betaFeedback

WorksMathematics

A characterization of Generalized functions of Bounded Deformation

Abstract

We show that Dal Maso's GBD space, introduced for tackling crack growth in linearized elasticity, can be defined by simple conditions in a finite number of directions of slicing.

Introduction

In [21] G. Dal Maso introduced the space of Generalized (Special) functions of Bounded Deformation (G(S)BDG(S)BD) to characterize the domain of a class of functionals associated to free discontinuity problems [22, 7] in linearized elasticity or elastoplasticity, in particular for the modeling of crack growth [25]. In these contexts the energy functionals depend on a variable u:Ω⊂Rd→Rdu:\Omega\subset\mathbb{R}^{d}\to\mathbb{R}^{d}, representing the displacement, through its symmetrized gradient e(u)=(Du+DuT)/2e(u)=(Du+Du^{T})/2 (the deformation) and its (d−1)(d-1)-dimensional jump set JuJ_{u} (the crack set), with a constraint imposing - in a weak sense - Dirichlet boundary conditions. Yet an integrability bound on uu in terms of a bound for such functionals is in general not available, so that it is not possible to control the deformation in the space of Radon measures.

An analogous issue arises both in the simplified antiplane shear setting, where the displacement is only in the vertical direction so it can be assumed scalar, and in the framework of finite elasticity or elastoplasticity, contexts in which the role of e(u)e(u) is played by the full gradient ∇u\nabla u. This motivated the introduction of the space G(S)BVG(S)BV [4, 5] as the space of functions which are in (S)BV(S)BV after composition ψ∘u\psi\circ u with C1C^{1} functions ψ\psi having ∇ψ\nabla\psi compactly supported.

However, e(ψ∘u)e(\psi\circ u) depends on the whole gradient of uu, even for uu smooth, which prevents from adopting a similar definition of G(S)BDG(S)BD. The solution found in [21] was to define this space by slicing: a measurable function u:Ω→Rdu:\Omega\to\mathbb{R}^{d} is in GBD(Ω)GBD(\Omega) if there exists a bounded Radon measure Λu\Lambda_{u} such that for every ξ∈Sd−1\xi\in\mathbb{S}^{d-1} and for Hd−1\mathcal{H}^{d-1}-a.e. z∈ξ⊥z\in\xi^{\perp} the slice s↦uzξ(s):=ξ⋅u(z+sξ)s\mapsto u_{z}^{\xi}(s):=\xi\cdot u(z+s\xi) has bounded variation in Ωzξ:={s∈R:z+sξ∈Ω}\Omega_{z}^{\xi}:=\{s\in\mathbb{R}:z+s\xi\in\Omega\} and

(μ^u)ξ(B):=∫ξ⊥((∣Dauzξ∣+∣Dcuzξ∣)(Bzξ)+∑s∈Juzξ∩Bzξ∣[uzξ](s)∣∧1) dHd−1(z)≤Λu(B)(\widehat{\mu}_{u})^{\xi}(B):=\int_{\xi^{\perp}}\left((|D^{a}u_{z}^{\xi}|+|D^{c}u_{z}^{\xi}|)(B_{z}^{\xi})+\sum_{s\in J_{u_{z}^{\xi}}\cap B_{z}^{\xi}}|[u_{z}^{\xi}](s)|\wedge1\right)\,d\mathcal{H}^{d-1}(z)\leq\Lambda_{u}(B)

for every B⊂ΩB\subset\Omega Borel, with Bzξ:={s∈R:z+sξ∈B}B_{z}^{\xi}:=\{s\in\mathbb{R}:z+s\xi\in B\}. Still in [21], it is shown that it is enough to have the bound for all ξ\xi in a dense subset of Sd−1\mathbb{S}^{d-1}. However, it is well known that a much simpler characterization exists for the space BDBD of functions of Bounded Deformation: a measurable uu is in BD(Ω)BD(\Omega) as soon as there is a basis (ei)i=1d(e_{i})_{i=1}^{d} of Rd\mathbb{R}^{d} such as the following holds: for any ξ∈V~:={ei:i=1,…,d}∪{ei+ej:1≤i<j≤d}\xi\in\widetilde{V}:=\{e_{i}:i=1,\ldots,d\}\cup\{e_{i}+e_{j}:1\leq i<j\leq d\} and for Hd−1\mathcal{H}^{d-1}-a.e. z∈ξ⊥z\in\xi^{\perp}, uzξ∈BV(Ωzξ)u_{z}^{\xi}\in BV(\Omega_{z}^{\xi}) and

∣Euξ⋅ξ∣(Ω)=∣D ξ(u⋅ξ)∣(Ω)=∫ξ⊥∣Duzξ∣(Ωzξ) dHd−1(z)<+∞,|Eu\xi\cdot\xi|(\Omega)=|D_{\,\xi}(u\cdot\xi)|(\Omega)=\int_{\xi^{\perp}}|Du_{z}^{\xi}|(\Omega_{z}^{\xi})\,d\mathcal{H}^{d-1}(z)<+\infty,

cf. [6 Section 3] and [32 Chap. II, Section 2.2] (see also [33, 32, 8, 23, 24] for general properties of BDBD functions). It is therefore natural to ask whether GBDGBD functions can, similarly, be described by conditions in a finite number of appropriate directions. Our main result answers this affirmatively.

We show that a measurable uu is in GBD(Ω)GBD(\Omega) as soon as there exists a basis (ei)i=1d(e_i)_{i=1}^d of Rd\mathbb{R}^d and d(d−1)/2d(d-1)/2 vectors ξi,j∈span⁡{ei±ej}\xi_{i,j} \in\operatorname{span}\{e_i \pm e_j\}, $1 \le i < j \le d$ such that for any ξ∈V={ei:i=1,…,d}∪{ξi,j:1≤i<j≤d}\xi\in V = \{e_i : i = 1,\ldots,d\} \cup\{\xi_{i,j} : 1 \le i < j \le d\} and for Hd−1\mathcal{H}^{d-1}-a.e. z∈ξ⊥z \in\xi^\perp, uzξ∈BV(Ωzξ)u_z^\xi\in BV(\Omega_z^\xi) and

Λuξ:=(μuξ)(Ω)<+∞.\Lambda_u^\xi:= (\mu_u^\xi)(\Omega) < +\infty.

In other words, we show that if we have a control on the d(d+1)/2d(d+1)/2 directions ξ∈V\xi\in V through Λuξ\Lambda_u^\xi, then the same control holds in fact for all ξ∈Sd−1\xi\in\mathbb{S}^{d-1}, and can be localized on Borel subsets B⊂ΩB \subset\Omega and controlled by a global finite measure Λu\Lambda_u.

Our strategy is to approximate any measurable uu with Λuξ\Lambda_u^\xi bounded for all ξ∈V\xi\in V, by a family of functions, each in a suitable finite dimensional space, satisfying a uniform bound in GBDGBD, according to the definition in [21] (Theorem 1). More precisely, for any ε>0\varepsilon> 0, we choose a family of discretizations of the domain into hypercubes of sidelength ε>0\varepsilon> 0 so that uu is approximated in measure by the functions defined in any hypercube as the dd-linear interpolation over the vertices (see Appendix § 4). Then we modify such functions, setting them to 0 on the “bad hypercubes”, which are such that there are ξ∈V\xi\in V and a couple of vertices joined by a segment parallel to ξ\xi that intersect the set where the jump of uu is larger than 1. Thanks to an averaging argument, we show that we can choose the discretizations in such a way that (i) the bound on Λu\Lambda_u gives a uniform control on the perimeter of the union of the “bad hypercubes” while (ii) outside this set, the symmetric gradient of the dd-linear approximation is controlled in L1L^1 by the finite differences along directions in VV, thus again by the bound on Λu\Lambda_u (see Appendix § 3).

From this approximation we derive (Corollary 1) that u∈GBDu \in GBD by employing a suitable compactness result for bounded sequences in GBDGBD that converge in measure [21, 27]. This strategy of discretization/reinterpolation for (G)BD(G)BD functions was first used in [11, 12, 31] to show approximation results in (G)SBD(G)SBD.

As a further application (Corollary 2), we prove that any sequence of measurable functions satisfying uniformly the control on (Λukξ)ξ∈Vk(\Lambda_{u_k}^\xi)_{\xi\in V_k} is precompact in GBDGBD, in the sense of the general compactness and closedness result in [19] (we refer also to [3] for a related compactness result in GBDGBD).

Eventually, we address an open question raised in the recent work [1] (see also [2]) concerning a nonlocal approximation of Griffith-type energies for brittle cracks (see e.g. [25, 13, 26, 30, 15, 20, 14, 16, 10, 29, 9, 28]). In some approximation regimes the limiting admissible deformation uu naturally belong to the space therein called GBVε(Ω;Rd)GBV^\varepsilon(\Omega;\mathbb{R}^d), namely it satisfies the bound

∫Sd−1μ^uξ(B) dHd−1(ξ)≤Λu(B)for B⊂Ω Borel,\int_{\mathbb{S}^{d-1}} \widehat{\mu}_u^\xi(B)\,\mathrm{d}\mathcal{H}^{d-1}(\xi) \le\Lambda_u(B) \quad\text{for } B \subset\Omega\text{ Borel},

for a suitable bounded Radon measure Λu\Lambda_u. We prove that GBVεGBV^\varepsilon coincide with GBDGBD (Corollary 3), in particular we give positive answer to a question in [1] concerning the rectifiability of the measure obtained by integrating over directions ξ\xi the jump parts of the measures (μ^uξ)ξ(\widehat{\mu}_u^\xi)_\xi.

In the same spirit, we prove (Proposition 2) a characterization of the space GSBDpGSBD^p (i.e. the GBDGBD functions with slices in SBVSBV, symmetrized gradient in LpL^p and jump set of finite Hd−1\mathcal{H}^{d-1}-measure, see e.g. [26]), with a slight adaptation of the arguments in the proof of Theorem 1, and the corresponding compactness (Corollary 4), employing the general compactness result [18], see also [17].

