The current value of the strain function is
\(E(t)\) while the strain history and some relative histories are given by
Equation (
2.1)
3 differs from (
2.1)
2 only by a change of notation, but is included to emphasize that the relative strain is not always defined with respect to the current time
\(t\). The simple property
$$ \frac{\textstyle \partial E_{r}^{t}(s)}{\textstyle \partial E(t)} = -1 $$
(2.2)
will be used later. It is assumed here that
$$ \lim _{s \rightarrow \infty } E^{t}(s) = \lim _{u \rightarrow -\infty } E(u) =0, $$
(2.3)
which simplifies certain formulae. Note also that
$$ \dot{E}^{t}(u) = \frac{\textstyle \partial }{\textstyle \partial t} E^{t}(u)= -\frac{\textstyle \partial }{\textstyle \partial u} E^{t}(u) = - \frac{\textstyle \partial }{\textstyle \partial u} E_{r}^{t}(u). $$
(2.4)
The stress function is given by
\(T(t) = \tilde{T}(E^{t},E(t))\), while we denote a particular free energy at time
\(t\) by
\(\psi (t) = \tilde{\psi }(E^{t},E(t))\), where
\(\tilde{T}(E^{t},E(t))\) and
\(\tilde{\psi }(E^{t},E(t))\) are understood to be functionals of
\(E^{t}\) and functions of
\(E(t)\). Certain properties of free energies were derived in [
2] and used to characterize such quantities in [
3‐
5] and elsewhere. They have been referred to as the Graffi [
6] definition of (or conditions for) a free energy and are given as follows. Let
\(\psi (t) = \tilde{\psi }(E^{t},E(t))\) be a free energy. Then
P1:
$$ \frac{\textstyle \partial }{\textstyle \partial E(t)}\tilde{\psi }(E^{t},E(t)) = \frac{\textstyle \partial }{\textstyle \partial E(t)}\psi (t) = T(t). $$
(2.5)
P2:
For any history
\(E^{t}\) $$ \tilde{\psi }(E^{t},E(t)) \geq \tilde{\phi }(E(t)) \;\; \mbox{or} \;\; \psi (t) \geq \phi (t), $$
(2.6)
where
\(\phi (t)> 0\) is the equilibrium value of the free energy
\(\psi (t) \), defined as
Thus, equality in (
2.6) is achieved for equilibrium conditions. Observe that the classical elastic energy
\(\phi (t)\) is always taken to be positive so that, from (
2.6), the quantity
\(\psi (t)\) also has this property.
P3:
It is assumed that
\(\psi \) is differentiable. For any
\((E^{t},E(t))\) we have the first law (balance of energy)
$$ \dot{\psi }(t) + D(t) = T(t)\dot{E}(t), $$
(2.7)
where
\(D(t)\geq 0\) is the rate of dissipation of energy associated with
\(\psi (t)\). This nonnegativity requirement on
\(D(t)\) is an expression of the second law.
The basic condition is P3. Relations P1 and P2 follow from P3.
Integrating (
2.7) over
\((-\infty ,t]\) yields that
$$ \begin{aligned} &\psi (t) + {\mathcal{D}}(t) = W(t), \quad \quad {\mathcal{D}}(t) \geq 0, \\ &{\mathcal{D}}(t) = \int _{-\infty }^{t}D(u) d u, \quad W(t) = \int _{- \infty }^{t}T(u) \dot{E}(u) d u. \end{aligned} $$
(2.8)
The quantity
\(W(t)\) is the work function, while
\({\mathcal{D}}(t)\) is the total dissipation resulting from the entire history of deformation of the body. It is assumed that the integrals in (
2.8) are finite. In particular, we must have
$$ \lim _{t \rightarrow -\infty } W(t) = 0, $$
(2.9)
with a similar assumption for
\({\mathcal{D}}(t)\). The time derivative of
\(\tilde{\psi }(E^{t},E(t))\) consists of the ordinary time derivative of the
\(E(t)\) dependence, giving, with the aid of (
2.5),
$$ T(t)\dot{E}(t) = \dot{W}(t), $$
(2.10)
and a functional derivative of the history dependence
\(E^{t}\), which yields the dissipation. Note that (
2.10) follows from (
2.8)
4.
Let us now briefly demonstrate that the work function also obeys P1, using a somewhat modified version of the argument in [
2], and P2. We can write
$$ \dot{W}(t) = \frac{\textstyle \partial }{\textstyle \partial E(t)}{ \widetilde{W}}(E^{t},E(t))\dot{E}(t) + \delta {\widetilde{W}}, $$
(2.11)
where the rightmost term is a Fréchet differential of
\({\widetilde{W}}\), defined within a suitable Hilbert space (for example [
1], page 104). Thus, (
2.10) can be written in the form
$$ \left [ \frac{\textstyle \partial }{\textstyle \partial E(t)}{ \widetilde{W}}(E^{t},E(t)) - {\widetilde{T}}(E^{t},E(t))\right ] \dot{E}(t) = -\delta {\widetilde{W}}. $$
(2.12)
The quantity
\(\dot{E}(t)\) can take arbitrary values, so that (
2.5) or P1 must hold for
\(\tilde{\psi }\) replaced by
\({\widetilde{W}}\). Thus, we have
$$ \frac{\textstyle \partial }{\textstyle \partial E(t)}\psi (t) = \frac{\textstyle \partial }{\textstyle \partial E(t)}W(t) = T(t), $$
(2.13)
giving
$$ {\widetilde{\mathcal{D}}}(E^{t},E(t)) = {\widetilde{\mathcal{D}}}(E^{t}). $$
(2.14)
It follows that
$$ \frac{\textstyle \partial }{\textstyle \partial E(t)} {\mathcal{D}}(t) = 0. $$
(2.15)
Also, the quantity
\(\delta {\widetilde{W}}\) must vanish.
2.1 Kernels of Free Energy Terms
The following functions are introduced:
The quantity
\(N\) is a positive integer, which may be infinite. If it is infinite, convergence assumptions must be included for the series involved. The superscript on
\(G\) indicates the total number of arguments. We shall argue below that, for even
\(k\), these may be the kernels of a free energy, while for odd
\(k\), a particular choice of these quantities may define the stress function derived from the free energy (see (
2.5)).
The quantities
\(G^{(k)}\) \(k=1,2, \ldots , N\) and related functions introduced below first enter the model by their occurrence in integrals of the form
$$ \begin{aligned} \int _{0}^{\infty }G^{(k)}&(u_{1},u_{2},\dots ,u_{k})f(u_{1})f(u_{2}) \dots f(u_{k}) d u_{1} d u_{2} \dots du_{k} \\ & =\int _{0}^{\infty }G^{(k)}({\mathbf {u}}^{(k)})\prod \limits _{i=1}^{k} f(u_{i}) d {\mathbf {u}}^{(k)}, \\ d{\mathbf {u}}^{(k)}& = d u_{1}d u_{2},\dots ,d u_{k},\,\,\,\,k=1,2,\dots , N, \end{aligned} $$
(2.18)
where the single integral sign here and below is understood to mean as many integral signs as there are
\(d u_{1}\),
\(d u_{2}\) etc. The function
\(f(u)\), which is always related to the strain history, can be chosen arbitrarily, subject to the requirement that the integral exists. Note that
$$ \begin{aligned} \int _{0}^{\infty }G^{(k)}&(u_{1},u_{2},\dots ,u_{k})f(u_{1})f(u_{2}) \dots f(u_{k}) d u_{1} d u_{2} \dots du_{k} \\ & = \int _{0}^{\infty }G^{(k)}(u_{2},u_{1},\dots ,u_{k})f(u_{1})f(u_{2}) \dots f(u_{k}) d u_{1} d u_{2} \dots du_{k}, \end{aligned} $$
(2.19)
which follows by using the standard device of renaming integration variables. A similar property holds for any permutation of
\(u_{1}, u_{2}, \dots , u_{k}\). The important property here is the complete symmetry of the product of
\(f(u_{i})\). Thus, only the totally symmetric part of the kernel contributes to the integral. For simplicity, we assume that
$$ G^{(k)}(u_{1},u_{2},\dots ,u_{k}) = G^{(k)}(u_{2},u_{1},\dots ,u_{k}), $$
(2.20)
and similarly for any other permutation. In other words, we take
\(G^{(k)}\) \(k=1,2, \ldots , N\) and similar related quantities to be completely symmetric in all their arguments.
For odd
\(k\), the functional of
\(f(u)\) given by (
2.18) changes sign if
\(f(u)\) is replaced by
\(-f(u)\), so that it cannot be positive definite or semi-definite for all choices of
\(f(u)\). If
\(k\) is even, then this non-negativity property can hold, provided certain restrictions are imposed on the kernel.
We only consider free energies consisting of terms with even
\(k\); the constraints on the kernels to guarantee non-negativity are assumed to hold. Each term separately will be non-negative. Equation (
2.5) then gives that the stress function will only consist of terms with odd
\(k\).
It is assumed that the quantities \(G^{(k)}({\mathbf {u}}^{(k)})\), similar kernels introduced below and indeed the strain histories, have differentiability properties as required in the various contexts discussed in this work.
The constants
\(G^{(k)}_{\infty }\) are given by
$$ G^{(k)}_{\infty }= \lim _{u_{j} \rightarrow \infty }G^{(k)}(u_{1},u_{2}, \dots ,u_{k}), $$
(2.21)
where
\(j\) is any integer in the set
\(\{1,2,\ldots ,k\}\). The fact that this quantity is the same for every
\(j\) is an expression of the complete symmetry of the dependence of
\(G^{(k)}\) on its parameters. We illustrate this with the case
\(k=2\), where
\(G^{(2)}(u_{1},u_{2})\) is converging to let us say
\(G_{1}\) if the first argument is very large and
\(G_{2}\) if the second is very large. The symmetry property
\(G^{(2)}(u_{1},u_{2})=G^{(2)}(u_{2},u_{1})\) gives that
\(G_{2}=G_{1}\).
The quantities
\(G^{(k)}_{\infty }\) are non-negative. Let us define
$$ \begin{aligned} {\widetilde{G}}^{(k)}({\mathbf {u}}^{(k)}) & = G^{(k)}({\mathbf {u}}^{(k)})-G^{(k)}_{\infty }, \\ {\mathcal{G}}^{(k)}({\mathbf {u}}^{(k)}) & = G^{(k)}_{123\dots k}({\mathbf {u}}^{(k)}), \,\,\,\,k=1,2,\dots , N, \end{aligned} $$
(2.22)
where a subscripted integer
\(j\) indicates partial differentiation with respect to the corresponding
\(u_{j}\). Thus
$$ {\mathcal{G}}^{(k)}({\mathbf {u}}^{(k)})= \partial _{1} \partial _{2} \dots \partial _{k} G^{(k)}({\mathbf {u}}^{(k)}), \,\,\,\,k=1,2,\dots , N, $$
(2.23)
where the operator
\(\partial _{j}\) is the partial derivative
\(\frac{\textstyle \partial }{\textstyle \partial u_{j}}\).
The convergence of the integral (
2.19)
1 is better at large
\(u_{j}\) if
\(G^{(k)}(u_{1},u_{2},\dots ,u_{k})\) is replaced by
\({\widetilde{G}}^{(k)}(u_{1},u_{2},\dots ,u_{k})\).
