In recent years anisotropic materials have been finding their way into aerospace applications traditionally

47th AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Confere 1-4 May 2006, Newport, Rhode Island AIAA 2006-2250 Anisotropic Materials which Can be Modeled by Polyconvex Strain Energy Functions Nayden Kambouchev, Raul Radovitzky, Javier Fernandez Department of Aeronautics and Astronautics Massachusetts Institute of Technology 77 Massachusetts Avenue, Cambridge, MA 02139-4307 Polyconvexity is a property of the strain energy function which guarantees existence of solutions to boundary value problems. It is a local condition which can be verified relatively easy. There is an ever growing set of known polyconvex functions which constantly expands the set of material models known to posses the property of polyconvexity. Finding the proper polyconvex function to fit to a newly developed material can be a difficult and tedious task. To alleviate this difficulty we introduce the concept of minimal polyconvex function set which guarantees that any real material with given anisotropy can be fitted to some strain energy function containing terms only from the minimal function set. In cases when the minimal function sets are not known, their closest analogs are listed together with suggestions on future developments which may allow their completion to minimal polyconvex function sets. I. Introduction In recent years anisotropic materials have been finding their way into aerospace applications traditionally reserved for metals. Single crystals, multi-dimensional fiber composites and polymers have replaced metals in the never-ending pursuit for lighter and stronger materials. Most of these new materials are capable of sustaining very large deformations inaccessible to ordinary metals and alloys. The properties of the material models, which guarantee existence of solutions to boundary value problems in large deformation regimes, are non-trivial. In small deformation regimes, the existence of solutions can be argued by continuity. In large deformation regimes such an argument cannot be applied and therefore mathematically sane models are required if the embarrassment of obtaining numerical solutions to problems which in reality do not have solutions is to be avoided. In elasticity, several known conditions on the strain energy functions guaranteeing existence of solutions have been proposed. 1 The simplest condition is convexity, but convexity implies not only existence, but also uniqueness and thus precludes many interesting physical phenomena such as buckling and phase transformations. A necessary and almost sufficient condition for the existence of minimizers to variational formulations is the condition of quasiconvexity in the sense of Morrey. 2 Unfortunately, the quasiconvexity condition is a global integral condition which is difficult to verify. A more attractive condition which comes between convexity and quasiconvexity is polyconvexity. 3 Polyconvexity is a local condition which implies quasiconvexity, but can be verified relatively easy and does not render impossible the physical phenomena of interest. Examples of isotropic strain energy functions which are polyconvex have been known for a long time. 3 For example, the Arruda-Boyce model of polymeric materials is polyconvex, while some of Rivlin s and Ogden s models do not satisfy the polyconvexity condition. 4 Recently, anisotropic polyconvex strain energy functions have attracted significant attention. A complete review with new results about transversely isotropic polyconvex energy functions has been given by Schröder and Neff. 5 In addition to being one of the simplest anisotropies, transverse isotropy is well suited for describing composites with fibers running in one or several directions. Models for orthotropic and cubic materials have also been proposed. 6, 7 Institute for Soldier Nanotechnology, Massachusetts Institute of Technology, 77 Massachusetts Avenue, NE47-587, Cambridge, MA 02139 Copyright 2006 by Massachusetts Institute of Technology. American Published by Institute the American of Aeronautics Institute of Aeronautics and Astronautics and Astronautics, Inc., with permission. 1 of 6

In this work we introduce the novel concept of minimal polyconvex function sets. These function sets can be used to determine general polyconvex strain energy functions (one for each anisotropy class) possessing the fewest number of free parameters to be matched with experimental data and having the following property: if there exists a polyconvex strain energy function which matches the material linear elasticity constants, then the general polyconvex strain energy function also matches these constants for some appropriately chosen values of its free parameters. This property allows for the search of polyconvex strain energy functions for a given material to be limited just to the functional form specified by the general polyconvex strain energy function describing the anisotropy of that material. In the next section we introduce the formal definition of polyconvexity along with other useful properties of the strain