Main results

Here Ω\Omega is a bounded1 open set of Rd\mathbb{R}^d, d≥2d \ge2. Given u:Ω→Rdu : \Omega\to\mathbb{R}^d, for any ξ∈Rd∖{0}\xi\in\mathbb{R}^d \setminus\{0\} and z∈ξ⊥z \in\xi^\perp, we define:

uzξ(s):=ξ⋅u(z+sξ)for s∈Ωzξ:={s∈R:z+sξ∈Ω}.u^\xi_z(s) := \xi\cdot u(z+s\xi) \quad\text{for } s \in\Omega^\xi_z := \{s \in\mathbb{R} : z+s\xi\in\Omega\}.

We denote by L0(Ω;Rm)L^0(\Omega;\mathbb{R}^m), m≥1m \ge1, the measurable functions from Ω\Omega into Rm\mathbb{R}^m and by vk→vv_k \to v in L0(Ω;Rm)L^0(\Omega;\mathbb{R}^m) we mean that vkv_k converge to vv in LdL^d-measure in Ω\Omega, for Ld\mathcal{L}^d the Lebesgue measure on Rd\mathbb{R}^d. We also let Q:=[0,1)dQ := [0,1)^d and, for any ε>0\varepsilon> 0,

Ωε:={x∈Ω:dist⁡(x,∂∗Ω)>dε}.\Omega_\varepsilon:= \{x \in\Omega: \operatorname{dist}(x,\partial^*\Omega) > \sqrt{d}\varepsilon\}.

We refer to [7] for general theory of BVBV functions. If u∈BVu \in BV we denote by DauD^a u and DcuD^c u the absolutely continuous and the Cantor part of the bounded Radon measure DuDu. Moreover, we refer to [6] for a careful treatment of the SBDSBD space and to [21] for the definition and the main properties of GBDGBD functions.

Theorem 1. Let u∈L0(Ω;Rd)u \in L^0(\Omega;\mathbb{R}^d) and assume there is an orthonormal basis (ei)i=1d(e_i)_{i=1}^d of Rd\mathbb{R}^d such that for any ε>0\varepsilon> 0 it holds that ξ∈V:={ei:i=1,…,d}∪{ei+ej:1≤i<j≤d}\xi\in V := \{e_i : i = 1,\ldots,d\} \cup\{e_i+e_j : 1 \le i < j \le d\}, and that for Hd−1\mathcal{H}^{d-1}-a.e. z∈ξ⊥z \in\xi^\perp, uzξ∈BV(Ωzξ)u^\xi_z \in BV(\Omega^\xi_z) and

Λuξ:=∫ξ⊥((∣Dauzξ∣+∣Dcuzξ∣)(Ωzξ)+∑s∈Juzξ[∣uzξ(s)∣∧1]) dHd−1(z)<+∞.(1)\Lambda^\xi_u := \int_{\xi^\perp} \left((|D^a u^\xi_z|+|D^c u^\xi_z|)(\Omega^\xi_z)+\sum_{s\in J_{u^\xi_z}}[|u^\xi_z(s)|\wedge1]\right)\,d\mathcal{H}^{d-1}(z)<+\infty. \tag*{(1)}

Then there exist a dimensional constant C>0C>0 and a family (uε)ε>0⊂SBV∞(Ω;Rd)(u^\varepsilon)_{\varepsilon>0}\subset SBV^\infty(\Omega;\mathbb{R}^d) such that, for any ε>0\varepsilon>0, Juε⊂∂BεJ_{u^\varepsilon}\subset\partial B^\varepsilon with BεB^\varepsilon union of finitely many cubes of sidelength ε\varepsilon included in Ω\Omega,

∫Ωε∣e(uε)∣ dx+Hd−1(∂∗Bε∩Ωε)≤CΛuV,ΛuV:=∑ξ∈VΛuξ(2)\int_{\Omega_\varepsilon}|e(u^\varepsilon)|\,dx+\mathcal{H}^{d-1}(\partial^*B^\varepsilon\cap\Omega_\varepsilon)\le C\Lambda^V_u,\qquad\Lambda^V_u:=\sum_{\xi\in V}\Lambda^\xi_u \tag*{(2)}

and uε→uu^\varepsilon\to u in L0(Ω;Rd)L^0(\Omega;\mathbb{R}^d).

Remark 1. In Theorem 1, we assumed (ei)i=1d(e_i)_{i=1}^d orthonormal for simplicity. The result would hold also for an arbitrary basis: this is easily checked by considering a matrix AA which sends (ei)i=1d(e_i)_{i=1}^d to an orthonormal basis and replacing uu with A−Tu(A−1⋅)A^{-T}u(A^{-1}\cdot). One could of course also consider the integrals in (1) over arbitrary spaces EξE_\xi with Eξ⊕Rξ=RdE_\xi\oplus\mathbb{R}\xi=\mathbb{R}^d rather than just ξ⊥\xi^\perp, see [21 Remark 4.11]. Additionally, as in [21 Remark 4.3], one could replace in (1) the threshold 1 (in [∣uzξ∣]∧1[|u^\xi_z|]\wedge1) by an arbitrary (ξ\xi-dependent) threshold level βξ>0\beta_\xi>0, since ∣x∣∧α≤(1∨α/β)(∣x∣∧β)|x|\wedge\alpha\le(1\vee\alpha/\beta)(|x|\wedge\beta) for any α,β>0\alpha,\beta>0. Finally one could consider, for some pairs (i,j)(i,j), slicing in the direction ei−eje_i-e_j rather than ei+eje_i+e_j.

From Theorem 1 we deduce the following three consequences. We refer to [1 Definition 2.3] for the definition of the space GBVE(Ω;Rd)GBV^E(\Omega;\mathbb{R}^d), characterized in Corollary 3.

Corollary 1. For any fixed basis (ei)i=1d(e_i)_{i=1}^d of Rd\mathbb{R}^d and V:={ei:i=1,…,d}∪{ei+ej:1≤i<j≤d}V:=\{e_i:i=1,\ldots,d\}\cup\{e_i+e_j:1\le i<j\le d\} it holds that

GBD(Ω)={u∈L0(Ω;Rd):ΛuV<+∞},(3)GBD(\Omega)=\{u\in L^0(\Omega;\mathbb{R}^d):\Lambda^V_u<+\infty\}, \tag*{(3)}

where, for any u∈L0(Ω;Rd)u\in L^0(\Omega;\mathbb{R}^d), ΛuV\Lambda^V_u is defined in (1), (2).

Corollary 2. Let (uk)k⊂GBD(Ω)(u_k)_k\subset GBD(\Omega) such that for every k∈N∖{0}k\in\mathbb{N}\setminus\{0\} there exists (ek)k(e^k)_k, ek:=(eik)i=1de^k:=(e_i^k)_{i=1}^d orthonormal basis of Rd\mathbb{R}^d with (recalling (1), (2))

sup⁡k∈NΛukVk<+∞,\sup_{k\in\mathbb{N}}\Lambda^{V_k}_{u_k}<+\infty,

for Vk:={eik:i=1,…,d}∪{eik+ejk:1≤i<j≤d}V_k := \{e_i^k : i = 1,\ldots,d\} \cup\{e_i^k + e_j^k : 1 \le i < j \le d\}. Then there exist a Caccioppoli partition P=(Pn)n\mathcal{P} = (P_n)_n of Ω\Omega, a sequence of piecewise rigid motions (ak)k(a_k)_k with

ak=∑n∈NaknχPn,(4)a_k = \sum_{n\in\mathbb{N}} a_k^n\chi_{P_n}, \tag*{(4)}
∣akn(x)−akn′(x)∣→+∞for a.e. x∈Ω, for all n≠n′,(5)|a_k^n(x)-a_k^{n'}(x)| \to+\infty\quad\text{for a.e. } x\in\Omega,\ \text{for all } n\ne n', \tag*{(5)}

and u∈GBD(Ω)u\in GBD(\Omega) such that, up to a (not relabelled) subsequence,

uk−ak→ua.e. in Ω,(6)u_k-a_k\to u \quad\text{a.e. in }\Omega, \tag*{(6)}
Hd−1(∂∗P∩Ω)≤lim⁡σ→+∞lim inf⁡k→∞Hd−1(Jukσ).(7)\mathcal{H}^{d-1}(\partial^*\mathcal{P}\cap\Omega)\le\lim_{\sigma\to+\infty}\liminf_{k\to\infty}\mathcal{H}^{d-1}(J_{u_k}^{\sigma}). \tag*{(7)}

where Jukσ:={x∈Juk:∣[uk](x)∣≥σ}J_{u_k}^{\sigma}:=\{x\in J_{u_k}:|[u_k](x)|\ge\sigma\}.

Corollary 3. It holds that

GBVE(Ω;Rd)=GBD(Ω).GBV^{\mathcal{E}}(\Omega;\mathbb{R}^d)=GBD(\Omega).

We now provide analogous results for the space GSBDpGSBD^p of GBDGBD functions with symmetrized gradient in LpL^p and jump set of finite Hd−1\mathcal{H}^{d-1}-measure, see e.g. [15].

Proposition 2. Let u∈L0(Ω;Rd)u\in L^0(\Omega;\mathbb{R}^d), p>1p>1, and assume there is an orthonormal basis (ei)i=1d(e_i)_{i=1}^d of Rd\mathbb{R}^d such that for any ε>0\varepsilon>0 it holds that ξ∈V:={ei:i=1,…,d}∪{ei+ej:1≤i<j≤d}\xi\in V:=\{e_i:i=1,\ldots,d\}\cup\{e_i+e_j:1\le i<j\le d\}, and almost z∈ξ⊥z\in\xi^\perp, uzξ∈SBVp(Ωzξ)u^\xi_z\in SBV^p(\Omega^\xi_z) with Dauzξ=(uzξ)′L1D^a u^\xi_z=(u^\xi_z)' \mathcal{L}^1 and

Λup,ξ:=∫ξ⊥(∫Ωzξ∣(uzξ)′∣p dt+H0(Juzξ))dHd−1(z)<+∞.(8)\Lambda_u^{p,\xi}:=\int_{\xi^\perp}\left(\int_{\Omega^\xi_z}|(u^\xi_z)'|^p\,dt+\mathcal{H}^0(J_{u^\xi_z})\right)d\mathcal{H}^{d-1}(z)<+\infty. \tag*{(8)}

Then there exist a constant C>0C>0 depending only on dd, pp, and a family (uε)ε>0⊂SBV∞(Ω;Rd)(u^\varepsilon)_{\varepsilon>0}\subset SBV^\infty(\Omega;\mathbb{R}^d) such that, for any ε>0\varepsilon>0, Juε⊂∂BεJ_{u^\varepsilon}\subset\partial B^\varepsilon with BεB^\varepsilon union of finitely many cubes of sidelength ε\varepsilon included in Ω\Omega,

∫Ωε∣e(uε)∣p dx+Hd−1(∂∗Bε∩Ωε)≤CΛup,V,Λup,V:=∑ξ∈VΛup,ξ,(9)\int_{\Omega_\varepsilon}|e(u^\varepsilon)|^p\,dx+\mathcal{H}^{d-1}(\partial^*B^\varepsilon\cap\Omega_\varepsilon)\le C\Lambda_u^{p,V},\qquad\Lambda_u^{p,V}:=\sum_{\xi\in V}\Lambda_u^{p,\xi}, \tag*{(9)}

and uε→u∈L0(Ω;Rd)u^\varepsilon\to u\in L^0(\Omega;\mathbb{R}^d). In particular u∈GSBDp(Ω)u\in GSBD^p(\Omega).

