M.A. A Biot - Fluence

..- ,I.l.uIY .&. ‘._. l.yy ...yv . . . . &._.,_) . _1_.._1 _.
New Variational-Lagrangian
Irreversible Thermodynamics with Application to
Viscous Flow, Reaction-Diffusion,
and Solid Mechanics
M. A. BIOT
Royal Academy of Belgium, Brussels, Belgium
I. Introduction ........................................ 2
II. Restructured Thermodynamics of Open Systems and the Concept of
Thermobaric Transfer. ................................. 6
III. New Chemical Thermodynamics .......................... 9
IV. Homogeneous Mixtures and Reformulation of the Gibbs-Duhem
Theorem .......................................... 13
V. Nontensorial Virtual Work Approach to Finite Strain and Stress ..... 16
VI. Thermodynamic Functions of Open Deformable Solids. ........... 19
VII. The Fluence Concept. ................................. 21
VIII. The Nature of Entropy Production and Its Evaluation ............ 23
IX. The Principle of Virtual Dissipation ........................ 30
X. General Lagrangian Equations ............................ 3.5
XI. Dynamics of Viscous Fluid Mixtures with Reaction-Diffusion and
Radiation Pressure. ................................... 36
XII. Dynamics of Solids with Elastoviscous Stresses and Heat Conduction,
and Thermoelasticity .................................. 44
XIII. Inhomogeneous Viscous Fluid with Convected Coordinates and Heat
Conduction. ........................................ 51
XIV. Lagrangian Equations of Heat Transfer and Their Mechanical
Interpretation, and a Mass Transfer Analogy .................. 56
xv. Deformable Solids with Thermomolecular Diffusion and Chemical
Reactions. ......................................... 59
XVI. Thermodynamics of Nonlinear Viscoelasticity and Plasticity with
Internal Coordinates and Heredity ......................... 62
XVII. Dynamics of a Fluid-Saturated Deformable Porous Solid with Heat and
Mass Transfer. ...................................... 64
XVIII. Linear Thermodynamics near Equilibrium .................... 68
XIX. Linear Thermodynamics of a Solid under Initial Stress. ........... 13
I
Copyright 6 1984 by Academic Press, Inc.
All rights of reproduction in any form reserved.
ISBN O-12-002024-6
2 M. A. Biot
XX. Linear Thermodynamics and Dissipative Structures near Unstable

Equ~ib~um.......................................... 78
XXI. Thermoelastic Creep Buckling ..... ............... 81
XXII. Lagrangian Formulation of Bifurcations. ..... ....... 82
XXIII. Generalized Stability Criteria for Time-Dependent Evolution Far from
Equilibrium. . . . . ... . .............. ....... 82
XXIV. Creep and Folding Instability of a Layered Viscous Solid. . . . . . . . . . . 85
XXV. Coupling of Subsystems and the Principle of Interconnection .... 86
XXVI. Completeness of the Description by Generalized Coordinates. Resolution
Threshold and Lagrangian Finite Element Methods . . . . . ...... 88
XXVII. Lagrangian Equations in Configuration Space. Internal Relaxation,
Order-Disorder Phenomena, and Quantum Kinetics . . . . . . ..... 89
References . . . ... ... .. .... ... ....... .. . 89
I. Introduction
Our purpose here is to present a comprehensive view of a distinct

approach to the thermodynamics of irreversible processes that is based on
a principle of virtual dissipation. It represents a generalization of d’Alem-
bert’s principle of classical mechanics to irreversible thermodynamics
and leads to equations of evolution of thermodynamic systems. They are
derived directly either as field equations or as Lagrangian equations.
This work originated in the years 1954-1955, first in the context of
linear phenomena, and developed gradually along with its applications
during the next 25 years. During this period it became evident that the
concepts and methods are interdisciplinary. They extend far beyond pure
mechanics and in particular should include chemistry. It therefore be-
came necessary to reexamine some of the foundations of classical thermo-
dynamics. This was accomplished more recently and has led to a restruc-
turing of the thermodynamics of open systems. The approach is based on
physical “thought experiment” procedures, in contrast with current stan-
dard formalism. It provides new definitions for the energy and entropy of
open systems that bypass the fundamental difficulties of the classical
Gibbs approach without recourse to quantum statistics. The well-known
Gibbs paradox is avoided, and the Gibbs chemical potential is replaced by
a “convective potential” that does not involve undetermined constants.
The Gibbs-Duhem theorem is reformulated accordingly. Results in
thermochemistry introduce the concept of “intrinsic heat of reaction,”
which is more representative of chemical energy than the concepts de-
rived from standard definitions. It leads to a new expression for the affin-
New Variational-Lagrangian Irreversible Thermodynamics 3
ity and to a rigorous and complete generalization of the Kirchhoff

formula.
Thus the work involves two distinct developments, constituted on the
one hand by a new approach to classical thermodynamics and on the
other by a variational-Lagrangian formulation of irreversible processes.
The first provides a sound basis for the second. The same logical se-
quence will be followed here, and many applications to mechanics and
coupled chemical reaction-diffusion will be presented.
Attention should be called to some of the more fundamental aspects of
the concept of virtual dissipation and the associated Lagrangian formula-
tion. The concept itself does not postulate any particular kinetics of the
phenomena involved. This provides great flexibility in the applications,
because one can introduce either a kinetics based on Onsager’s principle
or a more general one represented, for example, by nonlinear chemical
reactions or a non-Newtonian viscosity. The extreme generality of this
viewpoint opens a wide domain that does not exclude quantum kinetics.
The variational principle represents essentially a probing of the system
in the vicinity of a frozen state of evolution. It is accomplished by apply-
ing virtual changes that obey constraints of mass and energy conservation
and by evaluating the virtual entropy of an equivalent closed and adia-
batic system that represents the virtual entropy produced. In this process
the virtual work of the inertial forces is taken into account. This implies a
generalization of d’Alembert’s principle of mechanics to include the ther-
mal energy. A fundamental aspect of the principle of virtual work in
mechanics seems to have been overlooked. It is not merely a formal tool
but embodies a physical law, sometimes called the third law of mechan-
ics, namely that action and reaction forces are equal and opposite. This
leads to the vanishing of virtual work between frictionless or adherent
surfaces and the disappearance of interfacial forces from the equations.
Its generalization involves the continuity of mass and energy flux and
leads to a principle of interconnection of subsystems whereby the general-
ized interfacial forces, mechanical as well as thermodynamic, of interact-
ing subsystems are eliminated from global equations.
The formulation of the principle was based on the introduction of a
thermodynamic function that was shown by the author to be fundamental
in irreversible thermodynamics. It was initially referred to as “general-
ized free energy.” A formally similar expression was later given the name
“exergy” in the literature. We have retained this term for the more pre-
cise definition required by the present context, using the concept of
“hypersystem” as a basic model in the thermodynamics of open systems.
The viewpoint here is completely different from the purely formal con-
ventional procedures that start from the differential field operators and
4 M. A. Biot
then derive corresponding variational principles. This implies a prior

knowledge of the equations that govern the continuous field, and each
particular case must then be treated separately with its own boundary
conditions. In this context the deeper unitary physical insight is lost. By
contrast, the principle of virtual dissipation introduces fundamental phys-
ical invariants that lead at the same time to field differential equations of
evolution of the continuous system as well as to Lagrangian equations of
collective evolution with generalized coordinates. In some problems new
physical terms whose existence has been overlooked are thus obtained
and are revealed essentially by the variational approach. Prior knowledge
of the jield differential equations is thus not required.
One may analyze very complex systems made up of components of
physically very different natures. Lagrangian equations are then derived
for each subsystem by simplified analysis, and the subsystems are cou-
pled together using a principle of interconnection yielding Lagrangian
equations for the global system from which the interfacial coupling forces
have been eliminated. This includes modal synthesis.
Essentially the procedure also provides the foundation of a large uari-
ety offinite-element methods without requiring the use or knowledge of
the field equations of the continuum.
The ideas embodied in this way of thinking may be considered as the
natural extension to thermodynamics of those developed by Lagrange,
d’Alembert, and Rayleigh in the more restricted domain of mechanics.
The use of global generalized coordinates to describe the evolution of
complex mechanical structures has been common practice, particularly
among aeronautical engineers. A classic example is also provided by the
Lagrangian equations of motion of a rigid solid immersed in a perfect
incompressible fluid. The generalized coordinates are fundamentally
more general than those obtained by a choice of basis in functional
spaces. This is exemplified by the concept of “penetration depth” in
diffusion where, starting from a linear problem, a nonlinear equation is
obtained that is easier to solve than the original one. In analogy with La-
grangian mechanics, the generalized coordinates are chosen so that the
mechanical constraints as well as global or local conservation of mass and
energy are satisfied. This already solves part of the problem by the very
choice of coordinates and gives the method a remarkable accuracy with
very modest calculation requirements and rough approximations for as-
sumed distributions of the intensive variables. One may add that the La-
grangian formulation is another procedure for achieving what has been
one of the objectives of “computer algebra,” whereby complicated ana-
lytical expressions describing physical systems are manipulated by a com-
puter and simplified by elimination of negligible and physically nonrele-
vant tp.*ms. The Lagrangian equations directly provide such a formulation

leadi e: to simplified analytical results and improved physical insight. This
cha7 cter of simplicity is important, particularly in industrial problems,
where a good grasp of physical reality as well as reliability and low cost
are essential. The Lagrangian equations also lead naturally to general
methods of bifurcation and related stability analyses. New stability crite-
ria of evolution are obtained that are simpler and more general than those
& currently in use.
Some general remarks are in order regarding the use of spatial deriva-
tives. In continuum theories of matter, because of the discontinuous mo-
= lecular structure of matter, spatial derivatives are of only conventional
significance. From the viewpoint of the physicist, it is just as rigorous to
treat macroscopic problems dealing with matter by considering finite ele-
ments described by a finite number of coordinates, without using field
differential equations. The size of the finite element simply must not be
situated below a resolution threshold where the macroscopic laws break
down because of fluctuations and molecular scale effects. On the other
hand, the use of continuum models to represent material physical systems
constitutes an extrapolation beyond the validity of physical laws. Further-
more, they create spurious mathematical difficulties by forcing the intro-
duction of the concept of measure and associated properties of complete-
ness of representation that have no physical relevance. In fact, it has been
shown that much of the current fashionable continuum mechanics and
thermodynamics represents purely formal exercises without physical
foundation. These difficulties are eliminated if we use a description based
on finite elements of suitable size to represent the macroscopic physics
correctly. From this viewpoint, the corresponding Lagrangian equations
provide a rigorous description of the physical evolution.
We should add that purely formal methods often tend to mask the more
profound unitary aspect of the physical principles involved. It may even
be said that it is sometimes in the context of physical applications that
abstract generalizations are discovered. As an example, we may cite the
extension of variational methods in terms of symbolic Heaviside opera-
tors and corresponding convolutions, which has played the role of cata-
lyst in ulterior formal developments. The same remark can be made re-
garding the Lagrangian variational formulation, which has instigated
much formal work on finite-element methods.
The general ideas discussed stand in contrast with methods based on
prior knowledge of the field equations of evolution that derive variational
properties for each particular case by manipulation in the context of func-
tional spaces. Such formal methods have their place, but their role should
not be exclusive. Removed from their physical context they may lead to
6 M. A. Biot
serious errors and stand in the way of intuitive understanding of physical

problems as well as their analysis by simpler and more direct methods.
II. Restructured Thermodynamics of Open

Systems and the Concept of
Thermoharic Transfer
i
A new conceptual approach to the thermodynamics of open systems
has been developed (Biot, 1976a, 1977a) that avoids the difficulties inher-
ent in standard procedures without recourse to the axioms and ponderous
methods of quantum statistics. It replaces the classic formalism by an
operational approach using thought experiments on a physical model
called a hypersystem. This model is constituted by primary cells E Cr
(primary system), supply cells X C sk , and a thermal well TW. The supply
cells are large and rigid and contain pure substances denoted by k, all at
the same pressure p. and temperature To, whereas TW is a large rigid iso-
thermal reservoir at a temperature TOassumed for convenience to be the
same as in the supply cells.
For the present we consider an open primary cell CP constituted by a
fluid mixture. Its state is determined by its volume u, its temperature T,
and the masses Mk of each pure substance k added to it, starting from a
given initial state. An infinitesimal change of state may be obtained as
follows. We extract a mass dMk of pure substance adiabatically from the
supply cell Csk and compress and heat it gradually and reversibly to the
temperature T of CP and pressure pk so that it is in equilibrium with CP
through a semipermeable membrane. We then inject it reversibly into Cr
through the membrane. In addition we also inject into CP a quantity of
heat dh. This operation is called a thermobaric transfer. This type of
thermobaric transfer does not involve the thermal well TW. Another type
of thermobaric transfer that does involve TW will be considered later in
connection with the concept of exergy. We may perform this operation
for each pure substance and also increase the volume of the cell by the
amount du. These operations produce an increase of energy of the collec-
tive system Cp + x Csk equal to
d%=-pdv+xEkdMk+dh. (2.1)
k
We denote by p the total pressure acting on Cr, and Ek is given by
Ek = (2.2)
where pi, pi, T’, and d$ are, respectively, the pressure, density, temper-
ature, and specific entropy differential of the mass dM” along the path of
integration. The value of Ek includes the work of extraction of dMk from
Csk and injection into Cr , and T’ dfi is the heat injected into dMk at every
step of the thermobaric transfer.
Similarly, the increase of entropy of the subsystem Cr + c Csk is
PLT
3 dY = T Sk dM” + dhlT, Sk =
I
P,~~o
dG . (2.3)
We called pk the injection pressure and Ek and Sk, IXSpeCtiVdy, the in&X-
tion enthalpy and injection entropy. From Eq. (2.2) we derive the impor-
tant differential relation for each substance
dE:k = dpktpk + T dgk, (2.4)
where pk is the density at the temperature T and pressure pk.

By this thermobaric transfer process we may bring the cell Cp to any
desired temperature T and increase the masses of each substance in the
cell by arbitrary amounts Mk. Starting from a given initial state, the vari-
ables v, Mk, and T define the state of Cr. Because the Mk are the masses
extracted from the supply cells, they also determine the state of the sup-
ply cells. Hence the state of the subsystem CP + 2 Csk and its increase of
collective energy 011and entropy Y in the thermobaric transfer are deter-
mined by the same variables v, Mk, and T. As a consequence, we may
adopt % and Y as dejinitions of the energy and entropy of the primary cell
CP, keeping in mind that they refer to collective concepts.
To determine their values, we write the heat added dh in the form
dh = hkT dv + 2 hiT dMk + C,, dT. (2.5)

k
We call htT the heat of mixing at constant temperature and volume;

whereas C,, is the heat capacity of Cr at constant volume and composi-
tion. The differential coefficients h& and h’& may be evaluated without
calorimetric measurements by generalizing a procedure leading to the
classic Maxwell relations (Biot, 1982a). The result is obtained by substi-
tuting the value of dh from Eq. (2.5) into Eqs. (2.1) and (2.3) and noting
that d% and dY are exact differentials. We obtain the relations
-&(hkT - p) = $, (2.6)
=--1 =,lrl (2.7)

T dv ’
8 M. A. Biot
Elimination of X,,,,lau between these equations yields
The subscript urn indicates that the derivative is for u and Mk constant.
Similarly, we obtain the relations
(2.9) *
(2.10) c
Elimination of I~C,,I~IM~ between these equations yields

aEk
-- T aFk
z+T=o.
hk (2.11)
dT
Taking into account the differential relation (2.4), we derive
(2.12)
The derivatives in Eqs. (2.8) and (2.12) are obtained from relations
P = P(U, Mk, 0, (2.13)
Pk = Pk(% Mk, 0, (2.14)

which may be considered as generalized equations of state derived experi-
mentally or from kinetic theories. For perfect gas mixtures the injection
pressure is
Pk = Pyk, (2.15)
where yk is the molar fraction of substance k in the mixture.
With the value of dh from Eq. (2.5), the differential coefficients of d%
and d9 are now known functions of u, Mk, and T. We may then integrate
Eqs. (2.1) and (2.3) along any path, putting % = Y = 0 for the initial state.
We thus obtain
Q = %(u, Mk, T), Y = Y(u, Mk, T) (2.16)
for the collective energy and entropy of the open cell CP in terms of u, Mk,
and T.
By eliminating dh between Eqs. (2.1) and (2.3), we obtain
dQ=--pdu+~+kdMk+TdY, (2.17)
k
where
& = Ek - TFk (2.18)
is the convective potential (Biot, 1976a, 1977a). Relation (2.17) is analo-
gous to the Gibbs equation with & replacing the chemical potential. In
contrast with the classic procedure, Eq. (2.17) is not used to define & but
constitutes a theorem. In addition, &, Ek, Sk do not involve undetermined
9 constants as in the classic case. Note that these undetermined constants
are not eliminated by taking differentials, because we then obtain
_ d& = d& - T d$k - Sk dT, (2.19)
where, if we follow standard procedures, an undetermined constant still
remains for nonisothermal transformations in the coefficient Sk of dT. This
difficulty was already recognized by Gibbs himself (Gibbs, 1906) as well
as by others (see, e.g., Hatsopoulos and Keenan, 1965).
Gibbs’s paradox is also eliminated, as shown by Eq. (2.3), because for
identical substances the injection pressures I)k and hence also the injection
entropies Sk are the same, with the result that in Eq. (2.3) entropies
become additive. It is important to note that, when two components
become identical, Eq. (2.15) for the injection pressure in terms of the
molar fraction loses its validity.
This is consistent with the physical fact that in that case a semiperme-
able membrane loses its ability to distinguish between components.
III. New Chemical Thermodynamics
The concept of thermobaric transfer has provided a new approach to

chemical thermodynamics (Biot, 1976a, 1977a).
The first step is to introduce the new concept of intrinsic heat of reac-
tion defined as follows. A chemical reaction is measured by a reaction
coordinate .$ such that the masses of pure substances dmk produced by a
given chemical reaction d( are given by
dmk = vk d[, (3.1)
where r+ are constants characteristic of the reaction that satisfy the condi-
tion 2.k Vk= 0 of mass conservation. The coefficients are algebraic, so that
the masses produced may be positive or negative. Consider a reaction d,f
occurring in a rigid cell in such a way that the temperature is maintained
constant while the products of the reaction are extracted reversibly
through semipermeable membranes. Because the temperature and com-
10 M. A. Biot
position of the cell remain unchanged, its state and hence its pressure and
volume do not vary. The constancy of fhe temperature is obtained by
injecting into the cell a quantity of heat hr de. This defines the intrinsic
heat of reaction hT. How this is related to standard definitions can be
shown by reinjecting into the cell the products of the reaction either at
constant volume or at constant pressure, with T constant. We write
hur = hT + ck vk h:r, (3.2a)

x
hpr = ET + ck vk h;r,
where huT and hPT are the standard heats of reaction at constant tempera-
ture and, respectively, at constant volume and pressure. The heat of
mixing hiT at constant T and u is given by Eq. (2.12) and hiT denotes the
heat of mixing at constant temperature and pressure. By a method similar
to the previous derivation of h$T, we showed (Biot, 1982a) that the value
Of h;T iS
(3.3)
The derivative ~/cJTis for constant pressure and composition. For a per-
fect gas mixture, substitution of the value for& from (2.15) yields hiT = 0.
Hence in this case hpT k = 0, i.e., the standard and intrinsic heats of
reaction coincide.
Note that the intrinsic heat of reaction is more representative of chemi-
cal properties than standard concepts, because it does not involve the
heat of mixing or external work. For this reason its value is obtained by a
very general formula relating the two heats of reaction for two different
states. Consider a rigid cell in state 1 with values pr’, T, and another in
state 2 with values pp’, T2. A reaction d[ occurs in cell 1 and -d[ in cell 2,
while products of reaction are extracted from 1 and injected into 2 by
thermobaric transfer. The temperatures of the cells are kept constant by
injecting amounts of heat hg’ de and -h’,2’ d[, respectively, into each
cell. Conservation of energy implies the relation (Biot, 1977a) generalizing
Kirchhoff’s formula
(3.4)
Hence if we know the intrinsic heat of reaction for a single state, we may
derive its value (as well as the standard values huT and hpT) for any other
state of arbitrary composition and temperature.
In the preceding section we showed how to evaluate the energy and

entropy of an open cell by thermobaric transfer. The procedure may be
extended to the case where a chemical reaction also occurs in the cell.
This is accomplished by adding to the hypersystem a chemical supply cell
C,,, in which the reaction takes place and is in equilibrium at the tempera-
ture Teq, with injection pressures pkeq and an intrinsic heat of reaction h”+.
We consider a reaction d,f to occur in Cch while the products vkd,$ are
z extracted and injected into the primary cell CP by thermobaric transfer.
The temperature Teq of Cch is also kept constant by injecting the heat h”+
d.$. In this process the state of Cch does not vary. At the same time we
_ proceed as in the previous section by injecting masses dMk from Csk into
Cr, adding an amount of heat dh, and changing its volume by dv. The
change of state of the system through the reversible process of thermo-
baric transfer can be made to be the same as if a reaction dt had oc-
curred in the primary cell in addition to the other changes.
The increase of energy and entropy of the subsystem Cch + ECsk + Cp
is
d% = -p dv + c vk dt PkT dri + h”$ d.$

k I ~keqTeq
+zEkdMk+dh 9
k
with
dh = hkT dv + c h$(dMk + &‘kd[) + C,, dT. (3.7)

k
These results are obtained by adding to Eqs. (2.1), (2.3), and (2.5) the
terms due to the masses Vkd[ transferred and the heat hFq d[ injected into
Cch. We have also taken into account the fact that Cch is in equilibrium, so
no change of entropy occurs in the system due to the chemical reaction in
Cch .
The differentials (3.5) and (3.6) may be integrated from a given initial
state, yielding values
Q = Q(v, 6, Mk, T), 9 = Y(v, 5, Mk, T), (3.8)
functions of v, 8, Mk, and T, considered as state variables of the primary

cell CP. The values in Eq. (3.8) define the energy and entropy of CP in the
presence of a chemical reaction. Note that, starting from a given.initial
12 M. A. Biot
state 5 = 0, Mk = 0, the increase of masses of the various substances in

the cell is
mk = ,+.$ + Mk. (3.9)
However, the energy and entropy of the cell as dejined by Eqs. (3.5) and
(3.6) will vary euen if mk remains constant. This is due to the collective
definition of % and Y, which involves the subsystem Cch + c Csk + Cr.
On the other hand, in the equations of state,
P = P(u, mk, T), (3.10)
Pk = Pktv, mk, T) (3.11) 5
are functions Of Only u, mk, and T.

