Loss of polyconvexity by homogenization: a new example (Q2641920): Difference between revisions

The author studies the asymptotic behavior of the zero-energy deformations set of a periodic nonlinear composite material, and applies his investigations to show that polyconvex energies are not closed with respect to periodic homogenization. Let \(\mathbb M ^{s \times d}\) denote the set of \(s \times d\) real matrices. For \(\Lambda \in \mathbb M ^{s \times d}\), \(\widehat \Lambda\) denotes the list of all subdeterminants of \(\Lambda\); thus with \(\gamma(d,s)= \binom{d+s}{d} -1\), we can identify \(\widehat \Lambda\) with a point of \( \mathbb R ^{\gamma(d,s)}\). A function \({\mathcal W}: \mathbb M ^{s \times d} \rightarrow \mathbb R\) is polyconvex if there is a convex function \({\mathcal V}: \mathbb R ^{\gamma(d,s)} \rightarrow \mathbb R\) such that \({\mathcal W} (\Lambda)= {\mathcal V} (\widehat \Lambda)\) \((\Lambda \in \mathbb M ^{s \times d})\). A set \({\mathfrak A} \subseteq \mathbb M ^{s \times d}\) is polyconvex if there is a convex set \({\mathfrak C} \subseteq \mathbb R ^{\gamma(d,s)}\) such that \({\mathfrak A}= \{ \Lambda \in \mathbb M ^{s \times d}:\widehat \Lambda \in {\mathfrak C}\}\). For \(y \in \mathbb R ^{d}\), \(\left\langle y \right\rangle\) denotes the coordinatewise fractional part of \(y\). Fix \(d, n, s \in \mathbb N\) and \(p \in (1,+\infty)\), let \(\{P_{k}\}_{k < n}\) be a measurable partition of \([0,1)^{d}\) and let \({\mathfrak A}_{k} \subseteq \mathbb M ^{s \times d}\) \((k < n)\) be compact sets. By taking a sort of weak\(^{\star}\) scaling limit, one can associate to \(\{({\mathfrak A}_{k}, P_{k})\}_{k < n}\) the so-called homogenized set \({\mathfrak A}_{\text{hom}} \subseteq \mathbb M ^{s \times d}\). Similarly, if the continuous functions \({\mathcal W}_{k}:\mathbb M ^{s \times d} \rightarrow [0, +\infty)\) \((k < n)\), which usually stand for the energy densities of the components, satisfy certain \(p\)-dependent growth conditions, with \[ {\mathcal W}(y, \Lambda) = \sum _{k<n} \chi_{P_{k}}(y) {\mathcal W}_{k}(\Lambda) \] and \[ \widetilde {\mathcal W} (t,\Lambda) = \inf \left\{ \int _{(0,t)^{d}} {\mathcal W} (\left\langle y\right\rangle, \Lambda + \nabla \varphi (y)) \,dy:\varphi \in W_{0}^{1,p}\left((0,t)^{d}, \mathbb R ^{s}\right)\right\}, \] we can define the asymptotic homogenization for \(\{({\mathcal W}_{k}, P_{k})\}_{k < n}\) as \[ {\mathcal W}_{\text{hom}} (\Lambda) = \lim _{t \rightarrow + \infty} \frac{1}{t^{d}} \widetilde {\mathcal W} (t,\Lambda). \] The first result of the paper says that under appropriate assumptions, for \({\mathfrak A}_{k} = {\mathcal W}_{k}^{-1}(0)\) \((k < n)\) we have \({\mathfrak A}_{\text{hom}} = {\mathcal W}_{\text{hom}}^{-1}(0)\). The second result is an explicit example in the case \(d = n = s = 2\) for two polyconvex sets \({\mathfrak A}_{0}, {\mathfrak A}_{1}\) such that with the partition \(P_{0} = [0, 1/2)\times [0, 1)\) and \(P_{1} =[1/2, 1)\times [0, 1)\), \({\mathfrak A}_{\text{hom}}\) is not polyconvex. This example substantially differs from the first such example, due to \textit{A.\ Braides} [Arch. Ration. Mech. Anal. 127, No. 2, 183--190 (1994; Zbl 0811.73005)].