energy functions and proceed to formulate the procedure for fitting strain energy functions to material parameters which could have been obtained either experimentally or in any other scientifically sound manner. After defining minimal polyconvex function sets we continue with examples of such sets or their closest known analogs in three different spatial symmetries. The limitations of the sets are also discussed together with some suggestions for future developments. II. Strain Energy Density Functions for General Non-linear Elastic Materials The material objectivity condition that the material response needs to remain invariant under transformations from the special orthogonal group SO 3 (so that the response is independent of the orientation of the laboratory reference frame) implies that any acceptable strain energy function ψ : R 9 R has the form 1 ψ(f) = φ(f T F) = φ(c), (1) where F is the deformation gradient, C = F T F is the right Cauchy-Green tensor, R is the set of all real numbers and φ : R 9 R is another real valued function. We shall call both ψ and φ strain energy functions and use them interchangeably depending on which one has the more convenient arguments to work with. The second Piola-Kirchhoff stress tensor S and the tangent moduli C are given by 1 S = 2 φ C, (2) C = 4 2 φ C 2. (3) For a strain energy function to represent a real material it must satisfy two conditions: Stress free reference configuration i.e. S(1) = 0. (4) The tangent moduli C in absence of deformation must be the same as the conventional small strain moduli C i.e. C(1) = C. (5) Condition (4) imposes at most 6 constraints since S is symmetric, while condition (5) imposes at most 21 due to the various symmetries in the tangent moduli. 8 A fully anisotropic material will impose 6 + 21 = 27 constraints, but in the presence of material symmetries this number can be significantly reduced. For example, in the isotropic case to be discussed later there are 3 constraints one from the stress free requirement (4) and two from the tangent moduli matrix, equation (5). The strain energy function ψ : R 9 R is polyconvex 3 if and only if it is representable by a convex function ψ : R 19 R, such that ψ(f) = ψ(f, adj(f), det(f)). (6) It follows from this definition that polyconvexity is preserved under addition and multiplication by positive real numbers (but not under subtraction!). This property allows for a large variety of polyconvex strain energy functions to be assembled even from a very limited set of functions which are initially known to be polyconvex. Usually polyconvex strain energy functions are assembled from polyconvex monomials of tensorial invariants relevant to the specific material symmetry under consideration. 2 of 6

Let the strain energy function φ be a linear combination with non-negative coefficients of N functions φ n which are known to be polyconvex i.e. φ(c) = N α n φ n (C), (7) n=1 where α n 0. For each specific material we shall try to find appropriate constants α n such that the material can be represented by the polyconvex strain energy function φ. Conditions (4) and (5), respectively, take the form 4 2 N n=1 N n=1 φ n α n (1) = 0, (8) C 2 φ n α n (1) = C. (9) C2 For any given set of functions φ n the values of the derivatives in the previous two equations can be computed and therefore the question of interest is whether for a given set of constants C the system of equations (8,9) has a solution in non-negative real numbers α n. The number of unknowns N must be at least as large as the number of independent constraints, but not necessarily equal, and therefore this is a linear programming problem. If the problem has a solution, it can be found by standard techniques. 9 Therefore it would be desirable to have a set of functions {φ n } N n=1 such that this linear programming problem has a solution for all physically relevant elastic moduli matrices C. In addition, it is also desirable to have a set with a minimum total number of functions N in order to minimize the cost of the solution search and the model complexity. We call such a set a minimal set of polyconvex functions. A minimal set contains at least as many functions as the number of independent equations in the system (8, 9), but might contain more due to the requirement that the solution of the system must be in nonnegative numbers. We analyze the availability and the properties of the minimal polyconvex function sets in the next section. Satisfaction of equations (8,9) does not necessarily make the strain energy function φ a good representation of the real properties of the material of interest. In fact it only guarantees good agreement with the actual material response at small strains, but this is always a necessary first step towards a realistic nonlinear material model. For better fits at large strains one should impose additional conditions at deformation gradients F different than 1 making use of non-minimal polyconvex function sets. In the examples to follow the role of the functions φ n will be played primarily by monomials of tensorial invariants of the right Cauchy-Green stress tensor C and the structural