Consider the function space
\({\mathcal{F}}^{(k)}\), where
\(k\) is even, with a norm
$$ {\mathcal{N}}^{(k)}(f) = \int _{0}^{\infty }{\widetilde{G}}^{(k)}({\mathbf {u}}^{(k)}) \prod \limits _{i=1}^{k} f(u_{i})d {\mathbf {u}}^{(k)}\geq 0, \qquad { \mathcal{N}}^{(k)}(f) < \infty . $$
(2.24)
This is a non-negative, finite quantity, as indicated. We assume that the function
\(f\), which in this context is always related to strain history, belongs to the function space
$$ {\mathcal{F}}(f) = {\mathcal{F}}^{(0)} \cup {\mathcal{F}}^{(2)} \dots \cup {\mathcal{F}}^{(N)}, $$
(2.25)
where
\(N\) is even and
\({\mathcal{F}}^{(0)}\) contains functions without history dependence, specifically the quantities
\(\phi (t)\), defined by (
3.3) below, with
\(E(t)\) replaced by
\(f(t)\).
2.2 The Kernels of the Rate of Dissipation Terms
We also define the functions
$$ \begin{aligned} K^{(k)}({\mathbf {u}}^{(k)})& = \sum _{i=1}^{k} \partial _{i} G^{(k)}({\mathbf {u}}^{(k)}), \\ {\mathcal{K}}^{(k)}({\mathbf {u}}^{(k)})& = \sum _{i=1}^{k} \partial _{i} { \mathcal{G}}^{(k)}({\mathbf {u}}^{(k)}), \,\,\,\,k=2,4, \dots , 2 N, \end{aligned} $$
(2.26)
which will be shown to be the negative of the kernels making up the rate of dissipation
\(D(t)\), introduced in (
2.7). These kernels are defined only for the positive even integers. They have the property that
$$ K^{(k)}({\mathbf {u}}^{(k)}), {\mathcal{K}}^{(k)}({\mathbf {u}}^{(k)}) \rightarrow 0 \,\,\,\,\mbox{for every parameter} \,\,\,\,u_{i} \rightarrow \infty , \,\,\,\,i=1,2,\dots , k, $$
(2.27)
where
\(k=2,4, \dots , 2 N\).
A difficulty in constructing free energy functionals arises in making choices that ensure nonnegative functional forms both for the free energy and for the rate of dissipation. A method, proposed in the context of the linear model [
7], which renders this task more straightforward, is generalized below to higher terms. Instead of constructing the free energy and determining from this the rate of dissipation, which may not have the required nonnegativity, the procedure is reversed, which guarantees a satisfactory free energy functional [
7] (see also [
1], pg. 394). One chooses a nonnegative functional for the rate of dissipation. Formulae are presented below which give the associated free energy functional in terms of the dissipation rate kernel. It emerges that the resulting free energy has the required nonnegativity property.
There are two equivalent alternatives for the developments outlined below, the first being to use \({\widetilde{G}}^{(k)}({\mathbf {u}}^{(k)})\), \(K^{(k)}({\mathbf {u}}^{(k)})\), \(\dot{E}^{t}(s)\) and the second to use \({\mathcal{G}}^{(k)}({\mathbf {u}}^{(k)})\), \({\mathcal{K}}^{(k)}({\mathbf {u}}^{(k)})\), \(E^{t}_{r}(s)\). Both have been widely adopted in discussing the minimum and related free energies for linear constitutive materials. In the present work, we will mainly present formulae in both notations. Moving from one alternative to the other involves a series of integrations by parts.
We can in fact choose either alternative for each parameter
\(u_{i}\), rather than fix on the same choice for all
\(u_{i}\), as in (
3.1), below. For example, for
\(u_{i}\), we could switch from
\({\widetilde{G}}^{(k)}({\mathbf {u}}^{(k)})\),
\(K^{(k)}({\mathbf {u}}^{(k)})\),
\(\dot{E}^{t}(u_{i})\) to
\(\partial _{i} G^{(k)}({\mathbf {u}}^{(k)})\),
\(\partial _{i} K^{(k)}({\mathbf {u}}^{(k)})\),
\(E^{t}_{r}(u_{i})\).