On Scaling Properties for Two-State Problems and for a Singularly Perturbed T3\documentclass[12pt]{minimal}				\usepackage{amsmath}				\usepackage{wasysym}				\usepackage{amsfonts}				\usepackage{amssymb}				\usepackage{amsbsy}				\usepackage{mathrsfs}				\usepackage{upgreek}				\setlength{\oddsidemargin}{-69pt}				\begin{document}$T_{3}$\end{document} Structure

Bogdan Raiţă; Angkana Rüland; Camillo Tissot

doi:10.1007/s10440-023-00557-7

On Scaling Properties for Two-State Problems and for a Singularly Perturbed T3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$T_{3}$\end{document} Structure

Raiţă, Bogdan; Rüland, Angkana; Tissot, Camillo 2023-04-01 00:00:00 In this article we study quantitative rigidity properties for the compatible and incompatible two-state problems for suitable classes of A-free differential inclusions and for a singularly perturbed T structure for the divergence operator. In particular, in the compatible setting of the two-state problem we prove that all homogeneous, ﬁrst order, linear operators with afﬁne boundary data which enforce oscillations yield the typical -lower scaling bounds. As observed in Chan and Conti (Math. Models Methods Appl. Sci. 25(06):1091–1124, 2015) for higher order operators this may no longer be the case. Revisiting the example from Chan and Conti (Math. Models Methods Appl. Sci. 25(06):1091–1124, 2015), we show that this is reﬂected in the structure of the associated symbols and that this can be exploited for a new Fourier based proof of the lower scaling bound. Moreover, building on Rüland and Tribuzio (Arch. Ration. Mech. Anal. 243(1):401–431, 2022); Garroni and Nesi (Proc. R. Soc. Lond., Ser. A, Math. Phys. Eng. Sci. 460(2046):1789–1806, 2004, https://doi.org/10. 1098/rspa.2003.1249); Palombaro and Ponsiglione (Asymptot. Anal. 40(1):37–49, 2004), we discuss the scaling behavior of a T structure for the divergence operator. We prove that as in Rüland and Tribuzio (Arch. Ration. Mech. Anal. 243(1):401–431, 2022) this yields a non-algebraic scaling law. Keywords A-Free inclusions · Two-well problem · Divergence T · Phase transformation Mathematics Subject Classiﬁcation 35Q74 · 35F05 · 74N05 · 74G99 C. Tissot camillo.tissot@uni-heidelberg.de B. Raita ˘ bogdanraita@gmail.com A. Rüland rueland@uni-bonn.de Centro di Ricerca Matematica Ennio de Giorgi, Scuola Normale Superiore, Piazza dei Cavalieri, 3, 56126 Pisa, Italy Department of Mathematics, Universitatea Alexandru-Ioan Cuza, Bulevardul Carol I 11, 700506 Iasi, Romania Institut für Angewandte Mathematik, Ruprecht-Karls-Universität Heidelberg, Im Neuenheimer Feld 205, 69120 Heidelberg, Germany Institute for Applied Mathematics and Hausdorff Center for Mathematics, University of Bonn, Endenicher Allee 60, 53115 Bonn, Germany 5 Page 2of50 B.Raitae ˘ tal. 1Introduction Rigidity and ﬂexibility properties associated with (nonlinear) differential inclusions for the gradient have been objects of intensive study. They arise in a variety of applications, in- cluding the analysis of PDEs, e.g. the study of regularity of elliptic systems [5–7], ﬂuid dynamics [8, 9], geometry [10–13] and various settings in the materials sciences, e.g. the study of patterns in shape-memory alloys [14–19]. Motivated by applications of microstruc- tures in composites [20, 21], optimal design problems [22–24] and micromagnetics [25, 26] as well as by recent developments on more general differential inclusion problems [27–29], in this article we study two instances of quantitative rigidity and ﬂexibility properties of differential inclusions for more general operators. On the one hand, we consider constant- coefﬁcient, homogeneous, linear differential operators for which we discuss quantitative versions of the compatible and incompatible two-state problems. On the other hand, we investigate quantitative properties of a T structure for the divergence operator. 1.1 On Quantitative Results for the Two-Well Problem for A-Free Diﬀerential Inclusions A-free differential inclusions arise in many different settings, including linearized elasticity [30, 31], liquid crystal elastomers [32, 33] and the study of the Aviles-Giga functional [34, 35] to name just a few examples. They have been systematically investigated in the context of compensated compactness theory in classical works such as [36–39]but also in more recent literature on compensated compactness theory [40–46], in truncation results [47], in classical minimization and regularity questions in the calculus of variations [48–51]and in the context of ﬁne properties of such operators in borderline spaces [27, 52–54]. In the recent articles [27, 28] general A-free versions of the incompatible two-well problem in borderline spaces and the study of T structures have been initiated. Motivated by these applications, in the ﬁrst part of this article, we seek to study quantitative versions of the compatible and incompatible two-state problems. Before turning to the setting of general A-free differential inclusions, let us recall the analogous “classical” setting for the gradient: Inspired by problems from materials science and phase transformations, the exact and approximate rigidity properties of differential in- clusion problems for the gradient [14](seealso[18, 55–57]) with and without gauge invari- ance have been considered. For two energy wells without gauge invariances this amounts to the study of the differential inclusion ∇v ∈{A, B} in (1) d×d d for A, B ∈ R , A = B and ⊂ R a bounded Lipschitz domain. It is well-known that depending on the compatibility of the wells, a dichotomy arises: d×d • On the one hand, for incompatible wells, i.e. if B − A ∈ R is not a rank-one matrix, the differential inclusion (1)is rigid both for exact and approximate solutions: Indeed, if A, B are incompatible,(1) only permits solutions with constant deformation gradient, which is in the following referred to as the rigidity of the two-state problem for exact solutions. Moreover, in this setting one has that for sequences ∇u with dist(∇u , {A, B}) → 0in k k measure, it necessarily holds that along a subsequence ∇u → A or ∇u → B in measure. k k We will refer to this as rigidity of the two-state problem for approximate solutions. We remark that various far-reaching generalizations of these results have been ob- tained: For settings with SO(d) symmetries a quantitative version of such a result was Scaling for the Two-State Problem and a T Structure Page 3 of 50 5 deduced in [58]; a further proof was found in [59]. The one-state problem with continu- ous symmetry group was studied in [60, 61]. The article [27] investigates a similar problem for two incompatible wells for a general constant coefﬁcient, homogeneous, linear differential operator A(D), providing qualita- tive rigidity results for the associated exact and approximate differential inclusions, in- cluding L -based frameworks. d×d • On the other hand, if the wells are compatible, i.e. if B − A ∈ R is a rank-one matrix, then simple laminate solutions of (1) exist, in which the deformation gradient oscillates between the two ﬁxed values A, B and is a one-dimensional function depending only on the direction determined by the difference B − A. Due to the failure of rigidity on the exact level, also rigidity on the approximate level cannot be expected without additional regularization terms. In the ﬁrst part of this article we seek to consider quantitative, L -based variants of these type of results for more general, linear differential operators A(D). In this context, we will consider the following two guiding questions: • Quantitative incompatible rigidity. For a constant coefﬁcient, homogeneous, linear dif- ferential operator A(D) and two incompatible wells, i.e. A, B ∈ R such that B − A/ ∈ ,cf. (5) for the deﬁnition of the wave cone, do we have a quantitative rigidity result in terms of domain scaling for prescribed boundary data which are a convex combination of the two states? Here the dimension n depends on the operator A(D);for A(D) = curl (which corresponds to the gradient setting from (1) above) we would for instance consider n = d × d . More precisely, we seek to study the following question: Let A, B ∈ R be such that B − A/ ∈ .Isittruethat E (u, χ ) = |u − χ A − χ B| dx ≥ C(A, B, λ)||, el A B d n d if u : R → R , A(D)u = 0in R , χ := Aχ + Bχ ∈{A, B} in and χ ,χ ∈{0, 1} A B A B with χ + χ = 1 and if for some λ ∈ (0, 1) we have that u = F := λA + (1 − λ)B in A B λ R \ ? • Quantitative compatible rigidity. Letusnextconsider two compatible wells A, B ∈ R , i.e. let A, B ∈ R be such that B − A ∈ ,see (5) below, and let us again consider boundary data F := λA + (1 − λ)B for some λ ∈ (0, 1) as above and with the set of admissible functions given by D in (10). For a singularly perturbed energy similarly as in (15)isittruethatasin[62, 63] also in the setting of a more general constant coefﬁcient, homogeneous, linear differential operator A(D) the following bound holds 2/3 inf inf (E (u, χ ) + |∇χ |) ≥ C ? el χ ∈BV (;{A,B}) u∈D In what follows, we will formulate the set-up, the relevant operator classes and our results on these questions. 1.2 Formulation of the Two-State Problem for A-Free Operators in Bounded Domains Following [27, 28, 48], we consider a particular class of linear, homogeneous, constant- ∞ d n ∞ d m coefﬁcient operators. The operator A(D) : C (R ; R ) → C (R ; R ) of order k ∈ N is 5 Page 4of50 B.Raitae ˘ tal. giveninthe form A(D) := A ∂ , (2) |α|=k d m×n where α ∈ N denotes a multi-index of length |α|:= α and A ∈ R are constant j α j =1 matrices. Seeking to study microstructure, in the sequel we are particularly interested in non-elliptic operators. Here the operator A(D) is said to be elliptic if its symbol A(ξ ) := A ξ (3) |α|=k is injective for all ξ = 0. If A(D) is not elliptic, there exist vectors ξ ∈ R \{0} and μ ∈ R \{0} such that A(ξ )μ = 0. (4) The collection of these vectors μ ∈ R \{0} forms the wave cone associated with the operator A(D): := ker(A(ξ )). (5) d−1 ξ ∈S The relevance of the wave cone for compensated compactness and the existence of mi- crostructure is well-known. For instance, for any pair (μ, ξ ) as in (4) it is possible to obtain A-free simple laminate solutions. These are one-dimensional functions u(x) := μh(x · ξ), where h : R →{0, 1}, which obey the differential constraint A(D)u = 0 due to the choice of μ, ξ and u ∈{0,μ}. More generally, if k = 1, and for μ ∈ (see, for instance, [64]) it holds that u(x) := μh(x · ξ ,...,x · ξ ) is a solution to the differential equation A(D)u = 0 for vectors ξ ,...,ξ ∈ R \{0} forming a basis of the vectorspace V := ξ ∈ R : A(ξ )μ = 0 . (6) A,μ For the row-wise curl operator (n = d × d ) this is an at most one-dimensional space, while for the row-wise divergence operator (n = m × d ), it is a space of possibly higher dimension as V = ker μ, leading to substantially more ﬂexible solutions of the associated differen- div,μ tial inclusions than for the curl. In order to study microstructures arising as solutions to the two-state problem, for the above speciﬁed class of operators, analogously as in the gradient setting, we consider the following A-free differential inclusion with prescribed boundary values: u ∈ K in , (7) A(D)u = 0in R , n d with K ⊂ R ,and ⊂ R an open, bounded, simply connected domain, with appropriately prescribed boundary data. For the two-state problem we consider K ={A, B}⊂ R . Scaling for the Two-State Problem and a T Structure Page 5 of 50 5 Now, in analogy to the gradient setting, on the one hand, we call the differential inclusion for the two-state problem (7) incompatible if it is elliptic in the sense that B − A/ ∈ .In this case it is proved in [27] that both the exact and approximate differential inclusion (7) are rigid. We emphasize that incompatibility in particular excludes the presence of simple laminates. On the other hand, the differential inclusion (7)issaidtobe compatible if B − A ∈ . In this case, also in the setting of more general operators, a particular class of solutions to (7) consists of (generalized) simple laminates. Moreover, in the compatible setting, we further distinguish a particular case: We consider the subspace I := ker(A(ξ )) = ker(A ). (8) A α |α|=k ξ ∈R This is the space of values that are (algebraically) unconstrained by A(D), meaning that for 2 d all u ∈ L (R ; I ),wehavethat A(D)u = 0 without taking any regularity constraints on u. In the case of I ={0}, the operator A(D) belongs to the class of cocanceling operators, introduced in [65]. We seek to study both settings and the resulting microstructures quantitatively in the spirit of scaling results as, for instance, in the following non-exhaustive list involving differ- ent physical applications [2, 22, 62, 63, 66–77]. To this end, for ⊂ R an open, bounded, Lipschitz set, we introduce elastic and surface energies and consider their minimization for prescribed, not globally compatible boundary data F = λA + (1 − λ)B for some λ ∈ (0, 1), where again the set of states is given by K ={A, B}. Motivated by the applications from materials science, we study the following “elastic energy” E (u, χ ) := |u − χ | dx, (9) el which we minimize in the following admissible class of deformations 2 d n d d 2 u ∈ D := u ∈ L (R ; R ) : A(D)u = 0in R ,u = F in R \ ,χ ∈ L (; K). F λ λ loc (10) For ease of notation, here and in what follows, we often use the convention that χ := χ A + χ B with χ ,χ ∈ L (;{0, 1}) and χ + χ = 1in . Moreover, we further use the B A B A B notation E (χ ; F ) := inf E (u, χ ). el λ el u∈D In addition to the “elastic” energy contributions, we also introduce a surface energy con- tribution of the form E (χ ) := |∇χ |,χ ∈ BV (; K) (11) surf and consider the following singularly perturbed elastic energy for > 0 E (u, χ ) := E (u, χ ) + E (χ ), (12) el surf and correspondingly E (χ ; F ) := E (χ ; F ) + E (χ ) = inf E (u, χ ). λ el λ surf u∈D λ 5 Page 6of50 B.Raitae ˘ tal. We note that this can be deﬁned for an arbitrary set of states K and suitable boundary data F . With these quantities in hand, we can formulate the following quantitative rigidity results for the two-state problems: Theorem 1 Let d, n, m, k ∈ N, n> 1. Let ⊂ R be a bounded Lipschitz domain, and let A(D) be as in (2) with the wave cone given in (5) and let I be given in (8). Let A A n 2 A, B ∈ R and let χ = χ A + χ B ∈ L (;{A, B}). Further, let the elastic and surface A B energies E , E be given as in (9) and (11), respectively. For λ ∈ (0, 1) set F = λA + el surf λ (1 − λ)B ∈ R and consider D as in (10). The following results hold: (i) Incompatible case: Assume that B −A/ ∈ . Then there is a constant C = C(A, B) > 0 such that, inf inf E (u, χ ) ≥ C(min{λ, 1 − λ}) ||. el 2 u∈D χ ∈L (;{A,B}) F (ii) Compatible case: Assume that A(D) is one-homogeneous, i.e. k = 1 in (2), and that B − A ∈ \ I . Then, there exist C = C(A(D), A, B, , d, λ) > 0 and = A A 0 (A(D), A, B, , d, λ) > 0 such that for ∈ (0, ) 0 0 2/3 inf inf (E (u, χ ) + E (χ )) ≥ C . el surf χ ∈BV (;{A,B}) u∈D Furthermore, if we assume A(D) = div and =[0, 1] , then there exists a constant c = c(A, B, λ) > 0 such that for > 0 we also have the matching upper bound 2/3 inf inf (E (u, χ ) + E (χ )) ≤ c . el surf u∈D χ ∈BV (;{A,B}) (iii) Super-compatible case: Assume that A − B ∈ I . Then, inf inf (E (u, χ ) + E (χ )) = 0. el surf χ ∈BV (;{A,B}) u∈D We highlight that in our discussion of the compatible case, we have restricted ourselves to operators of order one. This is due to the fact that for higher order operators it is expected that more complicated microstructures may arise. This is also reﬂected in the Fourier space properties of the symbol A. We refer to Sect. 3.5, see Proposition 3.10, for a brief discussion of this, illustrating that the scaling may, in general, be no longer of the order in the higher order setting. Let us discuss a prototypical example of the above results: Example 1.1 As an example of the above differential inclusion, we consider the case in ∞ d m×d ∞ d m d which A(D) = div : C (R ; R ) → C (R ; R ) row-wise for matrix ﬁelds u : R → m×d R . In this case the boundary value problem under consideration turns into the following differential inclusion: u ∈{A, B} in , div u = 0in R , u = F in R \ , λ Scaling for the Two-State Problem and a T Structure Page 7 of 50 5 for some F := λA + (1 − λ)B , λ ∈ (0, 1). Such differential inclusions are related to appli- cations in shape-optimization as, for instance, in [22]. d m×d The divergence is applied row-wise to matrix ﬁelds u : R → R , i.e. A(D)u = m×d (∂ u)e with the matrix vector product. Hence A(ξ )M = Mξ for M ∈ R and thus j j j =1 the wave cone is given by m×d d ={M ∈ R : there is ξ ∈ R with Mξ = 0}. div Moreover the divergence is a cocanceling operator as I ={0}. div We emphasize that Example 1.1 is indeed a prototypical example and plays a central role in the study of ﬁrst order operators in that all ﬁrst order operators can be reduced to this model operator by a suitable linear transformation, see [28, Appendix] and also Sect. B below. We emphasize that this reduction is particularly useful if the differential inclusion is incompatible or if the boundary data are in \ I . As a consequence, quantitative lower A A bound estimates for incompatible differential inclusions for ﬁrst order operators, e.g. for T structures as qualitatively studied in [28], or for compatible, but not super-compatible boundary data can be deduced from the ones of the divergence operator (see Proposition B.2 and, in general, the discussion in Sect. B). A reduction to an equivalent problem for a mod- iﬁed operator and modiﬁed boundary data to the setting involving a cocanceling operator will be discussed in Sect. 3.4, see Proposition 3.8 and Corollary 3.9. 1.3 Quantitative Rigidity of a T Structure for the Divergence Operator In the second part of the article, building on the works [2–4] and motivated by the high- lighted considerations on the role of the divergence operator, we study the quantitative rigid- ity of the T conﬁguration u ∈{A ,A ,A } a.e. in , div u = 0in R , (13) 1 2 3 3 3 3×3 where =[0, 1] , u : R → R and ⎛ ⎞ − 00 ⎝ ⎠ A = 0 ,A = 0 0 ,A = Id . (14) 1 3×3 2 3 3×3 00 3 By virtue of the results from [3, 4] this problem is ﬂexible for approximate solutions but rigid on the level of exact solutions: • More precisely, on the level of exact solutions to (13), only constant solutions u ≡ A for j ∈{1, 2, 3} obey the differential inclusion. • Considering however approximate solutions, i.e. sequences (u ) such that k k∈N dist(u , {A ,A ,A }) → 0 in measure as k →∞, div u = 0for all k ∈ N, k 1 2 3 k there exists a sequence of approximate solutions (u ) for (13) such that there is no sub- k k∈N sequence which converges in measure to one of the constant deformations {A ,A ,A }. 1 2 3 Compared to the setting of the gradient, for the divergence operator rigidity for approximate solutions is already lost for the three-state problem (while this arises only for four or more states for the gradient [78, 79], see also [18] for further instances in which the Tartar square was found and used). 5 Page 8of50 B.Raitae ˘ tal. As in [2] we here study a quantitative version of the dichotomy between rigidity and ﬂexibility: We consider the singularly perturbed variant of (13) as in Sect. 1.2 ˆ ˆ E (u, χ ) := |u − χ | dx + |∇χ | (15) qc under the constraint u ∈ D ,cf. (10), with A(D) = div and where F ∈{A ,A ,A } (see F 1 2 3 Sect. 2 for the deﬁnition of the A-quasi-convexiﬁcation of a compact set). Here the function χ ∈ BV (;{A ,A ,A }) denotes the phase indicator of the “phases” A , A , A , respec- 1 2 3 1 2 3 tively. Adapting the ideas from [2] to the divergence operator in three dimensions, we prove the following scaling result: Theorem 2 Let =[0, 1] and let K ={A ,A ,A } with A , j ∈{1, 2, 3} given in (14), 1 2 3 j qc F ∈ K \ K, E be as in (15) above for the divergence operator A(D) = div, and consider D given analogously as (10). Then, there exist constants c = c(F ) > 0 and C = C(F ) > 1 such that for any ν ∈ (0, ) there is = (ν, F ) > 0 and c > 0 such that for ∈ (0, ) 0 0 ν 0 we have 1 1 −1 +ν 2 2 C exp(−c | log()| ) ≤ inf inf E (u, χ ) ≤ C exp(−c| log()| ). χ ∈BV (;K) u∈D Let us comment on this result: As in [2] we obtain essentially matching upper and lower scaling bounds with less than algebraic decay behavior as → 0, reﬂecting the inﬁnite order laminates underlying the T structure and the fact that the problem is “nearly” rigid. As in [2] a key step is the analysis of a “quantitative chain rule in a negative Sobolev space” which results from the interaction of Riesz type transforms and a nonlinearity originating from the “ellipticity” of the differential inclusion. Both in the upper and the lower bound, these estimates for the divergence operator however require additional care due to the three- dimensionality of the problem. In the upper bound construction this is manifested in the use of careful cut-off arguments; in the lower bound, a more involved iterative scheme has to be used to reduce the possible regions of concentration in Fourier space. Similarly, as in [76] the scaling law from Theorem 2 is obtained as a consequence of a rigidity estimate encoding both the rigidity and ﬂexibility of the T differential inclusion (in analogy to the T case from [76, Proposition 3]). Proposition 1.2 Let =[0, 1] and let K ={A ,A ,A } with A , j ∈{1, 2, 3} given in 1 2 3 j qc (14), F ∈ K \ K. Let χ denote the diagonal entries of the matrix ﬁeld χ ∈ BV (; K) and jj per denote by E (χ ; F) the periodic singularly perturbed energy (see (32) and Sect. 4.3.2). Then, there exists > 0 such that for any ν ∈ (0, 1) there is a constant c > 0 such that for 0 ν ∈ (0, ) we have 1 1 2 +ν per 2 2 χ − χ ≤ exp(c | log()| )E (χ ; F) . jj jj 2 3 ν L ([0,1] ) j =1 We emphasize that in parallel to the setting of the Tartar square, this estimate quanti- tatively encodes both rigidity and ﬂexibility of the differential inclusion, as it measures the distance to the constant state (and thus reﬂects rigidity of the exact differential inclusion) but also quantiﬁes the “price” for this in terms of a “high energy” scaling law (and thus reﬂects the underlying ﬂexibility of the approximate problem). Moreover, due to the ﬂexibility of the differential inclusion, we stress that such an estimate can only be inferred for a combination of elastic and surface energies. Scaling for the Two-State Problem and a T Structure Page 9 of 50 5 1.4 Outline of the Article The remainder of the article is structured as follows: After brieﬂy recalling relevant notation and facts on convex hulls related to the operator A(D), we ﬁrst discuss the compatible and incompatible two-state problems in Sect. 3. Here we begin by discussing the incompatible setting, which is a consequence of direct elliptic estimates in Sect. 3.2 andthenturnto the lower bounds for the compatible setting in Sect. 3.3. The super-compatible case is then treated in Sect. 3.4. It is also in this section that we discuss a reduction to cocanceling operators. We revisit the scaling of a prototypical higher order operator from [1] in Sect. 3.5 and explain how our scheme of deducing lower bounds also yields a quick Fourier based proof of the lower scaling bound from [1]. In Sect. 4 we then turn to the T differential inclusion, for which we ﬁrst prove upper bounds in Sect. 4.1 and then adapt the ideas from [2] to infer essentially matching lower bounds in Sect. 4.2. In the Appendix, we complement the general lower bounds for ﬁrst order differential op- erators with upper bounds for the speciﬁc case of the divergence operator (Sect. A). More- over, in Sect. B.1 we discuss the reduction to the divergence operator. 2Notation In this section we collect the notation which is used throughout the article. • For a set U,wedenoteby d the (possibly smoothed-out) distance to this set: d (x) = U U dist(x, U ) = inf |x − y| and denote by χ the (in some places smoothed-out) indicator y∈U U function of this set. • For a set U with ﬁnite measure and a function f : U → R we denote the mean by f = fdx . |U | U 2 d 2 d • For a function f ∈ L (T ) or f ∈ L (R ) we denote the Fourier transform by − −ik·x d F (f )(k) = f(k) = (2π) e f(x)dx (k ∈ Z ) or − −iξ ·x d F (f )(ξ ) = f(ξ) = (2π) e f(x)dx (ξ ∈ R ). ∞ d For a function h ∈ L (R ), we denote by h(D) the corresponding Fourier multiplier −1 h(D)f = F (h(·)f(·)). • We usually denote the phase indicator by χ ∈ L (; K), with the component functions qc qc given by χ = (χ ), i ∈{1,...,n}. Moreover, we use F ∈ K (where K is introduced below) as the exterior data. • The set of admissible ’deformations’ u for an operator of order k ≥ 1is given by A 2 d n d d D := D := {u ∈ L (R ; R ) : A(D)u = 0in R ,u = F in R \ }. Here the equa- F loc tion A(D)u = 0 is considered in a distributional sense. qc • As introduced in Sect. 1.2,for F ∈ K we consider the elastic energy with u ∈ D , χ ∈ L (; K): E (u, χ ) = |u − χ | dx, E (χ ; F) = inf E (u, χ ), (16) el el el u∈D 5 Page 10 of 50 B. Raitae ˘ tal. and, for χ ∈ BV (; K), the surface energy as the total variation norm: E (χ ) =∇χ = |∇χ |. (17) surf TV() • The total energy with > 0is given by E (u, χ ) = E (u, χ ) + E (χ ). el surf • We write f ∼ g if there are constants c, C > 0 such that cf ≤ g ≤ Cf . −1 −1 3 • We use the notation · for the homogeneous H semi-norm for f ∈ H (T ; R): ˙ −1 2 2 f = |f(k)| . −1 |k| k∈Z \{0} We further recall the notions of the -convex hull of a set (see [64]) and the A-quasi- convex hull of a set: n n • Let ⊂ R be the wave cone from (5)and let K ⊂ R be a compact set. For j ∈ N,we (j ) then deﬁne K as follows: (0) K := K, (j ) n (j −1) K := {M ∈ R : M = λA + (1 − λ)B : A, B ∈ K ,B − A ∈ ,λ ∈[0, 1]}. lc Moreover, we deﬁne the -convex hull K : lc (j ) K := K . j =0 We deﬁne the order of lamination of a matrix M ∈ R to be the minimal j ∈ N such that (j ) lc M ∈ K . In analogy to the gradient case, we will also refer to K as the laminar convex hull. qc n • We recall that the A-quasi-convex hull K of a compact set K ⊂ R is deﬁned by duality to A-quasi-convex functions [48]. We recall that the div-quasi-convex hull of the T ma- trices from (13), (14) has been explicitly characterized in [21, Theorem 2] to consist of the union of the closed triangle formed by the matrices S ,S ,S and the three “legs” formed 1 2 3 3×3 by the line segments A S . The matrices S ,S ,S ∈ R are introduced in Sect. 4 be- j j 1 2 3 3×3 low. Given the set {A ,A ,A }⊂ R from (13), (14), as in the gradient case, we denote 1 2 3 qc its A-quasi-convex hull by {A ,A ,A } . 1 2 3 3 Quantitative Results on the Two-State Problem In this section, we study quantitative versions of the two-state problem for general, linear, constant coefﬁcient, homogeneous operators, always considering the divergence operator as a particular model case. Building on the precise formulation of the problem from (7), in Sect. 3.1,weﬁrstcharac- terize the elastic energies in terms of the operator A(D). With this characterization in hand, using ellipticity, we next prove the quantitative L bounds in the incompatible two-well case (Sect. 3.2)and lower -scaling bounds for the compatible case with ﬁrst order, linear oper- ators (Sect. 3.3). In Sect. 3.4, we prove Theorem 1(iii) and deduce a reduction to cocanceling operators. In Sect. 3.5 we brieﬂy discuss the role of degeneracies in the symbol of the elastic Scaling for the Two-State Problem and a T Structure Page 11 of 50 5 energies which may arise for higher order, linear differential operators and which may thus lead to an alternative scaling behavior different from the -scaling bound. Finally, in Sect. A we complement the lower bounds from this section with matching up- per bounds for the special case of A(D) = div. Similar constructions are also known for the gradient, the symmetrized gradient and lower dimensional problems from micromagnetics [1, 66, 73, 74, 80]. 3.1 Elastic Energy Characterization We begin by recalling an explicit lower bound for the elastic energy in terms of the operator A(D) (see, for instance also [49, discussion before Lemma 1.17]) in an, for us, convenient form. In everything that follows, we will assume d> 1. Lemma 3.1 (Fourier characterization of the elastic energy) Let d, n, k ∈ N. Let ⊂ R be a bounded Lipschitz domain with associated indicator function χ and let K ⊂ R . Let A(D) = A ∂ be as in (2) with the symbol A, cf.(3), and E be as in (16) and F ∈ α el |α|=k qc 2 K . Then there is a constant c = c(A(D)) > 0 such that for any χ ∈ L (; K), extended by zero outside of , E (χ ; F) ≥ inf (u − F) − (χ − Fχ ) dx el A(D)u=0 d ∗ ∗ −1 = A(ξ ) A(ξ )A(ξ ) ) A(ξ )(χ ˆ − F χ ˆ ) dξ ≥ c A( )(χ ˆ − F χ ˆ ) dξ, |ξ | 2 d n d where the inﬁmum is taken over all u ∈ L (R ; R ) that fulﬁll A(D)u = 0 in R . loc Proof The proof follows from a projection argument in Fourier space. Step 1: Whole space extension, Fourier and pseudoinverse of the differential operator. We begin by transforming our problem to one on the whole space R , introducing the whole space extension w := u − F : ˆ ˆ 2 2 E (u, χ ) = u − χ dx = u − F − (χ − F) dx el = w − (χ − Fχ ) dx, where we have extended all functions in the integrand by zero outside of . In the following, we write χ ˜ := χ − Fχ . Fourier transforming the expression for the elastic energy then leads to E (u, χ ) = w ˆ − χ ˜ dξ. el Further, seeking to deduce a lower bound, we relax the boundary data for w, obtaining ˆ ˆ ˆ 2 2 2 E (χ ; F) = inf |u − χ | dx = inf |w −˜ χ | dx ≥ inf |ˆ w − χ ˜ | dξ. el u∈D w∈D d A(D)w=0 d F 0 R R 5 Page 12 of 50 B. Raitae ˘ tal. Minimizing the integrand w ˆ (and still denoting the minimizer by w ˆ ) for each ﬁxed mode ξ ∈ R \{0}, we infer that ˆ ˆ ˆ ˆ w( ˆ ξ ) − χ( ˜ ξ ) = χ( ˜ ξ ) − χ( ˜ ξ ) = ∗χ( ˜ ξ ) ker A(ξ ) ran A(ξ ) ∗ ∗ −1 = A(ξ ) A(ξ )A(ξ ) ) A(ξ )χ( ˜ ξ ) , whereweview A(ξ )A(ξ ) : ran A(ξ ) → ran A(ξ ) as an isomorphism. This directly implies the claimed lower bound for E (χ ; F) in terms of the symbol A and its pseudoinverse. el Step 2: Proof of the ﬁnal estimate. Now to show the ﬁnal estimate for the elastic energy, ∗ ∗ −1 d n we seek to bound |A(ξ ) (A(ξ )A(ξ ) ) A(ξ )x| for every ξ ∈ R , x ∈ R from below in terms of |A(ξ )x|. As the projection operator is zero-homogeneous, we can reduce to ξ ∈ d−1 d−1 d−1 S , and thus can use the continuity of S ξ → A(ξ ) and the compactness of S for n d−1 the desired bound: For any x ∈ R , ξ ∈ S it hence holds ∗ ∗ −1 A(ξ )x = A(ξ )A(ξ ) (A(ξ )A(ξ ) ) A(ξ )x ∗ ∗ −1 ≤ A(ξ ) (A(ξ )A(ξ ) ) A(ξ )x sup (|A(ξ )|). d−1 ξ ∈S As A(D) = 0, we have that 0 < sup d−1 |A(ζ )|≤ C< ∞. Dividing by this and plugging ζ ∈S this into the expression with the pseudoinverse, we obtain ∗ ∗ −1 E (u, χ ) ≥ A(ξ ) A(ξ )A(ξ ) ) A(ξ )χ ˜ dξ el 1 ξ ≥ A( )χ ˜ dξ, sup |A(ζ )| d |ξ | d−1 ζ ∈S which concludes the argument. We emphasize that we are relaxing the boundary conditions for w( ˆ ξ ) := χ( ˜ ξ ) ker A(ξ ) as we do not calculate the projection of χ onto D , hence the above Fourier bounds only provide lower bounds for the elastic energy. We apply the lower bound from Lemma 3.1 to the two-well problem: Corollary 3.2 Let d, n ∈ N. Let , A(D), A, E be as in Lemma 3.1. Consider K = el {A, B}⊂ R with F = λA + (1 − λ)B for λ ∈ (0, 1). Then there exists a constant 2 d C = C(A(D)) > 0 such that for any χ = χ A + χ B ∈ L (; K), extended to R by zero, A B it holds E (χ ; F ) ≥ C ((1 − λ)χ ˆ − λχ ˆ )A( )(A − B) dξ. el λ A B d |ξ | Proof Using the expression of F in terms of A, B , λ yields A − F = (1 − λ)(A − B) and λ λ B − F =−λ(A − B). Thus, the fact that χ = χ A + χ B and Lemma 3.1 imply λ A B E (χ ; F ) ≥ C A( )(χ ˆ − F χ ˆ ) dξ el λ λ d |ξ | = C ((1 − λ)χ ˆ − λχ ˆ )A( )(A − B) dξ. A B d |ξ | R Scaling for the Two-State Problem and a T Structure Page 13 of 50 5 Remark 3.3 (The divergence operator) As seen in Example 1.1,inthe caseof A(D) = div, ∞ d d×d we have for u ∈ C (R ; R ) (note that we chose square matrices out of simplicity) ∞ d d A(D)u = (∂ u)e ∈ C (R ; R ). i i i=1 With this we can calculate A(ξ )M = ξ Me = Mξ. i i i=1 ∗ d d×d This, in particular, shows that the adjoint operator is given by A(ξ ) : R → R with A(ξ ) x = x ⊗ ξ. ∗ 2 Therefore A(ξ )A(ξ ) x = (x ⊗ ξ)ξ =|ξ | x , and the projection in the lower bound for the elastic energy of Lemma 3.1 takes the desired form −1 ∗ ∗ A(ξ ) A(ξ )A(ξ ) A(ξ )M = (Mξ ⊗ ξ). |ξ | Furthermore it holds −1 ∗ ∗ A(ξ ) A(ξ )A(ξ ) A(ξ )M = A( )M . |ξ | 3.2 The Incompatible Two-Well Problem and Scaling As a ﬁrst application of the Fourier characterizations from the previous section, we prove a quantitative lower bound for the incompatible two-well problem. We emphasize that – as in [27] – this argument is an elliptic argument and thus can be applied to all linear, constant coefﬁcient homogeneous operators. Indeed, the following result holds: Proposition 3.4 Let d, n ∈ N. Let ⊂ R be a bounded Lipschitz domain, let A(D) be given in (2) and A, B ∈ R with B − A/ ∈ , cf.(5), further let F = λA + (1 − λ)B A λ for some λ ∈ (0, 1), and let E be as in (16) with D given in (10). Then there is C = el F C(A, B, A(D)) > 0, such that for any χ ∈ L (;{A, B}) inf E (u, χ ) ≥ C min{λ, 1 − λ} ||. el u∈D Proof By virtue of Corollary 3.2, we have the lower bound E (χ ; F ) ≥ C ((1 − λ)χ ˆ − λχ ˆ )A( )(A − B) dξ, el λ A B |ξ | where the constant only depends on the operator A(D). d d As (A − B) ∈ / ker A(ξ ) for all ξ ∈ R and thus |A(ξ )(A − B)| > 0for any ξ ∈ R , d−1 ξ by continuity of ξ → A(ξ ) and compactness of S , this implies |A( )(A − B)|≥ |ξ | C(A, B, A(D)) > 0. Hence, ˆ ˆ 2 2 2 2 E (u, χ ) ≥ C |(1 − λ)χ ˆ − λχ ˆ | dξ = C |(1 − λ)χ − λχ | dx el A B A B d d R R 5 Page 14 of 50 B. Raitae ˘ tal. ˆ ˆ 2 2 2 2 2 ≥ C ( (1 − λ) dx + λ dx) ≥ C min{λ, 1 − λ} ||, A B where := supp(χ ) ⊂ and := supp(χ ) ⊂ . A A B B Remark 3.5 We emphasize that this result can be viewed as an incompatible nucleation bound. 3.3 The Compatible Two-Well Case and Scaling We next turn to the setting of two compatible wells and restrict our attention to ﬁrst order operators. In this case, we claim the following -lower scaling bound. This is in analogy to the situation for the gradient which had ﬁrst been derived in the seminal works [62, 63]. Proposition 3.6 Let d, n ∈ N. Let ⊂ R be a bounded Lipschitz domain and let A(D) be a ﬁrst order operator as in (2). Let A, B ∈ R be such that B − A ∈ \ I where A A and I are given in (5) and (8), and deﬁne F := λA + (1 − λ)B for some λ ∈ (0, 1). A A λ Let E := E + E be given in (12) with D deﬁned in (10). Then, there exist C = el surf F C(A(D), A, B, , d, λ) > 0 and = (A(D), A, B, , d, λ) > 0 such that for any ∈ 0 0 (0, ) we have 2/3 ≤ C inf inf E (u, χ ). u∈D χ ∈BV (;{A,B}) In order to deal with the compatible case, we invoke the following (slightly generalized) auxiliary results from [2], see also [23, 71], which we formulate for a general Fourier mul- tiplier m: Lemma 3.7 (Elastic, surface and low frequency cut-off) Let d, n, N ∈ N. Let ⊂ R d N be a bounded Lipschitz domain. Let m : R → R be a linear map and denote by d d V := ker m R its kernel and by : R → V the orthogonal projection onto V . Let f ∈ BV (R ;{−λ, 0, 1 − λ}) for λ ∈ (0, 1) with f = 0 outside and f ∈{−λ, 1 − λ} in . Consider the elastic and surface energies given by ˆ ˆ ˜ ˆ ˜ E (f ) := |m( )f(ξ)| dξ, E (f ) := |∇f |. el surf d |ξ | Then the following results hold: (a) Low frequency elastic energy control. Let μ> 1, then there exists C = C(m, ) > 0 with 2 2 ˆ ˜ f ≤ Cμ E (f ). 2 d el L ({ξ ∈R :| (ξ )|≤μ}) (b) High frequency surface energy control. There exists C = C(d, λ) > 0 such that for μ> 0 it holds 2 −1 ˆ ˜ f ≤ Cμ (E (f ) + Per()). 2 d surf L ({ξ ∈R :|ξ |≥μ}) Proof of Lemma 3.7 Since the property (b) is directly analogous to the one from [2, Lemma 2], we only discuss the proof of (a) which requires some (slight) modiﬁcations with respect Scaling for the Two-State Problem and a T Structure Page 15 of 50 5 to [2](and[23]). We thus present the argument for this for completeness. We split ξ = ξ + ξ ,where ξ ∈ V = {0}, ξ ∈ V = ker m. With this in hand and by the linearity of m it holds ξ ξ + ξ ξ ξ m( ) = m( ) = m( ) = M |ξ | |ξ | |ξ | |ξ | for some matrix representation M of m. Hence, using ξ ⊥ ker m,there is c = c(m) > 0such that ˆ ˆ ξ ξ 2 2 ˜ ˆ ˆ E (f ) = |m( )f(ξ)| dξ ≥ c | f(ξ)| dξ. el d |ξ | d |ξ | R R With this in hand, we argue similarly as in [23]and [2]: For a, b ∈ R, a, b ≥ 0and the orthogonal splitting ξ = ξ + ξ from above, it holds that ˆ ˆ 1 ξ |ξ | ˜ ˆ ˆ E (f ) ≥ f dξ = |f | dξ el 2 2 c |ξ | |ξ | +|ξ | d d R R ≥ |f | dξ + 1 1 1 {|ξ |≥ , |ξ |≤ } a b ⎛ ⎞ (18) ˆ ˆ ˆ ⎜ ⎟ 2 2 ˆ ˆ = |f | dξ − |f | dξ dξ ⎝ ⎠ + 1 1 ⊥ 1 {|ξ |≤ } V {|ξ |< } b a ⎛ ⎞ ˆ ˆ dim V 1 2 2 2 ⎝ ˆ ˆ ⎠ ≥ |f | dξ − sup |f | dξ . + 1 ξ ∈V 1 ⊥ {|ξ |≤ } Using the notation f(ξ) = f(ξ ,ξ ), setting sup ⊥ |f(ξ ,ξ )| ⊥ ⊥ ξ ∈V dim V dim V +1 a := 2 sup , ˆ 2 |f(ξ ,ξ )| dξ ξ ∈V ∞ 1 and using Plancherel’s identity, the L − L bounds for the Fourier transform and Hölder’s inequality, we obtain that F f(·,ξ ) ξ 1 ⊥ ⊥ ⊥ ⊥ L (V ) dim V dim V +1 dim V +1 a ≤ 2 sup ≤ C()2 . F f(·,ξ ) ξ ∈V ξ 2 ⊥ L (V ) In particular, the constant a is well-deﬁned. Returning to (18), we consequently deduce that for b ∈ (0, 1) 2 2 ˜ ˆ E (f ) ≥ C(, m)b |f | dξ. el {ξ :|ξ |≤ } −1 d d 1 Choosing b = μ < 1 and noting that {ξ ∈ R :| (ξ )|≤ μ}={ξ ∈ R :|ξ |≤ } im- plies the claim. 5 Page 16 of 50 B. Raitae ˘ tal. With Lemma 3.7 in hand, we turn to the proof of the lower bound in Proposition 3.6: Proof of the lower bound in Proposition 3.6 Since B − A ∈ \ I , there exists ξ ∈ R \{0} A A such that A(ξ )(B − A) = 0. As A(D) is a ﬁrst order operator, we have that A(ξ ) is linear in ξ ∈ R . In particular, we have that the set, cf. (6), V := V ={ξ ∈ R : A(ξ )(B − A) = 0}={0} B−A A,B−A d ⊥ is a linear space. Rewriting R ξ = ξ + ξ with ξ ∈ V and ξ ∈ V as in the proof B−A B−A of Lemma 3.7, then Corollary 3.2 implies that E (χ ; F ) ≥ C |((1 − λ)χ ˆ − λχ ˆ )A( )(A − B)| dξ, el λ A B |ξ | ˆ ˆ (19) E (χ ) = |∇(χ − F )|= |∇((1 − λ)χ − λχ )(A − B)|dx surf λ A B ≥ C |∇[(1 − λ)χ − λχ ]|, A B for a constant C depending on the operator A(D) and on A − B . Now, setting m(ξ ) := A(ξ )(A − B) and f := (1 − λ)χ − λχ , yields the applicability A B of Lemma 3.7 with V = V R . This is the only place where we use the assumption B−A that A − B/ ∈ I . We deduce that by (19) and the decomposition of R into the two regions from Lemma 3.7 we have for μ> 1 2 2 2 f ≤ 2 χ (D)f +χ (D)f 2 {|ξ |≥μ} 2 {|ξ |≤μ} 2 L L L 2 −1 −1 −1 ≤ C μ E (χ ; F ) + (μ )E (χ ) + μ Per() , el λ surf where the constant C depends on A(D), A, B , , d , λ. Now choosing μ = > 1, noting −1 −1 − that then μ = for < 1, we obtain 2 1 2 − 3 3 f ≤ C E (χ ; F ) + Per() , 2 λ where E (χ ; F ) := E (χ ; F ) + E (χ ). Using the lower bound λ el λ surf ˆ ˆ 2 2 2 2 f = |(1 − λ)χ ˆ + λχ ˆ | dξ = |(1 − λ)χ + λχ | dx ≥ (min{λ, 1 − λ}) ||, 2 A B A B d d R R then implies that ≤ C(E (χ ; F ) + Per()). Finally, for ∈ (0, ) and = (A(D), A, B, , d, λ) > 0 sufﬁciently small, the perime- 0 0 0 ter contribution on the right hand side can be absorbed into the left hand side, which yields the desired result. Scaling for the Two-State Problem and a T Structure Page 17 of 50 5 3.4 The Super-Compatible Setting: Proof of Theorem 1(iii) and Reduction to Cocanceling Operators In this subsection we will show that if we are in the setting in which the estimates of the previous subsection degenerate, i.e., A(ξ )(A − B) = 0for all ξ ∈ R , then in fact there can be no non-trivial bound from below. We will however also show that, in general for pairwise not super-compatible wells, it is possible to reduce to an equivalent minimization problem in the setting of cocanceling operators for suitably modiﬁed boundary data. Proof of the super-compatible case in Theorem 1 It sufﬁces to give an upper bound construc- tion with zero total energy. To this end, we consider χ = Aχ and u = Aχ + F χ d and R \ observe that A(D)u = A(D)(u − B) = A(D)[(χ + λχ d )(A − B)]= 0. R \ As a consequence, u is admissible in the deﬁnition of the elastic energy and the elastic energy vanishes. Moreover, since χ = A in we also have the vanishing of the surface energy. This concludes the argument. We will show that for two not super-compatible wells, we can always assume that I ={0}, in which case we work in the class of cocanceling operators introduced by Van Schaftingen in [65]. Proposition 3.8 Let n, d, k ∈ N, let ⊂ R be a bounded Lipschitz domain, A(D) a differ- ential operator of order k as in (2) with I given in (8). For χ ∈ L (; K) for some compact n qc setofstates K ⊂ R let E (χ ; F) be as in (9) for F ∈ K , cf. Sect. 2. Then for the restricted el ∞ d ⊥ ∞ d m 2 operator A(D) : C (R ; I ) → C (R ; R ) there exists a function χ ∈ L (, K) A I such that A 2 E (χ ; F) = E (χ ; F ) = inf |u − χ | dx. el ⊥ ⊥ ⊥ ⊥ el u ∈D Here we denote the orthogonal projection of F onto I by F . n ⊥ Proof We use the orthogonal decomposition R = I ⊕ I to write u = u + u ,χ = χ + χ , ⊥ I ⊥ I d ⊥ d ⊥ with u : R → I , u : R → I , χ : → I , χ : → I . By orthogonality we can ⊥ I A ⊥ I A A A also split the elastic energy ˆ ˆ ˆ 2 2 2 E (u, χ ) = |u − χ | dx = |u − χ | dx + |u − χ | dx. el ⊥ ⊥ I I ∞ d ⊥ ∞ d m ˜ ˜ Deﬁning the restricted operator A(D) : C (R ; I ) → C (R ; R ), A(D)u := A(D)u and the restricted space of admissible functions as in (10) A 2 d ⊥ d d ˜ ¯ D := {u ∈ L (R ; I ) : A(D)u = 0in R ,u = F in R \ }, ⊥ ⊥ ⊥ ⊥ F loc A we see that for u ∈ D it holds u ∈ D , with F = ⊥ F ,and u = F − F outside . F ⊥ ⊥ I ⊥ F I A 5 Page 18 of 50 B. Raitae ˘ tal. Thus, after minimizing the elastic energy in u, it holds ˆ ˆ 2 2 inf E (u, χ ) = inf inf |u − χ | dx + |u − χ | dx. el ⊥ ⊥ I I ˜ 2 u∈D d ¯ c F A u ∈L (R ;I ),u =F −F in u ∈D I A I ⊥ ⊥ loc As we have seen in the proof of Theorem 1(iii), the second term involving u vanishes and hence, E (χ ; F) = E (χ ; F ). el ⊥ ⊥ el This reduces the elastic energy to the case of a cocanceling operator as indeed I = {0}. As the surface energy does not depend on the operator A(D), this result yields: E (χ ; F) = E (χ ; F ) + E (χ ). ⊥ ⊥ surf el As a corollary, we apply this to the N -well problem: Corollary 3.9 (Finitely many, pairwise not super-compatible wells) Under the same as- sumptions as in Proposition 3.8, with χ ∈ BV (; K), for the special case that K := {B ,...,B }⊂ R for N ∈ N such that B , j ∈{1,...,N }, are pairwise not super- 1 N j compatible, i.e. B − B ∈ / I for i = j , there exists a constant C = C(B ,...,B )> 1 i j A 1 n such that −1 C E (χ ) ≤ E (χ ) ≤ CE (χ ). surf ⊥ surf surf ⊥ In particular, it hence holds that E (χ ; F) ∼ E (χ ; F ). ⊥ ⊥ N N Proof Writing χ = B χ with χ ∈ BV (;{0, 1}), χ = 1in ,wecan j =1 j j j =1 j calculate d−1 ∗ ∗ |∇χ|= |B − B |H (∂ ∩ ∂ ), i j i j i<j d−1 ∗ ∗ |∇χ |= |(B − B ) |H (∂ ∩ ∂ ), ⊥ i j ⊥ i j i<j where we denote the reduced boundary of a set E with ﬁnite perimeter by ∂ E andusedthe notation from above for B ∈ R to write B = ⊥ B . By assumption, for i< j it holds B − B ∈ / I , and therefore also the projection satisﬁes i j A (B − B ) = 0. This implies |(B − B ) | > 0 for all tupels (i, j ) such that i< j and hence i j ⊥ i j ⊥ |B −B | i j −1 there is a constant 0 <C < <C such that |(B −B ) | i j ⊥ −1 0 <C |∇χ |≤|∇χ|≤ C|∇χ |. ⊥ ⊥ This together with Proposition 3.8 concludes the proof. Scaling for the Two-State Problem and a T Structure Page 19 of 50 5 We emphasize that Corollary 3.9 in particular holds in the context of Theorem 1.There- fore in the statement of Theorem 1(ii) we can assume without loss of generality that I ={0}. In fact, note that in the crucial bound of Corollary 3.2 we have A(ξ )(A − B) = A(ξ )(A − B ) = A(ξ )(A − B ) ⊥ ⊥ ⊥ ⊥ for |ξ|= 1. 3.5 Some Remarks on the Compatible Two-State Problem for Higher Order Operators We conclude our discussion of lower scaling bounds by commenting on the case of the com- patible two-well problem for higher order operators. Here the situation is still less transpar- ent, yet some remarks are possible. Indeed, on the one hand, it is known that, in general, for operators of order k ≥ 2the two-well problem does not have to scale with . In order to illustrate this, we consider the speciﬁc operator A(D) := curl curl. This operator is the annihilator of the symmetrized 1 t gradient e(v) := (∇v + (∇v) ). We consider the following quantitative two-state problem (for d = 2) E (v, χ ) := E (v, χ ) + E (χ ) el surf ˆ ˆ := e(v) − dx + |∇χ | (20) 01 + α(1 − 2χ) 2 2 [0,1] [0,1] A 2 with e(v) ∈ D , χ ∈ BV ([0, 1] ;{0, 1}), α ∈ (0, 1) and study the corresponding minimiza- tion problem with prescribed boundary data F := . Proposition 3.10 Let E (v, χ ) and F be as in (20). Then there exists = (α) > 0 such 0 0 that for ∈ (0, ) it holds that inf inf E (v, χ ) ∼ . 2 A χ ∈BV ([0,1] ;{0,1}) e(v)∈D We remark that this observation is not new; indeed, a geometrically nonlinear version of this had earlier been derived in [1, Theorem 1.2]. As observed in [1] the reason for the different scaling in Proposition 3.10 and [1, Theorem 1.2], compared to the more standard behavior from Theorem 1, consists of the higher degeneracy of the multiplier associ- ated with the energy which is manifested in the presence of only one possible normal in the (symmetrized) rank-one condition. On the level of the multiplier this can be seen as e ⊗e +e ⊗e 2 1 2 2 1 A(ξ )(e ⊗ e ) = ξ , opposed to A(ξ )( ) =−ξ ξ , has only one root of multiplic- 2 2 1 2 ity two instead of two roots of single multiplicity on the unit sphere. For convenience of the reader and in order to illustrate the robustness of the above approach within geometrically linear theories, we present an alternative short proof (of the lower bound) of Proposition 3.10 based on our Fourier theoretic framework. We note that in the geometrically linear setting this provides an alternative to the approach from [1] in which the lower bound for the en- ergy is deduced by a local “averaging” argument, considering the energy on representative domain patches with the expected scaling behavior. 5 Page 20 of 50 B. Raitae ˘ tal. Proof Step 1: Lower bound. We note that the lower bound for this setting directly follows from our arguments above: Indeed, for A − B = 2αe ⊗ e , we obtain that 2 2 A(ξ )(B − A) = 2αξ × ξ × (e ⊗ e ) = 2αξ . 2 2 With this in hand, an analogous argument as in Lemma 3.7 and, in particular, in (18) implies that 2 4 ˆ χ ≤ Cμ E (χ ; F), (21) el 2 2 L ({ξ ∈R :|k |≤μ}) where we have used that in this situation the multiplier is given by m(ξ ) = A(ξ )(B − A) ∼ ξ . The different exponent of μ in (21) (compared to the one from Lemma 3.7(a)) is a consequence of the degeneracy of the symbol m(ξ ) and the higher order of the operator curl curl (or put, more concretely, the quadratic dependence ξ ). Hence, replacing the bound from Lemma 3.7(a) by the one from (21) and carrying out the splitting as in the proof of Theorem 1, we obtain the following optimization problem: For f := 1 + α(1 − 2χ) 2 2 2 2 (1 − α) ≤f ≤χ (D)f +χ (D)f 2 {|ξ |≥μ} 2 {|ξ |≤μ} 2 L L L 4 −1 −1 −1 ≤ C μ E (χ ; F) + (μ )E (χ ) + μ Per() . el surf Choosing μ ∼ and rearranging the estimates then imply the claim. Step 2: Upper bound. The associated improved upper bound makes use of the vectorial 2/3 structure of the problem in contrast to the essentially scalar “standard construction” (see the arguments below). Recalling that the nonlinear construction from the proof of [1, Lemma 2.1] also yields a construction with the desired scaling for the geometrically lin- earized problem, we do not carry out the details of this but refer to [1, Lemma 2.1] for these. On the other hand, the arguments from [1, 73, 74, 81] show that still for A(D) = curl curl 2/3 if A − B = γ(e ⊗ e + e ⊗ e ) for γ ∈ R \{0}, then one recovers the scaling for the 1 2 2 1 symmetrized gradient differential inclusion. In this case, the symbol reads m(ξ ) = A(ξ )(B − A) ∼ ξ ξ and the operator is “less degenerate”. 1 2 We expect that the scaling behavior of general higher order operators is in many inter- esting settings directly linked to the degeneracy of the symbol A(ξ )(B − A).Weplanto explore this in future work. 4 Quantitative Rigidity of the T Structure from (13), (14)for A(D) = div In this section, we consider the T structure for the divergence operator introduced in (13), (14). The upper bound construction is given by an approximate solution of the type de- scribed in the introduction. The lower bound is motivated by the rigidity of exact solutions as outlined in Sect. 4.3.1. 4.1 The Upper Bound Construction – an Inﬁnite Order Laminate To begin with, we construct an inﬁnite order laminate similar to the one for the Tartar square (cf. [2, 82, 83] for quantitative versions of this). This is based on [3] and will yield the Scaling for the Two-State Problem and a T Structure Page 21 of 50 5 Fig. 1 The diagonal matrices A , A , A , S , S , S with the 2 3 1 2 3 dashed lines depicting the connections in the wave cone for A(D) = div and the Voronoi-regions of A shown by the lines. As shown in [21]the qc set K is given by the inner triangle formed by S S S and 1 2 3 the “legs” connecting S and A j j for j ∈{1, 2, 3} upper bound estimate from Theorem 2. We recall that for the divergence operator, instead of requiring rank-one connectedness for the existence of a laminate as for the curl, in our three-dimensional set-up we need rank-one or rank-two connectedness as can be seen from the wave cone for the divergence operator, cf. (5). As a consequence, for two matrices A, B ∈ 3×3 3 3×3 R such that rank(B − A) ≤ 2 there exists a piecewise constant map u : R → R such that u ∈{A, B} a.e. in and div u = 0. The lamination can be done in any direction of the kernel ker(B − A) = {0}. Considering now the matrices A , A , A given in (14), we observe that rank(A − A ) = 1 2 3 i j 3×3 3for i = j . Following [3], we introduce auxiliary matrices S ,S ,S ∈ R . 1 2 3 ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ 000 00 000 2 2 1 ⎝ ⎠ ⎝ ⎠ ⎝ ⎠ S = 0 0 ,S = 0 0 ,S = 0 0 . (22) 1 2 3 3 3 3 002 00 1 001 It then holds for i = 1, 2, 3that ker(S − A ) = span(e ), S = (A + S ), i i i i i+1 i+1 where A = A , S = S .Asprovedin[21, Theorem 2], the A-quasi-convex hull 4 1 4 1 qc {A ,A ,A } can be explicitly characterized as the convex hull of the matrices S , S , 1 2 3 1 2 S together with the “legs” given by the line segments A S for j ∈{1, 2, 3},cf. Fig. 1. 3 j j For simplicity and deﬁniteness, we ﬁrst assume, that u = F = S outside =[0, 1] .In this setting we prove the following energy estimate: Proposition 4.1 Let =[0, 1] , K ={A ,A ,A } for the particular choice of matrices in 1 2 3 −1 (14) and let E be as in (12) and ∈ (0, 1). Then for any r ∈ (0, ) with r ∈ 4N there are (m) 1,∞ 3 3×3 (m) (m) 3 sequences u ∈ W (R ; R ), such that div u = 0, u = F := S outside [0, 1] , (m) 3 and χ ∈ BV (R ; K), and a constant C = C(F ) > 1 with (m) (m) −m −k E (u ,χ ) ≤ C 2 + 2 r + r + . k=1 In order to achieve this, in the next subsections, we iteratively construct a higher and higher order laminate (depending on > 0). As in the setting of the Tartar square, we keep track of the surface and elastic energy contributions which arise in this process. 5 Page 22 of 50 B. Raitae ˘ tal. 4.2 Proof of the Upper Bound from Theorem 2 We split the proof of the upper bound from Theorem 2 into several steps which we will carry out in the next sections and then combine in Sect. 4.2.4. First, we start by a simple lamination of A and S to obtain regions in which u ∈ K 1 1 holds and to satisfy the exterior data condition u = F = S = (A + S ) outside .This 3 1 1 is followed by a similar construction replacing S by a lamination of A and S achieving 1 2 2 a second order laminate. Iterating the procedure of replacing S by a lamination of A j j +1 and S (with the convention that A = A , S = S ) yields Proposition 4.1. Finally, we j +1 4 1 4 1 optimize the parameter r and the number of iterations depending on in order to show the desired upper bound estimate in Proposition 4.3. As we will use a potential for the laminates, we will deﬁne a “proﬁle-function” once in a more general form and will then refer to this in our construction for the higher order laminates. Lemma 4.2 Let R =[a ,b ]×[a ,b ]×[a ,b ]⊂ R be an axis-parallel cuboid, then for 1 1 2 2 3 3 b −a j j any direction e with j ∈{1, 2, 3} and any scale r> 0 such that ∈ N there is a contin- 3×3 uous function f ( ·; R, r) : R → R satisfying the following properties: • The function f ( ·; R, r) only depends on the j -th coordinate x and is r -periodic. j j • It holds f (x; R, r) = 0 if x ∈ R is such that x = (x ,x ,x ) lies in one of the planes j 1 2 3 characterized by x ∈ a + rN . j j 0 S −A j j • The matrix-valued function curl f (x; R, r) only attains the values ± , where A , S j j j is given as in (14) and in (22), respectively. Furthermore, the volumes of the level sets are equal, i.e. |{x ∈ R : f (x; R, r) = S }| = |{x ∈ R : f (x; R, r) = A }| = |R|. j j j j Proof Without loss of generality by a translation, we may assume that R =[0,b ]×[0,b ]× 1 2 [0,b ] for b ,b ,b > 0. We consider the continuous one-periodic extension of the function 3 1 2 3 1 1 tt ∈[0, ), 2 2 h :[0, 1) → R,h(t) := 1 1 (1 − t) t ∈[ , 1). 2 2 Furthermore, we deﬁne the matrices ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ 00 0 001 01 0 2 2 ⎝ ⎠ ⎝ ⎠ ⎝ ⎠ M := 00 − ,M := 000 ,M := − 00 , 1 2 3 3 3 02 0 200 00 0 3×3 satisfying e × M = S − A . With this at hand we deﬁne f ( ·; R, r) : R → R : j j j j j f (x ,x ,x ; R, r) := rh( )M . j 1 2 3 j It follows directly, that f ( ·; R, r) is continuous, only depends on x ,is r -periodic, and j j x S −A j j j vanishes for x ∈ rN . Lastly, we note that curl f (x; R, r) = h ( )e × M ∈{± } j 0 j j j r 2 and that indeed |{x ∈ R : f (x; R, r) = S }| = |{x ∈ R : f (x; R, r) = A }| = |R|. j j j j 4.2.1 First Order Laminates 3 3×3 We use a potential v : R → R to construct our laminates attaining the prescribed exterior data, i.e. we consider the row-wise curl: u = curl v. Scaling for the Two-State Problem and a T Structure Page 23 of 50 5 1 1 As S = A + S and (S − A )e = 0 the ﬁrst order lamination is in the e -direction. 3 1 1 1 1 1 1 2 2 We seek to use Lemma 4.2 to construct v, but have to adapt the boundary condition. For 3 3×3 ˜ ˜ ˜ this we deﬁne a mapping S : R → R with curl S = S = F . A possible choice for S is 3 3 3 3 given by the following matrix-valued function ⎛ ⎞ 00 0 ˜ ⎝ ⎠ S (x) := 00 − x . 3 1 0 x 0 Furthermore, we also deﬁne the cut-off function 0 t< , 1 1 3 φ : R →[0, 1],φ(t) = 4t − t ∈[ , ], 2 8 8 1 t> . 1 −1 With this, we deﬁne the (continuous) potential for r ∈ (0, ), r ∈ 4N by using Lemma 4.2 for j = 1, R =[0, 1] : (1) 3×3 v : → R , (1) v (x) = φ( d (x))f (x; , r) + S (x), ∂ 1 3 where d (x) denotes a smoothed-out distance function to the boundary ∂. Without (1) 3 change of notation, we consider the (continuous) extension of v to R by S (x),which is (1) possible, as v (x) = S (x) on ∂. We then set (1) (1) u := curl v , and note that in it holds (1) (1) u (x) = curl v (x) d (x) x d (x) ∂ 1 ∂ = φ ( )h( )∇d (x) × M + φ( ) curl f (x; , r) + F ∂ 1 1 r r r d (x) x d (x) x 1 ∂ 1 ∂ 1 = φ ( )h( )∇d (x) × M + φ( )h ( )(S − A ) + (S + A ). ∂ 1 1 1 1 1 r r r r 2 (1) With these considerations, we have obtained the following properties: It holds div u = −1 3 (1) 3 (1) 3 (1) div(curl v ) = 0in R , u (x) = F in R \ and for x ∈ \ d ([0, r]) we have u ∈ (1) {A ,S }. In other words, our deformation u is a divergence-free function satisfying the 1 1 desired boundary conditions and which, outside of the cut-off region, is a solution to the (1) differential inclusion u ∈{A ,S }. 1 1 With the higher order laminates in mind we rephrase this using the decomposition of into the three disjoint parts consisting of the S -cells, the A -cells and the cut-off region. To 1 1 be more precise, we deﬁne d (x) (1) (1) R := {x ∈ : φ( ) = 1,u = S }, d (x) (1) (1) Q := {x ∈ : φ( ) = 1,u = A }, r 5 Page 24 of 50 B. Raitae ˘ tal. Fig. 2 The x = slice of the (1) projection χ ,bluerepresents A ,red A and green A .(Color 1 2 3 ﬁgure online) d (x) (1) C := {x ∈ : φ( )< 1}. (1) (1) (1) (1) (1) Indeed it holds = R ∪ Q ∪ C and |C |= 6 r and by Lemma 4.2 we know |R |≤ 1 1 ||= . 2 2 (1) (1) Choosing χ = u as the pointwise orthogonal (with a ﬁxed choice for the not (1) uniquely deﬁned points) projection of u onto K,see Fig. 2 for an illustration, up to a uniformly bounded constant, the elastic energy can be bounded by the measure of the region (1) in which u (x) = S and the cut-off region: (1) (1) (1) (1) 2 (1) (1) E (u ,χ ) = |u − χ | dx ≤ C |R |+|C | el 1 3r 1 ≤ C ||+ 6 ≤ C + 3r . 2 8 2 (1) Furthermore, the surface energy is bounded by counting the interfaces at which χ may 2 2 8 jump. This consists of at most interfaces in the interior and at most + 2 · 6 ≤ new r r r interfaces in the cut-off region. For r ∈ (0, ), the surface area is thus controlled by (1) (1) E (χ ) = |∇χ |≤ C . surf 4.2.2 Second Order Lamination After this ﬁrst order lamination, the differential inclusion u ∈ K with div u = 0 holds only (1) in Q . In order to further reduce the energy, we now replace each of the many cuboids (1) (1) 1 1 in R for which u = S = A + S by a lamination in the e -direction. For this, we 1 2 2 2 2 2 (1) modify the potential v in these regions: (1) 2 r −2 3 For x ∈ R and for r ∈ (0, ), r + ∈ N, we deﬁne with the help of Lemma 4.2 2 4r d (1) (x) (2) ∂R (1) 2 v (x) = φ( )f (x; R ,r ) + f (x; , r) + S (x) 2 1 3 (2) (1) (1) and v (x) = v (x) else. Here, and in the following, we use the notation f (x; R ,r) (1) for R which is not a cuboid but an union of disjoint cuboids and mean depending on x (2) (1) the corresponding connected cuboid. By this v deﬁnes a continuous map, as inside R Scaling for the Two-State Problem and a T Structure Page 25 of 50 5 (1) the cut-off attains the constant value one and therefore v (x) = f (x; , r) + S (x) for 1 3 (1) (2) (2) x ∈ ∂R . By construction, the map u := curl v is divergence free, i.e. the interfaces are compatible. (2) (2) As in the construction for ﬁrst order laminates we set χ = u , and deﬁne the sets d (x) (1) (2) (1) ∂R (2) R := {x ∈ R : φ( ) = 1,u = S }, d (x) (1) ∂R (2) (1) (2) Q := {x ∈ R : φ( ) = 1,u = A }, d (1) (x) ∂R (2) (1) C := {x ∈ R : φ( )< 1}. (1) (2) (2) (2) (2) 1 (1) Then indeed again we have the decomposition R = R ∪ Q ∪ C with |R |≤ |R | (2) 1 3 2 and |C |≤ r 6 as the volume of the cut-off region can be bounded by the number of r 8 (1) 2 cells in R times r times six times the area of the biggest face. (2) As the elastic energy vanishes in Q ,weobtain (2) (2) (2) (2) 2 (2) (2) (1) E (u ,χ ) = |u − χ | dx ≤ C |R |+|C |+|C | el 1 3 ≤ C + 3r + r . 4 2 Indeed, this follows from the fact that we have improved our deformation in half the volume (1) of the region in which u ∈ / K but have added a new cut-off region in each cuboid in which we do the second order lamination. For the surface energy it holds 10 1 10 r 10 5 (2) E (χ ) ≤ C( + ) = C( + ), surf 2 2 r r r 2 r r 4 10 1 as we add at most + 12 ≤ many new faces in each one of the many cuboids and as 2 2 r r each surface has a surface area of size at most . 