Corollary 4. With the notation of Proposition 2, let (uk)k⊂GSBDp(Ω)(u_k)_k\subset GSBD^p(\Omega), (ek)k(e^k)_k, (Vk)k(V_k)_k with ek:=(eik)i=1de^k:=(e_i^k)_{i=1}^d orthonormal basis of Rd\mathbb{R}^d, Vk:={eik:i=1,…,d}∪{eik+ejk:1≤i<j≤d}V_k:=\{e_i^k:i=1,\ldots,d\}\cup\{e_i^k+e_j^k:1\le i<j\le d\} be sequences such that

sup⁡k∈NΛukp,Vk<+∞.\sup_{k\in\mathbb{N}}\Lambda_{u_k}^{p,V_k}<+\infty.

Then there exist a Caccioppoli partition P=(Pn)n\mathcal{P}=(P_n)_n of Ω\Omega, a sequence of piecewise rigid motions (ak)k(a_k)_k with

ak=∑n∈NaknχPn,(10)a_k=\sum_{n\in\mathbb{N}}a_k^n\chi_{P_n}, \tag*{(10)}
∣akn(x)−akn′(x)∣→+∞for a.e. x∈Ω, for all n≠n′,(11)|a_k^n(x)-a_k^{n'}(x)|\to+\infty\quad\text{for a.e. }x\in\Omega,\ \text{for all }n\ne n', \tag*{(11)}

and u∈GSBDp(Ω)u\in GSBD^p(\Omega) such that, up to a (not relabelled) subsequence,

uk−ak→ua.e. in Ω,(12)u_k-a_k\to u \quad\text{a.e. in }\Omega, \tag*{(12)}
e(uk)⇀e(u)in Lp(Ω;Msymd×d),(13)e(u_k)\rightharpoonup e(u)\quad\text{in }L^p(\Omega;\mathbb{M}^{d\times d}_{\mathrm{sym}}), \tag*{(13)}
Hd−1(Ju∪∂∗P)≤lim inf⁡k→∞Hd−1(Juk).(14)\mathcal{H}^{d-1}(J_u\cup\partial^*\mathcal{P})\le\liminf_{k\to\infty}\mathcal{H}^{d-1}(J_{u_k}). \tag*{(14)}

From now on we prove the announced results, following the order of presentation.

Proof of Theorem 1. For ξ∈V\xi\in V and z∈ξ⊥z \in\xi^\perp such that uzξ∈BV(Ωzξ)u_z^\xi\in BV(\Omega_z^\xi), we denote by μzξ\mu_z^\xi the measure (on Ωzξ\Omega_z^\xi):

μzξ=∣Dauzξ∣+∣Dcuzξ∣+∑s∈Juzξ(∣[uzξ](s)∣∧1)H0⌞{s}.\mu_z^\xi= |D^a u_z^\xi| + |D^c u_z^\xi| + \sum_{s\in J_{u_z^\xi}} (|[u_z^\xi](s)| \wedge1)\mathcal{H}^0\llcorner\{s\}.

Fixed ε>0\varepsilon> 0, we evaluate the integral

εd−1∫Q∑ξ∈V∑i∈εZd(∣ξ⋅(u(εy+i+εξ)−u(εy+i))∣∧1) dy\varepsilon^{d-1}\int_Q \sum_{\xi\in V}\sum_{i\in\varepsilon\mathbb{Z}^d} (|\xi\cdot(u(\varepsilon y+i+\varepsilon\xi)-u(\varepsilon y+i))|\wedge1)\,dy

where here the sum is over i∈εZdi\in\varepsilon\mathbb{Z}^d such that εy+i∈Ω∩(Ω−εξ)\varepsilon y+i\in\Omega\cap(\Omega-\varepsilon\xi). Given ξ∈V\xi\in V, then the change of variable x=i+εyx=i+\varepsilon y shows that:

εd−1∫Q∑i∈εZd(∣ξ⋅(u(εy+i+εξ)−u(εy+i))∣∧1) dy=ε−1∫Ω∩(Ω−εξ)(∣ξ⋅(u(x+εξ)−u(x))∣∧1) dx.\varepsilon^{d-1}\int_Q \sum_{i\in\varepsilon\mathbb{Z}^d} (|\xi\cdot(u(\varepsilon y+i+\varepsilon\xi)-u(\varepsilon y+i))|\wedge1)\,dy = \varepsilon^{-1}\int_{\Omega\cap(\Omega-\varepsilon\xi)} (|\xi\cdot(u(x+\varepsilon\xi)-u(x))|\wedge1)\,dx.

and writing x=z+sξx=z+s\xi, z∈ξ⊥z\in\xi^\perp, s∈Ωzξ∩(Ωzξ−ε)s\in\Omega_z^\xi\cap(\Omega_z^\xi-\varepsilon) we find that this is also expressed as

ε−1∫ξ⊥∫Ωzξ∩(Ωzξ−ε)(∣uzξ(s+ε)−uzξ(s)∣∧1)∣ξ∣ ds dHd−1(z).\varepsilon^{-1}\int_{\xi^\perp}\int_{\Omega_z^\xi\cap(\Omega_z^\xi-\varepsilon)} (|u_z^\xi(s+\varepsilon)-u_z^\xi(s)|\wedge1)|\xi|\,ds\,d\mathcal{H}^{d-1}(z).

Now, since (for Hd−1\mathcal{H}^{d-1}-a.e. zz) uzξ∈BV(Ωzξ)u_z^\xi\in BV(\Omega_z^\xi), then for a.e. ss: either there is s′∈(s,s+ε)∩Juzξs'\in(s,s+\varepsilon)\cap J_{u_z^\xi} with ∣[uzξ](s)∣≥1|[u_z^\xi](s)|\ge1, and then clearly ∣uzξ(s+ε)−uzξ(s)∣∧1≤μzξ((s,s+ε))|u_z^\xi(s+\varepsilon)-u_z^\xi(s)|\wedge1\le\mu_z^\xi((s,s+\varepsilon)), or there isn’t, in which case, μzξ⌞(s,s+ε)=∣Duzξ∣⌞(s,s+ε)\mu_z^\xi\llcorner(s,s+\varepsilon)=|Du_z^\xi|\llcorner(s,s+\varepsilon), which implies that ∣uzξ(s+ε)−uzξ(s)∣≤μzξ((s,s+ε))|u_z^\xi(s+\varepsilon)-u_z^\xi(s)|\le\mu_z^\xi((s,s+\varepsilon)). It follows that

∫Ωzξ∩(Ωzξ−ε)(∣uzξ(s+ε)−uzξ(s)∣∧1)∣ξ∣ ds≤∣ξ∣∫Ωzξ∩(Ωzξ−ε)∫χ(s,s+ε)(t) dμzξ(t) ds≤ε∣ξ∣∫Ωzξdμzξ(t)\int_{\Omega_z^\xi\cap(\Omega_z^\xi-\varepsilon)} (|u_z^\xi(s+\varepsilon)-u_z^\xi(s)|\wedge1)|\xi|\,ds \le|\xi|\int_{\Omega_z^\xi\cap(\Omega_z^\xi-\varepsilon)}\int\chi_{(s,s+\varepsilon)}(t)\,d\mu_z^\xi(t)\,ds \le\varepsilon|\xi|\int_{\Omega_z^\xi}d\mu_z^\xi(t)

where we have used Fubini’s theorem and that χ(s,s+ε)(t)=χ(t−ε,t)(s)\chi_{(s,s+\varepsilon)}(t)=\chi_{(t-\varepsilon,t)}(s). Combining the previous estimates, we end up with

εd−1∫Q∑ξ∈V∑i∈εZd(∣ξ⋅(u(εy+i+εξ)−u(εy+i))∣∧1) dy≤∑ξ∈V∣ξ∣∫ξ⊥∫Ωzξdμzξ dHd−1(z)≤∑ξ∈V∣ξ∣Λuξ=:M.\varepsilon^{d-1}\int_Q \sum_{\xi\in V}\sum_{i\in\varepsilon\mathbb{Z}^d} (|\xi\cdot(u(\varepsilon y+i+\varepsilon\xi)-u(\varepsilon y+i))|\wedge1)\,dy \le\sum_{\xi\in V}|\xi|\int_{\xi^\perp}\int_{\Omega_z^\xi}d\mu_z^\xi\,d\mathcal{H}^{d-1}(z) \le\sum_{\xi\in V}|\xi|\Lambda_u^\xi=:M.

