Indexed text attachment
evolutionary-game-theory-in-higher-categories.tex
Source for Evolutionary Game Theory in Higher Categories
View paper · Download original file
Lines 629–728 of 3949. Source text is displayed literally.
629 \bigl(u_A(p,x_\varepsilon)-u_A(q,x_\varepsilon)\bigr)630\]631follows from affinity in the first argument. Hence every632share below \(\min\{1,\delta/(1+D)\}\) gives positive633invasion advantage independently of \(q\). This yields634a uniform invasion barrier. A uniform threshold gives635the pointwise thresholds of evolutionary stability636immediately.637\end{proof}638 639Two features of the proof have distinct roles. Compactness640of the unit directions controls approach to the critical641cone. Boundedness of the simplex controls the passage from642nearby populations to a common mutant share for every643\(q\). General strategy spaces can lose either property.644The finite-dimensional proof cannot then be transferred645without a replacement estimate.646 647\section{Interior residents and the quadratic test}648 649In the relative interior of \(\Delta(I)\), the cone \(C_p\)650is the whole tangent space \(T_I\). A linear form nonnegative651on a vector space is zero. The Nash condition therefore says652that \(Ap\) is constant in the type index. For an interior653Nash resident, \(K_p=T_I\), and evolutionary stability is654equivalent to the restriction of the symmetric part655\((A+A^{\mathsf T})/2\) being negative definite on \(T_I\).656Indeed \(v^{\mathsf T}Av\) only sees that symmetric part.657When \(A=A^{\mathsf T}\), this is the Hessian test for the658quadratic function \(x\mapsto x^{\mathsf T}Ax/2\) restricted659to the affine hull of the simplex. Its differential at a660Nash state vanishes along \(T_I\), and its exact expansion661has quadratic term \(Q_A/2\). When \(A\) is not symmetric,662the argument establishes the sign of the critical quadratic663form. The derivative of this same quadratic function664contains the symmetric part and in general differs from665the fitness vector \(Ax\).666 667Negative definiteness on \(T_I\) also implies uniqueness of668a symmetric Nash state whenever one exists. If \(p,q\)669both satisfy Nash, adding \(u_A(p,p)\geq u_A(q,p)\)670and \(u_A(q,q)\geq u_A(p,q)\) gives671\((p-q)^{\mathsf T}A(p-q)\geq0\). Since \(p-q\in T_I\),672this forces equality of the states under negative definiteness.673 674\section{Persistence in payoff families}675 676The interior test has a direct parameter consequence.677Let \(\Pi\) be the Euclidean projection onto \(T_I\).678If \(Q_A<0\) on \(T_I\setminus\{0\}\), the operator679\(\Pi A|_{T_I}\) is invertible. Indeed, a vector \(v\)680in its kernel satisfies \(v^{\mathsf T}Av=0\).681Injectivity then gives invertibility on the682finite-dimensional space \(T_I\).683For any fixed \(b\) in the affine hull of the simplex,684the unique solution of \(\Pi A p=0\) in that affine hull685is686\[687 p_A=b-(\Pi A|_{T_I})^{-1}\Pi A b .688\]689The expression varies continuously with \(A\) near a690matrix for which the operator is invertible.691Strict positivity of each coordinate is open. Negative692definiteness is also open, by taking the maximum of693\(Q_A\) over the compact unit sphere in \(T_I\)694when \(T_I\ne0\). Thus an interior ESS satisfying695this test persists through sufficiently small payoff696perturbations, and in that neighborhood the displayed697formula gives the unique Nash state. Relabeling commutes698with the zero-mean projection, and uniqueness makes699these locally computed states agree wherever the labels700change. This is an explicit instance of transport of a701solution in a parameter family.702 703\section{A local margin and continuation on a face}704 705The compact-direction argument yields a useful quantitative706version. If \(p\) is an ESS and the feasible unit set is707nonempty, there are constants \(c,\delta>0\) such that708\[709 u_A(p,x)-u_A(x,x)\ \geq\ c\|x-p\|^2710 \qquad(\|x-p\|<\delta).711\]712Indeed, if the critical unit set is nonempty, choose \(a>0\)713with \(Q_A\leq-a\) there. On the directions where714\(Q_A<-a/2\), the difference at \(v=rw\) is at least715\((a/2)r^2\). On the closed complement the minimum of716\(L_p\) is some \(m>0\). For717\(r<\min\{1,m/(2(1+M))\}\), the difference is at least718\((m/2)r\geq(m/2)r^2\). If either region is empty only719the other bound is needed. When the critical unit set720is empty the second estimate applies on all directions.721The lower bound is an estimate on payoffs. Its use722as an estimate on motion requires a dynamics whose723drift is related to those payoffs.724 725At a Nash point, let \(B_p=\{i:(Ap)_i=M_p\}\), where726\(M_p=\max_i(Ap)_i\). Since \(\sum_i v_i=0\), one has727\[728 L_p(v)=\sum_{i\notin B_p}\bigl(M_p-(Ap)_i\bigr)v_i .