Article

Mathematics and Mechanics of Solids

1–22

A variational approach to nonlinear © The Author(s) 2017

Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
electro-magneto-elasticity: DOI: 10.1177/1081286517696536
journals.sagepub.com/home/mms
Convexity conditions and existence
theorems
Miroslav Šilhavý
Institute of Mathematics of the CAS, Czech Republic

Received 22 August 2016; accepted 6 February 2017

Abstract
Electro- or magneto-sensitive elastomers are smart materials whose mechanical properties change instantly by the
application of an electric or magnetic field. This paper analyses the convexity conditions (quasiconvexity, polyconvexity,
ellipticity) of the free energy of such materials. These conditions are treated within the framework of the general
A-quasiconvexity theory for the constraints

curl F = 0, div d = 0, div b = 0, (∗)

where F is the deformation gradient, d is the electric displacement and b is the magnetic induction. If the energy
depends separately only on F, or on d, or on b, the A-quasiconvexity reduces, respectively, to Morrey’s quasiconvexity,
polyconvexity and ellipticity conditions or to convexity in d or in b. In the present case, the simultaneous occurrence of
F, d and b leads to the cross-phenomena: mechanic–electric, mechanic–magnetic and electro–magnetic.
The main results of the paper are as follows.
• In dimension 3 there are 32 linearly independent scalar A-affine functions (and 15 in dimension 2) corresponding
to the constraints (∗).
• Therefore, an energy function ψ(F, d, b) is A-polyconvex if and only if it is of the form

ψ(F, d, b) = (F, cof F, det F, d, b, Fd, Fb)

where  is a convex function (of 31 scalar variables). Apart from the expected terms F, cof F, det F, d and b, we
have the cross-effect terms Fd, Fb (and in dimension 2 also d × b).
• An existence theorem is proved for a state of minimum energy for a system consisting of an A-polyconvex
electro-magneto-elastic solid plus the vacuum electromagnetic field outside the body.

Keywords
Electrmechanical and magnetomechanical interactions, finite strain, constitutive equations, energy methods, variational
principles, instabilities, A-quasiconvexity, A-polyconvexity

1. Introduction
Electro- or magneto-sensitive elastomers are smart materials whose mechanical properties change instantly
by the application of an electric or magnetic field. The sensitivity to the electromagnetic fields is due to the

Corresponding author:
Corresponding author: Miroslav Šilhavý, Institute of Mathematics of the CAS, Žitná 25, 115 67 Prague 1, Czech Republic.
Email: [email protected]
2 Mathematics and Mechanics of Solids

manufacturing process in which some metallic electro- or magneto-sensitive inclusions (such as alumina par-
ticles or iron powder) are deposited in an elastomeric (usually rubber) matrix. If the fabrication process is
conducted under external electric or magnetic fields, it produces an alignment of the inclusions and conse-
quently an anisotropy; the latter is combined with large deformations of the matrix. One is thus faced with full
nonlinear couplings of the mechanical response with the electric and magnetic fields and also with an indirect
magneto–electric coupling.
As is well-known, for large deformations the well-posedness questions play an important role.
For nonlinear elastostatics, Ball [1] showed that Morrey’s quasiconvexity condition [2, 3] has a direct rele-
vance for the behaviour of the body; moreover, he recognized the importance of Morrey’s sufficient condition
for quasiconvexity [3, Theorem 4.4.10], for which he introduced the term polyconvexity. He showed that the
polyconvexity is compatible with the realistic constraint for the energy function ψ, viz.,

ψ(F) → ∞ as det F → 0

and leads to satisfactory existence theorems in nonlinear elasticity under realistic assumptions. These convexity
conditions are known to be one of the main guiding principles for the formation of the nonlinear constitutive
equations, e.g. [4–6].
In this paper I extend the quasiconvexity and polyconvexity notions to the completely coupled problem in
electro-magneto-rheological elastomers. There are new issues beyond the purely mechanical case, as we shall
see below.
Our choice of the basic variables in the constitutive equations are the deformation gradient F, the
(lagrangean) electric displacement d and the (lagrangean) magnetic induction b. Among the new issues that
we encounter is that the electromagnetic variables satisfy

div d = 0, div b = 0 (1)

identically as a counterpart of
curl F = 0 (2)
for the deformation gradient F. We make a full use of (1) and (2) by adopting the convexity theory under
differential constraints known as the A-quasiconvexity theory [7–13]. Central to the theory are the notions of
A-quasiconvex function, A-quasiaffine function and A-polyconvex function, see Definitions 5.1 and 5.4.
The novel feature of the present paper is that it adopts these “A-notions” to the full combined electro–
magneto–elastic interactions, that is, to the combinations of the constrains (1) and (2) (Definitions 6.1 and 6.4).
For brevity, in this special case we omit the modifier “A-” and use the terms quasiconvex function, quasiaffine
function and polyconvex function. The main results of the paper are as follows.

• In dimension 3 there are 32 linearly independent scalar quasiaffine functions; in dimension 2 there are 15
linearly independent scalar quasiaffine functions (see the lists in (35) in Theorem 6.3, below). Here ‘linear
independence’ means linear independence in the linear space of (unrestricted) functions f = f (F, d, b)
defined on the space Mn×n × Rn × Rn , n = 2 or 3, under standardly defined addition and multiplication
by scalars.1 The nontrivial proof is given in Section 7.2
• In dimension 3, an energy function ψ = ψ(F, d, b) of an electro-magneto-elastic material is polyconvex
if and only if it is of the form

ψ(F, d, b) = (F, cof F, det F, d, b, Fd, Fb) (3)

where  is a convex function (of 31 scalar variables). In dimension 2, ψ is polyconvex if and only if it is
of the form
ψ(F, d, b) = (F, det F, d, b, Fd, Fb, d × b) (4)
where  is a convex function (of 14 scalar variables); see Theorem 6.5, below. Apart from the expected
terms F, cof F, det F, d, and b, we have the cross-effect terms Fd, Fb (and in dimension 2 also d × b). This
description applies to materials of arbitrary symmetry.3
• An existence theorem is proved for a state of minimum energy for a system consisting of a polyconvex
electro-magneto-elastic solid plus the vacuum electromagnetic field outside the body; see Theorem 8.3,
below.
Šilhavý 3

To mention restricted cases (as opposed to the full case described above), we start with noting that in the
absence of electromagnetic phenomena the A-quasiconvexity under the constraint (2) reduces to the afore-
mentioned Morrey’s quasiconvexity. The convexity conditions for the electric or magnetic phenomena in rigid
bodies (no deformation) have been studied in [7, 8, 16–18]. These works show that the A-quasiconvexity under
(1)1 or under (1)2 reduces to the ordinary convexity.
The quasiconvexity for combinations of mechanical and magnetic phenomena has been discussed in [19] and
[20], but ignoring the constraint (1)2 , which substantially reduces the class of quasiconvex and polyconvex ener-
gies. The paper [21] briefly mentions, as an example, a combination of mechanical and magnetic phenomena in
two dimensions within a different framework, but without any further development.

Note. After the research presented in this paper had been completed, the author became aware of the
recent papers by Gil and Ortigosa [22–24],4 which postulate (3) also. The main motivation for (3) in [22-24]
comes from the electro-magneto-elastic ellipticity condition, which is implied by condition (3). Accordingly,
the developments and motivations of [22–24] are different from the present work since here I derive the poly-
convexity (3) from the general concepts of A-quasiconvexity theory (Theorem 6.5) and prove a result specific
to polyconvexity: the existence theorem (Theorem 8.3). The electro-magneto-elastic ellipticity condition figures
as a consequence of the A-quasiconvexity (Proposition 6.2, below) and the nontrivial proof of the form of the
polyconvexity presented in Section 7 has no counterpart in [22–24].
The paper is organized as follows. Section 2 describes the notation and presents the basic definitions from
the ordinary convexity. Section 3 gives a survey of the equilibrium and constitutive equations for the static
electro-magneto-elasticity. Formal aspects of the variational principle of the electro-magneto-elasticity (the total
energy, its first and second variations and the variational derivation of the equilibrium equations) are treated in
Section 4. The optional Section 5 introduces the A-quasiconvexity in the general case. A specialization of
the A-quasiconvexity to electro-magneto-elasticity is provided in Section 6. This central section can be read
independently of Section 5 since independent definitions are given therein. Section 7 provides the proof of
Theorem 6.3. Section 8 establishes an existence theorem. The remaining sections are appendices. Section A1.1
summarizes some results on the classical rank 1 convexity needed in our proofs. Section A2.1 collects the results
on the weak convergence necessary for the existence theorem.

2. Preliminaries: notation, brief introduction to convexity

We use the direct notation with the same conventions as in [25, 26]. The following sets are used throughout:
R̄ = R ∪ {∞} = the extended real line,
Rn = the n-dimensional euclidean space,
Zn = the set of all n-tuples of integers,
Mn×n = the space of all real n × n matrices,
Mn×n
+ = {F ∈ Mn×n : det F > 0}.
We interpret the matrices from Mn×n as second-order tensors on Rn . We denote by 1 ∈ Mn×n the unit matrix,
by a · b the usual scalar product of two vectors in Rn and by A · B := tr(AT B) the scalar product of tensors. We
−T
+ is given by cof F = (det F)F
recall that the tensor of cofactors of F ∈ Mn×n .
If n = 3, we define the vector product and the curl in the usual way as vectors in R3 while if n = 2 then both
the vector product and the curl are the numbers a × b = a1 b2 − a2 b1 , curl a = a2,1 − a1,2 .
Although the main theme of the paper is various weakened notions of convexity, an essential use is made
of the classical convexity. We refer to [27] and [28] for systematic expositions of the convexity theory; here we
only outline basic notions. A function f : X → R̄ on a vector space X is said to be convex if
 
f (1 − t)ξ1 + tξ2 ≤ (1 − t)f (ξ1 ) + tf (ξ2 ) (5)
for every ξ1 , ξ2 ∈ X and every t ∈ (0, 1). The function f is said to be affine if we have the equality sign in (5)
holding identically. f is affine if and only if there is a linear functional ϕ on X and a constant c ∈ R such that
 
f (ξ ) = ϕ, ξ + c (6)
 
for every ξ ∈ X where ϕ, ξ is the value of ϕ on ξ . If X = Rm then (6) reads f (ξ ) = ϕ · ξ + c where ϕ ∈ Rm .
4 Mathematics and Mechanics of Solids

We conclude this section by recording Jensen’s inequality [29, Theorem 4.80], which underlies the notion of
polyconvexity. If  : Rm → R̄ is a convex lowersemicontinuous function and Q = (0, 1)n then
  
(z(x)) d Vx ≥  z(x) d Vx (7)
Q Q

for any measurable map z : Q → R . m

3. Equilibrium and constitutive equations for electro-magneto-elasticity

With the exception of Section 5, we work in the space dimensions n = 2 or n = 3. Recall from the introduction
that the variables in the constitutive equations are the deformation gradient F ∈ Mn×n
+ , the referential electric
displacement d ∈ Rn and the magnetic induction b ∈ Rn . Throughout the section, ψ : Mn×n + ×R ×R → R
n n

denotes the energy function of an electro-magneto-elastic body and (F, d, b) is an element of M+ × R × Rn .

n×n n

The coupling between electricity, magnetism and nonlinear elasticity is well studied since the sixties of the
last century, as illustrated by the book expositions [30–36] and others. Our situation is purely static, so that
only the static form of Maxwell’s equations and the mechanical equilibrium of forces govern the behavior of the
body.

3.1. Equilibrium equations

Actual (“eulerian”) configuration. The basic electromagnetic variables are the electric and magnetic fields, the elec-
tric displacement and the magnetic induction, denoted, respectively, by E, H, D, B. The mechanical variables are
the Cauchy stress tensor T, the density of the body force g and the actual density of mass ρ. The equilibrium
equations are
Div D = 0, Div B = 0, Curl E = 0, Curl H = 0 on Rn , (8)
Div T + ρg = 0 on ω (9)
where Curl and Div denote the curl and divergence with respect to the actual position and ω is the actual
configuration of the body. In Section 4, the equilibrium equations will be derived from a variational principle.
The equations (8) and (9) are assumed to hold in the weak sense, which then includes the well-known jump
conditions for the electromagnetic variables on the boundary of the body. This is not repeated here. Furthermore,
below we shall consider only the Dirichlet boundary conditions for the deformation; thus there is no equation
for the surface traction on the boundary. Outside ω we have the ether relations
E = D, H = B; (10)
on ω, we have the constitutive relations for E and H to be discussed below.

3.2. Referential (“lagrangian”) configuration

We denote by  ⊂ Rn the reference configuration of the body and by y :  → Rn the deformation. We prescribe
the Dirichlet boundary conditions on ∂, that is,
y = ỹ on ∂ (11)
where ỹ : ∂ → Rn is a given function. We assume that ỹ can be extended to an equally denoted injective
function on Rn \ cl  such that det ∇ ỹ > 0 on Rn \ . For notational convenience we define the deformation
gradient F : Rn → Mn×n+ by
∇y on ,
F= (12)
∇ ỹ on Rn \ cl .
We now use the classical Piola transformation, [37, 38, Chapter I, 7.18–7.20] to introduce the referential
(lagrangean) quantities by
e = F T E, h = F T H, d = (cof F)T D, b = (cof F)T B on Rn ,
(13)
S = T cof F on ,
Šilhavý 5

where, of course the spatial variables E, . . . , T are now expressed as functions of the referential variable. The
referential forms of the equilibrium equations read

div d = 0, div b = 0, curl e = 0, curl h = 0 on Rn , (14)

div S + g = 0 on , (15)

where curl and div denote the referential forms of the curl and divergence, that is, the same differential operators
as Curl and Div, but with the derivatives with respect to the actual position replaced by the derivatives with
respect to referential position. The ether relations (10) read in terms of the referential variables as

e = F T Fd/ det F, h = F T Fb/ det F (16)

outside .

3.3. Constitutive relations

The density of the free energy is
ψ = ψ(F, d, b)
which is a twice continuously differentiable function with the domain

+ ×R ×R
Dn+ := Mn×n n n
(17)

which is a subset of
Dn := Mn×n × Rn × Rn . (18)
We have the potential relations
S = DF ψ, e = Dd ψ, h = Db ψ. (19)
We assume that ψ satisfies the principle of material frame indifference

ψ(QF, d, b) = ψ(F, d, b) (20)

for all (F, d, b) ∈ Dn+ and all proper orthogonal tensors Q. A standard argument shows that (20) implies the
symmetry of the stress,
SF T = FS T , T T = T.
Example 3.1 (Non-interacting matter). Consider an elastic body  in the external electromagnetic field but
suppose that there is no field-matter interaction. Therefore, the energy splits into the sum

ψ(F, d, b) = ψ1 (F) + ψ2 (F, d, b)

of the elastic energy ψ1 (F) and of the energy of the vacuum electromagnetic field ψ2 (F, d, b). In the reference
configuration , ψ2 is given by
1  
ψ2 (F, d, b) = (det F)−1 |Fd|2 + |Fb|2 . (21)
2
Indeed, passing from the reference variable x to the spatial variable y = y(x) and employing the transformation
rules (13) we obtain the vacuum energy of the electromagentic field, that is,
 
1   1  2 
(det F)−1 |Fd|2 + |Fb|2 d Vx = |D| + |B|2 d Vy
2  2 ω
where dx and dy are the referential and actual elements of volume if n = 3 or those of area if n = 2, D and B are
the spatial electric displacement and magnetic induction and ω = y() is the actual configuration of the body.
The potential relations (19)2,3 yield the ether relations (16); the stress relation (19)1 yields

S = S1 + S2 where S1 (F) = DF ψ1 (F), S2 (F, d, b) = DF ψ2 (F, d, b)

6 Mathematics and Mechanics of Solids

where S1 is the elastic stress while a calculation shows that S2 is given by

 1  
S2 (F, d, b) = (det F)−1 Fd ⊗ d + Fb ⊗ b − F −T |Fd|2 + |Fb|2 .
2
Let us show that
div S2 = 0 (22)
for any deformation y of  and any vector fields d and b that satisfy (14)1,2 on . Indeed, passing to the spatial
stress T2 = S2 cof F −1 , we obtain the vacuum Maxwell tensor
1 2 
T2 = D ⊗ D + B ⊗ B − |D| + |D|2 1
2
whose spatial divergence is known to vanish as a consequence of (8)1,2 :
Div T2 = 0.
The referential form (22) then follows by Piola’s transformation. The equilibrium equation (15) with the total
stress S then reduces to the equilibrium for the elastic stress
div S1 + g = 0 on .
We thus summarize that the total stress S is different form zero even in the (idealized) absence of matter as a
consequence of the geometric factors in (21); however, its divergence identically vanishes.

4. Variational principle
This section presents a preliminary analysis of a variational principle for an electro-magneto-elastic body. We
consider a state of minimum energy of the system consisting of an elastic body  interacting with the electro-
magnetic field inside  and the vacuum electromagnetic field in its exterior. Section 8 treats the same minimum
principle under natural, weakened assumptions on y, d, b which ensure the existence of a minimizer. As already
mentioned, the proof is currently available only for the Dirichlet data for the deformation. Even though the
considerations to be presented in this section can be carried out for the general boundary conditions, we assume
the Dirichlet data for notational simplicity also here.

4.1. The system and its states

We assume that the reference configuration  is bounded and has class C 2 boundary ∂. We denote by c =
Rn \ ( ∪ ∂) the complement of the body and by ν the outer normal to ∂.
By a state we mean any triplet σ = (y, d, b) of maps
y :  → Rn , d : Rn → Rn , b : Rn → Rn ;
these represent the deformation of the body and the referential electric displacement and magnetic induction,
respectively. We assume that:
(i) y is twice continuously differentiable with y and its derivatives up to the order 2 having continuous
extensions to the closure cl  of , and with det ∇y > 0 on cl ;
(ii) y satisfies Dirichlet’s boundary condition (11);
(iii) d and b are continuously differentiable in  and in c with d and b and their derivatives having continuous
extensions from  to cl  and from c to cl c ;
(iv) d and b satisfy
div d = 0, div b = 0 on  ∪ c ,
[[d ]] · ν = 0, [[b ]] · ν = 0 on ∂,
where [[· ]] is the jump across ∂.
We denote by S the set of all states.
Šilhavý 7

4.2. The total energy

The total energy of a state σ = (y, d, b) ∈ S is defined by
  
1  
E(σ ) = ψ(∇y, d, b) d Vx − g · y d Vx + J −1 |Fd|2 + |Fb|2 d Vx (23)
  2 c

where the deformation gradient outside  is defined by (12) using the extension ỹ on c which is fixed, and
J = det F. Following [39, Chapter 8], we note that the last term in (23) is independent of the choice of the
fictitious ‘deformation’ ỹ since it can be transformed into the vacuum energy

1  2 
|D| + |B|2 d Vy
2 ωc

as in Example 3.1, where ωc = Rn \ y() is the exterior of the actual configuration y().
The set δ S of admissible variations of state is the set of triplets (κ, δ, β) of infinitely differentiable functions
on Rn with values in Rn such that

κ = 0 on c , div δ = 0, div β = 0 in Rn (24)

and δ, β vanish outside some (varying) bounded subset of Rn . If σ = (y, d, b) is a state, we define the first and
second variations δ E(σ )[·] and δ 2 E(σ )[·] of energy at σ as linear and quadratic functionals on δ S by
  ⎫
δ E(σ )[κ, δ, β] = D ψ(∇y, d, b)[∇κ, δ, β] d Vx − g · κ d Vx ⎪
⎪
⎪
  ⎪
⎪
 ⎪
⎪
1   ⎪
⎪
+ −1
J (Fd · Fδ) + (Fb · Fβ) d Vx , ⎪
⎪
2 c ⎬
 ⎪
(25)
⎪
δ 2 E(σ )[κ, δ, β] = D2 ψ(∇y, d, b)[(∇κ, δ, β), (∇κ, δ, β)] d Vx ⎪ ⎪
⎪
 ⎪
⎪
 ⎪
⎪
  ⎪
⎪
+ −1
J |Fδ| + |Fβ| d Vx .
2 2 ⎭
c

4.3. Equilibrium states

A state σ = (y, d, b) ∈ S is said to be an equilibrium state if E(σ ) < ∞ and

E(σ ) ≤ E(σ̄ )

for all σ̄ ∈ S. Necessary conditions for the minimum are, standardly,

δ E(σ )[κ, δ, β] = 0, δ 2 E(σ )[κ, δ, β] ≥ 0 (26)

for each (κ, δ, β) ∈ δ S. Moreover, (26)1 is equivalent to the equilibrium conditions

⎫
div S + g = 0 in , ⎪
⎬
curl e = 0, curl h = 0 in  ∪  ,
c
(27)
⎪
⎭
[[e ]] × ν = 0, [[h ]] × ν = 0 on ∂,

where the associated stress S on  and the electric and magnetic fields e and h on the entire space are given by
⎧
⎨ S = DF ψ(F, d, b)
⎪ on 
e = Dd ψ(F, d, b), h = Db ψ(F, d, b) on 
⎪
⎩
e = J −1 F T Fd, h = J −1 F T Fb on c .
8 Mathematics and Mechanics of Solids

To derive (27), note that if (κ, δ, β) ∈ δ S, then

 
δ E(σ )[κ, δ, β] = (S · ∇κ − g · κ) d Vx + (e · δ + h · β) d Vx = 0 (28)
 Rn

by (26)1 . By (24)2,3 we may write δ = curl π , β = curl ρ; inserting this in to (28) and integrating by parts we
obtain  
(− div S − g) · κ d Vx + (π · curl e + ρ · curl h) d Vx = 0.
 Rn

The arbitrariness of κ, π , and ρ then gives (27). 

5. A-quasiconvexity: the general case

Our treatment of the convexity properties for the electro-magneto-elasticity is based on the A-quasiconvexity
theory, which includes the associated notions A-quasiaffinity, A-polyconvexity, -convexity and -ellipticity.
The A-quasiconvexity theory has been introduced in [7] and further developed in [8]. Closely related is the
compensated compactness theory [9, 10]. The reader is referred to [11–13] for more recent developments and
additional literature. This section discusses these notions from a general point of view; the specialization to
electro-magneto-elastic materials is the subject of the succeeding sections.

5.1. The differential operator A and the characteristic cone 

The following dimensions will be needed in the subsequent discussion:

n = the number of independent variables, x = (x1 , . . . , xn ),

d = the number of dependent variables, u = (u1 , . . . , ud ),
l = the number of differential constrains.
∞
Let Q = (0, 1)n be the unit cube, let Cper (Rn , Rd ) denote the set of all infinitely differentiable Q-periodic maps
u : Rn → Rd . We shall consider the first-order differential constraint Av = 0 on a map v ∈ C ∞ (Rn , Rd ) where


n
Av = A(i) v,i
i=1

with A(i) ∈ Lin(Rd , Rl ). For each η = (η1 , . . . , ηn ) ∈ Rn define


n
A(η) = ηi A(i) ,
i=1

which is an element of Lin(Rd , Rl ), and make the standing assumption that the rank of A(η) is the same for all
η = 0. We define the characteristic cone

 = {u ∈ Rd : A(η)u = 0 for some η ∈ Rn , η = 0}.

Definition 5.1. A continuous function f : Rd → R̄ is said to be

(i) A-quasiconvex if 
f (u + v(x)) d Vx ≥ f (u) (29)
Q
∞

for all u ∈ Rd and all v ∈ Cper (Rn , Rd ) such that Av = 0 on Rn and Q v d Vx = 0;
(ii) A-quasiaffine if it takes only finite values and both f and −f are A-quasiconvex;
Šilhavý 9

(iii) -convex if
f (tu1 + (1 − t)u2 ) ≤ tf (u1 ) + (1 − t)f (u2 )
for every t ∈ (0, 1) and u1 , u2 ∈ Rd such that u2 − u1 ∈ ;
(iv) -affine if it takes only finite values and both f and −f are -convex.
If f is continuously differentiable then the -convexity is equivalent to the -ellipticity
D2 f (u)(l, l) ≥ 0
for every u ∈ Rn and l ∈ .
Theorem 5.2. [8, Proposition 3.4] If f : Rd → R̄ is a continuous A-quasiconvex function then f is -convex;
consequently, if f is A-quasiaffine then f is -affine.

The following weak sequential lower semicontinuity theorem is the main motivation for the
A-quasiconvexity. We refer to Section A2.1 (below) for our conventions about the weak convergence. The
weak sequential lower semicontinuity is the basic ingredient of the direct method of the calculus of variations.
It should be also noted that for the proof of the existence of the minimizer in electro-magneto-elasticity in
Theorem 8.3 (below) the sequential lower semicontinuity theorem cannot be used as the hypothesis (30) is
inconsistent with the requirement ψ(F, d, b) → ∞ for det F → 0. One has to use the A-polyconvexity defined
below.
Theorem 5.3. [8, Theorem 3.7] Let 1 ≤ p < ∞ and suppose that f :  × Rd → [0, ∞) is a Carathéodory
integrand such that f (x, ·) is A-quasiconvex and
0 ≤ f (x, u) ≤ a(x)(1 + |u|p ) (30)
for all x ∈  and u ∈ Dd where a :  → [0, ∞) is a bounded function. If u and uk belong to Lp (, Rd ) and
satisfy
uk  u in Lp (, Rd ) and Auk → 0 in W −1,p (, Rd )
then  
lim inf f (x, uk (x)) d Vx ≥ f (x, u(x)) d Vx .
k→∞  

Definition 5.4. A continuous function f : Rd → R̄ is said to be A-polyconvex if there exists a finite number of
A-quasiaffine functions f1 , . . . , fm and a convex lowersemicontinuous function  : Rm → R̄ such that
f (u) = ( f1 (u), . . . , fm (u)) (31)
for each u ∈ Rd .
Theorem 5.5. [7, Corollary 2.5] Any A-polyconvex function is A-quasiconvex.

Proof Since the polyconvexity is central for the present paper, let us outline the proof. Thus let f = f (u) be as
in (31) and prove (29) for all u, v as in Definition 5.1(i). The quasiaffinity of f1 , . . . , fm means that

fi (u + v(x)) d Vx = fi (u), 1 ≤ i ≤ m,
Q

and hence the application of Jensen’s inequality (7) with z given by

    
z(x) = f1 u + v(x) , . . . , fm u + v(x)
gives  
    
f (u + v(x)) d Vx =  f1 u + v(x) , . . . , fm u + v(x) d Vx
Q Q  
     
≥ f1 u + v(x) d Vx , . . . , fm u + v(x) d Vx
Q Q

= ( f1 (u), . . . , fm (u)) = f (u). 

10 Mathematics and Mechanics of Solids

6. A specialization to electro-magneto-elasticity
We apply the formalism of the preceding section with n = 2 or n = 3, and with the identifications
v = (F, d, b)
where F is the deformation gradient, d the electric displacement and b the magnetic induction. In view of the
constraint det F > 0, we apply the A-quasiconvexity notions to functions f defined on the domain Dn+ , see (17).
To obtain an agreement with the general theory of the preceding section, where only functions f defined on the
entire Rd have been considered, we tacitly extend f : Dn+ → R̄ to the entire Dn from (18) by setting f equal to
∞ on Dn \ Dn+ .
The functions v of Section 5 will be identified with the triples (ϕ, δ, β) : Rn → Dn and the operator A with
A(ϕ, δ, β) = (curl ϕ, div δ, div β). (32)
Here curl of ϕ = [ϕij ]ni,j=1 is defined by


3 
2
(curl ϕ)il = ljk ϕij,k , (curl ϕ)i = jk ϕij,k ,
j,k=1 j,k=1

in dimensions n = 3 and n = 2, respectively, where i, l = 1, 2, 3 or i = 1, 2 and ijk and ij are the three- and
two-dimensional permutation symbols. Analogously, if η ∈ Rn , we define ϕ × η by

3 
2
(ϕ × η)il = ljk ϕij ηk , (ϕ × η)i = jk ϕij ηk
j,k=1 j,k=1

in dimensions n = 3 and n = 2, respectively.

To determine the characteristic cone  ≡ E corresponding to the system (32), we replace the partial
derivatives ∇ϕ, ∇δ, ∇β in (32) by the tensor products ϕ ⊗ η, δ ⊗ η, β ⊗ η where η ∈ Rn is an arbitrary nonzero
vector. This transforms (32) into
ϕ × η = 0, δ · η = 0, β · η = 0; (33)
noting that (33)1 is satisfied in and only if ϕ = ξ ⊗ η for some ξ ∈ Rn , one obtains
E = {(ξ ⊗ η, δ, β) ∈ Dn : ξ , δ, β, η ∈ Rn , δ · η = β · η = 0, η = 0}.
We now specialize the general definitions of Section 5 to the present case.
Definition 6.1. A continuous function f : Dn → R̄ is said to be
(i) quasiconvex at (F, d, b) ∈ Dn if

f (F + ϕ(x), d + δ(x), b + β(x)) d Vx ≥ f (F, d, b)
Q
∞
for each triplet (ϕ, δ, β) ∈ Cper (Rn , Dn ) satisfying

curl ϕ = 0, div δ = div β = 0 on Rn and (ϕ, δ, β) d Vx = 0;
Q

(ii) quasiconvex if it is quasiconvex at every point of Dn ;

(iii) quasiaffine if f takes only finite values, and both f and −f are quasiconvex;
(iv) E -convex if
f (F + tξ ⊗ η, d + tδ, b + tβ) ≤ (1 − t)f (F, d, b) + tf (F + ξ ⊗ η, d + δ, b + β)
for every t ∈ (0, 1) and (F, d, b) ∈ Dn and every ξ , δ, β, η ∈ Rn such that
δ · η = β · η = 0 and η = 0; (34)
Šilhavý 11

(v) E -affine if f takes only finite values, and both f and −f are E -convex.
Proposition 6.2. Let f : Dn+ → R be twice continuously differentiable.
(i) If f is quasiconvex at (F, d, b) ∈ Dn+ then f is elliptic at (F, d, b), that is,
D2 ψ(F, d, b)[(ξ ⊗ η, δ, β), (ξ ⊗ η, δ, β)] ≥ 0
for every ξ , δ, β, η ∈ Rn satisfying (34).
(ii) If f is quadratic, that is, if
f (F, d, b) = C[(F, d, b), (F, d, b)]
for every (F, d, b) ∈ D and some symmetric bilinear form C then f is quasiconvex at some point ⇔ f is
n

quasiconvex ⇔ f is elliptic at some point ⇔ f is elliptic at every point of Dn .

Here the second derivative D2 ψ(F, d, b)[·, ·] of ψ at (F, d, b) is interpreted as a bilinear form in the incremental
variable (ξ ⊗ η, δ, β). The proof of Proposition 6.2 is only a minor variation of van Hove’s original proof [40]
in the gradient case; it is therefore omitted.
Proposition 6.2 can be restated equivalently in terms of the second variation δ 2 E(σ ) of the total energy (see
(23) and (25)). Namely, the ellipticity of ψ at (F, d, b) is equivalent to the nonnegativity of the second variation
δ 2 E(σ ) at the homogeneous state with data (F, d, b) under the Dirichlet boundary conditions, that is,
δ 2 E(σ )[κ, δ, β] ≥ 0
for every triplet of infinitely differentiable functions κ, δ, β : Rn → Rn which vanish outside .
The main results of this paper are the following theorem and Theorem 6.5, below.
Theorem 6.3. A continuous function f : Dn → R is quasiaffine ⇔ f is E -affine ⇔ f is a linear combination,
with constant coefficients, of the following functions:

1, F, det F, d, b, Fd, Fb, d × b if n = 2,
(35)
1, F, cof F, det F, d, b, Fd, Fb if n = 3.

Here the expressions involving the variables F, d, b in (35) stand for the functions of (F, d, b) defined on the
domain Dn ; linear independence is understood in the linear space of functions f = f (F, d, b) on Dn under
standardly defined addition and multiplication by scalars.
Thus there are 15 linearly independent quasiaffine functions if n = 2 and 32 linearly independent quasiaffine
functions if n = 3, including constants. The proof of Theorem 6.3 is deferred to Section 7.
Definition 6.4. A continuous function f : Dn → R̄ is said to be polyconvex if there exists a finite number of
quasiaffine functions f1 , . . . , fm and a convex lowersemicontinuous function  : Rm → R̄ such that
f (F, d, b) = ( f1 (F, d, b), . . . , fm (F, d, b))
for each (F, d, b) ∈ Dn .
Theorem 6.3 has the following corollary.
Theorem 6.5. A continuous function f : Dn → R̄ is polyconvex if and only if f is of the following form:
f (F, d, b) = (∧(F, d, b))
for every (F, d, b) ∈ Dn+ , where we abbreviate

(F, det F, d, b, Fd, Fb, d × b) if n = 2,
∧(F, d, b) =
(F, cof F, det F, d, b, Fd, Fb) if n = 3
and where  is a convex lowersemicontinuous function on
⎧
⎨ M2×2 × R × (R2 )4 × R if n = 2,
D̃ =
n
⎩ M3×3 × M3×3 × R × (R3 )4 if n = 3.

Thus  is a function of 14 and 31 scalar variables, respectively.