4.2.3 Iteration: (m+1)-th Order Without loss of generality, we assume that the (m + 1)-th order lamination will be in e - direction, i.e. m = 3j for some j ∈ N. Else, we only have to adapt the corresponding roles of the directions. (m) We deﬁne iteratively the (m + 1)-th potential with the help of the sets R , for this we (m) (m) (m) set (for given v ,u = curl v ) d (x) (m−1) (m) (m−1) ∂R (m) R := {x ∈ R : φ( ) = 1,u = S }, d (x) (m−1) ∂R (m) (m−1) (m) Q := {x ∈ R : φ( ) = 1,u = A }, d (m−1) (x) ∂R (m) (m−1) C := {x ∈ R : φ( )< 1}. (m) (m+1) Inside R we then deﬁne v by d (m) (x) ∂R (m+1) (m) m+1 (k−1) k v (x) = φ( )f (x; R ,r ) + f (x; R ,r ) + S (x), 1 [k] 3 m+1 k=1 5 Page 26 of 50 B. Raitae ˘ tal. Fig. 3 One S -cell with the corresponding lamination in e -direction. In green/red/orange the inner boundary of the cut-off region is depicted. (Color ﬁgure online) ⎪ 1 k ≡ 1mod 3, [k]= 2 k ≡ 2mod 3, 3 k ≡ 0mod 3, (m+1) (m) (m) (m+1) and v (x) = v (x) else for x/ ∈ R . By induction we see that v is continuous, d (x) (m) (m) ∂R (m+1) (m+1) that for x ∈ R such that φ( ) = 1 it holds that u := curl v ∈{A ,S } m+1 1 1 (m−1) (m) (m) (m) (m) 1 (m−1) and that we have the decomposition R = R ∪ Q ∪ C with |R |≤ |R |, (m) −m r −m |C |≤ C2 = C2 r for a universal constant C> 0 independent of m, r.This m−1 −m (m) 2 follows from the fact, that the number of cuboids in R is bounded by C and m−1 m−2 m−3 r r r m−2 m−3 that the biggest face of each cuboid has an area of at most r r . As the previous cut-offs still contribute to the total energy, we obtain the following elastic (m+1) (m+1) energy bound (χ := u ) ⎛ ⎞ m+1 m+1 (m+1) (m+1) (m+1) (k) −m −j ⎝ ⎠ E (u ,χ ) ≤ C |R |∪ |C | = C 2 + r + 2 r . el k=1 j =0 (1) (1) (1) (m+1) This bound resembles the (iterative) decomposition of = R ∪ Q ∪ C = R ∪ m+1 m+1 m+1 (k) (k) m+1 (m+1) (k) Q ∪ C and the fact that u = χ in Q . k=1 k=1 k=1 For the surface energy, we again calculate the new contribution of the next order lam- ination and then sum over all previous ones. Each new face has a surface area of at most m m−1 m−2 m−2 r r r r . In each S -cell, we add new faces in the lamination and at most + 12 m+1 m+1 2 2 r r −m−1 ones for the cut-off, cf. Fig. 3. Since we have at most C new cells the surface m m−1 m−2 r r r energy is increased by m−2 m m−1 8 r r r 1 −m −m C2 ( + (surf. in cut-off )) ≤ C2 , m m−1 m−2 m+1 m+1 r r r r 2 2 r yielding the following overall surface energy bound: ⎛ ⎞ m+1 (m+1) −j −j ⎝ ⎠ E (χ ) ≤ C 2 r ≤ . surf m+1 j =1 For the total energy this implies ⎛ ⎞ (m+1) (m+1) −m −j ⎝ ⎠ E (u ,χ ) ≤ C 2 + r + 2 r + . m+1 j =1 Scaling for the Two-State Problem and a T Structure Page 27 of 50 5 With this we have shown the claimed upper bound and have thus concluded the proof of Proposition 4.1. 4.2.4 Combining the Estimates: Proof of the Upper Bound in Theorem 2 With the previous construction in hand, we conclude the upper estimate from Theorem 2: 3 3×3 Proposition 4.3 Let =[0, 1] , K ={A ,A ,A } for A ,A ,A ∈ R be given in (14), 1 2 3 1 2 3 qc let ∈ (0, 1), and let E be as in (12), F ∈ K \ K and D given in (10). Then there are constants c> 0 and C> 1, only depending on the boundary data F , such that inf inf E (u, χ ) ≤ C exp(−c| log | ). u∈D χ ∈BV (;K) Proof Step 1: Conclusion of the argument for F = S . Using the sequences constructed in the construction from Proposition 4.1, implies (m) (m) −m −k −m −m −m E (u ,χ ) ≤ C 2 + 2 r + r + r ≤ C 2 + r + r . k=1 m+1 Optimizing the value of r depending on > 0, we require that r ∼ . Finally, we balance the resulting contributions and seek for the optimal number of iterations m.Thisis 1 1 −m m+1 2 given by 2 ∼ ,thatis m ∼| log | . Plugging this into the upper bound results in (m) (m) E (u ,χ ) ≤ C exp − log(2)| log()| . This concludes the proof for the special case F = S . Step 2: Conclusion of the argument for a general boundary datum. The situation of other qc boundary data F ∈ K can be reduced to the one from Proposition 4.3 by at most two 3×3 further iterations. Indeed, a general matrix F ∈ R in the convex hull of S , S , S , can be 1 2 3 represented as F = λF + (1 − λ)F = λ(ν A + (1 − ν )S ) + (1 − λ)(ν A + (1 − ν )S ) 1 2 1 j 1 j 2 k 2 k with λ, ν ,ν ∈[0, 1], j, k ∈{1, 2, 3} and k = j . Hence, after two additional iterations com- 1 2 pared to the argument from above, we arrive at similar iterative procedures as in the previous subsections. In case that F is an element of one of the legs S A a single iteration sufﬁces j j to reduce the situation to the above argument. This proves the result for a general boundary qc condition F ∈ K \ K. 4.3 Proof of the Lower Bound In this section, we present the proof of the lower bound from Theorem 2. To this end, sim- ilarly as in [2], we mimic and quantify the analogous argument from the stress-free setting which we brieﬂy recall in the following Sect. 4.3.1 and for which we will provide a number of auxiliary results in Sect. 4.3.2. The main argument, given in Sect. 4.3.3, will then consist of a bootstrap strategy, similar to [2], in which we iteratively reduce the possible regions of mass concentration in Fourier space. 5 Page 28 of 50 B. Raitae ˘ tal. Contrary to the previous section, in what follows we will work in a periodic set-up. Since the energy contributions on periodic functions provides a lower bound on the energy contributions of functions with prescribed Dirichlet boundary conditions, we hence also obtain the desired lower bound for the setting of Dirichlet boundary conditions. Indeed, for the elastic energy this is immediate; for the surface energy there is at most an increase by a ﬁxed factor (see the discussion in Lemma 4.5 in Sect. 4.3.2). 4.3.1 The Stress-Free Argument We begin by recalling the argument for the rigidity of the exact inclusion, as we will mimic this on the energetic level. 3 3×3 Proposition 4.4 Let u :[0, 1] → R be a solution to the differential inclusion (13)-(14). Then, there exists j ∈{1, 2, 3} such that u ≡ A in [0, 1] . Proof In the exactly stress-free setting in which the differential inclusion is satisﬁed exactly, i.e. u ∈{A ,A ,A }, the observation that div u = 0and that u is a diagonal matrix leads to 1 2 3 the following three equations ∂ u = 0,∂ u = 0,∂ u = 0, 1 11 2 22 3 33 where u denote the diagonal components of the matrix u. As a consequence, jj u = f (x ,x ), u = f (x ,x ), u = f (x ,x ). 11 1 2 3 22 2 1 3 33 3 1 2 Next we note that the values of u determine the ones for u if j = k, i.e. there are functions jj kk h such that h (u ) = u . Hence, comparing the functions u and u ,weﬁrstobtain k,j k,j jj kk 11 22 that u and u can only be functions of x . Indeed it holds 11 22 3 ∂ u (x) = ∂ h (u (x)) = ∂ h (f (x ,x )) = 0, (23) 2 11 2 1,2 22 2 1,2 2 1 3 and analogously ∂ u (x) = 0. Comparing this to u , we obtain that all three functions must 1 22 33 be constant. Hence, any solution to the (exact) differential inclusion must be constant and u is equal to one of the three matrices A , A , A globally. The exact problem is hence 1 2 3 rigid. Using the ideas from [2], we seek to turn this into a corresponding scaling result. The main difference that arises can be seen in the qualitative rigidity argument above: Instead of comparing only two diagonal entries like in [2], we have to compare twice to deduce that the map is constant. This will be seen in the quantitative argument for the lower bound below. Whereas in [2] there are cones around a single axis (the diagonal entries only depend on one variable), we consider cones around a plane (the diagonal entries depend on two variables). Furthermore, the bootstrap argument will be slightly modiﬁed as it resembles the comparison of the diagonal entries in the qualitative argument given above. 4.3.2 Reduction to the Periodic Setting and Auxiliary Results for the Elastic Energy In this subsection, we provide a number of auxiliary results which we will exploit in the following bootstrap arguments for deducing the lower bound. As a ﬁrst step, we reduce to the situation of periodic deformations. Scaling for the Two-State Problem and a T Structure Page 29 of 50 5 3 qc Lemma 4.5 Let =[0, 1] , let K := {A ,A ,A } be as in (14) and F ∈ K \ K. Let 1 2 3 E (u, χ ) be given by (15) and set ˆ ˆ per 2 E (u, χ ) := |u − χ | dx + |∇χ |. 3 3 T T per 3 3×3 3 Let further D := {u : R → R : div u = 0 in R , u = F }, where u := u(x)dx . Assume that E (u, χ ) ≤ 1 and that there is > 0 such that for any ν ∈ (0, ) there is c > 0 0 ν such that for any ∈ (0, ) it holds that per +ν E (u, χ ) ≥ exp(−c | log()| ). (24) Then, there exists a constant C> 1 such that for ˜ =˜ (ν) > 0 sufﬁciently small 0 0 −1 +ν C exp(−c | log()| ) ≤ inf inf E (u, χ ) 3 u∈D χ ∈BV ([0,1] ;K) F for all ∈ (0, ˜ ). 3 3 Proof In order to infer the lower bound, we show that any function u :[0, 1] → R with constant boundary data can be associated with a suitable periodic function which has the boundary data of u as its mean value and satisﬁes related energy estimates. Indeed, for given u ∈ D , we view it as a function on T by restriction. By the prescribed boundary data it still satisﬁes the differential constraint and further the mean value property. Moreover, per inf E (u, χ ) := inf |u − χ | dx ≤ inf E (u, χ ). el el per per u∈D 3 F u∈D u∈D F F 3 3 Next, viewing χ :[0, 1] → K as a periodic function χ ˜ : T → K,weinfer that ˆ ˆ |∇χ ˜ |≤ |∇χ|+ C. 3 3 T [0,1] Now, due to (24), we obtain that +ν per exp(−c | log()| ) ≤ inf inf E (u, χ ) ≤ inf inf E (u, χ ) + C. per 3 3 u∈D χ ∈BV (T ;K) u∈D χ ∈BV ([0,1] ;K) F For ˜ (ν) > 0 sufﬁciently small, the last right hand side term may thus be absorbed into the left hand side, yielding the desired result. With this result in hand it sufﬁces to consider the periodic set-up in the remainder of this section. This will, in particular, allow us to rely on the periodic Fourier transform in deducing lower bounds for the elastic energy. In what follows all (semi-)norms will thus be considered on the torus. With a slight abuse of notation, we will often omit this dependence. per per qc Lemma 4.6 Let F ∈ K \K, where K ={A ,A ,A } with A in (14), and E and D be 1 2 3 j el F per 2 3 as in Lemma 4.5 and as in (10). Then, it holds for any χ ∈ L (T ; K) and for E (χ ; F) := el per inf u − χ dx u∈D T per i 2 2 E (χ ; F) = |ˆ χ | +|χ( ˆ 0) − F | , i,i el |k| i=1 k∈Z \{0} 5 Page 30 of 50 B. Raitae ˘ tal. where χ are the diagonal entries of χ and χ ˆ is the (discrete) Fourier transform of χ . i,i per Proof We ﬁrst calculate E (u, χ ) in Fourier space el per 2 2 E (u, χ ) = |u − χ | dx = |ˆ u −ˆ χ | , el k∈Z which allows us to characterize minimizers of the elastic energy. per In order to minimize this elastic energy in u ∈ D , u ˆ has to be the (pointwise) orthog- onal projection of χ ˆ onto the orthogonal complement of k, as the differential constraint div u = 0 reads iuk ˆ = 0 in Fourier space. Noting that the row-wise orthogonal projection of k k amatrix M onto span(k) can be written as (M) = (M ) ⊗ , the optimal u ˆ is given by |k| |k| k k u ˆ = ⊥ χ ˆ = (Id − )χ ˆ =ˆ χ − (χ ˆ ) ⊗ |k| |k| for any k ∈ Z \{0}. Returning to our energy, this yields k k per per 2 2 E (χ ; F) = E (u, χ ) = |ˆ χ − (χ ˆ ) ⊗ −ˆ χ | +|F −ˆ χ(0)| el el |k| |k| k∈Z \{0} 2 2 = |ˆ χ | +|χ( ˆ 0) − F | |k| k∈Z \{0} i 2 2 = |ˆ χ | +|χ( ˆ 0) − F | i,i |k| i=1 k∈Z \{0} and shows the claim. Next, following the ideas from [2], for μ, λ > 0, j = 1, 2, 3, we introduce the cones C ={k ∈ Z :|k |≤ μ|k|, |k|≤ λ}, (25) j,μ,λ j and their corresponding cut-off functions m (k) ∈ C (C \{0}; [0, 1]) fulﬁlling j,μ,λ j,2μ,2λ m = 1on C , supp(m (k)) ⊂ C and the decay properties in Marcinkiewicz’s j,μ,λ j,μ,λ j,μ,λ j,2μ,2λ multiplier theorem (see, for instance, [84, Corollary 6.2.5]). The corresponding cut-off mul- tiplier is thus deﬁned by −1 m (D)f = F (m (·)f(·)). (26) j,μ,λ j,μ,λ Furthermore, we use the following results which are shown in [2, Lemma 2, Lemma 3, and Corollary 1]. Following the conventions from [75], with slight abuse of notation compared to our setting in the ﬁrst part of the article, in the whole following section, we now use d to denote the degree of some suitable polynomials and no longer the dimension of the ambient space which in the whole section is simply ﬁxed to be equal to three. Lemma 4.7 Let β, δ, μ, λ > 0. Let m (D) denote the Fourier multipliers associated with i,μ,λ the cones C for i ∈{1, 2, 3} as deﬁned in (25) with the corresponding multipliers given i,μ,λ Scaling for the Two-State Problem and a T Structure Page 31 of 50 5 ∞ 3 3 in (26). Let f ∈ L (T ) ∩ BV (T ) for i = 1, 2, 3 and let h : R → R be nonlinear poly- i j,i nomials (of degree d ) with h (0) = 0 such that h (f ) = f for i = j . If j,i j,i i j 3 3 ∂ f ≤ δ, ∇f ≤ β, i i i TV ˙ −1 i=1 i=1 then there exist constants C = C(h , f ), C = C (h , f ,d), C = C (h , i,j i ∞ i,j i ∞ 0 0 i,j f ,d) > 0 such that for any γ ∈ (0, 1) we have for any i = j i ∞ 2 −2 −1 f − m (D)f ≤ C(μ δ + λ β), k k,μ,λ k 2 k=1 1−γ f − h (m (D)f ) 2 ≤ f − m (D)f , i i,j j,μ,λ j L j j,μ,λ j 2 12d L 2 −2 −1 1−γ −2 −1 h (m (D)f ) − m (D)f ≤ max{(μ δ + λ β) ,μ δ + λ β}. i,j j,μ,λ j i,μ,λ i 2 24d Here we choose the constants such that C > 2C + 2C + 3. Let us comment on these bounds: The functions f are representing the diagonal en- tries of the phase indicator χ ∈ BV ([0, 1] ; K). Thus the ﬁrst estimate corresponds to a ﬁrst frequency localization by exploiting the surface energy control for the high frequen- cies and the ellipticity of the elastic energy away from the cones C . It can be viewed j,μ,λ as a quantiﬁed version of the statement that u is a function only depending on x , x in 11 2 3 Proposition 4.4. The second estimate is a commutator bound that arises from the nonlinear relation h (f ) = f for i = j . The third estimate combines the ﬁrst two bounds. The sec- i,j j i ond and third estimate will form the core tool to iteratively decrease the Fourier support of the characteristic functions of our phase indicators. We will detail this in the remainder of the article. Remark 4.8 It is possible to make the mappings h for i, j ∈{1, 2, 3}, i = j explicit for the j,i choice of matrices A , A , A ,cf. (14). To this end, we may, for instance, consider 1 2 3 14 5 14 11 2 2 h (x) = x − x, h (x) = x − x, 1,2 1,3 9 9 3 3 21 17 21 23 2 2 h (x) = x − x, h (x) =− x + x, 2,1 2,3 4 4 2 2 7 19 7 25 2 2 h (x) =− x + x, h (x) =− x + x. 3,1 3,2 12 12 18 18 4.3.3 Comparison Argument in Fourier Space In this section, we carry out the iterative bootstrap argument which allows us to deduce the ﬁnal rigidity result. As a ﬁrst step of the bootstrap argument, we invoke the results from above which allow us to decrease the region of potential Fourier concentration from the cone C to a cone 1,μ,λ C with λ <λ. This resembles (23) in the exactly stress-free setting in a quanitiﬁed 1,μ,λ version. As f is determined by f and f with the help of h we can reduce λ, similarly 1 2 3 1,j as in our reduction of the dependences of u in Proposition 4.4. 11 5 Page 32 of 50 B. Raitae ˘ tal. Fig. 4 Illustration of the cones C (red) and C (blue), 1,μ,λ 2,μ,λ andalsoof C + C in 2,μ,λ 2,μ,λ Fourier space. (Color ﬁgure online) Lemma 4.9 Under the same conditions as in Lemma 4.7 and for 1 1 4dλμ λ> 0,μ ∈ (0, ), λ ∈ ( ,λ) √ √ 2 2 2 2d + 1 2 1 − 4μ it holds 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 L L L −2 −1 1−γ −2 −1 ≤ 10 max{(μ δ + λ β) ,μ δ + λ β}. 24d Here the constant C is chosen to be the same as in Lemma 4.7. Remark 4.10 We remark that the interval for λ is chosen such that it holds C \ 1,2μ,2λ 1 4dλμ C ⊂{max{|k |, |k |} > 4dμλ}∩ C and the one for μ such that <λ, 1,2μ,2λ 2 3 1,2μ,2λ 1−4μ i.e. the interval for λ is non-empty. Proof We ﬁrst observe that the Fourier transform of h (m (D)f ) is given by a convo- i,j j,μ,λ j lution of the function m f with itself. Hence, the support of h (m (D)f ) is con- j,μ,λ j i,j j,μ,λ j tained in the d -fold Minkowski-sum of C with itself. For j = 2, 3 it therefore holds j,2μ,2λ that F h (m (D)f ) (k) =0for |k | > 4dμλ. 1,j j,μ,λ j j Further we introduce the sets K ={|k | > 4dμλ}, K ={|k | > 4dμλ} and consider the 2 2 3 3 corresponding Fourier multipliers of the smoothed-out indicator functions χ (D), χ (D), K K 2 3 χ (D).Thisimpliesfor i = 2, 3 K ∪K 2 3 χ (D)h (m (D)f ) = 0. (27) K 1,i i,μ,λ i i Scaling for the Two-State Problem and a T Structure Page 33 of 50 5 With this we can show that the Fourier mass concentrates in a cone with smaller truncation parameter, depicted in Fig. 4: Indeed, in Fourier space χ ≤ χ + χ . Thus, K ∪K K K 2 3 2 3 2 2 χ (D)m (D)f ≤(χ (D) + χ (D))m (D)f K ∪K 1,μ,λ 1 2 K K 1,μ,λ 1 2 2 3 2 3 L L 2 2 ≤ 2χ (D)m (D)f + 2χ (D)m (D)f . K 1,μ,λ 1 2 K 1,μ,λ 1 2 2 3 L L (28) Invoking (27) together with the bounds from Lemma 4.7, then implies for i = 2, 3 2 2 χ (D)m (D)f =χ (D) m (D)f − h (m (D)f ) K 1,μ,λ 1 2 K 1,μ,λ 1 1,i i,μ,λ i 2 i i L L ≤m (D)f − h (m (D)f ) 1,μ,λ 1 1,i i,μ,λ i 2 −2 −1 1−γ −2 −1 ≤ max{(μ δ + λ β) ,μ δ + λ β}. 24d Combined with (28) this yields 2 −2 −1 1−γ −2 −1 χ (D)m (D)f ≤ 4 max{(μ δ + λ β) ,μ δ + λ β}. K ∪K 1,μ,λ 1 2 2 3 L 24d We observe that by the choice of the parameters C \ C ⊂ (K ∪ K ) ∩ C , 1,2μ,2λ 1,2μ,2λ 2 3 1,2μ,2λ and thus |m (k) − m (k)|≤ χ (k)m (k). Therefore, 1,μ,λ 1,μ,λ K ∪K 1,μ,λ 2 3 2 2 m (D)f − m (D)f ≤χ (D)m (D)f 1,μ,λ 1 1,μ,λ 1 2 K ∪K 1,μ,λ 1 2 2 3 L L −2 −1 1−γ −2 −1 ≤ 4 max{(μ δ + λ β) ,μ δ + λ β}. 24d In conclusion, (for C ≥ C ) 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 L L L 2 2 ≤f − m (D)f +f − m (D)f 2 2,μ,λ 2 2 3 3,μ,λ 3 2 L L 2 2 + 2f − m (D)f + 2m (D)f − m (D)f 1 1,μ,λ 1 2 1,μ,λ 1 1,μ,λ 1 2 L L −2 −1 −2 −1 1−γ −2 −1 ≤ 2C(μ δ + λ β) + 8 max{(μ δ + λ β) ,μ δ + λ β} 24d −2 −1 1−γ −2 −1 ≤ 10 max{(μ δ + λ β) ,μ δ + λ β}. 24d Let us stress that the decomposition into K and K is in analogy to the two compar- 2 3 isons from Proposition 4.4 in order to show that u is constant. The set K resembles the 11 2 comparison of u and u to show that u is constant in x and the set K resembles the 11 22 11 2 3 comparison of u and u to show that u does not depend on x . 11 33 11 3 Applying the previous result for all three directions simultaneously then yields the fol- lowing corollary which will serve as the induction basis for the subsequent inductive boot- strap argument. Corollary 4.11 (Induction Basis) Let β, δ, μ, λ > 0 and let C be the cones in (25) with i,μ,λ corresponding multiplier m (D), cf.(26), for i = 1, 2, 3. Further let f , h be functions i,μ,λ i j,i 5 Page 34 of 50 B. Raitae ˘ tal. as in Lemma 4.7. Let C > 0, γ ∈ (0, 1) and d ≥ 0 be the constants from Lemma 4.7. For 4dμλ 1 0 λ = λ> 0, μ ∈ (0, ), λ ∈ ( ,λ ) it holds that 0 1 0 2 2 2 1−4μ 2 2d +1 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 1 L 1 L 1 L 30C −2 −1 1−γ −2 −1 ≤ max{(μ δ + λ β) ,μ δ + λ β}. 24d With Corollary 4.11 in hand, we now iteratively further decrease the Fourier supports. To this end, we will invoke the commutator bounds from Lemma 4.7. Lemma 4.12 (Iteration process) Let β, δ > 0 and let C be the cones in (25) with cor- i,μ,λ responding multiplier m (D), cf.(26), for i = 1, 2, 3. Further let μ ∈ (0, ), i,μ,λ 2 2d +1 λ> 0, and let f , h be functions as in Lemma 4.7. Let C > 0, γ ∈ (0, 1) and d ≥ 0 i j,i 0 be the constants from Lemma 4.7. Let λ > 0 be a sequence for k ∈ N with λ = λ and k 0 4dμλ k−1 λ ∈ ( √ ,λ ). It then holds for every k ∈ N \{0} k k−1 2 1−4μ 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 k k k L L L 30C k −2 −1 (1−γ) −2 −1 ≤ max{(μ δ + λ β) ,μ δ + λ β}. 24d Proof We prove the statement by induction on k with the induction basis given by Corol- lary 4.11. Assume that for some arbitrary but ﬁxed k ∈ N it holds 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 k k k L L L (29) 30C k −2 −1 (1−γ) −2 −1 ≤ max{(μ δ + λ β) ,μ δ + λ β}. 24d Now for the induction step k → k + 1 we carry out the same argument as above to show 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 k+1 k+1 k+1 L L L k+1 (30) 30C k+1 −2 −1 (1−γ) −2 −1 ≤ max{(μ δ + λ β) ,μ δ + λ β}. 24d We argue as in the induction basis and present the calculations only for the ﬁrst term on the left hand side of (30). By the triangle inequality, f − m (D)f 1 1,μ,λ 1 2 k+1 2 2 ≤ 2f − m (D)f + 2m (D)f − m (D)f . 1 1,μ,λ 1 1,μ,λ 1 1,μ,λ 1 2 2 k L k k+1 L The ﬁrst contribution is already of the desired form. It thus remains to consider the sec- ond contribution. For this we consider an analogous argument as before: Let K := {|k | > k k 4dμλ }, K := {|k | > 4dμλ }. Then, since χ k (D)h (m (D)f ) = 0on K , k 3 k 1,j j,μ,λ j 3 K k j 2 2 m (D)f − m (D)f ≤χ k k (D)m (D)f 1,μ,λ 1 1,μ,λ 1 1,μ,λ 1 2 2 k k+1 L K ∪K k L 2 3 ≤ 2 χ k (D)(m (D)f − h (m (D)f )) 1,μ,λ 1 1,j j,μ,λ j 2 K k k j =2 Scaling for the Two-State Problem and a T Structure Page 35 of 50 5 ≤ 2 m (D)f − h (m (D)f ) 1,μ,λ 1 1,j j,μ,λ j 2 k k j =2 2 2 ≤ 4 [m (D)f − f +f − h (m (D)f ) ] 1,μ,λ 1 1 2 1 1,j j,μ,λ j 2 k k L L j =2 2 2 = 8m (D)f − f + 4 f − h (m (D)f ) . (31) 1,μ,λ 1 1 2 1 1,j j,μ,λ j 2 k k L L j =2 Again for the second right hand side contribution in (31) we use Lemma 4.7: 2 2 1−γ f − h (m (D)f ) ≤ (f − m (D)f ) . 1 1,j j,μ,λ j 2 j j,μ,λ j 2 k k L 24d L Using the inductive hypothesis (29), overall, we arrive at the following upper bound f − m (D)f 1 1,μ,λ 1 2 k+1 2 2 ≤ 2f − m (D)f + 16m (D)f − f 1 1,μ,λ 1 2 1,μ,λ 1 1 2 k L k L 2 1−γ + 8 (f − m (D)f ) j j,μ,λ j 2 24d ≤ 18f − m (D)f 1 1,μ,λ 1 2 1−γ C 30C k k −2 −1 (1−γ) −2 −1 + 16 ( ) max{(μ δ + λ β) ,μ δ + λ β} . 24d 24d γ γ Arguing symmetrically for f and f then yields: 2 3 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 k+1 k+1 k+1 L L L 2 2 2 ≤ 18(f − m (D)f +f − m (D)f +f − m (D)f ) 1 1,μ,λ 1 2 2,μ,λ 2 3 3,μ,λ 3 2 2 2 k L k L k L 1−γ C 30C k −2 −1 (1−γ) −2 −1 + 48 ( ) max{(μ δ + λ β) ,μ δ + λ β} 24d 24d γ γ 30C −2 −1 (1−γ) −2 −1 ≤ 18 max{(μ δ + λ β) ,μ δ + λ β} 24d 1−γ C 30C k k −2 −1 (1−γ) −2 −1 + 48 ( ) max{(μ δ + λ β) ,μ δ + λ β} . 24d 24d γ γ 2 0 Using that 2C ≤ C ,1 − γ ∈ (0, 1) and that ≥ 3, leads to 24d 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 k+1 k+1 k+1 L L L 30C k −2 −1 (1−γ) −2 −1 ≤ 18 max{(μ δ + λ β) ,μ δ + λ β} 24d C 30C k+1 0 0 −2 −1 (1−γ) −2 −1 1−γ + 24 max{(μ δ + λ β) ,(μ δ + λ β) } 24d 24d γ γ 5 Page 36 of 50 B. Raitae ˘ tal. 6C 30C k+1 0 0 k −2 −1 (1−γ) −2 −1 ≤ ( ) max{(μ δ + λ β) ,μ δ + λ β} 24d 24d γ γ 24C 30C k+1 0 0 k −2 −1 (1−γ) −2 −1 + ( ) max{(μ δ + λ β) ,μ δ + λ β} 24d 24d γ γ 30C k+1 k+1 −2 −1 (1−γ) −2 −1 = ( ) max{(μ δ + λ β) ,μ δ + λ β}. 24d This concludes the proof. Now with the inductive procedure of reducing regions of Fourier space concentration in hand, we turn to the proof of the lower bound in Theorem 2 and the argument for Proposi- tion 1.2. Proof of the lower bound in Theorem 2 and proof of Proposition 1.2 We argue in two steps, ﬁrst ﬁxing the free parameters and then exploiting the boundary conditions. Step 1: Choice of parameters and proof of Proposition 1.2. We seek to invoke per Lemma 4.12 expressing the bounds in terms of our energies, i.e. setting δ = E (χ ; F) := el per per −1 −2 inf E (u, χ ), β = E (χ ),and λ = μ . Moreover, we use the notation u∈D F el surf per per per E (χ ; F) := E (χ ; F) + E (χ ). (32) el surf By virtue of Lemma 4.6 we obtain that for f := χ indeed ∂ f ≤ δ. It follows i i,i i i −1 j =1 ˙ directly that also ∇f ≤ β and therefore the conditions in Lemma 4.7 are fulﬁlled. i TV i=1 As a consequence, the above iteration in Lemma 4.12 is applicable. With this in mind, we choose μ = for some α> 0 to be speciﬁed. Therefore, 2α−1 λ = λ = per per per and thus since without loss of generality E (χ ; F) := inf E (u, χ ) + E (χ ) ≤ 1 u∈D F el surf (having the upper bound from Proposition 4.3 in mind), we deduce 30C k 2 k −2α per (1−γ) f − m (D)f ≤ ( ) E (χ ; F) . (33) j j,μ,λ j 2 L 24d j =1 We further choose k k (2+k)α−1 λ = (2 2d + 1μ) λ = M , k 0 where M = M(d) = 2 2d + 1 > 2. This is admissible in the sense of the assumptions in λ 4dμ k 2 Lemma 4.9 since = 2 2d + 1μ ∈ ( , 1). λ 2 k−1 2 1−4μ Next, for α = α() ∈ (0, 1) and > 0 we choose k ∈ N to be given by log M 1 + (1 − 2α)| log |+ log 2 | log | k := ≤ . log M α| log |− log M α − | log | This ensures that λ ≤ . In what follows, we will choose the parameters , α such that log M α 4 ≤ which implies k ≤ . Exploiting the discrete Fourier transform, then yields | log | 2 α 2 2 2 f − m (D)f = |f (ξ )| =f − f , j j,μ,λ j 2 j j j 2 L L ξ ∈Z \{0} Scaling for the Two-State Problem and a T Structure Page 37 of 50 5 and hence, by (33) results in the estimate 30C 0 α 2 −2α per (1−γ) f − f ≤ E (χ ; F) . j j 2 24d j =1 4 α Using that (1 − γ) ≥ 1 − γ ,weset γ := ∈ (0, 1) which leads to the bound α 8 96d 1 2 24d − −2α per f − f ≤ 30C 8 α E (χ ; F) . j j 0 j =1 Next, we ﬁx the parameter α> 0: Observing that for any ν> 0 there exists α > 0such that for α ∈ (0,α ) it holds that 96d 96d − −1 −1 −1−ν −1−ν α = exp 96d log(α )α ≤ exp α = exp(C(ν)α ), (34) νe 2+ν we choose α =| log | for all ∈ (0, ) with > 0 still to be chosen. In particular, for 0 0 log M > 0 sufﬁciently small, such that | log | ≥ 2log M,(34) holds and also ≤ holds | log | 2 as required above. ν 1 As a consequence, for ν := ∈ (0, ) we arrive at 4+2ν 2 1 1 2 +ν per 2 2 f − f ≤ exp(c | log | )E (χ ; F) . j j 2 ν j =1 Step 2: Conclusion. In order to conclude the estimate, we derive a lower bound for f − f . For this we recall that f = χ is the j -th diagonal entry of the phase j j j j,j j =1 2 2 indicator χ and hence f − f =χ − χ . Thus, by the mean value condi- j j 2 2 j =1 L L per tion in D , per 2 2 2 | χ − F | =| χ − u | ≤ |u − χ | dx ≤ E (u, χ ), el and, furthermore, as the left hand side is independent of u, per | χ − F | ≤ E (χ ; F). el per per per Overall this implies for the total energy E (χ ; F) = E (χ ; F) + E (χ ) el surf ˆ ˆ ˆ 2 2 2 2 dist (F , K) ≤ |χ − F | dx ≤ 2 |χ − χ | dx + 2 | χ − F | dx 3 3 3 T T T 1 1 per +ν per 2 2 ≤ 2exp(c | log | )E (χ ; F) + 2E (χ ; F) el 1 1 1 +ν per per 2 2 2 ≤ 2exp(c | log | )E (χ ; F) + 2E (χ ; F) 1 1 +ν per 2 2 ≤ 4exp(c | log | )E (χ ; F) . per Finally, solving for E (χ ; F) shows the desired estimate per −4 +ν 4 E (χ ; F) ≥ 2 exp − 2c | log | dist (F , K). 5 Page 38 of 50 B. Raitae ˘ tal. Fig. A.1 Self-similar construction in Proposition A.3. (Color ﬁgure online) The desired claim follows by an application of Lemma 4.5. Appendix A: Branching, Upper Bound for the Two-State Problem for theDivergenceOperator We complement our lower bounds for the compatible two-well problem from Sect. 3 by an upper bound in the case of the divergence operator acting on matrix ﬁelds as introduced in Example 1.1. For simplicity, we only consider square matrices, i.e. m = d,and likebefore in Sect. 3 only consider the case d> 1 in the following. For earlier, closely related, three- dimensional constructions in the context of compliance minimization problems we refer to [24]. While the construction is by now rather “standard” [1, 62, 63, 80], our argument does provide a slightly different perspective, in that, in arbitrary dimension, we can ensure boundary conditions on all faces of the domain =[0, 1] (see the upper bound construc- tion for [76, Theorem 3] for a similar construction for the gradient). In deducing the upper bound for the divergence operator, we ﬁrst provide a construction in a unit cell (Lemma A.1) and iterate this construction (Proposition A.3). This yields a construction which attains the boundary data in two directions. In order to attain these also on the remaining sides we use the ﬂexibility of the wave cone for the divergence operator (Proposition A.5). We remark that in two dimensions there would be no modiﬁcation with respect to the gradient construction since there the curl and divergence only differ by a rotation of 90 degrees. In our unit cell branching construction, we do not work on the level of the potential, but directly consider the problem on the level of the wells. In this context, we recall the com- patibility conditions for laminates formed by the divergence operator which is determined d×d by the associated wave cone: For M ∈ R ,wehavethat M ∈ ker A(ξ ) if and only it holds Mξ = 0. With this in hand, we introduce an auxiliary matrix which will play an important role in d×d our construction: Let A, B ∈ R be such that (B − A)e = 0and let n = e + γ ν for a 1 1 2 unit vector ν perpendicular to e and for some γ ∈ R, γ = 0. We then deﬁne E by 1 2 2 ν E = γ (B − A)ν ⊗ e (A.1) ν 2 1 Scaling for the Two-State Problem and a T Structure Page 39 of 50 5 and the associated “perturbed” matrix A = A + E . By construction, this matrix obeys the ν ν identities ˜ ˜ (A − A)ν = 0,(B − A )n = 0. (A.2) ν ν ˜ ˜ This in particular allows for interfaces of A and A with normal ν and of A and B with ν ν normal n which we will use in our branching construction below. d−2 d×d Lemma A.1 For 0 <l <h ≤ 1, we deﬁne ω =[0,l]×[0,h]×[0, 1] . Let A, B ∈ R be such that (B − A)e = 0 and let F = λA + (1 − λ)B for some λ ∈ (0, 1). Then there exists 1 λ d d×d u : R → R such that div u = 0 in R , u = F for x ∈ (−∞, 0) ∪ (l, ∞). λ 1 Furthermore, there exist χ ∈ BV (ω;{A, B}) and a constant C = C(A, B) > 0 such that for any > 0 the localized energy can be bounded by ˆ ˆ 2 2 E (u, χ ; ω) := |u − χ | dx + |∇χ|≤ C(1 − λ) + 5h. ω ω Proof We consider the following partition of the domain ω into subdomains: λl λl λl (1 − λ)l ω ={x ∈ (0, )},ω ={x ∈ ( , + x )}, 1 1 2 1 2 2 2 2 2h λl (1 − λ)l (1 − λ)l (1 − λ)l ω ={x ∈ ( + x ,λl + x )},ω ={x ∈ (λl + x ,l)}. 