Andrej Cherkaev, cherk@math.utah.edu Optimal multimaterial composites: Bounds and structures The paper suggests a method for funding exact bounds for the effective conductivity moduli of multimaterial composites. These bounds expand and refine Hashin–Shtrikman and Nesi bounds. We prove that the fields in the materials within optimal structures vary in restricted domains and take this into account, obtaining more restricted bounds. The new bounds are solutions of a formulated relaxed finite-dimensional constrained optimization problem. For two-dimensional conducting three-material composites, bounds for effective conductivity are explicitly computed. These bounds are exact: Three-material isotropic microstructures of extremal conductivity are found that realize the bounds for all values of parameters. The optimal structures are laminates of a finite rank, their parameters vary with the volume fractions and they experience two topological transitions: For large values of material of minimal conductivity, its subdomain percolates (is connected), for intermediate values of that fraction, no material forms a connected domain, and for small values of that fraction, the domain of intermediate material percolates. Another type of isotropic optimal three–material structures is the “wheel assemblages” that replaces the Hashin–Shtrikman coated circles. Optimal multimaterial composites: Bounds and structures Andrej Cherkaev Department of Mathematics, University of Utah The talk is based on a joint project with Grzegorz Dzierżanowski (Warsaw Polytechnic University) and Nathan Briggs (University of Utah). An optimal design problem • Given three materials – an expensive and strong one, – a cheap and weak one, – void, find their layout in a fixed loaded from the boundary domain, that minimizes the stress energy, i.e. maximizes the stiffness with a fixed total cost (weight). If the layout is periodic, the problem is to find the best (stiffest) composite. The two-material problem has been solved decades ago (Hashin & Shtrikman 1963, Lurie & Ch 1982, Gibiansky & Ch 1986 (elasticity) by variational “translation” method” (Kohn and Strang, 1983, Lurie & Ch, 1986, Milton 1982, 2001) of bounds. Multimaterial design The standard variational method, however, does not give the correct result for multimaterial problem. Example (Hashin-Shtrikman bound, 2d conductivity) n 1 mi , kL k1 i1 ki k1 lim kL f (k1) m1 0 Optimal bounds for effective properties and structures of multimaterial composites were studied in the last 30 years, and some partial results were obtained in Milton 1983, Lurie & Ch., 1986, Milton & Kohn 1988, Nesi 1995, Albin, Ch, Nesi 2004, Liping Lui 2009. et al. This talk is based on the papers Ch, and Dzergzanovsky, 2013, Briggs, Ch, and Dzergzanovsky, 2013. Earlier development of the approach: Ch, 09, Ch. & Zhang, 2011, Ch. 2012 Design of minimal compliance • Given two materials – the expensive strong and the cheap weak ones -- and the void, find their layout in a fixed domain loaded from the boundary that minimizes the stress energy, i.e. maximizes the stiffness. • The problem is reduced to a variational with a nonconvex piecewise quadratic Lagrangian: inf F(,g)dx f (uT n) , : T , 0, F min(W i ( ) gi ), W i kiTr( 2 ) /2 k S u, RT R, W 22 (O). k1 k 2 k3 , g1 1, g3 0, g2 , Quasiconvex envelope – local problem of the best structure. 1 QF( ) inf | | F( )dx, dx, T , 0. is periodic. Due to non-convexity of F, the solution - optimal stress in a structure – oscillates in an infinite fine scale, alternating the values that are called supporting points or supports of the Young’s measure. 1 inf | | ( )dx, supporting points. Compatibility: The supporting points of the envelope must be compatible: Any two neighboring stresses satisfy det(1 2 ) 0. Fields in optimal structures and bounded and ordered Bounds problem multiwell variational problem inf u U F(u, )dx, F minW i (u,ki i i Materials (wells) is an optimal structure are identified by the range fields (gradients). Fields in materials are ordered and bounded Material 1 Material 2 Material 3 Alessandrini-Nesi inequality det e 0 s12 s22 d12 d22 (S 2 D2 ) x, if det E 0 0 An open question: The range of fields that support the quasiconvex envelope Ch, 2009: (for optimal isotropic composites) det(e(x) e(x') 0 if x 1, x' 3,det E 0 0 Comment on types of optimal partitions Fields in optimal structures are bounded. Therefore three or more materials cannot meet in isolated points Optimal structures either do not have meeting points (coated