Hence, there is a set Qε⊂QQ^\varepsilon\subset Q of measure at least 1/21/2 such that for any y∈Qεy\in Q^\varepsilon:

εd−1∑ξ∈V∑i∈εZd(∣ξ⋅(u(εy+i+εξ)−u(εy+i))∣∧1)≤2M.(15)\varepsilon^{d-1}\sum_{\xi\in V}\sum_{i\in\varepsilon\mathbb{Z}^d} (|\xi\cdot(u(\varepsilon y+i+\varepsilon\xi)-u(\varepsilon y+i))|\wedge1)\le2M. \tag*{(15)}

Let us choose yε∈Qε∩Qεy^\varepsilon\in Q^\varepsilon\cap Q_\varepsilon, where Qε⊂QQ_\varepsilon\subset Q is the set given by Proposition 7 in Appendix B. For ε>0\varepsilon>0 small enough and for any i∈εZdi\in\varepsilon\mathbb{Z}^d we denote by

Qεi:=εyε+i+ε[0,1)dQ_\varepsilon^i:=\varepsilon y^\varepsilon+i+\varepsilon[0,1)^d

the cube corresponding to ii, by

Vξ(Qεi):={j∈εZd:εyε+j+[0,ε)ξ is an edge of Qεi}V_\xi(Q_\varepsilon^i):=\{j\in\varepsilon\mathbb{Z}^d:\varepsilon y^\varepsilon+j+[0,\varepsilon)\xi\text{ is an edge of }Q_\varepsilon^i\}

the vertices of QεiQ_\varepsilon^i which are endpoints of edges parallel to ξ\xi (more precisely, we consider for any such edge the lowest endpoint, with respect to the order in Zd\mathbb{Z}^d, that is the one included in QεiQ_\varepsilon^i), and we define the function uεu^\varepsilon for every x∈Ωx \in\Omega as

  • the multilinear interpolation of the values of uu at the vertices of the cube QεiQ_\varepsilon^i which contains xx, if Qεi‾⊂Ω\overline{Q_\varepsilon^i} \subset\Omega and if for all ξ∈V\xi\in V and j∈Vξ(Qεi)j \in V_\xi(Q_\varepsilon^i)

∣ξ⋅(u(εyεi+j+εξ)−u(εyεi+j))∣≤1;\left|\xi\cdot\left(u(\varepsilon y_\varepsilon^i+j+\varepsilon\xi)-u(\varepsilon y_\varepsilon^i+j)\right)\right| \leq1;
  • 00, else.

Observe that uεu^\varepsilon is a polynomial in the cubes of the first type, so in particular it does not jump therein, that it is continuous accross a facet between two such cubes, and that the number of cubes of the second type is bounded by (2/ε)d−12M(2/\varepsilon)^{d-1}2M so that, denoting by BεB^\varepsilon the union of cubes of the second type, it holds that

Juε⊂∂Bε,Hd−1(∂Bε)≤CM,∣Bε∣≤CMε(16)J_{u^\varepsilon} \subset\partial B^\varepsilon,\qquad\mathcal{H}^{d-1}(\partial B^\varepsilon) \leq CM,\qquad|B^\varepsilon| \leq CM\varepsilon \tag*{(16)}

for a dimensional C>0C>0. In particular, the convergence properties of the multilinear interpolations stated in Proposition 7 also hold for uεu^\varepsilon, since they differ, in Ω\Omega, on a set of vanishing measure. In addition, if QεiQ_\varepsilon^i is a cube of the first type, we have that

∫Qεi∣e(uε)∣ dx≤Cεd−1∑ξ∈V∑j∈Vξ(Qεi)∣ξ⋅(u(εyεi+j+εξ)−u(εyεi+j))∣,\int_{Q_\varepsilon^i} |e(u^\varepsilon)|\,dx \leq C\varepsilon^{d-1}\sum_{\xi\in V}\sum_{j\in V_\xi(Q_\varepsilon^i)}\left|\xi\cdot\left(u(\varepsilon y_\varepsilon^i+j+\varepsilon\xi)-u(\varepsilon y_\varepsilon^i+j)\right)\right|,

by Proposition 4 (in Appendix § 3) for p=1p=1, and a change of variables. Therefore, summing the above two estimates over the cubes all included in Ω\Omega (which cover Ωε\Omega_\varepsilon), we find that, for ε\varepsilon small enough, uε∈SBV∞(Ω)u^\varepsilon\in SBV^\infty(\Omega) with

∫Ωε∣e(uε)∣ dx+Hd−1(Juε∩Ωε)≤2CM,\int_{\Omega_\varepsilon}|e(u^\varepsilon)|\,dx+\mathcal{H}^{d-1}(J_{u^\varepsilon}\cap\Omega_\varepsilon)\leq2CM,

for a dimensional constant C>0C>0, and this gives (2) recalling that M=∑ξ∈V∣ξ∣ΛuξM=\sum_{\xi\in V}|\xi|\Lambda_u^\xi.

Proof of Corollary 1. Let (uε)ε>0(u^\varepsilon)_{\varepsilon>0} be the sequence provided by Theorem 1 and fix Ω′⋐Ω\Omega'\Subset\Omega. By the slicing properties of BVBV functions and the bound (2) it follows that defining \Lambda^\varepsilon:=\int|e(u^\varepsilon)|\,dx+\mathcal{H}^{d-1}\mathbin{\vrule height 1.4ex depth -0.4ex width .07ex\vrule height .07ex depth -0.02ex width .7ex}\partial B^\varepsilon, then Λε(Ω′)≤CΛu\Lambda^\varepsilon(\Omega')\leq C\Lambda_u and one has, for all ξ∈Sd−1\xi\in\mathbb{S}^{d-1} and all Borel B⊂Ω′B\subset\Omega',

∫ξ⊥(∫Bzξ∩Juεξ∣e((uε)ξ)∣ dt+#(J(uε)ξ∩Bzξ)) dHd−1(z)≤Λε(B)(17)\int_{\xi^\perp}\left(\int_{B_z^\xi\cap J_{u^\varepsilon}^\xi}|e((u^\varepsilon)^\xi)|\,dt+\#(J_{(u^\varepsilon)^\xi}\cap B_z^\xi)\right)\,d\mathcal{H}^{d-1}(z)\leq\Lambda^\varepsilon(B) \tag*{(17)}

so that (uε)(u^\varepsilon) are equibounded in GBD(Ω′)GBD(\Omega'), in the sense of [21]. Moreover, as uε→uu^\varepsilon\to u in L0(Ω;Rd)L^0(\Omega;\mathbb{R}^d), by [19 Theorem 1.1] we conclude that u∈GBD(Ω′)u\in GBD(\Omega'). In fact, the convergence of uεu^\varepsilon to uu implies that the infinitesimal rigid motions can be taken as equal to 00 (alternatively, one could use [21 Corollary 11.2 and Theorem 11.3] together with the fact that since uε→uu^\varepsilon\to u in L0(Ω;Rd)L^0(\Omega;\mathbb{R}^d), there is an increasing function ψ0:R+→R+\psi_0:\mathbb{R}^+\to\mathbb{R}^+ with lim⁡s→+∞ψ0(s)=+∞\lim_{s\to+\infty}\psi_0(s)=+\infty such that ∥ψ0(uε)∥L1(Ω)\|\psi_0(u^\varepsilon)\|_{L^1(\Omega)} is uniformly bounded w.r.t. ε>0\varepsilon>0, see e.g. [27 Lemma 2.1]). Then we conclude that u∈GBD(Ω)u\in GBD(\Omega) by the arbitrariness of Ω′⋐Ω\Omega'\Subset\Omega.

Proof of Corollary 2. For any uku_k let (ukε)ε>0(u_k^\varepsilon)_{\varepsilon>0} be the approximating family provided by Theorem 1. Then for any k∈N∖{0}k\in\mathbb{N}\setminus\{0\} we can find εk>0\varepsilon_k>0 such that, for u~k:=ukεk\widetilde{u}_k:=u_k^{\varepsilon_k} it holds that u~k∈SBV∞(Ω;Rd)\widetilde{u}_k\in SBV^\infty(\Omega;\mathbb{R}^d) with u~k∈∂∗Bεk\widetilde{u}_k\in\partial^*B^{\varepsilon_k} and BεkB^{\varepsilon_k} union of finitely many cubes of sidelength εk\varepsilon_k included in Ω\Omega, such that

∥arctan⁡(u~k−uk)∥L1≤1k,\left\|\arctan(\tilde u_k-u_k)\right\|_{L^1}\leq\frac{1}{k},
∫Ωεk∣e(u~k)∣ dx+Hd−1(∂∗Bεk∩Ωεk)≤CΛukVk.(18)\int_{\Omega_{\varepsilon_k}} |e(\tilde u_k)|\,dx+\mathcal{H}^{d-1}\left(\partial^*B^{\varepsilon_k}\cap\Omega_{\varepsilon_k}\right)\leq C\Lambda_{u_k}^{\mathcal{V}_k}. \tag*{(18)}

Recalling (12), it follows that the sequence (u~k)k(\tilde u_k)_k satisfies the assumptions of [19 Theorem 1.1] on any Ωεk\Omega_{\varepsilon_k} for k≥k‾k\geq\overline{k}, and therefore by that compactness result and by a diagonal argument on Ωεk\Omega_{\varepsilon_k}, up to a (not relabelled) subsequence, there are a Caccioppoli partition P=(Pn)nP=(P_n)_n of Ω\Omega, a sequence of piecewise rigid motions (ak)k(a_k)_k satisfying (4), and u∈GBD(Ω)u\in GBD(\Omega) such that

u~k−ak→ua.e. in Ω.(19)\tilde u_k-a_k\to u\quad\text{a.e. in }\Omega. \tag*{(19)}

By the first in (13) we deduce (5a). At this stage, (5b) follows arguing exactly as in [19 Section 3.3: Lower semicontinuity]. □

Proof of Corollary 3. We prove that GBVE(Ω;Rd)⊂GBD(Ω)GBV^\mathcal{E}(\Omega;\mathbb{R}^d)\subset GBD(\Omega), the opposite inclusion being true by definition of GBD(Ω)GBD(\Omega) in [21]. The proof follows by averaging. We fix an orthonormal basis (ei)i=1d(e_i)_{i=1}^d, and let V^={ei,i=1,…,d;ei+ej2,1≤i<j≤d}\widehat V=\left\{e_i,i=1,\ldots,d;\frac{e_i+e_j}{\sqrt{2}},1\leq i<j\leq d\right\}. Then, we have that:

\fint_{SO(d)}\sum_{\xi\in\widehat V}\Lambda_u^{R\xi}\,d\mu(R)=\frac{d(d+1)}{2}\fint_{\mathbb{S}^{d-1}}\Lambda_u^\xi\,d\mathcal{H}^{d-1}(\xi).