Elimination of dh between Eqs. (3.5) and (3.6) yields
dQ=-pdv-Ad.$+~~kdMk+TdY, (3.12)
k
where
Equation (3.12) generalizes Eq. (2.17) in the presence of a chemical

reaction. When a reaction occurs adiabatically in a rigid closed cell we put
d% = dv = dMk = 0, and Eq. (3.12) yields
Ad4 = TdY. (3.14)
This corresponds to De Donder’s formula (1936) and shows that A is the

affinity whereas dY is the entropy produced by the reaction and defined
here precisely in a new way as a collective concept.
We may write the affinity in a way formally identical to a standard
expression by putting
h”Tq
= c
k
t+(/:TerdEk + F,(O)), (3.15)
ds; + S,(O) . (3.16)
The integrals at the lower limit are evaluated by extrapolating gaseous

classic properties of Ek and Fk to absolute zero, with Ek(o) and Sk(O) con-
stants assumed to be characteristic of the substance and independent of
any particular chemical reactions. We may consider these assumptions as
basic axioms. Substitution of the values from Eqs. (3.15) and (3.16) into
Eq. (3.13) yields
A = -2 vkpk, (3.17)
k
where
pk = ‘AT dEi - T j-;‘dF; + &k(O)- TS,#). (3.18)

I 0
The constants &k(O)and ~(0) are determined experimentally from chemi-

: cal reactions. This provides a novel definition of the chemical potentials
pk.
The values of &k(O)and Sk(O) may also be defined and evaluated by
. introducing the axioms of quantum statistics and physical properties of
matter in the cryogenic range. However, in practice, in most cases this
leads to many difficulties, as shown by Fowler and Guggenheim (1952).
According to expression (3.13) and the value of Pk from Eq. (3.1 l), the
affinity
A = A(v, mk, T) (3.19)
is a function of v, mk, and T. With the value of mk from Eq. (3.9) it also
becomes a function of v, [, Mk, and T:
A = A(v, 5, Mk, T). (3.20)
We considered a single chemical reaction, but the results are readily
generalized to multiple simultaneous reactions by adding a set of chemical
supply cells Cchp, one for each reaction p. The masses produced by the
reactions are then
dmk = 2 vkp dtp. (3.21)

P
With an affinity A, for each reaction, Eq. (3.12) becomes
dQ=-_pdv-~A,d~p+~~kdMk+TdY. (3.22)
P k
IV. Homogeneous Mixtures and Reformulation of

the Gibbs-Duhem Theorem
The new restructured thermodynamics by thermobaric transfer pro-

vides a novel evaluation of the thermodynamic functions of a cell contain-
ing a homogeneous mixture of fluids. To evaluate the thermodynamic
functions for this case we apply Eqs. (3.5) and (3.6) by writing them in the
form
14 M. A. Biot
dQ = -p du + qeq dt + c Fk dmk + dh. (4.1)

k
d9 = Y,, dt + q Sk dmk + dhlT, (4.2)
where we have put
%‘q = i”T - c vks;‘, y,, = g-7

-2 vkf;4,
(4.3) ;
k eq k
s;q =
and dmk = vk d.f + dMk. For simplicity we assume a single reaction, but
the results may be readily generalized to multiple reactions. We start with
a cell of zero volume. Its volume u is then increased gradually while
maintaining constant the pressure p, the temperature T, and the concen-
trations Of the VariOUS substances. Hence, PkEk and Sk ah0 remain con-
stant. The heat that must be injected is
dh = 2 hiTdmk, (4.5)
k
where hkT is the heat of mixing at constant pressure and temperature.

Substitution of the value of dh from Eq. (4.5) into Eqs. (4.1) and (4.2)
yields
d%= -pdv+%_,d~+~%kdmk, (4.6)

k
dY = Y,, da$ + x Ykdmk, (4.7)

k
with
%k = Ek + hiT, Ypk = Sk + hkTIT. (4.8) *
During the transformation considered here, the value of hiT remains con-
stant, hence also the values of %k and yk as well as p and qeq. Thus
integration of Eqs. (4.6) and (4.7) yields
(4.9)
!?+ = Y,qt + c Ykmk. (4. IO)

k
These values are the energy and entropy of a homogeneous cell contain-
ing masses mk = vkt + Mk of each substance.
For a pure gas mixture hET = 0, and without chemical reactions (< = 0)
we obtain
(4.11)
k
showing that in this case the enthalpy of the mixture is the sum of the
enthalpies of the individual components, and likewise for the entropy
(Biot, 1982a).
We note again that for identical components the injection pressures pk
are the same, hence also the values of Sk, so that Gibbs’s paradox is
avoided.
To reformulate the Gibbs-Duhem theorem in the present context, we
differentiate Eqs. (4.9) and (4.10). Taking into account the values of these
differentials from Eqs. (4.6) and (4.7), we derive
2 mk d%k = V dp, (4.12)

k
T mk dYk = 0. (4.13)
Hence we may also write
u dp = ; mk(d%k - T dYk)
= c mk d(%k - TYk) + 2 m,@k dT. (4.14)

k k
Using Eqs. (2.18) and (4.8) we also write

Qk- TYk=+k, (4.15)
and Eq. (4.14) becomes
v dp = c mk d& + Yc dT, (4.16)

k
where
9’ = ck mkYk. (4.17)
We call Yc the convective entropy (Biot, 1982a), because it represents the

value of the entropy obtained by pure convection of the masses mk into
the cell without chemical reaction.
Equation (4.16) constitutes a reformulation of the Gibbs-Duhem the-
orem (Biot, 1982a). It is expressed here in terms of C$kand the convective
16 M. A. Biot
entropy Yc instead of the chemical potentials & and the entropy used in
the classical form. The present formulation avoids the basic difJiculties of
the classical treatment, which involves undetermined constants in pk,
and in the entropy.
Expressions (4.9) and (4.10) are readily generalized for multiple reac-
tions as
Q = -pv + 2 %_lbtp+ T %kmk, (4.18)

P
Y=~Y&tp+~Ykmk, (4.19)
P k
where
tip=l = h”Tqp- T ,@zqp, Ygq = (heqplTeqp) - T ,+Seqp. (4.20)
The superscript p refers to values for the particular reaction p.
V. Nontensorial Virtual Work Approach to Finite

Strain and Stress
The irreversible thermodynamics developed here is based on a princi-

ple of virtual dissipation. It constitutes a generalization of the traditional
method of virtual work in classical mechanics. These methods in the field
of solid mechanics have been used extensively by the author since 1934
(see Biot, 1965b); more recently, it was also emphasized by Washizu
(1975). Because of the currently fashionable trends, the power and sim-
plicity of this method have generally been overlooked. The method pro-
vides an approach to finite strain that is more general than that derived
from tensor definitions and contains the latter as a particular case. At the
same time, it is flexible and ideally suited to describe physical properties
of anisotropic media such as laminated and fiber composites of technolog-
ical importance.
We start from a homogeneous deformation defined by the linear trans-
formation of the initial coordinates xi of the material points to
Xi = (&j + a&j, (5.1)
where 6, is the unit matrix and aij are nine coefficients. Summation is
implied for repeated tensor indices. We may consider the transformation
to be equivalent to the following process (Biot, 1965b). First, a pure solid
rotation is described by three degrees of freedom and written as
Xf = R+j. (5.2)
It is then followed by another transformation,
Xi = (6ij + &ij)Xj’ 9
(5.3)
where the coefficients are not independent but satisfy three constraints,
so that they contain only six degrees of freedom. For example, we may
adopt the three constraints
Eij = &ji (5.4)
and call eij the six finite strain components defining the deformation (5.3).
In this case, the transformation (5.3) is such that principal directions of
strain remain invariant (Biot, 1965b).
Other choices may be made. For example, the transformation (5.3) may
be chosen so that the material on the xi axis remains on that axis while the
material in the xix{ plane remains in that plane. In this case, z21 = &31= ~32
= 0, and the six remaining coefficients define the strain components.
Other definitions may be chosen suitable to the particular physical proper-
ties of the material (Biot, 1973a).
For the method to be complete we must be able to express the six strain
components cti in terms of the nine coefficients aij. We have shown that
this can easily be done to any order by a systematic procedure (Biot,
1965b). For example, to the second order, for the choice (5.4), we have
1 1
Eij = eij + j (f?ipOJpj + &jpWpi) + - WipWjp, (5.5)
2
where
1
eij = i (ati + Uji), Oij = 5 (au - Uji). (5.6)
In two dimensions, for a choice of strain components such that &21= 0,

the three strain components to the second order are (Biot, 1973b)
1 1
~11 = err + -2 a&, ~22 = e22 - z a2lCb2 + a21), (5.7)
1
~12 = e12 + 7 a2lh22 - ad. (5.8)
The second-order approximation is sufficient in a vast group of problems.

We may also choose Green’s definition of finite strain,
1
Eij = eij + - U,iU,j. (5.9)
2
18 M. A. Biot
This has the advantage of providing the exact functional dependence in

terms of a~. However, in practice this leads to difficulties because the
strain components are nonlinearly related to extension ratios and intro-
duce spurious purely geometric nonlinearities in the physical description.
The virtual work method readily provides the definition of six stress
components rti associated with the strain. A cubic element of unit size
oriented along the coordinate axes is subject to the transformation (5.3)
by forces applied on the six faces. A virtual deformation 8~~ produces a
virtual work
6W = rij8sij (5.10)
with six coefficients rti defining stress components conjugate to Ed. The
quantity 6W is a physical invariant, but the factors rij and 8~ need not be
tensors. The indices i and j indicate a summation to all six independent
terms whatever their nature. If cij is expressed as a function of ati, we may
write
(5.11)
where
TV = rFp &Ida, (5.12)
are the nine components of the Piola tensor. We may also consider the
stress aij per unit area after deformation, referred to the initial axes xi. A
virtual transformation after deformation may be written in terms of final
coordinates Xi as
axi = Fj aa.. (5.13)
U’
where 6Zij are nine suitable coefficients to be determined. The unit cube
after deformation has become a parallelepiped, and the virtual work of the
forces aij in the virtual transformation (5.13) is
6W = Jq Gi,, (5.14)
where .I is the Jacobian of the transformation of Xi to Xi. TO derive &i,,

consider the virtual transformation in terms of initial coordinates by sub-
stituting into Eq. (5.13) the value of Xj from Eq. (5.1). We obtain
6zi = CjkXk 6a,, cjk = 8jk + ajk; (5.15)
and from Eq. (5.1),
6xi = xk 6a&. (5.16)
Since Eqs. (5.15) and (5.16) are identical transformations, we derive

6aik = cjk 6iiy. (5.17)
With this value, Eq. (5.11) becomes

6W = TikCjk SZ,. (5.18)
Comparing Eqs. (5.14) and (5.18) yields
‘+ij = (l/J)Tikcjke (5.19)
Since the virtual work (5.14) must vanish for a pure rotation %ig = -Ssl,,
the tensor oV is symmetric and satisfies the three relations
Uij = Uji 3 (5.20)
which also expresses the conditions that the moments due to the nine
stress components Tti vanish.
Substitution in Eq. (5.19) of the values (5.12) of Tij yields
Cri = (IIJ)~p~Cjk d$,/&li~. (5.21)
This expression is valid for all definitions of r@,,, and the summation is
extended to all six components of the stress 7PVas defined above (Biot,
1981).
For a continuous deformation field, we write Xi = xi + Ui(x/), where the
displacement Ui is a function of the initial coordinates Xi. The local infini-
tesimal transformation
dXi = (6ij + auilaxj) dxj (5.22)
is the same as Eq. (5.1) provided we write
dUildXj = au. (5.23)
With ag representing the nine displacement gradients, the foregoing defi-
nitions of stress and strain remain valid as local values for a nonhomoge-
neous deformation.
VI. Thermodynamic Functions of Open

Deformable Solids
The collective energy and entropy per unit initial volume of a deforma-
ble solid based on virtual work and nontensorial concepts were derived
earlier (Biot, 1981). The results may be readily extended to an open solid
with chemical reactions by proceeding as in Section III. We consider
initially a cube of unit size. A homogeneous deformation is defined by the
20 M. A. Biot
six strain components aij under the corresponding stresses TV. The solid
contains substances in solution. The increment of mass of each substance
is
dmk = Yk ds$ + dMk, (6.1)
where Vkde is due to a chemical reaction and dMk is the mass acquired
from an external source. For simplicity we assume a single reaction,
because results are readily generalized for multiple reactions. An infini-
tesimal change is defined by dq, dt, dMk and the temperature change dT.
The corresponding changes in energy and entropy are obtained by ther-
mobaric transfer and their values are derived from Eqs. (3.5) and (3.6) by
replacing -p du by rij 8~~. This yields
d% = rij dq + c vk df IkITe,dak + h”T de + T Ek dMk + dh, (6.2)

k I
Fl”T”
dF;+ydt+cFkdMk+$.
eq k
The value from Eq. (3.7) must also be changed to
dh = hLT dsij + c h$ (,+ dt + dMk) + C,, dT. (6.4)

k
The terms hiT dq represent the heat that must be injected when applying
a deformation dq under the constraint dmk = dT = 0. Similarly, h$ is the
heat of mixing, where hfT dMk is the heat injected when adding a mass
dMk under the constraint dsti = dT = 0. The heat capacity C,, is for de, =
dmk = 0. We may show that h$,T and h$ may be determined without
calorimetric measurements when we know the equations of state
rij = r&&u, mk, T), (6.5)
Pk = Pk(&pv, mkv 77, (6.6)
i.e., when we know the stress rij and the injection pressures pk as func-
tions of the strain F~, the masses mk of each substance added by chemical
reaction and transport, and the temperature T. From the conditions that
Eqs. (6.2) and (6.3) are exact differentials, proceeding exactly as before in
deriving Eqs. (2.8) and (2.12), we obtain
h$r = - T(c%,lc3T),, , (6.7)
hfr = - TIP,&PklaT)m, (6.8)
where the subscripts indicate that the partial derivatives are for aij and mk
constant.
The coefficients of the differentials in expressions (6.2) and (6.3) of d’%

and dY are now known as functions of Ed, mk = Vkt + Mk, and T.
Integration along any arbitrary path with initial conditions % = Y = 0
yields the values
011= Q&i, 5, Mk, T), (6.9)
9 = Y(E~, 5, Mk, T) (6.10)
as functions of the state variables Ed, 5, Mk, and T.
Replacing -p dv by rij dEij, Eq. (3.22) is generalized to
d~=qd~G-~Apd&,+~~~dMk+TdY. (6.11)
P P
VII. The Fluence Concept
In a fixed coordinate system xi, a rate of mass flow or mass flux of a

given substance is represented by a Cartesian vector &lf. The component
tif is the mass flux across a unit area normal to xi : conservation of mass is
expressed by
Mk = -a~~ldXi. (7.1)
The total mass entering a domain R is then
(7.2)
where n; is the unit normal to the boundary A. Since kf represents a time

derivative, we may integrate the relations with respect to the time with
zero initial values. Hence
Mk = -aMflax;, (7.3)
where Mf is called the mass Jluence. It constitutes a Cartesian vector.
In deformable systems, a more general definition of the fluence has
been found extremely useful (Biot, 1977b). We consider an area initially
equal to unity and normal to the xi axis. This area is then transported and
deformed by the coordinate transformation
Xj = Xi(xi, t). (7.4)
We now call A!lf the mass flux across this transformed area. Obviously the
mass flowing into a transformed element initially of unit volume is given
22 M. A. Biot
by (7.1), where i6lf represents the more general mass flux. Hence Eqs.
(7.1)-(7.3) retain exactly the same form with this generalized definition of
the mass flux &l: and mass fluence M$. In this case, however, Zt%fis not a
Cartesian vector. We denote by &flk the corresponding Cartesian compo-
nents, i.e., tiik represents the mass flux per unit area in the space Xi. An
initial surface A is transformed into A’ by transformation (7.4). The mass
flux through it may be expressed in two different forms as
IA n;r~ni dA =
I A’
n;lj”C~inr dA’, (7.5)
where the second integral is expressed in terms of the coordinates Xi and