tensors describing the material anisotropy. The structural tensors M 1, M 2,... of a given spatial symmetry completely characterize it and provide a convenient mathematical tool for generation of strain energy functions respecting the symmetry. More specifically, the effect of the structural tensors on the strain energy function can be captured in the following two statements: The relation φ(c, M n ) = φ(q T CQ, Q M n ), where the componentwise definition of the star operation is (Q M n ) ik = Q ij...q kl (M n ) jl, holds for all elements of the special orthogonal group, Q SO 3 making φ an isotropic scalar-valued tensor function. The relation φ(c, M n ) = φ(q T CQ, M n ) holds for all elements of the material symmetry group, Q G M making φ an anisotropic scalar-valued tensor function. The structural tensors of all spatial symmetries have been derived by Zheng. 10 For example, the structural tensor for a transversely isotropic material is M t = a a, (10) where a is a unit vector in the preferred direction, 5 usually chosen to coincide with one of the coordinate vectors e 1, e 2 and e 3. An example of a polyconvex function respecting the transverse isotropic symmetry is φ 0 = J4 k, with k 1, (11) where J 4 = C : M t. 5 It can be shown that all functions respecting a given symmetry can be written as functions of the invariants of the right Cauchy-Green stress tensor and the structural tensors corresponding to the symmetry of interest. 11 3 of 6

III. Examples A. Isotropic Material Isotropic materials do not require structural tensors for their representation 11 invariants are available: and therefore only three I 1 = tr(c), (12) I 2 = tr(cof(c)), (13) I 3 = det(c). (14) Examples of possible polyconvex functions of these invariants are I k 1, I k 2, I 1 3 and log(i 3 ) (Hartmann and Neff 4 give a large list of additional possibilities). Equations (8) imposes one linearly independent constraint, while equation (9) places two: isotropic materials have two independent elastic constants, for example, C 1111 and C 1122 because the third nonzero constant is dependent on the other two: C 1112 = 1 2 (C 1111 C 1122 ). It can be easily verified that the set of functions φ 1 = I 1, φ 2 = I 3 and φ 3 = log(i 3 ) guarantees the existence of non-negative α n, n = 1, 2, 3 as materials of practical interest always satisfy 0 ν = C 1122 C 1122 + C 1111 1 2 where ν is the Poisson ratio. As this set contains as many functions as the number of constrains, it is a minimal polyconvex function set. This minimal set is not unique and other minimal sets exist, for example, φ 1 = I 1, φ 2 = I 1 log(i 3 ) and φ 3 = I 3 log(i 3 ). The implications of the existence of the polyconvex function set above are non-trivial: for every isotropic material there exist a polyconvex strain energy function capturing correctly the material behavior at least in small deformations such that all proper boundary value problems (even large deformation problems) have a solution. The strain energy function is of the type (15) φ = α 1 I 1 + α 2 I 2 + α 3 ( log(i 3 )), (16) but other function sets may lead to better representations for specific materials especially at large deformations. This will also be true for the strain energy functions proposed in the other examples to follow. B. Cubic Material A single structural tensor for cubic symmetry is the fourth order tensor M c, 7 M c = e 1 e 1 e 1 e 1 + e 2 e 2 e 2 e 2 + e 3 e 3 e 3 e 3, (17) which gives in addition to the invariants (12,13,14), the three invariants L 4 = C : M c : C, (18) L 5 = C : M c : C 2, (19) L 6 = C 2 : M c : C 2. (20) The authors have been able to demonstrate that the family L k 4, k 1 is polyconvex. 7 Direct evaluation of the derivatives of L k 4 reveals that any function set containing isotropic functions and powers of L 4 can be replaced by the set φ 1 = I 1, φ 2 = I 3, φ 3 = log(i 3 ) and φ 4 = L 4. This replacement is done in the sense that if the set provides a solution with nonnegative α n for given values of the elastic constants C 1111, C 1122 and C 1212 then the set φ n, n = 1, 2, 3, 4 also provides non-negative solution α n, n = 1, 2, 3, 4 for the same elastic constants. The set φ n, n = 1, 2, 3, 4 is minimal in the sense that it is minimal for all sets of polyconvex functions with cubic symmetry known to the authors (it contains four functions as many as the number of constraints). Unfortunately this set (and therefore any known set of polyconvex functions) does not guarantee solution with nonnegative α n for all possible values of the elastic constants. Specifically, we have shown that a function φ with nonnegative coefficients α n may be obtained 7 if 1 2 A = 2C 1212 C 1111 C 1122 1, (21) 4 of 6