(for μ\mu the Haar measure over SO(d)SO(d)), which is finite whenever u∈GBVE(Ω;Rd)u\in GBV^\mathcal{E}(\Omega;\mathbb{R}^d). We then pick any rotation for which ∑ξ∈RV^Λuξ\sum_{\xi\in R\widehat V}\Lambda_u^\xi is less than average and consider the basis (Rei)i=1d(Re_i)_{i=1}^d in Theorem 1. We conclude since ΛuR(ei+ej)=2ΛuR(ei+ej)/2\Lambda_u^{R(e_i+e_j)}=\sqrt{2}\Lambda_u^{R(e_i+e_j)/\sqrt{2}}. □

Proof of Proposition 2. We argue as in the proof of Theorem 1, with the measures μzξ\mu_z^\xi replaced by the measures

(μzp)ξ:=∣(uzξ)p∣ dL1+H0⌞Juzξ,(\mu_z^p)^\xi:=|(u_z^\xi)^p|\,d\mathcal{L}^1+\mathcal{H}^0\mathbin{\llcorner}J_{u_z^\xi},

thus evaluating

εd−1∫Q∑ξ∈V∑i∈εZd(∣ξ⋅(u(εy+i+εξ)−u(εy+i))∣p∧1) dy≤∑ξ∈V∫ξ⊥∫Ωεdμz,ξp dHd−1(z)≤Λup,V.\varepsilon^{d-1}\int_Q\sum_{\xi\in\mathcal{V}}\sum_{i\in\varepsilon\mathbb{Z}^d}\left(|\xi\cdot(u(\varepsilon y+i+\varepsilon\xi)-u(\varepsilon y+i))|^p\wedge1\right)\,dy\leq\sum_{\xi\in\mathcal{V}}\int_{\xi^\perp}\int_{\Omega_\varepsilon}d\mu_{z,\xi}^p\,d\mathcal{H}^{d-1}(z)\leq\Lambda_u^{p,\mathcal{V}}.

Therefore the proof proceeds as done for Theorem 1, first choosing suitably yεy^\varepsilon from the previous inequality and Proposition 7 in Appendix B, and then replacing ∣ξ⋅(u(εyε+i+εξ)−u(εyε+i))∣|\xi\cdot(u(\varepsilon y^\varepsilon+i+\varepsilon\xi)-u(\varepsilon y^\varepsilon+i))| by ∣ξ⋅(u(εyε+i+εξ)−u(εyε+i))∣p|\xi\cdot(u(\varepsilon y^\varepsilon+i+\varepsilon\xi)-u(\varepsilon y^\varepsilon+i))|^p, to define the approximating functions uεu^\varepsilon converging in measure to uu. We still control BεB^\varepsilon, the union of cubes of the second type, as in (11); on any Q‾ε=εyε+i+ε[0,1)d\overline{Q}_\varepsilon=\varepsilon y^\varepsilon+i+\varepsilon[0,1)^d of the first type, we have that

∫Q‾ε∣e(uε)∣p dx≤Cεd−1∑ξ∈V∣ξ⋅(u(εyε+i+εξ)−u(εyε+i))∣p\int_{\overline{Q}_\varepsilon}|e(u^\varepsilon)|^p\,dx\leq C\varepsilon^{d-1}\sum_{\xi\in\mathcal{V}}|\xi\cdot(u(\varepsilon y^\varepsilon+i+\varepsilon\xi)-u(\varepsilon y^\varepsilon+i))|^p

by Proposition 4 for p>1p>1 and a change of variables. Summing up we obtain the bound (7). □

Proof of Corollary 4. For any uku_k let (ukε)ε>0(u_k^\varepsilon)_{\varepsilon>0} be the approximating family provided by Proposition 2. Then for any k∈N∖{0}k\in\mathbb{N}\setminus\{0\} we can find εk>0\varepsilon_k>0 such that, for u~k:=ukεk\tilde u_k:=u_k^{\varepsilon_k} it holds that u~k∈SBV∞(Ω;Rd)\widetilde{u}_k \in SBV^\infty(\Omega;\mathbb{R}^d) with Ju~k⊂∂∗BεkJ_{\widetilde{u}_k} \subset\partial^* B^{\varepsilon_k} and BεkB^{\varepsilon_k} union of finitely many cubes of sidelength εk\varepsilon_k included in Ω\Omega, such that

∥arctan⁡(u~k−uk)∥L1≤1k,∫Ωεk∣e(u~k)∣p dx+Hd−1(∂∗Bεk∩Ωεk)≤CΛukp,Vk.(20)\|\arctan(\widetilde{u}_k-u_k)\|_{L^1} \le\frac{1}{k}, \qquad \int_{\Omega_{\varepsilon_k}} |e(\widetilde{u}_k)|^p\,dx+\mathcal{H}^{d-1}(\partial^*B^{\varepsilon_k}\cap\Omega_{\varepsilon_k})\le C\Lambda^{p,V_k}_{u_k}. \tag*{(20)}

Then (u~k)k(\widetilde{u}_k)_k satisfies the assumptions of [18 Theorem 1.1] on any Ωεk\Omega_{\varepsilon_k} for k≥k‾k\ge\overline{k}, and therefore by that compactness result and by a diagonal argument on Ωεk\Omega_{\varepsilon_k}, up to a (not relabelled) subsequence, there are a Caccioppoli partition P=(Pn)n\mathcal{P}=(P_n)_n of Ω\Omega, a sequence of piecewise rigid motions (ak)k(a_k)_k satisfying (4), and u∈GSBDp(Ω)u\in GSBD^p(\Omega) such that

u~k−ak→ua.e. in Ω.(21)\widetilde{u}_k-a_k\to u \quad\text{a.e. in }\Omega. \tag*{(21)}

By the first in (15) we deduce (9a). Eventually, we get (9b) and (9c) by [21 Theorem 11.3] (recall [27 Lemma 2.1]) applied to (uk−ak)k(u_k-a_k)_k.

Discrete energy estimate

In what follows, (ei)i=1d(e_i)_{i=1}^d denotes the canonical basis of Rd\mathbb{R}^d.

Lemma 3. Consider the unit cube Q=[0,1]d⊂RdQ=[0,1]^d\subset\mathbb{R}^d. Let v∈(Rd){0,1}dv\in(\mathbb{R}^d)^{\{0,1\}^d} be given at all vertices of QQ such that vi(x+ei)=vi(x)v_i(x+e_i)=v_i(x) for any x∈{0,1}dx\in\{0,1\}^d with xi=0x_i=0 and vi(x+ei+ej)+vj(x+ei+ej)=vi(x)+vj(x)v_i(x+e_i+e_j)+v_j(x+e_i+e_j)=v_i(x)+v_j(x) for any x∈{0,1}dx\in\{0,1\}^d with xi=xj=0x_i=x_j=0. For x∈Qx\in Q, we also denote by v(x)v(x) the multilinear interpolation of the values vv at the vertices (affine on each [x,x+ei][x,x+e_i] for any x∈Qx\in Q with xi=0x_i=0). Then e(v)=0e(v)=0 in QQ (so that, in fact, vv is affine with skew-symmetric gradient).

Proof. We first assume d=2d=2. Then,

v(x)=(1−x1)(1−x2)v(0,0)+(1−x1)x2v(0,1)+x1(1−x2)v(1,0)+x1x2v(1,1).v(x)=(1-x_1)(1-x_2)v(0,0)+(1-x_1)x_2v(0,1)+x_1(1-x_2)v(1,0)+x_1x_2v(1,1).

One easily sees that ∂1v1=∂2v2=0\partial_1v_1=\partial_2v_2=0 everywhere. One has

∂1v2(x)+∂2v1(x)=(1−x2)(v2(1,0)−v2(0,0))+x2(v2(1,1)−v2(0,1))+(1−x1)(v1(0,1)−v1(0,0))+x1(v1(1,1)−v1(1,0)).\partial_1v_2(x)+\partial_2v_1(x)=(1-x_2)(v_2(1,0)-v_2(0,0))+x_2(v_2(1,1)-v_2(0,1)) +(1-x_1)(v_1(0,1)-v_1(0,0))+x_1(v_1(1,1)-v_1(1,0)).

Since by assumption v1(0,0)=v1(1,0)v_1(0,0)=v_1(1,0), v1(0,1)=v1(1,1)v_1(0,1)=v_1(1,1), v2(0,0)=v2(0,1)v_2(0,0)=v_2(0,1), v2(1,0)=v2(1,1)v_2(1,0)=v_2(1,1), this is also:

∂1v2(x)+∂2v1(x)=(1−x2)(v2(1,1)−v2(0,0))+x2(v2(1,1)−v2(0,0))+(1−x1)(v1(1,1)−v1(0,0))+x1(v1(1,1)−v1(0,0))=v2(1,1)−v2(0,0)+v1(1,1)−v1(0,0)=0,\begin{aligned} \partial_1v_2(x)+\partial_2v_1(x)=(1-x_2)(v_2(1,1)-v_2(0,0))+x_2(v_2(1,1)-v_2(0,0)) +(1-x_1)(v_1(1,1)-v_1(0,0))+x_1(v_1(1,1)-v_1(0,0)) \\ &= v_2(1,1)-v_2(0,0)+v_1(1,1)-v_1(0,0)=0, \end{aligned}

where the last equality follows again by assumption. Hence e(v)=0e(v)=0.

Now, consider d≥3d\ge3: then for any {i,j}⊂{1,…,d}\{i,j\}\subset\{1,\ldots,d\}, the case d=2d=2 shows that e(v)i,j(x)=0e(v)_{i,j}(x)=0 for any xx with (xi,xj)∈[0,1]2(x_i,x_j)\in[0,1]^2 and xk∈{0,1}x_k\in\{0,1\}, k∉{i,j}k\notin\{i,j\}. Since at any other x∈Qx\in Q, e(v)i,j(x)e(v)_{i,j}(x) is a convex interpolation of those values, we find that e(v)i,j=0e(v)_{i,j}=0 everywhere. Similarly, e(v)i,i(x)=∂ivi(x)=0e(v)_{i,i}(x)=\partial_i v_i(x)=0 for any x∈Qx\in Q and i∈{1,…,d}i\in\{1,\ldots,d\}. Hence e(v)=0e(v)=0.

