Professional Documents
Culture Documents
Edited by
CRISTIAN DASCALU
University J. Fourier, Grenoble, France
GÉRARD A. MAUGIN
University Pierre et Marie Curie, Paris, France
and
CLAUDE STOLZ
Ecole Polytechnique, Palaiseau, France
Ó 2008 Springer
All Rights Reserved. No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical, including photocopying, recording or by any
information storage and retrieval system, without written permission from the copyright owner.
Springer.com
Table of Contents
Preface 1
C. Dascalu and G.A. Maugin
Reciprocity in fracture and defect mechanics 3–11
R. Kienzler
Configurational forces and gauge conditions in electromagnetic bodies 13–19
C. Trimarco
The anti-symmetry principle for quasi-static crack propagation in Mode III 21–33
G.E. Oleaga
Configurational balance and entropy sinks 35–43
M. Epstein
Application of invariant integrals to the problems of defect identification 45–54
R.V. Goldstein, E.I. Shifrin and P.S. Shushpannikov
On application of classical Eshelby approach to calculating effective elastic moduli of 55–66
dispersed composites
K.B. Ustinov and R.V. Goldstein
Material forces in finite elasto-plasticity with continuously distributed dislocations 67–81
S. Cleja-Ţigoiu
Distributed dislocation approach for cracks in couple-stress elasticity: shear modes 83–102
P.A. Gourgiotis and H.G. Georgiadis
Bifurcation of equilibrium solutions and defects nucleation 103–107
C. Stolz
Theoretical and numerical aspects of the material and spatial settings in nonlinear 109–116
electro-elastostatics
D.K. Vu and P. Steinmann
Energy-based r-adaptivity: a solution strategy and applications to fracture mechanics 117–132
M. Scherer, R. Denzer and P. Steinmann
Variational design sensitivity analysis in the context of structural optimization and 133–155
configurational mechanics
D. Materna and F.-J. Barthold
An anisotropic elastic formulation for configurational forces in stress space 157–161
A. Gupta and X. Markenscoff
Conservation laws, duality and symmetry loss in solid mechanics 163–172
H.D. Bui
Phase field simulation of domain structures in ferroelectric materials within the context of 173–180
inhomogeneity evolution
R. Müller, D. Gross, D. Schrade and B.X. Xu
An adaptive singular finite element in nonlinear fracture mechanics 181–190
R. Denzer, M. Scherer and P. Steinmann
Moving singularities in thermoelastic solids 191–198
A. Berezovski and G.A. Maugin
Dislocation tri-material solution in the analysis of bridged crack in anisotropic bimaterial 199–217
half-space
T. Profant, O. Ševec9ek, M. Kotoul and T. Vysloužil
Study of the simple extension tear test sample for rubber with Configurational Mechanics 219–225
E. Verron
Stress-driven diffusion in a deforming and evolving elastic circular tube of single 227–234
component solid with vacancies
C.H. Wu
Mode II intersonic crack propagation in poroelastic media 235–267
E. Radi and B. Loret
Material forces for crack analysis of functionally graded materials in adaptively refined 269–283
FE-meshes
R. Mahnken
A multiscale approach to damage configurational forces 285–294
C. Dascalu and G. Bilbie
Preface
C. Dascalu · G. A. Maugin
The volume presents recent developments in the This special issue aims at bringing together recent
theory of defects and the mechanics of material forces. developments in Material Mechanics and the more clas-
Most of the contributions were presented at the Inter- sical Defect Mechanics approaches. The contributions
national Symposium on Defect and Material Forces are highlighting recent research on topics like: fracture
(ISDMM2007), held in Aussois, France, March 25– and damage, electromagnetoelasticity, plasticity, dis-
29, 2007. tributed dislocations, thermodynamics, poroelasticity,
Originated in the works of Eshelby, the Material or generalized continua, structural optimization, conser-
Configurational Mechanics experienced a remarkable vation laws and symmetries, multiscale approaches and
revival over the last two decades. When the mechanics numerical solution strategies.
of continua is fully expressed on the material manifold, We expect the present volume to be a valuable
it captures the material inhomogeneities. The driving resource for researchers in the field of Mechanics of
(material) forces on inhomogeneities appear naturally Defects in Solids.
in this framework and are requesting for constitutive We dedicate this special issue to the memory of
modeling of the evolution of inhomogeneities through the late Professor George Herrmann (1921–2007).
kinetic laws. G. Herrmann was a prestigious scientist, well-known
In this way, a general scheme for describing struc- in the international mechanics community. In the last
tural changes in continua is obtained. The Eshelbian years, he was an active researcher in the field of Mate-
mechanics formulation comes up with a unifying treat- rial Mechanics. George Herrmann supported the orga-
ment of different phenomena like fracture and damage nization and registered for attending the ISDMM2007
evolution, phase transitions, plasticity and dislocation Symposium in Aussois, before his sudden dead on
motion, etc. January 7, 2007. None of us will forget his passion for
mechanical sciences, his enthusiasm and generosity.