12 Mathematics and Mechanics of Solids

7. Proof of Theorem 6.3

Recall from Definition 6.1 that a continuous function f : Dn → R is quasiaffine if

f (F + ϕ, d + δ, b + β) d Vx = f (F, d, b) (36)
Q

∞
for each triplet (ϕ, δ, β) ∈ Cper (Rn , Dn ) satisfying

curl ϕ = 0, div δ = div β = 0 on R and n
(ϕ, δ, β) d Vx = 0, (37)
Q

and that f is E -affine if

f (F + tξ ⊗ η, d + tδ, b + tβ) = (1 − t)f (F, d, b) + tf (F + ξ ⊗ η, d + δ, b + β) (38)

for every t ∈ (0, 1) and (F, d, b) ∈ Dn and every ξ , δ, β, η ∈ Rn such that

δ · η = β · η = 0 and η = 0. (39)

The proof of Theorem 6.3 is divided into several lemmas. We start with the analysis of the separate E -
affinity with respect to the variables F, d and b. The cross effects will be analysed subsequently. We refer to
Section A1.1 for the rank 1 affinity which underlies item (i) of the following result and many points in the
subsequent treatment.
Lemma 7.1. Let f : Dn → R be a E -affine function. Then

(i) for each d, b ∈ Rn the function f (·, d, b) is rank 1 affine, that is, it is a linear combination, with coefficients
depending on d, b, of the functions occurring in (58);
(ii) for each F ∈ Mn×n the function f (F, ·, ·) is a linear combination, with coefficients depending on F, of the
functions
1, d, b, d × b if n = 2,
1, d, b if n = 3.

Proof (i) Fixing d, b ∈ Rn , taking δ = β = 0 in (38) and denoting g(·) = f (·, d, b) we obtain inequality (57)
with the equality sign for every t ∈ (0, 1), every F ∈ Mn×n and every ξ , η ∈ Rn . Thus f (·, d, b) is rank 1 affine
and Lemma A1.2 yields the assertion.
(ii) Employing (38) with ξ = β = 0 and noting that there always exists an η ∈ Rn , η = 0, such that
δ · η = β · η = 0, we obtain

f (F, d + tδ, b) = (1 − t)f (F, d, b) + tf (F, d + δ, b)

for every F, d, b and δ. Thus f (F, ·, b) is affine and hence

f (F, d, b) = (F, b) + (F, b) · d,

for each (F, d, b) ∈ Dn , where (F, b) ∈ R and (F, b) ∈ Rn . Repeating the same argument for δ = 0, β
arbitrary, we obtain
f (F, d, b) = (F, d) + ζ (F, d) · b,
(F, d, b) ∈ Dn , where (F, d) ∈ R and ζ (F, d) ∈ Rn . Thus

(F, d) + ζ (F, d) · b = (F, b) + (F, b) · d.

Since the left-hand side is affine in b at any fixed d, we see that the functions  and  must be affine functions
of b as well, that is,
(F, b) = c2 (F) · b + c4 (F), (F, b) = c1 (F) + A(F)b,
Šilhavý 13

b ∈ Rn , where c2 (F), c1 (F) ∈ Rn , c4 (F) ∈ R and A(F) ∈ Mn×n . Hence

f (F, d, b) = c1 (F) · d + c2 (F) · b + A(F)b · d + c4 (F). (40)

To complete the proof, we return to (38), this time with ξ = 0, so that we have

f (F, d + tδ, b + tβ) = (1 − t)f (F, d, b) + tf (F, d + δ, b + β) (41)

for every t ∈ (0, 1), every (F, d, b) ∈ Dn and every δ, β, η ∈ Rn such that (39) holds. This gives

A(F)(b + tβ) · (d + tδ) = (1 − t)A(F)b · d + tA(F)(b + β) · (d + δ).

The left-hand side contains a quadratic term (i.e. the coefficient of t2 ) which is equal to A(F)β · δ and hence we
have to have
A(F)β · δ = 0 (42)
for every δ, β such that δ · η = β · η = 0 for some η = 0.
If n = 3, then for a given pair (δ, β) there always exists a η = 0 such that δ · η = β · η = 0. Hence (41)
asserts that f (F, ·, ·) is affine. Thus the bilinear term A(F)b · d in (40) must vanish and hence f (F, ·, ·) is of the
form asserted in (ii).
If n = 2 then for a given pair (δ, β) there exists a η = 0 such that δ · η = β · η = 0 if and only if δ and β are
parallel, that is, δ × β = 0. Thus (42) requires

A(F)b · d = c3 (F)(d × b)

for all d, b ∈ R2 and some c3 (F) ∈ R. Then (40) gives the asserted form. 
We are about to pass to the cross effects. In view of the results of Lemma 7.1 it suffices to consider functions
of very special forms considered in Lemmas 7.2–7.5, as explained in the proof of Lemma 7.6.
Lemma 7.2. Let f : Dn → R be given by

f (F, d, b) = (F) · d

(F, d, b) ∈ Dn where  is a linear transformation from Mn×n into Rn , written F → (F). Then f is E -affine if
and only if f is of the form
f (F, d, b) = Fd · c
for all (F, d, b) ∈ Dn and some c ∈ Rn .

Proof The choice F = 0, d = b = 0 in (38) yields

t2 (ξ ⊗ η) · δ = t(ξ ⊗ η) · δ (43)

for every t ∈ (0, 1) and every ξ , δ, η ∈ Rn such that

δ · η = 0, η = 0. (44)

Thus (ξ ⊗ η) · δ = 0 for every ξ , δ, η ∈ Rn such that (44) holds. Consequently,

(ξ ⊗ η) = m(ξ )η

ξ , η ∈ Rn where m(ξ ) ∈ R. The linearity in ξ requires m(ξ ) = c · ξ for some c ∈ Rn and all ξ ∈ Rn ; thus
(ξ ⊗ η) = (c · ξ )η = (ξ ⊗ η)T c for all ξ , η ∈ Rn . Since every A ∈ Mn×n is a sum of tensor products ξ ⊗ η, the
linearity of (·) yields (A) = AT c for all A ∈ Mn×n . Hence

f (F, d, b) = (A) · d = F T c · d ≡ Fd · c.

This completes the proof of the direct implication; the proof of the converse implication is straightforward and
the details are omitted. 
14 Mathematics and Mechanics of Solids

Lemma 7.3. Let n = 3 and let f : D3 → R be given by

f (F, d, b) = (cof F) · d

(F, d, b) ∈ Dn where  is a linear transformation from M3×3 into R3 , written A → (A). Then f is E -affine if
and only if f = 0 identically.

Proof We apply (43) with F = 1, d = 0 and ξ , δ, η ∈ R3 as in (44). Using the formula

 
cof(1 + ξ ⊗ η) = (1 + ξ · η)1 − η ⊗ ξ

one finds that (38) is equivalent to

   
t2  (ξ · η)1 − tη ⊗ ξ · δ = t (ξ · η)1 − η ⊗ ξ · δ

This requires  
 (ξ · η)1 − η ⊗ ξ · δ = 0
for every ξ , δ, η ∈ R3 as above. Hence

((η · ξ )1 − η ⊗ ξ ) = m(ξ )η

ξ , η ∈ R3 where m(ξ ) ∈ R. The linearity in ξ provides m(ξ ) = −c · ξ for some c ∈ R3 and all ξ ; hence

((η · ξ )1 − η ⊗ ξ ) = −(c · ξ )η = −(η ⊗ ξ )c.

Setting a := (1), we obtain

(η ⊗ ξ ) = (η · ξ )a + (η ⊗ ξ )c.
Since every A ∈ M3×3 is a sum of tensor products η ⊗ ξ , the linearity of (·) yields

(A) = (tr A)a + Ac

for each A ∈ M3×3 . The consistency requires a = (1) = 3a + c and hence

1
(A) = − (tr A)c + Ac.
2
To complete the proof, let us show that c = 0. Let η ∈ R3 be any unit vector, and apply (38) with F = 1 − η ⊗ η,
d = 0, δ ∈ R3 satisfying δ · η = 0. Using

cof(F + tη ⊗ η) = tF + η ⊗ η

one finds that equation (38) reads

 
t tFc − (1 + t/2)c · δ = −tc · δ/2;

the arbitrariness of t then leads to the unique consequence

c·δ =0

for any δ such that δ · η = 0 for some unit vector η. Taking η such that η · c = 0, we can take δ = c to obtain
c = 0. 
Lemma 7.4. Let f : Dn → R be given by

f (F, d, b) = (det F)c · d

for every (F, d, b) ∈ Dn where c ∈ Rn is a constant. If f is E -affine then f = 0 identically.

Šilhavý 15

Proof We apply (38) with F = 1, d = 0 and ξ , δ, η ∈ Rn as in (44). This gives

t(1 + tξ · η)(c · d) = t(c · d) + (1 − t)(1 + ξ · η)(c · d)

and clearly this can hold only if c = 0. 