Proposition 4. *Consider the unit cube QQ, v∈(Rd){0,1}dv\in(\mathbb{R}^d)^{\{0,1\}^d} given at all vertices of QQ, and the dd-linear interpolation of vv inside QQ. Then for every p∈[1,+∞)p\in[1,+\infty) there is a constant C>0C>0

depending on dd and pp such that:

∫Q∣e(v)∣p dx≤C(∑i=1d∑x∈{0,1}dxi=0∣vi(x+ei)−vi(x)∣p+∑i,j=1d∑x∈{0,1}dxi=xj=0∣vi(x+ei+ej)+vj(x+ei+ej)−vi(x)−vj(x)∣p)(22)\int_Q |e(v)|^p\,dx \le C\left(\sum_{i=1}^{d}\sum_{\substack{x\in\{0,1\}^d\\x_i=0}} |v_i(x+e_i)-v_i(x)|^p +\sum_{i,j=1}^{d}\sum_{\substack{x\in\{0,1\}^d\\x_i=x_j=0}} |v_i(x+e_i+e_j)+v_j(x+e_i+e_j)-v_i(x)-v_j(x)|^p\right) \tag*{(22)}

Proof. If not, there is a sequence (vn)n≥1(v_n)_{n\ge1} which satisfies the reverse inequality, with CC replaced with nn, arbitrarily large. Renormalizing, we may assume that ∥e(vn)∥Lp(Q)=1\|e(v_n)\|_{L^p(Q)}=1 and thus that the right-hand side of (17), evaluated for vnv_n, goes to zero as n→∞n\to\infty. By Poincaré-Korn inequality, there is a sequence rigid motions (affine functions an(x)=Anx+bna_n(x)=A_nx+b_n with skew-symmetric gradient AnA_n) such that:

∫Q∣vn(x)−an(x)∣p dx≤c∫Q∣e(vn)∣p dx=c\int_Q |v_n(x)-a_n(x)|^p\,dx \le c\int_Q |e(v_n)|^p\,dx=c

where cc depends only on the dimension and on pp. Letting vn′=vn−anv'_n=v_n-a_n, it is bounded in Lp(Q)L^p(Q). Since it is, in fact, finite-dimensional (it is a polynomial of degree at most dd), it is relatively compact and, up to a subsequence, converges to a limit v′v' with ∥e(v′)∥Lp(Q)=1\|e(v')\|_{L^p(Q)}=1. Yet the right-hand side of (17), for v′v', vanishes, so that v′v' satisfies the assumptions of Lemma 3 and we deduce e(v′)=0e(v')=0. This is a contradiction, hence the corollary is true. An alternative proof consists in arguing that both left-hand and right-hand sides of (17) define ppth-powers of norms on the finite dimensional space (Rd){0,1}d/∼(\mathbb{R}^d)^{\{0,1\}^d}/\sim where u∼vu\sim v if and only if the dd-linear interpolation of u−vu-v is an infinitesimal rigid motion.

Convergence of Discretizations

Let Q(x)=χ[−1/2,1/2)d(x)Q(x)=\chi_{[-1/2,1/2)^d}(x), Δ(x)=∏i=1d(1−∣xi∣)+\Delta(x)=\prod_{i=1}^{d}(1-|x_i|)^+, for x=(x1,…,xd)∈Rdx=(x_1,\ldots,x_d)\in\mathbb{R}^d, and for every k∈Nk\in\mathbb{N} let Tk:Rm→RmT_k:\mathbb{R}^m\to\mathbb{R}^m be defined by

Tk(t)=t(1∧k∣t∣).T_k(t)=t\left(1\wedge\frac{k}{|t|}\right).

We remark that for u∈L1(Rd;Rm)u\in L^1(\mathbb{R}^d;\mathbb{R}^m) it holds that for any R>0R>0,

lim⁡ℓ→∞∣{x∈BR(0):∣u(x)∣>ℓ}∣=∣⋂ℓ>0{x∈BR(0):∣u(x)∣>ℓ}∣=∣{x∈BR(0):∣u(x)∣=+∞}∣=0.(23)\lim_{\ell\to\infty}\left|\{x\in B_R(0):|u(x)|>\ell\}\right| =\left|\bigcap_{\ell>0}\{x\in B_R(0):|u(x)|>\ell\}\right| =\left|\{x\in B_R(0):|u(x)|=+\infty\}\right|=0. \tag*{(23)}

Moreover, for any ε>0\varepsilon>0 and y∈Qy\in Q we denote by

Dyε:={εy+z:z∈εZd},D_y^\varepsilon:=\{\varepsilon y+z:z\in\varepsilon\mathbb{Z}^d\},

the discretization of Rd\mathbb{R}^d formed by cubes (with the same orientation) of sidelength ε\varepsilon and anchor point yy and by uyε:Rd→Rmu_y^\varepsilon:\mathbb{R}^d\to\mathbb{R}^m the corresponding discretized function defined by

uyε(x):=∑xˉ∈Dyεu(xˉ)Δ(x−xˉε)for all x∈Rd.(24)u_y^\varepsilon(x):=\sum_{\bar{x}\in D_y^\varepsilon}u(\bar{x})\Delta\left(\frac{x-\bar{x}}{\varepsilon}\right)\quad\text{for all }x\in\mathbb{R}^d. \tag*{(24)}

In the following notation for the norms, we denote by L1(Bk)L^1(B_k) the space L1(Bk(0);Rm)L^1(B_k(0);\mathbb{R}^m), with Bk≡Bk(0)⊂RdB_k\equiv B_k(0)\subset\mathbb{R}^d the ball of center $0andradiusand radiusk.

Lemma 5. Let u∈L0(Rd;Rm)u \in L^0(\mathbb{R}^d;\mathbb{R}^m) and assume that for all ε>0\varepsilon> 0 there exists yε∈Qy_\varepsilon\in\mathbb{Q} such that for any k∈N∖{0}k \in\mathbb{N} \setminus\{0\}

lim⁡ε→0∥∑xˉ∈DyεεTk(u(xˉ))Q(−xˉε)−Tk(u)∥L1(Bk)+∥∑xˉ∈DyεεTk(u(xˉ))Δ(−xˉε)−Tk(u)∥L1(Bk)=0.(25)\lim_{\varepsilon\to0} \left\| \sum_{\bar{x} \in\mathcal{D}_{y_\varepsilon}^{\varepsilon}} T_k(u(\bar{x}))Q\left(\frac{-\bar{x}}{\varepsilon}\right) - T_k(u) \right\|_{L^1(B_k)} + \left\| \sum_{\bar{x} \in\mathcal{D}_{y_\varepsilon}^{\varepsilon}} T_k(u(\bar{x}))\Delta\left(\frac{-\bar{x}}{\varepsilon}\right) - T_k(u) \right\|_{L^1(B_k)} = 0. \tag*{(25)}

Then

lim⁡ε→0∥Tk(uyεε)−Tk(u)∥L1(Bk)=0for any k∈N∖{0}.(26)\lim_{\varepsilon\to0} \left\| T_k(u_{y_\varepsilon}^{\varepsilon}) - T_k(u) \right\|_{L^1(B_k)} = 0 \quad\text{for any } k \in\mathbb{N} \setminus\{0\}. \tag*{(26)}

Proof. Let us fix k∈N∖{0}k \in\mathbb{N} \setminus\{0\}. Observe that if ℓ>k\ell> k, since Tk=Tk∘TℓT_k = T_k \circ T_\ell and TkT_k is 11-Lipschitz,

∥Tk(∑xˉTℓ(u(xˉ))Δ(−xˉε))−Tk(u)∥L1(Bk)≤∥∑xˉTℓ(u(xˉ))Δ(−xˉε)−Tℓ(u)∥L1(Bk)→0(27)\left\| T_k\left(\sum_{\bar{x}} T_\ell(u(\bar{x}))\Delta\left(\frac{-\bar{x}}{\varepsilon}\right)\right) - T_k(u) \right\|_{L^1(B_k)} \leq\left\| \sum_{\bar{x}} T_\ell(u(\bar{x}))\Delta\left(\frac{-\bar{x}}{\varepsilon}\right) - T_\ell(u) \right\|_{L^1(B_k)} \to0 \tag*{(27)}

as ε→0\varepsilon\to0, thanks to (25). This means we just need to bound

∥Tk(uyεε)−Tk(∑xˉTℓ(u(xˉ))Δ(−xˉε))∥L1(Bk).\left\| T_k(u_{y_\varepsilon}^{\varepsilon}) - T_k\left(\sum_{\bar{x}} T_\ell(u(\bar{x}))\Delta\left(\frac{-\bar{x}}{\varepsilon}\right)\right) \right\|_{L^1(B_k)}.

Observe that this quantity is zero, except possibly in the cubes with center a point xˉ\bar{x} such that ∣u(xˉ)∣>ℓ|u(\bar{x})| > \ell, in which case it is bounded (in norm) by 2k2k. Hence we get the bound

∥Tk(uyεε)−Tk(∑xˉTℓ(u(xˉ))Δ(−xˉε))∥L1(Bk)≤2k(2ε)d#{xˉ∈Dyεε∩Bk+εd:∣u(xˉ)∣>ℓ}.(28)\left\| T_k(u_{y_\varepsilon}^{\varepsilon}) - T_k\left(\sum_{\bar{x}} T_\ell(u(\bar{x}))\Delta\left(\frac{-\bar{x}}{\varepsilon}\right)\right) \right\|_{L^1(B_k)} \leq2k(2\varepsilon)^d \#\{\bar{x} \in\mathcal{D}_{y_\varepsilon}^{\varepsilon} \cap B_{k+\varepsilon\sqrt{d}} : |u(\bar{x})| > \ell\}. \tag*{(28)}

For any x∈xˉ+[−ε/2,ε/2)dx \in\bar{x}+[-\varepsilon/2,\varepsilon/2)^d with ∣u(xˉ)∣>ℓ|u(\bar{x})| > \ell, either ∣u(x)∣>ℓ/2|u(x)| > \ell/2, or

∣Tℓ(u(xˉ))−Tℓ(u(x))∣≥∣Tℓ(u(xˉ))∣−∣Tℓ(u(x))∣=ℓ−∣u(x)∣≥ℓ2.|T_\ell(u(\bar{x})) - T_\ell(u(x))| \geq|T_\ell(u(\bar{x}))| - |T_\ell(u(x))| = \ell- |u(x)| \geq\frac{\ell}{2}.