C; is the cofactor of aXi/aFj in the Jacobian
J’ = detIdxilZj/. (7.6)
Obviously, in the second integral of Eq. (7.5) the integrand k;Cji is the
Cartesian component &fIk of the mass flux in the space Xi. Hence
nj,!” = ~Q!c!. (7.7)
J JI’
This relation may be given another useful form by considering the linear
transformation of dfi into dxi and solving for dfi. We find
axilaxj = CjiIJ’. (7.8)
Hence
n;r,!” = Jln;rjk afi/ax.
(7.9)
J’
Multiplying this equation by axjldfi, we obtain (Biot, 1977b)

n;lj” = JM Ik dXj/d~i, (7.10)
where
J = l/J’ = detlaFi/axjI. (7.11)
These results show the mass flux, as defined above for a deformable
coordinate system, to be a relative contravariant vector (Sokolnikoff,
1951, see pp. 58, 72). We shall refer to &ff and Mf, respectively, as the
contravariant mass flux and the contravariant mass jluence.
Similarly, consider the local Cartesian heat flux BI at Xi and a contrava-
riant heat flux Hi defined as the rate of heat flow across a transformed
surface initially equal to unity and initially normal to the Xi axis. They
satisfy the same relations as Eqs. (7.9) and (7.10), namely
The time integral of I!Ziis the contravariant heatfluence Hi. We may write
a conservation condition analogous to Eq. (7.3) as
h = -aHiIaXi, (7.13)
where h is the heat acquired by a deformed element of unit initial volume.
The entropy flux across an area is defined as (Biot, 1977b)
(7.14)
Its time integral is the entropy fluence Si. They are either Cartesian or
contravariant components.
The fluence fields Mf are state variables, because they determine Mk by
Eq. (7.3). However, the fluence fields Hi and Si are not state variables.
Other fluence fields that are state variables may be introduced by put-
ting
9 = -aStldxi, % = -aSd:laxi, (7.15)
where ST is called a pseudo-entropy fluence and ST a pseudo-energy
jluence. Using such concepts, the state of a system may be partially
described by fluence vectors as further clarified below in several applica-
tions.
VIII. The Nature of Entropy Production

and Its Evaluation
In order to formulate the principle of virtual dissipation in the next

section, we must evaluate the entropy produced by an irreversible pro-
cess. Entropy production is due to various phenomena, in particular
thermomolecular diffusion, viscosity, and chemical reactions. We shall
first consider the entropy produced by diffusion, using a derivation that
brings out its physical significance (Biot, 1982b). Consider a medium that
may or may not be deformable. The entropy increase per unit initial
volume as given by Eq. (2.3) is written in variational form
69 = $ Sk 6Mk + 6hlT, (8.1)
where 6h is an amount of heat added reversibly, so as to produce the

same change of state as the actual irreversible process. This amount of
heat may be written
6h = 6hP - d 6HJaxi, (8.2)
24 M. A. Biot
where -d 6HilaXi is the amount of heat added by the actual irreversible

heat flow. We assume that this amount is added reversibly along with an
additional heat 6hP so that the total generates the actual irreversible
change ofstate. The heat 6hP defines the heat produced by the irreversibi-
lity and corresponds to Clausius’ uncompensated heat. From Eq. (7.3) we
also write
6Mk = -a 6MFldxi. (8.3)
In these equations, Hi and MC are fluence concepts either Cartesian or
contravariant, as defined in Section VII, and are valid in a deformable
system in terms of initial coordinates xi. We substitute the values (8.2)
and (8.3) into Eq. (8.1). The result may be written in the form
(8.4)
where
SSi = 5 Sk 8M$ + 6HiIT (8.5)

is the variation of entropy fluence as defined by Eq. (7.14), and
Equation (8.4), when integrated over a volume R with ni 6Hi = ni 6Mi

= ni 6Si = 0 at the boundary, yields
I, SY dfl = I,, Gs*~~ dfk. (8.7)
Hence 8s*TMis the virtual entropy produced per unit initial volume. It has
three terms. The first is due to the masses convected 6Mf, the second is
due to the heat 6hP produced by the irreversibility, and the third is due to
the heat flow 6Hi across a temperature gradient.
It remains to evaluate the heat produced 6hP. This is obtained by con-
sidering energy conservation. Using the value (8.2) of ah, the variation of
energy (2.1) per unit initial volume with 6v = 0 is
&?L=~EkBMk+6hP-&6Hi. (8.8)
I
Substitution of Eq. (8.3) for 6Mk and integration over a domain fl yields
+ T 26M) + 6hP do. (8.9)

1
On the other hand, conservation of energy is expressed by
(8.10)
This expression is based on the significance of (& Ek 6Mf + GHi)ni as

representing the variation of diffusive energy flowing across a surface,
and while %i is the body force per unit mass and a: the acceleration of
each substance. Note that Eq. (8.10) is actually based on a generalized
d’Alembert principle, where -a; plays the role of a body force per unit
mass. Comparing the values (8.9) and (&lo), which are valid for an arbi-
trary domain a, we derive for the heat produced
ShP = C(%; - a: - ark/axi) 6Mf. (8.11)

k
This provides a physical interpretation of 6hP as the heat produced by the

irreversible mechanical work. When it is substituted in the variational
relation (8.6), taking into account relation (2.4), we obtain
T Gs*~~ = T Xi” 6Mf + Xi’ 6H;, (8.12)
where
(8.13)
are dissipative disequilibrium forces. They are functions of the mass

fluxes i$f,” and the heat flux fii. We write
xi” = %/(tij, Eij), XT = $RT(n;rjl, fij). (8.14)
It should be understood that these relations also depend on the local state
variables, although this is not formulated explicitly. They embody the
irreversible kinetics of the system and may be obtained either experimen-
tally or theoretically from molecular kinetics. With the values (8.14),
26 M. A. Biot
equations (8.12) may be written
T 8s*TM = T P&f’SMF + %T SHi. (8.15)
In many problems it is convenient to use the entropy flux Si instead of the

heat flux fii. Solving Eq. (7.14) for I;ri and Eq. (8.5) for 6Hi, and substitut-
ing these values into Eq. (8.15), we obtain
T ~~~~~~= F G!il~”
6M,k + 3:’ 6Si, (8.16)
where
aFk(h/, Sj) = GRf- TFk%T, (8.17)
~~T(n;rj’, Sj) = Twit. (8: 18)
The present derivation of entropy production is fully general and does not
assume linearity or any dependence on Onsager’s (1930, 1931) principle.
For example, we may include nonlinear diffusion of a non-Newtonian
fluid through a porous medium.
In the case of a linear dependence of ‘%T on the fluxes and in the
absence of a temperature gradient we obtain from (8.14) the linear rela-
tions
%T(n;r,l, Ejj)= 0, (8.19)
where Hi is the coupled heat flux due solely to the mass flux. If we solve
these equations for fii, the coefficients of fi,” constitute what is generally
called the heat of transport.
Consider now the entropy produced in a primary cell by a chemical
reaction. This is the entropy increase of a cell for du = dMk = d% = 0. In
this case Eq. (3.22) is applicable and is written in variational form as
T i3~*~~= T A, S&, , (8.20)
where 8s*Chis the entropy produced by the reactions S&,. The values of
the affinities are functions Of u, mk, T,
A, = A,(u, mk, T), (8.21)
where mk = x, V&p + Mk is the total mass of each substance added to

the cell by reaction and transport. An expression similar to Eq. (8.14) is
obtained by introducing the kinetics of the various reactions. The rates of
reaction [, are functions of u, mk, and T,
&I = fp(U, mk, T). (8.22)

Using Eqs. (8.21) and (8.22) it is easily shown that A, is of the form
A, = aP(lP, u, mk, T). (8.23)

The affinities A, are similar to the disequilibrium forces X/ and Xir and in
analogy with (8.14); they are expressed by SRPin terms of the irreversible
kinetics. Equation (8.20) becomes
T 8s*Ch = c CR,,St,. (8.24)

P
The total entropy production 6s” due to simultaneous diffusion and chem-
ical reactions, obtained by adding Eqs. (8.16) and (8.24), is given by
(8.25)
This expression is applicable to a deformable cell with fluence concepts

for M! and Si. It yields the virtual entropy production per unit initial
volume. For an actual irreversible transformation we replace the varia-
tions by time derivatives and Eq. (8.25) becomes
(8.26)
This expression represents the rate of dissipation per unit initial volume
and is positive. Note that Eq. (8.26) is a consequence of relation (8.2.5),
which is more general and concerns a virtual change, whereas the actual
change in Eq. (8.26) is a particular case.
When in addition to being linear the relations between the rates and the
dissipative forces satisfy Onsager’s reciprocity relations (1930, 193 I),
they may be written
C!J$= aCP/a&_ 3;” = $JrM/an;l” I 9 %TT = d91TM18Si, (8.27)
where ‘Sch is a quadratic form in [, and GJTMa quadratic form in i# and
Si, with coefficients functions of the local state.
Applying Euler’s theorem, we write Eq. (8.26) as
+ 2 C C~Ai"Sj + ThiiSiSj, (8.28)

k
where G!Z= ‘GBch + STM is a combined dissipation function for chemical

reactions and thermomolecular diffusion.
Other types of entropy production, such as that due to viscous effects,
will be discussed below in the applications.
28 M. A. Biot
A fundamental entropy balance equation may be expressed in terms of

entropy production by adding 6SSchin Eq. (8.4). The total entropy varia-
tion is then
69 = - & 6Si + as*, 6s” = &*rM + &s*ch. (8.29)

I
We replace the variations by time derivatives and write
9 = -aSjlaXi + S*. (8.30)
If we integrate this equation with respect to time, assuming zero initial
values, we obtain the entropy balance equation
Y = -dSilaXi + S”. (8.31)
It should be understood that in problems where we choose a nonzero
initial Y, we must simply subtract this value of Y in Eq. (8.31).
The entropy production has been expressed here in terms of 4,) M/, and
Si. While cP and Mf are state variables, the entropy fluence Si is not. In
order to avoid this difficulty, it was shown (Biot, 1981, 1982b) that we may
introduce the pseudofluence concepts (7.15) by proceeding as follows. We
note that because Eq. (8.10) is valid for an arbitrary domain, we have
Using Eqs. (8.13) and (8.14) and the definition (7.15) of !+:, this yields
with
From Eqs. (8.14), we may obtain fii in terms of X,! and h,“. Substituting
these values of rii into CR”determines the latter in terms of local state
variables, their gradients and i@. Thus in the quantity
&k = ~~“c~;) (8.35)
the only fluence fields are the state variables Mj" . With gi(Xj, xj’) repre-
senting a fluence field due to a unit concentrated source at x,?, Eq. (8.33)
may be written
6Hi = 8%: - J$ Ek SMF + J,- gi(X,, ~7) TJ;” 6Mj da+, (8.36)
where the integral is extended to the space a+ of x,?. Considering the

value (8.5) of 6Si and (2.18) of I#Q,this may be written
T 6Si = SST - T 4k 6M/ + I,,+ gi(x,, x:) T Ak SM: dlR+. (8.37)
With time derivatives, this becomes
(8.38)
Thus 6Si and Si are now expressed in terms of the fluence vectors S,? and
M!, which are state variables.
An important simplification occurs in problems where the effects of
inertia and body forces are negligible, and more generally ifjjb is negligi-
ble. In that case, withAk = 0, Eq. (8.33) yields
(8.39)
where
S?PF= SSFi = C Ek 6M/ + 6Hi (8.40)

k
or
8i = c Ekn;ri”+ Eii. (8.41)

k
Under these conditions 9: coincides with the actual physical energy&x

Si due to molecular and heat diffusion.
Relations (8.37) and (8.38) become
T 6Si = 69i - C ok 6Mik, (8.42)

k
T,$, = si - c $,n;ri”. (8.43)

k
In many problems where fi” remains small these simplified results are
applicable.
Another possibility is to introduce the pseudo-entropy fluence S: de-
fined by Eq. (7.15). Equation (8.29) becomes
(8.44)
Hence
30 M. A. Biot
6Sj = SST + I,,+gi(X/, XT)6S* dR’ (8.45)
and
Si = ST + I,- gi(X,, xT)S* dQ+.

In many problems we may neglect the entropy s*TMproduced by thermo-
molecular diffusion. In that case, we write Eqs. (8.45) and (8.46) as
(8.47)
Thus 6Si and Si are now expressed in terms of state variables ST and ep.
Another important simplification also applies for problems where the
entropy produced does not contribute significantly to the total value of the
entropy Y. In that case we may write
Y = -dSf/dXi, (8.49)
and Si itself becomes a state variable. This is the case for quasi-reversible
processes and in linear thermodynamics.
IX. The Principle of Virtual Dissipation
Consider again a hypersystem constituted by a collection of primary

cells (a primary system), a collection of supply cells including chemical
cells, and a thermal well. In the process of thermobaric transfer as defined
above, the thermal well did not play any role, because the source of the
heat injected in the primary cells and the transferred masses is not speci-
fied. We now introduce a modi$ed thermobaric transfer, in which the heat
injected is provided by reversible heat pumps extracting heat from the
thermal well. The mechanical work necessary to inject an amount of heat
dh into an element at the temperature T is
dw = 8/T dh, (9.1)

where
O=T-TO (9.2)
and TOis the thermal well temperature. A thermobaric reversible transfor-
mation of the hypersystem is now accomplished entirely through mechan-
ical work performed on the system, without any exchange of matter or

heat with the environment. For such a transformation, the increase of
energy of the hypersystem is equal to the mechanical work dW performed
on the system. This may be written
dU - To dS = dW, (9.3)
where dU is the energy of the primary system defined above as a collec-

tive concept that includes the energy of the supply cells. The term -T,-, !S
is the energy acquired by the thermal well; its increase of entropy -dS,
because of reversibility, must be equal and opposite in sign to the entropy
increase dS of the primary system. We remember that dS is defined as a
collective concept that includes the supply cells. We may write Eq. (9.3)
as
dV = dW, (9.4)
where
V= U- T,,S (9.5)
is a thermodynamic function shown by the author (Biot, 1954, 1955) to be
the key concept that plays a fundamental role in irreversible thermody-
namics. It was initially referred to as the “generalized free energy” be-
cause it coincides with the Helmholtz definition for the particular case of
isothermal transformations. The term “exergy” was later introduced by
others to designate a formally similar expression without defining more
precisely the significance of U and S in the general case of open systems
in terms of collective concepts, as we have done here. However, we shall
keep the name exergy for the more general concept represented by Eq.
(9.5).
Further physical insight is provided by considering the exergy of a
single primary cell:
V = % - T&f. (9.6)
In order to simplify the writing, and without loss of generality, we may
consider a rigid cell without chemical reactions. Using the values (2.1)
and (2.3), with dv = 0, we may write the differential of the exergy as (Biot,
1976a, 1977a)
dY- = c $I~dMk + BIT dh, (9.7)
where
(9.8)
.?k = (9.9)
32 M. A. Biot
In a reversible transformation, d”lr is the increase of energy of the hyper-

system, and (BIT) dh is the work accomplished by the heat pump in order
to inject a quantity of heat dh into the primary cell. Hence the remaining
term & dMk is the work required to bring the mass to equilibrium with the
primary cell and inject it into the cell. We have called $k the thermobaric
potential. In terms of the convective potential (2.18), we write
$k = & + @?k. (9.10)
Note that $k is defined purely in terms of mechanical work, which is not

the case for +k, which involves the concept of entropy.
Until now we have been dealing with reversible transformations. How-
ever, the value of V is determined by the state variables, whether they
follow a reversible or irreversible evolution. We therefore consider varia-
tions of the state variables for an irreversible change of state. These
variations are arbitrary except for one condition. They must satisfy the
constraint that the flow of heat and matter between cells is continuous.
For a continuous system considered as a collection of infinitesimal cells,
this means that the heat and mass fluence field must satisfy the constraint
of continuity. At the boundary of the primary system, heat and matter
may be injected into it by a modified thermobaric transfer within the
hypersystem. Hence the variation of the hypersystem occurs without ex-
change of matter and heat with the environment. As a consequence,
conservation of energy for the hypersystem is expressed by
6U + TO6&w = 6W, (9.11)
where 6Srw is the variation of entropy of the thermal well and TOi3STw its
variation of energy. This relation differs from Eq. (9.3) in that the trans-
formation is irreversible and we may not put 6s~~ = -6s. With the value
(9.5) of V we may write Eq. (9.11) as
6V+ T&Y” = 6W, (9.12)
where
6s” = &srw + 6s (9.13)
is the variation of entropy produced in the hypersystem. In terms of the

local entropy produced as*, as evaluated in Section VIII, it is expressed
by
TO6s” = T,, I, as* da. (9.14)
An important step results from a generalized interpretation of the value

of 6W, the virtual work of the external forces acting on the hypersystem.
It represents the work 6WM of the mechanical forces acting on the pri-
mary system plus the work 6 WTH accomplished in the thermobaric trans-
fer of heat and mass injected at the boundary. In addition, we generalize
d’Alembert’s principle by including in the external work the virtual work
-CiZi 6qi of the reversed inertial forces due to mass accelerations in the
primary system at any particular instant of the evolution. The generaliza-
tion involved here implies the validity of d’Alembert’s principle where U
is a thermodynamic energy involving heat and not simply a mechanical
potential as in classical mechanics. Accordingly, we may write
(9.15)
where
awrn = - I,ic $k 6MI + BIT 6Hi)ni dA (9.16)
represents the virtual work due to the thermobaric transfer of heat

6Hini and masses 6Mfni at the boundary. Note that for a deforming
boundary, Mf and Hi are the contravariant mass and heat fluences as
defined in Section VII, whereas the surface integral is evaluated at the
initial boundary A. Note also that Eq. (9.16) may be expressed in terms of
the entropy fluence Si at the boundary by substituting into Eq. (9.16) the
value of 6Hi extracted from Eq. (8.5). This yields
6wTu = - J’,(C +k 6Mt + 0 6Si)n; dA. (9.17)
k
We have called GWTH the virtual work of thermodynamic forces at the
boundary.
With the value (9.15) of 6W substituted in Eq. (9.12), we obtain
C Zi 6qi + 6V + To 6s” = 6WM + GWTH. (9.18)
This constitutes the generalized principle of virtual dissipation (Biot,

1955, 1975, 1976b, 1982b) validfor arbitrary variations, provided continu-
ity of heat and mass flux is preserved.
If the system is in a potential field such as a gravity field, we denote by
G the potential energy in this field. By including -6G in the mechanical
work of external forces, we may write Eq. (9.18) as
C Zi 69, + 69 + To 6S* = 6WM + 6WoTH, (9.19)
where 9 = V + G (Biot, 1975, 1976b) is called the mixed collective poten-

tial, while 6WM denotes the work of forces other than gravity acting
34 M. A. Biot
directly on the primary system. The term
6WGTH = -
JA(T pk 6Mf + 8 SSi)ni dA (9.20)
now includes the work against the gravity field in the boundary thermo-
baric transfer. We have put
‘Pk = $k + %, (9.21)
where % is the gravity potential field per unit mass chosen so that the
supply cells are on the surface % = 0. We have called $0kthe mixed
convective potential.
With time derivatives instead of variations, Eq. (9.14) yields the total
rate of dissipation as
T,$* = To I, S* da (9.22)
It is inteRSting to compare T,+* and TS*. For eXaII@e, if aSk/dXi =

aT/axi = 0, writing Eq. (8.6) with time derivatives yields
7’s* = jlr (9.23)
where itp At is the heat produced by the irreversibility in the time interval
At. If we have a thermal well at a lower temperature T,, this heat is not
entirely lost, because we may recover the mechanical work
oT-TT hPAt (9.24)
through a heat engine. The actual loss of useful energy is therefore
(T,/T)hP At = T,i* At. (9.25)
For this reason we have called TP the intrinsic dissipation, and ToS* the
relative dissipation (Biot, 1975, 1976b).
In the modified thermobaric transfer heat is pumped from the thermal
well and injected into the various elements and cells of the hypersystem.
The process involves only mechanical work (given by Eq. 9.1) and pro-
vides purely mechanical definitions of qk as well as the exergy and 6 WTH.
The pump may use matter, subject to a Carnot cycle. However, it is of
interest to point out (Biot, 1976a) that use of the Carnot cycle may be
avoided by using pure heat as blackbody radiation extracted from the
thermal well and compressed adiabatically to the required temperature of
injection.
X. General Lagrangian Equations
Consider a system going through an irreversible evolution. We shall

assume that its state may be described at every instant by a finite set of
parameters qi, unknown functions of time called generalized coordinates.
In particular, this may be accomplished by the use of fluence fields as
state variables either exactly or approximately. Such fluence fields, along
with material displacements and reaction coordinates, are then consid-
ered to be given functions of the initial coordinates and generalized coor-
dinates qi to be determined. Arbitrary variations 6qi generate field varia-
tions that may be called holonomic and may satisfy identically the
condition of field continuity. We may write
(10.1)
(10.2)
SWM+ SW” = 2 Qi Sqi. (10.3)
Substitution of these values into Eq. (9.19), which expresses the principle
of virtual dissipation, considering that 6qi is arbitrary, yields
Zi + 69’laqi + Ri = Qi, (10.4)
where Zi are generalized inertia forces, Ri are generalized dissipative
forces, and Qi are driving forces of a mixed mechanical and thermody-
namic nature due to environmental conditions (Biot, 1975, 1976b). These
equations are the general Lagrangian equations of evolution of irrevers-
ible thermodynamics. For systems that are quasi-reversible such that
local states do not deviate much from a local equilibrium satisfying Onsa-
ger’s (1930, 193 1) principle, the total virtual dissipation may be expressed
in terms of a dissipation function. For example, when the entropy produc-
tion is due to thermomolecular diffusion and chemical reactions near local
equilibrium, Eqs. (8.25), (8.27), and (9.14) yield
(10.5)
If c,,, Mf , and Si are expressed as functions of qi, this may be written

36 M. A. Biot
where D is the total dissipation function
D=/$%!Cl=;.S* (10.7)
and 9 is given by Eq. (8.28). Expression (10.7) is a quadratic form in 4;

with coefficients depending on 9;. The Lagrangian equations (10.4) for
quasi-reversible evolution become
Zi + d?PPldqi+ aD/aGi = Qie (10.8)
This form of the Lagrangian equations was initially derived in the linear
context (Biot, 1954, 1955, 1956a).
In many problems it is possible to express the generalized inertial
forces in the classical form
(10.9)
where 3 denotes the kinetic energy expressed as a function of qi and Gi. In

this case the variational principle may be written in the Hamiltonian form
1,[6(Y + 9’) + T (R; - Qi)6qi]dt = 0.