3 1 2 2 4 1 2 2 2h 2h 2h Based on this we deﬁne Ax ∈ ω , Ax ∈ ω ∪ ω , 1 3 u(x) = Bx ∈ ω ∪ ω , χ(x) = 2 4 Bx ∈ ω ∪ ω , ⎩ 2 4 A + E x ∈ ω , e 3 (1−λ)l where E is givenin(A.1)for n = e − e . We highlight that u is independent of x e 1 2 k 2h for k ≥ 3. By deﬁnition of E , the characterization of the wave cone (5) for the divergence operator and the remarks on laminates in (A.2), this deﬁnes an divergence-free mapping. Further, as (B − F )e = (A − F )e = 0, the exterior data are attained in x ∈ (−∞, 0) ∪ λ 1 λ 1 1 (l, ∞). To calculate the energy, we observe, that the only contribution to the elastic energy is given in ω . Hence, ˆ ˆ 2 2 E (u, χ ; ω) := |u − χ | dx = |A + E − A| dx el e ω ω 2 3 (1 − λ) λl 2 2 =|E | |ω |=|(B − A)e | . e 3 2 8h As the surface energy is determined by the interfaces between ω and ω ,weobtain(l< h) j k " " 2 2 2 (1 − λ) l (1 − λ) E (χ ; ω) := |∇χ|= h + 2 + h ≤ h + 2h + 1 ≤ 5h. surf 4 4 ω 5 Page 40 of 50 B. Raitae ˘ tal. Fig. A.2 Setting of Lemma A.2 This shows the claim. As for analogous constructions for the gradient, we will use this unit cell as a building block in order to achieve a self-similar construction attaining the boundary data on two directions. For this to be admissible in the sense of an A-free map, we rely on the following lemma. It shows that for ﬁrst order operators corners in which interfaces meet do not give rise to singularities. ∞ d n ∞ d m Lemma A.2 Let d, n, m ∈ N and let A(D) = A ∂ : C (R ; R ) → C (R ; R ) be j j j =1 a homogeneous, linear operator of degree one with symbol A given in (3) and let , j = 1,...,l, be a polygonal set (the set is deﬁned as the intersection of half spaces) with outer unit normal n such that • R = , j =1 • the two sets , have one common face (l + 1 = 1), j j +1 • and such that they meet in one point, i.e. ={x }, j 0 j =1 cf. Fig. A.2. Assume further that B ∈ R , j = 1,...,l, are such that B − B ∈ ker A(n ). j j j +1 j Then the map u(x) = B for x ∈ j j is A-free. Proof First we note that u is indeed well-deﬁned by the properties of , further we notice n d ∞ d m that for M ∈ R and U ⊂ R a Lipschitz domain it holds for ϕ ∈ C (R ; R ) ˆ ˆ ˆ d d ∗ t t d−1 M · (A(D)) ϕdx = M · ∂ (A ϕ)dx = M · (A ϕ)n dH k k k k U U ∂U k=1 k=1 d−1 = A(n)M · ϕdH . ∂U With this it holds ˆ ˆ ˆ l l ∗ ∗ d−1 u · (A(D)) ϕdx = B · (A(D)) ϕdx = A(n )B · ϕdH . j j j R ∂ j j j =1 j =1 Scaling for the Two-State Problem and a T Structure Page 41 of 50 5 Moreover on ∂ ∩ ∂ it holds that n =−n and thus by the assumptions on B j j +1 j +1 j j ˆ l ˆ ∗ d−1 u · (A(D)) ϕdx = A(n )(B − B ) · ϕdH = 0. j j j +1 R ∂ ∩∂ j j +1 j =1 As ϕ was arbitrary the claim follows. With Lemma A.2 in hand, we now iterate the unit cell-construction from Lemma A.1,as illustrated in Fig. A.1. d n Proposition A.3 Let d, n ∈ N. Let =[0, 1] , let A, B ∈ R be such that B − A ∈ , cf. div (5), and let F = λA + (1 − λ)B for some λ ∈ (0, 1). Let E be as in (12). Then there exist u : d d×d d 2 R → R and χ ∈ BV (;{A, B}) with div u = 0 in R and u = F for (x ,x )/ ∈[0, 1] λ 1 2 such that for any ∈ (0, 1) and any N ∈ N E (u, χ ) ≤ C( + N ) for some constant C = C(A, B, λ) > 0. Proof Without loss of generality we may assume (B − A)e = 0, i.e. B − A ∈ ker A(e ) for 1 1 1 1 A(D) = div. Let θ ∈ ( , ). We argue symmetrically in the upper and lower half of the cube, 4 2 1 d−2 i.e. we give the construction of u on [0, 1]×[ , 1]×[0, 1] and deﬁne u on the lower half by symmetry. We deﬁne for N ∈ N and for j ∈ N θ 1 1 − θ y = 1 − ,l = ,h = y − y = θ . j j j j +1 j 2 2 N 2 Furthermore, let j ∈ N be the maximal j ∈ N such that l <h .Weset 0 j j d−2 ω = (kl ,y ) +[0,l ]×[0,h ] ×[0, 1] , j,k j j j j j j +1 for k ∈{0, 1,...,N 2 − 1}, j ∈{0, 1,...,j };for k ∈{0, 1,...,N 2 − 1}, j = j + 1we 0 0 set j +1 d−2 ω = (kl ,y ) +[0,l ]×[0, ] ×[0, 1] . j +1,k j +1 j +1 j +1 0 0 0 0 d−2 Let u , χ in [0,l ]×[0,h ]×[0, 1] be given by Lemma A.1 for j = 1,...,j .Further, j j j j 0 in the layer j = j + 1 we interpolate with the desired boundary data by a cut-off argument: To this end, we introduce the cut-off function φ :[0, ∞) →[0, 1] and the proﬁle h :[0, 1]→ [0, ∞) by setting ⎪ 1 t ∈[0, ], (1 − λ)t t ∈[0,λ), 1 3 φ(t) = −4t + 3 t ∈ ( , ), h(t ) = 2 4 λ(1 − t) t ∈[λ, 1]. 0 t ≥ , j +1 d−2 We consider the function u ˜ :[0,l ]×[0, ]×[0, 1] deﬁned via j +1 j 0 0 2 2l 2x x j +1 2 1 u ˜ (x) =− φ h ((A − B)e ) ⊗ e j +1 2 1 j +1 j +1 0 0 θ θ l j +1 0 5 Page 42 of 50 B. Raitae ˘ tal. 2x x 2 1 + φ h (A − B) + F . j +1 θ l j +1 The associated phase indicator is deﬁned by χ (x) = χ (x )A + χ (x )B. j +1 (0,λl ) 1 (λl ,l ) 1 0 j j j 0 0 0 We note that χ (x) = h ( )(A − B) + F and moreover for x < it holds u ˜ (x) = j +1 λ 2 j +1 0 l 2 0 j +1 χ (x) and for x > correspondingly u ˜ (x) = F . Furthermore, for x ∈{0,l },we j +1 2 j +1 λ 1 j +1 0 0 0 x 2x x 1 2 1 know h( ) = 0 and thus (u ˜ (x) − F )e = φ( )h (A − B)e = 0. j +1 λ 1 1 0 j +1 l 0 l j +1 θ j +1 0 0 With the help of this construction we meet the prescribed data for x ≥ 1, and we can deﬁne u in the upper half of the full cube: u (x − (kl ,y )) x ∈ ω , j j j j,k u(x) = u ˜ (x − (kl ,y )) x ∈ ω . j +1 j j +1 j +1,k 0 0 0 0 For the lower half of the cube we argue similarly, mirroring the unit cell construction of Lemma A.1, i.e. instead of E we consider E .Wedeﬁne χ in [0, 1] analogously. e −e 2 2 We note, that this deﬁnes a divergence free mapping, as all the laminations are in com- patible directions as (B − A)e = 0 and by the choice of E in (A.1). Lemma A.2 shows, 1 e that this function is divergence-free even thought interfaces meet in corners. Moreover, we j +1 can bound the energy in the ω cells for any k ∈{1,...,N 2 − 1}: j +1,k 2l 2x x j +1 2 1 2 0 |˜ u (x) − χ (x)| = φ h ((A − B)e ⊗ e ) j +1 j +1 2 1 0 0 j +1 j +1 0 0 θ θ l j +1 2x x 2 1 − (φ − 1)h (A − B) j +1 θ l j +1 2x j +1 2 0 2 ≤ C(A, B, λ) φ + 1 , 2j +2 j +1 0 0 θ θ j +1 and, since l ≥ h and θ ∼ h , j +1 j +1 j +1 0 0 0 ˆ ˆ j +1 2 0 2 j +1 |˜ u − χ (x)| dx ≤ Cl ( φ (t )dt + θ ) j +1 j +1 j +1 0 0 0 j +1 j +1 θ 0 0 d−2 θ [0,l ]×[0, ]×[0,1] 0 j +1 0 2 j +1 0 j +1 ≤ C( + l θ ) j +1 j +1 j +1 ≤ C . j +1 Furthermore, the surface energy is bounded by E (χ ; ω ) ≤ C(A, B, λ)h . surf j +1 j +1,k j +1 0 0 0 and can extend it to be F for (x ,x )/ ∈ Overall, we have a function deﬁned on [0, 1] λ 1 2 [0, 1] . The energy then can be bounded by j +1 j +1 0 N 2 −1 0 3 E (u, χ ) ≤ 2 E (u ,χ ; ω ) ≤ C N 2 ( + h ) j j j,k j j =0 k=0 j =0 Scaling for the Two-State Problem and a T Structure Page 43 of 50 5 j +1 j +1 0 2 0 h 1 1 j j j j ≤ C ( + ) = C ( ) + N (2θ) h l N 4θ j j j =0 j =0 ≤ C( + N ). Remark A.4 We remark that in the situation of the divergence operator, there are situations d×d with substantially more ﬂexibility than for the gradient: If for the two wells A, B ∈ R it does not only hold that (B − A)e = 0 but also that (B − A)e = 0, there would not be 1 2 any elastic energy contribution involved. In this situation, for the above construction, we would only have contributions to the surface energy, as then E = 0 and hence A + E = e e 2 2 A ∈{A, B}. In particular, for boundary data which are only attained on two directions this would yield a linear scaling law in . For curl free mappings as in gradient inclusions, this is not possible, as the direction of lamination is unique in that case, i.e. V is at most rot,λ one-dimensional. As a last auxiliary step towards the upper bound construction from Theorem 1,inorder to achieve the exterior data on all sides of the unit cube, we adapt the branching construction similarly as in [75], as the construction from Proposition A.3 does not yet satisfy F at, e.g., x = 0. Thus, we combine Proposition A.3 with a further domain splitting for which we split [0, 1] into different regions. In each region, we prescribe a different direction for the branching construction from Proposition A.3. To this end, we use that the choice of e in the above results was arbitrary and we also can choose any other direction e for j ∈{2,...,d}. Combined with compatibility conditions at the resulting interfaces, this will allow us to deduce the desired branching construction. Proposition A.5 Under the same assumptions as in Proposition A.3 there exist u : R → d×d R , χ ∈ BV (;{A, B}) and a constant C = C(A, B, λ) > 0 such that u ∈ D for F = F λ λA + (1 − λ)B (λ ∈ (0, 1)) and for any ∈ (0, 1) and N ∈ N it holds E (u, χ ) ≤ C( + N ). Proof For simplicity, we ﬁrst carry out the details for the case d = 3 and then only comment on the changes in the case of arbitrary dimension. We split [0, 1] into the following four parts and use different branching directions in each part: Let 1 1 1 1 1 ± 3 ={x ∈[0, 1] :±(x − ) ≥ 0, −|x − |≤ −|x − |}, 2 2 3 2 2 2 2 2 1 1 1 1 1 ={x ∈[0, 1] :±(x − ) ≥ 0, −|x − |≤ −|x − |}, 3 3 2 2 2 2 2 2 + − and consider the upper ( )and lower( ) halves separately as in the proof of Proposi- j j tion A.3. + + Next, we deﬁne u in and u in using Proposition A.3: The function u is given 2 3 2 2 3 by the function from Proposition A.3 above, whereas u is obtained from u by exchanging 3 2 roles of e and e , i.e. the branching is done in e direction and we use E instead of E . 2 3 3 e e 3 2 For ν ∈{e ,e } the error matrix E is givenin(A.1). We then deﬁne the overall deformation 2 3 ν + + u in ∪ by 2 3 u (x) x ∈ , u(x) = u (x) x ∈ . 3 5 Page 44 of 50 B. Raitae ˘ tal. Fig. A.3 The three-dimensional branching construction to achieve the boundary data on all sides. The shaded regions are the diagonal interface at x = x 2 3 with the interfaces of A + E and A + E marked in orange − − This construction is depicted in Fig. A.3. As above u is deﬁned in the lower halves , 2 3 by symmetry. We claim that this overall construction is divergence-free. In the individual regions and this follows by Proposition A.3. It thus remains to discuss the compatibility at the interface x = x . Since all other values of u are given by (matching domains in which) 2 3 u ∈{A, B}, it sufﬁces to discuss the compatibility of the error matrices E and E at this e e 2 3 interface. To this end, we however note that (E − E )ζ = 0for all ζ ∈ span(e ,e ), i.e. e e 2 3 3 2 also this interface is admissible. This shows that u indeed deﬁnes a divergence free map. The upper bound for the elastic and surface energies from Proposition A.3 remains valid, thus yielding the claimed estimate which concludes the proof of the proposition. In order to show the d -dimensional result, we split [0, 1] into 2d − 2 regions := d 1 1 1 1 1 {x ∈[0, 1] :±(x − ) ≥ 0, −|x − |= min −|x − |} for j = 2,...,d and j j 2≤k≤d k 2 2 2 2 2 argue as above. Finally, with Proposition A.5 in hand, we immediately obtain the proof of the upper bound construction from Theorem 1 for the divergence operator. Proof of the upper bound in Theorem 1 In order to deduce the upper bound of Theorem 1,we choose N ∼ , which shows the claim. Remark A.6 (Generalizations) Building on the ideas from the gradient case and the ones from above one can formulate (rather restrictive) conditions, allowing for similar constructions for more general linear, constant coefﬁcient differential operators. A key difﬁculty here consists in the “hard form” of the prescribed boundary conditions. When considering “softer forms” of these, as for instance in [66], constructions for general constant coefﬁcient operators with the desired boundary conditions would be feasible under much more general conditions by using Fourier theoretic arguments as in [74]. We do not pursue these ideas here but postpone this to possible future work. Scaling for the Two-State Problem and a T Structure Page 45 of 50 5 Appendix B: On the Role of the Divergence Operator Following [28], in this section we highlight the relevance of the divergence operator which also partially motivates our discussion of the scaling law for the T problem from Sect. 1.3. To this end, we ﬁrst recall that considering any ﬁrst order homogeneous constant co- j ∞ d n ∞ d m efﬁcient differential operator A(D) = A ∂ : C (R ; R ) → C (R ; R ), we can j =1 n m×d rewrite A(D)u = div ω (u) with a linear map ω : R → R . Indeed, let us deﬁne 1 1 n m×d ω : R → R ,ω (x) = A x . 1 1 k i=1,...,m, ik j =1,...,d k=1 ∞ d n Let u ∈ C (R ; R ), then it holds A(D)u = (div ◦ω )(u) for the row-wise divergence. α α Indeed this can be generalized for higher order operators A(D)u = A ∂ u,where |α|=k α m×n A ∈ R are coefﬁcient matrices, as follows. For this we denote the space of symmetric k d d tensors on R by Sym(R ,k).Let the k-th order divergence be given as k ∞ d m d ∞ d m div : C (R ; R ⊗ Sym(R ,k)) → C (R ; R ), (B.3) (div u) := ∂ ...∂ u . j i i ji ...i 1 k 1 k 1≤i ≤···≤i ≤d 1 k For k = 1 this is exactly the row-wise divergence as mentioned above. Remark B.1 This deﬁnition is natural in that sense that this operator (up to a sign) is the adjoint of the k-th derivative D . n m d The linear map ω : R → R ⊗ Sym(R ,k) then takes the form l=1 (ω (x)) := A x (B.4) k ji ...i 1 k ∞ d n and by this choice it holds A(D)u = (div ◦ω )(u) for any u ∈ C (R ; R ) and ker ω = k k I = ker A . In what follows, we omit the k dependence of ω = ω in the notation. A k |α|=k With this in hand, it is possible to bound the energy for a general homogeneous linear operator A(D) (of order k) by the corresponding energy for the (k-th order) divergence (cf. [28, Appendix] for the corresponding qualitative result in the case k = 1). n d Proposition B.2 Let d, n, k ∈ N, K ⊂ R , and let ⊂ R be a bounded Lipschitz domain. Let A(D) be a k-th order homogeneous linear differential as in (2) and the elastic and sur- face energies be given by (16) and (17). Moreover let ω = ω be the linear transformation in (B.4) and div the generalized k-th order divergence in (B.3) with the corresponding en- k k div div ergies E , E . Then there exist constants C ,C > 0 such that for any χ ∈ BV (; K), 1 2 el surf F ∈ R div E (χ ; F) ≥ C E (ω(χ ); ω(F )), el 1 el div E (χ ) ≥ C E (ω(χ )). surf 2 surf Moreover if A(D) is cocanceling (and thus ω is injective), it also holds for all u ∈ D , χ ∈ BV (; K) (cf.(10)) k k div div E (u, χ ) + E (χ ) ∼ E (ω(u), ω(χ )) + E (ω(χ )). el surf surf 5 Page 46 of 50 B. Raitae ˘ tal. Proof In order to obtain the desired result, we use the pointwise bound |u − χ|≥ C|ω(u) − ω(χ)| and consider the adapted boundary data: For any u ∈ D with D denoting the set F F k d from (10), the composition ω(u) satisﬁes div ω(u) = A(D)u = 0in R and ω(u) = ω(F ) d div in R \ . In other words, it holds that ω(u) ∈ D for the divergence operator and bound- ω(F ) k k div div ary data ω(F ). For the elastic energy (denoting by E , D the energy and domain for el ω(F ) the divergence operator) this implies ˆ ˆ 2 2 div E (u, χ ) = |u − χ | dx ≥ C |ω(u) − ω(χ)| dx = CE (ω(u), ω(χ )), (B.5) el el ˆ ˆ 2 2 E (χ ; F) ≥ C inf |ω(u) − ω(χ)| dx ≥ C inf |ω(u) − ω(χ)| dx. el u∈D F div u:ω(u)∈D ω(F ) We emphasize that, in general, this only yields lower bound inequalities since ω is possibly not injective and thus there may be deformations u with A(D)u = 0and u = F outside but still fulﬁlling ω(u) = ω(F ) outside (see the example in Sect. B.1 below). Replacing d m d div now ω(u) by a general function w : R → R ⊗ Sym(R ,k) such that w ∈ D yields ω(F ) 2 div E (χ ; F) ≥ C inf |w − ω(χ)| dx = CE (ω(χ ); ω(F )). el el div w∈D ω(F ) Furthermore, as |∇χ|≥ c|∇(ω(χ ))|, we can also bound the surface energy ˆ ˆ div E (χ ) = |∇χ|≥ c |∇(ω(χ ))|= cE (ω(χ )). (B.6) surf surf In the case of a cocanceling operator ω is injective and we also have the bounds |u − χ|≤ C|ω(u) − ω(χ)|, |∇χ|≤ C|∇(ω(χ ))|, thus in (B.5)and (B.6) also the matching upper bounds hold, which concludes the proof. As a consequence, lower bounds for the divergence operator often also imply lower bounds for more general operators. A particular setting (see [28]) for instance arises in the three state problem with K ={A ,A ,A } being such that A − A ∈ / . In this case also 1 2 3 j k A ω(K) consists of three states which is a result of the fact that the kernel of ω is given by d d ker(ω) = ker A = ker A(e ) = I . j A j =1 j =1 In particular, if we ﬁnd a T structure for a general linear, homogeneous, constant coefﬁcient, ﬁrst order differential operator A(D) such that it is mapped to the T structure in Sect. 4,we can exploit the same lower bound as for the divergence operator. In addition to the relevance of the divergence operator for applications, this argument serves as an additional motivation for focusing particularly on the divergence operator in this article. Moreover with Proposition 3.8 in mind, also for pairwise non super-compatible wells, we can assume without loss of generality that I ={0} and thus ω in injective. B.1 Comparison of the Two-State Problem for the Divergence Operator In this section, we discuss the comparison between the general two-state problem for linear, homogeneous differential operators and the one for the (k-th order) divergence operator. In particular, this yields yet another proof of the compatible case in Theorem 1. Scaling for the Two-State Problem and a T Structure Page 47 of 50 5 In the calculations from the ﬁrst part of Sect. B, we notice that for an injective map ω we can also bound the quantities E (u, χ ), E (χ ) from above with the corresponding term el surf in which u, χ are replaced by ω(u), ω(χ); hence for I ={0} it holds div E (u, χ ) ∼ E (ω(u), ω(χ )). We here emphasize that this only holds on the level of ﬁxed u, χ and that after the mini- mization in u, this does not necessarily yield a two-sided comparison of the energies any more. Indeed, while the lower bound estimates always hold (cf. Proposition B.2), this may not be true for the upper bound estimates. In fact, even if I ={0}, we can at the moment div not exclude that there may be w ∈ D \ ω(D ). We postpone a further discussion of this ω(F ) to future work. The advantage of I ={0} is that we do not lose wells in that sense that for I ={0} A A also ker ω ={0} and thus, ω is injective. As seen above in Sect. 3.4 for two wells A, B ∈ R , A − B/ ∈ I we can restrict to A(D) which fulﬁlls I ={0}. This implies that for A ˜ two compatible wells, which are not super-compatible, in deducing lower scaling bounds, we can use the corresponding lower bounds of the divergence operator as we do not lose information. Example B.3 In concluding this section, we give an example of an operator which is not cocanceling. Considering d = 2, n = 3, m = 1and A(D)u = ∂ u + ∂ u , 1 2 2 3 3 1×2 implies that ω : R → R , ω(x) = (x ,x ) and ker(ω) = span(e ) = I . 2 3 1 A 2 3 The reduced operator A(D) would act on mappings taking values only in {0}× R ⊂ R . Acknowledgements A.R. and C.T. gratefully acknowledge support through the Heidelberg STRUCTURES Excellence Cluster which is funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foun- dation) under Germany’s Excellence Strategy EXC 2181/1 - 390900948. Funding Note Open Access funding enabled and organized by Projekt DEAL. Data Availability Data sharing not applicable to this article as no datasets were generated or analysed during the current study. Declarations Competing Interests The authors have no competing interests to declare that are relevant to the content of this article. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. 5 Page 48 of 50 B. Raitae ˘ tal. References 1. Chan, A., Conti, S.: Energy scaling and branched microstructures in a model for shape-memory alloys with SO(2) invariance. Math. Models Methods Appl. Sci. 25(06), 1091–1124 (2015) 2. Rüland, A., Tribuzio, A.: On the energy scaling behaviour of a singularly perturbed Tartar square. Arch. Ration. Mech. Anal. 243(1), 401–431 (2022) 3. Garroni, A., Nesi, V.: Rigidity and lack of rigidity for solenoidal matrix ﬁelds. Proc. R. Soc. Lond., Ser. A, Math. Phys. Eng. Sci. 460(2046), 1789–1806 (2004). https://doi.org/10.1098/rspa.2003.1249 4. Palombaro, M., Ponsiglione, M.: The three divergence free matrix ﬁelds problem. Asymptot. Anal. 40(1), 37–49 (2004) 5. Müller, S., Šverák, V.: Convex integration for Lipschitz mappings and counterexamples to regularity. Ann. Math. 157(3), 715–742 (2003) 6. Kirchheim, B., Müller, S., Šverák, V.: Studying nonlinear PDE by geometry in matrix space. In: Geo- metric Analysis and Nonlinear Partial Differential Equations, pp. 347–395. Springer, Berlin (2003) 7. Székelyhidi, L. Jr: The regularity of critical points of polyconvex functionals. Arch. Ration. Mech. Anal. 172(1), 133–152 (2004) 8. De Lellis, C., Székelyhidi, L. Jr.: The Euler equations as a differential inclusion. Ann. Math. 170(3), 1417–1436 (2009) 9. De Lellis, C., Székelyhidi, L. Jr.: The h-principle and the equations of ﬂuid dynamics. Bull. Am. Math. Soc. 49(3), 347–375 (2012) 10. Kuiper, N.H.: On C -isometric imbeddings. II. In: Indagationes Mathematicae (Proceedings), vol. 58, pp. 683–689. Elsevier, Amsterdam (1955) 11. Nash, J.: C isometric imbeddings. Ann. Math. 60(3), 383–396 (1954) 1,α 12. Conti, S., De Lellis, C., Székelyhidi, L. Jr.: h-principle and rigidity for C isometric embeddings. In: Nonlinear Partial Differential Equations, pp. 83–116. Springer, Berlin (2012) 13. De Lellis, C., Székelyhidi, L. Jr.: On h-principle and Onsager’s conjecture. Newsl. - Eur. Math. Soc. 95, 19–24 (2015) 14. Ball, J.M., James, R.D.: Fine phase mixtures as minimizers of energy. In: Analysis and Continuum Mechanics, pp. 647–686. Springer, Berlin (1989) 15. Ball, J.M., James, R.D.: Proposed experimental tests of a theory of ﬁne microstructure and the two-well problem. Philos. Trans. R. Soc. Lond. A, Math. Phys. Eng. Sci. 338(1650), 389–450 (1992) 16. Bhattacharya, K., Firoozye, N.B., James, R.D., Kohn, R.V.: Restrictions on microstructure. Proc. R. Soc. Edinb., Sect. A, Math. 124(5), 843–878 (1994) 17. Müller, S., Šverák, V.: Convex integration with constraints and applications to phase transitions and partial differential equations. J. Eur. Math. Soc. 1(4), 393–422 (1999) 18. Müller, S.: Variational models for microstructure and phase transitions. In: Calculus of Variations and Geometric Evolution Problems, pp. 85–210. Springer, Berlin (1999) 19. Rüland, A.: The cubic-to-orthorhombic phase transition: rigidity and non-rigidity properties in the linear theory of elasticity. Arch. Ration. Mech. Anal. 221(1), 23–106 (2016) 20. Palombaro, M.: Rank-(n − 1) convexity and quasiconvexity for divergence free ﬁelds. Adv. Calc. Var. 3(3), 279–285 (2010) 21. Palombaro, M., Smyshlyaev, V.P.: Relaxation of three solenoidal wells and characterization of extremal three-phase H-measures. Arch. Ration. Mech. Anal. 194(3), 775–822 (2009) 22. Kohn, R.V., Wirth, B.: Optimal ﬁne-scale structures in compliance minimization for a uniaxial load. Proc. R. Soc. A, Math. Phys. Eng. Sci. 470(2170), 20140432 (2014) 23. Kohn, R.V., Wirth, B.: Optimal ﬁne-scale structures in compliance minimization for a shear load. Com- mun. Pure Appl. Math. 69(8), 1572–1610 (2016) 24. Potthoff, J., Wirth, B.: Optimal ﬁne-scale structures in compliance minimization for a uniaxial load in three space dimensions (2021). arXiv:2111.06910 25. Kohn, R.V., DeSimone, A., Otto, F., Müller, S.: Recent analytical developments in micromagnetics. Sci. Hyst. 2, 269–381 (2006) 26. DeSimone, A., Knüpfer, H., Otto, F.: 2-d stability of the Néel wall. Calc. Var. Partial Differ. Equ. 27(2), 233–253 (2006) 27. De Philippis, G., Palmieri, L., Rindler, F.: On the two-state problem for general differential operators. Nonlinear Anal. 177, 387–396 (2018) 28. Sorella, M., Tione, R.: The four-state problem and convex integration for linear differential operators (2021). arXiv:2107.10785 29. Skipper, J.W.D., Wiedemann, E.: Lower semi-continuity for A-quasiconvex functionals under convex restrictions (2019). arXiv:1909.11543 30. Bhattacharya, K.: Microstructure of Martensite. Oxford Series on Materials Modelling. Oxford Univer- sity Press, London (2003) Scaling for the Two-State Problem and a T Structure Page 49 of 50 5 31. Bhattacharya, K.: Comparison of the geometrically nonlinear and linear theories of martensitic transfor- mation. Contin. Mech. Thermodyn. 5(3), 205–242 (1993) 32. Warner, M., Terentjev, E.M.: Liquid Crystal Elastomers, vol. 120. Oxford University Press, London (2007) 33. Cesana, P., Della Porta, F., Rüland, A., Zillinger, C., Zwicknagl, B.: Exact constructions in the (non- linear) planar theory of elasticity: from elastic crystals to nematic elastomers. Arch. Ration. Mech. Anal. 237(1), 383–445 (2020) 34. Lamy, X., Lorent, A., Peng, G.: Rigidity of a non-elliptic differential inclusion related to the Aviles–Giga conjecture. Arch. Ration. Mech. Anal. 238(1), 383–413 (2020) 35. Lamy, X., Lorent, A., Peng, G.: On a generalized Aviles-Giga functional: compactness, zero-energy states, regularity estimates and energy bounds (2022). arXiv:2203.05418 36. Tartar, L.: Compensated compactness and applications to partial differential equations. In: Nonlinear Analysis and Mechanics: Heriot-Watt Symposium, vol. 4, pp. 136–212 (1979) 37. DiPerna, R.J.: Compensated compactness and general systems of conservation laws. Trans. Am. Math. Soc. 292(2), 383–420 (1985) 38. Tartar, L.: The compensated compactness method applied to systems of conservation laws. In: Systems of Nonlinear Partial Differential Equations, pp. 263–285. Springer, Berlin (1983) 39. Murat, F., Tartar, L.: H-convergence. In: Topics in the Mathematical Modelling of Composite Materials, pp. 21–43. Springer, Berlin (2018) 40. Müller, S.: Rank-one convexity implies quasiconvexity on diagonal matrices. Int. Math. Res. Not. 1999(20), 1087–1095 (1999) 2×2 41. Faraco, D., Székelyhidi, L. Jr.: Tartar’s conjecture and localization of the quasiconvex hull in R .Acta Math. 200(2), 279–305 (2008) 42. Guerra, A., Rai¸ ta, B., Schrecker, M.R.I.: Compensated compactness: continuity in optimal weak topolo- gies. J. Funct. Anal. 283(7), 109596 (2022) 43. Arroyo-Rabasa, A., De Philippis, G., Rindler, F.: Lower semicontinuity and relaxation of linear-growth integral functionals under PDE constraints. Adv. Calc. Var. 13(3), 219–255 (2020) 44. Rai¸ ta, ˘ B.: Potentials for A-quasiconvexity. Calc. Var. Partial Differ. Equ. 58(3), 105 (2019) 45. Guerra, A., Rai¸ ta, ˘ B., Schrecker, M.: Compensation phenomena for concentration effects via nonlinear elliptic estimates (2021). arXiv:2112.10657 46. Rai¸ ta, ˘ B.: A simple construction of potential operators for compensated compactness (2021). arXiv:2112. 47. Behn, L., Gmeineder, F., Schiffer, S.: On symmetric div-quasiconvex hulls and divsym-free L - truncations (2021). arXiv:2108.05757 48. Fonseca, I., Müller, S.: A-Quasiconvexity, lower semicontinuity, and young measures. SIAM J. Math. Anal. 30(6), 1355–1390 (1999) 49. Kristensen, J., Rai¸ ta, ˘ B.: An introduction to generalized Young measures. Max-Planck-Institut für Math- ematik in den Naturwissenschaften Leipzig 45 (2020) 50. Conti, S., Gmeineder, F.: A-Quasiconvexity and Partial Regularity (2020). arXiv:2009.13820 51. Gmeineder, F., Lewintan, P., Neff, P.: Optimal incompatible Korn-Maxwell-Sobolev inequalities in all dimensions (2022). arXiv:2206.10373 52. De Philippis, G., Rindler, F.: On the structure of A-free measures and applications. Ann. Math. 184(3), 1017–1039 (2016) 53. Arroyo-Rabasa, A., De Philippis, G., Hirsch, J., Rindler, F.: Dimensional estimates and rectiﬁability for measures satisfying linear PDE constraints. Geom. Funct. Anal. 29(3), 639–658 (2019) 54. Breit, D., Diening, L., Gmeineder, F.: On the trace operator for functions of bounded A-variation. Anal. PDE 13(2), 559–594 (2020) 55. Dacorogna, B.: Direct Methods in the Calculus of Variations, vol. 78. Springer, Berlin (2007) 56. Dacorogna, B., Marcellini, P.: Implicit Partial Differential Equations, vol. 37. Springer, Berlin (2012) 57. Pedregal, P.: Parametrized Measures and Variational Principles. Springer, Berlin (1997) 58. Chaudhuri, N., Müller, S.: Rigidity estimate for two incompatible wells. Calc. Var. Partial Differ. Equ. 19(4), 379–390 (2004) 59. De Lellis, C., Székelyhidi, L. Jr.: Simple proof of two-well rigidity. C. R. Math. 343(5), 367–370 (2006) 60. Friesecke, G., James, R.D., Müller, S.: A theorem on geometric rigidity and the derivation of nonlinear plate theory from three-dimensional elasticity. Commun. Pure Appl. Math. 55(11), 1461–1506 (2002) 61. Lamy, X., Lorent, A., Peng, G.: Quantitative rigidity of differential inclusions in two dimensions (2022). arXiv:2208.08526 62. Kohn, R.V., Müller, S.: Branching of twins near an austenite—twinned-martensite interface. Philos. Mag. A 66(5), 697–715 (1992) 63. Kohn, R.V., Müller, S.: Surface energy and microstructure in coherent phase transitions. Commun. Pure Appl. Math. 47(4), 405–435 (1994) 5 Page 50 of 50 B. Raitae ˘ tal. 64. Barroso, A.C., Matias, J., Santos, P.M.: Differential inclusions and A-quasiconvexity. Mediterr. J. Math. 3(14), 1–14 (2017) 65. Van Schaftingen, J.: Limiting sobolev inequalities for vector ﬁelds and canceling linear differential op- erators. J. Eur. Math. Soc. 15(3), 877–921 (2013) 66. Choksi, R., Kohn, R.V., Otto, F.: Domain branching in uniaxial ferromagnets: a scaling law for the minimum energy. Commun. Math. Phys. 201(1), 61–79 (1999) 67. Conti, S., Diermeier, J., Melching, D., Zwicknagl, B.: Energy scaling laws for geometrically linear elas- ticity models for microstructures in shape memory alloys. ESAIM Control Optim. Calc. Var. 26, 115 (2020) 68. Kohn, R.V.: Energy-driven pattern formation. In: International Congress of Mathematicians, vol. 1, pp. 359–383 (2007) 69. Conti, S., Kohn, R.V., Misiats, O.: Energy minimizing twinning with variable volume fraction, for two nonlinear elastic phases with a single rank-one connection. Math. Models Methods Appl. Sci. 32(08), 1671–1723 (2022). https://doi.org/10.1142/S0218202522500397 70. Knüpfer, H., Kohn, R.V.: Minimal energy for elastic inclusions. Proc. R. Soc. A, Math. Phys. Eng. Sci. 467(2127), 695–717 (2011) 71. Knüpfer, H., Kohn, R.V., Otto, F.: Nucleation barriers for the cubic-to-tetragonal phase transformation. Commun. Pure Appl. Math. 66(6), 867–904 (2013) 72. Rüland, A.: A rigidity result for a reduced model of a cubic-to-orthorhombic phase transition in the geometrically linear theory of elasticity. J. Elast. 123(2), 137–177 (2016) 73. Capella, A., Otto, F.: A quantitative rigidity result for the cubic-to-tetragonal phase transition in the geometrically linear theory with interfacial energy. Proc. R. Soc. Edinb., Sect. A, Math. 142(2), 273–327 (2012) 74. Capella, A., Otto, F.: A rigidity result for a perturbation of the geometrically linear three-well problem. Commun. Pure Appl. Math. 62(12), 1632–1669 (2009) 75. Rüland, A., Tribuzio, A.: On the energy scaling behaviour of singular perturbation models involving higher order laminates (2021). arXiv:2110.15929 76. Rüland, A., Tribuzio, A.: On scaling laws for multi-well nucleation problems without gauge invariances (2022). arXiv:2206.05164 77. Rüland, A., Taylor, J.M., Zillinger, C.: Convex integration arising in the modelling of shape-memory alloys: some remarks on rigidity, ﬂexibility and some numerical implementations. J. Nonlinear Sci. 29(5), 2137–2184 (2019) 78. Tartar, L.: Some remarks on separately convex functions. In: Microstructure and Phase Transition, pp. 191–204. Springer, Berlin (1993) 79. Chlebík, M., Kirchheim, B.: Rigidity for the four gradient problem. J. Reine Angew. Math. 2002(551), 1–9 (2002) 80. Otto, F., Viehmann, T.: Domain branching in uniaxial ferromagnets: asymptotic behavior of the energy. Calc. Var. Partial Differ. Equ. 38(1), 135–181 (2010) 81. Diermeier, J.: Domain branching in geometrically linear elasticity (2013) 82. Winter, M.: An example of microstructure with multiple scales. Eur. J. Appl. Math. 8(2), 185–207 (1997) 83. Chipot, M.: The appearance of microstructures in problems with incompatible wells and their numerical approach. Numer. Math. 83(3), 325–352 (1999) 84. Grafakos, L.: Classical Fourier Analysis. Graduate Texts in Mathematics, vol. 249. Springer, Berlin (2014) Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional afﬁliations. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Acta Applicandae Mathematicae Springer Journals http://www.deepdyve.com/lp/springer-journals/on-scaling-properties-for-two-state-problems-and-for-a-singularly-0k8c4esrCs