Hence (for ℓ\ell large enough),

εd#{xˉ∈Dyεε∩Bk+εd:∣u(xˉ)∣>ℓ}\varepsilon^d \#\{\bar{x} \in\mathcal{D}_{y_\varepsilon}^{\varepsilon} \cap B_{k+\varepsilon\sqrt{d}} : |u(\bar{x})| > \ell\}
≤∣{x∈Bk+εd:∣u(x)∣≥ℓ2}∣+∣{x∈Bℓ:∣∑xˉTℓ(u(xˉ))Q(x−xˉε)−Tℓ(u(x))∣≥ℓ2}∣\leq\left|\left\{x \in B_{k+\varepsilon\sqrt{d}} : |u(x)| \geq\frac{\ell}{2}\right\}\right| + \left|\left\{x \in B_\ell: \left|\sum_{\bar{x}} T_\ell(u(\bar{x}))Q\left(\frac{x-\bar{x}}{\varepsilon}\right) - T_\ell(u(x))\right| \geq\frac{\ell}{2}\right\}\right|
≤∣{x∈Bk+εd:∣u(x)∣>ℓ2}∣+2ℓ∫Bℓ∣∑xˉTℓ(u(xˉ))Q(x−xˉε)−Tℓ(u(x))∣ dx.\leq\left|\left\{x \in B_{k+\varepsilon\sqrt{d}} : |u(x)| > \frac{\ell}{2}\right\}\right| + \frac{2}{\ell}\int_{B_\ell}\left|\sum_{\bar{x}} T_\ell(u(\bar{x}))Q\left(\frac{x-\bar{x}}{\varepsilon}\right) - T_\ell(u(x))\right|\,dx.

We deduce from (25), (27), and (28) that for any ℓ>k\ell> k,

lim sup⁡ε→0∥Tk(uyεε)−Tk(u)∥L1(Bk)≤2d+1k∣{x∈Bk+εd:∣u(x)∣>ℓ2}∣.\limsup_{\varepsilon\to0} \left\| T_k(u_{y_\varepsilon}^{\varepsilon}) - T_k(u) \right\|_{L^1(B_k)} \leq2^{d+1}k \left|\left\{x \in B_{k+\varepsilon\sqrt{d}} : |u(x)| > \frac{\ell}{2}\right\}\right|.

Sending ℓ→∞\ell\to\infty, by (23) we obtain (26).

Recalling the notation (24), the following lemma holds.

Lemma 6. Let u∈L0(Rd;Rm)u \in L^0(\mathbb{R}^d;\mathbb{R}^m) and (uyεε)ε>0(u_{y_\varepsilon}^{\varepsilon})_{\varepsilon>0} be such that (26) holds. Then

uyεε→ulocally in L0(Rd;Rm).(29)u_{y_\varepsilon}^{\varepsilon} \to u \quad\text{locally in } L^0(\mathbb{R}^d;\mathbb{R}^m). \tag*{(29)}

Proof. Fix η>0\eta> 0, R>0R > 0. For any xx and any k∈N∖{0}k \in\mathbb{N}\setminus\{0\}, k≥2ηk \geq2\eta, it holds that if ∣uyεε(x)−u(x)∣>η|u_{y_\varepsilon}^{\varepsilon}(x)-u(x)|>\eta then either ∣u(x)∣>k/2|u(x)|>k/2, or ∣Tk(uyεε(x))−Tk(u(x))∣>η|T_k(u_{y_\varepsilon}^{\varepsilon}(x))-T_k(u(x))|>\eta. Hence, for k≥Rk \geq R,

∣{x∈BR(0):∣uyεε(x)−u(x)∣>η}∣≤∣{x∈BR(0):∣u(x)∣>k2}∣+1η∥Tk(uyεε)−Tk(u)∥L1(Bk).\left|\{x \in B_R(0): |u_{y_\varepsilon}^{\varepsilon}(x)-u(x)|>\eta\}\right| \leq\left|\{x \in B_R(0): |u(x)|>\frac{k}{2}\}\right|+\frac{1}{\eta}\left\|T_k(u_{y_\varepsilon}^{\varepsilon})-T_k(u)\right\|_{L^1(B_k)}.

Hence, for any fixed kk,

lim sup⁡ε→0∣{x∈BR(0):∣uyεε(x)−u(x)∣>η}∣≤∣{x∈BR(0):∣u(x)∣>k2}∣.\limsup_{\varepsilon\to0}\left|\{x \in B_R(0): |u_{y_\varepsilon}^{\varepsilon}(x)-u(x)|>\eta\}\right| \leq\left|\{x \in B_R(0): |u(x)|>\frac{k}{2}\}\right|.

The conclusion follows letting k→∞k \to\infty.

Proposition 7. For any u∈L0(Rd;Rm)u \in L^0(\mathbb{R}^d;\mathbb{R}^m) and for all ε>0\varepsilon> 0 there exists a measurable set Qε⊂Q=[0,1)dQ_\varepsilon\subset Q = [0,1)^d such that both lim⁡ε→0∣Qε∣=1\lim_{\varepsilon\to0}|Q_\varepsilon|=1 and uyεε→uu_{y_\varepsilon}^\varepsilon\to u locally in L0(Rd;Rm)L^0(\mathbb{R}^d;\mathbb{R}^m) if yε∈Qεy_\varepsilon\in Q_\varepsilon for every ε>0\varepsilon> 0.

Proof. By Lemma 5 and Lemma 6, it is enough to find a set QεQ_\varepsilon such that (20) holds for any choice of yε∈Qεy_\varepsilon\in Q_\varepsilon. For y∈Qy \in Q, we let:

Φε(y)=∑k≥12−k∫Bk∣∑x∈DyεTk(u(x‾))Q(x−x‾ε)−Tk(u)∣+∣∑x∈DyεTk(u(x‾))Δ(x−x‾ε)−Tk(u)∣ dx.\Phi_\varepsilon(y)=\sum_{k\geq1}2^{-k}\int_{B_k}\left|\sum_{x\in D^\varepsilon_y}T_k(u(\overline{x}))Q\left(\frac{x-\overline{x}}{\varepsilon}\right)-T_k(u)\right|+\left|\sum_{x\in D^\varepsilon_y}T_k(u(\overline{x}))\Delta\left(\frac{x-\overline{x}}{\varepsilon}\right)-T_k(u)\right|\,dx.

For any bounded function with compact support v∈L∞(Rd)v\in L^\infty(\mathbb{R}^d), and a continuous function ψ:Rd→[0,1]\psi:\mathbb{R}^d\to[0,1] with compact support such that ∑z∈Zdψ(x−z)=1\sum_{z\in\mathbb{Z}^d}\psi(x-z)=1 for any x∈Rdx\in\mathbb{R}^d, one has

∫Q∣∑x∈Dyεv(x‾)ψ(x−x‾ε)−v(x)∣ dx dy=∫∣∫Q∑x∈Dyε(v(x‾)−v(x))ψ(x−x‾ε) dy∣ dx\int_Q\left|\sum_{x\in D^\varepsilon_y}v(\overline{x})\psi\left(\frac{x-\overline{x}}{\varepsilon}\right)-v(x)\right|\,dx\,dy=\int\left|\int_Q\sum_{x\in D^\varepsilon_y}\left(v(\overline{x})-v(x)\right)\psi\left(\frac{x-\overline{x}}{\varepsilon}\right)\,dy\right|\,dx
≤y′=εy∫∫Q∑z∈εZd∣v(y′+z)−v(x)∣ψ(x−z−y′ε)ε−d dy′ dx=∫∫ε−dψ(x−y′ε)∣v(y′)−v(x)∣ dy′ dx\mathrel{\leq}^{y'=\varepsilon y}\int\int_{Q}\sum_{z\in\varepsilon\mathbb{Z}^d}|v(y'+z)-v(x)|\psi\left(\frac{x-z-y'}{\varepsilon}\right)\varepsilon^{-d}\,dy'\,dx=\int\int\varepsilon^{-d}\psi\left(\frac{x-y'}{\varepsilon}\right)|v(y')-v(x)|\,dy'\,dx
=x−y′=εξ∫ψ(ξ)∥v(⋅−εξ)−v∥L1 dξ.\mathrel{=}^{x-y'=\varepsilon\xi}\int\psi(\xi)\|v(\cdot-\varepsilon\xi)-v\|_{L^1}\,d\xi.

where we have set ψε:=ψ(⋅/ε)ε−d\psi_\varepsilon:=\psi(\cdot/\varepsilon)\varepsilon^{-d}. Hence, for v=Tk(u)χBkv=T_k(u)\chi_{B_k} and ψ=Q,Δ\psi=Q,\Delta,

∫QΦε(y) dy≤∑k≥12−k∫(Q(ξ)+Δ(ξ))∥Tk(u)(⋅−εξ)−Tk(u)∥L1(Bk) dξ.\int_Q\Phi_\varepsilon(y)\,dy\leq\sum_{k\geq1}2^{-k}\int(Q(\xi)+\Delta(\xi))\|T_k(u)(\cdot-\varepsilon\xi)-T_k(u)\|_{L^1(B_k)}\,d\xi.

Since the terms in the sum above are uniformly bounded (by 4k2−k(1+∣Bk∣)4k2^{-k}(1+|B_k|) for ε\varepsilon small enough) and go to zero, we deduce that

lim⁡ε→0∫QΦε(y) dy=0.\lim_{\varepsilon\to0}\int_Q\Phi_\varepsilon(y)\,dy=0.

We can then choose Qε:={y∈Q:Φε(y)≤∫QΦε}Q_\varepsilon:=\{y\in Q:\Phi_\varepsilon(y)\leq\sqrt{\int_Q\Phi_\varepsilon}\}, which is such that ∣Q∖Qε∣≤∫QΦε|Q\setminus Q_\varepsilon|\leq\sqrt{\int_Q\Phi_\varepsilon}.