C. Dascalu (B)
Laboratoire Sols Solides Structures, Université Joseph
Fourier, Grenoble, Domaine Universitaire, B.P. 53,
38041 Grenoble cedex 9, France
e-mail: cristian.dascalu@hmg.inpg.fr
G. A. Maugin
Université Pierre et Marie Curie, Institut Jean Le Rond
d’Alembert, Case 152, 4 place Jussieu,
75252 Paris cedex 05, France
e-mail: gam@ccr.jussieu.fr
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 1
doi: 10.1007/978-1-4020-6929-1_1, © Springer Science+Business Media B.V. 2008
Reciprocity in fracture and defect mechanics
R. Kienzler
Abstract For defects in solids, when displaced within The dot marks the scalar product between the two
the material, reciprocity relations have been established vectors. In double-indexed terms, the first index indi-
recently similar to the theorems attributed to Betti and cates the position at which the quantity is measured
Maxwell. These theorems are applied to crack- and (effect), and the second index indicates the cause due
defect-interaction problems. to which this quantity occurs.
To reach a scalar version of Betti’s theorem the dis-
Keywords Reciprocity · Fracture · Defect placement component of u12 in the direction of F1 is
interaction · Material forces P and the component of u in the direc-
introduced as u 12 21
P (cf. Fig. 1). With the magnitude F
tion of F2 as u 21 1
and F2 of F1 and F2 , respectively, it is
1 Introduction
P
F1 u 12 = F2 u 21
P
. (2)
When treating problems of linear elastic systems, such Since P
u 12
is proportional to F2 and P
u 21
is proportional
as beams, frames or two- and three-dimensional contin- to F1 influence coefficient may be defined as
uous elastic solids, the reciprocity theorems associated
with the names of Betti and Maxwell have proven to
P
u 12 = δ12 F2 , (3a)
be quite valuable. In its simplest form, Betti’s theorem P
u 21 = δ21 F1 , (3b)
states that if a linear elastic body is supported properly
and according to Marguerre (1962), Maxwell’s theo-
such that rigid body displacements are precluded and
rem states
if an external force F1 at point 1 which produces a dis-
placement u21 at some other point 2, then a force F2 δ12 = δ21 . (4)
at 2 would produce a displacement u12 at 1 where (cf.
The reciprocity relations are based on the result that the
e.g., Marguerre 1962)
energy stored in an elastic body after application of two
F1 · u12 = F2 · u21 . (1) forces is independent of their sequence of application,
The contents of the present paper has been developed together
and equals the external work done on the body. Various
with Prof. Dr. Dr. h. c. George Herrmann, Stanford University, applications of these theorems are to be found in , e.g.,
California, who passed away on January 7, 2007. Timoshenko and Goodier (1970), Barber (2002).
During the recent decades a new topic has emerged
R. Kienzler (B)
Department of Production Engineering,
in mechanics of elastically deformable media which
University of Bremen, Bremen, Germany is variously described as Defect Mechanics, Fracture
e-mail: rkienzler@uni-bremen.de Mechanics, Configurational Mechanics, Mechanics in
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 3
doi: 10.1007/978-1-4020-6929-1_2, © Springer Science+Business Media B.V. 2007
4 R. Kienzler
q
F1
F 1
F2
2
1 1
u12P 2
P
B12 2
u12 P
B12 P
u21 u21 B21 B21
N N
a1
a2
b
2 Interacting cracks
a1 a2
Interacting cracks have been the object of research over
several years (cf., e.g., Erdogan 1962; Panasyuk et al.
1977; Gross 1982). The material forces at crack tips are
a1 b a2
usually calculated from the path-independent J inte-
gral (Rice 1968). Reciprocity relations are, therefore, Fig. 4 Single crack (a), two neighboured cracks (b)
concerned with the change of the J integral due to the
translation of some other defect, e.g., the change of and loaded by an axial force in tension as depicted in
length of a crack 2 in the neighbourhood of the origi- Fig. 3.
nal crack 1. Usually, the solution of crack-interaction If a single crack is considered first, the key idea is
problems involve either some advances analytical tools that the energy-release rate is proportional to the length
or an extended numerical investigation. Based on the of the shaded strip in Fig. 4a, i.e.,
strength-of-materials theories, a simple first estimate √ √
for interacting edge cracks in an elastic bar under ten- G = J ∝ 2a 2; J = C2a 2. (9)
sion have been given recently by Rohde and Kienzler The constant C, say, would be proportional to the square
(2005) and Rohde et al. (2005), and the reciprocity rela- of the applied load, inverse proportional to Young’s
tions are applied within this simplified problem setting. modulus, and would combine some information about
In Rohde and Kienzler (2005) and Rohde et al. (2005) the geometry of the bar.
bars are investigated with a set of 2 × 2 edge cracks In the presence of a second crack, cf. Fig. 4b, both
symmetrically positioned with respect to its length axis strips can not develop completely due to shielding such
6 R. Kienzler
b x1
1
2 M
J1H
L
J1D
1
x1
J 2D
(ξi + λi ) − (ξi ) E ∗ b2 cos ϕ 1
lim = −Ji . (13) J1 = − + 1 + 2 sin ϕ , (14a)
2
λi →0 λi 4πr0 ε3 ε2 − 1
E ∗ b2 sin ϕ 1
J2 = − + 2 sin ϕ ,
2
(14b)
As usual, the energy-release rate is calculated by means 4πr0 ε3 ε2 − 1
of a path-independent contour integral involving the E ∗ b2 sin ϕ cos ϕ
energy-momentum tensor (cf., e.g., Maugin 1993; L=− , (14c)
4π ε2
Gurtin 2000; Kienzler and Herrmann 2000). The con-
E ∗ b2 1
tour of integration is arbitrary as long as the same de- M=+ + 1 + sin ϕ ,
2
(14d)
4π ε2 ε2 − 1
fect, either the hole or the inclusion, is included. In
addition, two further path-independent integrals have with E ∗ = E for plane stress, E ∗ = E/(1 − ν 2 ) for
been introduced (Günther 1962; Knowles and Stern- plane strain, Young’s modulus E and Poisson’s ratio
berg 1972; Budiansky and Rice 1973) designated as ν. As in Physical Space, free body diagrams can be
L and M integrals. The L-integral is a material vector sketched in Material Space. This is shown in Fig. 7,
moment (r × J) and calculates the energy-release rate where for the virial M, in the sketch a pressure-like
due to a rotation of the defect ϕ → ϕ + ω. It is path symbol and in the equation the symbol ⊗ is used.
independent if the material is isotropic. The M integral Likewise, material equilibrium conditions apply,
is a material scalar moment (r · J) and calculates the where the expanding-moment condition (“Fliehmo-
energy-release rate due to a self-similar expansion of ment”, cf. Schweins 1849) is normally not used nei-