Loading next page...

Publisher: Springer Journals
Copyright: Copyright © The Author(s) 2023
ISSN: 0167-8019
eISSN: 1572-9036
DOI: 10.1007/s10440-023-00557-7
Publisher site: See Article on Publisher Site

Abstract

In this article we study quantitative rigidity properties for the compatible and incompatible two-state problems for suitable classes of A-free differential inclusions and for a singularly perturbed T structure for the divergence operator. In particular, in the compatible setting of the two-state problem we prove that all homogeneous, ﬁrst order, linear operators with afﬁne boundary data which enforce oscillations yield the typical -lower scaling bounds. As observed in Chan and Conti (Math. Models Methods Appl. Sci. 25(06):1091–1124, 2015) for higher order operators this may no longer be the case. Revisiting the example from Chan and Conti (Math. Models Methods Appl. Sci. 25(06):1091–1124, 2015), we show that this is reﬂected in the structure of the associated symbols and that this can be exploited for a new Fourier based proof of the lower scaling bound. Moreover, building on Rüland and Tribuzio (Arch. Ration. Mech. Anal. 243(1):401–431, 2022); Garroni and Nesi (Proc. R. Soc. Lond., Ser. A, Math. Phys. Eng. Sci. 460(2046):1789–1806, 2004, https://doi.org/10. 1098/rspa.2003.1249); Palombaro and Ponsiglione (Asymptot. Anal. 40(1):37–49, 2004), we discuss the scaling behavior of a T structure for the divergence operator. We prove that as in Rüland and Tribuzio (Arch. Ration. Mech. Anal. 243(1):401–431, 2022) this yields a non-algebraic scaling law. Keywords A-Free inclusions · Two-well problem · Divergence T · Phase transformation Mathematics Subject Classiﬁcation 35Q74 · 35F05 · 74N05 · 74G99 C. Tissot camillo.tissot@uni-heidelberg.de B. Raita ˘ bogdanraita@gmail.com A. Rüland rueland@uni-bonn.de Centro di Ricerca Matematica Ennio de Giorgi, Scuola Normale Superiore, Piazza dei Cavalieri, 3, 56126 Pisa, Italy Department of Mathematics, Universitatea Alexandru-Ioan Cuza, Bulevardul Carol I 11, 700506 Iasi, Romania Institut für Angewandte Mathematik, Ruprecht-Karls-Universität Heidelberg, Im Neuenheimer Feld 205, 69120 Heidelberg, Germany Institute for Applied Mathematics and Hausdorff Center for Mathematics, University of Bonn, Endenicher Allee 60, 53115 Bonn, Germany 5 Page 2of50 B.Raitae ˘ tal. 1Introduction Rigidity and ﬂexibility properties associated with (nonlinear) differential inclusions for the gradient have been objects of intensive study. They arise in a variety of applications, in- cluding the analysis of PDEs, e.g. the study of regularity of elliptic systems [5–7], ﬂuid dynamics [8, 9], geometry [10–13] and various settings in the materials sciences, e.g. the study of patterns in shape-memory alloys [14–19]. Motivated by applications of microstruc- tures in composites [20, 21], optimal design problems [22–24] and micromagnetics [25, 26] as well as by recent developments on more general differential inclusion problems [27–29], in this article we study two instances of quantitative rigidity and ﬂexibility properties of differential inclusions for more general operators. On the one hand, we consider constant- coefﬁcient, homogeneous, linear differential operators for which we discuss quantitative versions of the compatible and incompatible two-state problems. On the other hand, we investigate quantitative properties of a T structure for the divergence operator. 1.1 On Quantitative Results for the Two-Well Problem for A-Free Diﬀerential Inclusions A-free differential inclusions arise in many different settings, including linearized elasticity [30, 31], liquid crystal elastomers [32, 33] and the study of the Aviles-Giga functional [34, 35] to name just a few examples. They have been systematically investigated in the context of compensated compactness theory in classical works such as [36–39]but also in more recent literature on compensated compactness theory [40–46], in truncation results [47], in classical minimization and regularity questions in the calculus of variations [48–51]and in the context of ﬁne properties of such operators in borderline spaces [27, 52–54]. In the recent articles [27, 28] general A-free versions of the incompatible two-well problem in borderline spaces and the study of T structures have been initiated. Motivated by these applications, in the ﬁrst part of this article, we seek to study quantitative versions of the compatible and incompatible two-state problems. Before turning to the setting of general A-free differential inclusions, let us recall the analogous “classical” setting for the gradient: Inspired by problems from materials science and phase transformations, the exact and approximate rigidity properties of differential in- clusion problems for the gradient [14](seealso[18, 55–57]) with and without gauge invari- ance have been considered. For two energy wells without gauge invariances this amounts to the study of the differential inclusion ∇v ∈{A, B} in (1) d×d d for A, B ∈ R , A = B and ⊂ R a bounded Lipschitz domain. It is well-known that depending on the compatibility of the wells, a dichotomy arises: d×d • On the one hand, for incompatible wells, i.e. if B − A ∈ R is not a rank-one matrix, the differential inclusion (1)is rigid both for exact and approximate solutions: Indeed, if A, B are incompatible,(1) only permits solutions with constant deformation gradient, which is in the following referred to as the rigidity of the two-state problem for exact solutions. Moreover, in this setting one has that for sequences ∇u with dist(∇u , {A, B}) → 0in k k measure, it necessarily holds that along a subsequence ∇u → A or ∇u → B in measure. k k We will refer to this as rigidity of the two-state problem for approximate solutions. We remark that various far-reaching generalizations of these results have been ob- tained: For settings with SO(d) symmetries a quantitative version of such a result was Scaling for the Two-State Problem and a T Structure Page 3 of 50 5 deduced in [58]; a further proof was found in [59]. The one-state problem with continu- ous symmetry group was studied in [60, 61]. The article [27] investigates a similar problem for two incompatible wells for a general constant coefﬁcient, homogeneous, linear differential operator A(D), providing qualita- tive rigidity results for the associated exact and approximate differential inclusions, in- cluding L -based frameworks. d×d • On the other hand, if the wells are compatible, i.e. if B − A ∈ R is a rank-one matrix, then simple laminate solutions of (1) exist, in which the deformation gradient oscillates between the two ﬁxed values A, B and is a one-dimensional function depending only on the direction determined by the difference B − A. Due to the failure of rigidity on the exact level, also rigidity on the approximate level cannot be expected without additional regularization terms. In the ﬁrst part of this article we seek to consider quantitative, L -based variants of these type of results for more general, linear differential operators A(D). In this context, we will consider the following two guiding questions: • Quantitative incompatible rigidity. For a constant coefﬁcient, homogeneous, linear dif- ferential operator A(D) and two incompatible wells, i.e. A, B ∈ R such that B − A/ ∈ ,cf. (5) for the deﬁnition of the wave cone, do we have a quantitative rigidity result in terms of domain scaling for prescribed boundary data which are a convex combination of the two states? Here the dimension n depends on the operator A(D);for A(D) = curl (which corresponds to the gradient setting from (1) above) we would for instance consider n = d × d . More precisely, we seek to study the following question: Let A, B ∈ R be such that B − A/ ∈ .Isittruethat E (u, χ ) = |u − χ A − χ B| dx ≥ C(A, B, λ)||, el A B d n d if u : R → R , A(D)u = 0in R , χ := Aχ + Bχ ∈{A, B} in and χ ,χ ∈{0, 1} A B A B with χ + χ = 1 and if for some λ ∈ (0, 1) we have that u = F := λA + (1 − λ)B in A B λ R \ ? • Quantitative compatible rigidity. Letusnextconsider two compatible wells A, B ∈ R , i.e. let A, B ∈ R be such that B − A ∈ ,see (5) below, and let us again consider boundary data F := λA + (1 − λ)B for some λ ∈ (0, 1) as above and with the set of admissible functions given by D in (10). For a singularly perturbed energy similarly as in (15)isittruethatasin[62, 63] also in the setting of a more general constant coefﬁcient, homogeneous, linear differential operator A(D) the following bound holds 2/3 inf inf (E (u, χ ) + |∇χ |) ≥ C ? el χ ∈BV (;{A,B}) u∈D In what follows, we will formulate the set-up, the relevant operator classes and our results on these questions. 1.2 Formulation of the Two-State Problem for A-Free Operators in Bounded Domains Following [27, 28, 48], we consider a particular class of linear, homogeneous, constant- ∞ d n ∞ d m coefﬁcient operators. The operator A(D) : C (R ; R ) → C (R ; R ) of order k ∈ N is 5 Page 4of50 B.Raitae ˘ tal. giveninthe form A(D) := A ∂ , (2) |α|=k d m×n where α ∈ N denotes a multi-index of length |α|:= α and A ∈ R are constant j α j =1 matrices. Seeking to study microstructure, in the sequel we are particularly interested in non-elliptic operators. Here the operator A(D) is said to be elliptic if its symbol A(ξ ) := A ξ (3) |α|=k is injective for all ξ = 0. If A(D) is not elliptic, there exist vectors ξ ∈ R \{0} and μ ∈ R \{0} such that A(ξ )μ = 0. (4) The collection of these vectors μ ∈ R \{0} forms the wave cone associated with the operator A(D): := ker(A(ξ )). (5) d−1 ξ ∈S The relevance of the wave cone for compensated compactness and the existence of mi- crostructure is well-known. For instance, for any pair (μ, ξ ) as in (4) it is possible to obtain A-free simple laminate solutions. These are one-dimensional functions u(x) := μh(x · ξ), where h : R →{0, 1}, which obey the differential constraint A(D)u = 0 due to the choice of μ, ξ and u ∈{0,μ}. More generally, if k = 1, and for μ ∈ (see, for instance, [64]) it holds that u(x) := μh(x · ξ ,...,x · ξ ) is a solution to the differential equation A(D)u = 0 for vectors ξ ,...,ξ ∈ R \{0} forming a basis of the vectorspace V := ξ ∈ R : A(ξ )μ = 0 . (6) A,μ For the row-wise curl operator (n = d × d ) this is an at most one-dimensional space, while for the row-wise divergence operator (n = m × d ), it is a space of possibly higher dimension as V = ker μ, leading to substantially more ﬂexible solutions of the associated differen- div,μ tial inclusions than for the curl. In order to study microstructures arising as solutions to the two-state problem, for the above speciﬁed class of operators, analogously as in the gradient setting, we consider the following A-free differential inclusion with prescribed boundary values: u ∈ K in , (7) A(D)u = 0in R , n d with K ⊂ R ,and ⊂ R an open, bounded, simply connected domain, with appropriately prescribed boundary data. For the two-state problem we consider K ={A, B}⊂ R . Scaling for the Two-State Problem and a T Structure Page 5 of 50 5 Now, in analogy to the gradient setting, on the one hand, we call the differential inclusion for the two-state problem (7) incompatible if it is elliptic in the sense that B − A/ ∈ .In this case it is proved in [27] that both the exact and approximate differential inclusion (7) are rigid. We emphasize that incompatibility in particular excludes the presence of simple laminates. On the other hand, the differential inclusion (7)issaidtobe compatible if B − A ∈ . In this case, also in the setting of more general operators, a particular class of solutions to (7) consists of (generalized) simple laminates. Moreover, in the compatible setting, we further distinguish a particular case: We consider the subspace I := ker(A(ξ )) = ker(A ). (8) A α |α|=k ξ ∈R This is the space of values that are (algebraically) unconstrained by A(D), meaning that for 2 d all u ∈ L (R ; I ),wehavethat A(D)u = 0 without taking any regularity constraints on u. In the case of I ={0}, the operator A(D) belongs to the class of cocanceling operators, introduced in [65]. We seek to study both settings and the resulting microstructures quantitatively in the spirit of scaling results as, for instance, in the following non-exhaustive list involving differ- ent physical applications [2, 22, 62, 63, 66–77]. To this end, for ⊂ R an open, bounded, Lipschitz set, we introduce elastic and surface energies and consider their minimization for prescribed, not globally compatible boundary data F = λA + (1 − λ)B for some λ ∈ (0, 1), where again the set of states is given by K ={A, B}. Motivated by the applications from materials science, we study the following “elastic energy” E (u, χ ) := |u − χ | dx, (9) el which we minimize in the following admissible class of deformations 2 d n d d 2 u ∈ D := u ∈ L (R ; R ) : A(D)u = 0in R ,u = F in R \ ,χ ∈ L (; K). F λ λ loc (10) For ease of notation, here and in what follows, we often use the convention that χ := χ A + χ B with χ ,χ ∈ L (;{0, 1}) and χ + χ = 1in . Moreover, we further use the B A B A B notation E (χ ; F ) := inf E (u, χ ). el λ el u∈D In addition to the “elastic” energy contributions, we also introduce a surface energy con- tribution of the form E (χ ) := |∇χ |,χ ∈ BV (; K) (11) surf and consider the following singularly perturbed elastic energy for > 0 E (u, χ ) := E (u, χ ) + E (χ ), (12) el surf and correspondingly E (χ ; F ) := E (χ ; F ) + E (χ ) = inf E (u, χ ). λ el λ surf u∈D λ 5 Page 6of50 B.Raitae ˘ tal. We note that this can be deﬁned for an arbitrary set of states K and suitable boundary data F . With these quantities in hand, we can formulate the following quantitative rigidity results for the two-state problems: Theorem 1 Let d, n, m, k ∈ N, n> 1. Let ⊂ R be a bounded Lipschitz domain, and let A(D) be as in (2) with the wave cone given in (5) and let I be given in (8). Let A A n 2 A, B ∈ R and let χ = χ A + χ B ∈ L (;{A, B}). Further, let the elastic and surface A B energies E , E be given as in (9) and (11), respectively. For λ ∈ (0, 1) set F = λA + el surf λ (1 − λ)B ∈ R and consider D as in (10). The following results hold: (i) Incompatible case: Assume that B −A/ ∈ . Then there is a constant C = C(A, B) > 0 such that, inf inf E (u, χ ) ≥ C(min{λ, 1 − λ}) ||. el 2 u∈D χ ∈L (;{A,B}) F (ii) Compatible case: Assume that A(D) is one-homogeneous, i.e. k = 1 in (2), and that B − A ∈ \ I . Then, there exist C = C(A(D), A, B, , d, λ) > 0 and = A A 0 (A(D), A, B, , d, λ) > 0 such that for ∈ (0, ) 0 0 2/3 inf inf (E (u, χ ) + E (χ )) ≥ C . el surf χ ∈BV (;{A,B}) u∈D Furthermore, if we assume A(D) = div and =[0, 1] , then there exists a constant c = c(A, B, λ) > 0 such that for > 0 we also have the matching upper bound 2/3 inf inf (E (u, χ ) + E (χ )) ≤ c . el surf u∈D χ ∈BV (;{A,B}) (iii) Super-compatible case: Assume that A − B ∈ I . Then, inf inf (E (u, χ ) + E (χ )) = 0. el surf χ ∈BV (;{A,B}) u∈D We highlight that in our discussion of the compatible case, we have restricted ourselves to operators of order one. This is due to the fact that for higher order operators it is expected that more complicated microstructures may arise. This is also reﬂected in the Fourier space properties of the symbol A. We refer to Sect. 3.5, see Proposition 3.10, for a brief discussion of this, illustrating that the scaling may, in general, be no longer of the order in the higher order setting. Let us discuss a prototypical example of the above results: Example 1.1 As an example of the above differential inclusion, we consider the case in ∞ d m×d ∞ d m d which A(D) = div : C (R ; R ) → C (R ; R ) row-wise for matrix ﬁelds u : R → m×d R . In this case the boundary value problem under consideration turns into the following differential inclusion: u ∈{A, B} in , div u = 0in R , u = F in R \ , λ Scaling for the Two-State Problem and a T Structure Page 7 of 50 5 for some F := λA + (1 − λ)B , λ ∈ (0, 1). Such differential inclusions are related to appli- cations in shape-optimization as, for instance, in [22]. d m×d The divergence is applied row-wise to matrix ﬁelds u : R → R , i.e. A(D)u = m×d (∂ u)e with the matrix vector product. Hence A(ξ )M = Mξ for M ∈ R and thus j j j =1 the wave cone is given by m×d d ={M ∈ R : there is ξ ∈ R with Mξ = 0}. div Moreover the divergence is a cocanceling operator as I ={0}. div We emphasize that Example 1.1 is indeed a prototypical example and plays a central role in the study of ﬁrst order operators in that all ﬁrst order operators can be reduced to this model operator by a suitable linear transformation, see [28, Appendix] and also Sect. B below. We emphasize that this reduction is particularly useful if the differential inclusion is incompatible or if the boundary data are in \ I . As a consequence, quantitative lower A A bound estimates for incompatible differential inclusions for ﬁrst order operators, e.g. for T structures as qualitatively studied in [28], or for compatible, but not super-compatible boundary data can be deduced from the ones of the divergence operator (see Proposition B.2 and, in general, the discussion in Sect. B). A reduction to an equivalent problem for a mod- iﬁed operator and modiﬁed boundary data to the setting involving a cocanceling operator will be discussed in Sect. 3.4, see Proposition 3.8 and Corollary 3.9. 1.3 Quantitative Rigidity of a T Structure for the Divergence Operator In the second part of the article, building on the works [2–4] and motivated by the high- lighted considerations on the role of the divergence operator, we study the quantitative rigid- ity of the T conﬁguration u ∈{A ,A ,A } a.e. in , div u = 0in R , (13) 1 2 3 3 3 3×3 where =[0, 1] , u : R → R and ⎛ ⎞ − 00 ⎝ ⎠ A = 0 ,A = 0 0 ,A = Id . (14) 1 3×3 2 3 3×3 00 3 By virtue of the results from [3, 4] this problem is ﬂexible for approximate solutions but rigid on the level of exact solutions: • More precisely, on the level of exact solutions to (13), only constant solutions u ≡ A for j ∈{1, 2, 3} obey the differential inclusion. • Considering however approximate solutions, i.e. sequences (u ) such that k k∈N dist(u , {A ,A ,A }) → 0 in measure as k →∞, div u = 0for all k ∈ N, k 1 2 3 k there exists a sequence of approximate solutions (u ) for (13) such that there is no sub- k k∈N sequence which converges in measure to one of the constant deformations {A ,A ,A }. 1 2 3 Compared to the setting of the gradient, for the divergence operator rigidity for approximate solutions is already lost for the three-state problem (while this arises only for four or more states for the gradient [78, 79], see also [18] for further instances in which the Tartar square was found and used). 5 Page 8of50 B.Raitae ˘ tal. As in [2] we here study a quantitative version of the dichotomy between rigidity and ﬂexibility: We consider the singularly perturbed variant of (13) as in Sect. 1.2 ˆ ˆ E (u, χ ) := |u − χ | dx + |∇χ | (15) qc under the constraint u ∈ D ,cf. (10), with A(D) = div and where F ∈{A ,A ,A } (see F 1 2 3 Sect. 2 for the deﬁnition of the A-quasi-convexiﬁcation of a compact set). Here the function χ ∈ BV (;{A ,A ,A }) denotes the phase indicator of the “phases” A , A , A , respec- 1 2 3 1 2 3 tively. Adapting the ideas from [2] to the divergence operator in three dimensions, we prove the following scaling result: Theorem 2 Let =[0, 1] and let K ={A ,A ,A } with A , j ∈{1, 2, 3} given in (14), 1 2 3 j qc F ∈ K \ K, E be as in (15) above for the divergence operator A(D) = div, and consider D given analogously as (10). Then, there exist constants c = c(F ) > 0 and C = C(F ) > 1 such that for any ν ∈ (0, ) there is = (ν, F ) > 0 and c > 0 such that for ∈ (0, ) 0 0 ν 0 we have 1 1 −1 +ν 2 2 C exp(−c | log()| ) ≤ inf inf E (u, χ ) ≤ C exp(−c| log()| ). χ ∈BV (;K) u∈D Let us comment on this result: As in [2] we obtain essentially matching upper and lower scaling bounds with less than algebraic decay behavior as → 0, reﬂecting the inﬁnite order laminates underlying the T structure and the fact that the problem is “nearly” rigid. As in [2] a key step is the analysis of a “quantitative chain rule in a negative Sobolev space” which results from the interaction of Riesz type transforms and a nonlinearity originating from the “ellipticity” of the differential inclusion. Both in the upper and the lower bound, these estimates for the divergence operator however require additional care due to the three- dimensionality of the problem. In the upper bound construction this is manifested in the use of careful cut-off arguments; in the lower bound, a more involved iterative scheme has to be used to reduce the possible regions of concentration in Fourier space. Similarly, as in [76] the scaling law from Theorem 2 is obtained as a consequence of a rigidity estimate encoding both the rigidity and ﬂexibility of the T differential inclusion (in analogy to the T case from [76, Proposition 3]). Proposition 1.2 Let =[0, 1] and let K ={A ,A ,A } with A , j ∈{1, 2, 3} given in 1 2 3 j qc (14), F ∈ K \ K. Let χ denote the diagonal entries of the matrix ﬁeld χ ∈ BV (; K) and jj per denote by E (χ ; F) the periodic singularly perturbed energy (see (32) and Sect. 4.3.2). Then, there exists > 0 such that for any ν ∈ (0, 1) there is a constant c > 0 such that for 0 ν ∈ (0, ) we have 1 1 2 +ν per 2 2 χ − χ ≤ exp(c | log()| )E (χ ; F) . jj jj 2 3 ν L ([0,1] ) j =1 We emphasize that in parallel to the setting of the Tartar square, this estimate quanti- tatively encodes both rigidity and ﬂexibility of the differential inclusion, as it measures the distance to the constant state (and thus reﬂects rigidity of the exact differential inclusion) but also quantiﬁes the “price” for this in terms of a “high energy” scaling law (and thus reﬂects the underlying ﬂexibility of the approximate problem). Moreover, due to the ﬂexibility of the differential inclusion, we stress that such an estimate can only be inferred for a combination of elastic and surface energies. Scaling for the Two-State Problem and a T Structure Page 9 of 50 5 1.4 Outline of the Article The remainder of the article is structured as follows: After brieﬂy recalling relevant notation and facts on convex hulls related to the operator A(D), we ﬁrst discuss the compatible and incompatible two-state problems in Sect. 3. Here we begin by discussing the incompatible setting, which is a consequence of direct elliptic estimates in Sect. 3.2 andthenturnto the lower bounds for the compatible setting in Sect. 3.3. The super-compatible case is then treated in Sect. 3.4. It is also in this section that we discuss a reduction to cocanceling operators. We revisit the scaling of a prototypical higher order operator from [1] in Sect. 3.5 and explain how our scheme of deducing lower bounds also yields a quick Fourier based proof of the lower scaling bound from [1]. In Sect. 4 we then turn to the T differential inclusion, for which we ﬁrst prove upper bounds in Sect. 4.1 and then adapt the ideas from [2] to infer essentially matching lower bounds in Sect. 4.2. In the Appendix, we complement the general lower bounds for ﬁrst order differential op- erators with upper bounds for the speciﬁc case of the divergence operator (Sect. A). More- over, in Sect. B.1 we discuss the reduction to the divergence operator. 2Notation In this section we collect the notation which is used throughout the article. • For a set U,wedenoteby d the (possibly smoothed-out) distance to this set: d (x) = U U dist(x, U ) = inf |x − y| and denote by χ the (in some places smoothed-out) indicator y∈U U function of this set. • For a set U with ﬁnite measure and a function f : U → R we denote the mean by f = fdx . |U | U 2 d 2 d • For a function f ∈ L (T ) or f ∈ L (R ) we denote the Fourier transform by − −ik·x d F (f )(k) = f(k) = (2π) e f(x)dx (k ∈ Z ) or − −iξ ·x d F (f )(ξ ) = f(ξ) = (2π) e f(x)dx (ξ ∈ R ). ∞ d For a function h ∈ L (R ), we denote by h(D) the corresponding Fourier multiplier −1 h(D)f = F (h(·)f(·)). • We usually denote the phase indicator by χ ∈ L (; K), with the component functions qc qc given by χ = (χ ), i ∈{1,...,n}. Moreover, we use F ∈ K (where K is introduced below) as the exterior data. • The set of admissible ’deformations’ u for an operator of order k ≥ 1is given by A 2 d n d d D := D := {u ∈ L (R ; R ) : A(D)u = 0in R ,u = F in R \ }. Here the equa- F loc tion A(D)u = 0 is considered in a distributional sense. qc • As introduced in Sect. 1.2,for F ∈ K we consider the elastic energy with u ∈ D , χ ∈ L (; K): E (u, χ ) = |u − χ | dx, E (χ ; F) = inf E (u, χ ), (16) el el el u∈D 5 Page 10 of 50 B. Raitae ˘ tal. and, for χ ∈ BV (; K), the surface energy as the total variation norm: E (χ ) =∇χ = |∇χ |. (17) surf TV() • The total energy with > 0is given by E (u, χ ) = E (u, χ ) + E (χ ). el surf • We write f ∼ g if there are constants c, C > 0 such that cf ≤ g ≤ Cf . −1 −1 3 • We use the notation · for the homogeneous H semi-norm for f ∈ H (T ; R): ˙ −1 2 2 f = |f(k)| . −1 |k| k∈Z \{0} We further recall the notions of the -convex hull of a set (see [64]) and the A-quasi- convex hull of a set: n n • Let ⊂ R be the wave cone from (5)and let K ⊂ R be a compact set. For j ∈ N,we (j ) then deﬁne K as follows: (0) K := K, (j ) n (j −1) K := {M ∈ R : M = λA + (1 − λ)B : A, B ∈ K ,B − A ∈ ,λ ∈[0, 1]}. lc Moreover, we deﬁne the -convex hull K : lc (j ) K := K . j =0 We deﬁne the order of lamination of a matrix M ∈ R to be the minimal j ∈ N such that (j ) lc M ∈ K . In analogy to the gradient case, we will also refer to K as the laminar convex hull. qc n • We recall that the A-quasi-convex hull K of a compact set K ⊂ R is deﬁned by duality to A-quasi-convex functions [48]. We recall that the div-quasi-convex hull of the T ma- trices from (13), (14) has been explicitly characterized in [21, Theorem 2] to consist of the union of the closed triangle formed by the matrices S ,S ,S and the three “legs” formed 1 2 3 3×3 by the line segments A S . The matrices S ,S ,S ∈ R are introduced in Sect. 4 be- j j 1 2 3 3×3 low. Given the set {A ,A ,A }⊂ R from (13), (14), as in the gradient case, we denote 1 2 3 qc its A-quasi-convex hull by {A ,A ,A } . 1 2 3 3 Quantitative Results on the Two-State Problem In this section, we study quantitative versions of the two-state problem for general, linear, constant coefﬁcient, homogeneous operators, always considering the divergence operator as a particular model case. Building on the precise formulation of the problem from (7), in Sect. 3.1,weﬁrstcharac- terize the elastic energies in terms of the operator A(D). With this characterization in hand, using ellipticity, we next prove the quantitative L bounds in the incompatible two-well case (Sect. 3.2)and lower -scaling bounds for the compatible case with ﬁrst order, linear oper- ators (Sect. 3.3). In Sect. 3.4, we prove Theorem 1(iii) and deduce a reduction to cocanceling operators. In Sect. 3.5 we brieﬂy discuss the role of degeneracies in the symbol of the elastic Scaling for the Two-State Problem and a T Structure Page 11 of 50 5 energies which may arise for higher order, linear differential operators and which may thus lead to an alternative scaling behavior different from the -scaling bound. Finally, in Sect. A we complement the lower bounds from this section with matching up- per bounds for the special case of A(D) = div. Similar constructions are also known for the gradient, the symmetrized gradient and lower dimensional problems from micromagnetics [1, 66, 73, 74, 80]. 3.1 Elastic Energy Characterization We begin by recalling an explicit lower bound for the elastic energy in terms of the operator A(D) (see, for instance also [49, discussion before Lemma 1.17]) in an, for us, convenient form. In everything that follows, we will assume d> 1. Lemma 3.1 (Fourier characterization of the elastic energy) Let d, n, k ∈ N. Let ⊂ R be a bounded Lipschitz domain with associated indicator function χ and let K ⊂ R . Let A(D) = A ∂ be as in (2) with the symbol A, cf.(3), and E be as in (16) and F ∈ α el |α|=k qc 2 K . Then there is a constant c = c(A(D)) > 0 such that for any χ ∈ L (; K), extended by zero outside of , E (χ ; F) ≥ inf (u − F) − (χ − Fχ ) dx el A(D)u=0 d ∗ ∗ −1 = A(ξ ) A(ξ )A(ξ ) ) A(ξ )(χ ˆ − F χ ˆ ) dξ ≥ c A( )(χ ˆ − F χ ˆ ) dξ, |ξ | 2 d n d where the inﬁmum is taken over all u ∈ L (R ; R ) that fulﬁll A(D)u = 0 in R . loc Proof The proof follows from a projection argument in Fourier space. Step 1: Whole space extension, Fourier and pseudoinverse of the differential operator. We begin by transforming our problem to one on the whole space R , introducing the whole space extension w := u − F : ˆ ˆ 2 2 E (u, χ ) = u − χ dx = u − F − (χ − F) dx el = w − (χ − Fχ ) dx, where we have extended all functions in the integrand by zero outside of . In the following, we write χ ˜ := χ − Fχ . Fourier transforming the expression for the elastic energy then leads to E (u, χ ) = w ˆ − χ ˜ dξ. el Further, seeking to deduce a lower bound, we relax the boundary data for w, obtaining ˆ ˆ ˆ 2 2 2 E (χ ; F) = inf |u − χ | dx = inf |w −˜ χ | dx ≥ inf |ˆ w − χ ˜ | dξ. el u∈D w∈D d A(D)w=0 d F 0 R R 5 Page 12 of 50 B. Raitae ˘ tal. Minimizing the integrand w ˆ (and still denoting the minimizer by w ˆ ) for each ﬁxed mode ξ ∈ R \{0}, we infer that ˆ ˆ ˆ ˆ w( ˆ ξ ) − χ( ˜ ξ ) = χ( ˜ ξ ) − χ( ˜ ξ ) = ∗χ( ˜ ξ ) ker A(ξ ) ran A(ξ ) ∗ ∗ −1 = A(ξ ) A(ξ )A(ξ ) ) A(ξ )χ( ˜ ξ ) , whereweview A(ξ )A(ξ ) : ran A(ξ ) → ran A(ξ ) as an isomorphism. This directly implies the claimed lower bound for E (χ ; F) in terms of the symbol A and its pseudoinverse. el Step 2: Proof of the ﬁnal estimate. Now to show the ﬁnal estimate for the elastic energy, ∗ ∗ −1 d n we seek to bound |A(ξ ) (A(ξ )A(ξ ) ) A(ξ )x| for every ξ ∈ R , x ∈ R from below in terms of |A(ξ )x|. As the projection operator is zero-homogeneous, we can reduce to ξ ∈ d−1 d−1 d−1 S , and thus can use the continuity of S ξ → A(ξ ) and the compactness of S for n d−1 the desired bound: For any x ∈ R , ξ ∈ S it hence holds ∗ ∗ −1 A(ξ )x = A(ξ )A(ξ ) (A(ξ )A(ξ ) ) A(ξ )x ∗ ∗ −1 ≤ A(ξ ) (A(ξ )A(ξ ) ) A(ξ )x sup (|A(ξ )|). d−1 ξ ∈S As A(D) = 0, we have that 0 < sup d−1 |A(ζ )|≤ C< ∞. Dividing by this and plugging ζ ∈S this into the expression with the pseudoinverse, we obtain ∗ ∗ −1 E (u, χ ) ≥ A(ξ ) A(ξ )A(ξ ) ) A(ξ )χ ˜ dξ el 1 ξ ≥ A( )χ ˜ dξ, sup |A(ζ )| d |ξ | d−1 ζ ∈S which concludes the argument. We emphasize that we are relaxing the boundary conditions for w( ˆ ξ ) := χ( ˜ ξ ) ker A(ξ ) as we do not calculate the projection of χ onto D , hence the above Fourier bounds only provide lower bounds for the elastic energy. We apply the lower bound from Lemma 3.1 to the two-well problem: Corollary 3.2 Let d, n ∈ N. Let , A(D), A, E be as in Lemma 3.1. Consider K = el {A, B}⊂ R with F = λA + (1 − λ)B for λ ∈ (0, 1). Then there exists a constant 2 d C = C(A(D)) > 0 such that for any χ = χ A + χ B ∈ L (; K), extended to R by zero, A B it holds E (χ ; F ) ≥ C ((1 − λ)χ ˆ − λχ ˆ )A( )(A − B) dξ. el λ A B d |ξ | Proof Using the expression of F in terms of A, B , λ yields A − F = (1 − λ)(A − B) and λ λ B − F =−λ(A − B). Thus, the fact that χ = χ A + χ B and Lemma 3.1 imply λ A B E (χ ; F ) ≥ C A( )(χ ˆ − F χ ˆ ) dξ el λ λ d |ξ | = C ((1 − λ)χ ˆ − λχ ˆ )A( )(A − B) dξ. A B d |ξ | R Scaling for the Two-State Problem and a T Structure Page 13 of 50 5 Remark 3.3 (The divergence operator) As seen in Example 1.1,inthe caseof A(D) = div, ∞ d d×d we have for u ∈ C (R ; R ) (note that we chose square matrices out of simplicity) ∞ d d A(D)u = (∂ u)e ∈ C (R ; R ). i i i=1 With this we can calculate A(ξ )M = ξ Me = Mξ. i i i=1 ∗ d d×d This, in particular, shows that the adjoint operator is given by A(ξ ) : R → R with A(ξ ) x = x ⊗ ξ. ∗ 2 Therefore A(ξ )A(ξ ) x = (x ⊗ ξ)ξ =|ξ | x , and the projection in the lower bound for the elastic energy of Lemma 3.1 takes the desired form −1 ∗ ∗ A(ξ ) A(ξ )A(ξ ) A(ξ )M = (Mξ ⊗ ξ). |ξ | Furthermore it holds −1 ∗ ∗ A(ξ ) A(ξ )A(ξ ) A(ξ )M = A( )M . |ξ | 3.2 The Incompatible Two-Well Problem and Scaling As a ﬁrst application of the Fourier characterizations from the previous section, we prove a quantitative lower bound for the incompatible two-well problem. We emphasize that – as in [27] – this argument is an elliptic argument and thus can be applied to all linear, constant coefﬁcient homogeneous operators. Indeed, the following result holds: Proposition 3.4 Let d, n ∈ N. Let ⊂ R be a bounded Lipschitz domain, let A(D) be given in (2) and A, B ∈ R with B − A/ ∈ , cf.(5), further let F = λA + (1 − λ)B A λ for some λ ∈ (0, 1), and let E be as in (16) with D given in (10). Then there is C = el F C(A, B, A(D)) > 0, such that for any χ ∈ L (;{A, B}) inf E (u, χ ) ≥ C min{λ, 1 − λ} ||. el u∈D Proof By virtue of Corollary 3.2, we have the lower bound E (χ ; F ) ≥ C ((1 − λ)χ ˆ − λχ ˆ )A( )(A − B) dξ, el λ A B |ξ | where the constant only depends on the operator A(D). d d As (A − B) ∈ / ker A(ξ ) for all ξ ∈ R and thus |A(ξ )(A − B)| > 0for any ξ ∈ R , d−1 ξ by continuity of ξ → A(ξ ) and compactness of S , this implies |A( )(A − B)|≥ |ξ | C(A, B, A(D)) > 0. Hence, ˆ ˆ 2 2 2 2 E (u, χ ) ≥ C |(1 − λ)χ ˆ − λχ ˆ | dξ = C |(1 − λ)χ − λχ | dx el A B A B d d R R 5 Page 14 of 50 B. Raitae ˘ tal. ˆ ˆ 2 2 2 2 2 ≥ C ( (1 − λ) dx + λ dx) ≥ C min{λ, 1 − λ} ||, A B where := supp(χ ) ⊂ and := supp(χ ) ⊂ . A A B B Remark 3.5 We emphasize that this result can be viewed as an incompatible nucleation bound. 3.3 The Compatible Two-Well Case and Scaling We next turn to the setting of two compatible wells and restrict our attention to ﬁrst order operators. In this case, we claim the following -lower scaling bound. This is in analogy to the situation for the gradient which had ﬁrst been derived in the seminal works [62, 63]. Proposition 3.6 Let d, n ∈ N. Let ⊂ R be a bounded Lipschitz domain and let A(D) be a ﬁrst order operator as in (2). Let A, B ∈ R be such that B − A ∈ \ I where A A and I are given in (5) and (8), and deﬁne F := λA + (1 − λ)B for some λ ∈ (0, 1). A A λ Let E := E + E be given in (12) with D deﬁned in (10). Then, there exist C = el surf F C(A(D), A, B, , d, λ) > 0 and = (A(D), A, B, , d, λ) > 0 such that for any ∈ 0 0 (0, ) we have 2/3 ≤ C inf inf E (u, χ ). u∈D χ ∈BV (;{A,B}) In order to deal with the compatible case, we invoke the following (slightly generalized) auxiliary results from [2], see also [23, 71], which we formulate for a general Fourier mul- tiplier m: Lemma 3.7 (Elastic, surface and low frequency cut-off) Let d, n, N ∈ N. Let ⊂ R d N be a bounded Lipschitz domain. Let m : R → R be a linear map and denote by d d V := ker m R its kernel and by : R → V the orthogonal projection onto V . Let f ∈ BV (R ;{−λ, 0, 1 − λ}) for λ ∈ (0, 1) with f = 0 outside and f ∈{−λ, 1 − λ} in . Consider the elastic and surface energies given by ˆ ˆ ˜ ˆ ˜ E (f ) := |m( )f(ξ)| dξ, E (f ) := |∇f |. el surf d |ξ | Then the following results hold: (a) Low frequency elastic energy control. Let μ> 1, then there exists C = C(m, ) > 0 with 2 2 ˆ ˜ f ≤ Cμ E (f ). 2 d el L ({ξ ∈R :| (ξ )|≤μ}) (b) High frequency surface energy control. There exists C = C(d, λ) > 0 such that for μ> 0 it holds 2 −1 ˆ ˜ f ≤ Cμ (E (f ) + Per()). 2 d surf L ({ξ ∈R :|ξ |≥μ}) Proof of Lemma 3.7 Since the property (b) is directly analogous to the one from [2, Lemma 2], we only discuss the proof of (a) which requires some (slight) modiﬁcations with respect Scaling for the Two-State Problem and a T Structure Page 15 of 50 5 to [2](and[23]). We thus present the argument for this for completeness. We split ξ = ξ + ξ ,where ξ ∈ V = {0}, ξ ∈ V = ker m. With this in hand and by the linearity of m it holds ξ ξ + ξ ξ ξ m( ) = m( ) = m( ) = M |ξ | |ξ | |ξ | |ξ | for some matrix representation M of m. Hence, using ξ ⊥ ker m,there is c = c(m) > 0such that ˆ ˆ ξ ξ 2 2 ˜ ˆ ˆ E (f ) = |m( )f(ξ)| dξ ≥ c | f(ξ)| dξ. el d |ξ | d |ξ | R R With this in hand, we argue similarly as in [23]and [2]: For a, b ∈ R, a, b ≥ 0and the orthogonal splitting ξ = ξ + ξ from above, it holds that ˆ ˆ 1 ξ |ξ | ˜ ˆ ˆ E (f ) ≥ f dξ = |f | dξ el 2 2 c |ξ | |ξ | +|ξ | d d R R ≥ |f | dξ + 1 1 1 {|ξ |≥ , |ξ |≤ } a b ⎛ ⎞ (18) ˆ ˆ ˆ ⎜ ⎟ 2 2 ˆ ˆ = |f | dξ − |f | dξ dξ ⎝ ⎠ + 1 1 ⊥ 1 {|ξ |≤ } V {|ξ |< } b a ⎛ ⎞ ˆ ˆ dim V 1 2 2 2 ⎝ ˆ ˆ ⎠ ≥ |f | dξ − sup |f | dξ . + 1 ξ ∈V 1 ⊥ {|ξ |≤ } Using the notation f(ξ) = f(ξ ,ξ ), setting sup ⊥ |f(ξ ,ξ )| ⊥ ⊥ ξ ∈V dim V dim V +1 a := 2 sup , ˆ 2 |f(ξ ,ξ )| dξ ξ ∈V ∞ 1 and using Plancherel’s identity, the L − L bounds for the Fourier transform and Hölder’s inequality, we obtain that F f(·,ξ ) ξ 1 ⊥ ⊥ ⊥ ⊥ L (V ) dim V dim V +1 dim V +1 a ≤ 2 sup ≤ C()2 . F f(·,ξ ) ξ ∈V ξ 2 ⊥ L (V ) In particular, the constant a is well-deﬁned. Returning to (18), we consequently deduce that for b ∈ (0, 1) 2 2 ˜ ˆ E (f ) ≥ C(, m)b |f | dξ. el {ξ :|ξ |≤ } −1 d d 1 Choosing b = μ < 1 and noting that {ξ ∈ R :| (ξ )|≤ μ}={ξ ∈ R :|ξ |≤ } im- plies the claim. 5 Page 16 of 50 B. Raitae ˘ tal. With Lemma 3.7 in hand, we turn to the proof of the lower bound in Proposition 3.6: Proof of the lower bound in Proposition 3.6 Since B − A ∈ \ I , there exists ξ ∈ R \{0} A A such that A(ξ )(B − A) = 0. As A(D) is a ﬁrst order operator, we have that A(ξ ) is linear in ξ ∈ R . In particular, we have that the set, cf. (6), V := V ={ξ ∈ R : A(ξ )(B − A) = 0}={0} B−A A,B−A d ⊥ is a linear space. Rewriting R ξ = ξ + ξ with ξ ∈ V and ξ ∈ V as in the proof B−A B−A of Lemma 3.7, then Corollary 3.2 implies that E (χ ; F ) ≥ C |((1 − λ)χ ˆ − λχ ˆ )A( )(A − B)| dξ, el λ A B |ξ | ˆ ˆ (19) E (χ ) = |∇(χ − F )|= |∇((1 − λ)χ − λχ )(A − B)|dx surf λ A B ≥ C |∇[(1 − λ)χ − λχ ]|, A B for a constant C depending on the operator A(D) and on A − B . Now, setting m(ξ ) := A(ξ )(A − B) and f := (1 − λ)χ − λχ , yields the applicability A B of Lemma 3.7 with V = V R . This is the only place where we use the assumption B−A that A − B/ ∈ I . We deduce that by (19) and the decomposition of R into the two regions from Lemma 3.7 we have for μ> 1 2 2 2 f ≤ 2 χ (D)f +χ (D)f 2 {|ξ |≥μ} 2 {|ξ |≤μ} 2 L L L 2 −1 −1 −1 ≤ C μ E (χ ; F ) + (μ )E (χ ) + μ Per() , el λ surf where the constant C depends on A(D), A, B , , d , λ. Now choosing μ = > 1, noting −1 −1 − that then μ = for < 1, we obtain 2 1 2 − 3 3 f ≤ C E (χ ; F ) + Per() , 2 λ where E (χ ; F ) := E (χ ; F ) + E (χ ). Using the lower bound λ el λ surf ˆ ˆ 2 2 2 2 f = |(1 − λ)χ ˆ + λχ ˆ | dξ = |(1 − λ)χ + λχ | dx ≥ (min{λ, 1 − λ}) ||, 2 A B A B d d R R then implies that ≤ C(E (χ ; F ) + Per()). Finally, for ∈ (0, ) and = (A(D), A, B, , d, λ) > 0 sufﬁciently small, the perime- 0 0 0 ter contribution on the right hand side can be absorbed into the left hand side, which yields the desired result. Scaling for the Two-State Problem and a T Structure Page 17 of 50 5 3.4 The Super-Compatible Setting: Proof of Theorem 1(iii) and Reduction to Cocanceling Operators In this subsection we will show that if we are in the setting in which the estimates of the previous subsection degenerate, i.e., A(ξ )(A − B) = 0for all ξ ∈ R , then in fact there can be no non-trivial bound from below. We will however also show that, in general for pairwise not super-compatible wells, it is possible to reduce to an equivalent minimization problem in the setting of cocanceling operators for suitably modiﬁed boundary data. Proof of the super-compatible case in Theorem 1 It sufﬁces to give an upper bound construc- tion with zero total energy. To this end, we consider χ = Aχ and u = Aχ + F χ d and R \ observe that A(D)u = A(D)(u − B) = A(D)[(χ + λχ d )(A − B)]= 0. R \ As a consequence, u is admissible in the deﬁnition of the elastic energy and the elastic energy vanishes. Moreover, since χ = A in we also have the vanishing of the surface energy. This concludes the argument. We will show that for two not super-compatible wells, we can always assume that I ={0}, in which case we work in the class of cocanceling operators introduced by Van Schaftingen in [65]. Proposition 3.8 Let n, d, k ∈ N, let ⊂ R be a bounded Lipschitz domain, A(D) a differ- ential operator of order k as in (2) with I given in (8). For χ ∈ L (; K) for some compact n qc setofstates K ⊂ R let E (χ ; F) be as in (9) for F ∈ K , cf. Sect. 2. Then for the restricted el ∞ d ⊥ ∞ d m 2 operator A(D) : C (R ; I ) → C (R ; R ) there exists a function χ ∈ L (, K) A I such that A 2 E (χ ; F) = E (χ ; F ) = inf |u − χ | dx. el ⊥ ⊥ ⊥ ⊥ el u ∈D Here we denote the orthogonal projection of F onto I by F . n ⊥ Proof We use the orthogonal decomposition R = I ⊕ I to write u = u + u ,χ = χ + χ , ⊥ I ⊥ I d ⊥ d ⊥ with u : R → I , u : R → I , χ : → I , χ : → I . By orthogonality we can ⊥ I A ⊥ I A A A also split the elastic energy ˆ ˆ ˆ 2 2 2 E (u, χ ) = |u − χ | dx = |u − χ | dx + |u − χ | dx. el ⊥ ⊥ I I ∞ d ⊥ ∞ d m ˜ ˜ Deﬁning the restricted operator A(D) : C (R ; I ) → C (R ; R ), A(D)u := A(D)u and the restricted space of admissible functions as in (10) A 2 d ⊥ d d ˜ ¯ D := {u ∈ L (R ; I ) : A(D)u = 0in R ,u = F in R \ }, ⊥ ⊥ ⊥ ⊥ F loc A we see that for u ∈ D it holds u ∈ D , with F = ⊥ F ,and u = F − F outside . F ⊥ ⊥ I ⊥ F I A 5 Page 18 of 50 B. Raitae ˘ tal. Thus, after minimizing the elastic energy in u, it holds ˆ ˆ 2 2 inf E (u, χ ) = inf inf |u − χ | dx + |u − χ | dx. el ⊥ ⊥ I I ˜ 2 u∈D d ¯ c F A u ∈L (R ;I ),u =F −F in u ∈D I A I ⊥ ⊥ loc As we have seen in the proof of Theorem 1(iii), the second term involving u vanishes and hence, E (χ ; F) = E (χ ; F ). el ⊥ ⊥ el This reduces the elastic energy to the case of a cocanceling operator as indeed I = {0}. As the surface energy does not depend on the operator A(D), this result yields: E (χ ; F) = E (χ ; F ) + E (χ ). ⊥ ⊥ surf el As a corollary, we apply this to the N -well problem: Corollary 3.9 (Finitely many, pairwise not super-compatible wells) Under the same as- sumptions as in Proposition 3.8, with χ ∈ BV (; K), for the special case that K := {B ,...,B }⊂ R for N ∈ N such that B , j ∈{1,...,N }, are pairwise not super- 1 N j compatible, i.e. B − B ∈ / I for i = j , there exists a constant C = C(B ,...,B )> 1 i j A 1 n such that −1 C E (χ ) ≤ E (χ ) ≤ CE (χ ). surf ⊥ surf surf ⊥ In particular, it hence holds that E (χ ; F) ∼ E (χ ; F ). ⊥ ⊥ N N Proof Writing χ = B χ with χ ∈ BV (;{0, 1}), χ = 1in ,wecan j =1 j j j =1 j calculate d−1 ∗ ∗ |∇χ|= |B − B |H (∂ ∩ ∂ ), i j i j i<j d−1 ∗ ∗ |∇χ |= |(B − B ) |H (∂ ∩ ∂ ), ⊥ i j ⊥ i j i<j where we denote the reduced boundary of a set E with ﬁnite perimeter by ∂ E andusedthe notation from above for B ∈ R to write B = ⊥ B . By assumption, for i< j it holds B − B ∈ / I , and therefore also the projection satisﬁes i j A (B − B ) = 0. This implies |(B − B ) | > 0 for all tupels (i, j ) such that i< j and hence i j ⊥ i j ⊥ |B −B | i j −1 there is a constant 0 <C < <C such that |(B −B ) | i j ⊥ −1 0 <C |∇χ |≤|∇χ|≤ C|∇χ |. ⊥ ⊥ This together with Proposition 3.8 concludes the proof. Scaling for the Two-State Problem and a T Structure Page 19 of 50 5 We emphasize that Corollary 3.9 in particular holds in the context of Theorem 1.There- fore in the statement of Theorem 1(ii) we can assume without loss of generality that I ={0}. In fact, note that in the crucial bound of Corollary 3.2 we have A(ξ )(A − B) = A(ξ )(A − B ) = A(ξ )(A − B ) ⊥ ⊥ ⊥ ⊥ for |ξ|= 1. 3.5 Some Remarks on the Compatible Two-State Problem for Higher Order Operators We conclude our discussion of lower scaling bounds by commenting on the case of the com- patible two-well problem for higher order operators. Here the situation is still less transpar- ent, yet some remarks are possible. Indeed, on the one hand, it is known that, in general, for operators of order k ≥ 2the two-well problem does not have to scale with . In order to illustrate this, we consider the speciﬁc operator A(D) := curl curl. This operator is the annihilator of the symmetrized 1 t gradient e(v) := (∇v + (∇v) ). We consider the following quantitative two-state problem (for d = 2) E (v, χ ) := E (v, χ ) + E (χ ) el surf ˆ ˆ := e(v) − dx + |∇χ | (20) 01 + α(1 − 2χ) 2 2 [0,1] [0,1] A 2 with e(v) ∈ D , χ ∈ BV ([0, 1] ;{0, 1}), α ∈ (0, 1) and study the corresponding minimiza- tion problem with prescribed boundary data F := . Proposition 3.10 Let E (v, χ ) and F be as in (20). Then there exists = (α) > 0 such 0 0 that for ∈ (0, ) it holds that inf inf E (v, χ ) ∼ . 2 A χ ∈BV ([0,1] ;{0,1}) e(v)∈D We remark that this observation is not new; indeed, a geometrically nonlinear version of this had earlier been derived in [1, Theorem 1.2]. As observed in [1] the reason for the different scaling in Proposition 3.10 and [1, Theorem 1.2], compared to the more standard behavior from Theorem 1, consists of the higher degeneracy of the multiplier associ- ated with the energy which is manifested in the presence of only one possible normal in the (symmetrized) rank-one condition. On the level of the multiplier this can be seen as e ⊗e +e ⊗e 2 1 2 2 1 A(ξ )(e ⊗ e ) = ξ , opposed to A(ξ )( ) =−ξ ξ , has only one root of multiplic- 2 2 1 2 ity two instead of two roots of single multiplicity on the unit sphere. For convenience of the reader and in order to illustrate the robustness of the above approach within geometrically linear theories, we present an alternative short proof (of the lower bound) of Proposition 3.10 based on our Fourier theoretic framework. We note that in the geometrically linear setting this provides an alternative to the approach from [1] in which the lower bound for the en- ergy is deduced by a local “averaging” argument, considering the energy on representative domain patches with the expected scaling behavior. 5 Page 20 of 50 B. Raitae ˘ tal. Proof Step 1: Lower bound. We note that the lower bound for this setting directly follows from our arguments above: Indeed, for A − B = 2αe ⊗ e , we obtain that 2 2 A(ξ )(B − A) = 2αξ × ξ × (e ⊗ e ) = 2αξ . 2 2 With this in hand, an analogous argument as in Lemma 3.7 and, in particular, in (18) implies that 2 4 ˆ χ ≤ Cμ E (χ ; F), (21) el 2 2 L ({ξ ∈R :|k |≤μ}) where we have used that in this situation the multiplier is given by m(ξ ) = A(ξ )(B − A) ∼ ξ . The different exponent of μ in (21) (compared to the one from Lemma 3.7(a)) is a consequence of the degeneracy of the symbol m(ξ ) and the higher order of the operator curl curl (or put, more concretely, the quadratic dependence ξ ). Hence, replacing the bound from Lemma 3.7(a) by the one from (21) and carrying out the splitting as in the proof of Theorem 1, we obtain the following optimization problem: For f := 1 + α(1 − 2χ) 2 2 2 2 (1 − α) ≤f ≤χ (D)f +χ (D)f 2 {|ξ |≥μ} 2 {|ξ |≤μ} 2 L L L 4 −1 −1 −1 ≤ C μ E (χ ; F) + (μ )E (χ ) + μ Per() . el surf Choosing μ ∼ and rearranging the estimates then imply the claim. Step 2: Upper bound. The associated improved upper bound makes use of the vectorial 2/3 structure of the problem in contrast to the essentially scalar “standard construction” (see the arguments below). Recalling that the nonlinear construction from the proof of [1, Lemma 2.1] also yields a construction with the desired scaling for the geometrically lin- earized problem, we do not carry out the details of this but refer to [1, Lemma 2.1] for these. On the other hand, the arguments from [1, 73, 74, 81] show that still for A(D) = curl curl 2/3 if A − B = γ(e ⊗ e + e ⊗ e ) for γ ∈ R \{0}, then one recovers the scaling for the 1 2 2 1 symmetrized gradient differential inclusion. In this case, the symbol reads m(ξ ) = A(ξ )(B − A) ∼ ξ ξ and the operator is “less degenerate”. 1 2 We expect that the scaling behavior of general higher order operators is in many inter- esting settings directly linked to the degeneracy of the symbol A(ξ )(B − A).Weplanto explore this in future work. 4 Quantitative Rigidity of the T Structure from (13), (14)for A(D) = div In this section, we consider the T structure for the divergence operator introduced in (13), (14). The upper bound construction is given by an approximate solution of the type de- scribed in the introduction. The lower bound is motivated by the rigidity of exact solutions as outlined in Sect. 4.3.1. 4.1 The Upper Bound Construction – an Inﬁnite Order Laminate To begin with, we construct an inﬁnite order laminate similar to the one for the Tartar square (cf. [2, 82, 83] for quantitative versions of this). This is based on [3] and will yield the Scaling for the Two-State Problem and a T Structure Page 21 of 50 5 Fig. 1 The diagonal matrices A , A , A , S , S , S with the 2 3 1 2 3 dashed lines depicting the connections in the wave cone for A(D) = div and the Voronoi-regions of A shown by the lines. As shown in [21]the qc set K is given by the inner triangle formed by S S S and 1 2 3 the “legs” connecting S and A j j for j ∈{1, 2, 3} upper bound estimate from Theorem 2. We recall that for the divergence operator, instead of requiring rank-one connectedness for the existence of a laminate as for the curl, in our three-dimensional set-up we need rank-one or rank-two connectedness as can be seen from the wave cone for the divergence operator, cf. (5). As a consequence, for two matrices A, B ∈ 3×3 3 3×3 R such that rank(B − A) ≤ 2 there exists a piecewise constant map u : R → R such that u ∈{A, B} a.e. in and div u = 0. The lamination can be done in any direction of the kernel ker(B − A) = {0}. Considering now the matrices A , A , A given in (14), we observe that rank(A − A ) = 1 2 3 i j 3×3 3for i = j . Following [3], we introduce auxiliary matrices S ,S ,S ∈ R . 1 2 3 ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ 000 00 000 2 2 1 ⎝ ⎠ ⎝ ⎠ ⎝ ⎠ S = 0 0 ,S = 0 0 ,S = 0 0 . (22) 1 2 3 3 3 3 002 00 1 001 It then holds for i = 1, 2, 3that ker(S − A ) = span(e ), S = (A + S ), i i i i i+1 i+1 where A = A , S = S .Asprovedin[21, Theorem 2], the A-quasi-convex hull 4 1 4 1 qc {A ,A ,A } can be explicitly characterized as the convex hull of the matrices S , S , 1 2 3 1 2 S together with the “legs” given by the line segments A S for j ∈{1, 2, 3},cf. Fig. 1. 3 j j For simplicity and deﬁniteness, we ﬁrst assume, that u = F = S outside =[0, 1] .In this setting we prove the following energy estimate: Proposition 4.1 Let =[0, 1] , K ={A ,A ,A } for the particular choice of matrices in 1 2 3 −1 (14) and let E be as in (12) and ∈ (0, 1). Then for any r ∈ (0, ) with r ∈ 4N there are (m) 1,∞ 3 3×3 (m) (m) 3 sequences u ∈ W (R ; R ), such that div u = 0, u = F := S outside [0, 1] , (m) 3 and χ ∈ BV (R ; K), and a constant C = C(F ) > 1 with (m) (m) −m −k E (u ,χ ) ≤ C 2 + 2 r + r + . k=1 In order to achieve this, in the next subsections, we iteratively construct a higher and higher order laminate (depending on > 0). As in the setting of the Tartar square, we keep track of the surface and elastic energy contributions which arise in this process. 5 Page 22 of 50 B. Raitae ˘ tal. 4.2 Proof of the Upper Bound from Theorem 2 We split the proof of the upper bound from Theorem 2 into several steps which we will carry out in the next sections and then combine in Sect. 4.2.4. First, we start by a simple lamination of A and S to obtain regions in which u ∈ K 1 1 holds and to satisfy the exterior data condition u = F = S = (A + S ) outside .This 3 1 1 is followed by a similar construction replacing S by a lamination of A and S achieving 1 2 2 a second order laminate. Iterating the procedure of replacing S by a lamination of A j j +1 and S (with the convention that A = A , S = S ) yields Proposition 4.1. Finally, we j +1 4 1 4 1 optimize the parameter r and the number of iterations depending on in order to show the desired upper bound estimate in Proposition 4.3. As we will use a potential for the laminates, we will deﬁne a “proﬁle-function” once in a more general form and will then refer to this in our construction for the higher order laminates. Lemma 4.2 Let R =[a ,b ]×[a ,b ]×[a ,b ]⊂ R be an axis-parallel cuboid, then for 1 1 2 2 3 3 b −a j j any direction e with j ∈{1, 2, 3} and any scale r> 0 such that ∈ N there is a contin- 3×3 uous function f ( ·; R, r) : R → R satisfying the following properties: • The function f ( ·; R, r) only depends on the j -th coordinate x and is r -periodic. j j • It holds f (x; R, r) = 0 if x ∈ R is such that x = (x ,x ,x ) lies in one of the planes j 1 2 3 characterized by x ∈ a + rN . j j 0 S −A j j • The matrix-valued function curl f (x; R, r) only attains the values ± , where A , S j j j is given as in (14) and in (22), respectively. Furthermore, the volumes of the level sets are equal, i.e. |{x ∈ R : f (x; R, r) = S }| = |{x ∈ R : f (x; R, r) = A }| = |R|. j j j j Proof Without loss of generality by a translation, we may assume that R =[0,b ]×[0,b ]× 1 2 [0,b ] for b ,b ,b > 0. We consider the continuous one-periodic extension of the function 3 1 2 3 1 1 tt ∈[0, ), 2 2 h :[0, 1) → R,h(t) := 1 1 (1 − t) t ∈[ , 1). 2 2 Furthermore, we deﬁne the matrices ⎛ ⎞ ⎛ ⎞ ⎛ ⎞ 00 0 001 01 0 2 2 ⎝ ⎠ ⎝ ⎠ ⎝ ⎠ M := 00 − ,M := 000 ,M := − 00 , 1 2 3 3 3 02 0 200 00 0 3×3 satisfying e × M = S − A . With this at hand we deﬁne f ( ·; R, r) : R → R : j j j j j f (x ,x ,x ; R, r) := rh( )M . j 1 2 3 j It follows directly, that f ( ·; R, r) is continuous, only depends on x ,is r -periodic, and j j x S −A j j j vanishes for x ∈ rN . Lastly, we note that curl f (x; R, r) = h ( )e × M ∈{± } j 0 j j j r 2 and that indeed |{x ∈ R : f (x; R, r) = S }| = |{x ∈ R : f (x; R, r) = A }| = |R|. j j j j 4.2.1 First Order Laminates 3 3×3 We use a potential v : R → R to construct our laminates attaining the prescribed exterior data, i.e. we consider the row-wise curl: u = curl v. Scaling for the Two-State Problem and a T Structure Page 23 of 50 5 1 1 As S = A + S and (S − A )e = 0 the ﬁrst order lamination is in the e -direction. 3 1 1 1 1 1 1 2 2 We seek to use Lemma 4.2 to construct v, but have to adapt the boundary condition. For 3 3×3 ˜ ˜ ˜ this we deﬁne a mapping S : R → R with curl S = S = F . A possible choice for S is 3 3 3 3 given by the following matrix-valued function ⎛ ⎞ 00 0 ˜ ⎝ ⎠ S (x) := 00 − x . 3 1 0 x 0 Furthermore, we also deﬁne the cut-off function 0 t< , 1 1 3 φ : R →[0, 1],φ(t) = 4t − t ∈[ , ], 2 8 8 1 t> . 1 −1 With this, we deﬁne the (continuous) potential for r ∈ (0, ), r ∈ 4N by using Lemma 4.2 for j = 1, R =[0, 1] : (1) 3×3 v : → R , (1) v (x) = φ( d (x))f (x; , r) + S (x), ∂ 1 3 where d (x) denotes a smoothed-out distance function to the boundary ∂. Without (1) 3 change of notation, we consider the (continuous) extension of v to R by S (x),which is (1) possible, as v (x) = S (x) on ∂. We then set (1) (1) u := curl v , and note that in it holds (1) (1) u (x) = curl v (x) d (x) x d (x) ∂ 1 ∂ = φ ( )h( )∇d (x) × M + φ( ) curl f (x; , r) + F ∂ 1 1 r r r d (x) x d (x) x 1 ∂ 1 ∂ 1 = φ ( )h( )∇d (x) × M + φ( )h ( )(S − A ) + (S + A ). ∂ 1 1 1 1 1 r r r r 2 (1) With these considerations, we have obtained the following properties: It holds div u = −1 3 (1) 3 (1) 3 (1) div(curl v ) = 0in R , u (x) = F in R \ and for x ∈ \ d ([0, r]) we have u ∈ (1) {A ,S }. In other words, our deformation u is a divergence-free function satisfying the 1 1 desired boundary conditions and which, outside of the cut-off region, is a solution to the (1) differential inclusion u ∈{A ,S }. 1 1 With the higher order laminates in mind we rephrase this using the decomposition of into the three disjoint parts consisting of the S -cells, the A -cells and the cut-off region. To 1 1 be more precise, we deﬁne d (x) (1) (1) R := {x ∈ : φ( ) = 1,u = S }, d (x) (1) (1) Q := {x ∈ : φ( ) = 1,u = A }, r 5 Page 24 of 50 B. Raitae ˘ tal. Fig. 2 The x = slice of the (1) projection χ ,bluerepresents A ,red A and green A .(Color 1 2 3 ﬁgure online) d (x) (1) C := {x ∈ : φ( )< 1}. (1) (1) (1) (1) (1) Indeed it holds = R ∪ Q ∪ C and |C |= 6 r and by Lemma 4.2 we know |R |≤ 1 1 ||= . 2 2 (1) (1) Choosing χ = u as the pointwise orthogonal (with a ﬁxed choice for the not (1) uniquely deﬁned points) projection of u onto K,see Fig. 2 for an illustration, up to a uniformly bounded constant, the elastic energy can be bounded by the measure of the region (1) in which u (x) = S and the cut-off region: (1) (1) (1) (1) 2 (1) (1) E (u ,χ ) = |u − χ | dx ≤ C |R |+|C | el 1 3r 1 ≤ C ||+ 6 ≤ C + 3r . 2 8 2 (1) Furthermore, the surface energy is bounded by counting the interfaces at which χ may 2 2 8 jump. This consists of at most interfaces in the interior and at most + 2 · 6 ≤ new r r r interfaces in the cut-off region. For r ∈ (0, ), the surface area is thus controlled by (1) (1) E (χ ) = |∇χ |≤ C . surf 4.2.2 Second Order Lamination After this ﬁrst order lamination, the differential inclusion u ∈ K with div u = 0 holds only (1) in Q . In order to further reduce the energy, we now replace each of the many cuboids (1) (1) 1 1 in R for which u = S = A + S by a lamination in the e -direction. For this, we 1 2 2 2 2 2 (1) modify the potential v in these regions: (1) 2 r −2 3 For x ∈ R and for r ∈ (0, ), r + ∈ N, we deﬁne with the help of Lemma 4.2 2 4r d (1) (x) (2) ∂R (1) 2 v (x) = φ( )f (x; R ,r ) + f (x; , r) + S (x) 2 1 3 (2) (1) (1) and v (x) = v (x) else. Here, and in the following, we use the notation f (x; R ,r) (1) for R which is not a cuboid but an union of disjoint cuboids and mean depending on x (2) (1) the corresponding connected cuboid. By this v deﬁnes a continuous map, as inside R Scaling for the Two-State Problem and a T Structure Page 25 of 50 5 (1) the cut-off attains the constant value one and therefore v (x) = f (x; , r) + S (x) for 1 3 (1) (2) (2) x ∈ ∂R . By construction, the map u := curl v is divergence free, i.e. the interfaces are compatible. (2) (2) As in the construction for ﬁrst order laminates we set χ = u , and deﬁne the sets d (x) (1) (2) (1) ∂R (2) R := {x ∈ R : φ( ) = 1,u = S }, d (x) (1) ∂R (2) (1) (2) Q := {x ∈ R : φ( ) = 1,u = A }, d (1) (x) ∂R (2) (1) C := {x ∈ R : φ( )< 1}. (1) (2) (2) (2) (2) 1 (1) Then indeed again we have the decomposition R = R ∪ Q ∪ C with |R |≤ |R | (2) 1 3 2 and |C |≤ r 6 as the volume of the cut-off region can be bounded by the number of r 8 (1) 2 cells in R times r times six times the area of the biggest face. (2) As the elastic energy vanishes in Q ,weobtain (2) (2) (2) (2) 2 (2) (2) (1) E (u ,χ ) = |u − χ | dx ≤ C |R |+|C |+|C | el 1 3 ≤ C + 3r + r . 4 2 Indeed, this follows from the fact that we have improved our deformation in half the volume (1) of the region in which u ∈ / K but have added a new cut-off region in each cuboid in which we do the second order lamination. For the surface energy it holds 10 1 10 r 10 5 (2) E (χ ) ≤ C( + ) = C( + ), surf 2 2 r r r 2 r r 4 10 1 as we add at most + 12 ≤ many new faces in each one of the many cuboids and as 2 2 r r each surface has a surface area of size at most . 4.2.3 Iteration: (m+1)-th Order Without loss of generality, we assume that the (m + 1)-th order lamination will be in e - direction, i.e. m = 3j for some j ∈ N. Else, we only have to adapt the corresponding roles of the directions. (m) We deﬁne iteratively the (m + 1)-th potential with the help of the sets R , for this we (m) (m) (m) set (for given v ,u = curl v ) d (x) (m−1) (m) (m−1) ∂R (m) R := {x ∈ R : φ( ) = 1,u = S }, d (x) (m−1) ∂R (m) (m−1) (m) Q := {x ∈ R : φ( ) = 1,u = A }, d (m−1) (x) ∂R (m) (m−1) C := {x ∈ R : φ( )< 1}. (m) (m+1) Inside R we then deﬁne v by d (m) (x) ∂R (m+1) (m) m+1 (k−1) k v (x) = φ( )f (x; R ,r ) + f (x; R ,r ) + S (x), 1 [k] 3 m+1 k=1 5 Page 26 of 50 B. Raitae ˘ tal. Fig. 3 One S -cell with the corresponding lamination in e -direction. In green/red/orange the inner boundary of the cut-off region is depicted. (Color ﬁgure online) ⎪ 1 k ≡ 1mod 3, [k]= 2 k ≡ 2mod 3, 3 k ≡ 0mod 3, (m+1) (m) (m) (m+1) and v (x) = v (x) else for x/ ∈ R . By induction we see that v is continuous, d (x) (m) (m) ∂R (m+1) (m+1) that for x ∈ R such that φ( ) = 1 it holds that u := curl v ∈{A ,S } m+1 1 1 (m−1) (m) (m) (m) (m) 1 (m−1) and that we have the decomposition R = R ∪ Q ∪ C with |R |≤ |R |, (m) −m r −m |C |≤ C2 = C2 r for a universal constant C> 0 independent of m, r.This m−1 −m (m) 2 follows from the fact, that the number of cuboids in R is bounded by C and m−1 m−2 m−3 r r r m−2 m−3 that the biggest face of each cuboid has an area of at most r r . As the previous cut-offs still contribute to the total energy, we obtain the following elastic (m+1) (m+1) energy bound (χ := u ) ⎛ ⎞ m+1 m+1 (m+1) (m+1) (m+1) (k) −m −j ⎝ ⎠ E (u ,χ ) ≤ C |R |∪ |C | = C 2 + r + 2 r . el k=1 j =0 (1) (1) (1) (m+1) This bound resembles the (iterative) decomposition of = R ∪ Q ∪ C = R ∪ m+1 m+1 m+1 (k) (k) m+1 (m+1) (k) Q ∪ C and the fact that u = χ in Q . k=1 k=1 k=1 For the surface energy, we again calculate the new contribution of the next order lam- ination and then sum over all previous ones. Each new face has a surface area of at most m m−1 m−2 m−2 r r r r . In each S -cell, we add new faces in the lamination and at most + 12 m+1 m+1 2 2 r r −m−1 ones for the cut-off, cf. Fig. 3. Since we have at most C new cells the surface m m−1 m−2 r r r energy is increased by m−2 m m−1 8 r r r 1 −m −m C2 ( + (surf. in cut-off )) ≤ C2 , m m−1 m−2 m+1 m+1 r r r r 2 2 r yielding the following overall surface energy bound: ⎛ ⎞ m+1 (m+1) −j −j ⎝ ⎠ E (χ ) ≤ C 2 r ≤ . surf m+1 j =1 For the total energy this implies ⎛ ⎞ (m+1) (m+1) −m −j ⎝ ⎠ E (u ,χ ) ≤ C 2 + r + 2 r + . m+1 j =1 Scaling for the Two-State Problem and a T Structure Page 27 of 50 5 With this we have shown the claimed upper bound and have thus concluded the proof of Proposition 4.1. 4.2.4 Combining the Estimates: Proof of the Upper Bound in Theorem 2 With the previous construction in hand, we conclude the upper estimate from Theorem 2: 3 3×3 Proposition 4.3 Let =[0, 1] , K ={A ,A ,A } for A ,A ,A ∈ R be given in (14), 1 2 3 1 2 3 qc let ∈ (0, 1), and let E be as in (12), F ∈ K \ K and D given in (10). Then there are constants c> 0 and C> 1, only depending on the boundary data F , such that inf inf E (u, χ ) ≤ C exp(−c| log | ). u∈D χ ∈BV (;K) Proof Step 1: Conclusion of the argument for F = S . Using the sequences constructed in the construction from Proposition 4.1, implies (m) (m) −m −k −m −m −m E (u ,χ ) ≤ C 2 + 2 r + r + r ≤ C 2 + r + r . k=1 m+1 Optimizing the value of r depending on > 0, we require that r ∼ . Finally, we balance the resulting contributions and seek for the optimal number of iterations m.Thisis 1 1 −m m+1 2 given by 2 ∼ ,thatis m ∼| log | . Plugging this into the upper bound results in (m) (m) E (u ,χ ) ≤ C exp − log(2)| log()| . This concludes the proof for the special case F = S . Step 2: Conclusion of the argument for a general boundary datum. The situation of other qc boundary data F ∈ K can be reduced to the one from Proposition 4.3 by at most two 3×3 further iterations. Indeed, a general matrix F ∈ R in the convex hull of S , S , S , can be 1 2 3 represented as F = λF + (1 − λ)F = λ(ν A + (1 − ν )S ) + (1 − λ)(ν A + (1 − ν )S ) 1 2 1 j 1 j 2 k 2 k with λ, ν ,ν ∈[0, 1], j, k ∈{1, 2, 3} and k = j . Hence, after two additional iterations com- 1 2 pared to the argument from above, we arrive at similar iterative procedures as in the previous subsections. In case that F is an element of one of the legs S A a single iteration sufﬁces j j to reduce the situation to the above argument. This proves the result for a general boundary qc condition F ∈ K \ K. 4.3 Proof of the Lower Bound In this section, we present the proof of the lower bound from Theorem 2. To this end, sim- ilarly as in [2], we mimic and quantify the analogous argument from the stress-free setting which we brieﬂy recall in the following Sect. 4.3.1 and for which we will provide a number of auxiliary results in Sect. 4.3.2. The main argument, given in Sect. 4.3.3, will then consist of a bootstrap strategy, similar to [2], in which we iteratively reduce the possible regions of mass concentration in Fourier space. 5 Page 28 of 50 B. Raitae ˘ tal. Contrary to the previous section, in what follows we will work in a periodic set-up. Since the energy contributions on periodic functions provides a lower bound on the energy contributions of functions with prescribed Dirichlet boundary conditions, we hence also obtain the desired lower bound for the setting of Dirichlet boundary conditions. Indeed, for the elastic energy this is immediate; for the surface energy there is at most an increase by a ﬁxed factor (see the discussion in Lemma 4.5 in Sect. 4.3.2). 4.3.1 The Stress-Free Argument We begin by recalling the argument for the rigidity of the exact inclusion, as we will mimic this on the energetic level. 3 3×3 Proposition 4.4 Let u :[0, 1] → R be a solution to the differential inclusion (13)-(14). Then, there exists j ∈{1, 2, 3} such that u ≡ A in [0, 1] . Proof In the exactly stress-free setting in which the differential inclusion is satisﬁed exactly, i.e. u ∈{A ,A ,A }, the observation that div u = 0and that u is a diagonal matrix leads to 1 2 3 the following three equations ∂ u = 0,∂ u = 0,∂ u = 0, 1 11 2 22 3 33 where u denote the diagonal components of the matrix u. As a consequence, jj u = f (x ,x ), u = f (x ,x ), u = f (x ,x ). 11 1 2 3 22 2 1 3 33 3 1 2 Next we note that the values of u determine the ones for u if j = k, i.e. there are functions jj kk h such that h (u ) = u . Hence, comparing the functions u and u ,weﬁrstobtain k,j k,j jj kk 11 22 that u and u can only be functions of x . Indeed it holds 11 22 3 ∂ u (x) = ∂ h (u (x)) = ∂ h (f (x ,x )) = 0, (23) 2 11 2 1,2 22 2 1,2 2 1 3 and analogously ∂ u (x) = 0. Comparing this to u , we obtain that all three functions must 1 22 33 be constant. Hence, any solution to the (exact) differential inclusion must be constant and u is equal to one of the three matrices A , A , A globally. The exact problem is hence 1 2 3 rigid. Using the ideas from [2], we seek to turn this into a corresponding scaling result. The main difference that arises can be seen in the qualitative rigidity argument above: Instead of comparing only two diagonal entries like in [2], we have to compare twice to deduce that the map is constant. This will be seen in the quantitative argument for the lower bound below. Whereas in [2] there are cones around a single axis (the diagonal entries only depend on one variable), we consider cones around a plane (the diagonal entries depend on two variables). Furthermore, the bootstrap argument will be slightly modiﬁed as it resembles the comparison of the diagonal entries in the qualitative argument given above. 4.3.2 Reduction to the Periodic Setting and Auxiliary Results for the Elastic Energy In this subsection, we provide a number of auxiliary results which we will exploit in the following bootstrap arguments for deducing the lower bound. As a ﬁrst step, we reduce to the situation of periodic deformations. Scaling for the Two-State Problem and a T Structure Page 29 of 50 5 3 qc Lemma 4.5 Let =[0, 1] , let K := {A ,A ,A } be as in (14) and F ∈ K \ K. Let 1 2 3 E (u, χ ) be given by (15) and set ˆ ˆ per 2 E (u, χ ) := |u − χ | dx + |∇χ |. 3 3 T T per 3 3×3 3 Let further D := {u : R → R : div u = 0 in R , u = F }, where u := u(x)dx . Assume that E (u, χ ) ≤ 1 and that there is > 0 such that for any ν ∈ (0, ) there is c > 0 0 ν such that for any ∈ (0, ) it holds that per +ν E (u, χ ) ≥ exp(−c | log()| ). (24) Then, there exists a constant C> 1 such that for ˜ =˜ (ν) > 0 sufﬁciently small 0 0 −1 +ν C exp(−c | log()| ) ≤ inf inf E (u, χ ) 3 u∈D χ ∈BV ([0,1] ;K) F for all ∈ (0, ˜ ). 3 3 Proof In order to infer the lower bound, we show that any function u :[0, 1] → R with constant boundary data can be associated with a suitable periodic function which has the boundary data of u as its mean value and satisﬁes related energy estimates. Indeed, for given u ∈ D , we view it as a function on T by restriction. By the prescribed boundary data it still satisﬁes the differential constraint and further the mean value property. Moreover, per inf E (u, χ ) := inf |u − χ | dx ≤ inf E (u, χ ). el el per per u∈D 3 F u∈D u∈D F F 3 3 Next, viewing χ :[0, 1] → K as a periodic function χ ˜ : T → K,weinfer that ˆ ˆ |∇χ ˜ |≤ |∇χ|+ C. 3 3 T [0,1] Now, due to (24), we obtain that +ν per exp(−c | log()| ) ≤ inf inf E (u, χ ) ≤ inf inf E (u, χ ) + C. per 3 3 u∈D χ ∈BV (T ;K) u∈D χ ∈BV ([0,1] ;K) F For ˜ (ν) > 0 sufﬁciently small, the last right hand side term may thus be absorbed into the left hand side, yielding the desired result. With this result in hand it sufﬁces to consider the periodic set-up in the remainder of this section. This will, in particular, allow us to rely on the periodic Fourier transform in deducing lower bounds for the elastic energy. In what follows all (semi-)norms will thus be considered on the torus. With a slight abuse of notation, we will often omit this dependence. per per qc Lemma 4.6 Let F ∈ K \K, where K ={A ,A ,A } with A in (14), and E and D be 1 2 3 j el F per 2 3 as in Lemma 4.5 and as in (10). Then, it holds for any χ ∈ L (T ; K) and for E (χ ; F) := el per inf u − χ dx u∈D T per i 2 2 E (χ ; F) = |ˆ χ | +|χ( ˆ 0) − F | , i,i el |k| i=1 k∈Z \{0} 5 Page 30 of 50 B. Raitae ˘ tal. where χ are the diagonal entries of χ and χ ˆ is the (discrete) Fourier transform of χ . i,i per Proof We ﬁrst calculate E (u, χ ) in Fourier space el per 2 2 E (u, χ ) = |u − χ | dx = |ˆ u −ˆ χ | , el k∈Z which allows us to characterize minimizers of the elastic energy. per In order to minimize this elastic energy in u ∈ D , u ˆ has to be the (pointwise) orthog- onal projection of χ ˆ onto the orthogonal complement of k, as the differential constraint div u = 0 reads iuk ˆ = 0 in Fourier space. Noting that the row-wise orthogonal projection of k k amatrix M onto span(k) can be written as (M) = (M ) ⊗ , the optimal u ˆ is given by |k| |k| k k u ˆ = ⊥ χ ˆ = (Id − )χ ˆ =ˆ χ − (χ ˆ ) ⊗ |k| |k| for any k ∈ Z \{0}. Returning to our energy, this yields k k per per 2 2 E (χ ; F) = E (u, χ ) = |ˆ χ − (χ ˆ ) ⊗ −ˆ χ | +|F −ˆ χ(0)| el el |k| |k| k∈Z \{0} 2 2 = |ˆ χ | +|χ( ˆ 0) − F | |k| k∈Z \{0} i 2 2 = |ˆ χ | +|χ( ˆ 0) − F | i,i |k| i=1 k∈Z \{0} and shows the claim. Next, following the ideas from [2], for μ, λ > 0, j = 1, 2, 3, we introduce the cones C ={k ∈ Z :|k |≤ μ|k|, |k|≤ λ}, (25) j,μ,λ j and their corresponding cut-off functions m (k) ∈ C (C \{0}; [0, 1]) fulﬁlling j,μ,λ j,2μ,2λ m = 1on C , supp(m (k)) ⊂ C and the decay properties in Marcinkiewicz’s j,μ,λ j,μ,λ j,μ,λ j,2μ,2λ multiplier theorem (see, for instance, [84, Corollary 6.2.5]). The corresponding cut-off mul- tiplier is thus deﬁned by −1 m (D)f = F (m (·)f(·)). (26) j,μ,λ j,μ,λ Furthermore, we use the following results which are shown in [2, Lemma 2, Lemma 3, and Corollary 1]. Following the conventions from [75], with slight abuse of notation compared to our setting in the ﬁrst part of the article, in the whole following section, we now use d to denote the degree of some suitable polynomials and no longer the dimension of the ambient space which in the whole section is simply ﬁxed to be equal to three. Lemma 4.7 Let β, δ, μ, λ > 0. Let m (D) denote the Fourier multipliers associated with i,μ,λ the cones C for i ∈{1, 2, 3} as deﬁned in (25) with the corresponding multipliers given i,μ,λ Scaling for the Two-State Problem and a T Structure Page 31 of 50 5 ∞ 3 3 in (26). Let f ∈ L (T ) ∩ BV (T ) for i = 1, 2, 3 and let h : R → R be nonlinear poly- i j,i nomials (of degree d ) with h (0) = 0 such that h (f ) = f for i = j . If j,i j,i i j 3 3 ∂ f ≤ δ, ∇f ≤ β, i i i TV ˙ −1 i=1 i=1 then there exist constants C = C(h , f ), C = C (h , f ,d), C = C (h , i,j i ∞ i,j i ∞ 0 0 i,j f ,d) > 0 such that for any γ ∈ (0, 1) we have for any i = j i ∞ 2 −2 −1 f − m (D)f ≤ C(μ δ + λ β), k k,μ,λ k 2 k=1 1−γ f − h (m (D)f ) 2 ≤ f − m (D)f , i i,j j,μ,λ j L j j,μ,λ j 2 12d L 2 −2 −1 1−γ −2 −1 h (m (D)f ) − m (D)f ≤ max{(μ δ + λ β) ,μ δ + λ β}. i,j j,μ,λ j i,μ,λ i 2 24d Here we choose the constants such that C > 2C + 2C + 3. Let us comment on these bounds: The functions f are representing the diagonal en- tries of the phase indicator χ ∈ BV ([0, 1] ; K). Thus the ﬁrst estimate corresponds to a ﬁrst frequency localization by exploiting the surface energy control for the high frequen- cies and the ellipticity of the elastic energy away from the cones C . It can be viewed j,μ,λ as a quantiﬁed version of the statement that u is a function only depending on x , x in 11 2 3 Proposition 4.4. The second estimate is a commutator bound that arises from the nonlinear relation h (f ) = f for i = j . The third estimate combines the ﬁrst two bounds. The sec- i,j j i ond and third estimate will form the core tool to iteratively decrease the Fourier support of the characteristic functions of our phase indicators. We will detail this in the remainder of the article. Remark 4.8 It is possible to make the mappings h for i, j ∈{1, 2, 3}, i = j explicit for the j,i choice of matrices A , A , A ,cf. (14). To this end, we may, for instance, consider 1 2 3 14 5 14 11 2 2 h (x) = x − x, h (x) = x − x, 1,2 1,3 9 9 3 3 21 17 21 23 2 2 h (x) = x − x, h (x) =− x + x, 2,1 2,3 4 4 2 2 7 19 7 25 2 2 h (x) =− x + x, h (x) =− x + x. 3,1 3,2 12 12 18 18 4.3.3 Comparison Argument in Fourier Space In this section, we carry out the iterative bootstrap argument which allows us to deduce the ﬁnal rigidity result. As a ﬁrst step of the bootstrap argument, we invoke the results from above which allow us to decrease the region of potential Fourier concentration from the cone C to a cone 1,μ,λ C with λ <λ. This resembles (23) in the exactly stress-free setting in a quanitiﬁed 1,μ,λ version. As f is determined by f and f with the help of h we can reduce λ, similarly 1 2 3 1,j as in our reduction of the dependences of u in Proposition 4.4. 11 5 Page 32 of 50 B. Raitae ˘ tal. Fig. 4 Illustration of the cones C (red) and C (blue), 1,μ,λ 2,μ,λ andalsoof C + C in 2,μ,λ 2,μ,λ Fourier space. (Color ﬁgure online) Lemma 4.9 Under the same conditions as in Lemma 4.7 and for 1 1 4dλμ λ> 0,μ ∈ (0, ), λ ∈ ( ,λ) √ √ 2 2 2 2d + 1 2 1 − 4μ it holds 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 L L L −2 −1 1−γ −2 −1 ≤ 10 max{(μ δ + λ β) ,μ δ + λ β}. 24d Here the constant C is chosen to be the same as in Lemma 4.7. Remark 4.10 We remark that the interval for λ is chosen such that it holds C \ 1,2μ,2λ 1 4dλμ C ⊂{max{|k |, |k |} > 4dμλ}∩ C and the one for μ such that <λ, 1,2μ,2λ 2 3 1,2μ,2λ 1−4μ i.e. the interval for λ is non-empty. Proof We ﬁrst observe that the Fourier transform of h (m (D)f ) is given by a convo- i,j j,μ,λ j lution of the function m f with itself. Hence, the support of h (m (D)f ) is con- j,μ,λ j i,j j,μ,λ j tained in the d -fold Minkowski-sum of C with itself. For j = 2, 3 it therefore holds j,2μ,2λ that F h (m (D)f ) (k) =0for |k | > 4dμλ. 1,j j,μ,λ j j Further we introduce the sets K ={|k | > 4dμλ}, K ={|k | > 4dμλ} and consider the 2 2 3 3 corresponding Fourier multipliers of the smoothed-out indicator functions χ (D), χ (D), K K 2 3 χ (D).Thisimpliesfor i = 2, 3 K ∪K 2 3 χ (D)h (m (D)f ) = 0. (27) K 1,i i,μ,λ i i Scaling for the Two-State Problem and a T Structure Page 33 of 50 5 With this we can show that the Fourier mass concentrates in a cone with smaller truncation parameter, depicted in Fig. 4: Indeed, in Fourier space χ ≤ χ + χ . Thus, K ∪K K K 2 3 2 3 2 2 χ (D)m (D)f ≤(χ (D) + χ (D))m (D)f K ∪K 1,μ,λ 1 2 K K 1,μ,λ 1 2 2 3 2 3 L L 2 2 ≤ 2χ (D)m (D)f + 2χ (D)m (D)f . K 1,μ,λ 1 2 K 1,μ,λ 1 2 2 3 L L (28) Invoking (27) together with the bounds from Lemma 4.7, then implies for i = 2, 3 2 2 χ (D)m (D)f =χ (D) m (D)f − h (m (D)f ) K 1,μ,λ 1 2 K 1,μ,λ 1 1,i i,μ,λ i 2 i i L L ≤m (D)f − h (m (D)f ) 1,μ,λ 1 1,i i,μ,λ i 2 −2 −1 1−γ −2 −1 ≤ max{(μ δ + λ β) ,μ δ + λ β}. 24d Combined with (28) this yields 2 −2 −1 1−γ −2 −1 χ (D)m (D)f ≤ 4 max{(μ δ + λ β) ,μ δ + λ β}. K ∪K 1,μ,λ 1 2 2 3 L 24d We observe that by the choice of the parameters C \ C ⊂ (K ∪ K ) ∩ C , 1,2μ,2λ 1,2μ,2λ 2 3 1,2μ,2λ and thus |m (k) − m (k)|≤ χ (k)m (k). Therefore, 1,μ,λ 1,μ,λ K ∪K 1,μ,λ 2 3 2 2 m (D)f − m (D)f ≤χ (D)m (D)f 1,μ,λ 1 1,μ,λ 1 2 K ∪K 1,μ,λ 1 2 2 3 L L −2 −1 1−γ −2 −1 ≤ 4 max{(μ δ + λ β) ,μ δ + λ β}. 24d In conclusion, (for C ≥ C ) 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 L L L 2 2 ≤f − m (D)f +f − m (D)f 2 2,μ,λ 2 2 3 3,μ,λ 3 2 L L 2 2 + 2f − m (D)f + 2m (D)f − m (D)f 1 1,μ,λ 1 2 1,μ,λ 1 1,μ,λ 1 2 L L −2 −1 −2 −1 1−γ −2 −1 ≤ 2C(μ δ + λ β) + 8 max{(μ δ + λ β) ,μ δ + λ β} 24d −2 −1 1−γ −2 −1 ≤ 10 max{(μ δ + λ β) ,μ δ + λ β}. 24d Let us stress that the decomposition into K and K is in analogy to the two compar- 2 3 isons from Proposition 4.4 in order to show that u is constant. The set K resembles the 11 2 comparison of u and u to show that u is constant in x and the set K resembles the 11 22 11 2 3 comparison of u and u to show that u does not depend on x . 11 33 11 3 Applying the previous result for all three directions simultaneously then yields the fol- lowing corollary which will serve as the induction basis for the subsequent inductive boot- strap argument. Corollary 4.11 (Induction Basis) Let β, δ, μ, λ > 0 and let C be the cones in (25) with i,μ,λ corresponding multiplier m (D), cf.(26), for i = 1, 2, 3. Further let f , h be functions i,μ,λ i j,i 5 Page 34 of 50 B. Raitae ˘ tal. as in Lemma 4.7. Let C > 0, γ ∈ (0, 1) and d ≥ 0 be the constants from Lemma 4.7. For 4dμλ 1 0 λ = λ> 0, μ ∈ (0, ), λ ∈ ( ,λ ) it holds that 0 1 0 2 2 2 1−4μ 2 2d +1 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 1 L 1 L 1 L 30C −2 −1 1−γ −2 −1 ≤ max{(μ δ + λ β) ,μ δ + λ β}. 24d With Corollary 4.11 in hand, we now iteratively further decrease the Fourier supports. To this end, we will invoke the commutator bounds from Lemma 4.7. Lemma 4.12 (Iteration process) Let β, δ > 0 and let C be the cones in (25) with cor- i,μ,λ responding multiplier m (D), cf.(26), for i = 1, 2, 3. Further let μ ∈ (0, ), i,μ,λ 2 2d +1 λ> 0, and let f , h be functions as in Lemma 4.7. Let C > 0, γ ∈ (0, 1) and d ≥ 0 i j,i 0 be the constants from Lemma 4.7. Let λ > 0 be a sequence for k ∈ N with λ = λ and k 0 4dμλ k−1 λ ∈ ( √ ,λ ). It then holds for every k ∈ N \{0} k k−1 2 1−4μ 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 k k k L L L 30C k −2 −1 (1−γ) −2 −1 ≤ max{(μ δ + λ β) ,μ δ + λ β}. 24d Proof We prove the statement by induction on k with the induction basis given by Corol- lary 4.11. Assume that for some arbitrary but ﬁxed k ∈ N it holds 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 k k k L L L (29) 30C k −2 −1 (1−γ) −2 −1 ≤ max{(μ δ + λ β) ,μ δ + λ β}. 24d Now for the induction step k → k + 1 we carry out the same argument as above to show 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 k+1 k+1 k+1 L L L k+1 (30) 30C k+1 −2 −1 (1−γ) −2 −1 ≤ max{(μ δ + λ β) ,μ δ + λ β}. 24d We argue as in the induction basis and present the calculations only for the ﬁrst term on the left hand side of (30). By the triangle inequality, f − m (D)f 1 1,μ,λ 1 2 k+1 2 2 ≤ 2f − m (D)f + 2m (D)f − m (D)f . 1 1,μ,λ 1 1,μ,λ 1 1,μ,λ 1 2 2 k L k k+1 L The ﬁrst contribution is already of the desired form. It thus remains to consider the sec- ond contribution. For this we consider an analogous argument as before: Let K := {|k | > k k 4dμλ }, K := {|k | > 4dμλ }. Then, since χ k (D)h (m (D)f ) = 0on K , k 3 k 1,j j,μ,λ j 3 K k j 2 2 m (D)f − m (D)f ≤χ k k (D)m (D)f 1,μ,λ 1 1,μ,λ 1 1,μ,λ 1 2 2 k k+1 L K ∪K k L 2 3 ≤ 2 χ k (D)(m (D)f − h (m (D)f )) 1,μ,λ 1 1,j j,μ,λ j 2 K k k j =2 Scaling for the Two-State Problem and a T Structure Page 35 of 50 5 ≤ 2 m (D)f − h (m (D)f ) 1,μ,λ 1 1,j j,μ,λ j 2 k k j =2 2 2 ≤ 4 [m (D)f − f +f − h (m (D)f ) ] 1,μ,λ 1 1 2 1 1,j j,μ,λ j 2 k k L L j =2 2 2 = 8m (D)f − f + 4 f − h (m (D)f ) . (31) 1,μ,λ 1 1 2 1 1,j j,μ,λ j 2 k k L L j =2 Again for the second right hand side contribution in (31) we use Lemma 4.7: 2 2 1−γ f − h (m (D)f ) ≤ (f − m (D)f ) . 1 1,j j,μ,λ j 2 j j,μ,λ j 2 k k L 24d L Using the inductive hypothesis (29), overall, we arrive at the following upper bound f − m (D)f 1 1,μ,λ 1 2 k+1 2 2 ≤ 2f − m (D)f + 16m (D)f − f 1 1,μ,λ 1 2 1,μ,λ 1 1 2 k L k L 2 1−γ + 8 (f − m (D)f ) j j,μ,λ j 2 24d ≤ 18f − m (D)f 1 1,μ,λ 1 2 1−γ C 30C k k −2 −1 (1−γ) −2 −1 + 16 ( ) max{(μ δ + λ β) ,μ δ + λ β} . 