If one chooses yε∈Qεy_\varepsilon\in Q_\varepsilon, then for each kk, both L1L^1 norms in (20) are bounded by 2k∫QΦε2^k\sqrt{\int_Q\Phi_\varepsilon}, which goes to zero. Then we conclude by Lemma 6. □

Remark 2. Proposition 7 implies that uyεεu_{y_\varepsilon}^\varepsilon converge to uu a.e. in Rd\mathbb{R}^d, up to a subsequence εn↓0\varepsilon_n\downarrow0. We remark that such convergence may be recovered directly, without using Lemma 6, by modifying the argument in the proof of Proposition 7 as follows. With the notation in the proof of Proposition 7, first let εn\varepsilon_n such that ∫QΦεn≤2−n\sqrt{\int_Q\Phi_{\varepsilon_n}}\leq2^{-n}. Then, letting Qn:=⋂h≥nQεhQ_n:=\bigcap_{h\geq n}Q_{\varepsilon_h}, we find that ∣Q∖Qn∣≤2−n+1|Q\setminus Q_n|\leq2^{-n+1} and for any y∈Qny\in Q_n and h≥nh\geq n, Φεh(y)≤2−h\Phi_{\varepsilon_h}(y)\leq2^{-h}. Therefore, for Q~:=⋃n≥1Qn\widetilde{Q}:=\bigcup_{n\geq1}Q_n, we find that ∣Q~∣=1|\widetilde{Q}|=1 and for any y∈Q~y\in\widetilde{Q}, Φεh(y)≤2−h\Phi_{\varepsilon_h}(y)\leq2^{-h} for hh large enough (depending on yy).

Next, once the main result is proven, we can choose a (further) subsequence εk↓0\varepsilon_k\downarrow0 such that for each kk,

∥Tk(uyεkεk)−Tk(u)∥L1(Bk)≤2−k.\|T_k(u_{y_{\varepsilon_k}}^{\varepsilon_k})-T_k(u)\|_{L^1(B_k)}\leq2^{-k}.

Then, ∑k∣(Tk(uyεkεk(x))−Tk(u(x)))χBk(x)∣<+∞\sum_k|(T_k(u_{y_{\varepsilon_k}}^{\varepsilon_k}(x))-T_k(u(x)))\chi_{B_k}(x)|<+\infty for a.e. x∈Rdx\in\mathbb{R}^d, which shows that

uyεkεk→ua.e. in Rd as k→+∞.u_{y_{\varepsilon_k}}^{\varepsilon_k}\to u\quad\text{a.e. in }\mathbb{R}^d\text{ as }k\to+\infty.

As a consequence, if one considers instead a subsequence in Proposition 7, one can replace QεQ_{\varepsilon} by the set Q~\widetilde{Q} of full measure, independent on ε\varepsilon, and assume that the convergence is almost everywhere.

Acknowledgements

Acknowledgements. V.C. acknowledges the support of the MUR - PRIN project 2022J4FYNJ CUP B53D23009320006 “Variational methods for stationary and evolution problems with singularities and interfaces”, PNRR Italia Domani, funded by the European Union via the program NextGenerationEU. V.C. is member of the Gruppo Nazionale per l’Analisi Matematica, la Probabilità e le loro Applicazioni (INdAM-GNAMPA). Most of this work was done while A.C. was visiting the Dipartimento di Matematica, Università di Roma–La Sapienza, thanks to the “Dipartimento di Eccellenza Project CUP B83C23001390001”.

References

CEREMADE, CNRS, Université Paris-Dauphine, Univ. PSL, France and Mokaplan, INRIA Paris. Email address, Antonin Chambolle: [email protected]

Dipartimento di Matematica Guido Castelnuovo, Piazzale Aldo Moro 5, 00185 Roma, Italy. Email address, Vito Crismale: [email protected]

References

  1. [1]S. Almi, E. Davoli, A. Kubin, and E. Tasso, On De Giorgi’s conjecture of nonlocal approximations for free-discontinuity problems: The symmetric gradient case, Preprint 2024, https://arxiv.org/abs/2410.23908.
  2. [2]S. Almi and E. Tasso, A general criterion for jump set slicing and applications, Preprint 2022, https://arxiv.org/abs/2212.09822.
  3. [3]S. Almi and E. Tasso, A new proof of compactness in G(S)BD, Advances in Calculus of Variations, (2022).arxiv.org/abs/2104.13141
  4. [4]L. Ambrosio, Existence theory for a new class of variational problems, Arch. Rational Mech. Anal., 111 (1990), pp. 291–322.DOI
  5. [5]L. Ambrosio, On the lower semicontinuity of quasiconvex integrals in SBV(Ω,Rᵏ), Nonlinear Anal., 23 (1994), pp. 405–425.DOI
  6. [6]L. Ambrosio, A. Coscia, and G. Dal Maso, Fine properties of functions with bounded deformation, Arch. Rational Mech. Anal., 139 (1997), pp. 201–238.DOI
  7. [7]L. Ambrosio, N. Fusco, and D. Pallara, Functions of bounded variation and free discontinuity problems, Oxford Mathematical Monographs, The Clarendon Press, Oxford University Press, New York, 2000.
  8. [8]J.-F. Babadjian, Traces of functions of bounded deformation, Indiana Univ. Math. J., 64 (2015), pp. 1271–1290.arxiv.org/abs/1308.5497
  9. [9]J.-F. Babadjian, F. Iurlano, and A. Lemenant, Partial regularity for the crack set minimizing the two-dimensional Griffith energy, J. Eur. Math. Soc. (JEMS), 24 (2022), pp. 2443–2492.
  10. [10]F. Cagnetti, A. Chambolle, and L. Scardia, Korn and Poincaré–Korn inequalities for functions with a small jump set, Math. Ann., 383 (2022), pp. 1179–1216.DOI
  11. [11]A. Chambolle, An approximation result for special functions with bounded deformation, J. Math. Pures Appl. (9), 83 (2004), pp. 929–954.DOI
  12. [12]———, Addendum to: “An approximation result for special functions with bounded deformation” [J. Math. Pures Appl. (9) 83 (2004), no. 7, 929–954; mr2074682], J. Math. Pures Appl. (9), 84 (2005), pp. 137–146.
  13. [13]A. Chambolle, S. Conti, and G. Francfort, Korn-Poincaré inequalities for functions with a small jump set, Indiana Univ. Math. J., 65 (2016), pp. 1373–1399.
  14. [14]A. Chambolle, S. Conti, and F. Iurlano, Approximation of functions with small jump sets and existence of strong minimizers of Griffith’s energy, J. Math. Pures Appl. (9), 128 (2019), pp. 119–139.arxiv.org/abs/1710.01929
  15. [15]A. Chambolle and V. Crismale, A density result in GSBDᵖ with applications to the approximation of brittle fracture energies, Arch. Ration. Mech. Anal., 232 (2019), pp. 1329–1378.DOI
  16. [16]———, Existence of strong solutions to the Dirichlet problem for the Griffith energy, Calc. Var. Partial Differential Equations, 58 (2019), pp. 27, No. 136.DOI
  17. [17]———, Compactness and lower semicontinuity in GSBD, J. Eur. Math. Soc. (JEMS), 23 (2021), pp. 701–719.arxiv.org/abs/1802.03302
  18. [18]———, Equilibrium configurations for nonhomogeneous linearly elastic materials with surface discontinuities, Ann. SNS Pisa Cl. Sci., 24 (2023), pp. 1575–1610.arxiv.org/abs/2006.00480
  19. [19]———, A general compactness theorem in G(S)BD, Indiana Univ. Math. J., Preprint 2022, in press (2025).arxiv.org/abs/2210.04355
  20. [20]S. Conti, M. Focardi, and F. Iurlano, Existence of strong minimizers for the Griffith static fracture model in dimension two, Ann. Inst. H. Poincaré C Anal. Non Linéaire, 36 (2019), pp. 455–474.arxiv.org/abs/1611.03374
  21. [21]G. Dal Maso, Generalised functions of bounded deformation, J. Eur. Math. Soc. (JEMS), 15 (2013), pp. 1943–1997.DOI
  22. [22]E. De Giorgi and L. Ambrosio, New functionals in the calculus of variations, Atti Accad. Naz. Lincei Rend. Cl. Sci. Fis. Mat. Nat. (8), 82 (1988), pp. 199–210 (1989).
  23. [23]G. De Philippis and F. Rindler, On the structure of A-free measures and applications, Ann. of Math. (2), 184 (2016), pp. 1017–1039.arxiv.org/abs/1601.06543
  24. [24]———, Fine properties of functions of bounded deformation, special issue on “Variational models in elasticity”, AIMS Math. Eng., 2 (2020).
  25. [25]G. A. Francfort and J.-J. Marigo, Revisiting brittle fracture as an energy minimization problem, J. Mech. Phys. Solids, 46 (1998), pp. 1319–1342.DOI
  26. [26]M. Friedrich, A piecewise Korn inequality in SBD and applications to embedding and density results, SIAM J. Math. Anal., 50 (2018), pp. 3842–3918.arxiv.org/abs/1604.08416
  27. [27]———, A compactness result in GSBD^p and applications to Γ-convergence for free discontinuity problems, Calc. Var. Partial Differential Equations, 58 (2019), pp. 31, No. 86.
  28. [28]M. Friedrich, C. Labourie, and K. Stinson, On regularity for griffith almost-minimizers in the plane, Preprint 2023, https://arxiv.org/abs/2310.07670.
  29. [29]M. Friedrich, M. Perugini, and F. Solombrino, Lower semicontinuity for functionals defined on piecewise rigid functions and on GSBD, J. Funct. Anal., 280 (2021), pp. 45, No. 108929.arxiv.org/abs/2002.08133
  30. [30]M. Friedrich and F. Solombrino, Quasistatic crack growth in 2d-linearized elasticity, Ann. Inst. H. Poincaré Anal. Non Linéaire, 35 (2018), pp. 27–64.arxiv.org/abs/1604.08338
  31. [31]F. Iurlano, A density result for GSBD and its application to the approximation of brittle fracture energies, Calc. Var. Partial Differential Equations, 51 (2014), pp. 315–342.DOI
  32. [32]R. Temam, Mathematical problems in plasticity, Gauthier-Villars, Paris, 1985. Translation of Problèmes mathématiques en plasticité. Gauthier-Villars, Paris, 1983.
  33. [33]R. Temam and G. Strang, Duality and relaxation in the variational problem of plasticity, J. Mécanique, 19 (1980), pp. 493–527.

Paper details

vlnc:W_cchx9neb
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 8

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents