No label defined (Q4938013)

The following heat conduction problem is considered with a mixed nonhomogeneous boundary condition \[ \begin{gathered} \frac{\partial u}{\partial t} -\operatorname{div} (\mathcal A \operatorname{grad}u)=f \quad \text{in} (0,T) \times \Omega,\tag{1} \\ \alpha u+n^T \mathcal A \operatorname{grad} u =g \quad \text{on} (0,T)\times \partial \Omega,\tag{2} \\ u(x,0)=u^0(x),\quad x\in \Omega, \tag{3} \end{gathered} \] where the matrix \(\mathcal A=\mathcal A(x)=(a_{ij}(x))^d_{i,j=1}\) is supposed to be symmetric and uniformly positive definite, \(a_{ij}\) and \(\alpha\) are assumed to be bounded measurable functions. In the hypotheses \(a_{ij}\in L^\infty(\Omega)\), \(\alpha\in L^\infty(\partial\Omega)\), \(u^0\in H^1(\Omega)\cap C(\bar \Omega)\), \(f\in C([0,T],L^2(\Omega))\) it is useful to give the weak formulation of the problem (1)--(3): find \(u\in H^1((0,T),H^1(\Omega))\) (weak solution) which satisfies \[ (u'(t),v)+a(u,v)=F(t;v), \quad \forall v\in H^1(\Omega), \tag{4} \] for a.e. fixed \(t\in (0,T)\) and (3) for a.e. \(x\in \Omega\), where \(u'=\partial u/\partial t\), \[ a(v,w)=(\mathcal A \operatorname{grad}v,\operatorname{grad} w) +\langle \alpha v,w\rangle \] and \[ F(t;v)=(f(\cdot,t),v)+\langle g(\cdot,t)v\rangle. \] The semidiscrete Galerkin approximation of the problem (1)--(3) consists of finding \(u_h\in H^1(0,T),V_h)\) which satisfies \[ \begin{gathered} (u'_h(t),v_h)+a(u_h,v_h)=F(t;v_h),\;\forall v_h\in V_h,\;\text{ for a.e.} t\in (0,T) \tag{5} \\ (u_h(0),v_h)=(u^0,v_h),\;\forall v_h\in V_h, \tag{6} \end{gathered} \] where \(V_h=\{v_h\in H^1(\Omega)|v_h|_k\in Pr (k)\), \(\forall k\in \mathcal T_h\}\), \(\mathcal T_h\) is a triangulation of \(\bar \Omega\) consisting of elements \(k_i\) and having certain standard properties and \[ P:H^1(\Omega)\to V_h, \quad Pw=w_h \] is the so-called elliptic projection. Firstly, the authors show that if \(u\) and \(u_h\) are the solutions of (4) and (5), respectively, then an estimate of the form \[ \|u(t)-u_h(t)\|\leq \|u_h(0)-u^0\|+Ch^r\left(\|u^0\|_r+\int^t_0\|u'\|_r ds\right) \] holds. The semidiscretization in \(V_h\) results in the second part the authors consider the numerical solution of the semidiscrete problem via the so-called \(\theta\)-method and they prove the second order of the convergence for the Crank-Nicolson-Galerkin scheme. The main result states that for the global error of the fully discretized Galerkin method with \(\theta\in [0.5,1]\) an estimate of the form \[ \begin{gathered} \|U^n-u(t_n)\|\leq C_{20}\|u_h(0)-u^0\|+C_{21} h^r\left(\|u^0\|_r+\int_0^{t_n} \|u'(s)\|_r ds\right)\\ +C_{22}\tau^2\int^{t_n}_0\|u'''(s)\|_r ds +\tau \int^{t_n}_0\|Lu'(s)\|_r ds \end{gathered} \] holds, where \(Lu=\operatorname{div}( \mathcal A (x),\operatorname{grad} u)\) and \(U^n\) is the approximation in \(V_h\) of \(u(t)\) at \(t_n=n\tau\), \(n=1,2,\dots,\tau\), being the time steps.