24d 24d γ γ Arguing symmetrically for f and f then yields: 2 3 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 k+1 k+1 k+1 L L L 2 2 2 ≤ 18(f − m (D)f +f − m (D)f +f − m (D)f ) 1 1,μ,λ 1 2 2,μ,λ 2 3 3,μ,λ 3 2 2 2 k L k L k L 1−γ C 30C k −2 −1 (1−γ) −2 −1 + 48 ( ) max{(μ δ + λ β) ,μ δ + λ β} 24d 24d γ γ 30C −2 −1 (1−γ) −2 −1 ≤ 18 max{(μ δ + λ β) ,μ δ + λ β} 24d 1−γ C 30C k k −2 −1 (1−γ) −2 −1 + 48 ( ) max{(μ δ + λ β) ,μ δ + λ β} . 24d 24d γ γ 2 0 Using that 2C ≤ C ,1 − γ ∈ (0, 1) and that ≥ 3, leads to 24d 2 2 2 f − m (D)f +f − m (D)f +f − m (D)f 1 1,μ,λ 1 2 2 2,μ,λ 2 2 3 3,μ,λ 3 2 k+1 k+1 k+1 L L L 30C k −2 −1 (1−γ) −2 −1 ≤ 18 max{(μ δ + λ β) ,μ δ + λ β} 24d C 30C k+1 0 0 −2 −1 (1−γ) −2 −1 1−γ + 24 max{(μ δ + λ β) ,(μ δ + λ β) } 24d 24d γ γ 5 Page 36 of 50 B. Raitae ˘ tal. 6C 30C k+1 0 0 k −2 −1 (1−γ) −2 −1 ≤ ( ) max{(μ δ + λ β) ,μ δ + λ β} 24d 24d γ γ 24C 30C k+1 0 0 k −2 −1 (1−γ) −2 −1 + ( ) max{(μ δ + λ β) ,μ δ + λ β} 24d 24d γ γ 30C k+1 k+1 −2 −1 (1−γ) −2 −1 = ( ) max{(μ δ + λ β) ,μ δ + λ β}. 24d This concludes the proof. Now with the inductive procedure of reducing regions of Fourier space concentration in hand, we turn to the proof of the lower bound in Theorem 2 and the argument for Proposi- tion 1.2. Proof of the lower bound in Theorem 2 and proof of Proposition 1.2 We argue in two steps, ﬁrst ﬁxing the free parameters and then exploiting the boundary conditions. Step 1: Choice of parameters and proof of Proposition 1.2. We seek to invoke per Lemma 4.12 expressing the bounds in terms of our energies, i.e. setting δ = E (χ ; F) := el per per −1 −2 inf E (u, χ ), β = E (χ ),and λ = μ . Moreover, we use the notation u∈D F el surf per per per E (χ ; F) := E (χ ; F) + E (χ ). (32) el surf By virtue of Lemma 4.6 we obtain that for f := χ indeed ∂ f ≤ δ. It follows i i,i i i −1 j =1 ˙ directly that also ∇f ≤ β and therefore the conditions in Lemma 4.7 are fulﬁlled. i TV i=1 As a consequence, the above iteration in Lemma 4.12 is applicable. With this in mind, we choose μ = for some α> 0 to be speciﬁed. Therefore, 2α−1 λ = λ = per per per and thus since without loss of generality E (χ ; F) := inf E (u, χ ) + E (χ ) ≤ 1 u∈D F el surf (having the upper bound from Proposition 4.3 in mind), we deduce 30C k 2 k −2α per (1−γ) f − m (D)f ≤ ( ) E (χ ; F) . (33) j j,μ,λ j 2 L 24d j =1 We further choose k k (2+k)α−1 λ = (2 2d + 1μ) λ = M , k 0 where M = M(d) = 2 2d + 1 > 2. This is admissible in the sense of the assumptions in λ 4dμ k 2 Lemma 4.9 since = 2 2d + 1μ ∈ ( , 1). λ 2 k−1 2 1−4μ Next, for α = α() ∈ (0, 1) and > 0 we choose k ∈ N to be given by log M 1 + (1 − 2α)| log |+ log 2 | log | k := ≤ . log M α| log |− log M α − | log | This ensures that λ ≤ . In what follows, we will choose the parameters , α such that log M α 4 ≤ which implies k ≤ . Exploiting the discrete Fourier transform, then yields | log | 2 α 2 2 2 f − m (D)f = |f (ξ )| =f − f , j j,μ,λ j 2 j j j 2 L L ξ ∈Z \{0} Scaling for the Two-State Problem and a T Structure Page 37 of 50 5 and hence, by (33) results in the estimate 30C 0 α 2 −2α per (1−γ) f − f ≤ E (χ ; F) . j j 2 24d j =1 4 α Using that (1 − γ) ≥ 1 − γ ,weset γ := ∈ (0, 1) which leads to the bound α 8 96d 1 2 24d − −2α per f − f ≤ 30C 8 α E (χ ; F) . j j 0 j =1 Next, we ﬁx the parameter α> 0: Observing that for any ν> 0 there exists α > 0such that for α ∈ (0,α ) it holds that 96d 96d − −1 −1 −1−ν −1−ν α = exp 96d log(α )α ≤ exp α = exp(C(ν)α ), (34) νe 2+ν we choose α =| log | for all ∈ (0, ) with > 0 still to be chosen. In particular, for 0 0 log M > 0 sufﬁciently small, such that | log | ≥ 2log M,(34) holds and also ≤ holds | log | 2 as required above. ν 1 As a consequence, for ν := ∈ (0, ) we arrive at 4+2ν 2 1 1 2 +ν per 2 2 f − f ≤ exp(c | log | )E (χ ; F) . j j 2 ν j =1 Step 2: Conclusion. In order to conclude the estimate, we derive a lower bound for f − f . For this we recall that f = χ is the j -th diagonal entry of the phase j j j j,j j =1 2 2 indicator χ and hence f − f =χ − χ . Thus, by the mean value condi- j j 2 2 j =1 L L per tion in D , per 2 2 2 | χ − F | =| χ − u | ≤ |u − χ | dx ≤ E (u, χ ), el and, furthermore, as the left hand side is independent of u, per | χ − F | ≤ E (χ ; F). el per per per Overall this implies for the total energy E (χ ; F) = E (χ ; F) + E (χ ) el surf ˆ ˆ ˆ 2 2 2 2 dist (F , K) ≤ |χ − F | dx ≤ 2 |χ − χ | dx + 2 | χ − F | dx 3 3 3 T T T 1 1 per +ν per 2 2 ≤ 2exp(c | log | )E (χ ; F) + 2E (χ ; F) el 1 1 1 +ν per per 2 2 2 ≤ 2exp(c | log | )E (χ ; F) + 2E (χ ; F) 1 1 +ν per 2 2 ≤ 4exp(c | log | )E (χ ; F) . per Finally, solving for E (χ ; F) shows the desired estimate per −4 +ν 4 E (χ ; F) ≥ 2 exp − 2c | log | dist (F , K). 5 Page 38 of 50 B. Raitae ˘ tal. Fig. A.1 Self-similar construction in Proposition A.3. (Color ﬁgure online) The desired claim follows by an application of Lemma 4.5. Appendix A: Branching, Upper Bound for the Two-State Problem for theDivergenceOperator We complement our lower bounds for the compatible two-well problem from Sect. 3 by an upper bound in the case of the divergence operator acting on matrix ﬁelds as introduced in Example 1.1. For simplicity, we only consider square matrices, i.e. m = d,and likebefore in Sect. 3 only consider the case d> 1 in the following. For earlier, closely related, three- dimensional constructions in the context of compliance minimization problems we refer to [24]. While the construction is by now rather “standard” [1, 62, 63, 80], our argument does provide a slightly different perspective, in that, in arbitrary dimension, we can ensure boundary conditions on all faces of the domain =[0, 1] (see the upper bound construc- tion for [76, Theorem 3] for a similar construction for the gradient). In deducing the upper bound for the divergence operator, we ﬁrst provide a construction in a unit cell (Lemma A.1) and iterate this construction (Proposition A.3). This yields a construction which attains the boundary data in two directions. In order to attain these also on the remaining sides we use the ﬂexibility of the wave cone for the divergence operator (Proposition A.5). We remark that in two dimensions there would be no modiﬁcation with respect to the gradient construction since there the curl and divergence only differ by a rotation of 90 degrees. In our unit cell branching construction, we do not work on the level of the potential, but directly consider the problem on the level of the wells. In this context, we recall the com- patibility conditions for laminates formed by the divergence operator which is determined d×d by the associated wave cone: For M ∈ R ,wehavethat M ∈ ker A(ξ ) if and only it holds Mξ = 0. With this in hand, we introduce an auxiliary matrix which will play an important role in d×d our construction: Let A, B ∈ R be such that (B − A)e = 0and let n = e + γ ν for a 1 1 2 unit vector ν perpendicular to e and for some γ ∈ R, γ = 0. We then deﬁne E by 1 2 2 ν E = γ (B − A)ν ⊗ e (A.1) ν 2 1 Scaling for the Two-State Problem and a T Structure Page 39 of 50 5 and the associated “perturbed” matrix A = A + E . By construction, this matrix obeys the ν ν identities ˜ ˜ (A − A)ν = 0,(B − A )n = 0. (A.2) ν ν ˜ ˜ This in particular allows for interfaces of A and A with normal ν and of A and B with ν ν normal n which we will use in our branching construction below. d−2 d×d Lemma A.1 For 0 <l <h ≤ 1, we deﬁne ω =[0,l]×[0,h]×[0, 1] . Let A, B ∈ R be such that (B − A)e = 0 and let F = λA + (1 − λ)B for some λ ∈ (0, 1). Then there exists 1 λ d d×d u : R → R such that div u = 0 in R , u = F for x ∈ (−∞, 0) ∪ (l, ∞). λ 1 Furthermore, there exist χ ∈ BV (ω;{A, B}) and a constant C = C(A, B) > 0 such that for any > 0 the localized energy can be bounded by ˆ ˆ 2 2 E (u, χ ; ω) := |u − χ | dx + |∇χ|≤ C(1 − λ) + 5h. ω ω Proof We consider the following partition of the domain ω into subdomains: λl λl λl (1 − λ)l ω ={x ∈ (0, )},ω ={x ∈ ( , + x )}, 1 1 2 1 2 2 2 2 2h λl (1 − λ)l (1 − λ)l (1 − λ)l ω ={x ∈ ( + x ,λl + x )},ω ={x ∈ (λl + x ,l)}. 3 1 2 2 4 1 2 2 2h 2h 2h Based on this we deﬁne Ax ∈ ω , Ax ∈ ω ∪ ω , 1 3 u(x) = Bx ∈ ω ∪ ω , χ(x) = 2 4 Bx ∈ ω ∪ ω , ⎩ 2 4 A + E x ∈ ω , e 3 (1−λ)l where E is givenin(A.1)for n = e − e . We highlight that u is independent of x e 1 2 k 2h for k ≥ 3. By deﬁnition of E , the characterization of the wave cone (5) for the divergence operator and the remarks on laminates in (A.2), this deﬁnes an divergence-free mapping. Further, as (B − F )e = (A − F )e = 0, the exterior data are attained in x ∈ (−∞, 0) ∪ λ 1 λ 1 1 (l, ∞). To calculate the energy, we observe, that the only contribution to the elastic energy is given in ω . Hence, ˆ ˆ 2 2 E (u, χ ; ω) := |u − χ | dx = |A + E − A| dx el e ω ω 2 3 (1 − λ) λl 2 2 =|E | |ω |=|(B − A)e | . e 3 2 8h As the surface energy is determined by the interfaces between ω and ω ,weobtain(l< h) j k " " 2 2 2 (1 − λ) l (1 − λ) E (χ ; ω) := |∇χ|= h + 2 + h ≤ h + 2h + 1 ≤ 5h. surf 4 4 ω 5 Page 40 of 50 B. Raitae ˘ tal. Fig. A.2 Setting of Lemma A.2 This shows the claim. As for analogous constructions for the gradient, we will use this unit cell as a building block in order to achieve a self-similar construction attaining the boundary data on two directions. For this to be admissible in the sense of an A-free map, we rely on the following lemma. It shows that for ﬁrst order operators corners in which interfaces meet do not give rise to singularities. ∞ d n ∞ d m Lemma A.2 Let d, n, m ∈ N and let A(D) = A ∂ : C (R ; R ) → C (R ; R ) be j j j =1 a homogeneous, linear operator of degree one with symbol A given in (3) and let , j = 1,...,l, be a polygonal set (the set is deﬁned as the intersection of half spaces) with outer unit normal n such that • R = , j =1 • the two sets , have one common face (l + 1 = 1), j j +1 • and such that they meet in one point, i.e. ={x }, j 0 j =1 cf. Fig. A.2. Assume further that B ∈ R , j = 1,...,l, are such that B − B ∈ ker A(n ). j j j +1 j Then the map u(x) = B for x ∈ j j is A-free. Proof First we note that u is indeed well-deﬁned by the properties of , further we notice n d ∞ d m that for M ∈ R and U ⊂ R a Lipschitz domain it holds for ϕ ∈ C (R ; R ) ˆ ˆ ˆ d d ∗ t t d−1 M · (A(D)) ϕdx = M · ∂ (A ϕ)dx = M · (A ϕ)n dH k k k k U U ∂U k=1 k=1 d−1 = A(n)M · ϕdH . ∂U With this it holds ˆ ˆ ˆ l l ∗ ∗ d−1 u · (A(D)) ϕdx = B · (A(D)) ϕdx = A(n )B · ϕdH . j j j R ∂ j j j =1 j =1 Scaling for the Two-State Problem and a T Structure Page 41 of 50 5 Moreover on ∂ ∩ ∂ it holds that n =−n and thus by the assumptions on B j j +1 j +1 j j ˆ l ˆ ∗ d−1 u · (A(D)) ϕdx = A(n )(B − B ) · ϕdH = 0. j j j +1 R ∂ ∩∂ j j +1 j =1 As ϕ was arbitrary the claim follows. With Lemma A.2 in hand, we now iterate the unit cell-construction from Lemma A.1,as illustrated in Fig. A.1. d n Proposition A.3 Let d, n ∈ N. Let =[0, 1] , let A, B ∈ R be such that B − A ∈ , cf. div (5), and let F = λA + (1 − λ)B for some λ ∈ (0, 1). Let E be as in (12). Then there exist u : d d×d d 2 R → R and χ ∈ BV (;{A, B}) with div u = 0 in R and u = F for (x ,x )/ ∈[0, 1] λ 1 2 such that for any ∈ (0, 1) and any N ∈ N E (u, χ ) ≤ C( + N ) for some constant C = C(A, B, λ) > 0. Proof Without loss of generality we may assume (B − A)e = 0, i.e. B − A ∈ ker A(e ) for 1 1 1 1 A(D) = div. Let θ ∈ ( , ). We argue symmetrically in the upper and lower half of the cube, 4 2 1 d−2 i.e. we give the construction of u on [0, 1]×[ , 1]×[0, 1] and deﬁne u on the lower half by symmetry. We deﬁne for N ∈ N and for j ∈ N θ 1 1 − θ y = 1 − ,l = ,h = y − y = θ . j j j j +1 j 2 2 N 2 Furthermore, let j ∈ N be the maximal j ∈ N such that l <h .Weset 0 j j d−2 ω = (kl ,y ) +[0,l ]×[0,h ] ×[0, 1] , j,k j j j j j j +1 for k ∈{0, 1,...,N 2 − 1}, j ∈{0, 1,...,j };for k ∈{0, 1,...,N 2 − 1}, j = j + 1we 0 0 set j +1 d−2 ω = (kl ,y ) +[0,l ]×[0, ] ×[0, 1] . j +1,k j +1 j +1 j +1 0 0 0 0 d−2 Let u , χ in [0,l ]×[0,h ]×[0, 1] be given by Lemma A.1 for j = 1,...,j .Further, j j j j 0 in the layer j = j + 1 we interpolate with the desired boundary data by a cut-off argument: To this end, we introduce the cut-off function φ :[0, ∞) →[0, 1] and the proﬁle h :[0, 1]→ [0, ∞) by setting ⎪ 1 t ∈[0, ], (1 − λ)t t ∈[0,λ), 1 3 φ(t) = −4t + 3 t ∈ ( , ), h(t ) = 2 4 λ(1 − t) t ∈[λ, 1]. 0 t ≥ , j +1 d−2 We consider the function u ˜ :[0,l ]×[0, ]×[0, 1] deﬁned via j +1 j 0 0 2 2l 2x x j +1 2 1 u ˜ (x) =− φ h ((A − B)e ) ⊗ e j +1 2 1 j +1 j +1 0 0 θ θ l j +1 0 5 Page 42 of 50 B. Raitae ˘ tal. 2x x 2 1 + φ h (A − B) + F . j +1 θ l j +1 The associated phase indicator is deﬁned by χ (x) = χ (x )A + χ (x )B. j +1 (0,λl ) 1 (λl ,l ) 1 0 j j j 0 0 0 We note that χ (x) = h ( )(A − B) + F and moreover for x < it holds u ˜ (x) = j +1 λ 2 j +1 0 l 2 0 j +1 χ (x) and for x > correspondingly u ˜ (x) = F . Furthermore, for x ∈{0,l },we j +1 2 j +1 λ 1 j +1 0 0 0 x 2x x 1 2 1 know h( ) = 0 and thus (u ˜ (x) − F )e = φ( )h (A − B)e = 0. j +1 λ 1 1 0 j +1 l 0 l j +1 θ j +1 0 0 With the help of this construction we meet the prescribed data for x ≥ 1, and we can deﬁne u in the upper half of the full cube: u (x − (kl ,y )) x ∈ ω , j j j j,k u(x) = u ˜ (x − (kl ,y )) x ∈ ω . j +1 j j +1 j +1,k 0 0 0 0 For the lower half of the cube we argue similarly, mirroring the unit cell construction of Lemma A.1, i.e. instead of E we consider E .Wedeﬁne χ in [0, 1] analogously. e −e 2 2 We note, that this deﬁnes a divergence free mapping, as all the laminations are in com- patible directions as (B − A)e = 0 and by the choice of E in (A.1). Lemma A.2 shows, 1 e that this function is divergence-free even thought interfaces meet in corners. Moreover, we j +1 can bound the energy in the ω cells for any k ∈{1,...,N 2 − 1}: j +1,k 2l 2x x j +1 2 1 2 0 |˜ u (x) − χ (x)| = φ h ((A − B)e ⊗ e ) j +1 j +1 2 1 0 0 j +1 j +1 0 0 θ θ l j +1 2x x 2 1 − (φ − 1)h (A − B) j +1 θ l j +1 2x j +1 2 0 2 ≤ C(A, B, λ) φ + 1 , 2j +2 j +1 0 0 θ θ j +1 and, since l ≥ h and θ ∼ h , j +1 j +1 j +1 0 0 0 ˆ ˆ j +1 2 0 2 j +1 |˜ u − χ (x)| dx ≤ Cl ( φ (t )dt + θ ) j +1 j +1 j +1 0 0 0 j +1 j +1 θ 0 0 d−2 θ [0,l ]×[0, ]×[0,1] 0 j +1 0 2 j +1 0 j +1 ≤ C( + l θ ) j +1 j +1 j +1 ≤ C . j +1 Furthermore, the surface energy is bounded by E (χ ; ω ) ≤ C(A, B, λ)h . surf j +1 j +1,k j +1 0 0 0 and can extend it to be F for (x ,x )/ ∈ Overall, we have a function deﬁned on [0, 1] λ 1 2 [0, 1] . The energy then can be bounded by j +1 j +1 0 N 2 −1 0 3 E (u, χ ) ≤ 2 E (u ,χ ; ω ) ≤ C N 2 ( + h ) j j j,k j j =0 k=0 j =0 Scaling for the Two-State Problem and a T Structure Page 43 of 50 5 j +1 j +1 0 2 0 h 1 1 j j j j ≤ C ( + ) = C ( ) + N (2θ) h l N 4θ j j j =0 j =0 ≤ C( + N ). Remark A.4 We remark that in the situation of the divergence operator, there are situations d×d with substantially more ﬂexibility than for the gradient: If for the two wells A, B ∈ R it does not only hold that (B − A)e = 0 but also that (B − A)e = 0, there would not be 1 2 any elastic energy contribution involved. In this situation, for the above construction, we would only have contributions to the surface energy, as then E = 0 and hence A + E = e e 2 2 A ∈{A, B}. In particular, for boundary data which are only attained on two directions this would yield a linear scaling law in . For curl free mappings as in gradient inclusions, this is not possible, as the direction of lamination is unique in that case, i.e. V is at most rot,λ one-dimensional. As a last auxiliary step towards the upper bound construction from Theorem 1,inorder to achieve the exterior data on all sides of the unit cube, we adapt the branching construction similarly as in [75], as the construction from Proposition A.3 does not yet satisfy F at, e.g., x = 0. Thus, we combine Proposition A.3 with a further domain splitting for which we split [0, 1] into different regions. In each region, we prescribe a different direction for the branching construction from Proposition A.3. To this end, we use that the choice of e in the above results was arbitrary and we also can choose any other direction e for j ∈{2,...,d}. Combined with compatibility conditions at the resulting interfaces, this will allow us to deduce the desired branching construction. Proposition A.5 Under the same assumptions as in Proposition A.3 there exist u : R → d×d R , χ ∈ BV (;{A, B}) and a constant C = C(A, B, λ) > 0 such that u ∈ D for F = F λ λA + (1 − λ)B (λ ∈ (0, 1)) and for any ∈ (0, 1) and N ∈ N it holds E (u, χ ) ≤ C( + N ). Proof For simplicity, we ﬁrst carry out the details for the case d = 3 and then only comment on the changes in the case of arbitrary dimension. We split [0, 1] into the following four parts and use different branching directions in each part: Let 1 1 1 1 1 ± 3 ={x ∈[0, 1] :±(x − ) ≥ 0, −|x − |≤ −|x − |}, 2 2 3 2 2 2 2 2 1 1 1 1 1 ={x ∈[0, 1] :±(x − ) ≥ 0, −|x − |≤ −|x − |}, 3 3 2 2 2 2 2 2 + − and consider the upper ( )and lower( ) halves separately as in the proof of Proposi- j j tion A.3. + + Next, we deﬁne u in and u in using Proposition A.3: The function u is given 2 3 2 2 3 by the function from Proposition A.3 above, whereas u is obtained from u by exchanging 3 2 roles of e and e , i.e. the branching is done in e direction and we use E instead of E . 2 3 3 e e 3 2 For ν ∈{e ,e } the error matrix E is givenin(A.1). We then deﬁne the overall deformation 2 3 ν + + u in ∪ by 2 3 u (x) x ∈ , u(x) = u (x) x ∈ . 3 5 Page 44 of 50 B. Raitae ˘ tal. Fig. A.3 The three-dimensional branching construction to achieve the boundary data on all sides. The shaded regions are the diagonal interface at x = x 2 3 with the interfaces of A + E and A + E marked in orange − − This construction is depicted in Fig. A.3. As above u is deﬁned in the lower halves , 2 3 by symmetry. We claim that this overall construction is divergence-free. In the individual regions and this follows by Proposition A.3. It thus remains to discuss the compatibility at the interface x = x . Since all other values of u are given by (matching domains in which) 2 3 u ∈{A, B}, it sufﬁces to discuss the compatibility of the error matrices E and E at this e e 2 3 interface. To this end, we however note that (E − E )ζ = 0for all ζ ∈ span(e ,e ), i.e. e e 2 3 3 2 also this interface is admissible. This shows that u indeed deﬁnes a divergence free map. The upper bound for the elastic and surface energies from Proposition A.3 remains valid, thus yielding the claimed estimate which concludes the proof of the proposition. In order to show the d -dimensional result, we split [0, 1] into 2d − 2 regions := d 1 1 1 1 1 {x ∈[0, 1] :±(x − ) ≥ 0, −|x − |= min −|x − |} for j = 2,...,d and j j 2≤k≤d k 2 2 2 2 2 argue as above. Finally, with Proposition A.5 in hand, we immediately obtain the proof of the upper bound construction from Theorem 1 for the divergence operator. Proof of the upper bound in Theorem 1 In order to deduce the upper bound of Theorem 1,we choose N ∼ , which shows the claim. Remark A.6 (Generalizations) Building on the ideas from the gradient case and the ones from above one can formulate (rather restrictive) conditions, allowing for similar constructions for more general linear, constant coefﬁcient differential operators. A key difﬁculty here consists in the “hard form” of the prescribed boundary conditions. When considering “softer forms” of these, as for instance in [66], constructions for general constant coefﬁcient operators with the desired boundary conditions would be feasible under much more general conditions by using Fourier theoretic arguments as in [74]. We do not pursue these ideas here but postpone this to possible future work. Scaling for the Two-State Problem and a T Structure Page 45 of 50 5 Appendix B: On the Role of the Divergence Operator Following [28], in this section we highlight the relevance of the divergence operator which also partially motivates our discussion of the scaling law for the T problem from Sect. 1.3. To this end, we ﬁrst recall that considering any ﬁrst order homogeneous constant co- j ∞ d n ∞ d m efﬁcient differential operator A(D) = A ∂ : C (R ; R ) → C (R ; R ), we can j =1 n m×d rewrite A(D)u = div ω (u) with a linear map ω : R → R . Indeed, let us deﬁne 1 1 n m×d ω : R → R ,ω (x) = A x . 1 1 k i=1,...,m, ik j =1,...,d k=1 ∞ d n Let u ∈ C (R ; R ), then it holds A(D)u = (div ◦ω )(u) for the row-wise divergence. α α Indeed this can be generalized for higher order operators A(D)u = A ∂ u,where |α|=k α m×n A ∈ R are coefﬁcient matrices, as follows. For this we denote the space of symmetric k d d tensors on R by Sym(R ,k).Let the k-th order divergence be given as k ∞ d m d ∞ d m div : C (R ; R ⊗ Sym(R ,k)) → C (R ; R ), (B.3) (div u) := ∂ ...∂ u . j i i ji ...i 1 k 1 k 1≤i ≤···≤i ≤d 1 k For k = 1 this is exactly the row-wise divergence as mentioned above. Remark B.1 This deﬁnition is natural in that sense that this operator (up to a sign) is the adjoint of the k-th derivative D . n m d The linear map ω : R → R ⊗ Sym(R ,k) then takes the form l=1 (ω (x)) := A x (B.4) k ji ...i 1 k ∞ d n and by this choice it holds A(D)u = (div ◦ω )(u) for any u ∈ C (R ; R ) and ker ω = k k I = ker A . In what follows, we omit the k dependence of ω = ω in the notation. A k |α|=k With this in hand, it is possible to bound the energy for a general homogeneous linear operator A(D) (of order k) by the corresponding energy for the (k-th order) divergence (cf. [28, Appendix] for the corresponding qualitative result in the case k = 1). n d Proposition B.2 Let d, n, k ∈ N, K ⊂ R , and let ⊂ R be a bounded Lipschitz domain. Let A(D) be a k-th order homogeneous linear differential as in (2) and the elastic and sur- face energies be given by (16) and (17). Moreover let ω = ω be the linear transformation in (B.4) and div the generalized k-th order divergence in (B.3) with the corresponding en- k k div div ergies E , E . Then there exist constants C ,C > 0 such that for any χ ∈ BV (; K), 1 2 el surf F ∈ R div E (χ ; F) ≥ C E (ω(χ ); ω(F )), el 1 el div E (χ ) ≥ C E (ω(χ )). surf 2 surf Moreover if A(D) is cocanceling (and thus ω is injective), it also holds for all u ∈ D , χ ∈ BV (; K) (cf.(10)) k k div div E (u, χ ) + E (χ ) ∼ E (ω(u), ω(χ )) + E (ω(χ )). el surf surf 5 Page 46 of 50 B. Raitae ˘ tal. Proof In order to obtain the desired result, we use the pointwise bound |u − χ|≥ C|ω(u) − ω(χ)| and consider the adapted boundary data: For any u ∈ D with D denoting the set F F k d from (10), the composition ω(u) satisﬁes div ω(u) = A(D)u = 0in R and ω(u) = ω(F ) d div in R \ . In other words, it holds that ω(u) ∈ D for the divergence operator and bound- ω(F ) k k div div ary data ω(F ). For the elastic energy (denoting by E , D the energy and domain for el ω(F ) the divergence operator) this implies ˆ ˆ 2 2 div E (u, χ ) = |u − χ | dx ≥ C |ω(u) − ω(χ)| dx = CE (ω(u), ω(χ )), (B.5) el el ˆ ˆ 2 2 E (χ ; F) ≥ C inf |ω(u) − ω(χ)| dx ≥ C inf |ω(u) − ω(χ)| dx. el u∈D F div u:ω(u)∈D ω(F ) We emphasize that, in general, this only yields lower bound inequalities since ω is possibly not injective and thus there may be deformations u with A(D)u = 0and u = F outside but still fulﬁlling ω(u) = ω(F ) outside (see the example in Sect. B.1 below). Replacing d m d div now ω(u) by a general function w : R → R ⊗ Sym(R ,k) such that w ∈ D yields ω(F ) 2 div E (χ ; F) ≥ C inf |w − ω(χ)| dx = CE (ω(χ ); ω(F )). el el div w∈D ω(F ) Furthermore, as |∇χ|≥ c|∇(ω(χ ))|, we can also bound the surface energy ˆ ˆ div E (χ ) = |∇χ|≥ c |∇(ω(χ ))|= cE (ω(χ )). (B.6) surf surf In the case of a cocanceling operator ω is injective and we also have the bounds |u − χ|≤ C|ω(u) − ω(χ)|, |∇χ|≤ C|∇(ω(χ ))|, thus in (B.5)and (B.6) also the matching upper bounds hold, which concludes the proof. As a consequence, lower bounds for the divergence operator often also imply lower bounds for more general operators. A particular setting (see [28]) for instance arises in the three state problem with K ={A ,A ,A } being such that A − A ∈ / . In this case also 1 2 3 j k A ω(K) consists of three states which is a result of the fact that the kernel of ω is given by d d ker(ω) = ker A = ker A(e ) = I . j A j =1 j =1 In particular, if we ﬁnd a T structure for a general linear, homogeneous, constant coefﬁcient, ﬁrst order differential operator A(D) such that it is mapped to the T structure in Sect. 4,we can exploit the same lower bound as for the divergence operator. In addition to the relevance of the divergence operator for applications, this argument serves as an additional motivation for focusing particularly on the divergence operator in this article. Moreover with Proposition 3.8 in mind, also for pairwise non super-compatible wells, we can assume without loss of generality that I ={0} and thus ω in injective. B.1 Comparison of the Two-State Problem for the Divergence Operator In this section, we discuss the comparison between the general two-state problem for linear, homogeneous differential operators and the one for the (k-th order) divergence operator. In particular, this yields yet another proof of the compatible case in Theorem 1. Scaling for the Two-State Problem and a T Structure Page 47 of 50 5 In the calculations from the ﬁrst part of Sect. B, we notice that for an injective map ω we can also bound the quantities E (u, χ ), E (χ ) from above with the corresponding term el surf in which u, χ are replaced by ω(u), ω(χ); hence for I ={0} it holds div E (u, χ ) ∼ E (ω(u), ω(χ )). We here emphasize that this only holds on the level of ﬁxed u, χ and that after the mini- mization in u, this does not necessarily yield a two-sided comparison of the energies any more. Indeed, while the lower bound estimates always hold (cf. Proposition B.2), this may not be true for the upper bound estimates. In fact, even if I ={0}, we can at the moment div not exclude that there may be w ∈ D \ ω(D ). We postpone a further discussion of this ω(F ) to future work. The advantage of I ={0} is that we do not lose wells in that sense that for I ={0} A A also ker ω ={0} and thus, ω is injective. As seen above in Sect. 3.4 for two wells A, B ∈ R , A − B/ ∈ I we can restrict to A(D) which fulﬁlls I ={0}. This implies that for A ˜ two compatible wells, which are not super-compatible, in deducing lower scaling bounds, we can use the corresponding lower bounds of the divergence operator as we do not lose information. Example B.3 In concluding this section, we give an example of an operator which is not cocanceling. Considering d = 2, n = 3, m = 1and A(D)u = ∂ u + ∂ u , 1 2 2 3 3 1×2 implies that ω : R → R , ω(x) = (x ,x ) and ker(ω) = span(e ) = I . 2 3 1 A 2 3 The reduced operator A(D) would act on mappings taking values only in {0}× R ⊂ R . Acknowledgements A.R. and C.T. gratefully acknowledge support through the Heidelberg STRUCTURES Excellence Cluster which is funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foun- dation) under Germany’s Excellence Strategy EXC 2181/1 - 390900948. Funding Note Open Access funding enabled and organized by Projekt DEAL. Data Availability Data sharing not applicable to this article as no datasets were generated or analysed during the current study. Declarations Competing Interests The authors have no competing interests to declare that are relevant to the content of this article. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. 5 Page 48 of 50 B. Raitae ˘ tal. References 1. Chan, A., Conti, S.: Energy scaling and branched microstructures in a model for shape-memory alloys with SO(2) invariance. Math. Models Methods Appl. Sci. 25(06), 1091–1124 (2015) 2. Rüland, A., Tribuzio, A.: On the energy scaling behaviour of a singularly perturbed Tartar square. Arch. Ration. Mech. Anal. 243(1), 401–431 (2022) 3. Garroni, A., Nesi, V.: Rigidity and lack of rigidity for solenoidal matrix ﬁelds. Proc. R. Soc. Lond., Ser. A, Math. Phys. Eng. Sci. 460(2046), 1789–1806 (2004). https://doi.org/10.1098/rspa.2003.1249 4. Palombaro, M., Ponsiglione, M.: The three divergence free matrix ﬁelds problem. Asymptot. Anal. 40(1), 37–49 (2004) 5. Müller, S., Šverák, V.: Convex integration for Lipschitz mappings and counterexamples to regularity. Ann. Math. 157(3), 715–742 (2003) 6. Kirchheim, B., Müller, S., Šverák, V.: Studying nonlinear PDE by geometry in matrix space. In: Geo- metric Analysis and Nonlinear Partial Differential Equations, pp. 347–395. Springer, Berlin (2003) 7. Székelyhidi, L. Jr: The regularity of critical points of polyconvex functionals. Arch. Ration. Mech. Anal. 172(1), 133–152 (2004) 8. De Lellis, C., Székelyhidi, L. Jr.: The Euler equations as a differential inclusion. Ann. Math. 170(3), 1417–1436 (2009) 9. De Lellis, C., Székelyhidi, L. Jr.: The h-principle and the equations of ﬂuid dynamics. Bull. Am. Math. Soc. 49(3), 347–375 (2012) 10. Kuiper, N.H.: On C -isometric imbeddings. II. In: Indagationes Mathematicae (Proceedings), vol. 58, pp. 683–689. Elsevier, Amsterdam (1955) 11. Nash, J.: C isometric imbeddings. Ann. Math. 60(3), 383–396 (1954) 1,α 12. Conti, S., De Lellis, C., Székelyhidi, L. Jr.: h-principle and rigidity for C isometric embeddings. In: Nonlinear Partial Differential Equations, pp. 83–116. Springer, Berlin (2012) 13. De Lellis, C., Székelyhidi, L. Jr.: On h-principle and Onsager’s conjecture. Newsl. - Eur. Math. Soc. 95, 19–24 (2015) 14. Ball, J.M., James, R.D.: Fine phase mixtures as minimizers of energy. In: Analysis and Continuum Mechanics, pp. 647–686. Springer, Berlin (1989) 15. Ball, J.M., James, R.D.: Proposed experimental tests of a theory of ﬁne microstructure and the two-well problem. Philos. Trans. R. Soc. Lond. A, Math. Phys. Eng. Sci. 338(1650), 389–450 (1992) 16. Bhattacharya, K., Firoozye, N.B., James, R.D., Kohn, R.V.: Restrictions on microstructure. Proc. R. Soc. Edinb., Sect. A, Math. 124(5), 843–878 (1994) 17. Müller, S., Šverák, V.: Convex integration with constraints and applications to phase transitions and partial differential equations. J. Eur. Math. Soc. 1(4), 393–422 (1999) 18. Müller, S.: Variational models for microstructure and phase transitions. In: Calculus of Variations and Geometric Evolution Problems, pp. 85–210. Springer, Berlin (1999) 19. Rüland, A.: The cubic-to-orthorhombic phase transition: rigidity and non-rigidity properties in the linear theory of elasticity. Arch. Ration. Mech. Anal. 221(1), 23–106 (2016) 20. Palombaro, M.: Rank-(n − 1) convexity and quasiconvexity for divergence free ﬁelds. Adv. Calc. Var. 3(3), 279–285 (2010) 21. Palombaro, M., Smyshlyaev, V.P.: Relaxation of three solenoidal wells and characterization of extremal three-phase H-measures. Arch. Ration. Mech. Anal. 194(3), 775–822 (2009) 22. Kohn, R.V., Wirth, B.: Optimal ﬁne-scale structures in compliance minimization for a uniaxial load. Proc. R. Soc. A, Math. Phys. Eng. Sci. 470(2170), 20140432 (2014) 23. Kohn, R.V., Wirth, B.: Optimal ﬁne-scale structures in compliance minimization for a shear load. Com- mun. Pure Appl. Math. 69(8), 1572–1610 (2016) 24. Potthoff, J., Wirth, B.: Optimal ﬁne-scale structures in compliance minimization for a uniaxial load in three space dimensions (2021). arXiv:2111.06910 25. Kohn, R.V., DeSimone, A., Otto, F., Müller, S.: Recent analytical developments in micromagnetics. Sci. Hyst. 2, 269–381 (2006) 26. DeSimone, A., Knüpfer, H., Otto, F.: 2-d stability of the Néel wall. Calc. Var. Partial Differ. Equ. 27(2), 233–253 (2006) 27. De Philippis, G., Palmieri, L., Rindler, F.: On the two-state problem for general differential operators. Nonlinear Anal. 177, 387–396 (2018) 28. Sorella, M., Tione, R.: The four-state problem and convex integration for linear differential operators (2021). arXiv:2107.10785 29. Skipper, J.W.D., Wiedemann, E.: Lower semi-continuity for A-quasiconvex functionals under convex restrictions (2019). arXiv:1909.11543 30. Bhattacharya, K.: Microstructure of Martensite. Oxford Series on Materials Modelling. Oxford Univer- sity Press, London (2003) Scaling for the Two-State Problem and a T Structure Page 49 of 50 5 31. Bhattacharya, K.: Comparison of the geometrically nonlinear and linear theories of martensitic transfor- mation. Contin. Mech. Thermodyn. 5(3), 205–242 (1993) 32. Warner, M., Terentjev, E.M.: Liquid Crystal Elastomers, vol. 120. Oxford University Press, London (2007) 33. Cesana, P., Della Porta, F., Rüland, A., Zillinger, C., Zwicknagl, B.: Exact constructions in the (non- linear) planar theory of elasticity: from elastic crystals to nematic elastomers. Arch. Ration. Mech. Anal. 237(1), 383–445 (2020) 34. Lamy, X., Lorent, A., Peng, G.: Rigidity of a non-elliptic differential inclusion related to the Aviles–Giga conjecture. Arch. Ration. Mech. Anal. 238(1), 383–413 (2020) 35. Lamy, X., Lorent, A., Peng, G.: On a generalized Aviles-Giga functional: compactness, zero-energy states, regularity estimates and energy bounds (2022). arXiv:2203.05418 36. Tartar, L.: Compensated compactness and applications to partial differential equations. In: Nonlinear Analysis and Mechanics: Heriot-Watt Symposium, vol. 4, pp. 136–212 (1979) 37. DiPerna, R.J.: Compensated compactness and general systems of conservation laws. Trans. Am. Math. Soc. 292(2), 383–420 (1985) 38. Tartar, L.: The compensated compactness method applied to systems of conservation laws. In: Systems of Nonlinear Partial Differential Equations, pp. 263–285. Springer, Berlin (1983) 39. Murat, F., Tartar, L.: H-convergence. In: Topics in the Mathematical Modelling of Composite Materials, pp. 21–43. Springer, Berlin (2018) 40. Müller, S.: Rank-one convexity implies quasiconvexity on diagonal matrices. Int. Math. Res. Not. 1999(20), 1087–1095 (1999) 2×2 41. Faraco, D., Székelyhidi, L. Jr.: Tartar’s conjecture and localization of the quasiconvex hull in R .Acta Math. 200(2), 279–305 (2008) 42. Guerra, A., Rai¸ ta, B., Schrecker, M.R.I.: Compensated compactness: continuity in optimal weak topolo- gies. J. Funct. Anal. 283(7), 109596 (2022) 43. Arroyo-Rabasa, A., De Philippis, G., Rindler, F.: Lower semicontinuity and relaxation of linear-growth integral functionals under PDE constraints. Adv. Calc. Var. 13(3), 219–255 (2020) 44. Rai¸ ta, ˘ B.: Potentials for A-quasiconvexity. Calc. Var. Partial Differ. Equ. 58(3), 105 (2019) 45. Guerra, A., Rai¸ ta, ˘ B., Schrecker, M.: Compensation phenomena for concentration effects via nonlinear elliptic estimates (2021). arXiv:2112.10657 46. Rai¸ ta, ˘ B.: A simple construction of potential operators for compensated compactness (2021). arXiv:2112. 47. Behn, L., Gmeineder, F., Schiffer, S.: On symmetric div-quasiconvex hulls and divsym-free L - truncations (2021). arXiv:2108.05757 48. Fonseca, I., Müller, S.: A-Quasiconvexity, lower semicontinuity, and young measures. SIAM J. Math. Anal. 30(6), 1355–1390 (1999) 49. Kristensen, J., Rai¸ ta, ˘ B.: An introduction to generalized Young measures. Max-Planck-Institut für Math- ematik in den Naturwissenschaften Leipzig 45 (2020) 50. Conti, S., Gmeineder, F.: A-Quasiconvexity and Partial Regularity (2020). arXiv:2009.13820 51. Gmeineder, F., Lewintan, P., Neff, P.: Optimal incompatible Korn-Maxwell-Sobolev inequalities in all dimensions (2022). arXiv:2206.10373 52. De Philippis, G., Rindler, F.: On the structure of A-free measures and applications. Ann. Math. 184(3), 1017–1039 (2016) 53. Arroyo-Rabasa, A., De Philippis, G., Hirsch, J., Rindler, F.: Dimensional estimates and rectiﬁability for measures satisfying linear PDE constraints. Geom. Funct. Anal. 29(3), 639–658 (2019) 54. Breit, D., Diening, L., Gmeineder, F.: On the trace operator for functions of bounded A-variation. Anal. PDE 13(2), 559–594 (2020) 55. Dacorogna, B.: Direct Methods in the Calculus of Variations, vol. 78. Springer, Berlin (2007) 56. Dacorogna, B., Marcellini, P.: Implicit Partial Differential Equations, vol. 37. Springer, Berlin (2012) 57. Pedregal, P.: Parametrized Measures and Variational Principles. Springer, Berlin (1997) 58. Chaudhuri, N., Müller, S.: Rigidity estimate for two incompatible wells. Calc. Var. Partial Differ. Equ. 19(4), 379–390 (2004) 59. De Lellis, C., Székelyhidi, L. Jr.: Simple proof of two-well rigidity. C. R. Math. 343(5), 367–370 (2006) 60. Friesecke, G., James, R.D., Müller, S.: A theorem on geometric rigidity and the derivation of nonlinear plate theory from three-dimensional elasticity. Commun. Pure Appl. Math. 55(11), 1461–1506 (2002) 61. Lamy, X., Lorent, A., Peng, G.: Quantitative rigidity of differential inclusions in two dimensions (2022). arXiv:2208.08526 62. Kohn, R.V., Müller, S.: Branching of twins near an austenite—twinned-martensite interface. Philos. Mag. A 66(5), 697–715 (1992) 63. Kohn, R.V., Müller, S.: Surface energy and microstructure in coherent phase transitions. Commun. Pure Appl. Math. 47(4), 405–435 (1994) 5 Page 50 of 50 B. Raitae ˘ tal. 64. Barroso, A.C., Matias, J., Santos, P.M.: Differential inclusions and A-quasiconvexity. Mediterr. J. Math. 3(14), 1–14 (2017) 65. Van Schaftingen, J.: Limiting sobolev inequalities for vector ﬁelds and canceling linear differential op- erators. J. Eur. Math. Soc. 15(3), 877–921 (2013) 66. Choksi, R., Kohn, R.V., Otto, F.: Domain branching in uniaxial ferromagnets: a scaling law for the minimum energy. Commun. Math. Phys. 201(1), 61–79 (1999) 67. Conti, S., Diermeier, J., Melching, D., Zwicknagl, B.: Energy scaling laws for geometrically linear elas- ticity models for microstructures in shape memory alloys. ESAIM Control Optim. Calc. Var. 26, 115 (2020) 68. Kohn, R.V.: Energy-driven pattern formation. In: International Congress of Mathematicians, vol. 1, pp. 359–383 (2007) 69. Conti, S., Kohn, R.V., Misiats, O.: Energy minimizing twinning with variable volume fraction, for two nonlinear elastic phases with a single rank-one connection. Math. Models Methods Appl. Sci. 32(08), 1671–1723 (2022). https://doi.org/10.1142/S0218202522500397 70. Knüpfer, H., Kohn, R.V.: Minimal energy for elastic inclusions. Proc. R. Soc. A, Math. Phys. Eng. Sci. 467(2127), 695–717 (2011) 71. Knüpfer, H., Kohn, R.V., Otto, F.: Nucleation barriers for the cubic-to-tetragonal phase transformation. Commun. Pure Appl. Math. 66(6), 867–904 (2013) 72. Rüland, A.: A rigidity result for a reduced model of a cubic-to-orthorhombic phase transition in the geometrically linear theory of elasticity. J. Elast. 123(2), 137–177 (2016) 73. Capella, A., Otto, F.: A quantitative rigidity result for the cubic-to-tetragonal phase transition in the geometrically linear theory with interfacial energy. Proc. R. Soc. Edinb., Sect. A, Math. 142(2), 273–327 (2012) 74. Capella, A., Otto, F.: A rigidity result for a perturbation of the geometrically linear three-well problem. Commun. Pure Appl. Math. 62(12), 1632–1669 (2009) 75. Rüland, A., Tribuzio, A.: On the energy scaling behaviour of singular perturbation models involving higher order laminates (2021). arXiv:2110.15929 76. Rüland, A., Tribuzio, A.: On scaling laws for multi-well nucleation problems without gauge invariances (2022). arXiv:2206.05164 77. Rüland, A., Taylor, J.M., Zillinger, C.: Convex integration arising in the modelling of shape-memory alloys: some remarks on rigidity, ﬂexibility and some numerical implementations. J. Nonlinear Sci. 29(5), 2137–2184 (2019) 78. Tartar, L.: Some remarks on separately convex functions. In: Microstructure and Phase Transition, pp. 191–204. Springer, Berlin (1993) 79. Chlebík, M., Kirchheim, B.: Rigidity for the four gradient problem. J. Reine Angew. Math. 2002(551), 1–9 (2002) 80. Otto, F., Viehmann, T.: Domain branching in uniaxial ferromagnets: asymptotic behavior of the energy. Calc. Var. Partial Differ. Equ. 38(1), 135–181 (2010) 81. Diermeier, J.: Domain branching in geometrically linear elasticity (2013) 82. Winter, M.: An example of microstructure with multiple scales. Eur. J. Appl. Math. 8(2), 185–207 (1997) 83. Chipot, M.: The appearance of microstructures in problems with incompatible wells and their numerical approach. Numer. Math. 83(3), 325–352 (1999) 84. Grafakos, L.: Classical Fourier Analysis. Graduate Texts in Mathematics, vol. 249. Springer, Berlin (2014) Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional afﬁliations.