where A is the material anisotropy ratio. 12 Interestingly this property is satisfied by V, Nb, Cr, Mo, W, all of which have body-centered cubic lattices and belong to groups Vb and VIb of the periodic table. These materials whose elastic constants satisfy this property are guaranteed to have a polyconvex strain energy function of the type φ = α 1 I 1 + α 2 I 2 + α 3 ( log(i 3 )) + α 4 L 4. (22) As a significant number of materials of practical interest have anisotropy ratios outside of the range (21) there is a need for new polyconvex functions. These new functions must contain at least one of the invariants L 4, L 5 and L 6. C. Transversely Isotropic Material As stated earlier, the structural tensor for transverse isotropy is M t = a a where a is a unit vector in the preferred direction. We choose a = e 1 for convenience. The additional invariants are: 5 J 4 = C : M t (23) J 5 = C 2 : M t. (24) Schröder and Neff 5 give a large list of polyconvex terms involving J 4 and J 5 such as J4 2, K 1 = J 5 I 1 J 4 + I 2 and K 3 = I 1 J 4 J 5. Equation (8) imposes two conditions on the value of α n (one in the preferred direction and one in a direction perpendicular to it), while equation (9) provides another five because a linear transversely isotropic material has five independent elastic constants. Therefore a minimal set of functions will consist of at least 7 functions. Currently there is no such set known to the authors. Furthermore, a significant number of the polyconvex terms listed by Schröder and Neff are independent in the sense that their contribution to the system of equations (8,9) cannot be replaced by contributions of several other terms with positive coefficients. An example of possible elimination of dependent terms would be the following: if the function set contains all of K 1, K1 2 and K3 1 then any term αk2 1 with α > 0 can be replaced with αk 1 + (α/3)k1 3 because (K 2 1) C 2 (K 2 1) C 2 = (K 1) C + 1 3 = 2 (K 1 ) C 2 + 1 3 (K 3 1) C 2 (K 3 1) C 2, (25). (26) Even though eliminations of this kind led to a small four member set in the case of cubic anisotropy, in the case of transverse isotropy only few eliminations can be performed and the remaining set contains a large number of functions. The set of the first 34 functions proposed by Schröder and Neff, 5 which is circumstantially not guaranteed to be large enough, can thus be reduced it to an irreducible set with 15 members. In addition, there seems to be no simple condition such as (15) or (21) in the case of transverse isotropy. IV. Conclusions Minimal polyconvex function sets provide a convenient systematic method for fitting of experimental material parameters to models which ensure existence of solutions of boundary value problems. While such sets are known to exist for isotropic materials, the existence of minimal function sets for anisotropic materials is an open question. Even for cubic anisotropy, which is one of the anisotropies closest to full isotropy, the additional condition (21), which must be satisfied by the material parameters for the fit to be successful, is too restrictive and does not cover the whole range of materials of practical interest. The main reason for that is hidden in the limited number of known polyconvex functions for each anisotropy which does not allow for the construction of a sufficiently rich function search space. Although the set of known polyconvex functions for transverse isotropy is large, it is unclear and not easy to check whether this set can cover the whole range of allowable values of the material parameters. This points to the need of developing new families of polyconvex functions for each type of anisotropy. 5 of 6

Aknowledgments This research was supported by the U.S. Army through the Institute for Soldier Nanotechnologies, under Contract DAAD-19-02-D-0002 with the U.S. Army Research Office. The content does not necessarily reflect the position of the Government, and no official endorsement should be inferred. References 1 Marsden, J. and Hughes, T., Mathematical Foundations of Elasticity, Dover Publications, 1993. 2 Morrey, C., Multiple Integrals in Calculus of Variations, Springer, 1966. 3 Ball, J., Convexity Conditions and Existence Theorems in Nonlinear Elasticity, Archive for Rational Mechanics and Analysis, Vol. 63, 1977, pp. 337 403. 4 Hartmann, S. and Neff, P., Polyconvexity of Generalized Polynomial-type Hyperelastic Strain Energy Functions for Near-incompressibility, International Journal of Solids and Structures, Vol. 40, 2003, pp. 2767 2791. 5 Schröder, J. and Neff, P., Invariant Formulation of Hyperelastic Transverse Isotropy Based on Polyconvex Free Energy Functions, International Journal of Solids and Structures, Vol. 40, 2003, pp. 401 445. 6 Itskov, M. and Nuri, A., A Class of Orthotropic and Transversely Isotropic Hyperelastic Constitutive Models based on a Polyconvex Strain Energy Function, International Journal of Solids and Structures, Vol. 41, 2004, pp. 337 403. 7 Kambouchev, N., Fernandez, J., and Radovitzky, R., Polyconvex Model for Materials with Cubic Anisotropy, In preparation. 8 Musgrave, M., Crystal Acoustics, Holden-Day, 1970. 9 Bertsimas, D. and Tsitsiklis, J., Introduction to Linear Optimization, Athena Scientific Press, 1997. 10 Zheng, Q. and Spencer, A., Tensors which Characterize Anisotropies, International Journal of Engineering Science, Vol. 31, 1993, pp. 679 693. 11 Weyl, J., The Classical Groups. Their Invariants and Representation, Princeton University Press, 1946. 12 Hirth, J. and Lothe, J., Theory of Dislocations, John Willey and Sons, 1982. 6 of 6