1: \begin{definition}\longthmtitle{Incentive to deviate from

2:     \baalgo}\label{def:incentive-deviate}

3:   Let $r > 0$ and assume that the stepsizes for any execution

4:   of~\eqref{eq:deviation-dyn} satisfy the hypotheses of

5:   Theorem~\ref{th:convergence}. Then, the generator $\nt \in

6:   \until{N}$ has an \emph{incentive to deviate} from the \baalgo if

7:   there exists an execution of~\eqref{eq:deviation-dyn} and $l \in

8:   \integerspositive$ such that

9:     \begin{align}

10:       u_{\nt}(b_{\nt}(k), x_{\nt}^{\opt}(k)) > u_{\nt}^{\max},

11:       \label{eq:strong-dev-crit}

12:     \end{align}

13:     for all $k \ge l$, where 

14:     \begin{align} 

15:       u_{\nt}^{\max} := \max \Bigl\{u_{\nt}(b_{\nt}, &x_{\nt}^\opt(b))

16:       \Bigl|  \norm{b-b^*} \le \Bigl(1+\frac{B(r)}{2 a_{\max}} \Bigr) r

17:       \notag

18:       \\

19:       & \quad \text{ and } x^\opt (b) \in \solone(b)

20:     \Bigr\}. \label{eq:umax}

21:   \end{align}

22: %    The generator $\nt$ is said to have a \emph{weak incentive to

23: %    deviate} from the \baalgo if there exists an execution

24: %    of~\eqref{eq:deviation-dyn} along which 

25: %    \begin{align}

26: %      \limsup_{k \to \infty}  u_{\nt} (b_{\nt}(k), x_{\nt}^{\opt}(b(k)))

27: %      > u_{\nt}^{\max}. \label{eq:weak-dev-crit}

28: %    \end{align}

29: \end{definition}

30: