Let ( U , β ˉ 0 , β ˉ 1 ) (U,\bar{\beta}_0,\bar{\beta}_1) ( U , β ˉ 0 , β ˉ 1 ) be twice continuously differentiable affine-controlled rate data on l l l states with control set A \mathcal{A} A and derivative bound K 0 K_0 K 0 , and let Δ l \Delta^l Δ l be the probability simplex . For each ordered pair ( σ , γ ) (\sigma,\gamma) ( σ , γ ) with σ ≠ γ \sigma\neq\gamma σ = γ in { 1 , … , l } \{1,\dots,l\} { 1 , … , l } let β 0 ( σ , γ , ⋅ ) \beta_0(\sigma,\gamma,\cdot) β 0 ( σ , γ , ⋅ ) and β 1 ( σ , γ , ⋅ ) \beta_1(\sigma,\gamma,\cdot) β 1 ( σ , γ , ⋅ ) denote the restrictions of β ˉ 0 ( σ , γ , ⋅ ) \bar{\beta}_0(\sigma,\gamma,\cdot) β ˉ 0 ( σ , γ , ⋅ ) and β ˉ 1 ( σ , γ , ⋅ ) \bar{\beta}_1(\sigma,\gamma,\cdot) β ˉ 1 ( σ , γ , ⋅ ) to Δ l \Delta^l Δ l , and let β ˉ 1 k \bar{\beta}_1^k β ˉ 1 k denote the k k k -th component of β ˉ 1 \bar{\beta}_1 β ˉ 1 for k ∈ { 1 , … , m } k\in\{1,\dots,m\} k ∈ { 1 , … , m } . Write ∣ ⋅ ∣ |\cdot| ∣ ⋅ ∣ for the Euclidean norm , x ⋅ y x\cdot y x ⋅ y for the dot product , and ⋅ \sqrt{\cdot} ⋅ for the nonnegative square root . Put
R = sup α ∈ A ∣ α ∣ , Λ = l m K 0 , K = K 0 ( 1 + m ( R + 1 ) ) , R=\sup_{\alpha\in\mathcal{A}}|\alpha|,\qquad \Lambda=\sqrt{l}\,\sqrt{m}\,K_0,\qquad K=K_0\big(1+\sqrt{m}\,(R+1)\big), R = α ∈ A sup ∣ α ∣ , Λ = l m K 0 , K = K 0 ( 1 + m ( R + 1 ) ) ,
let
V = { α ∈ R m : ρ ( α ) < 1 } , V=\{\alpha\in\mathbb{R}^m:\ \rho(\alpha)<1\}, V = { α ∈ R m : ρ ( α ) < 1 } ,
where ρ ( α ) \rho(\alpha) ρ ( α ) denotes the distance from α \alpha α to A \mathcal{A} A , and define
β ˉ ( σ , γ , Σ , α ) = β ˉ 0 ( σ , γ , Σ ) + β ˉ 1 ( σ , γ , Σ ) ⋅ α ( Σ ∈ U , α ∈ V ) . \bar{\beta}(\sigma,\gamma,\Sigma,\alpha)=\bar{\beta}_0(\sigma,\gamma,\Sigma)+\bar{\beta}_1(\sigma,\gamma,\Sigma)\cdot\alpha\qquad(\Sigma\in U,\ \alpha\in V). β ˉ ( σ , γ , Σ , α ) = β ˉ 0 ( σ , γ , Σ ) + β ˉ 1 ( σ , γ , Σ ) ⋅ α ( Σ ∈ U , α ∈ V ) .
1. (Restriction.) ( β 0 , β 1 ) (\beta_0,\beta_1) ( β 0 , β 1 ) is an affine-controlled transition-rate family on l l l states with control set A \mathcal{A} A and Lipschitz constant Λ \Lambda Λ . Let β \beta β and B B B denote the transition-rate family and the rate bound determined by ( β 0 , β 1 ) (\beta_0,\beta_1) ( β 0 , β 1 ) as in the lemma on affine-controlled data .
2. (Control neighbourhood.) V V V is an open , convex , bounded subset of Euclidean space R m \mathbb{R}^m R m with A ⊆ V \mathcal{A}\subseteq V A ⊆ V , and ∣ α ∣ ≤ R + 1 |\alpha|\le R+1 ∣ α ∣ ≤ R + 1 for every α ∈ V \alpha\in V α ∈ V .
3. (Extension.) ( U , V , β ˉ ) (U,V,\bar{\beta}) ( U , V , β ˉ ) is a twice continuously differentiable extension of β \beta β with derivative bound K K K .
4. (Affine dependence on the control.) Points of U × V U\times V U × V are written x = ( Σ , α ) x=(\Sigma,\alpha) x = ( Σ , α ) with coordinates indexed as in that definition, and ∂ i \partial_i ∂ i denotes the partial derivative with respect to the coordinate x i x_i x i . For every ordered pair ( σ , γ ) (\sigma,\gamma) ( σ , γ ) with σ ≠ γ \sigma\neq\gamma σ = γ , all k , k ′ ∈ { 1 , … , m } k,k'\in\{1,\dots,m\} k , k ′ ∈ { 1 , … , m } , every i ∈ { 1 , … , l } i\in\{1,\dots,l\} i ∈ { 1 , … , l } and every ( Σ , α ) ∈ U × V (\Sigma,\alpha)\in U\times V ( Σ , α ) ∈ U × V ,
∂ l + k β ˉ ( σ , γ , Σ , α ) = β ˉ 1 k ( σ , γ , Σ ) , ∂ l + k ′ ∂ l + k β ˉ ( σ , γ , Σ , α ) = 0 , ∂ l + k ∂ i β ˉ ( σ , γ , Σ , α ) = ∂ i β ˉ 1 k ( σ , γ , Σ ) . \partial_{l+k}\bar{\beta}(\sigma,\gamma,\Sigma,\alpha)=\bar{\beta}_1^k(\sigma,\gamma,\Sigma),\qquad \partial_{l+k'}\partial_{l+k}\bar{\beta}(\sigma,\gamma,\Sigma,\alpha)=0,\qquad \partial_{l+k}\partial_i\bar{\beta}(\sigma,\gamma,\Sigma,\alpha)=\partial_i\bar{\beta}_1^k(\sigma,\gamma,\Sigma). ∂ l + k β ˉ ( σ , γ , Σ , α ) = β ˉ 1 k ( σ , γ , Σ ) , ∂ l + k ′ ∂ l + k β ˉ ( σ , γ , Σ , α ) = 0 , ∂ l + k ∂ i β ˉ ( σ , γ , Σ , α ) = ∂ i β ˉ 1 k ( σ , γ , Σ ) .