutionary-game-theory-in-higher-categories.tex
Source for Evolutionary Game Theory in Higher Categories
View paper · Download original file
Lines 521–620 of 3949. Source text is displayed literally.
521and need not reduce to checking only its vertices.522 523If \(p\) is strictly Nash, its payoff gap on the region524of competitors at distance at least \(r>0\) from \(p\)525has a positive minimum whenever that region is nonempty.526This follows from continuity on a closed subset of the527compact simplex. There is generally no positive minimum528over every distinct competitor, since a sequence of529competitors can approach \(p\). The existence of a uniform530invasion barrier is instead a uniform statement about531the relative contributions of the first and second532payoff differences. The next chapter establishes it by533examining the directions in which the first difference534vanishes.535\par536\chapter{Local superiority and mixed strategies}537\label{ch:superiority}538 539The invasion test fixes a competitor and then varies its share.540Local superiority compares the resident with every nearby541population against that population itself. In finite-dimensional542matrix games, critical directions at the resident relate these543two descriptions. They also isolate the part of the payoff matrix544that controls stability of a mixed resident. The equivalence with545a uniform invasion barrier is discussed in546\cite[Ch.~2]{alexander} and \cite[Ch.~2]{weibull}.547 548\section{The critical cone}549 550Write \(T_I=\{v:\sum_i v_i=0\}\). For \(p\in\Delta(I)\)551let \(C_p=\{v\in T_I:v_i\geq0\text{ if }p_i=0\}\).552Every displacement \(q-p\) belongs to \(C_p\), and every553\(v\in C_p\) is an actual displacement up to multiplication by554a small positive scalar. Introduce the linear and quadratic forms555\[556 L_p(v)=-v^{\mathsf T}Ap,\qquad Q_A(v)=v^{\mathsf T}Av557\]558and the critical cone \(K_p=C_p\cap\ker L_p\).559The Nash condition is exactly \(L_p\geq0\) on \(C_p\).560In one direction one scales each feasible vector to an actual561displacement and applies Nash. In the other direction one562applies the inequality to \(q-p\) for arbitrary \(q\).563 564If \(q=p+v\), the two terms in the invasion criterion are565\(d_0=L_p(v)\) and \(d_1=L_p(v)-Q_A(v)\).566It follows from Theorem~\ref{thm:invasion-criterion} and567scaling of the quadratic form that568\[569 p\text{ is an ESS}570 \quad\iff\quad L_p|_{C_p}\geq0,\qquad571 Q_A(v)<0\quad(0\ne v\in K_p).572\]573The critical cone consists of directions with no initial loss574against the resident. The quadratic condition applies only575on that cone. At a pure resident with a unique best response,576for example, there is no nonzero critical direction.577At a boundary resident with additional tied best responses,578the cone can include directions that add previously absent579types. Restricting the calculation to the resident's support580can omit those directions.581 582\section{Uniform superiority}583 584Call \(p\) locally superior if \(u_A(p,x)>u_A(x,x)\)585for every \(x\ne p\) sufficiently near \(p\) in the simplex.586A uniform invasion barrier is one positive threshold that587works for all distinct competitors in the definition of588evolutionary stability.589 590\begin{theorem}591\label{thm:uniform-ess}592For a matrix game on a finite type set, evolutionary593stability, local superiority, and the existence of a594uniform invasion barrier are equivalent.595\end{theorem}596 597\begin{proof}598Suppose \(p\) is an ESS. For \(v=x-p\) the superiority599difference is600\[601 u_A(p,x)-u_A(x,x)=L_p(v)-Q_A(v).602\]603If there is no nonzero feasible direction, the local604condition is vacuous. Otherwise put605\(S=C_p\cap\{v:\|v\|=1\}\), a compact set. On its606critical subset \(S\cap K_p\), the quadratic form is607negative. The relatively open subset of \(S\) on which608\(Q_A<0\) contains this critical set. On its complement,609if nonempty, \(L_p\) has a strictly positive minimum \(m\)610by compactness and the fact that its zeros are critical.611Also \(|Q_A|\) has a finite maximum \(M\) on \(S\).612For \(w\in S\) and \(r>0\) we have613\[614 L_p(rw)-Q_A(rw)=r\bigl(L_p(w)-rQ_A(w)\bigr).615\]616It is positive wherever \(Q_A(w)<0\) because \(L_p\geq0\).617On the complementary subset it is positive for618\(0<r<m/(1+M)\). This proves local superiority, including619the case in which the complementary subset is empty.620