Lemma 7.5. Let n = 2 and let f : D2 → R be given by

f (F, d, b) = m(F)(d × b)

for every (F, d, b) ∈ Dn where m : M2×2 → R is a rank 1 affine function. Then f is E -affine if and only if

f (F, d, b) = c(d × b) (45)

for all (F, d, b) ∈ D2 and some c ∈ R.

Proof By Lemma A1.2 we have m(F) = A · F + b det F + c for each F ∈ M2×2 where A ∈ M2×2 and b, c ∈ R
are constants. Hence
f (F, d, b) = (A · F + b det F + c)(d × b).
Let η ∈ R2 be any unit vector, and let β = η⊥ and λ > 0. Let us write the equality (38) with F = λ1, d = η,
b = 0, ξ ∈ R2 arbitrary, δ = 0. This gives
   
t A · (λ1 + tξ ⊗ η) + bλ2 1 + tλ−1 (ξ · η) + c (η × η⊥ )
= (1 − t)f (F, d, b) + tf (F + ξ ⊗ η, d, b + β).

The quadratic term (i.e. the coefficient of t2 ) on the left-hand side is

A · (ξ ⊗ η) + bλ(ξ · η).

This term must vanish. The arbitrariness of λ, ξ , η then gives A = 0, b = 0. Thus we have (45). Conversely, if
f is given by (45), then clearly, f is E -affine. 
Lemma 7.6. A continuous function f : Dn → R is E -affine if and only if f is a linear combination, with
constant coefficients, of the functions occurring in (35).

Proof Let f be E -affine. By Lemma 7.1(ii) then

f (F, d, b) = c0 (F) + c1 (F) · d + c2 (F) · b + c3 (F)(d × b)

for each (F, d, b) ∈ Dn where the last term must be omitted if n = 3. By item (i) of the same lemma then
f (·, d, b) is a rank 1 affine function for each d, b ∈ Rn . The independence of d, b and d × b then implies that
each of the coefficients c0 to c3 are rank 1 affine functions. Lemma A1.2 then asserts that c0 and c3 are exactly of
the form described in (58). Since c1 and c2 are vector valued functions, a componentwise application of Lemma
A1.2 gives
c1 (F) = c + (F) + (cof F) + d det F
for every F ∈ Mn×n , where c ∈ Rn and ,  are linear transformations from Mn×n to Rn with  = 0 if n = 2.
A similar form applies to c2 . Using the just described forms of c0 (F) to c3 (F) and collecting some of the terms
of the same type into one, it is found that
f (F, d, b) = m0 (F, d, b) + M1 · cof F
+ 1 (F) · d + 2 (F) · b + 1 (cof F) · d + 2 (cof F) · b (46)
+ (m2 · d + m3 · b) det F + m4 (F)(d × b)

for every (F, d, b) ∈ Dn where m0 is an affine function of F, d, b, the tensor M1 is in Mn×n with M1 = 0 if
n = 2, the objects i , i (i = 1, 2) are linear transformations from Mn×n into Rn with i = 0 if n = 2, the
vectors mi ∈ Rn (i = 1, 2) are constants and m4 is a rank 1 affine function.
16 Mathematics and Mechanics of Solids

We shall now make use of the full power of the E -affinity equality (38) (so far only various particular cases
have been used). Inserting the form of f from (46) into (38) and noting that the affine function m0 and the term
M1 · cof F trivially satisfy that equality, we see that we have to require that the function f1 given by

f1 (F, d, b) = 1 (F) · d + 2 (F) · b + 1 (cof F) · d + 2 (cof F) · b

+ (m2 · d + m3 · b) det F + m4 (F)(d × b)

has to satisfy (38). Then of course the terms of different order in F have to satisfy the equality individually as
well as the terms with d and b. Thus each of the functions

1 (F) · d, 2 (F) · b,

1 (cof F) · d, 2 (cof F) · b, mi · d det F, m4 (F)(d × b)

must be E -affine. By Lemma 7.2 then 1 (F) · d = Fd · c1 , 2 (F) · b = Fb · c2 where ci ∈ Rn are constants, by
Lemma 7.3 then i = 0, i = 1, 2, by Lemma 7.4 mi = 0 and by Lemma 7.5 m4 is constant. The asserted form
of f follows. The converse implication is immediate. 
We conclude this section with the following converse statement.
Lemma 7.7. Each function from the list (35) is quasiaffine.

Proof We have to prove that any function f from the list (35) satisfies the equality (36) for each triplet (ϕ, δ, β) ∈
∞
Cper (Rn , Dn ) satisfying (37).
Consider first the case n = 3. The system (37) implies that there are functions ω, π, ρ ∈ C ∞ (R3 , R3 ), such
that ϕ = ∇ω, δ = curl π , β = curl ρ; equation (36) then reads

f (F + ∇ω, d + curl π , b + curl ρ) d Vx = f (F, d, b). (47)
Q

An argument described in [41, Remark, p. 141] shows that it suffices to verify (47) only for ω, π , ρ ∈
C0∞ (Q, R3 ). The verification of (47) for the functions (F, d, b) → F, cof F, det F is standard, see, for example,
[1]. Consider now the functions (F, d, b) → d, b. Then (47) reads

(d + curl π ) d Vx = d
Q

and a similar equation for b, which is true since

 
curl π d Vx = π × ν dAx = 0
Q ∂Q

by Gauss’ theorem (where ν is the normal to ∂Q) and π vanishes on ∂Q. Finally consider the functions
(F, d, b) → Fd, Fb. We have

    
Q (F + ∇ω)(d + δ) d Vx = div (Fx + ω) ⊗ (d + δ) d Vx
Q
 
= (Fx + ω) (d + δ) · n dAx
∂Q
= Fx (d · n) dAx (since δ = 0 on ∂Q)
 ∂Q
 
= div (Fx) ⊗ d d Vx = Fd.
Q

This completes the proof for n = 3. The case n = 2 is similar; the details are omitted. 
Šilhavý 17

8. Existence theorem
The present section deals with the existence of minimum energy states for the energy E of a polyconvex solid
and the surrounding vacuum electromagnetic field. As is usual, the state space S from Section 4 has to be
enlarged as described in Definition 8.1 (below). Let us recall the definitions of Dn+ and Dn in (17) and (18).
The existence theory for a purely elastic material with a polyconvex energy is well understood [1, 42–44].
The corresponding part of the proof is based on the sequential weak continuity of the cofactor and determinant.
The additional electromagnetic variables d and b interact with the mechanical variable in the nonlinear terms
Fd and Fb and in dimension 2 we have also electrical–magnetic interactions d × b. These terms are sequentially
weakly continuous as well, but this time one has to use the div–curl lemma. We summarize the results on the
weak convergence and weak continuity in Section A2.1, below.
Definition 8.1. Let  ⊂ Rn be a bounded open set with Lipschitz boundary. We denote by S the set of all
triplets (y, d, b) ∈ W 1,1 (, Rn ) × L1 (, Rn ) × L1 (, Rn ) such that

div d = 0, div b = 0 on Rn ,
(48)
y = ỹ on ∂

in the sense of distributions and in the sense of traces, respectively. Here ỹ : ∂ → Rn is a prescribed function.
As in Section 3.2, we assume that ỹ can be extended to an equally denoted injective function on ∂ ∪ c ; in the
present section we assume that
c · 1 ≥ J −1 F T F ≥ c−1 · 1 (49)
on c where F is as in (12) and c is a positive constant.
Definition 8.2. The total energy of a state σ = (y, d, b) ∈ S is defined by the original formula (23) where
ψ : Dn+ → R is the energy, which is assumed to be continuous and bounded from below and where we assume
that the body force g is in L∞ (, Rn ) for notational simplicity.
Theorem 8.3. Let (49) hold and let p, r and s be numbers satisfying

2 ≤ p < ∞, 1/r + 1/p ≤ 1, 1/s + 1/p ≤ 1,

and additionally
let 1/r + 1/s ≤ 1 if n = 2,
let q be a number satisfying 3/2 ≤ q < ∞ if n = 3.
Extend the energy function ψ : Dn+ → R to ψ̃ : Dn → R by setting ψ̃(F, d, b) = ∞ if det F ≤ 0 and assume
that the following conditions hold:
(i) there exists a continuous convex and bounded from below function  : D̃n → R ∪ {∞} such that

ψ̃(F, d, b) = (∧(F, d, b)) (50)

for every (F, d, b) ∈ Dn ;

(ii) we have
c(|F|p + |d|r + |b|s ) + d if n = 2,
ψ(F, d, b) ≥
c(|F| + | cof F| + |d| + |b| ) + d if n = 3
p q r s

for some c > 0, d ∈ R and all (F, d, b) ∈ Dn+ .

