Variational framework for structure-preserving electromagnetic particle-in-cell methods (Q2147459)

The authors consider the Vlasov equations \(\partial _{t}f_{s}(t,x,v)+v\cdot \nabla _{x}f_{s}(t,x,v)+\frac{q_{s}}{m_{s}}(E(t,x)+v\times B(t,x))\cdot \nabla _{v}f_{s}(t,x,v)=0\), where \(f_{s}\) is the distribution function, \( m_{s}\) the mass and \(q_{s}\) the charge of the particle species \(s\). The electric and magnetic fields are solutions of the Maxwell's equations \( \partial _{t}E(t,x)=\mathrm{curl}\ B(t,x)-J(t,x)\), \(\partial _{t}B(t,x)=- \mathrm{curl}\ E(t,x)\), \(\mathrm{div}\ E(t,x)=\rho (t,x)\), \(\mathrm{div}\ B(t,x)=0\), the charge and current densities being respectively defined as \(\rho (t,x)=\sum_{s}q_{s}\int_{ \mathbb{R}^{3}}f_{s}(t,x,v)dv\), \(J(t,x)=\sum_{s}q_{s}\int_{\mathbb{R} ^{3}}vf_{s}(t,x,v)dv\). The authors introduce the characteristics trajectories solutions of \(\frac{d}{dt}X(t)=V(t)\), \(\frac{d}{dt}V(t)=\frac{ q_{s}}{m_{s}}(E(t,X(t))+V(t)\times B(t,X(t)))\). They then introduce the de Rham sequence \(H^{1}(\mathbb{R}^{3})\overset{\mathrm{grad}}{\rightarrow}H(\mathrm{curl};\mathbb{R}^{3})\overset{\mathrm{curl}}{\rightarrow }H(\mathrm{div}; \mathbb{R}^{3})\overset{\mathrm{div}}{\rightarrow }L^{2}(\mathbb{R}^{3})\) that they generalize with spaces \(V^{0},V^{1},V^{3}\), and they build a diagram introducing operators \(\Pi ^{i}\) from the infinite-dimensional spaces \(V^{l}\) into the discrete ones \(V_{h}^{l}\). They suppose that this diagram commutes and that the spaces \(V^{l}\) are translation invariant function or distribution spaces. They define the notion of admissible shape function. They consider Low's Lagrangian functional \(\mathcal{L}=\sum_{s}\int_{\Omega \times \mathbb{R} ^{3}}f_{s}(t_{0},z_{0})((m_{s}V+q_{s}A(t,X))\cdot X^{\prime }-(\frac{m_{s}}{2 }V^{2}+q_{s}\phi (t,X))dz_{0}+\frac{1}{2}\int_{\Omega }\left\vert \mathrm{grad}\ \phi (t,x)+A^{\prime }(t,x)\right\vert ^{2}dx-\frac{1}{2}\int_{\Omega }\left\vert \mathrm{curl}\ A(t,x)\right\vert ^{2}dx\) and its discrete version \( \mathcal{L}_{h}=\sum_{p=1}^{N}w_{p}((m_{s}V_{p}+q_{s}A^{S}(X_{p}))\cdot X_{p}^{\prime }-(\frac{m_{s}}{2}V_{p}^{2}+q_{s}\phi ^{S}(X_{p}))+\frac{1}{2} \int_{\Omega }\left\vert \mathrm{grad}_{w}\phi _{h}(x)+A_{h}^{\prime }(x)\right\vert ^{2}dx-\frac{1}{2}\int_{\Omega }\left\vert \mathrm{curl} _{w}A_{h}(x)\right\vert ^{2}dx\), for discrete variables \(X_{p},X_{p}^{\prime },V_{p}\), which are collections of trajectories and \(A_{h},A_{h}^{\prime }\in V_{h}^{2}\), \(\phi _{h}\in V_{h}^{3}\). The discrete action functional is defined as \(\mathcal{S}_{h}(X_{N},V_{N},\phi _{h},A_{h})=\int_{0}^{T} \mathcal{L}_{h}(X_{N},\frac{d}{dt}X_{N},V_{N},\phi _{h},A_{h},\partial _{t}A_{h})dt\). The authors then derive discrete Ampère and Faraday equations, completed with two discrete Gauss laws, and they build a discrete Hamiltonian and a discrete bracket which is proved to be a non-canonical Poisson bracket. The discrete Ampère and Faraday equations are proved to be a Poisson system and to preserve the discrete Hamiltonian and the variational Gauss laws. Because this scheme is not momentum-preserving, the authors introduce a modified scheme. Coming back to the above scheme and the Ampère and Faraday equations, the authors deduce a system of ordinary differential equations. In the last parts of their paper, the authors focus on a special case considering B-splines and they present results of numerical simulations.