For a quasi-reversible evolution we may write Eq. (10.8) as
aD/aGi = Xi, (10.11)
with Xi = Q; - Zi - dPP/aqi. This equation shows that in any given state,
when we consider all possible generalized velocity fields 4; satisfying the
constraint
C XiCji = const., (10.12)
I
the actual velocity of evolution minimizes the dissipation function D. In
view of relation (10.7), the rate of total entropy production is also a
minimum (Biot, 1955, 1976b). As a simple physical illustration, consider a
fluid under gravity seeping through a porous medium. At a given instant,
of all possible velocity fields with the same rate of descent of the center of
mass, the actual one minimizes the dissipation.
XI. Dynamics of Viscous Fluid Mixtures with

Reaction-Diffusion and Radiation Pressure
Field and Lagrangian equations of viscous fluid mixtures with chemical

reactions, thermomolecular diffusion, and radiation pressure have been
derived directly from the principle of virtual dissipation (Biot, 1979,

1982a). The results provide new and powerful methods in a large variety
of technological problems as well as in stellar dynamics, in particular in
the analysis of stability and oscillations of self-gravitating bodies.
The coordinate system xi is Cartesian and fixed. The fluid flow through
it is described by a Cartesian mass flux klf of each substance in the
mixture. In addition, we consider a Cartesian entropy flux Si. The fluid is
in a gravity field. Because the mixture is homogeneous we may apply the
concepts of Section XIII, where a cell of unit volume is created by ther-
mobaric transfer from an initial state of zero volume at constant tempera-
ture T and pressure p. The energy Q and entropy Y per unit volume are
then obtained from Eqs. (4.9) and (4.10) by putting v = 1. Actually, to
conform with our general procedure, we add suitable constants so that
Q = Y = 0 at the initial time t = 0. The values mk represent the masses of
each substance per unit volume in the mixture. They may be written as
(11.1)
where mOkare the initial masses at t = 0. We also put tp = 0 at t = 0 and

assume the chemical reactions to be initiated at that instant, whereas Mk
are the masses per unit volume added by convection for t > 0. As before,
we may write the mass conservation equation (7.3) as
Mk = -aMhI&.
I I? (11.2)
where MI is the Cartesian mass fluence.
In the present case, the entropy production is due to three causes: the
chemical reactions, the thermomolecular diffusion, and the viscosity. The
first two have been evaluated and expressed by Eq. (8.25). In order to
evaluate the entropy produced by the viscosity we consider the viscous
stresses due to strain rates. The strain rate of a mixture is expressed in
terms of the barycentric velocity vi defined by
(11.3)
where p is the density of the mixture. The viscous stresses are then
written
oij = ql(uij + Vji) + rj2 aijV//, Vij = dVi/aXj, (11.4)
where r), and r/2 are viscosity coefficients depending on the local state of
the mixture. These stresses may be written in the form
‘+ij = lYSP/avij, (11.5)
where
38 M. A. Biot
(11.6)
the viscous dissipation function, is a quadratic form in vij.

When varying the mass fluence 6Mf, we produce a virtual displacement
derived from Eq. (11.3) as
(11.7)
and a virtual work

6hP = cu d8uildxj. (11.8)
The variation 6hP is the virtual heat produced and is analogous to the
value in Eq. (8.2), namely, the heat to be added reversibly in order to
obtain the same change of state as was due to the irreversible process.
Hence the virtual entropy Ss*” produced by the viscous stresses is given
by
T 6s”” = 6hP = r~u&kilaxj. (11.9)
We shall assume that the mass and heat flux obey locally linear laws
and Onsager’s principle. In that case the virtual entropy production due to
thermomolecular diffusion as derived from Eqs. (8.26) and (8.27) is given
by
T aS*TM = (11.10)
with a dissipation function BTM to be specified below. For the dissipation

due to the chemical reactions we retain the general form of Eq. (8.24).
The total virtual entropy production is then obtained by adding the
values of Eqs. (8.24), (1 I .9), and (11.10). We write
asTM
T &Y* = C CR,,Sgp + ~0 Z& 6Ui + T $ SM,+ + _ 6Si. (11.11)
P J I as;
For a domain IR under gravity the mixed collective potential is
9 = I,(V + p%) dfi (11.12)
where % is the gravity potential field and Yf the exergy per unit volume.
The fixed-coordinate system defines cells whose volume is not varied.
Hence from Eqs. (3.22) and (9.6) with Sv = 0, we obtain
i!% = 6% - T,, 69 = - c A, S(, + c & 6Mk + 8 6Y (11.13)

P k
Also Sp = & 6Mk, and we may write
@j”= \@
P
A, atp+ T (dk 6Mk + 8 SLY) dR, (11.14)
where C&is the mixed convective potential (9.21).

The virtual work of the inertial forces is evaluated by assuming that all
accelerations are the same and equal to the barycentric value
LZi= dVi/dt + Vj dVi/dXj (11.15)
as determined by vi. This amounts to neglecting the inertial forces due to
the relative velocities of diffusion. We write
C Ii?@,= I,,aiC 6MfdlR. (11.16)

I k
We now apply the principle of virtual dissipation (9.19), considering

variations inside the domain a. In this case 6 WM = 6 WTH = 0. With the
values (11.14) and (11.16) and the value (8.29) for 6Y, the principle of
virtual dissipation becomes
I,(ai~~M~-CAp65,+C~k6Mk-B~6Si+T6s*)d~=0. (11.17)
P I
We then introduce the value (11.11) of T 6s” and the value (11.2) for Mk.
After integration by parts we equate to zero the coefficients of the arbi-
trary variations and obtain
(11.18)
aTlax.I + XZJTM/&$,
I = 0> (11.19)
-A, + CRp= 0. (11.20)
These are the field dynamical equations of the viscous fluid mixture with
reaction-diffusion. They bring out a fundamental coupling between diffu-
sion and the viscous stress gradient aa,laxj. They contain the unknowns
tPb, Mf, and Si. According to Eqs. (8.31) and (11.2), they determine the
state of the system as a function of time, if we may neglect the contribu-
tion of the entropy produced s* to the value Y of the entropy. However, if
this is not the case, we may determine s* by adding the auxiliary equation
Ti* = c aPtp + 29” + 2QTM, (11.21)

P
which expresses the rate of dissipation and is obtained from Eq. (11.11)
by replacing the variations by time derivatives.
40 M. A. Biot
Use of s* and the auxiliary equation (11.21) may be avoided by using

pseudofluence vectors such as SP”+or YP+discussed in Section VIII as state
variables. For example, in the present case, if we use ST, Eq. (8.33) must
be completed to take into account an additional heat source Uij dauilaxj
corresponding to the energy of viscous dissipation. Hence expressions
(8.37) and (8.38) for T 6s and TSi remain valid, provided we add terms oij
asuilaxj and gij aUi/axj in the volume integrals. We obtain
+
f
n+ gi(X[,
(
Xj+) T jjk SMjk+ Uij $) J
Lift+ (11.22)
+ (11.23)
With these values the field equations become integro-differential equa-

tions, except if the integral terms in Eqs. (11.22) and (11.23) are negligible,
which is usually the case.
An energy Jlux theorem was derived (Biot, 1979, 1982a) from the field
equations (11.18), (11.19), and (11.20). This may be shown by adding
these equations after multiplying the first set by k,“, the second set by Si,
and the third by 4,. We obtain
(11.24)
where
(11.24a)
k
represents the total energy flux. Note that in this expression the quantity
& n;ri”E, + tii is the diffusive energy flux (8.41).
We have not yet mentioned the important fact that the dissipation func-
tion QTM for thermomolecular diffusion must be invariant under transla-
tion. It was shown (Biot, 1982a) that this condition is satisfied if we put
gTM = ; ;C’ku;kujk+ ; ch cku,f$;S, (11.25)
where
u!k
I
= 2l;i!/m,
I
- &k/m
I k, UkS
I = tiftmk - SiIY’y (11.26)
and Yc is equal to the value (4.17). The coefficients Cik and Ck are func-
tions of the local state, and the values of Ck are chosen such that the
coefficient of $ is T/2k where k is the local thermal conductivity of the
mixture. The dissipation function (11.25) satisfies the identity
ck (11.27)
Using this relation, we may verify that the field equations (11.18) and
(11.19) satisfy the total momentum balance (Biot, 1982a) by multiplying
Eq. (11.18) by mk and Eq. (11.19) by Yc. Adding the results and taking
into account the identity (11.27) and the modified Gibbs-Duhem theorem
(4.16) for v = 1, we obtain
pai - lYU,lCYXj + dpldXi + p i3%ldXi = 0. (11.28)
This result expresses the total momentum balance.

As shown in a detailed discussion (Biot, 1982a), the effect of radiation
pressure may be taken into account in the present analysis by including it
in the total pressure p and the injection pressures pk as generalized equa-
tions of state. It was pointed out that kinetic and radiation pressures may
not be additive in dense mixtures because the radiation group velocity in
dense matter is smaller than the velocity of light (Brillouin, 1930).
The foregoing results neglect the accelerations due to diffusion. Equa-
tions that avoid this simplification and introduce partial viscous stresses
were also derived (Biot, 1979) in the absence of chemical reactions. The
latter condition is required if relative accelerations due to diffusion are to
retain any physical meaning, because chemical reactions imply a form of
impact and coalescence between molecular flows that are not taken into
account in the present theory.
Lagrangian equations of evolution are obtained directly from the princi-
ple of virtual dissipation. The state of the system may be described by 5,
and the fluence fields M/ and 9: expressed in terms of generalized coor-
dinates qi as
5p = 5&l? q2 *** X/h (11.29)
Mi” = Mik(qi, q2, **. Xl), (11.30)

?J: = a:(qi, q2, ..f Xl). (11.31)
We derive the Lagrangian equations using the general procedure of Sec-

tion X. Instead of ST we may also use the fluence fields 9i, ST, or Si
according to the particular approximations involved (see Section VII).
The virtual work of the inertial forces is
42 M. A. Biot
Jnaj2 k
SMjkdfl = ~Z;6q;. (11.32)
Hence the generalized inertial force is
(11.33)
The virtual dissipation is
T,, 6s” = I, To 6s” da = R, Sqi, (11.34)
where as* is given by Eq. (11.11). Hence

Ri = RFh + Rv + RT”. (11.35)
The term
(11.36)
is the chemical dissipative force. It can be shown that the viscous dissipa-
tive force is
(11.37)
where $9’ has the value in Eq. (11.6). The dissipation function Dv is a
quadratic form in 4;. The value, of %,rM is the dissipative force due to
thermomolecular diffusion. It is obtained by writing the variational rela-
tion
replacing 6Si and si by the values obtained from Eqs. (8.37) and (8.38).
This provides the exact value of RTM as a nonlinear function of Gi. How-
ever in most problems we may neglect the integral in the values (8.37) and
(8.38). In this case we may write
6lDTM
RTM
I
= - DTM = -rbc#M &
(11.38a)
&ji ’ I
s1 Tt
where DTM is a quadratic form in Gi representing the total dissipation of

thermomolecular diffusion.
Determination of the generalized driving force Qi is obtained from Eq.
(10.3), written in the form
~Q~6qi=6W”-IA(~~xSM:+B6Si)nidA, (11.39)
where A is the boundary of R. On the right is the virtual work of all the
mechanical forces. The integral represents the work due to thermobaric
transfer at the boundary considered as fixed, whereas 6WM represents the
virtual work of all other forces. Special care must be exercised here by
noting that in the absence of viscous stresses 6WM = 0. In this case the
virtual work due to boundary displacement 6ui against the local equilib-
rium pressure p is already included in cpkSM/ of the surface integral.
Hence the remaining work 6WM is due only to the viscous stresses and is
written
6WM = j-* ojlnl 6Uj dA. (11.40)
This is also verified using a lengthier procedure (Biot, 1979, 1982a).

We substitute this value into Eq. (11.39) and use the values in Eqs.
(11.22) and (11.23) for 65’; and Si. Again we may generally neglect the
volume integrals in these expressions. Coefficients of 6qi yield Qi.
With these results the Lagrangian equations are written
Zi + RFh + aD/aQi + &P’/aqi = Qi, (11.41)
where D = Dv + DTM is the dissipation function due to viscous and

thermomolecular dissipation. For a quasi-reversible evolution including
chemical reactions we may neglect s* in describing the system and use the
entropy fluence Si as a state variable instead of ?P:. In that case the
dissipation is expressed by a single dissipation function and the Lagran-
gian equations assume the simpler form [Eq. (10.8)].
It was also shown (Biot, 1979, 1982a) that the inertial force may be
expressed in terms of the kinetic energy
(11.42)
as
V,V, aMj nj dA, (I 1.43)

Qi
where
A,i= Indj!$dbl, Mi=CM;,

I k
(11.44)
and
44 M. A. Biot
1
SBi =JjW;j, Wij = 3 (Uij - Uji). (11.45)
It is interesting to note that
SeiUi= 0, T JtAiGi= 0. (11.46)
Hence & represents a generalized inertial force normal to the velocity,

corresponding to a Magnus effect.
XII. Dynamics of Solids with Elastoviscous

Stresses and Heat Conduction,
and Thermoelasticity
The dynamics of a deformable solid will now be considered in the

particular case where rate-dependent viscous stresses are present in addi-
tion to elastic stresses, with simultaneous heat conduction across the
deforming medium. The problem was treated in detail earlier (Biot, 1976b,
1981) and we shall present here a short and simplified version of the
results.
The field is described by mass point displacements Ui such that
Xi = Xi + Ui(Xj, t), (12.1)
where a solid mass point with initial coordinate Xi is displaced to a point
with coordinate Xi. We have seen (Section V) that the local deformation
may be defined as an affine transformation relative to rotating axes by six
components &ij, which are assumed to be known functions of the displace-
ment gradients auilaxi. The corresponding stress components TVare then
derived from the principle of virtual work. In the absence of molecular
diffusion and chemical reactions, the energy % and entropy Y of the solid
per unit initial volume are obtained from Eqs. (6.2) and (6.3) as
d% = (T; + hf) dr, + C, dT, (12.2)
dY = (l/T)(h$dq + C, dT), (12.3)
where Q-:is the elastic part of the stress for reversible deformations and C,
is the heat capacity of the solid per unit initial volume with constant
strain. The value of h$ is given by Eq. (6.7) without calorimetric measure-
ments as
h$ = - T(d$dT), , (12.4)
New Variational-Lagrangian Irreversible Thermodynamics 4.5
where
r; = T&, T, x,) (12.5)
represents the local equations of state of the solid with the elastic stresses
expressed in terms of strain and temperature. The subscript E in Eq. (12.4)
indicates constant strain. Since the solid may be intrinsically nonhomoge-
neous, r: may depend also on the initial coordinates Xi.
Integration of Eqs. (12.2) and (12.3) yields %, Y, and ‘V = Q - TOY as
functions of Eij, T and xi.
The entropy production arises from the viscous stresses and the ther-
mal conduction. This provides another example of two distinct types of
entropy production, the first being due to an irreversible production of
heat while the second is not.
In order to apply the principle of virtual dissipation we must evaluate
the virtual entropy production. We therefore replace the differentials by
variations in Eqs. (12.2) and (12.3). They yield
6% = r; 8~ + T 69. (12.6)
For an irreversible transformation, conservation of energy is expressed
by
6% = (7; + rz) a&ij - d 6HildXi, (12.7)
where ri is the additional viscous stress and 6Hi is the variation of con-
travariant heat fluence.
We recall that fii is the heat flux across a deforming material area
initially normal to xi and equal to unity, whereas Hi is its time integral.
Equating (12.6) and (12.7) yields
69 = 6s” - d GSildXi 3 (12.8)
where
T 6s” = TV
u &s..
rJ - aTlax.I 6s. I? 65‘; = 6HiIT. (12.9)
We recognize the term (-dT/axi) 6Si already derived in (8.6) for thermal
diffusion. The first term may be interpreted as due to heat production
(Biot, 1981, 1982b) by considering an adiabatic transformation where con-
servation of energy requires
&?L = (7: + r$) SEij. (12.10)
Obviously this is equivalent to a reversible transformation where an
amount of heat 6hP = TzSEij is added in order to reproduce the same
change of state as in the irreversible process. The heat corresponds to the
concept of uncompensated heat of Clausius and illustrates the physical
46 M. A. Biot
difference between the two types of entropy production in Eq. (12.9). The
same distinction was discussed above in Section VIII.
As in the case of thermomolecular diffusion (see Section VIII) we now
introduce the kinetics of irreversibility. We write
r; = &v, EPV,T, XI), (12.11)

-dTlaxi = hofij, (12.12)
where
A, = A&+, T, a> (12.13)
is the thermal resistivity of the deformable solid expressed in terms of the

contravariant heat flux Eij. It is a function of the local state variable Ed,,
and T. If the material is nonhomogeneous, it is also a function of the initial
coordinates xi. The viscous stress Q-;, as a function of the rate of deforma-
tion Eij, embodies the kinematics of the mechanical irreversibility. Note
that ry and
-1/T aTIaxi = A,Sj, (12.14)
with si = fii/T, play the same role as the dissipative disequilibrium forces
aP, a?“, and Ch$’ in Eq. (8.25) for the case of thermomolecular diffusion
with chemical reactions. This can be seen by writing the virtual dissipa-
tion (12.9) in terms of rate variables in the form
T 6~” = T$ 6&u + TAGS, SSj, (12.15)

where Q-;is the rate function (12.11). Comparing with Eq. (8.25), we see
that the viscous stress pi may be considered as the tensor equivalent of
the afjkity C!Q,.
Field equations may be obtained readily for the dynamics of the de-
formable solid by applying the principle of virtual dissipation (9.19). We
vary the displacement Ui and the entropy fluence Si. The virtual work of
the inertial forces is
(12.16)
where the volume integral is for the initial domain CRof the space Xi and p
is the initial density. The variation of 9 is
SY + p a%(Xi)/aXi 6Ui dR, (12.17)

6y = InI I
where I is the gravity potential field per unit mass at the point ;Fi= Xi +
Ui. The exergy variation per unit initial volume is
8~=6~-T”69=7~6e,+86Y (12.18)
(fJ = T - To). The variations are applied only inside the domain a, and
with the values (12.16), (12.17), and (12.18), the variational principle
(9.19) is written
.(piii 6Ui + 7B 6&u + 8 6Y + p a%ldxi 6ui + To a~*) dR = 0. (12.19)

I
Equation (8.29) is quite general, and with the value (12.15) it yields
69 = -a 6SilaXi + 6s”. (12.20)
Hence Eq. (12.19) becomes
We replace T 6s* by its value (12.15) and write
= a.9,,laa,j
8&/.&V 6aij, 6aij = d8uildXj. (12.22)
After integration by parts, we equate to zero the coefficients of the arbi-
trary variations 6Ui, 6Si and obtain
(12.23)
g + TAySj = 0. (12.24)
I
These are the field equations for the deformable solid with elastic and
nonlinear viscous stresses and with thermal conduction.
In solving the problem we may consider ui and St as unknowns that
determine the state through Eq. (8.30) if we also determine the entropy
produced s*. This may be done by adding to the field equations the auxil-
iary equation
TS” = ~~6~ + TAuSiSj (12.25)
derived from Eq. (12.15). It extends Meixner’s (1941) result to deformable
solids with viscosity. Another procedure in analogy with the use of Eq.
(8.36) for thermomolecular diffusion is to introduce a pseudoenergy
fluence 9: defined by Eq. (7.15). The energy balance equation (12.7) is
then written
-di!@:laXi = ($ + T;) 8~ - dMIi/aXi. (12.26)
Hence
48 M. A. Biot
6H; = T 6Si = 8%: + lfl+ g(x/ 9 x:)(7; + ~z) 8~0 do+. (12.27)
This becomes
TS, = 9: + I,+ g(x/ x1+)(7: + ~;)sGediR+.

3 (12.28)
With this value of Si Eqs. (12.23) and (12.24) become integro-differential

equations in Ui and ST, which are now state variables. A similar proce-
dure has also been discussed in an earlier paper (Biot, 1981).
If the irreversible process is quasi-reversible, we may neglect the con-
tribution of s* to the value of Y and use Si as a state variable. In that case
Onsager’s (1930, 1931) reciprocity relations are valid, and we may write
7; = d9”/dbij, aT/aXi = -tBTIdSi, (12.29)
where 9” is a quadratic form in .+ with coefficients that are functions of
the local state variables eij and T, and
9’ = i TAoSiSj, for A, = Aji, (12.30)
is a purely thermal dissipation function. The virtual dissipation (12.15)

now becomes
T 6S* = d~“fd~, 8Eij + &hTIaSi 6Si (12.31)
and leads to the field equations
Pfii - 6 (7:” 2) - & (g $$ + p g = 0, (12.32)
g+z=o.
I I
(12.33)
The contravariant thermal resistivity A, may be expressed in terms of

the local Cartesian thermal conductivity kc as follows. With the Cartesian
resistivity
[A,1= W,l-‘, (12.34)

we write the heat conduction law in the form
-dTlt+Fi = X,fii) (12.35)
where fij is the Cartesian heat flux in the Xi coordinates. By using Eqs.
(7.12) we introduce the contravariant heat flux and obtain
--aTId.& = J’h,@, axjlax,. (12.36)

Multiplying both sides by aXi/ax, this yields
(12.37)
Comparing with Eq. (12.12), we see that
(12.38)
showing the covariant nature of A,. Note that A,, is a function only of the
strain Ed, T, and Xi, and hence represents an intrinsic physical property,
whereas A, is referred to the fixed axes xi and depends also on the rotation
of the material.
Lagrangian equations are immediately obtained from the principle of
virtual dissipation. For simplicity, we shall assume that the entropy pro-
duced s* does not contribute significantly to the state variables, so that
the system may be described in terms of generalized coordinates qi by the
displacement field Ui and the contravariant entropy fluence Si. We write
uj = uj(41, q2, *** XI)5 (12.39)

Sj = sj(41 3 q2 3 -*. Xl). (12.40)
We apply arbitrary variations 6qi. The variation of the mixed collective

potential is obtained from its value CP(qJ expressed as a function of qi.
By a classical procedure the virtual work of the inertia forces yields
Tzisqi= [g(g) -g] *qi, (12.41)
where
+’ n ptiiicidC! (12.42)
I
2
is the kinetic energy, and 0 the initial domain.

We shall assume that the thermal conduction satisfies the reciprocity
relations A, = Aji while the viscous stress remains a general nonlinear
function rT(.sPY,.k,, , T). In this case, according to Eqs. (12.15) and (12.29),
the virtual dissipation is
T 6~” = T: 8~0 + (&31T/~Si)GSi. (12.43)
For the whole system we write
(12.44)
50 M. A. Biot
where
(12.45)
The total thermal dissipation function DT is a quadratic form in & with

coefficients being functions of qi. It is obtained by writing
By noting that aSj/aqi = aSj/ag;, we derive
(12.47)
where
DT = I,, : ST da. (12.48)
Finally we express the virtual work of the mechanical and thermodynamic

forces applied at the boundary A. This is
6 WM + 6 WH = C Qi 6qi = I, (A 8Uj - 8 SSj nj) dA. (12.49)

i
The integral is extended to the boundary A in the initial space xi, while5 is
the boundary force per unit initial area at the deformed boundary. Values
of 8 and SSj are also at the deformed boundary. The generalized mechani-
cal and thermodynamic forces at the boundary are therefore
(12.50)
With 69(qJ and the values (12.41), (12.44), and (12.50), the principle of
virtual dissipation (9.19) leads to the Lagrangian equations
(12.51)
The term R? is the generalized dissipative force due to the viscous

stresses; it embodies the most general case where these stresses are non-
linear functions of the strain &iiand the strain rate .+. If these stresses are
linear functions of the strain rate and satisfy the Onsager reciprocity
relations
ar,,/a&j = ar,/l%,, ) (12.52)
they may be expressed in terms of a dissipation function 9” that is a

quadratic form in .Q. This leads to the Lagrangian equations
(12.53)
with a total dissipation function
(12.54)
which represents the combined viscous and thermal dissipation.

We have simplified the Lagrangian formulation by neglecting s* as a
state variable. The accuracy may be improved in several ways. One way
is to evaluate s* as a function of time from the first approximation and
introduce it in the value (8.31) of Y, which is now written
Y = - a(s; + AS;)laxi + s*, (12.55)
whereas the displacement is Ui + Aui . The new unknowns are the correc-
tions ASi and Aui to the first approximations Si and Ui. The corrections are
expressed in terms of generalized coordinates as Eqs. (12.39) and (12.40),
and the corresponding Lagrangian equations contain the time explicitly.
Another procedure is to use the pseudo energy fluence %“: and Eqs.
(12.27) and (12.28) for 6Si and $i. The procedure is analogous to the one
explained in Section VIII and discussed more extensively elsewhere
(Biot, 1981).
If there are no viscous stresses the solid is thermoelastic. The field
equations are obtained from Eqs. (12.23) and (12.24) by putting T: = 0.
The Lagrangian equations are obtained from Eq. (12.51) by putting Rv =
0. In this case the exergy ‘Y becomes the thermoelastic potential (Biot,
1973a).
XIII. Inhomogeneous Viscous Fluid with

Convected Coordinates
and Heat Conduction
The general case of a deforming solid, analyzed in the previous section,

has a number of interesting applications that are better discussed sepa-
rately. In particular, we shall consider the case of a compressible viscous
fluid, Newtonian or non-Newtonian, with thermal conduction. For the
homogeneous fluid, this may also be treated from an Eulerian viewpoint
as in Section XI for the more general case of a mixture. The purpose here
52 M. A. Biot
is to develop equations using convected coordinates Xi = xi + Uifor a fluid,

which may be inhomogeneous. The energy 011and entropy Y per unit
initial volume are functions
% = %(J, Ty xi), Y = 9(Jy T, Xi) (13.1)
of the temperature T and the Jacobian J (7.1 l), which represents the ratio
of final to initial volume of each fluid element. If the fluid is nonhomoge-
neous, %. and Y are also functions of the initial coordinates xi.
The thermal dissipation function per unit initial volume is derived from
Eq. (12.30) as follows. The thermal conductivity
k = k(J, T, Xi) (13.2)
remains isotropic and is a function of J, T, and Xi. The Cartesian thermal

resistivity is
A, = 6,jk. (13.3)
By substituting this value in the covariant thermal resistivity (12.38) we

obtain
A,, = 1 ax,8% (13.4)

!I
Jk dxi axj *
The thermal dissipation function per unit initial volume (12.30) is there-
fore
(13.5)
in terms of the contravariant entropy flux Si.

In order to derive the dissipation due to the fluid viscosity for a Newto-
nian fluid we consider the stress components aij in terms of the velocity
gradients in the Xi space. They are
(13.6)
where
71 = ql(J, T, -4, ~2 = ~(59 T, xi)
are viscosity coefficients, functions of the local state variables J, T and
the initial coordinates Xi. If there is no bulk viscosity we put
2
3 7)l + r/2 = 0. (13.7)
The virtual dissipation due to the viscosity per unit initial volume is
T 6~“” = Juip %&la?, = Jai+ dxjlaZ, 6aij (13.8)
where
6ati = d&.dJdXj. (13.9)
We may express the derivative axi/aXj in terms of aSSi/axjby writing the

affine transformation from dxj into dli and solving the linear equations for
dxi. The result is similar to Eq. (7.8). We obtain
dXildEj = CjilJ, (13.10)
where C, is the cofactor of Zi$ilaxj in the Jacobian J. Hence the virtual
dissipation (13.8) becomes
T 6~“’ = Cpjcip 6ati = fl. 6ai . (13.11)
The physical significance of this expression is brought out by comparison
with the virtual work (5.12). Obviously,
TT = CpjaiF (13.12)
is the Piola viscous stress, i.e., the Cartesian components of the viscous
forces acting on areas initially equal to unity and normal to xj.
Similarly, we may express the velocity gradients in Eq. (13.6) in terms
of partial derivatives with respect to xi. We note that tii = Uiand write
dUi &ii ax, Cjv .
aq=ax,bg=~aiv, (13.13)
where
&ij = dlii/dXj. (13.14)
We substitute the value (13.13) into expression (13.6) and changej into p,
thus obtaining
rip = (~llJ)(Cpv& + C&p) + (s2/J)&~Cd/v. (13.15)
With this value, the Piola stress (13.12) becomes
Ti = (~~IJ)(C,“C~jki, + CpjC’,kp,) + (~2/J)GipC~jC’~,&~,. (13.16)
This may be written as
Ti = (~lIJ)(C~vC&~+ C+$iu)bpv+ (~lJ)C,Cpv~pv. (13.17)
By putting
By = (~llJ)(C/uC$i~ + CpjCiv>
+ (~/J)c~cpv, (13.18)
54 44. A. Biot
the stress (13.17) becomes
Hence, Onsagers reciprocity relations By = Biv are verified for TT . This

is as should be because it is an invariant property, already verified by the
Newtonian stresses (13.6) (Biot, 1976b).
The rate of dissipation per unit initial volume is obtained from Eq.
(13.11) by replacing the variations by time derivatives. We obtain
Ti*” = T;.& = 2sv = B~~,,~.., (13.20)
where Qv is a quadratic form in kij. Because of the reciprocity property
(13.19), we may write
T; = aWl&i,. (13.21)
The exergy per unit initial volume is
V = “Ir(J, Ty Xi); (13.22)
it is a function of the Jacobian .I, the temperature T, and the initial coordi-
nates xi. Considering YVas a function of .I and Y, its variation is
6r/‘ = cYtf/aJ 6J + t’ 69. (13.23)
Obviously, the local equilibrium pressure is
P = 8tf/dJ (13.24)
for a reversible slow deformation. This result corresponds to equation
(12.6). We may also write
$V = Tf 6a, + 196S, (13.25)
where
TF = P 8Jlaa,, (13.26)
is the Piola stress for the elastic pressure P.
By applying the principle of virtual dissipation with arbitrary variations
6ui inside a domain R and proceeding as in the preceding section for the
solid, we obtain the field equations
.. d a%
Wi - G, <TF+ Tz) + p axi = 0, (13.27)
dTldXi + d9TlaSi= 0, (13.28)
where % is the gravity potential, %I*is given by Eq. (13.5), and p is the
initial density. An auxiliary equation for S* is
TS* = 2(%” + aT). (13.29)
Under the assumption that Si may be used as a state variable, the

Lagrangian equations have the same form as Eq. (12.53), where the val-
ues (13.20) and (13.5) are now substituted in Eq. (12.54) for the dissipation
function.
For a non-Newtonian fluid, the viscous stress uij is expressed as
u~ij= F,Q + Fze,$ + Fjeiet, (13.30)

where
(13.31)
and F1, F2, F3 are functions of the three invariants e&, ekei, e;ejke;i.
In the general case they are also functions of T and xi if the fluid is
nonisothermal and nonhomogeneous. A very simple proof of this formula
has been given (Biot, 1976b). The Piola viscous stress TT in this case
cannot generally be derived from a dissipation function as for the Newto-
nian case (13.21). However, it is given by the same formula (13.12) in
terms of the stress Q. The field equations are then the same as Eqs.
(13.27) and (13.28). However, the Lagrangian equations become
(13.32)
where
derived from Eq. (13.11)) is the generalized non-Newtonian viscous force

and
(13.33)
is the thermal dissipation function.

A special case of practical interest is that of a viscous incompressible
solid undergoing slow deformations under isothermal conditions (T = TO).
Inertial forces are negligible, and we put 3 = 0. The material may be
inhomogeneous, either continuous or composed of different adherent
solids. The potential C!?depends only on gravity, and its value is
9 = J,p%(Zi) do. (13.34)

56 M. A. Biot
The dissipation function D depends only on the viscosity. By assuming

Newtonian viscosity and putting T = TO, we derive
(13.35)
The displacement field Ui is represented by generalized coordinates qi as
uj = uj(91, q2, ...v x/). (13.36)
The dissipation function (13.35) is then a quadratic form
(13.37)
where the coefficients bu are functions of the generalized coordinates qi .

The Lagrangian equations become
a9’ldqi + aDlag, = Qi. (13.38)
The generalized driving forces Qi are obtained by putting 8 = 0 in the

surface integral (12.50), where5 are the forces at the boundary per unit
initial area.
XIV. Lagrangian Equations of Heat Transfer and

Their Mechanical Interpretation, and a Mass
Transfer Analogy
The particular case of pure heat conduction is obtained by considering a

rigid solid (Ed = 0). An important feature here is that the heatfluence Hi
becomes a state variable. The heat content, or energy per unit volume, is
then
%_= h = -aHilaxi. (14.1)
The total exergy of the domain R is now
(14.2)
where Y is a function of h. The dissipation function is
(14.3)
The thermal time history of the domain is determined by the fluence field
Hi in terms of a finite number of generalized coordinates whose evolution
is governed by the Lagrangian equations
aV/dqi + aD/aGi = Qi. (14.4)
The thermal generalized force Qi is obtained as a particular case of Eq.
(12.50) by putting 6Ui = 0 SO that
(14.5)
It is worth noting the physical significance of these equations. The purely

thermal exergy Yf represents the work to be accomplished mechanically
by thermobaric transfer to bring a unit volume of solid to a given tempera-
ture. The heat is transferred by a heat pump from the thermal well at a
temperature To to the solid at a different temperature T. Similarly, the
generalized thermal force (14.5) is the work accomplished mechanically in
order to inject heat at the boundary by the same heat pumping process.
On the other hand, the dissipation function (14.3) represents the rate of
loss of mechanical availability of the thermal energy due to thermal con-
duction. Thus the Lagrangian equations (14.4) for heat conduction are
given a purely mechanical interpretation as related to an availability
balance.
An important formal simplification is immediately evident in the linear
theory where 8 is small and T = To. In this case l/T,, is eliminated from the
equations as a common factor. It was found that the factor l/T0 may also
be eliminated in the general nonlinear case. This is shown by applying the
principle of virtual dissipation with arbitrary variations inside the domain
R. This leads to
(,($6h + $ hoHi SHj) da = 0.
After integration by parts using the relation a(e/T)axi = -(T0/T2) de/dxi

and taking out the factor T,-,lT*, we obtain
Another integration by parts then yields
I,,cH 6h + AgCij 6Hi) da = - I, 8 6Hini dA, (14.8)
which is the variational principle derived earlier (see Biot, 1970). It leads
to the Lagrangian equations (14.4) with a thermal potential V, a dissipa-
58 M. A. Biot
tion function D, and a generalized thermal force Qj, which are now ex-
pressed as
(14.9)
Applications of this Lagrangian approach to heat transfer were developed

in great detail in a monograph (Biot, 1970). The Lagrangian equations for
conduction and simultaneous convection in an incompressible fluid of
prescribed motion were also obtained. The heat flux in this case is
hi = Ji + Uih, (14.10)
where ui is the velocity of the medium. The dissipation function is now
written
D = 4 I,, hoJiJj d!Fl. (14.11)
This result may be derived directly (Biot, 1970) or also as a particular case
of the general treatment of Section XI when restricted to pure heat trans-
fer for a given velocity field of the fluid.
Attention should be called to special formulations such as that of asso-
ciatedJluencefields, which lead to the use of scalar temperature fields as
unknowns instead of Hi, and the treatment of boundary heat transfer to a
moving fluid using the concept of a trailing function (Biot, 1970). The
latter eliminated the inconsistencies of standard methods based on local
heat transfer coefficients. Collective analysis by Lagrangian equations is
ideally suited to the unified treatment of heat transfer in mixed systems
constituted by solids and moving fluids.
A highly useful concept that was derived from the Lagrangian formula-
tion is that of penetration depth, which yields immediately the heat
fluence due to sudden temperature rise at the boundary. It can be used as
a basic tool simplifying the formulation of very complex problems.
Application of the Lagrangian equations has brought simplification and
physical insight in many problems of heat transfer (Lardner, 1963, 1967;
Prasad and Agrawal, 1972, 1974; Chung and Yeh, 1975; Yeh and Chung,
1977).
Many types of finite element methods may also be derived directly from
the Lagrangian equations. For example, we may treat as generalized co-
ordinates the fluence vectors located at the vertices of a grid, using linear
or polynomial interpolations of these values to represent the complete
field. The method is systematic and should be highly accurate, because

heat flux continuity is preserved (see Section XXVI).
It should also be noted that the Lagrangian equations are derived di-
rectly from the principle of virtual dissipation, using the dissipation func-
tion as a basic invariant, without requiring any prior knowledge of the
field differential equations. This is in contrast with formal methods based
on functional space theories. In addition, the variational principle yields
directly the field differential equations in any coordinate system. The
invariance of the variational Lagrangian formulations also brings to light
unifying fundamental properties of the governing equations (Lonngren
and Hsuan, 1978).
The Lagrangian variational equations of heat transfer are completely
isomorphic to those of pure isothermal mass transfer (Biot, 1970). The
heat fluence field Hi is simply replaced in all equations by the mass fluence
Mi. Field and Lagrangian equations for nonlinear problems of mass trans-
fer are thus readily obtained, as exemplified by the case of a moving
boundary (Senf, 1981).
XV. Deformable Solids with Tbermomolecular

Diffusion and Chemical Reactions
As another example we shall consider the deformation of a nonhomoge-

neous solid, which contains a number of substances in solution. Thermal
and molecular diffusion of the substances are induced by the deformation
as well as their concentration and thermal gradients. Simultaneous chemi-
cal reactions between the dissolved substances may also occur. For sim-
plicity we shall neglect the inertial forces and assume creeping deforma-
tion. The energy Q and entropy Y per unit initial volume are obtained by
integrating Eqs. (6.2) and (6.3) along an arbitrary path. Although they are
formulated for a single reaction, the results are readily generalized to
multiple reactions tP, as was done for the value (6.11). We thus obtain
Q = Wij, 5,) Mk, T, xl), (15.1)
Y = Y(Q, Z, 9 Mk, T, x/l (15.2)
as functions of the deformation &ii, the chemical coordinates tP, the

masses Mk added by diffusion, the temperature T, and the initial coordi-
nates xi. The latter dependence represents the inhomogeneity. From
60 M. A. Biot
Eq. (15.2) we may obtain T as a function of Y, Q, 5,) M’, and xi. Thus the
exergy
=V= Q_l- T&f’ (15.3)
per unit initial volume becomes a function of these variables. In a gravity
field %(Xi) the mixed collective potential of a solid occupying the initial
domain R is
dMl‘
y=Jp+( mo + m)%(Lf;)] da, m=-CL
k axj ’
(15.4)
where mo is the initial mass per unit initial volume and m is the mass
added by the contravariant fluences Mf. We recall that kl: is the mass flux
of the dissolved substance k through a deforming solid area initially equal
to unity and initially normal to the xi axis. From Eqs. (6.11) and (15.3) we
obtain
6~=rij~~ij+fC~8Mk+OSY--AAp~~p. (15.5)
k P
This result extends the value (12.18) to the solid with molecular diffusion
and chemical reactions.
The virtual dissipation is due only to the chemical reactions and the
thermomolecular diffusion. It has a form similar to Eq. (8.25).
where i@ and Si are now contravariant fluxes. The affinity CRnp

is a function
of tp, tJp, EC, Mk, T, and Xi. The dissipation function ?JTM for thermo-
molecular diffusion is a quadratic form in k$ and Si whose coefficients are
functions &, q, Mk, T, and Xi.
We now apply the principle of virtual dissipation with arbitrary varia-
tions, St,, 6ui, 6MF, 6Si inside the domain R of the solid. Proceeding as
for the solid in Section XII we derive the field equations
(15.7)
d(Pkldxi + &STMIa~f= 0, (15.8)