If S contains an element of finite total energy then there exists a σ = (y, d, b) ∈ S such that

E(σ ) ≤ E(σ̄ )

for all σ̄ ∈ S; each such a σ satisfies

det ∇y > 0 for almost every point of .

18 Mathematics and Mechanics of Solids

Note that the continuity of , the definition of ψ̃ and (50) imply that ψ(F, d, b) → ∞ if det F → 0.
Proof Let n = 3 and assume that q > 3/2 rather than q ≥ 3/2 to simplify the matters; the case q = 3/2 is
similar but slightly more complicated (cf. [45, Proof of Theorem 5. 1, Case 2] for purely elastic bodies).
In the proof, we are going to apply Propositions A2.1 and A2.2, below. These propositions involve hypothe-
ses on the exponents; we leave to the reader to verify that the hypotheses of the present theorem on p, q, r and s
are chosen exactly to satisfy the hypotheses of Propositions A2.1 and A2.2.
Let σk = (yk , dk , bk ) ∈ S be a minimizing sequence and write Fk = ∇yk for brevity. The coercivity condition
(ii) implies that the sequence yk is bounded in W 1,p (, R3 ), the sequence cof Fk is bounded in Lq (, M3×3 ), the
restrictions of dk and bk to  are bounded in Lr (, R3 ) and Ls (, R3 ), respectively, and the restrictions of dk
and bk to c are bounded in L2 (c , R3 ). The reflexivity of these spaces implies that it is possible to extract a
subsequence of the sequence σk = (yk , dk , bk ), again denoted by σk , such that

yk  y in W 1,p (, R3 ),

Lr (, R3 ),
dk  d in
L2 (c , R3 ),

Ls (, R3 ),
bk  b in
L2 (c , R3 )

for some (y, d, b) the indicated spaces. Proposition A2.1 then implies that

cof Fk  cof F in Lq (, M3×3 ), (51)

det Fk  det F in L2q/3 (). (52)

Furthermore, the components of the vector Fd with a general F = ∇y and a general d are Fi · d where Fi =
(Fi1 , Fi2 , Fi3 ) and since
curl Fi = 0, div d = 0,
the div–curl lemma Proposition A2.2 implies

Fk dk  Fd in L1 (, R3 ) and similarly Fk bk  Fb in L1 (, R3 ).

To summarize, we have
∧(Fk , dk , bk )  ∧(F, d, b) in L1 () (53)
and hence Proposition A2.3 gives
 
lim inf (∧(Fk , dk , bk )) dVx ≥ (∧(F, d, b)) dVx .
k→∞  

This can be rewritten as  

lim inf ψ̃(Fk , dk , bk ) dVx ≥ ψ̃(F, d, b) dVx . (54)
k→∞  

The integral on the right-hand side is finite and the fact that ψ̃ = ∞ if det F ≤ 0, we see that det ∇y > 0 for
almost every x ∈ . Furthermore, the weak form of (48)1 reads
 
dk · ∇φ d Vx = 0, bk · ∇φ d Vx = 0
R3 R3

for each indefinitely integrable function φ : R3 → R with compact support. The convergence indicated in
(53)4,5 then yields
 
d · ∇φ d Vx = 0, b · ∇φ d Vx = 0,
R3 R3
Šilhavý 19

that is,
div d = 0, div b = 0 R3 in the sense of distributions.
As one also finds that yk = ỹ on ∂ implies y = ỹ on ∂, we see that the triplet σ = (y, d, b) belongs to S.
Also, trivially in view of (49),
 
−1
lim inf J (|Fd k | + |Fbk | ) d Vx ≥
2 2
J −1 (|Fd|2 + |Fb|2 ) d Vx , (55)
k→∞ c c
 
lim yk · g d Vx = y · g d Vx . (56)
k→∞  

Inequalities (54)–(56) can be collected to show that

lim inf E(σk ) ≥ E(σ )

k→∞

which shows that σ is the required minimizer. This completes the proof in the case n = 3.
The proof is similar if n = 2. Instead of Proposition A2.1 one has to use the simpler result of Reshetnyak
[46, 47] and Ball [1] that yk  y in W 1,p (, R2 ) with p > 2 implies det Fk  F in Lp/2 (); moreover, one
more use of the div–curl lemma Proposition A2.2 is needed to show that dk  d in Lr (, R2 ) and bk  b in
Ls (, R2 ) implies that dk × bk  d × b in L1 (). For this, one has to identify the sequence gk of Proposition
A2.2 with (d2,k , −d1,k ) so that div dk = 0 reads curl gk = 0. 

Acknowledgement
I thank M. Itskov for fruitful discussions.

Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The
support of the institutional research plan RVO 67985840 is gratefully acknowledged.

Notes
1. See Section 2 for the notation.
2. The paper [14] proves a general result which yields the electro-magneto-elastic quasiaffine functions as a special case.
3. A companion paper [15] treats isotropic polyconvex electro-magneto-elastic bodies.
4. I thank M. Itskov for drawing my attention to these papers.

References
[1] Ball, JM. Convexity conditions and existence theorems in nonlinear elasticity. Arch Rational Mech Anal 1977; 63: 337–403.
[2] Morrey Jr, CB. Quasi-convexity and the lower semicontinuity of multiple integrals. Pacific J Math 1952; 2: 25–53.
[3] Morrey Jr, CB. Multiple integrals in the calculus of variations. New York: Springer, 1966.
[4] Schröder, J, and Neff, P. Invariant formulation of hyperelastic transverse isotropy. Int J Solid Struct 2003; 40: 401–445.
[5] Steigmann, DJ. Frame-invariant polyconvex strain-energy functions for some anisotropic solids. Math Mech Solids 2003; 8:
497–506.
[6] Itskov, M, and Aksel, N. A class of orthotropic and transversely isotropic hyperelastic constitutive models based on a polyconvex
strain energy function. Int J Solids Struct 2004; 41: 3833–3848.
[7] Dacorogna, B. Weak continuity and weak lower semicontinuity of non-linear functionals. Berlin: Springer, 1982.
[8] Fonseca, I, and Müller, S. A-quasiconvexity, lower semicontinuity, and Young measures. SIAM J Math Anal 1999; 30: 1355–1390.
[9] Murat, F. Compacité par compensation. Ann Scuola Normale Sup Pisa Cl Sci S IV 1978; 5: 489–507.
[10] Murat, F. Compacité par compensation: condition nécessaire et suffisante de continuité faible sous une hypothése de rang
constant. Ann Scuola Normale Sup Pisa Cl Sci S IV 1981; 8: 68–102.
[11] Braides, A, Fonseca, I, and Leoni, G. A-quasiconvexity: relaxation and homogenization. ESAIM Contr Optim Ca 2000; 5:
539–577.
[12] Krämer, J, Krömer, S, Kružzík, M, et al. A-quasiconvexity at the boundary and weak lower semicontinuity of integral functionals.
Adv Calc Var 2015; 10: 49–67.
20 Mathematics and Mechanics of Solids