spheres) Or the meeting points are dense (laminates) Expected structure of optimal fields • 1. Piece-wise constancy of the fields -- all structures are compared and “the best” is chosen. The stress depends on only several parameters and does not depend on a structural point. It is either piece-wise constant, or it varies within a manifold so that this does not change the effective stiffness. • 2. This constancy forces a structure to become fractals with infinitely thin ligaments, etc. We a priori expect that the optimal solution does not exist but the minimizing sequences tend to a distribution and the fields become differentially constrained Young’s measures. • The boundary conditions are satisfied: det A B 0, det SA A SB B 0. Technique for calculating/estimating the quasiconvex envelope Upper bound: Necessary conditions on fields: Structural (Weierstrass type) variations – K.Lurie Direct methods: Building minimizing sequences. Lower bound: Sufficient conditions: Translation method + inequalities To find the lower bound for the quasiconvex envelope, we -- replace differential constrains on the minimizers with several weaker algebraic conditions and -- constract the convex envelope, or solve a finite-dimensional constrained optimization problem Sufficient conditions: Translation method + Point-wise constraints • Constraints (A) det det : T , 0, is 1- periodic W i ( ) t det = T (ki I tT) (B) , C() : det( - ) = 0 - compatibility (C) If k1 k2 ..., then det( - 0 ) 0 1 - optimality cond. The algebraic expression for the lower bound depends on -- which constraints (B) and (C) are active -- singularity of the matrix (ki I tT) for some i To constraint (B) Necessary condition for for all supporting points (B) , C(/ ) : det( - ) = 0 det 0 on the lines of discontinuity at a multiscale boundary, k c i i , c i 1, c i 0, i i i:ik i:ik To constraint (c) Conditions on supports in relation to the average field (C) If k1 k2 ..., then det( - 0 ) 0 1 - optimality cond. • Example: if a strong material envelopes the structure of forms an “exterior layer”, the inequality is obvious. Admissible region of stress in mat. 1 Optimal structures Knowing the positions of optimal fields (supports of Young measure), we build laminate structure that possess these fields in the layers. In the construction, we satisfy the boundary conditions between the layers. Example. Two materials 1 Zones optimality of the pure materials and zone of non-quasiconvexity Observe, that both inequalities (B) and (C) are “naturally” satisfied in optimal structures 2 a)-c): the trace of stress tensor in P1 is constant everywhere Tr (x) 21 x 1 c) The stress in P2 is isotropic d) One component of stress in P1 and P2 is equal to the external stress. (x) 2I x 2 det( (x) 0 ) 0 x . Optimal geometries are not unique Optimal three-material structures • Below, we present the types of optimal structures, in dependence of their volume fraction and anisotropy of the external field. • Each composites minimizes the stored stress energy in a given average stress field. • The optimality of the structures is proven by sufficient conditions (bounds). • The results for general elasticity case (two constants) were obtained in 2013. • Optimal three-material structures depend to the anisotropy of the external field and to the volume fractions. • The structure is independent of the magnitude of the external field. • In each point of an optimal structure, the stress satisfies sufficient conditions of optimality. Color code: orange – strong material blue -- weak material, white -- void. Moderate m1, isotropic Equivalent optimal structures Periodic assemblage Stress in P1 is always in rank-one connection with the stress in P3 Stress in P2 is isotropic det( (x)) 0 x 1, (x) I x 2 Moderate m1, anisotropic stress det( (x)) 0 x 1, det( (x) 0 ) 0 x 2 Large m1, close to isotropy (Hashin-Shtrikman or Translation bounds) Isotropic stress or pure shear Vertical stress is bigger The structures are not unique. However, in all optimal structures -- the trace of the stress is constant in P1 -- the stress is isotropic in P2. Tr (x) 21 x 1, (x) 2I x 2 W1() t1 det = T (k1I t1T) Large m1, anisotropic The stress in P1 either is in rank-one connection with P3, or with rank-one connection with external field det( (x)) det((x) 0 ) 0 x 1 A periodic assemblage A different type of structure and a different expression for the optimal energy Small m1, isotropic stress det( (x)) 0 x 1, Tr((x)) 22 x 1 Small m1, anisotropy varies Both types of structure correspond to