Local linearization method
≈ 39 min read
Names in other languages
No names in other languages are held for this entry yet.
Article bodies are English and are not machine-translated. A name in another language is not a translation of the article.
In numerical analysis, the local linearization (LL) method is a general strategy for designing numerical integrators for differential equations based on a local (piecewise) linearization of the given equation on consecutive time intervals. The numerical integrators are then iteratively defined as the solution of the resulting piecewise linear equation at the end of each consecutive interval. The LL method has been developed for a variety of equations such as the ordinary, delayed, random and stochastic differential equations. The LL integrators are key component in the implementation of inference methods for the estimation of unknown parameters and unobserved variables of differential equations given time series of (potentially noisy) observations. The LL schemes are ideals to deal with complex models in a variety of fields as neuroscience, finance, forestry management, control engineering, mathematical statistics, etc.
In numerical analysis, the local linearization (LL) method is a general strategy for designing numerical integrators for differential equations based on a local (piecewise) linearization of the given equation on consecutive time intervals. The numerical integrators are then iteratively defined as the solution of the resulting piecewise linear equation at the end of each consecutive interval. The LL method has been developed for a variety of equations such as the ordinary, delayed, random and stochastic differential equations. The LL integrators are key component in the implementation of inference methods for the estimation of unknown parameters and unobserved variables of differential equations given time series of (potentially noisy) observations. The LL schemes are ideals to deal with complex models in a variety of fields as neuroscience, finance, forestry management, control engineering, mathematical statistics, etc. High-order local linearization (HOLL) method is a generalization of the Local Linearization method oriented to obtain high-order integrators for differential equations that preserve the stability and dynamics of the linear equations. The integrators are obtained by splitting, on consecutive time intervals, the solution x of the original equation in two parts: the solution z of the locally linearized equation plus a high-order approximation of the residual \mathbf{r}= \mathbf{x}-\mathbf{z} . \frac{d\mathbf{x}\left( t\right) }{dt}=\mathbf{f}\left( t,\mathbf{x}\left( t\right) \right) ,\qquad t\in \left[ t_{0},T\right], \qquad \qquad \qquad \qquad (4.1) with initial condition \mathbf{x}(t_{0})=\mathbf{x}_{0} , where \mathbf{f} is a differentiable function. Let \left( t\right) _{h}=\{t_{n}:n=0,..,N\} be a time discretization of the time interval [t_{0},T] with maximum stepsize h such that t_{n} and h_{n}=t_{n+1}-t_{n}\leq h . After the local linearization of the equation (4.1) at the time step t_{n} the variation of constants formula yields \mathbf{x}(t_n+h)=\mathbf{x}(t_n)+\mathbf{\phi }(t_n,\mathbf{x}(t_n);h)+\mathbf{r}(t_n,\mathbf{x}(t_n);h), where \mathbf{\phi }(t_n,\mathbf{z}_n;h)=\int\limits_0^h e^{\mathbf{f}_{\mathbf{x}}(t_n,\mathbf{z}_n) (h-s)}(\mathbf{f} (t_n, \mathbf{z}_n) + \mathbf{f}_t(t_n,\mathbf{z}_n) s) \, ds \qquad results from the linear approximation, and \mathbf{r}(t_n,\mathbf{z}_n;h)=\int\limits_0^h e^{\mathbf{f}_{\mathbf{x}} (t_n,\mathbf{z}_n) (h-s)}\mathbf{g}_n(s,\mathbf{x}(t_n+s)) \, ds, \qquad \qquad \qquad (4.2) is the residual of the linear approximation. Here, \mathbf{f}_{\mathbf{x}} and \mathbf{f}_{t} denote the partial derivatives of f with respect to the variables x and t, respectively, and \mathbf{g}_n(s,\mathbf{u})=\mathbf{f}(s,\mathbf{u})-\mathbf{f}_{\mathbf{x}}(t_n,\mathbf{z}_n) \mathbf{u}-\mathbf{f}_t (t_n,\mathbf{z}_n) (s-t_n)-\mathbf{f}(t_n,\mathbf{z}_n) +\mathbf{f}_{\mathbf{x}}(t_n,\mathbf{z}_n)\mathbf{z}_n. \mathbf{z}_{n+1}=\mathbf{z}_n+\mathbf{\phi}(t_n,\mathbf{z}_n;h_n), \qquad \text{ with } \quad \mathbf{z}_0=\mathbf{x}_0. \qquad \qquad \qquad \qquad (4.3) The Local Linear discretization (4.3) converges with order 2 to the solution of nonlinear ODEs, but it match the solution of the linear ODEs. The recursion (4.3) is also known as Exponential Euler discretization. \mathbf{z}_{n+1}=\mathbf{z}_n+\mathbf{\phi}(t_n,\mathbf{z}_n;h_n) + \widetilde{\mathbf{r}}(t_n,\mathbf{z}_n;h_n),\qquad \text{ with } \quad \mathbf{z}_0=\mathbf{x}_0, \qquad \qquad \qquad(4.4) where \tilde{r} is an order \alpha (> 2) approximation to the residual r (i.e., \left\vert \mathbf{r}(t_n, \mathbf{z}_n;h)-\widetilde{\mathbf{r}}(t_n,\mathbf{z}_n;h)\right\vert \propto h^{\alpha+1 }). The HOLL discretization (4.4) converges with order \alpha to the solution of nonlinear ODEs, but it match the solution of the linear ODEs. HOLL discretizations can be derived in two ways: 1) (quadrature-based) by approximating the integral representation (4.2) of r; and 2) (integrator-based) by using a numerical integrator for the differential representation of r defined by \frac{d\mathbf{r}(t) }{dt} = \mathbf{q}(t_n,\mathbf{z}_n;t \mathbf{,\mathbf{r}}(t) \mathbf), \qquad \text{ with } \qquad \mathbf{r}(t_n) =\mathbf{0,} \qquad \qquad \qquad (4.5) for all t\in \lbrack t_k,t_{k+1}] , where \mathbf{q}(t_{n},\mathbf{z}_{n};s\mathbf{,\xi })=\mathbf{f}(s,\mathbf{z}_{n}+ \mathbf{\phi }\left( t_{n},\mathbf{z}_{n};s-t_{n}\right) +\mathbf{\xi })- \mathbf{f}_{\mathbf{x}}(t_{n},\mathbf{z}_{n})\mathbf{\phi } ( t_n, \mathbf{z}_n;s-t_n) -\mathbf{f}_t( t_n,\mathbf{z}_n) (s-t_n)-\mathbf{f}( t_n,\mathbf{z}_n). HOLL discretizations are, for instance, the followings: Locally Linearized Runge Kutta discretization \qquad \mathbf{z}_{n+1}=\mathbf{z}_n+\mathbf{\phi }(t_n,\mathbf{z}_n;h_n)+h_n \sum_{j=1}^s b_j \mathbf{k}_j,\quad \text{ with } \quad \mathbf{k}_i = \mathbf{q}(t_n,\mathbf{z}_n;\text{ }t_n + c_i h_n \mathbf{,} \mathbf{} h_n \sum_{j=1}^{i-1} a_{ij}\mathbf{k}_j), which is obtained by solving (4.5) via a s-stage explicit Runge–Kutta (RK) scheme with coefficients \mathbf{c}=\left[ c_{i}\right] , \mathbf{A}=\left[ a_{ij}\right] \quad and \quad \mathbf{b}=\left[ b_{j}\right] . Local linear Taylor discretization \mathbf{z}_{n+1}=\mathbf{z}_n+\mathbf{\phi}(t_n,\mathbf{z}_n;h_n)+\int_0^{h_n}e^{(h_n-s) \mathbf{f}_{\mathbf{x}} \left( t_n,\mathbf{z}_n\right) } \sum_{j=2}^p\frac{\mathbf{c}_{n,j}}{j!} s^j \, ds,\text{ with } \mathbf{c}_{n,j}=\left( \frac{d^{j+1}\mathbf{x}(t)}{dt^{j+1}}-\mathbf{f}_{\mathbf{x}} (t_n,\mathbf{z}_n) \frac{d^{j}\mathbf{x}(t) }{dt^j}\right) \mid _{t=\mathbf{z}_n}, which results from the approximation of \mathbf{g}_{n} in (4.2) by its order-p truncated Taylor expansion. Multistep-type exponential propagation discretization \mathbf{z}_{n+1}=\mathbf{z}_n+\mathbf{\phi}(t_n,\mathbf{z}_n;h)+h\sum_{j=0}^{p-1}\gamma_j\nabla^j\mathbf{g}_n(t_n,\mathbf{z }_{n}), \quad with \quad \gamma_j =(-1)^j \int\limits_0^1 e^{(1-\theta) h\mathbf{f}_{\mathbf{x}} (t_n,\mathbf{z}_n) } \left( \begin{array}{c} -\theta \\ j \end{array} \right) d\theta , which results from the interpolation of \mathbf{g}_{n} in (4.2) by a polynomial of degree p on t_{n},\ldots, t_{n-p+1} , where \nabla ^{j}\mathbf{g}_{n}(t_{m},\mathbf{z}_{m}) denotes the j-th backward difference of \mathbf{g}_{n}(t_{m},\mathbf{z}_{m}) . Runge Kutta type Exponential Propagation discretization \mathbf{z}_{n+1}=\mathbf{z}_n+\mathbf{\phi}(t_n,\mathbf{z}_n ;h) + h\sum_{j=0}^{p-1} \gamma _{j,p}\nabla^j \mathbf{g}_n (t_n,\mathbf{z}_n),\quad \text{ with } \quad \gamma_{j,p} = \int\limits_0^1 e^{(1-\theta )h\mathbf{f}_{\mathbf{x}} (t_n,\mathbf{z}_n) } \left( \begin{array}{c} \theta p\\ j \end{array} \right) d\theta , which results from the interpolation of \mathbf{g}_{n} in (4.2) by a polynomial of degree p on t_{n},\ldots, t_{n}+(p-1)h/p , Linealized exponential Adams discretization \mathbf{z}_{n+1}=\mathbf{z}_n+\mathbf{\phi }(t_n,\mathbf{z}_n;h)+h\sum_{j=1}^{p-1}\sum_{l=1}^j\frac{\gamma _{j+1}}{l} \nabla^l\mathbf{g}_n(t_n,\mathbf{z}_{n}),\quad \text{ with } \quad \gamma_{j+1}=(-1)^{j+1} \int\limits_0^1e^{(1-\theta )h\mathbf{f}_{\mathbf{x}} \left( t_n,\mathbf{z}_n\right) }\theta \left( \begin{array}{c} -\theta \\ j \end{array} \right) d\theta , which results from the interpolation of \mathbf{g}_{n} in (4.2) by a Hermite polynomial of degree p on t_{n},\ldots, t_{n-p+1} . \phi_j(\mathbf{A},h)=\int\limits_0^h e^{(h-s)\mathbf{A}} s^{j-1} \, ds,\qquad j=1,2\ldots, where A is a d × d matrix. Every numerical implementation \mathbf{y}_n of the LL (or of a HOLL) \mathbf{z}_{n} of any order is generically called Local Linearization scheme. \sum\nolimits_{i=1}^l \phi_i(\mathbf{A},h)\mathbf{a}_i = \mathbf{L}e^{h\mathbf{H}}\mathbf{r}, where \mathbf{a}_i are d-dimensional vectors, \mathbf{H}= \begin{bmatrix} \mathbf{A} & \mathbf{v}_{l} & \mathbf{v}_{l-1} & \cdots & \mathbf{v}_{1} \\ \mathbf{0} & \mathbf{0} & 1 & \cdots & 0 \\ \mathbf{0} & \mathbf{0} & 0 & \ddots & 0 \\ \vdots & \vdots & \vdots & \ddots & 1 \\ \mathbf{0} & \mathbf{0} & 0 & \cdots & 0 \end{bmatrix} \in \mathbb{R}^{(d+l)\times (d+l)}, \mathbf{L}=[\mathbf{I} \quad \mathbf{0}_{d\times l}] , \mathbf{r}=[\mathbf{0}_{1\times (d+l-1)}\quad1]^{\intercal}, \mathbf{v}_{i}=\mathbf{a}_{i}(i-1)! , being \mathbf{I} the d-dimensional