[13] Matias, J, Morandotti, M, and Santos, PM. Homogenization of functionals with linear growth in the context of A-quasiconvexity.
Appl Math Opt 2015; 72: 523–547.
[14] Šilhavý, M. The polyconvexity for functions of several closed differential forms, with applications to electro-magneto-
elastostatics. 2016.
[15] Šilhavý, M. Isotropic polyconvex electro-magneto-elastic bodies. Math Mech Solids 2017.
[16] Tartar, L. Compensated compactness and applications to partial differential equations. In: Nonlinear Analysis and Mechanics:
Heriot-Watt Symposium (ed R Knops), 1979, 136–212. Harlow: Longman.
[17] Pedregal, P. Parametrized measures and variational principles. Basel: Birkhäuser, 1997.
[18] Dacorogna, B, and Fonseca, I. A-B quasiconvexity and implicit partial differential equations. Calculus of Variations and Partial
Differential Equations 2002; 14: 115–149.
[19] Kankanala, SV, and Triantafyllidis, N. On finitely strained magnetorheological elastomers. J Mech Phys Solid 2004; 52: 2869–
2908.
[20] Itskov, M and Khiem, VN. A polyconvex anisotropic free energy function for electro- and magneto-rheological elastomers. Math
Mech Solids 2016; 21: 1126–1137.
[21] Foss, M, and Randriampiry, N. Some two-dimensional A-quasiaffine functions. Contemp Math 2010; 514: 133–141. (see also
http://bookstore.ams.org/conm-514#sthash.5RscJU7u.dpuf).
[22] Gil, AJ, and Ortigosa, R. A new framework for large strain electromechanics based on convex multi-variable strain energies:
Variational formulation and material characterisation. Comput Methods Appl Mech Eng 2016; 302: 293–328.
[23] Ortigosa, R, and Gil, AJ. A new framework for large strain electromechanics based on convex multi-variable strain energies:
Conservation laws and hyperbolicity and extension to electro-magnetomechanics. Comput Method Appl M, September 2016;
309: 2020150242.
[24] Ortigosa, R, and Gil, AJ. A new framework for large strain electromechanics based on convex multi-variable strain energies:
Finite Element discretisation and computational implementation. Comput Methods Appl Mech Eng 2016; 302: 329–360.
[25] Truesdell, C, and Noll, W. The non-linear field theories of mechanics. In: Fluegge, S (ed.) Handbuch der Physik III/3. Berlin:
Springer, 1965.
[26] Šilhavý, M. The mechanics and thermodynamics of continuous media. Berlin: Springer, 1997.
[27] Rockafellar, RT. Convex analysis. Princeton, NJ: Princeton University Press, 1970.
[28] Ekeland, I, and Temam, R. Convex analysis and variational problems. Philadelphia, PA: SIAM, 1999.
[29] Fonseca, I, and Leoni, G. Modern methods in the calculus of variations: Lp spaces. New York: Springer, 2007.
[30] Truesdell, C, and Toupin, RA. Classical field theories of mechanics. In: Fluegge, S (ed.) Handbuch der Physik III/1. Berlin:
Springer, 1960.
[31] Brown, WF. Magnetoelastic interactions. Berlin: Springer, 1966.
[32] Hutter, K, van de Ven, AAF, and Ursescu, A. Electromagnetic field matter interactions in thermoelastic solids and viscous fluids
(Lecture Notes in Physics, vol. 710). Berlin: Springer, 2006.
[33] Maugin, G. Continuum mechanics of electromagnetic solids. Amsterdam: North-Holland, 1988.
[34] Eringen, AC, and Maugin, GA. Electrodynamics of continua I. New York: Springer, 1990.
[35] Eringen, AC, and Maugin, GA. Electrodynamics of continua II. New York: Springer, 1990.
[36] Kovetz, A. Electromagnetic theory. Oxford: Oxford University Press, 2000.
[37] Hutter, K. On thermodynamics and thermostatics of viscous thermoelastic solids in the electromagnetic fields. A Lagrangian
formulation. Arch Rational Mech Anal 1975; 58: 339–368.
[38] Marsden, JE, and Hughes, TJR. Mathematical foundations of elasticity. New York: Dover Publications, 1994.
[39] Dorfmann, L, and Ogden, RW. Nonlinear theory of electroelastic and magnetoelastic interactions. New York: Springer, 2014.
[40] Van Hove, L. Sur l’extension de la condition de Legendre du calcul des variations aux intégrales multiples à plusieurs fonctions
inconnues. Proc Koninkl Ned Akad Wetenschap 1947; 50: 18–23.
[41] Ball, JM, Currie, JC, and Olver, PJ. Null lagrangians, weak continuity and variational problems of any order. J Funct Anal 1981;
41: 135–174.
[42] Giaquinta, M, Modica, G, and Souček, J. Cartesian currents, weak diffeomorphisms and existence theorems in non-linear
elasticity. Arch Rational Mech Anal 1989; 106: 97–160.
[43] Müller, S, Tang, Q, and Yan, BS. On a new class of elastic deformations not allowing for cavitation. Ann Inst H Poincaré, Analyse
non linéaire 1994; 11: 217–243.
[44] Giaquinta, M, Modica, G, and Souček, J. Cartesian currents in the calculus of variations I, II. Berlin: Springer, 1998.
[45] Müller, S. Higher integrability of determinant and weak convergence in L1 . J Reine Angew Math 1990; 412: 20–34.
[46] Reshetnyak, YuG. General theorems on semicontinuity and on convergence with a functional. Siberian Math J 1967; 8: 801–816.
[47] Reshetnyak, YuG. Stability theorems for mappings with bounded excursion. Siberian Math J 1968: 9: 499–512.
[48] Müller, S. Variational models for microstructure and phase transitions. In: Hildebrandt, S and Struwe, M (eds) Calculus of
variations and geometric evolution problems (Cetraro, 1996) (Lecture Notes in Mathematics, vol. 1713). Berlin: Springer, 1999,
85–210.
[49] Ericksen, JL. Nilpotent energies in liquid crystal theory. Arch Rational Mech Anal 1962; 10: 189–196.
[50] Edelen, DGB. The null set of the Euler–Lagrange operator. Arch Rational Mech Anal 1962; 11: 117–121.
[51] Ball, JM, and Murat, F. W 1,p -quasiconvexity and variational problems for multiple integrals. J Funct Anal 1984; 58: 225–253.
Šilhavý 21

Appendix 1
A1.1. Rank 1 convex and rank 1 affine functions
The reader is referred to [1, 7, 48] for the following notions.
Definition A1.1. Let g : Mn×n → R̄.
(i) g is said to be rank 1 convex if
g(F + tξ ⊗ η) ≤ (1 − t)g(F) + tg(F + ξ ⊗ η) (57)
for every t ∈ (0, 1), every F ∈ Mn×n and every ξ , η ∈ Rn .
(ii) g is said to be rank 1 affine if it takes only finite values and (57) holds with the equality sign for every t, F,
ξ , and η as in (i).

The following result is standard.

Lemma A1.2. [49,50] and [1, Theorem 4.1] A continuous function g : Mn×n → R is rank 1 affine if and only
if g is a linear combination, with constant coefficients, of the functions

1, F, det F, if n = 2,
(58)
1, F, cof F, det F if n = 3.

Appendix 2
A2.1. Weak convergence
We here gather some basic facts about maps that are continuous under the weak convergence.
Let 1 ≤ p ≤ ∞ and let θk and θ be measurable functions on open subset  of Rn . In this situation, we define
the following three types of weak convergence:
θk  θ Lp (, Rm ), (59)
∗
θk  θ M(, R ), m
(60)
θk  θ W 1,p (, Rm ) (61)
which mean, respectively:
• that θ, θk ∈ Lp (, Rm ) and  
θk · φ d Vx → θ · φ d Vx (62)
 
for each φ ∈ Lq (, Rm ) where 1/p + 1/q = 1; we then say that the sequence θk converges weakly to θ in
Lp (, Rm );
• that θ, θk ∈ L1 (, Rm ) and (62) holds for each continuous function φ : Rn → Rm which vanishes outside
; we then say that the sequence θk converges weak∗ to θ in the sense of measures;
• that θ, θk ∈ W 1,p (, Rm ) and θk  θ in Lp (, Rm ) and ∇θk  ∇θ in Lp (, Mm×n ); we then say that θk
converges weakly to θ in W 1,p (, Rm ).
If p = ∞, we should actually write ∗ instead of  and speak about the weak∗ convergence; however, this is
consistently ignored here.
Proposition A2.1. (Müller et al. [43]) Let  be a bounded open subset of Rn where n is arbitrary, let
p ≥ n − 1, q > n/(n − 1)
and let y, yk ∈ W 1,p (, Rn ) satisfy
yk  y in W 1,p (, Rn ), (63)
cof Fk is bounded in Lq (, Mn×n ) (64)
22 Mathematics and Mechanics of Solids

where F = ∇y, Fk = ∇yk . Then

cof Fk  cof F in Lq (, Mn×n ), (65)

det Fk  det F in Lr (), r = q(n − 1)/n. (66)

If q = n/(n − 1) and det Fk ≥ 0 then instead of (66) we have

det Fk  det F in L1 (K),

for all compact subsets K ⊂ .

Proposition A2.2. (Murat [9], Tartar [16]) Let  ⊂ Rn be open bounded and let 1 < p, q < ∞ satisfy
1/p + 1/q = 1. Suppose d, dk ∈ Lp (; Rn ), g, gk ∈ Lq (, Rn ) are sequences such that

dk  d in Lp (, Rn )
gk  g in Lq (, Rn ),
div dk → div d in W −1,1 (),
curl gk → curl g in W −1,1 ().

Then
dk · gk  d · g in M().

Here W −1,1 () is the dual of W01,∞ ().

Proposition A2.3. (Reshetnyak [46], Ball and Murat [51]) Let  : Rm → R̄ be convex, lower semicontinuous
and bounded below. Let θ, θk ∈ L1 (, Rm ) with θk ∗ θ in the sense of measures. Then
 
lim inf (θk ) d Vx ≥ (θ) d Vx .
j→∞