8Ttdx. I + &!hTM/aS~
I = 0? (15.9)
-A,, + 91p = 0, (15.10)
where C$?k
is the mixed convective potential (9.21). The symmetric struc-
ture and simplicity of these equations is worth noting in view of the
complexity of the physics. We may consider 5,) Ui, MF, and Si as the field
unknowns, describing the evolution of the state of the system. However if
the contribution from s* is significant in determining the state of the
system we may add an auxiliary equation such as Eq. (11.21) for the rate
of dissipation or proceed as in Sections XI and XII by introducing a
pseudo-energy fluence St in the field equations.
By neglecting s* in first approximation we may describe the system in
terms of generalized coordinates qi as
uj = uj(41, q2, *+*945 (15.11)
M! = Mj(qr, q2, . . . . 4, (15.12)
sj = sj<Sl, q2, .*.FX/)7 (15.13)
5p = &&?I, q2, ***,XI). (15.14)

By applying the principle of virtual dissipation with arbitrary variations
6qi, we derive the Lagrangian equations
d’??‘/aqi+ Ri + aDTMI&ji = Qi, (15.15)
where
(15.16)
is similar to Eq. (11.36) and represents the generalized chemical dissipa-

tive force or affinity. The dissipation function for thermomolecular diffu-
sion is
DTM ;/$3TMdfi
= _ (15.17)
and the generalized driving force
is due to forces fi applied at the boundary per unit initial area and to
thermodynamic forces of the environment on the open system. This ex-
pression contains the particular case [Eq. (12.50)] derived above.
For the case where chemical reactions are not far from equilibrium, the
evolution is quasi-reversible, and we may introduce a chemical dissipa-
tion function
(15.19)
62 M. A. Biot
where GPh is a quadratic form in &, such that

?hP = d?Pl&,. (15.20)
In this case the Lagrangian equations (15.15) become
a9’/dqi + aDId& = Qi, (15.21)
where
D = Dch + DTM (15.22)
is the total dissipation function. Note the formal identity of (15.21) with
(13.38) for the creeping viscous solid.
XVI. Thermodynamics of Nonlinear

Viscoelasticity and Plasticity with Internal
Coordinates and Heredity
A general theory of linear viscoelasticity based on the Lagrangian equa-

tions of irreversible thermodynamics was established (Biot, 1954, 1955,
1956a, 1958). Its characteristic feature is the use of the concept of internal
coordinates to represent heredity properties. The same Lagrangian for-
mulation provides a natural extension to nonlinear viscoelasticity (Biot,
1976b). We shall consider the linear case in a subsequent section dealing
with general linear phenomena and present here the thermodynamic the-
ory of nonlinear viscoelasticity.
In order to formulate the stress-strain relations of the solid we apply a
finite homogeneous deformation defined by six strain components cij to a
solid element that is initially a cube of unit size. The six strain compo-
nents may be defined by any of the various ways described in Section V.
As already pointed out, they are not necessarily tensor components. The
six stress components rii are corresponding forces applied to the faces of
the solid element and defined by the principle of virtual work. Obviously
the solid element represents a domain fi as considered in Sections XII and
XV, where eij plays the role of generalized coordinates qi while rij plays
the role of generalized forces Qi applied externally to the system. In
addition to external degrees of freedom represented by Ed, there may be
internal ones represented by internal coordinates qs, for which there are
no external driving forces (Qs = 0).
Such will be the case if, for example, the solid contains substances in
solution that react chemically. Internal coordinates may also correspond
to internal fluence fields due to thermomolecular diffusion if the material
possesses a microstructure of inhomogeneities.
Nonlinear viscoelasticity may be defined by the property that the sys-

tem is quasi-reversible, i.e., that the irreversibility is represented by a
dissipation function that is a quadratic form in the time derivatives of the
generalized coordinates. The stress-strain relations are then obtained by
writing the Lagrangian equations of the system. We write the exergy of
the solid element as
v = V(&ij, Y, 4s), (16.1)
where Y is the entropy of the solid element and qs represents a very large
number of internal coordinates. The dissipation function
D = D(.Q, Eij, Y, Yi, qs, ci,) (16.2)
is a quadratic form in Eij, Y, and &. The Lagrangian equations of the
element are
al’/&, + aDI&, = rti, (16.3)
aVIaY + aDl& = 8, (16.4)
aV/aq, + aDI&& = 0. (16.5)
By referring to Eq. (8.31), we note that because of the assumption of
quasi-reversibility, S* is negligible; hence this equation, integrated over
the domain Q of the element, shows that Y is the surface integral of the
fluence Si at the boundary. Hence Y is a state variable whose conjugate
generalized force is 8. As a consequence, T dY represents the amount of
heat provided reversibly to the solid element. Equations (16.5) are linear
in es. They may be integrated for qs, which then become functionals
4s = sS[&j(t)V WI1 (16.6)
of q(t) and Y(t). When these functionals are substituted into Eqs. (16.3)
and (16.4), these equations provide the stress-strain relations for nonlin-
ear thermoviscoelasticity. Note that this result takes into account the
temperature change and heat injection at the rate T$‘, thus yielding a
heredity of the specific heat.
Plasticity may be handled in the same way (Biot, 1976b). However in
this case the deformation is not quasi-reversible, and we may not assume
the existence of a dissipation function. The internal coordinates for a
plastic material are represented by dislocation slips qt. The double sub-
script defines the orientation of the dislocation and the slip direction. For
simplicity we assume constant temperature T = To. The exergy
v = V(&jj, 4;) (16.7)
is a function of the deformation and the internal coordinates. The virtual
dissipation is
64 M. A. Biot
To as* = R,; Sq,;, (16.8)
where Ri is the dissipative force acting on the dislocations. It may be

expressed as a nonlinear rate function
(16.9)
The corresponding Lagrangian equations
aV/a&~ = rij, (16.10)

aV/aq; + R; = 0 (16.11)
represent the stress-strain relations for plastic deformation with internal
coordinates q$ and heredity. Comparison with Eq. (15.16) shows that RG
is the tensor equivalent of the affinity.
Strain hardening is represented by freezing an increasing number of
internal coordinates qi as the deformation proceeds. For a material that
becomes weaker as the deformation increases, the opposite procedure is
used by releasing an increasing number of internal coordinates. This prop-
erty may be assimilated to internal failures. It is of particular interest for
fiber composites.
For a viscoelastic continuum we may also derive field equations from
the variational principle (Biot, 1976b). The result is entirely similar to
Eqs. (15.7) and (15.9). However for a solid that is not centrosymmetric we
must take into account possible coupling terms of the type of EijSP and
&$, in the dissipation function. This leads to a coupling between the
temperature gradient and the rate of deformation. These more general
results are easily obtained by straightforward application of the varia-
tional procedures described above.
XVII. Dynamics of a Fluid-Saturated Deformable

Porous Solid with Heat and Mass Transfer
The principle of virtual dissipation has been used to develop the field
and Lagrangian equations for the dynamics of a porous solid with fluid
saturation of the pores and heat transfer by convection and conduction.
We shall follow the procedures developed earlier for the isothermal non-
inertial case (Biot, 1972) and later for the nonisothermal dynamic case
(Biot, 1977b).
For the noninertial case the problem is similar to that of a solid with a
single substance in solution, and in many ways it may be considered as a
particular case of the one treated in Section XV. However in the present
case we shall also take into account the inertial forces due to the motion of
the porous frame and the motion of the fluid relative to the frame. We
start with a unit initial volume of the material, initially of total mass mo.
As the material is deformed, the exergy per unit initial volume is
Y = Y(E~, m, T, XI), (17.1)
where cij are the six strain components as defined in Section V, and
m = -aMilaXi (17.2)
is the mass of fluid added in the pores per unit initial volume. The con-
travariant fluence Mj is the mass of fluid that has been flowing through a
material surface of the porous frame initially equal to unity and initially
normal to Xi. A material point of the solid frame initially of coordinates Xi
is displaced to a point of coordinates Xi = xi + Ui. If the porous material,
which is approximated as a continuum, is inhomogeneous, Y is also a
function of the initial coordinates xi. Using procedures similar to those in
Section XV, we may evaluate the entropy Y of the element in terms of .Q,
m, T, and xI. Hence we express T replacing T by Y as one of the state
variables. We write
‘Y = Y(eij, m, 9, XI), (17.3)
with the property analogous to Eq. (15.5):
(17.4)
where 8 = T - To. The fluid pressure pf in the pores in the present case is
equal to the injection pressure, which was denoted by ps for the case of a
substance in solution. The corresponding injection enthalpy Fr and injec-
tion entropy Ff are given by Eqs. (2.2) and (2.4) as
(17.5)
where pf , pE T’, and F; are the pressure, density temperature, and specific
entropy of the fluid along the path of thermobaric transfer from a supply
cell of fluid at the pressure p. and temperature To. The convective poten-
tial is
d, = Fr - Ts;-, (17.6)
and the mixed collective potential is now
9 = I,, [T + (m. + m)%(F)] dfl, (17.7)
where % is the gravity potential.

66 M. A. Biot
We may write the rate of dissipation per unit initial volume as

TS* = 29, (17.8)
where the dissipation function
(17.9)
is a quadratic form in &i and Si with coefficients functions of the local

state variables .sij and T and the initial coordinates Xi. The contravariant
entropy flux is
S = Srh!.li+ EiiIT. (17.10)
It embodies the convective heat transfer due to &‘i and the heat flux &i
due to conduction. For ai = 0 the dissipation is due entirely to heat
conduction and the dissipation reduces to the term &TA,SiSj already ob-
tained above (12.30) with a covariant thermal resistivity A,. For Eii = 0
the dissipation function involves only hi and leads to Darcy’s law gener-
alized to a nonisotropic deformable medium.
In order to evaluate the inertial forces, we consider the momentum of
the masses of solid and fluid of an element initially of unit volume and
mass mo. This momentum is
Ati = (R&J+ ?YZ)tii+ Q/J, (17.11)
where ti/ is the mass flux of pore fluid relative to the solid frame per unit
final area in the space Xi. The Jacobian (7.11) is the volume J of the
element. We may express ti/ in terms of the contravariant mass flux hi
using relations (7.9) and (7.11):
(17.12)
Hence
Ati = (mo + RZ)tii + tij CY.iT~laXj. (17.13)
Because the element of volume J is not of constant mass, the resultant of

all inertial forces is aAti/dt plus an additional term due to the rate of flow of
momentum out of J. If we neglect the square of the relative velocity of the
pore fluid then the resultant of the inertial forces is
9ii = dJtd/lli/dt + d(tiitij)lC3Xj. (17.14)
The virtual work of the inertial forces for the porous solid, when we vary
6ui and 6Mi, is
C, Zi 69,
1
= I,,9i SUMda + lE iii 6M/ dY, dX2 dZ, 3 (17.15)
where n is the domain R in the coordinates xi.

In evaluating the second integral we have assumed that the acceleration
of the fluid particles may be approximated as iii. On the other hand, in
variational form (7.12) is written
J SM/ = aZi/axj 6Mj. (17.16)
Hence (7.15) becomes
T Zi Sqi = I,(gi Sui + iii aXi/aXj6Mj) dCk. (17.17)
By varying Sui and 6Mi inside the domain fi and applying the principle
of virtual dissipation, we proceed as in Sections XII and XV. This leads to
the field equations
(17.18)
dcplaxi + f39/dtii = -lij ajsjlaxi, (17.19)

aTtaxi + &W&!?i = 0, (17.20)
where cp = a7flam + % = C#I+ 93, T/,, = aV’/as,, , and ati = &Jaxj. The
equations are for the field components Mi and Si. If we wish to take into
account the entropy produced s* as contributing to the state variables we
add the auxiliary equations (17.8).
Lagrangian equations are also obtained directly from the principle of
virtual dissipation by describing the fields Ui, Mi, and Si in terms of gener-
alized coordinates, as in Eqs. (15.11), (15.12), and (1.5.13), assuming Si to
be a state variable as an approximation. The Lagrangian equations are
written
Zi + aDlag, + &!P/aqi = Q;, (17.21)
with the generalized inertial force
+ ii (17.22)
the dissipation function
(17.23)
and the generalized driving force at the boundary A

68 M. A. Biot
(17.24)
The forces fj are applied at the boundary per unit initial area.
A semilinear theory of deformation of porous solids was also developed
for the particular case where linear dependence is assumed between fluid
pressure and microscopic volume changes while bulk deformations re-
main nonlinear (Biot, 1973~).
The problem of fluid flow through a rigid porous solid with heat transfer
was also discussed in detail (Biot, 1978), including phase changes from
vapor to liquid and multiple diffusion channels due to surface adsorption.
The results lead to a large variety of possible applications in geothermal
and aquifer problems with heat and fluid flow including associated sub-
sidence.
XVIII. Linear Thermodynamics near Equilibrium
Perturbations of a system near equilibrium are described by small

changes of the state variables. We denote by 8 the first-order increase of
the initial uniform temperature T,,. Other first-order perturbations are Mk,
the mass increase of each molecular species per unit volume due to con-
vection, the entropy Y, and the reaction coordinates t,, with initial zero
values. The affinities also vanish at equilibrium, and their perturbations
A, are first-order quantities.
There are two important simplifications in the linear theory. First con-
sider the linearized value of the first-order entropy obtained from Eq.
(2.3),
9 = T F&,fk + h/T,, (18.1)
where h is the heat acquired per unit volume and Fokis the initial value of
the injection entropy in the initial equilibrium state. Since Y and Mk are
state variables, the heat injected h is also a state variable.
Another simplification, which is a consequence of the first, is the use of
the thermobaric potential $k instead of the convective potential pk. For
example, we may write Eq. (15.5) for the more general case of a solid, in
differential form, as
dSr = rij dq - 2 A,, d&, + c +k dMk + (O/TO) dh, (18.2)

P k
where dh is now a state variable. The heat fluence Hi defined by

h = -aHi/E+Xi (18.3)
is now also a state variable. By introducing the constant factor l/T0 we

may use as state variable the thermal entropy sT. We write
ST = h/To = -&STIdXi, ST = Hi/To, (18.4)
with a thermal entropy fluence ST. These quantities were introduced in

the earlier developments of the theory (Biot, 1956b). Finally, we may of
course use the total entropy fluence
(18.5)
The fact that it is a first-order state variable is also a consequence of Eq.

(8.31), namely,
Y = -&SJaXi + S*. (18.6)
Since the entropy produced s* is now of the second order, it may be

neglected in a first-order description of the state of the system.
Another point of importance in the linear theory is the dual role played
by the energy % and entropy Y. This may be illustrated in the simple case
of linear thermoelasticity (Biot, 1956b, 1981). For this case the linear
equations of state are
7..
0 = C!!“&
rJ I*v- PijO, (18.7)
where C[” are the isothermal moduli of linear elasticity. The energy and
entropy per unit volume are obtained by applying Eqs. (6.2) and (6.3) for
the particular case d,f = dMk = 0. They become
d% = rij de, + dh, (18.8)
dY = dhlT. (18.9)
By using the equations of state (18.7) in (6.7), the value (6.4) of dh be-
comes
dh = T& deij + c do, (18.10)
where c is the heat capacity per unit volume at constant strain, which we
shall assume constant. We integrate the values (18.8) and (18.9) along a
path, first for 8 = 0 and then for deij = 0. Keeping only first- and second-
order terms, we derive
Q = $CcL”&..& + T,,sT, (18.11)
rJ u P
9 = sT - $(ce*/T;), (18.12)
with
70 M. A. Biot
TosT = h = TO&sij + CO. (18.13)
We see that sT is a first-order state variable, but it does not represent the
correct entropy to the second order. The thermoelastic potential or ex-
ergy becomes (Biot, 1981)
‘V = ‘Q - TOY = B(C;V~ti~pv+ cd*/T,,). (18.14)
It is interesting to note that in this expression the first-order terms drop

out. Hence the entropy plays here a dual role, first as a linearized state
variable sT and second as a value Y that includes second-order terms.
A general discussion of the linear theory is particularly fruitful in La-
grangian form. Small perturbations from equilibrium may be described by
expressing the fluence and displacement fields as linear functions of gen-
eralized coordinates qi. For example, we may represent material point
displacements as
(18.15)
where uji(xl) are fixed displacement fields. Fluence fields Mj” and Sj as well
as <,, are represented similarly. Application of the principle of virtual
dissipation then yields the linear Lagrangian equations for the perturba-
tions qi. They are
(18.16)
with a kinetic energy
(18.17)
a dissipation function
(18.18)
and a mixed collective potential
(18.19)
The constant coefficients are symmetric:

mu = Wlji, bu = bji 7 au = Uji, (18.20)
while 9 and D are positive definite.

Note that the generalized coordinates constitute a complete and accu-

rate representation of the physical system and are not essentially “trial
functions.” This can be seen by using as generalized coordinates values
of the fields at the vertices of a grid system of finite elements sufficiently
small so that macroscopic laws are still valid while fluctuations at the
molecular scale do not yet enter into plcly (see Section XXVI below).
The Lagrangian equations (18.16) thus govern and unify a vast domain
of linear physics obeying a single universal mathematical formalism. In
particular, in the absence of inertial and gravitational forces we may write
dV/dqi + dD/dGi= Qi, (18.21)

aV/& + aD/a& = 0, (18.22)
where qS are a large number of internal coordinates, while Qi are external
driving forces. Consider the case where the equilibrium is stable. Then V
is also nonnegative. We then solve Eqs. (18.22) for qs and substitute these
values into Eq. (18.21). The forces Qi are then expressed as (Biot, 1954)
Qi = C Zvqj> (18.23)
j
with
Z(j=Zji=CDg&+Dij+D~p,
s s
where DC, D, and Di are nonnegative symmetric matrices and r, 2 0. The

quantity p must be interpreted as p = iw for the case where qi varies
proportionally to the harmonic function of time exp(pt).
For a nonperiodic dependence on time starting at t = 0 in a quiescient
system, it is easy to show that Zti may be interpreted as an operator where
df
&f(t) = em” s’e H’ -$ dt’,
0
PfW = z. (18.25)
This interpretation is completely general provided we introduce general-

ized Dirac functions (Biot, 1970).
The linear Lagrangian equations (18.21) and (18.22) were applied to
derive the stress-strain relations of linear viscoelasticity with heredity
(Biot, 1954, 1955, 1956a). By proceeding as in the more general nonlinear
case in Section XVI, we consider the driving forces Qi to represent
stresses rij applied to a unit cube of solid, and cij to represent the associ-
ated response qi. Equations (18.23) in this case are written
7.. = C’tLVE (18.26)
rJ rJ PJ’
72 M. A. Biot
with the operators
(18.27)
These operators satisfy the symmetry properties @’ = &iv = &” = &.