the same bound det( (x)) 0 x 1, Tr( (x)) 22 x 1 W2 () t2 det = T (k2I t2T) Structures at a glance Close to isotropy large m1 moderate m1 small m1 Anisotropic Optimal structures are not unique: Assemblage elements, “wheels”, vary depending on the fractions of materials. Ch. 2011 Assemblage elements: the topology depends on m1 m1>m11 Hashin Shtrikman bound. k1 (white) is connected m12<m1<m11 No connected phases m1<m12 k2 (gray) is connected Distribution of the structures • An optimal distribution of the materials in an optimal structure depends on the costs of the materials. We assume that cost g1 of P1 is equal one, • g1=1, • cost of P3 (void) is equal zero, g3=0, • and the cost g2 of P2 varies. • The problem is to minimize the stored energy plus the total cost by optimization of the structures’ volume fractions and anisotropy level. min W str (,m1,m2 ) m1 gm2 m1,m2 • It the stress is very high, only stiff material P1 is optimal (m1=1). Multimaterial happen when the stress density is low enough and it is beneficially to mix the materials. Distribution of optimal structures (general case) Notice that the optimal structures for more intensive fields correspond to P1-P2 composites. After the first transition, P1 always enters the 1-2-3 structures (gray areas) via 1-3 laminate. Taxi cab geometry of the quasiconvex envelope These rules can be heuristically generalized to more materials, to 3d problem, to polycrystalls, etc. We would obtain the upper bound (perhaps exact) of the quasiconvex envelope Supporting sets: boundaries of domains of optimality of the pure materials 2 . 1 1 2 3 Obtained by structural variations and verified by sufficient conditions 1 2 Region of optimality of the second well Dependence on the cost g. The region is bounded by two symmetric ellipses and two dependent on g symmetric hyperbolas with the independent of g asymptotes. Asymptotic: “blue” region of optimality of pure second material shrinks to a point: Hashin-Shtrikman case. The graph shows optimal structures in dependence of eigenvalues of the external stress field. Large stresses correspond to the optimality of pure stiff material. Gray fields show three–material structures. First transition • The effective stiffness of an optimal 1-3 structure in an isotropic field is equal to the stiffness of P2, the costs are equal too. W 2 (I) g2 QW13 (I) m1, g2 2k1 k1 k2 = = = Optimal structures at this point are not unique. Well 2 touches the 1-3-quasiconvex envelope Volume fraction of m2 immediately after the transition The amount of P2 depends on the relative area of the 13 optimal structure that can be occupied with the isotropic field of a given density. This area can be filled by P2 without changing the total stiffness and cost. m2 decreases when the external field is too large, too small, or too anisotropic. It is equal one in the point where the second well touches the quasiconvex envelope The second asymptotic: fields and geometries These structures correspond to a range of total volume fractions. Second transition • After the first transition, P1 always enters the1-2-3 structure through a 13 laminate. • Second transition occurs when the effective stiffness and cost of a 1-3laminate structure in a unidirectional stress becomes equal to the stiffness and cost of 1-3-laminate. = Optimal structures are not unique. In 1-2-3 laminate, one can replace any portion of P2 with an equivalent 1-3 laminate Low cost of P2: fields and geometries After the second transition, well 2 separates two components of 1-3quasiconvex envelope. The optimality domain of the pure P2 cuts the domain of the optimal composites into two. After the transition, the three materials do not meet in an optimal structure. For smaller fields, P2 is an enveloping material (P3 – inclusions) , for larger fields P2 forms inclusions, P1 –an envelope. Quasiconvex envelope. First transition. Quasiconvex envelope, first quadrant Nonlinear part of the envelope 1 4k1 | 1 2 | 2 2k1 (| 1 | | 2 |), 2k1(| 1 | | 2 |) 2 1, QW 2 1 2k1 (12 2 2 ) otherwise First quadrant T 2k1 (1 2 ) 1, i 0 1 4k1R R 2 2k1 I, 2 2k1 2k1 (1 2 ) 1, i 0 Tr 4k1Tr 4 2k1 Euler - Lagrange equation is identically zero. Cantilever, loaded at the middle (computations by Grzegorz Dzierżanowski and Nathan Briggs) We expect novel unexpected suboptimal designs when penalties for mixing is introduced. Conclusion Quotient of design problems: Multimaterial design -----------------------------Two-material design Color TV = ------------------------Black-and-white TV Thank you!