Journal

Acta Applicandae Mathematicae – Springer Journals

Published: Apr 1, 2023

Keywords: A\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal{A}$\end{document}-Free inclusions; Two-well problem; Divergence T3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$T_{3}$\end{document}; Phase transformation; 35Q74; 35F05; 74N05; 74G99

References

Energy scaling and branched microstructures in a model for shape-memory alloys with SO(2)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$SO(2)$\end{document} invariance

Chan, A.; Conti, S.
On the energy scaling behaviour of a singularly perturbed Tartar square

Rüland, A.; Tribuzio, A.
Rigidity and lack of rigidity for solenoidal matrix fields

Garroni, A.; Nesi, V.
The three divergence free matrix fields problem

Palombaro, M.; Ponsiglione, M.
Convex integration for Lipschitz mappings and counterexamples to regularity

Müller, S.; Šverák, V.
Studying nonlinear PDE by geometry in matrix space

Kirchheim, B.; Müller, S.; Šverák, V.
The regularity of critical points of polyconvex functionals

Székelyhidi, L.
The Euler equations as a differential inclusion

De Lellis, C.; Székelyhidi, L.
The h\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$h$\end{document}-principle and the equations of fluid dynamics

De Lellis, C.; Székelyhidi, L.
On C1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$C^{1}$\end{document}-isometric imbeddings. II

Kuiper, N.H.
C1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$C^{1}$\end{document} isometric imbeddings

Nash, J.
h-principle and rigidity for C1,α\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$C^{1, \alpha }$\end{document} isometric embeddings

Conti, S.; De Lellis, C.; Székelyhidi, L.
On h-principle and Onsager’s conjecture

De Lellis, C.; Székelyhidi, L.
Fine phase mixtures as minimizers of energy

Ball, J.M.; James, R.D.
Proposed experimental tests of a theory of fine microstructure and the two-well problem

Ball, J.M.; James, R.D.
Restrictions on microstructure

Bhattacharya, K.; Firoozye, N.B.; James, R.D.; Kohn, R.V.
Convex integration with constraints and applications to phase transitions and partial differential equations

Müller, S.; Šverák, V.
Variational models for microstructure and phase transitions

Müller, S.
The cubic-to-orthorhombic phase transition: rigidity and non-rigidity properties in the linear theory of elasticity

Rüland, A.
Rank-(n−1)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$(n-1)$\end{document} convexity and quasiconvexity for divergence free fields

Palombaro, M.
Relaxation of three solenoidal wells and characterization of extremal three-phase H-measures

Palombaro, M.; Smyshlyaev, V.P.
Optimal fine-scale structures in compliance minimization for a uniaxial load

Kohn, R.V.; Wirth, B.
Optimal fine-scale structures in compliance minimization for a shear load

Kohn, R.V.; Wirth, B.
Recent analytical developments in micromagnetics

Kohn, R.V.; DeSimone, A.; Otto, F.; Müller, S.
2-d stability of the Néel wall

DeSimone, A.; Knüpfer, H.; Otto, F.
On the two-state problem for general differential operators

De Philippis, G.; Palmieri, L.; Rindler, F.
Comparison of the geometrically nonlinear and linear theories of martensitic transformation

Bhattacharya, K.
Exact constructions in the (non-linear) planar theory of elasticity: from elastic crystals to nematic elastomers

Cesana, P.; Della Porta, F.; Rüland, A.; Zillinger, C.; Zwicknagl, B.
Rigidity of a non-elliptic differential inclusion related to the Aviles–Giga conjecture

Lamy, X.; Lorent, A.; Peng, G.
Compensated compactness and applications to partial differential equations

Tartar, L.
Compensated compactness and general systems of conservation laws

DiPerna, R.J.
The compensated compactness method applied to systems of conservation laws

Tartar, L.
H-convergence

Murat, F.; Tartar, L.
Rank-one convexity implies quasiconvexity on diagonal matrices

Müller, S.
Tartar’s conjecture and localization of the quasiconvex hull in R2×2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathbb{R}^{2\times 2}$\end{document}

Faraco, D.; Székelyhidi, L.
Compensated compactness: continuity in optimal weak topologies

Guerra, A.; Raiţă, B.; Schrecker, M.R.I.
Lower semicontinuity and relaxation of linear-growth integral functionals under PDE constraints

Arroyo-Rabasa, A.; De Philippis, G.; Rindler, F.
Potentials for A\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal{A}$\end{document}-quasiconvexity

Raiţă, B.
A\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal{A}$\end{document}-Quasiconvexity, lower semicontinuity, and young measures

Fonseca, I.; Müller, S.
On the structure of A\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal{A}$\end{document}-free measures and applications

De Philippis, G.; Rindler, F.
Dimensional estimates and rectifiability for measures satisfying linear PDE constraints

Arroyo-Rabasa, A.; De Philippis, G.; Hirsch, J.; Rindler, F.
On the trace operator for functions of bounded A\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathbb{A}$\end{document}-variation

Breit, D.; Diening, L.; Gmeineder, F.
Rigidity estimate for two incompatible wells

Chaudhuri, N.; Müller, S.
Simple proof of two-well rigidity

De Lellis, C.; Székelyhidi, L.
A theorem on geometric rigidity and the derivation of nonlinear plate theory from three-dimensional elasticity

Friesecke, G.; James, R.D.; Müller, S.
Branching of twins near an austenite—twinned-martensite interface

Kohn, R.V.; Müller, S.
Surface energy and microstructure in coherent phase transitions

Kohn, R.V.; Müller, S.
Differential inclusions and A\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$\mathcal{A}$\end{document}-quasiconvexity

Barroso, A.C.; Matias, J.; Santos, P.M.
Limiting sobolev inequalities for vector fields and canceling linear differential operators

Van Schaftingen, J.
Domain branching in uniaxial ferromagnets: a scaling law for the minimum energy

Choksi, R.; Kohn, R.V.; Otto, F.
Energy scaling laws for geometrically linear elasticity models for microstructures in shape memory alloys

Conti, S.; Diermeier, J.; Melching, D.; Zwicknagl, B.
Energy-driven pattern formation

Kohn, R.V.
Energy minimizing twinning with variable volume fraction, for two nonlinear elastic phases with a single rank-one connection

Conti, S.; Kohn, R.V.; Misiats, O.
Minimal energy for elastic inclusions

Knüpfer, H.; Kohn, R.V.
Nucleation barriers for the cubic-to-tetragonal phase transformation

Knüpfer, H.; Kohn, R.V.; Otto, F.
A rigidity result for a reduced model of a cubic-to-orthorhombic phase transition in the geometrically linear theory of elasticity

Rüland, A.
A quantitative rigidity result for the cubic-to-tetragonal phase transition in the geometrically linear theory with interfacial energy

Capella, A.; Otto, F.
A rigidity result for a perturbation of the geometrically linear three-well problem

Capella, A.; Otto, F.
Convex integration arising in the modelling of shape-memory alloys: some remarks on rigidity, flexibility and some numerical implementations

Rüland, A.; Taylor, J.M.; Zillinger, C.
Some remarks on separately convex functions

Tartar, L.
Rigidity for the four gradient problem

Chlebík, M.; Kirchheim, B.
Domain branching in uniaxial ferromagnets: asymptotic behavior of the energy

Otto, F.; Viehmann, T.
An example of microstructure with multiple scales

Winter, M.
The appearance of microstructures in problems with incompatible wells and their numerical approach

Chipot, M.

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Abstract

Journal

Recommended Articles

References

Our policy towards the use of cookies