They are the same as the symmetry relations satisfied by t;e ela&
moduli in classical linear elasticity. They also formally coincide with the
elastic moduli obtained for various cases of geometric symmetry, such as
isotropy, cubic symmetry, etc. As a consequence, a principle of uis-
coelastic correspondence was obtained (Biot, 1954, 1955, 1956a, 1958)
whereby all formulas of linear elasticity are immediately applicable to
linear viscoelasticity by simply replacing the elastic moduli by the opera-
tors (18.27).
Another important application of linear thermodynamics is to porous
media including viscoelastic behavior of the solid component. The
stresses rij and the fluid pore pressure pf are considered as driving forces
Q;, while the response qi is the bulk solid strain .sijand the volume of fluid
5 which has entered the pores. Its value is 5 = mlpf , where m is the mass
of fluid added in the pores per unit initial volume of the bulk material and
pr is the fluid density. By applying the genera1 solution (18.23) we obtain
the stress-strain relations (Biot, 1962)
7P = f$YEjj + Li& (18.28)
pf = I%& + i&y, (18.29)
where the operators ar, i&, and & are of the type (18.27). Note that the
heredity properties are due not only to viscoelastitiity of the solid itself
but also involve interactions of a very genera1 relaxation type between the
fluid and the solid, such as fluid squirting in microcracks and between
grains, microthermoelasticity, mutual solubility, adsorption, and surface
diffusion as well as chemical reactions.
A principle of viscoelastic correspondence is also valid for porous me-
dia whereby all formulas of the purely elastic theory are valid for the
viscoelastic case if one simply replaces the elastic’coefficients by the
corresponding operators of the stress-strain relations (18.28) and (18.29)
(Biot, 1962).
The linear thermodynamic theory has also been applied to piezoelectric
crystals with thermal dissipation (Mindlin, 1961, 1974) by adding suitable
electrostatic terms.
XIX. Linear Thermodynamics of a Solid under

Initial Stress
The theory of small perturbations of a system initially in thermody-

namic equilibrium is quite general and is applicable to the case where the
system is in equilibrium in an initial state of stress. However the state of
initial stress must be taken into account in the evaluation of the thermody-
namic functions. The mechanics of initially stressed continua including
elastic and viscoelastic properties was treated extensively in a monograph
(Biot, 1965b). The theory was later extended on the basis of thermody-
namics to include thermomolecular diffusion and chemical reactions
(Biot, 1977~).
We shall present a summary of this completely general case. Attention
is called to its formal applicability to porous media under initial stress
insofar as we may identify seepage with thermomolecular diffusion. The
problem of porous solids under initial stress was also treated earlier in a
somewhat different context (Biot, 1963).
The initial Cartesian stress components are denoted by S, . It is impor-
tant to note that the system may lie in a gravity field, so that equilibrium
does not imply uniformity of the initial injection pressure pOk. As a conse-
quence, the initial values Fok, EON,and I/J~~of the injection entropy, the
injection enthalpy, and the thermobaric potential may vary from point to
point. The initial equilibrium temperature To is of course uniform. When
the system is disturbed from equilibrium, the stress becomes
9-g = tij i- sg. (19.1)

The stress is defined here as in Section V on the basis of virtual work and
may be nontensorial, so that tij is the increase of rij per unit initial area.
The thermobaric potential becomes
$k = +Ok + A’&. (19.2)
The perturbation involves a small displacement field uj of the solid corre-

sponding small strain components
eij = f(aij + aji), aij = dLlilC?Xj. (19.3)
An important point here is that the linear strain components (19.3) are not
sufficient to establish a linearized theory under initial stress. Actually we
must develop the actual strain .aijto the second order. We shall write
(19.4)
74 M. A. Biot
where vij is a second-order quantity in au. For example, in the case (5.5),
we obtain
r)ij = t(eipopj + &jfiWpi + Wp6Jjp)y (19.5)
and in the two-dimensional case (5.7)
roll = 8&, r/22 = -ta21(&2 + a2lh
712 = b2l(a22 - all). (19.6)
The perturbation from equilibrium is described by the displacement

field Ui of the medium, the masses Mk of various substances acquired by
diffusion per unit initial volume, small chemical reactions sP, and temper-
ature increments 8. As already pointed out, in the case of the general
linearized theory we may use instead of 0 the quantity sT = h/To as a state
variable, where h is the heat acquired per unit initial volume. According
to Eq. (18.4), we also introduce the thermal entropy fluence ST as a state
variable. Another simplification is also to use the thermobaric potential I,!&
instead of &. We shall first evaluate the exergy M per unit initial volume.
Its differential (18.2) for the solid under initial stress is
dv = (tij + s,) ds, - c A, d&, + =+,k + L\$!/k)

dMk + 8 dsT (19.7)
P
or
d7r = dW + s, ds, + c ,j,okdMk, (19.8)

k
where
dW = tij ds, - 2 A, d&, + 2 Lb/&dMk + 8 dsT (19.9)

P k
The exergy V is a function of Q, 5,) Mk, and sT. Because S, dcij + & qbOk
dMk is an exact differential, dW is also an exact differential. Furthermore
we may consider au as independent state variables. In a linear theory we
need evaluate @V only to the second order, hence we may replace Ed by eij
in the value of dW. We write
dW = tij deti - 2 A, dtp + c A$I dMk + 8 dsT. (19.10)

P k
Integration yields the quadratic form
W = W(e,, tp, M”, sT) (19.11)
in eti, tp, Mk, and sT with the property

tij = dW/de,, -A,, = aWla&,, A@ = dWlaMk, 0 = awlcw-.

(19.12)
The potential energy in the gravity field is
m. + C Mk %(Fi) dQ, (19.13)

G=ln( k 1
where %(Fi) is the gravity potential at the displaced point Xi = xi + ui and

mo is the initial density. We develop %(Fi) to the second order in ui and
write
%(Xi) = %O+ %;U; + i%‘ijUiUj> (19.14)
with
%o = %Xi), %i = d%(Xi)ldXi 3 %ij = C32%(X;)ldXi dXj . (19.15)
Hence to the second order except for a constant
G=I,F
mo%iui + (%o + %iUi)C Mk + i mo%ouiuj dfk.
k 1 (19.16)
We now apply the principle of virtual dissipation (9.19). With the value
(19.8) of dY it becomes
C Zi 6qi + h, (6W + S, 8~0 + 2 @OkSM”) dlR + 6G + To 6s”

1 k
- C (@Ok+ AI//k+ %i + %iUi)%6MF - Oni SST dA, (19.17)

k 1
where Afi is the increment of forcesfi per unit initial area at the boundary
57 is the thermal entropy fluence (18.5).
When applied to the initial state of equilibrium under initial stress, the
variational principle (19.17) yields
6Mk + mo%i 6Ui dlR

S, 6eij + C I,JJO~
I(n k 1
= I, [A a& - F (‘$Ok + %o) ni SMf] dA. (19.18)
We now subtract equation (19.18) from (19.17) and obtain
2 Zi 6qi + 69 + To 6S*
I
zz AA 6Ui - x (A$k + %iUJ Izi 6M/ - 8ni SST dA

k 1 (19.19)
76 M. A. Biot
where
‘W + S,qi + %iuiC Mk + i mo%guiuj dR (19.20)

k
plays the role of an incremental mixed potential already encountered in

the quoted monograph (Biot, 1965b) for the less general case. The varia-
tional principle (19.19) represents the particular form of the principle of
virtual dissipation for a system under initial stress. In a uniform gravity
field YQ= 0, and %i is a constant equal to the acceleration of gravity. The
virtual work of the inertial forces is
+ 2
k
m&iii + af) 6&
1
da, (19.21)
where mokis the initial mass of substance k per unit volume of solid and a:
its acceleration relative to the solid. We denote by SUP the virtual dis-
placement of the substance associated with 6M/. We have in the linear
case
a? = ti//mok, 6U; = #&%?‘lok. (19.22)
With these values, (19.21) may be written in the form
(19.23)
where
(19.24)
is the kinetic energy per unit volume.

Finally we must evaluate the virtual dissipation. In the linearized the-
ory we may assume the validity of Onsager’s principle. Hence the dissipa-
tion is derived from a dissipation function. We write
where
9 = Gj@ + QrM (19.26)
The first term Sch is a quadratic form in [,, and represents the dissipation
due to chemical reactions. The second term represents the dissipation due
to thermomolecular diffusion. It is a quadratic form in A$: and ST analo-

gous to (17.9) with constant coefficients.
We now substitute the values (19.9), (19.20), (19.23), and (19.25) into
the variational principle (19.19) and vary ui, tp, Mf, and ST arbitrarily
inside the domain. Proceeding as before in the nonlinear problems (Sec-
tions XI, XII, XV), using integration by parts, we obtain the field equa-
tions
(19.27)
aTlaxi + aWa,$T
= 0, (19.29)
-A, + aWa& = 0, (19.30)
where
AKi= mo%guj
+ %iC Mk, (19.31)
k
AqA = A$k + %iUi. (19.32)

The quantity AYk represents the increase of a mixed thermobaric poten-
tial +k + % at the displaced point, as already considered in earlier work
(Biot, 1963) for the initially stressed porous solid. The unknowns in the
field equations are Ui, so, Mf, and ST. They determine the state of the
system to the first order as functions of time.
Lagrangian equations are also obtained directly from the principle of
virtual dissipation (19.20). Using representations of the fields ui, 5,) Mf,
and ST as linear functions of generalized coordinates qi of the type (18.15),
we derive
(19.33)
where
(19.34)
is the total kinetic energy and
D = I, 9 dQ (19.35)
is the dissipation function. The generalized driving force is

78 M. A. Biot
Note that it is an incremental quantity defined by the increments A$, Aqk,

and 8 at the boundary A. The values of 5, D, and 9 are quadratic forms
with constant coefficients, formally the same as (18.17), (KM), and
(18.19).
The Lagrangian equations (18.16) and (19.33), with or without initial
stress, are identical, as they should be, because they both govern the
same fundamental physics of perturbations of a thermodynamic system
near equilibrium. The difference lies in the particular evaluation of 9 and
Qi for each case. A difference also appears in the nature of the equilibrium
state, which may be stable or unstable, as will be discussed in the next
section.
As already mentioned, the theory is directly applicable to fluid satu-
rated porous media under initial stress. The only modification is in the
kinetic energy, which is written
(19.37)
where u is a factor taking into account the distribution of the microveloc-

ity field of the fluid in the pores. The same microvelocity field is also taken
into account in the evaluation of the dissipation function associated with
the viscous fluid seepage (Biot, 1963).
A further generalization of the concept of generalized inertia1 forces is
obtained by the introduction of uiscodynamic operators (Biot, 1962,
1976b).
XX. Linear Thermodynamics and Dissipative

Structures near Unstable Equilibrium
Thermodynamic equilibrium may be stable or unstable. There are many

cases in nature where the equilibrium is unstable. In particular this may be
the case for systems under initial stresses, which may be due to gravity or
externally applied forces. Such systems are governed by the linear La-
grangian equations (18.16). They exhibit important physical properties of
a very genera1 nature, which do not seem to have been recognized. One of
these is the appearance of regular spatial distribution of the unstable
perturbations that may be called dissipative structures and are not bifur-
cations. The erroneous notion that such structures require the system to
be nonlinear and far from equilibrium has been propagated by some

currently fashionable schools.
Another important property of linear instability is its nonoscillatory
character. The properties of such unstable systems near equilibrium may
be derived in complete generality by the linear Lagrangian equations
(18.16) after putting Qi = 0, i.e., assuming no perturbations of the applied
mechanical or thermodynamic forces at the boundary. The instability is
governed by the Lagrangian equations
(20.1)
or, explicitly,
(20.2)
Solutions of these equations are of the type exp(pt) where the p are
characteristic roots of the system. The roots p are either real or complex
conjugate and the system is stable if the real part of each root is negative.
However if there are roots with positive real parts the system is unstable.
A fundamental theorem has been established that states that unstable
roots are all real, hence that the instability is always nonoscillatory (Biot,
1965b, 1974). To show this, we consider a root p and its complex conju-
gate p*. The roots satisfy the equations
C(P’miiqj + pbbqj + aUqi) = 0, (20.3)
x(p*2mvqjx. + p*b,qf + aijqF) = 0. (20.4)

.i
We multiply Eqs. (20.3) by qT and Eqs. (20.4) by qi and add the results.
Taking into account the symmetry properties mij = mji, b, = bji , and aij =
aji, we obtain
~(p2m~ + pbg + av)qiqT = 0, (20.5)

ij
~(p*2m~ + p*bi + au)qjqT = 0. (20.6)

ij
The difference of these two equations yields
(p - P*)[(P + P*) T muqjq? + C bcqjq:] = 0. (20.7)

ij
80 M. A. Biot
By their physical nature the kinetic energy 3 and dissipation function D,

as represented by the quadratic forms (18.17) and (18.18), are positive
definite. Hence
2ij mij@qi > 0, C

ij
biqj*qi > 0. (20.8)
If the solution is unstable,
P + P* ’ 0, (20.9)
and Eq. (20.7) cannot be verified unless p = p*, i.e., unless the root is
real.
Thus near unstable equilibrium the perturbations in the linear range are
nonoscillatory and proportional to increasing exponentials. Note that this
includes dynamical systems with inertial forces.
Another property is derived by assuming an unstable solution with real
p. Equation (20.5) shows that in this case we must have
C a@$& < 0. (20.10)

ij
Hence no instability is possible if 9 = ~X~a~~qi~iis positive definite. This

constitutes a fundamental stability criterion.
An important example of linear instability is provided by a layer of
viscous medium resting on a rigid base and surmounted by another vis-
cous medium of higher density. Due to gravity forces, the system is under
initial stress. The interface is unstable and shows a wavy structure of
given wavelength and amplitude growing exponentially with time (Biot,
196513)as illustrated in Fig. 1. The appearance of such dissipative struc-
tures in linear thermodynamics near equilibrium is quite general. In the
context of geophysics and the formation of salt domes, such structures
have been analyzed in detail (Biot and Ode, 1965; Biot, 1966).
FIG. 1. Viscous layer (1) surmounted by a

denser viscous layer (2) under gravity. Unstable
waviness of increasing amplitude (3) appears at the
interface.
New Variational-Lagrangian Zrrreversible Thermodynamics 81
XXI. Thermoelastic Creep Buckling
Physical insight into the thermodynamics of systems near a state of

unstable equilibrium is provided by the case of a purely thermoelastic
continuum under initial stress. As a simple example, consider a straight
elastic rod initially under axial compression. It will tend to buckle. If the
thermal conductivity is very small, the buckling load will be determined
by the adiabatic elastic coefficients. On the other hand, if thermal conduc-
tivity is very large, the buckling load will be determined by the isothermal
elastic coefficients. Hence we may distinguish between isothermal and
adiabatic buckling. In the case where the axial load is between the iso-
thermal and adiabatic value the thermal conduction will determine the
rate at which the buckling instability appears. According to the general
theorem of the previous section, the buckling that includes the effect of
the inertial forces will be nonoscillatory, and all buckling modes will
exhibit exponentially increasing amplitudes. If the axial load barely ex-
ceeds the isothermal value, the buckling amplitude will increase slowly
and constitute a form of creep motion. In this case, the rate of creep is
limited and dominated by the thermal conduction. We are dealing here
with an example of creep instability that does not involve any viscosity
and is entirely of thermodynamic nature. When we increase the axial load
the rate of buckling increases until it is dominated by the inertial forces
and becomes a dynamic buckling. By idealizing the case for a massless
material we see that creep buckling will occur between two critical buck-
ling loads, a lower one for isothermal buckling and a higher one for adia- 9
batic buckling for which the rate of buckling becomes infinite. In the range
between these two loads the massless rod exhibits a finite rate of creep.
These effects were brought to light in some earlier work and discussed in
the context of a complete analogy with the buckling of a porous elastic
medium saturated with a massless fluid (Biot, 1963, 1964). The two phe-
nomena are isomorphic and belong to the same underlying theory of
instability of linear thermodynamic systems governed by the general La-
grangian equations (18.16). In the thermoelastic case, the dissipation
function is due to thermal diffusion, whereas in the case of a porous
medium it is due to viscous forces generated by fluid seepage between
pores obeying Darcy’s law.
The phenomenon was analyzed in more detail in later work in the
context of general three-dimensional thermoelasticity. In particular, it
was pointed out that for an isolated system the instability may be consid-
ered as occurring because the unstable state of equilibrium corresponds
to a minimum value of the entropy of the whole system (Biot, 1973a,
1974).
82 M. A. Biot
XXII. Lagrangian Formulation of Bifurcations
By their very nature, the generalized coordinates may be used to de-

scribe departures from a given time-dependent evolution of a system. For
example, we may write the displacement field of a solid as
ui = kc41 7 q2, -..Y X1, t), (22.1)

where q, = q2 = *.a = 0 corresponds to a given time-dependent evolution.
Note that in principle the dependence may be chosen arbitrarily, so that
the case qi = 0 does not necessarily represent a solution of the equations
of evolution. The principle of virtual dissipation is applicable to this more
general case, where variations 6qi are applied to any state of the system as
frozen at a particular instant t. Hence Lagrangian equations may be de-
rived that govern the departures from a given arbitrary evolution as mea-
sured by the generalized coordinates qi.
Of considerable interest is the case where qi = 0 represents a solution of
equations governing the system and corresponds to an actual physical
evolution. It is immediately evident that the Lagrangian equations in this
case provide a clue to solutions that represent branching or bifurcations
away from the case qi = 0. In particular, this provides a powerful method
of testing the stability of a given evolution. There are many advantages
associated with the Lagrangian formulation. One is the possibility of
studyingfinite departures from a given evolution in contrast to linearized
methods of infinitesimal perturbations. Another is that a finite number of
#generalized coordinates may be used, with resulting simplification in the
numerical or analytical treatment.
To be mentioned also is the probing of the accuracy of a given solution
by testing the magnitude of possible departures.
XXIII. Generalized Stability Criteria for

Time-Dependent Evolution Far
from Equilibrium
The Lagrangian equations may be applied to provide very general sta-

bility criteria for the time-dependent physical evolution of a system that is
not near equilibrium and for which linear thermodynamics is not applica-
ble. In order to illustrate the method we consider the case of a system for
which the inertial forces are negligible. Putting Zi = 0, the Lagrangian
equations (10.4) become
d9Pldqi + R; = Qi, (23.1)
where Ri is a generalized dissipative force and Qi the generalized driving

forces. Such equations govern (for example) coupled thermomolecular
diffusion and chemical reactions in a gravity field. We denote by
9i = (oi(t> (23.2)
a time-dependent solution; a perturbed solution is given as
qi = Pi(t) + Aqi, (23.3)
where Aqi is a small perturbation. The driving force Qi is given and main-
tained unperturbed. Substitution in Eqs. (23. l), neglecting higher-order
terms, yields
c- j
a29
Qi Nj
Aqj + ARi = 0. (23.4)
The perturbed dissipative force is
ARi = 2 ($ Aqj + aR'A(ij. (23.5)

j J %j 1
The linear differential equations (23.4) in Aqidetermine the time-depen-

dent evolution of the perturbations Aqi. The coefficients of these equa-
tions are generally functions of time. If the perturbations Aqi tend to zero
with time the evolution is stable. The mathematical theory of linear differ-
ential equations provides stability criteria for the solutions of the pertur-
bation equations (23.4). When considered in the context of irreversible
thermodynamics, special stability criteria may be obtained.
Consider, for example, the steady-state evolution already discussed
(Biot, 1976b). We write the perturbation equations (23.4) in the form
2 SQgAqj+ 2 %~A&= 0, (23.6)

j j
where
In addition, we assume that the system is quasi-reversible. In this case the

dissipative forces are derived from a dissipation function
(23.7)
84 M. A. Biot
which is a positive quadratic form in ki with coefficients by depending on

4;. We derive
Ri = aDld+ip $33,= bg = bji, (23.8)
and Eqs. (23.6) become
C bi AGj = 0. (23.9)
j
By multiplying these equations by Aqi and adding the results, we obtain
C &j Aqi Aqj + C 6, Aqi Ail, = 0. (23.10)

ij ij
We note that for a steady state evolution by is constant. Hence we may
write
2 bg Aqi AQj = i (AD), (23.11)

ij
with the positive quadratic form
AD=ic b,AqiAq. (23.12)

0
Equation (23.10) is now
2 tie Aqi Aqj + $ (AD) = 0. (23.13)

ij
If we assume & to be a positive definite matrix, i.e., if
C d, Aqi Aqj > 0, (23.14)

ij
then Eq. (23.13) shows that AD must decrease with time. Because it is
positive definite, the values of Aqi must also tend to zero. Hence the
inequality (23.14) constitutes a fundamental stability criterion.
If the inequality (23.14) is not verified, instability may arise. It may be
due to the negative nature of 8*9/dqi aqj, in which case it is analogous to
the instability of a linear system near equilibrium considered in Section
20. The instability may also be due to the negative nature of (dR;/dqj +
dRj/dqi), which arises essentially from the nonequilibrium state of evolu-
tion. Such a case of instability was illustrated by the example of an em-
bedded viscous layer (Biot, 1976b), as recalled in more detail in the next
section. In addition, in this example & = &ji, which implies that the
instability is nonoscillatory.
The stability criterion (23.14) differs fundamentally from those pres-

ently in vogue (Glansdorff and Prigogine, 1971) by its generality and sim-
plicity as well as by the physical insight provided.
XXIV. Creep and Folding Instability of a

