Optimal control for a coupled spin-polarized current and magnetization system (Q2673497)

The authors consider the coupled system between the Landau-Lifshitz-Gilbert equation \(\partial _{t}m=-\gamma _{0}m\times (H_{eff}(m)+cs)+\alpha m\times \partial _{t}m\), posed in \((0,T)\times D\), and the diffusion equation \( \partial _{t}s=-\nabla \cdot J-\frac{2D_{0}}{\lambda _{1}}s-\frac{2D_{0}}{ \lambda _{2}}s\times m\), posed in \((0,T)\times \overline{D}\), where \(D\) and \( \overline{D}\) are bounded and Lipschitz domains of \(\mathbb{R}^{d}\) (\(d=1,2,3 \)) with \(D\subset \overline{D}\), \(m\) being the magnetization, \(s\) the spin accumulation and the spin current \(J\) a \(3\times 3\) matrix written as \(J= \frac{\beta ^{\prime }\mu _{\beta }}{e}m\otimes j-D_{0}(\nabla s-\beta \beta ^{\prime }m\otimes (\nabla s\cdot m))\). The boundary condition \(\partial _{\nu }m=0\) is imposed on \((0,T)\times \partial D\), the boundary condition \( \partial _{\nu }s=0\) is imposed on \((0,T)\times \partial \overline{D}\) and the solutions start from initial conditions \(m_{0}\) in \(D\) and \(s_{0}\) in \( \overline{D}\), respectively. The purpose of the paper is to prove the existence of an optimal control to the functional \(F(\pi )=\frac{\kappa }{2} \int_{0}^{T}\left\Vert m-\overline{m}\right\Vert _{L^{2}}^{2}dt+\frac{1}{2} \int_{0}^{T}(\delta _{1}\left\Vert j\right\Vert _{L^{2}(\overline{D} )}^{2}+\delta _{2}\left\Vert \nabla j\right\Vert _{L^{2}(\overline{D} )}^{2}+\delta _{3}\left\Vert \Delta j\right\Vert _{L^{2}(\overline{D} )}^{2})dt+\frac{1-\kappa }{2}\Psi (m(T))\), where \(\overline{m}:[0,T]\times D\rightarrow \mathbb{S}^{2}\) is a given function, \(\pi =(m,s,j)\), \(\kappa ,\delta _{1},\delta _{2},\delta _{3}\) are nonnegative constants and \(\Psi (m(T))\) represents the terminal payoff. The authors first define the notion of weak solution to the above coupled system as a couple \((m,s)\in L^{\infty }(\mathbb{H}^{1}(D))\times (L^{\infty }(\mathbb{L}^{2}(\overline{D}))\cap L^{2}(\mathbb{H}^{2}(\overline{D})))\) such that \(\partial _{t}m\in L^{2}((0,T)\times D)\), \(\left\vert m\right\vert =1\) a.e. in \((0,T)\times D\), \(\partial _{t}s\in L^{2}(\mathbb{H}^{-1}(\overline{D}))\) and which satisfies a variational formulation. They here assume that \(s_{0}\in \mathbb{H}^{1}( \overline{D})\), \(m_{0}\in \mathbb{H}^{1}(D)\) with \(\left\vert m_{0}\right\vert =1\) a.e. in \(D\). They prove the existence of a weak solution for any \(j\in L^{2}(\mathbb{H}^{1}(\overline{D}))\), assuming some hypotheses on the data. They then prove the existence of a weak optimal control \(\pi ^{\ast }\) and they derive the first-order optimality conditions in the 1D case. They prove some regularity properties of the weak solution and of the optimal control. They here use a discretization technique with respect to time and a Lagrange multiplier theorem. The last part of the paper presents the results of numerical simulations performed using the FEniCS automated code generation system. They indeed write variational formulations of the two above equations, they use a backward Euler scheme for the time discretization, they introduce finite element spaces, and they use Newton's scheme to solve the nonlinear equation at each time step. The authors consider nine examples of domains, or initial data, or targets.