A new error bound for reduced basis approximation of parabolic partial differential equations (Q765759)

The authors consider a general linear parabolic equation \(\dot{u} (t)+Au(t)=g(t)\) where \(A\) is a linear operator in \(\mathcal{L}(V,V') \) where \(V\subset H\subset V'\) are Hilbert spaces and \(g\in L_{2}(0,T;V')\). They write the space-time variational formulation of this problem: Find \(u\in \mathcal{X}=L_{2}(0,T;V)\cap H_{0}^{1}(0,T,V')\) such that \(b(u,v)=f(v)\) for every \(v\in \mathcal{ Y}=L_{2}(0,T)\otimes V\). Here \[ b(w,v)=\int_{0}^{T}(\langle u(t),v(t)\rangle _{V' \times V}+a(w(t),v(t)))dt \] and \[ f(t)=\int_{0}^{T}\langle g(t),v(t)\rangle _{V' \times V}dt , \] where \(a\) is the bilinear form associated to the linear operator \(A\). The authors first prove an inf-sup lower bound \[ \inf_{w\in \mathcal{X} }\sup_{y\in \mathcal{Y}}\frac{b(w,v)}{\| w\| _{\mathcal{X} }\| v\| _{\mathcal{Y}}}\geq \beta ^{LB} \] for some constant \( \beta ^{LB}\) which can be expressed in terms of the continuity and coercivity constants of \(a\). Then the authors introduce a Petrov-Galerkin approximation associated to finite-dimensional subspaces \(\mathcal{X} _{\delta }\subset \mathcal{X}\) and \(\mathcal{Y}_{\delta }\subset \mathcal{Y}\). This approximation can be associated to a Crank-Nicolson scheme. The authors prove an inf-sup lower bound for this approximation. Finally the authors consider the case where \(A=A(\mu )=-\Delta u+\mu _{1}\beta (x)\cdot \nabla u+\mu _{2}u,\) where \(\mu _{1}\) and \(\mu _{2}\) are real parameters. They introduce a reduced basis approximation for the Crank-Nicolson scheme and they prove an error estimate for this reduced basis approximation. The note ends with a short description of the numerical results the authors obtain for different values of \(\mu _{1}\) and \(\mu _{2}\).