Layered Viscous Solid
Consider a solid viscous layer embedded in a large solid viscous me-

dium of much lower viscosity. We assume incompressible media. A strain
rate is imposed upon this system corresponding to uniform compressive
strain parallel to the axis of the layer. For example, it may be compressed
by two rigid frictionless planes normal to the layer, whose distance de-
creases with time. Obviously if the geometry is perfect, the layer will
remain straight and be uniformly compressed. However, it is known
(Biot, 1965b, 1976b) that the evolution is unstable, and if there are small
initial perturbations of the geometry, they will grow and develop into a
sinusoidal buckling of wavelength
L = 2nh m, (24.1)
where h is the layer thickness, r)l its viscosity, and r) the viscosity of the
embedding medium. The result was verified experimentally. The regular
pattern of wavelength L is obviously a dissipative structure due to an
unstable state of evolution away from equilibrium. It has important impli-
cations in geology.
In the context of the general stability theory in the previous section, the
folding represents a perturbation of a state of evolution. The unperturbed
state of evolution is a uniform compression where the layer remains per-
fectly straight. We have assumed isothermal deformation and neglected
gravity forces. The Lagrangian equations of creeping motion of the solid
in this case become extremely simple. They are
aDl8di = 0, (24.2)
where D is a quadratic form in 4i with coefficients depending on q;. With
9 = 0, the stability criterion is given by Eq. (23.14). It can be shown that it
is not verified in this case for perturbations Aqi from the uniform compres-
sion. The instability and Eq. (24.1) for the buckling wavelength have also
been derived by this method of perturbation of the Lagrangian equations
(Biot, 1976b).
Another example is provided by the internal folding of a stack of vis-
cous layers of alternately high and low viscosity. When the system is
86 44. A. Biot
FIG. 2. Stack of alternate layers of large and small viscosity between rigid planes subject
to a compression P. Note the appearance of internal folding of wavelength L.
compressed in the direction of the layers, an internal folding develops as

shown in Fig. 2, with a wavelength
L = 1.9 m, (24.3)
where h is the thickness of the more viscous layers and H the total
thickness of the stack (Biot, 1965a, 1967).
These instability problems may also be considered from the standpoint
of bifurcation, as discussed in Section XXII. In this case the uniform
compression is represented by a steady state where the generalized coor-
dinates are known functions of time qf = pi(t). Any deviation from this
steady state may then be represented by new generalized coordinates qi
that represent departures from the steady state. We then evaluate the
dissipation function D in terms of qf and qi. It will be a function that is
linear and quadratic in 4i with coefficients functions of qf and qi. The
Lagrangian equations are of the same form as Eq. (24.2) with coefficients
that may now be functions of time. We note that they express minimum
dissipation under the constraint qf = pi(t).
A final remark is in order here in connection with Helmholtz’s theorem,
which states that under creeping flow conditions a viscous fluid tends to a
stable steady-state flow (see Lamb, 1932, p. 619). This is not in contradic-
tion with our results, because Helmholtz’s theorem applies only to a fluid
of uniform viscosity. This is not the case for an embedded layer or a stack
of layers with two different viscosities.
XXV. Coupling of Subsystems and the Principle

of Interconnection
In many problems, we deal with complex systems made up of separate

components that differ from each other by their physical nature. Each of
these components or subsystems may be analyzed separately, and La-

grangian equations may be obtained that govern its behavior. The analysis
of each subsystem may usually be achieved by simple methods adapted to
its particular physical nature in terms of a small number of generalized
coordinates that determine the field distribution of mass and energy
fluence, material displacements, and reaction coordinates. The evolution
of each subsystem (s) is governed by Lagrangian equations
II”’ + RI”’ + $$“‘/aqi = Qjsjint + Qjs)-, (25.1)
where Qpjint are the driving forces on the subsystems at a coupling inter-
face while Q?jext are the driving forces external to the combined system.
We may add Eqs. (25.1) for all subsystems. In this process the interfacial
forces Qjsjintmay be grouped in pairs where they are equal and opposite in
sign at each interface. Hence
c Qpt = 0
s
and
(25.3)
We thus obtain unified Lagrangian equations for the combined system

from which interfacial forces have been eliminated.
The method constitutes a generalization of a process of elimination of
interfacial forces in classical mechanics by the method of virtual work.
We note that in this classical context it is nothing but the expression of
the third law of mechanics whereby action is equal and opposite to reac-
tion. Its extension to thermodynamic forces is evident from expression
(9.17), in which the entropy and mass fluences are continuous at the
boundary while the outward normal directions ni of the subsystems are
equal and opposite at the interface.
A special remark is in order for mechanical systems that are not adher-
ent at interfaces. In this case the interfacial virtual work vanishes only if
there is no friction. However, in the context of the more general thermo-
dynamic formulation we may consider the solids to be adherent by con-
sidering that one of the surfaces is constituted by a thin, deformable,
adherent skin where the shear deformation generates the friction forces
and entropy production. By this artljke, Coulomb friction may be in-
cluded in the general formulation.
The principle of interconnection is applicable to a wide variety of prob-
lems. It was discussed in the particular context of heat transfer (Biot,
1970). There are also a number of problems that have been treated in the
past without realizing that the methods involved are particular cases of
88 M. A. Biot
such a broad unifying principle. For example, in aeronautical structural

analysis and aeroelasticity during the years 1942-1945 it became common
practice to consider normal modes of subsystems as generalized coordi-
nates and interconnect the subsystems by modal synthesis.
Another example is in classical mechanics, where Lagrangian equa-
tions are obtained for the motion of rigid solids in a perfect incompressible
fluid (see Lamb, 1932, p. 160). These equations may be derived by using
the interconnection principle. Lagrangian equations are obtained sepa-
rately for the motion of the solids and that of the fluid due to generalized
interfacial forces that are equal and opposite. By interconnection, dynam-
ical equations are obtained for the solids that embody implicitly the dy-
namics of the surrounding fluid.
Many more general problems suggest themselves here. Among others
we may cite those of interaction between elastic solids and compressible
fluids with or without viscosity.
XXVI. Completeness of the Description by

Generalized Coordinates. Resolution
Threshold and Lagrangian Finite
Element Methods
The present treatment of irreversible thermodynamics emphasizes the

description of a complex system as an assemblage of cells. From a funda-
mental viewpoint, the size of these cells may be extremely small while
remaining above a resolution threshold, below which the statistical aver-
age definition of temperature and entropy breaks down and fluctuations
enter into play. The cells are finite in number and determined by a finite
number of generalized coordinates, which, as pointed out earlier (Biot,
1970), provide a complete physical description from the macroscopic
viewpoint. As a consequence, the corresponding Lagrangian equations
also describe rigorously the evolution of the system.
It is important to note that use of continuum models is an extrapolation,
beyond the validity of pk.ysical laws, which introduces spurious difficul-
ties in terms of completeness in the context of the mathematical concepts
of measure, continuous sets, and functional space theories. Recent work
by Woods (1981) has demonstrated the lack of physical validity of much of
the current fashionable formalism of continuum mechanics and thermo-
dynamics.
The Lagrangian equations also provide the foundation of a large variety
of finite element methods where the state of finite cells is described by
generalized coordinates as values of scalar and vector fields at grid verti-

ces, linear or quadratic interpolation providing values in the cells in terms
of these generalized coordinates. The corresponding Lagrangian equa-
tions for the discrete variables are then obtained directly without prior
knowledge of the field differential equations.
XXVII. Lagrangian Equations in Configuration

Space. Internal Relaxation,
Order-Disorder Phenomena, and
Quantum Kinetics
The concept of threshold minimum size of cells, as described above for

physical space, may be extended to subspaces in the abstract multidimen-
sional configurational thermodynamic space (Biot, 1982b). For example,
we may consider the translational and vibrational degrees of freedom of
gas molecules as constituting distinct subspaces with their own entropy
and temperature. The state of an assembly of such subsystems is then
determined by mass and energy fluence between them. Following the
same procedures as used for cells in physical space, completely general
Lagrangian equations of evolution may be obtained with exchanges repre-
sented by internal fluence coordinates. In particular, this approach is
applicable to internal relaxation effects in gases.
This procedure with Lagrangian equations and internal coordinates is
also implicit in the general thermodynamic theory of relaxation and hered-
ity in viscoelasticity (Biot, 1954).
The concept of subspace cells may be extended to quantum levels with
their own temperatures and entropy. The kinetics of exchanges obeyed by
fluence coordinates is then obtained from transition probabilities in quan-
tum kinetics and statistics.
The same procedure may also be used for order-disorder phenomena in
metal alloys where the order-disorder state is described by internal
fluence coordinates.
REFERENCES
Biot, M. A. (1954). Theory of stress-strain relations in anisotropic viscoelasticity and relax-

ation phenomena. J. Appl. Phys. 25, 1385-1391.
Biot, M. A. (1955). Variational principles in irreversible thermodynamics with application to
viscoelasticity. Phys. Reu. 97, 14.
Biot, M. A. (1956a). Variational and Lagrangian methods in viscoelasticity. In “Deforma-
90 M. A. Biot
tion and Flow of Solids” (IUTAM Colloquium, Madrid, 1955), pp. 251-263. Springer,
Berlin.
Biot, M. A. (1956b). Thermoelasticity and irreversible thermodynamics. J. Appt. Phys. 27,
240-253.
Biot, M. A. (1958). Linear thermodynamics, and the mechanics of solids, Proc. Third U.S.
Nut. Gong. Appl. Mech., pp. l-18. ASME, New York.
Biot, M. A. (1962). Generalized theory of acoustic propagation in porous dissipative media.
J. Acoust. Sot. Am., 34, 1254-1264.
Biot, M. A. (1963). Theory of stability and consolidation of a porous medium under initial
stress. J. Math Mech. 12, 521-542.
Biot, M. A. (1964). Theory of buckling of a porous slab and its thermoelastic analogy. Trans.
ASME, Ser. E 31, 194-198.
Biot, M. A. (1965a). Further development of the theory of internal buckling of multilayers.
Bull. Geol. Sot. Am. 76, 833-840.
Biot, M. A. (1965b). “Mechanics of Incremental Deformations.” Wiley, New York.
Biot, M. A. (1966). Three-dimensional gravity instability derived from two-dimensional
solutions. Geophysics 31, 153-166.
Biot, M. A. (1967). Rheological stability with couple stresses and its application to geologi-
cal folding. Proc. R. Sot. Ser. A, 298, 402-423.
Biot, M. A. (1970). “Variational principle in heat transfer.” Oxford Press.
Biot, M. A. (1972). Theory of finite deformations of porous solids, Indiana Univ. Math. J.
21, 597-620.
Biot, M. A. (1973a). Nonlinear thermoelasticity, irreversible thermodynamics and elastic
instability. Zndiana Univ. Math. J. 23, 310-335.
Biot, M. A. (1973b). Buckling and dynamics of multilayered and laminated plates under
initial stress. Znt. J. Solids Struct. 10, 419-451.
Biot, M. A. (1973~). Nonlinear and semilinear rheology of porous solids. J. Geophys. Res.
78, 4924-4937.
Biot, M. A. (1974). Thermoelastic buckling. An unstable thermodynamic equilibrium at
minimum entropy. Bull. Cl. Sci. Acad. R. Belg. 60, 116-140.
Biot, M. A. (1975). A virtual dissipation principle and Lagrangian equations in nonlinear
irreversible thermodynamics. Bull. Cl. Sci. Acad. R. Belg. 61, 6-30.
Biot, M. A. (1976a). New chemical thermodynamics of open systems. Thermobaric poten-
tial, a new concept, Bull. Cl. Sci. Acad, R. Belg. 62, 239-258; Erratum 62, 678.
Biot, M. A. (197613).Variational-Lagrangian irreversible thermodynamics of nonlinear ther-
morheology. Q. Appl. Math. 34, 213-248.
Biot, M. A. (1977a). New fundamental concepts and results in thermodynamics with chemi-
cal applications, Chem. Phys. 22, 183-198.
Biot, M. A. (1977b). Variational-Lagrangian thermodynamics of nonisothermal finite strain
mechanics of porous solids and thermomolecular diffusion. Znt. J. Solids Struct. 13,
579-597.
Biot, M. A. (1977~). Variational-Lagrangian irreversible thermodynamics of initially
stressed solids with thermomolecular diffusion and chemical reactions. J. Mech. Phys.
Solids 25, 289-307; Errata 26, 79.
Biot, M. A. (1978). Variational irreversible thermodynamics of heat and mass transfer in
porous solids: new concepts and methods. Q. Appl. Math. 36, 19-38.
Biot, M. A. (1979). New variational-Lagrangian thermodynamics of viscous fluid mixtures
with thermomolecular diffusion, Proc. R. Sot. London, Ser. A. 365, 467-494.
Biot, M. A. (1981). Generalized Lagrangian thermodynamics of thermorheology J. Thermal
Stresses 4, 293-320.
Biot, M. A. (1982a). Thermodynamic principle of virtual dissipation and the dynamics of

physical-chemical fluid mixtures including radiation pressure, Q. Appl. Math. 39, 517-
540.
Biot, M. A. (1982b). Generalized Lagrangian equations of nonlinear reaction-diffusion,
Chem. Whys. 66, 11-26.
Biot, M. A., and Ode, H. (1965). Theory of gravity instability with variable overburden and
compaction. Geophysics 30, 213-227.
Brillouin, L. (1930). “Les Statistiques Quantiques et Leurs Applications.” Presses Univer-
sitaires de France, Paris.
Chung, B. T. F., and Yeh, L. T. (1975). Solidification and melting of materials subject to
convection and radiation. J. Spacecr. Rockets 12, 329-333.
De Donder, T. (1936). “L’AffinitC.” Gauthier-Villars, Paris.
Fowler, R., and Guggenheim, E. A. (1952). “Statistical Thermodynamics.” Cambridge
Univ. Press, England.
Gibbs, J. W. (1906). “Thermodynamics I,” Longmans, London.
Glansdorff, P., and Prigogine, I. (1971). “Structure, Stabilite et Fluctuations.” Masson,
Paris.
Hatsopoulos, G. N., and Keenan, J. H. (1965). “Principles of General Thermodynamics.”
Wiley, New York.
Lamb, H. (1932). “Hydrodynamics.” Dover, New York.
Lardner, T. J. (1963). Biot’s variational principle in heat conduction. AZAA J. 1,196-206.
Lardner, T. J. (1967). Approximate solutions to phase change problems, AZAA J. 5, 2079-
2080.
Lonngren, K. E., and Hsuan, H. C. S. (1978). A consequence of the invariance of Biot’s
variational principle in thermal conduction. J. Math. Phys. 19, 357-358.
Meixner, J. (1941). Ziir Thermodynamik der Thermodiffusion. Ann. Phys. 39, 333-356.
Mindlin, R. D. (1961). On the equations of motion of piezoelectric crystals. In “Problems in
Continuum Mechanics,” pp. 282-290. Sot. Ind. Appl. Math. Philadelphia, Pennsyl-
vania.
Mindlin, R. D. (1974). Equations of high frequency vibrations of thermopiezoelectric crys-
tals. Znt. J. Solids Struct. 10,625-637.
Onsager, L. (1930). Reciprocal relations in irreversible processes I. Phys. Rev. 37,405-426.
Onsager, L. (1931). Reciprocal relations in irreversible processes II. Phys. Rev. 37, 2265-
2279.
Prasad, A., and Agrawal, H. C. (1972). Biot’s variational principle for a Stefan problem.
AZAA J. 10, 325-327.
Prasad, A., and Agrawal, H. C. (1974). Biot’s variational principle for aerodynamic ablating
melting solids. AZAA J. 12, 250-252.
Senf, L. (1981). A special case of diffusion with moving boundary. Znt. J. Heat Mass
Transfer 24, 1903-1905.
Sokolnikoff, I. S. (1951). “Tensor Analysis.” Wiley, New York.
Washizu, K. (1975). “Variational Methods in Elasticity and Plasticity.” Pergamon, Oxford.
Woods, L. C. (1981). On the local form of the second law of thermodynamics in continuum
mechanics. Q. Appl. Math. 39, 119-126.
Yeh, L. T., and Chung, B. T. F. (1977). Phase change in a radiating medium with variable
thermal properties. J. Spacecr. Rockets, 14, 178-182.

M.A. A Biot - Fluence

Uploaded by

Copyright:

Available Formats

M.A. A Biot - Fluence

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

M.A. A Biot - Fluence

Uploaded by

Copyright:

Available Formats

..- ,I.l.uIY .&. ‘._. l.yy ...yv . . . . &._.,_) . _1_.._1 _.

Royal Academy of Belgium, Brussels, Belgium

XX. Linear Thermodynamics and Dissipative Structures near Unstable

Our purpose here is to present a comprehensive view of a distinct

ity and to a rigorous and complete generalization of the Kirchhoff

then derive corresponding variational principles. This implies a prior

vant tp.*ms. The Lagrangian equations directly provide such a formulation

serious errors and stand in the way of intuitive understanding of physical

II. Restructured Thermodynamics of Open

We denote by p the total pressure acting on Cr, and Ek is given by

dE:k = dpktpk + T dgk, (2.4)

where pk is the density at the temperature T and pressure pk.

dh = hkT dv + 2 hiT dMk + C,, dT. (2.5)

We call htT the heat of mixing at constant temperature and volume;

=--1 =,lrl (2.7)

Elimination of X,,,,lau between these equations yields

Elimination of I~C,,I~IM~ between these equations yields

Taking into account the differential relation (2.4), we derive

Pk = Pk(% Mk, 0, (2.14)

III. New Chemical Thermodynamics

The concept of thermobaric transfer has provided a new approach to

hur = hT + ck vk h:r, (3.2a)

In the preceding section we showed how to evaluate the energy and

d% = -p dv + c vk dt PkT dri + h”$ d.$

dh = hkT dv + c h$(dMk + &‘kd[) + C,, dT. (3.7)

Q = Q(v, 6, Mk, T), 9 = Y(v, 5, Mk, T), (3.8)

functions of v, 8, Mk, and T, considered as state variables of the primary

state 5 = 0, Mk = 0, the increase of masses of the various substances in

Pk = Pktv, mk, T) (3.11) 5

are functions Of Only u, mk, and T.

Equation (3.12) generalizes Eq. (2.17) in the presence of a chemical

Ad4 = TdY. (3.14)

This corresponds to De Donder’s formula (1936) and shows that A is the

ds; + S,(O) . (3.16)

The integrals at the lower limit are evaluated by extrapolating gaseous

pk = ‘AT dEi - T j-;‘dF; + &k(O)- TS,#). (3.18)

The constants &k(O)and ~(0) are determined experimentally from chemi-

dmk = 2 vkp dtp. (3.21)

With an affinity A, for each reaction, Eq. (3.12) becomes

IV. Homogeneous Mixtures and Reformulation of

The new restructured thermodynamics by thermobaric transfer pro-

dQ = -p du + qeq dt + c Fk dmk + dh. (4.1)

d9 = Y,, dt + q Sk dmk + dhlT, (4.2)

where we have put

%‘q = i”T - c vks;‘, y,, = g-7

where hkT is the heat of mixing at constant pressure and temperature.

d%= -pdv+%_,d~+~%kdmk, (4.6)

dY = Y,, da$ + x Ykdmk, (4.7)

%k = Ek + hiT, Ypk = Sk + hkTIT. (4.8) *

!?+ = Y,qt + c Ykmk. (4. IO)

2 mk d%k = V dp, (4.12)

Hence we may also write

= c mk d(%k - TYk) + 2 m,@k dT. (4.14)

Using Eqs. (2.18) and (4.8) we also write

v dp = c mk d& + Yc dT, (4.16)

We call Yc the convective entropy (Biot, 1982a), because it represents the

Q = -pv + 2 %_lbtp+ T %kmk, (4.18)

tip=l = h”Tqp- T ,@zqp, Ygq = (heqplTeqp) - T ,+Seqp. (4.20)

The superscript p refers to values for the particular reaction p.

V. Nontensorial Virtual Work Approach to Finite

69 = - & 6Si + as, 6s” = &rM + &s*ch. (8.29)