On the mathematical structure and solvability of certain Hilbert space optimization problems in data-driven elasticity (Q6898223)

The authors consider a 1D elastic body occupying the closure of \(\Omega =(0,L)\), with deformation field \(u\), strain field \(e\), stress field \(s\), submitted to an external force field \(q\), and subject to boundary conditions. Given a Gelfand triple \(U\subset H\cong H^{\ast }\subset U^{\ast }\), the optimization problem under consideration is: \(\inf_{(x,z)\in X\times Z}\frac{k}{2}\left\Vert e-\widetilde{e}\right\Vert _{H}^{2}+\frac{1}{2k} \left\Vert s-\widetilde{s}\right\Vert _{H}^{2}\) such that \(e-Bu-p=0\in H\), \( B^{\ast }s-q=0\in U^{\ast }\), where \(X=U\times H\times H\), \(Z=H\times H\), \( B:H\rightarrow H\) is the bounded linear operator \(Bu=u^{\prime }\) (weak derivative), \(B^{\ast }\) the adjoint of \(B\), \(k>0\) a constant providing unit consistency, and \((\widetilde{e},\widetilde{s})\) are given in the set \( \mathcal{D}\) of auxiliary fields chosen from a given closed data set \( \mathfrak{D}\) of strain-stress pairs in \(\mathbb{R}^{2}\) as: \(\mathcal{D} _{d}=\{(\widetilde{e},\widetilde{s})\in H\times H:(\widetilde{e}(x), \widetilde{s}(x))\in \mathfrak{D}\), \(\forall x\in \Omega \}\), or \(\mathcal{D} _{c}=\{(\widetilde{e},\widetilde{s})\in H\times H:g(\widetilde{e}(x), \widetilde{s}(x))=0\), \(\forall x\in \Omega \}\), if \(\mathfrak{D}\) is defined by some constitutive relationship \(g\). The authors prove though direct computations that this problem is equivalent to the reduced problem \(\inf_{z\in Z}\frac{k}{2}\left\Vert P_{N}\widetilde{e}-\ell _{e}\right\Vert _{H}^{2}+\frac{1}{2k}\left\Vert P_{M}\widetilde{s}-\ell _{s}\right\Vert _{H}^{2}\), such that \((\widetilde{e},\widetilde{s})\in \mathcal{D}\), with \( \ell _{e}\in N\), \(\ell _{s}\in M\), in the following sense: (1) Given any pair \(z=(\widetilde{e},\widetilde{s})\) in \(\mathcal{D}\), fixing \(z\) in the original problem yields a unique minimizer \(x^{\ast }=(u^{\ast },e^{\ast },s^{\ast })\) in \(X\) with identical objective values of the original and reduced problems; (2) The original problem admits a minimizer \((x^{\ast },z^{\ast })\) in \(X\times Z\) if and only if the reduced problem admits a minimizer \(z^{\ast }\) in \(Z\). Here, \(M\) is the image of the injective operator \(B\), which is the closed subspace of zero mean functions in \(H\) with codimension one, \(N\) the kernel of the surjective adjoint operator \( B^{\ast }\), which is the subspace of constant functions, \(P_{M}\), \(P_{N}\) the projections associated with \(M\), \(N\), \(\ell _{e}=P_{N}\mathcal{E} _{u_{0},u_{L}}^{\prime }\), \(\ell _{s}=((B^{\ast }B)^{-1})B^{\ast })^{\ast }q\), with \(\mathcal{E}_{u_{0},u_{L}}(x)=\frac{u_{L}-e^{L}u_{0}}{e^{-L}-e^{L}} e^{-x}+\frac{e-^{L}u_{0}-u_{L}}{e^{-L}-e^{L}}e^{x}\). A generalized solution to the original or reduced problems is partially thermomechanically consistent if \((\widetilde{e}^{\ast },\widetilde{s}^{\ast })_{H}\geq 0\). A generalized solution is completely thermomechanically consistent if, in addition, \((e^{\ast },s^{\ast })_{H}\geq 0\). In the case of a linear constitutive manifold \(g(\widetilde{e},\widetilde{s})=\widetilde{s}-c \widetilde{e}\), where \(c\) is a positive constant and auxiliary fields in \( \mathcal{D}_{c}\), the authors prove that the problem \N\[\inf_{(x,z)\in X\times Z}\frac{k}{2}\left\Vert e-\widetilde{e}\right\Vert _{H}^{2}+\frac{1}{2k} \left\Vert s-\widetilde{s}\right\Vert _{H}^{2},\] \Nsuch that \(e-Bu-p=0\in H\), \( B^{\ast }s-q=0\in U^{\ast }\), \(\widetilde{s}-c\widetilde{e}=0\in H\), has a unique strong solution which is completely thermomechanically consistent. They first prove that the reduced problem \N\[\inf_{z\in Z}\frac{k}{2} \left\Vert P_{N}\widetilde{e}-\ell _{e}\right\Vert _{H}^{2}+\frac{1}{2k} \left\Vert P_{M}\widetilde{s}-\ell _{s}\right\Vert _{H}^{2},\] \Nsuch that \( \widetilde{s}-c\widetilde{e}=0\in H\), has a unique generalized solution, and they use properties of the operators. They give the expression of this solution in the special case where \(c=k\). Considering the case of auxiliary fields in \(\mathcal{D}_{d}\) with any closed data set \(\mathfrak{D}\), and assuming that \(\mathfrak{D}\) is finite, the authors prove the equivalence between the original and reduced associated problems. They prove that if \( \mathfrak{s}=c\mathfrak{e}\) for \((\mathfrak{s},\mathfrak{e})\in \mathfrak{D}\), or if \(\mathfrak{D}\) is symmetric, the original problem has a generalized solution. Finally considering mixed boundary conditions \(u(0)=u_{0}\) and \( s(L)=t\), the authors prove that the associated optimization problem with auxiliary fields in \(\mathcal{D}_{d}\) or \(\mathcal{D}_{c}\) has a generalized solution.