identity matrix. If \mathbf{P}_{p,q}(2^{-k}\mathbf{H}h) denotes the (p; q)-Padé approximation of e^{2^{-k}\mathbf{H}h} and k is the smallest natural number such that |2^{-k}\mathbf{H}h|\leq \frac{1}{2}, then \left\vert \sum\nolimits_{i=1}^l \phi_i (\mathbf{A},h)\mathbf{a}_{i}- \mathbf{L}\left( \mathbf{\mathbf{P}}_{p,q}(2^{-k}\mathbf{H}h)\right)^{2^k} \mathbf{r}\right\vert \varpropto h^{p+q+1}. If \mathbf{\mathbf{k}}_{m,k}^{p,q}(h,\mathbf{H},\mathbf{r}) denotes the (m; p; q; k) Krylov-Padé approximation of e^{h\mathbf{H}}\mathbf{r} , then \left\vert \sum\nolimits_{i=1}^{l}\phi _{i}(\mathbf{A},h)\mathbf{a}_i - \mathbf{L\mathbf{k}}_{m,k}^{p,q}(h,\mathbf{H},\mathbf{r})\right\vert \varpropto h^{\min ({m,p+q+1})}, where m \leq d is the dimension of the Krylov subspace. \mathbf{y}_{n+1}=\mathbf{y}_{n}+\mathbf{L}(\mathbf{P}_{p,q}(2^{-k_{n}} \mathbf{M}_{n}h_{n}))^{2^{k_{n}}}\mathbf{r,} \qquad \qquad (4.6) where the matrices \mathbf{M}_n , L and r are defined as \mathbf{M}_n = \begin{bmatrix} \mathbf{f}_{\mathbf{x}}(t_n,\mathbf{y}_n) & \mathbf{f}_t(t_n,\mathbf{y}_n) & \mathbf{f}(t_n,\mathbf{y}_n) \\ 0 & 0 & 1 \\ 0 & 0 & 0 \end{bmatrix} \in \mathbb{R}^{(d+2)\times (d+2)}, \mathbf{L}=\left[ \begin{array}{ll} \mathbf{I} & \mathbf{0}_{d\times 2} \end{array} \right] and \mathbf{r}^{\intercal }=\left[ \begin{array}{ll} \mathbf{0}_{1\times (d+1)} & 1 \end{array} \right] with p+q>1 . For large systems of ODEs \mathbf{y}_{n+1}=\mathbf{y}_{n}+\mathbf{L\mathbf{k}} _{m_{n},k_{n}}^{p,q}(h_{n},\mathbf{M}_{n},\mathbf{r})\mathbf{,}\qquad \text{ with } \qquad m_{n}>2. \mathbf{y}_{n+1}=\mathbf{y}_n+\mathbf{L}_{1}(\mathbf{P}_{p,q}(2^{-k_n} \mathbf{T}_n h_n))^{2^{k_n}}\mathbf{r}_1 \mathbf{,} \qquad \qquad (4.7) where for autonomous ODEs the matrices \mathbf{T}_{n}, \mathbf{L}_{1} and \mathbf{r}_{1} are defined as \mathbf{T}_{n}=\left[ \begin{array}{cccc} \mathbf{f}_{\mathbf{x}}(\mathbf{y}_{n}) & (\mathbf{I}\otimes \mathbf{f} ^{\intercal }(\mathbf{y}_{n}))\mathbf{f}_{\mathbf{xx}}(\mathbf{y}_{n}) \mathbf{f}(\mathbf{y}_{n}) & \mathbf{0} & \mathbf{f}(\mathbf{y}_{n}) \\ 0 & 0 & 0 & 0 \\ 0 & 0 & 0 & 1 \\ 0 & 0 & 0 & 0 \end{array} \right] \in \mathbb{R}^{(d+3)\times (d+3)}, \mathbf{L}_{1}=\left[ \begin{array}{ll} \mathbf{I} & \mathbf{0}_{d\times 3} \end{array} \right] \quad and \quad \mathbf{r}_{1}^{\intercal }=\left[ \begin{array}{ll} \mathbf{0}_{1\times (d+2)} & 1 \end{array} \right] . Here, \mathbf{f}_{\mathbf{xx}} denotes the second derivative of f with respect to x, and p + q > 2. For large systems of ODEs \mathbf{y}_{n+1}=\mathbf{y}_{n}+\mathbf{L\mathbf{k}} _{m_{n},k_{n}}^{p,q}(h_{n},\mathbf{T}_{n},\mathbf{r})\mathbf{,}\qquad \text{ with } \qquad m_{n}>3. \mathbf{y}_{n+1}=\mathbf{y}_{n}+\mathbf{u}_{4}+\frac{h_{n}}{6}(2\mathbf{k} _{2}+2\mathbf{k}_{3}+\mathbf{k}_{4}),\quad \qquad \qquad (4.8) where \mathbf{u}_{j}=\mathbf{L}(\mathbf{P}_{p,q}(2^{-\kappa _{j}}\mathbf{M} _{n}c_{j}h_{n}))^{2^{\kappa _{j}}}\mathbf{r} and \mathbf{k}_{j}=\mathbf{f}\left( t_{n}+c_{j}h_{n},\mathbf{y}_{n}+\mathbf{u} _{j}+c_{j}h_{n}\mathbf{k}_{j-1}\right) -\mathbf{f}\left( t_{n},\mathbf{y} _{n}\right) -\mathbf{f}_{\mathbf{x}}\left( t_{n},\mathbf{y}_{n}\right) \mathbf{u}_{j}\ -\mathbf{f}_{t}\left( t_{n},\mathbf{y}_{n}\right) c_{j}h_{n}, with \mathbf{k}_{1}\equiv \mathbf{0}, c=\left[ \begin{array}{cccc} 0 & \frac{1}{2} & \frac{1}{2} & 1 \end{array} \right] , and p + q > 3. For large systems of ODEs, the vector \mathbf{u}_{j} in the above scheme is replaced by \mathbf{u}_{j}=\mathbf{L\mathbf{k} }_{m_{j},k_{j}}^{p,q}(c_{j}h_{n},\mathbf{M}_{n},\mathbf{r}) with m_j > 4. \mathbf{y}_{n+1}=\mathbf{y}_n+\mathbf{u}_s+h_n\sum_{j=1}^s b_j \mathbf{k}_j \qquad \text{ and } \qquad \widehat{\mathbf{y}}_{n+1}=\mathbf{y}_n+\mathbf{u}_s+h_n\sum_{j=1}^s \widehat{b}_j \mathbf{k}_j,\quad \qquad \qquad (4.9) where s = 7 is the number of stages, \mathbf{k}_j=\mathbf{f(}t_n+c_j h_n,\mathbf{y}_n+\mathbf{u}_j + h_n \sum_{i=1}^{s-1} a_{j,i} \mathbf{k}_i)-\mathbf{f}\left( t_n, \mathbf{y}_n\right) -\mathbf{f}_{\mathbf{x}}\left( t_n,\mathbf{y}_n\right) \mathbf{u}_j\ -\mathbf{f}_t\left( t_n,\mathbf{y}_n \right) c_j h_n, with \mathbf{k}_{1}\equiv \mathbf{0} , and a_{j,i}, b_{j}, \widehat{b}_j \quad and \quad c_j are the Runge–Kutta coefficients of Dormand and Prince and p + q > 4. The vector \mathbf{u}_j in the above scheme is computed by a Padé or Krylor–Padé approximation for small or large systems of ODE, respectively. , and the order-4 LLRK4 schemes (4.8) with step size h=1/2, and p=q=6. By construction, the LL and HOLL discretizations inherit the stability and dynamics of the linear ODEs, but it is not the case of the LL schemes in general. With p\leq q\leq p+2 , the LL schemes (4.6)-(4.9) are A-stable. With q = p + 1 or q = p + 2, the LL schemes (4.6)–(4.9) are also L-stable. For linear ODEs, the LL schemes (4.6)-(4.9) converge with order p + q. In addition, with p = q = 6 and m_{n} = d, all the above described LL schemes yield to the ″exact computation″ (up to the precision of the floating-point arithmetic) of linear ODEs on the current personal computers. This includes stiff and highly oscillatory linear equations. Moreover, the LL schemes (4.6)-(4.9) are regular for linear ODEs and inherit the symplectic structure of Hamiltonian harmonic oscillators. These LL schemes are also linearization preserving, and display a better reproduction of the stable and unstable manifolds around hyperbolic equilibrium points and periodic orbits that other numerical schemes with the same stepsize. For instance, Figure 1 shows the phase portrait of the ODEs \begin{align} & \frac{dx_1}{dt} =-2x_1+x_2+1-\mu f(x_1,\lambda) \qquad \qquad (4.10) \\[6pt] & \frac{dx_{2}}{dt} = x_1-2x_2+1-\mu f(x_2,\lambda) \qquad \qquad \quad (4.11) \end{align} with f(u,\lambda) =u(1+u+\lambda u^2)^{-1} , \mu =15 and \lambda =57 , and its approximation by various schemes. This system has two stable stationary points and one unstable stationary point in the region 0\leq x_1,x_2\leq 1 . \frac{d\mathbf{x}(t) }{dt} = \mathbf{f}(t,\mathbf{x}(t) ,\mathbf{x}_t(-\tau_1),\ldots ,\mathbf{x}_t(-\tau_m)),\qquad t\in[t_0,T] , \qquad\qquad (5.1) with m constant delays \tau_i>0 and initial condition \mathbf{x}_{t_0}(s)=\mathbf{\varphi}(s) for all s\in[-\tau,0], where f is a differentiable function, \mathbf{x}_t:[-\tau,0] \longrightarrow \mathbb{R}^d is the segment function defined as \mathbf{x}_t(s):=\mathbf{x}(t+s),\text{ } s\in[-\tau,0], for all t\in[t_0,T], \mathbf{\varphi }:[-\tau,0] \longrightarrow \mathbb{R}^d is a given function, and \tau = \max \left\{ \tau_1,\ldots,\tau_m\right\} . \mathbf{z}_{n+1}=\mathbf{z}_n+\Phi (t_n,\mathbf{z}_n,h_n;\widetilde{\mathbf{z}}_{t_n}^1, \ldots,\widetilde{\mathbf{z}}_{t_n}^m), \qquad \qquad (5.2) where \Phi (t_n,\mathbf{z}_n,h_n;\widetilde{\mathbf{z}}_{t_n}^1, \ldots, \widetilde{\mathbf{z}}_{t_{n}}^{m}) = \int\limits_0^{h_n}e^{\mathbf{A}_n(h_n-u)} \left[\sum\limits_{i=1}^m \mathbf{B}_n^i (\widetilde{\mathbf{z}}_{t_n}^i (u-\tau_i) -\widetilde{\mathbf{z}}_{t_n}^i (-\tau_i) )+\mathbf{d}_n\right] \, du + \int \limits_0^{h_n}\int\limits_0^u e^{\mathbf{A}_n(h_n-u)}\mathbf{c}_n \, dr \, du \widetilde{\mathbf{z}}_{t_n}^i:\left[ -\tau_i,0\right] \longrightarrow \mathbb{R}^d is the segment function defined as \widetilde{\mathbf{z}}_{t_n}^i(s):=\widetilde{\mathbf{z}}^i(t_n+s), \text{ } s\in [-\tau_i,0] , and \widetilde{\mathbf{z}}^i:\left[ t_n-\tau_i,t_n\right] \longrightarrow \mathbb{R}^d is a suitable approximation to \mathbf{x}(t) for all t\in \lbrack t_n-\tau_i,t_n] such that \widetilde{\mathbf{z}}^i(t_n)=\mathbf{z}_n. Here, \mathbf{A}_n=\mathbf{f}_x(t_n,\mathbf{z}_n,\widetilde{\mathbf{z}}_{t_n}^1(-\tau_1),\ldots,\widetilde{\mathbf{z}}_{t_n}^m(-\tau_d)), \text{ }\mathbf{B}_n^i=\mathbf{f}_{x_t(-\tau_i)}(t_n,\mathbf{z}_n,\widetilde{\mathbf{z}}_{t_n}^1(-\tau_1),\ldots,\widetilde{\mathbf{z}}_{t_n}^m(-\tau_d)) are constant matrices and \mathbf{c}_n=\mathbf{f}_t(t_n,\mathbf{z}_n,\widetilde{\mathbf{z}}_{t_n}^1 (-\tau_1),\ldots,\widetilde{\mathbf{z}}_{t_n}^m(-\tau_d)) \text{ and }\mathbf{d}_n=\mathbf{f(}t_n,\mathbf{z}_n,\widetilde{\mathbf{z}}_{t_n}^1(-\tau_1),\ldots,\widetilde{\mathbf{z}}_{t_n}^m(-\tau_d)) are constant vectors. \mathbf{f}_{t}, \mathbf{f}_{x} \quad and \quad \mathbf{f} _{x_{t}(-\tau _{i})} denote, respectively, the partial derivatives of f with respect to the variables t and x, and \mathbf{x}_{t}(-\tau _{i}) . The Local Linear discretization (5.2) converges to the solution of (5.1) with order \alpha =\min \{2,r\}, if \widetilde{\mathbf{z}}_{t_{n}}^{i} approximates \mathbf{z}_{t_{n}}^{i} with order r \quad (i.e., \left\vert \mathbf{z}_{t_{n}}^{i}\mathbf{(}u-\tau _{i} \mathbf{)}-\widetilde{\mathbf{z}}_{t_{n}}^{i}\mathbf{(}u-\tau _{i}\mathbf{)} \right\vert \propto h_{n}^{r} for all u\in \lbrack 0,h_{n}]) . [[File:Figure DDE.png|thumb|450x450px|Fig. 2 Approximate paths of the Marchuk et al. (1991) antiviral immune model described by a stiff system of ten-dimensional nonlinear DDEs with five time delays: top, continuous Runge–Kutta (2,3) scheme; bottom, LL scheme (5.3). Step-size h = 0.01 fixed, and p = q = 6. Depending on the approximations \widetilde{\mathbf{z}}_{t_{n}}^{i} and on the algorithm to compute \mathbf{\phi } different Local Linearizations schemes can be defined. Every numerical implementation \mathbf{y}_{n} of a Local Linear discretization \mathbf{z}_{n} is generically called local linearization scheme. \mathbf{y}_{n+1}=\mathbf{y}_{n}+\mathbf{L}(\mathbf{P}_{p,q}(2^{-k_{n}} \mathbf{M}_{n}h_{n}))^{2^{k_{n}}}\mathbf{r,} \quad \qquad (5.3) where the matrices \mathbf{M}_{n}, \mathbf{L} and \mathbf{r} are defined as \mathbf{M}_{n}= \begin{bmatrix} \mathbf{A}_{n} & \mathbf{c}_{n}+\sum\limits_{i=1}^{m}\mathbf{B}_{n}^{i} \mathbf{\alpha }_{n}^{i} & \mathbf{d}_{n} \\ 0 & 0 & 1 \\ 0 & 0 & 0 \end{bmatrix} \in \mathbb{R}^{(d+2)\times (d+2)}, \mathbf{L}=\left[ \begin{array}{ll} \mathbf{I} & \mathbf{0}_{d\times 2} \end{array} \right] and \mathbf{r}^{\intercal }=\left[ \begin{array}{ll} \mathbf{0}_{1\times (d+1)} & 1 \end{array} \right], h_{n}\leq \tau , and p+q>1 . Here, the matrices \mathbf{A}_{n} , \mathbf{B}_{n}^{i} , \mathbf{c}_{n} and \mathbf{d}_{n} are defined as in (5.2), but replacing \mathbf{z} by \mathbf{y} and \mathbf{\alpha }_{n}^{i}=(\mathbf{y}(t_{n+1}-\tau _{i})-\mathbf{y} (t_{n}-\tau _{i}))/h_{n}, where \mathbf{y}\left( t\right) =\mathbf{y}_{n_{t}}+\mathbf{L}(\mathbf{P}_{p,q}(2^{-k_{n}} \mathbf{M}_{n_{t}}(t-t_{n_{t}})))^{2^{k_{n}}}\mathbf{r}, with n_{t}=\max \{n=0,1,2,...,:t_{n}\leq t\text{ and }t_{n}\in \left( t\right) _{h}\} , is the Local Linear Approximation to the solution of (5.1) defined through the LL scheme (5.3) for all t\in \lbrack t_{0},t_{n}] and by \mathbf{y}\left( t\right) =\mathbf{\varphi }\left( t\right) for t\in \left[ t_{0}-\tau ,t_{0}\right] . For large systems of DDEs \mathbf{y}_{n+1}=\mathbf{y}_{n}+\mathbf{L\mathbf{k}} _{m_{n},k_{n}}^{p,q}(h_{n},\mathbf{M}_{n},\mathbf{r})\quad and \quad \mathbf{y}\left( t\right) =\mathbf{y}_{n_{t}}+\mathbf{L\mathbf{k}} _{m_{n_{t}},k_{n_{t}}}^{p,q}(t-t_{n_{t}},\mathbf{M}_{n_{t}},\mathbf{r}), with p+q>1 and m_{n}>2 . Fig. 2 Illustrates the stability of the LL scheme (5.3) and of that of an explicit scheme of similar order in the integration of a stiff system of DDEs. \frac{d\mathbf{x}\left( t\right) }{dt}=\mathbf{f}(\mathbf{x}(t),\mathbf{\xi } (t)),\quad t\in \left[ t_{0},T\right] ,\qquad \qquad \qquad (6.1) with initial condition \mathbf{x}(t_0)=\mathbf{x}_0, where \mathbf{ \xi } is a k-dimensional separable finite continuous stochastic process, and f is a differentiable function. Suppose that a realization (path) of \mathbf{\xi } is given. \mathbf{z}_{n+1}=\mathbf{z}_{n}+\mathbf{\phi }(t_{n},\mathbf{z}_{n};h_{n}), \qquad \text{ with } \qquad \mathbf{z}_{0}=\mathbf{x}_{0}, where \mathbf{\phi }(t_n,\mathbf{z}_n;h_n)=\int\limits_0^{h_n} e^{\mathbf{f}_{\mathbf{x}} (\mathbf{z}_n,\mathbf{\xi}(t_n)) (h_n-u)}(\mathbf{f(z}_{n},\mathbf{\xi }(t_{n}))+\mathbf{f}_{\mathbf{\xi}}(\mathbf{z}_n,\mathbf{\xi }(t_{n}))(\widetilde{\mathbf{\xi }}(t_{n}+u)-\widetilde{\mathbf{\xi }}(t_n))) \, du and \widetilde{\mathbf{\xi }} is an approximation to the process \mathbf{ \xi } for all t\in \left[ t_{0},T\right]. Here, \mathbf{f}_{x} and \mathbf{f}_{\xi } denote the partial derivatives of \mathbf{f} with respect to \mathbf{x} and \xi , respectively. \mathbf{y}_{n+1}=\mathbf{y}_n+\mathbf{L}(\mathbf{P}_{p,q}(2^{-k_n} \mathbf{M}_n h_{n}))^{2^{k_n}}\mathbf{r,} \quad where the matrices \mathbf{M}_{n}, \quad \mathbf{L} \quad and \quad \mathbf{r} are defined as \mathbf{M}_{n}=\left[ \begin{array}{ccc} \mathbf{f}_{\mathbf{x}}\left( \mathbf{y}_{n},\mathbf{\xi }(t_{n})\right) & \mathbf{f}_{\mathbf{\xi }}(\mathbf{y}_{n},\mathbf{\xi }(t_{n})(\mathbf{\xi } (t_{n+1})-\mathbf{\xi }(t_{n}))/h_{n} & \mathbf{f}\left( \mathbf{y}_{n}, \mathbf{\xi }(t_{n})\right) \\ 0 & 0 & 1 \\ 0 & 0 & 0 \end{array} \right] \mathbf{L}=\left[ \begin{array}{ll} \mathbf{I} & \mathbf{0}_{d\times 2} \end{array} \right] , \mathbf{r}^{\intercal }=\left[ \begin{array}{ll} \mathbf{0}_{1\times (d+1)} & 1 \end{array} \right] , and p+q>1. For large systems of RDEs, \mathbf{y}_{n+1}=\mathbf{y}_{n}+\mathbf{L\mathbf{k}} _{m_{n},k_{n}}^{p,q}(h_{n},\mathbf{M}_{n},\mathbf{r}),\quad p+q>1 \quad and \quad m_{n}>2. The convergence rate of both schemes is min\{2,2\gamma \} , where is \gamma the exponent of the Holder condition of \mathbf{\xi } . Figure 3 presents the phase portrait of the RDE \frac{dx_{1}}{dt} =-x_{2}+\left( 1-x_{1}^{2}-x_{2}^{2}\right) x_{1}\sin (w^{H}(t))^{2}, \quad \qquad x_{1}(0)=0.8 \qquad (6.2) \frac{dx_{2}}{dt} =x_{1}+(1-x_{1}^{2}-x_{2}^{2})x_{2}\sin (w^{H}(t))^{2}, \qquad \qquad x_{2}(0)=0.1, \qquad (6.3) and its approximation by two numerical schemes, where w^{H} denotes a fractional Brownian process with Hurst exponent H=0.45. d\mathbf{x}(t)=\mathbf{f}(t,\mathbf{x}(t))dt+\sum\limits_{i=1}^{m}\mathbf{g} _{i}(t)d\mathbf{w}^{i}(t),\quad t\in \left[ t_{0},T\right] , \qquad \qquad \qquad (7.1) with initial condition \mathbf{x}(t_{0})=\mathbf{x}_{0} , where the drift coefficient \mathbf{f} and the diffusion coefficient \mathbf{g}_{i} are differentiable functions, and \mathbf{w=(\mathbf{w}}^{1},\ldots ,\mathbf{w} ^{m}\mathbf{)} is an m-dimensional standard Wiener process. For a time discretization \left( t\right) _{h} , the order- \mathbb{\gamma } (=1,1.5) Strong Local Linear discretization of the solution of the SDE (7.1) is defined by the recursive relation \mathbf{z}_{n+1}=\mathbf{z}_{n}+\mathbf{\phi }_{\mathbb{\gamma }}(t_{n}, \mathbf{z}_{n};h_{n})+\mathbf{\xi }(t_{n},\mathbf{z}_{n};h_{n}),\quad with \quad \mathbf{z}_{0}=\mathbf{x}_{0}, where \mathbf{\phi }_{\mathbb{\gamma }}(t_{n},\mathbf{z}_{n};\delta )=\int_{0}^{\delta }e^{\mathbf{f}_{\mathbf{x}}(t_{n},\mathbf{y}_{n})(\delta -u)}(\mathbf{f(}t_{n},\mathbf{z}_{n})+\mathbf{a}^{\mathbb{\gamma }}(t_{n}, \mathbf{z}_{n})u)du and \mathbf{\xi} \left( t_n,\mathbf{z}_n;\delta \right) = \sum\limits_{i=1}^m \int\nolimits_{t_n}^{t_n+\delta} e^{\mathbf{f}_{\mathbf{x}} (t_n,\mathbf{z}_n)(t_n+\delta -u)}\mathbf{g}_i(u) \, d\mathbf{w}^i(u). Here, \mathbf{a}^{\mathbb{\gamma}}(t_n,\mathbf{z}_n)= \left\{ \begin{array}{cl} \mathbf{f}_t(t_n,\mathbf{z}_n) & \text{for } \qquad \mathbb{\gamma }=1 \\ \mathbf{f}_t(t_n,\mathbf{z}_n) +\frac{1}{2} \sum\limits_{j=1}^m ( \mathbf{I}\otimes \mathbf{g}_j^\intercal (t_n)) \mathbf{f}_{\mathbf{xx}}(t_n, \mathbf{z}_n)\mathbf{g}_j (t_n) & \text{for } \quad \mathbb{\gamma}=1.5, \end{array} \right. \mathbf{f}_{\mathbf{x}}, \mathbf{f}_t denote the partial derivatives of \mathbf{f} with respect to the variables \mathbf{x} and t, respectively, and \mathbf{f}_{\mathbf{xx}} the Hessian matrix of \mathbf{f} with respect to \mathbf{x} . The strong Local Linear discretization \mathbf{z}_{n+1} converges with order \mathbb{\gamma } (= 1, 1.5) to the solution of (7.1). d\mathbf{r}(t) =\mathbf{q}_\gamma (t_n,\mathbf{z}_n;t \mathbf{,\mathbf{r}}(t)) \, dt + \sum\limits_{i=1}^m \mathbf{g}_i(t) \, d\mathbf{w}^i(t)\mathbf{,}\qquad \mathbf{r}(t_n) = \mathbf{0} for all t\in \lbrack t_n,t_{n+1}] , where \mathbf{q}_\gamma (t_n,\mathbf{z}_n;s\mathbf{,\xi })=\mathbf{f}(s,\mathbf{z}_n+\mathbf{\phi}_\gamma (t_n,\mathbf{z}_n;s-t_n) +\mathbf{\xi })-\mathbf{f}_{\mathbf{x}}(t_n,\mathbf{z}_n)\mathbf{\phi}_\gamma(t_n,\mathbf{z}_n;s-t_n) - \mathbf{a}^\gamma (t_n,\mathbf{z}_n) (s-t_n)-\mathbf{f}(t_n,\mathbf{z}_n). A high-order local linear discretization of the SDE (7.1) at each point t_{n+1}\in(t)_h is then defined by the recursive expression \mathbf{z}_{n+1}=\mathbf{z}_n+\mathbf{\phi}_\gamma (t_n,\mathbf{z}_n;h_n)+\widetilde{\mathbf{r}}(t_n,\mathbf{z}_n;h_n),\qquad \text{ with } \qquad \mathbf{z}_0=\mathbf{x}_0, where \widetilde{\mathbf{r}} is a strong approximation to the residual \mathbf{r} of order \alpha higher than 1.5. The strong HOLL discretization \mathbf{z}_{n+1} converges with order \alpha to the solution of (7.1). \mathbf{y}_{n+1}=\mathbf{y}_n+\mathbf{L}(\mathbf{P}_{p,q}(2^{-k_n} \mathbf{M}_{n}h_{n}))^{2^{k_{n}}}\mathbf{r+}\sum\limits_{i=1}^m \mathbf{g}_i(t_n)\Delta \mathbf{w}_n^i,\quad \qquad \qquad (7.2) where the matrices \mathbf{M}_{n} , \mathbf{L} and \mathbf{r} are defined as in (4.6), \Delta \mathbf{w}_{n}^{i} is an i.i.d. zero mean Gaussian random variable with variance h_{n} , and p + q > 1. For large systems of SDEs, in the above scheme (\mathbf{P}_{p,q}(2^{-k_n}\mathbf{M}_{n}h_{n}))^{2^{k_n}}\mathbf{r} is replaced by \mathbf{\mathbf{k}}_{m_{n}, k_n}^{p,q}(h_n,\mathbf{M}_n,\mathbf{r}) . \mathbf{y}_{n+1} =\mathbf{y}_n+\mathbf{L}(\mathbf{P}_{p,q}(2^{-k_n} \mathbf{M}_n h_n))^{2^{k_n}}\mathbf{r}+\sum\limits_{i=1}^m\left( \mathbf{g}_i(t_n)\Delta \mathbf{w}_n^i \mathbf{f}_{\mathbf{x}}(t_n,\widetilde{\mathbf{y}}_n)\mathbf{g}_i(t_n)\Delta \mathbf{z}_n^i+\frac{d\mathbf{g}_i(t_n)}{dt} (\Delta \mathbf{w}_{n}^{i}h_{n}-\Delta \mathbf{z}_{n}^{i})\right) , \qquad \qquad (7.3) where the matrices \mathbf{M}_{n} , \mathbf{L} and \mathbf{r} are defined as \mathbf{M}_n= \begin{bmatrix} \mathbf{f}_{\mathbf{x}}(t_n,\mathbf{y}_n) & \mathbf{f}_{t}(t_n,\mathbf{y}_n)+\frac{1}{2}\sum\limits_{j=1}^{m}\left( \mathbf{I}\otimes \mathbf{g}_j^\intercal (t_n) \right) \mathbf{f}_{ \mathbf{xx}}(t_n,\mathbf{y}_n)\mathbf{g}_j(t_n) & \mathbf{f}(t_{n},\mathbf{y}_n) \\ 0 & 0 & 1 \\ 0 & 0 & 0 \end{bmatrix} \in \mathbb{R}^{(d+2)\times (d+2)}, \mathbf{L}=\left[ \begin{array}{ll} \mathbf{I} & \mathbf{0}_{d\times 2} \end{array} \right] , \mathbf{r}^{\intercal }=\left[ \begin{array}{ll} \mathbf{0}_{1\times (d+1)} & 1 \end{array} \right] , \Delta \mathbf{z}_{n}^{i} is a i.i.d. zero mean Gaussian random variable with variance E\left( (\Delta \mathbf{z}_{n}^{i})^{2}\right) = \frac{1}{3}h_{n}^{3} and covariance E(\Delta \mathbf{w}_{n}^{i}\Delta \mathbf{z}_{n}^{i})=\frac{1}{2}h_{n}^{2} and p+q>1 . For large systems of SDEs, in the above scheme (\mathbf{P}_{p,q}(2^{-k_{n}}\mathbf{M}_{n}h_{n}))^{2^{k_{n}}}\mathbf{r} is replaced by \mathbf{\mathbf{k}} _{m_{n},k_{n}}^{p,q}(h_{n},\mathbf{M}_{n},\mathbf{r}) . \mathbf{y}_{t_{n+1}} =\mathbf{y}_{n}+\mathbf{L}(\mathbf{P}_{p,q}(2^{-k_{n}} \mathbf{M}_{n}h_{n}))^{2^{k_{n}}}\mathbf{r}+\sum\limits_{j=1}^{m}\mathbf{g} _{j}\left( t_{n}\right) \Delta \mathbf{w}_{n}^{j}+\sum\limits_{j=1}^{m} \mathbf{f}_{\mathbf{x}}(t_{n},\mathbf{y}_{n})\mathbf{g}_{j}\left( t_{n}\right) \widetilde{J}_{\left( j,0\right) } +\sum\limits_{j=1}^{m}\frac{d\mathbf{g}_{_{j}}}{dt}\left( t_{n}\right) \widetilde{J}_{\left( 0,j\right) } \qquad \qquad +\sum\limits_{j_{1},j_{2}=1}^{m}\left( \mathbf{I}\otimes \mathbf{g}_{j_{2}}^{\intercal }\left( t_{n}\right) \right) \mathbf{f}_{\mathbf{xx}}(t_{n},\mathbf{y}_{n})\mathbf{g} _{j_{1}}\left( t_{n}\right) \widetilde{J}_{\left( j_{1},j_{2},0\right), }\qquad \qquad (7.4) where \mathbf{M}_{n} , \mathbf{L} , \mathbf{r} and \Delta \mathbf{w}_{n}^{i} are defined as in the order-1 SLL schemes, and \widetilde{J}_{\alpha } is order 2 approximation to the multiple Stratonovish integral J_{\alpha } . with 10000 simulations of the SLL1 scheme with h=1/2 and p=q=6. For SDEs with a single Wiener noise (m=1) \mathbf{y}_{t_{n+1}}=\mathbf{y}_{n}+\widetilde{\mathbf{\phi }}(t_{n},\mathbf{ y}_{n};h_{n})+\frac{h_{n}}{2}\left( \mathbf{k}_{1}+\mathbf{k}_{2}\right) + \mathbf{g}\left( t_{n}\right) \Delta w_{n}+\frac{\left( \mathbf{g}\left( t_{n+1}\right) -\mathbf{g}\left( t_{n}\right) \right) }{h_{n}}J_{\left( 0,1\right) } \quad (7.5) \quad \quad \quad where \mathbf{k}_{1} =\mathbf{f}(t_{n}+\frac{h_{n}}{2},\mathbf{y}_{n}+\widetilde{ \mathbf{\phi }}(t_{n},\mathbf{y}_{n};\frac{h_{n}}{2})+\gamma _{+})-\mathbf{f} _{\mathbf{x}}(t_{n},\mathbf{y}_{n})\widetilde{\mathbf{\phi }}(t_{n},\mathbf{y }_{n};\frac{h_{n}}{2})-\mathbf{f}\left( t_{n},\mathbf{y}_{n}\right) -\mathbf{ f}_{t}\left( t_{n},\mathbf{y}_{n}\right) \frac{h_{n}}{2}, \mathbf{k}_{2} =\mathbf{f}(t_{n}+\frac{h_{n}}{2},\mathbf{y}_{n}+\widetilde{ \mathbf{\phi }}(t_{n},\mathbf{y}_{n};\frac{h_{n}}{2})+\gamma _{-}) -\mathbf{f} _{\mathbf{x}}(t_{n},\mathbf{y}_{n})\widetilde{\mathbf{\phi }}(t_{n},\mathbf{y }_{n};\frac{h_{n}}{2}) -\mathbf{f}\left( t_{n},\mathbf{y}_{n}\right) -\mathbf{f}_{t}\left( t_{n},\mathbf{y}_{n}\right) \frac{h_{n}}{2}, with \gamma _{\pm }=\frac{1}{h_{n}}\mathbf{g}\left( t_{n}\right) \Bigl( \widetilde{J}_{\left( 1,0\right) }\pm \sqrt{2\widetilde{J}_{\left(1,1,0\right) }h_{n}- \widetilde{J}_{\left( 1,0\right) }^{2}} \Bigr) . Here, \widetilde{\mathbf{\phi }}(t_{n},\mathbf{y}_{n};h_{n})=\mathbf{L}(\mathbf{P}_{p,q}(2^{-k_{n}}\mathbf{M}_{n}h_{n}))^{2^{k_{n}}}\mathbf{r} for low dimensional SDEs, and \widetilde{\mathbf{\phi }}(t_{n},\mathbf{y}_{n};h_{n})=\mathbf{L\mathbf{k}}_{m_{n},k_{n}}^{p,q}(h_{n},\mathbf{M}_{n}, \mathbf{r}) for large systems of SDEs, where \mathbf{M}_{n} , \mathbf{L} , \mathbf{r} , \Delta \mathbf{w}_{n}^{i} and \widetilde{J}_{\alpha } are defined as in the order-2 SLL-Taylor schemes, p+q>1 and m_{n}>2 . dx(t)=y(t)dt, & x_{1}(0)=0.01 \\ dy(t)=-(\omega ^{2}x(t)+\epsilon x^{4}(t))dt+\sigma dw_{t}, & x_{1}(0)=0.1, \end{array} \qquad \qquad (7.6) and their approximations by two numerical schemes. d\mathbf{x}(t)=\mathbf{f}(t,\mathbf{x}(t))dt+\sum\limits_{i=1}^{m}\mathbf{g} _{i}(t)d\mathbf{w}^{i}(t),\qquad t\in \left[ t_{0},T\right] , \qquad \qquad (8.1) with initial condition \mathbf{x}(t_{0})=\mathbf{x}_{0} , where the drift coefficient \mathbf{f} and the diffusion coefficient \mathbf{g}_{i} are differentiable functions, and \mathbf{w=(\mathbf{w}}^{1},\ldots ,\mathbf{w}^{m}\mathbf{)} is an m-dimensional standard Wiener process. For a time discretization \left( t\right) _{h} , the order- \mathbb{\beta } (=1,2) Weak Local Linear discretization of the solution of the SDE (8.1) is defined by the recursive relation \mathbf{z}_{n+1}=\mathbf{z}_{n}+\mathbf{\phi }_{\mathbb{\beta }}(t_{n}, \mathbf{z}_{n};h_{n})+\mathbf{\eta }(t_{n},\mathbf{z}_{n};h_{n}),\quad with \quad \mathbf{z}_{0}=\mathbf{x}_{0}, where \mathbf{\phi }_{\mathbb{\beta }}(t_{n},\mathbf{z}_{n};\delta )=\int_{0}^{\delta }e^{\mathbf{f}_{\mathbf{x}}(t_{n},\mathbf{z}_{n})(\delta -u)}(\mathbf{f(}t_{n},\mathbf{z}_{n})+\mathbf{b}^{\mathbb{\beta }}(t_{n}, \mathbf{z}_{n})u)du with \mathbf{b}^{\mathbb{\beta }}(t_{n},\mathbf{z}_{n})= \begin{cases} \mathbf{f}_{t}(t_{n},\mathbf{z}_{n}) & \text{for }\mathbb{\beta }=1 \\ \mathbf{f}_{t}(t_{n},\mathbf{z}_{n})+\frac{1}{2}\sum \limits_{j=1}^{m}\left( \mathbf{I}\otimes \mathbf{g} _{j}^{\intercal }\left( t_{n}\right) \right) \mathbf{f}_{\mathbf{xx}}(t_{n}, \mathbf{z}_{n})\mathbf{g}_{j}\left( t_{n}\right) & \text{for }\mathbb{\beta } =2, \end{cases} and \mathbf{\eta }(t_{n},\mathbf{z}_{n};\delta ) is a zero mean stochastic process with variance matrix \mathbf{\Sigma }(t_{n},\mathbf{z}_{n};\delta )=\int\limits_{0}^{\delta }e^{ \mathbf{f}_{\mathbf{x}}(t_{n},\mathbf{z}_{n})(\delta -s)}\mathbf{G}(t_{n}+s) \mathbf{G}^{\intercal }(t_{n}+s)e^{\mathbf{f}_{\mathbf{x}}^{\intercal }(t_{n},\mathbf{z}_{n})(\delta -s)}ds. Here, \mathbf{f}_{\mathbf{x}} , \mathbf{f}_{t} denote the partial derivatives of \mathbf{f} with respect to the variables \mathbf{x} and t, respectively, \mathbf{f}_{\mathbf{xx}} the Hessian matrix of \mathbf{f} with respect to \mathbf{x} , and \mathbf{G}(t)=[\mathbf{g}_{1}(t),\ldots, \mathbf{g}_{m}(t)] . The weak Local Linear discretization \mathbf{z}_{n+1} converges with order \mathbb{\beta } (=1,2) to the solution of (8.1). \mathbf{y}_{n+1}=\mathbf{y}_{n}+\mathbf{B}_{14}+(\mathbf{B}_{12}\mathbf{B} _{11}^{\intercal })^{1/2}\mathbf{\xi }_{n} where, for SDEs with autonomous diffusion coefficients, \mathbf{B}_{11} , \mathbf{B}_{12} and \mathbf{B}_{14} are the submatrices defined by the partitioned matrix \mathbf{B}=\mathbf{P}_{p,q}(2^{-k_{n}}\mathcal{M}_{n}h_{n}))^{2^{k_{n}}} , with \mathcal{M}_{n}=\left[ \begin{array}{cccc} \mathbf{f}_{\mathbf{x}}(t_{n},\mathbf{y}_{n}) & \mathbf{GG}^{\intercal } & \mathbf{f}_{t}(t_{n},\mathbf{y}_{n}) & \mathbf{f}(t_{n},\mathbf{y}_{n}) \\ \mathbf{0} & -\mathbf{f}_{\mathbf{x}}^{\intercal }(t_{n},\mathbf{y}_{n}) & \mathbf{0} & \mathbf{0} \\ \mathbf{0} & \mathbf{0} & 0 & 1 \\ \mathbf{0} & \mathbf{0} & 0 & 0 \end{array} \right] \in \mathbb{R}^{(2d+2)\times (2d+2)}, and \{\mathbf{\xi }_{n}\} is a sequence of d-dimensional independent two-points distributed random vectors satisfying P(\xi _{n}^{k}=\pm 1)=\frac{1}{2} . \mathbf{y}_{n+1}=\mathbf{y}_{n}+\mathbf{B}_{16}+(\mathbf{B}_{14}\mathbf{B} _{11}^{\intercal })^{1/2}\mathbf{\xi }_{n}, where \mathbf{B}_{11} , \mathbf{B}_{14} and \mathbf{B}_{16} are the submatrices defined by the partitioned matrix \mathbf{B}=\mathbf{P} _{p,q}(2^{-k_{n}}\mathcal{M}_{n}h_{n}))^{2^{k_{n}}} with \mathcal{M}_{n}=\left[ \begin{array}{cccccc} \mathbf{J} & \mathbf{H}_{2} & \mathbf{H}_{1} & \mathbf{H}_{0} & \mathbf{a} _{2} & \mathbf{a}_{1} \\ \mathbf{0} & -\mathbf{J}^{\intercal } & \mathbf{I} & \mathbf{0} & \mathbf{0} & \mathbf{0} \\ \mathbf{0} & \mathbf{0} & -\mathbf{J}^{\intercal } & \mathbf{I} & \mathbf{0} & \mathbf{0} \\ \mathbf{0} & \mathbf{0} & \mathbf{0} & -\mathbf{J}^{\intercal } & \mathbf{0} & \mathbf{0} \\ \mathbf{0} & \mathbf{0} & \mathbf{0} & \mathbf{0} & 0 & 1 \\ \mathbf{0} & \mathbf{0} & \mathbf{0} & \mathbf{0} & 0 & 0 \end{array} \right] \in \mathbb{R}^{(4d+2)\times (4d+2)}, \mathbf{J}=\mathbf{f}_{\mathbf{x}}(t_{n},\mathbf{y}_{n})\qquad \mathbf{a}_{1}=\mathbf{f}(t_{n},\mathbf{y}_{n})\qquad \mathbf{a} _{2}=\mathbf{f}_{t}(t_{n},\mathbf{y}_{n})+\frac{1}{2}\sum\limits_{i=1}^{m}( \mathbf{I}\otimes (\mathbf{g}^{i}(t_{n}))^{\intercal })\mathbf{f}_{\mathbf{xx }}(t_{n},\mathbf{y}_{n})\mathbf{g}^{i}(t_{n}) and \mathbf{H}_{0}=\mathbf{G}(t_{n})\mathbf{G}^{\intercal }(t_{n})\qquad \mathbf{H}_{1}=\mathbf{G}(t_{n})\frac{d\mathbf{G}^{\intercal }(t_{n})}{dt} +\frac{d\mathbf{G}(t_{n})}{dt}\mathbf{G}^{\intercal }(t_{n})\qquad \mathbf{H}_{2}=\frac{d\mathbf{G}(t_{n})}{dt}\frac{d\mathbf{G}^{\intercal }(t_{n})}{dt}\text{.}
Stability and dynamics
By construction, the weak LL discretizations inherit the stability and dynamics of the linear SDEs, but it is not the case of the weak LL schemes in general. WLL schemes, with p\leq q\leq p+2, preserve the first two moments of the linear SDEs, and inherits the mean-square stability or instability that such solution may have. This includes, for instance, the equations of coupled harmonic oscillators driven by random force, and large systems of stiff linear SDEs that result from the method of lines for linear stochastic partial differential equations. Moreover, these WLL schemes preserve the ergodicity of the linear equations, and are geometrically ergodic for some classes of nonlinear SDEs. For nonlinear SDEs with small noise (i.e., (8.1) with \mathbf{g}_{i}(t)\approx 0 ), the solutions of these WLL schemes are basically the nonrandom paths of the LL scheme (4.6) for ODEs plus a small disturbance related to the small noise. In this situation, the dynamical properties of that deterministic scheme, such as the linearization preserving and the preservation of the exact solution dynamics around hyperbolic equilibrium points and periodic orbits, become relevant for the mean of the WLL scheme. For instance, Fig. 5 shows the approximate mean of the SDE dx=-t^{2}x\text{ }dt+\frac{3}{2(t+1)}e^{-t^{3}/3}\text{ }dw_{t},\qquad \qquad x(0)=1, \qquad \quad(8.2) computed by various schemes.
Text derived from https://en.wikipedia.org/w/index.php?oldid=1314094892 under CC BY-SA 4.0; modifications have been made.