the defect, here r0 → αr0 . L and M depend on the ther in physical nor in material space
choice of the point of reference. Choosing for L the → : J1H − J1D = 0 (15a)
origin of the coordinate system and for M the center
↑ : J2H− J2D = 0 (15b)
of the hole, the path-independent integrals have been
evaluated (Kienzler and Herrmann 2000; Kienzler and ∩ : L + ξ1 J2H − ξ2 J1H = 0 (15c)
Kordisch 1990) and are given as ⊗ : M + ξ1 J1D + ξ2 J2D = 0 (15d)
8 R. Kienzler
x2 • material translation λ1
1
E ∗ b2
λ1 J1 = +λ1
d d 4πr02 ε4
d
2 cos 2 ϕ 6 cos 2 ϕ − 1
× +
(ε − 1)
2 2 (ε2 − 1)
x1
−3(4 cos 4 ϕ − 6 cos2 ϕ + 1) (20a)
Fig. 8 Material translation λ1 of the hole
E ∗ b2 sin ϕ cos ϕ 1
λ1 J2 = +λ1
2πr0 ε2 4 (ε − 1)2
2
The change of material dynamic quantities Ji , L , M
3
due to the small material kinematic quantities λi , ω, α + 2 + 6 sin 2 ϕ (20b)
can be calculated by differentiation (ε − 1)
E ∗ b2 sin ϕ
∂( ) λ1 L = −λ1
λi ( ) = λi + 0 λi2 (16) 4πr0 ε 3
∂λi
and by geometrical considerations. × 1 − 4 cos 2 ϕ (20c)
From Fig. 8, e.g., we read off
E ∗ b2 cos ϕ
λ1 M = −λ1
d̄ cos(ϕ − dϕ) = d cos ϕ + λ1 , (17a) 2πr0 ε 3
d̄ sin(ϕ − dϕ) = d sin ϕ (17b) ε4
× + 2 sin 2 ϕ (20d)
(ε2 − 1)2
and after linearization (dϕ 1, λ1 d) we find
λ1
ε → ε̄ = ε + cos ϕ (18a) • material translation λ2
r0
λ1 sin ϕ
ϕ → ϕ̄ = ϕ − . (18b)
r0 ε E ∗ b2 sin ϕ cos ϕ 1
λ2 J1 = +λ2
Inserting in (16) for i = 1 and applying the chain role 2πr02 ε4 (ε2 − 1)2
leads to
3
+ + 6 sin 2 ϕ (21a)
∂( ) cos ϕ ∂( ) sin ϕ (ε2 − 1)
λ1 ( ) = − λ1 , (19a)
∂ε r0 ∂ϕ r0 ε E ∗ b2
λ2 J2 = +λ2
and in a similar way we obtain 4πr02 ε4
∂( ) sin ϕ ∂( ) cos ϕ 2 sin 2 ϕ 6 sin 2 ϕ − 1
λ2 ( ) = + λ2 . (19b) × +
∂ε r0 ∂ϕ r0 ε (ε − 1)
2 2 ε2 − 1
Due to a rotation of the hole around the dislocation, −6 sin 2 ϕ(1 − 2 sin2 ϕ) (21b)
ϕ → ϕ + ω, and a self-similar expansion of the hole,
r0 → αr0 we find E ∗ b2 cos ϕ
λ2 L = −λ2
∂( ) 4πr0 ε 3
ω ( ) = ω, (19c)
∂ϕ × 1 − 4 sin 2 ϕ (21c)
∂( ) E ∗ b2 sin ϕ
α ( ) = ε α. (19d) λ2 M = −λ2
∂ε 2πr0 ε 3
Using Eq. 19, the change of the relevant quantities can ε4
× + 1 − 2 cos 2
ϕ (21d)
easily be calculated with the results (ε2 − 1)2
Reciprocity in fracture and defect mechanics 9
• material rotation ω J2 J2
E ∗ b2 sin ϕ
ω J1 = +ω
4πr0 ε3 x2
J1 J1
1
× 2 + 3(1 − 2 cos 2 ϕ) (22a)
ε −1
2 r0
E ∗ b2 cos ϕ
ω J2 = +ω
4πr0 ε 3
d
1
× − 2 − 6 sin 2 ϕ (22b)
ε −1
x1
E b2
∗ L L 1
ω L = −ω 1 − 2 cos 2 ϕ
4π ε 2
Fig. 9 Free-body diagram after material rotation ω
E ∗ b2
= +ω cos 2ϕ (22c)
4π ε2
E ∗ b2 Due to (15c) and neglection of terms of 0(ω2 ) we arrive
ω M = +ω sin ϕ cos ϕ at
2π ε2
E ∗ b2 ω L + r0 ε(cos ϕω J2 − ω sin ϕ J2
=ω sin 2ϕ (22d)
4π ε2 − sin ϕω J1 − ω cos ϕ J1 ) = 0. (25c)
• material self-similar expansion α
In a similar way we obtain
E ∗ b2 cos ϕ
α J1 = α λ1 L + ξ1 λ1 J2 + λ1 J2 − ξ2 λ1 J1 = 0, (25a)
2πr0 ε3
ε4
× − 2 − 2 sin ϕ
2
(23a)
(ε − 1)2 λ2 L + ξ1 λ2 J2 − λ2 J1 − ξ2 λ2 J1 = 0, (25b)
E ∗ b2 sin ϕ
α J2 = α
2πr0 ε 3
α L + r0 ε(cos ϕα J2 − sin ϕα J1 = 0. (25d)
ε4
× − 2 − 1 + 2 cos 2
ϕ (23b) The virial equations are
(ε − 1)2
E ∗ b2 λ1 M + λ1 J1 + ξ1 λ1 J1 + ξ2 λ1 J2 = 0, (26a)
α L = −α sin ϕ cos ϕ
2π ε2 λ2 M + λ2 J2 + ξ1 λ2 J1 + ξ2 λ2 J2 = 0, (26b)
E ∗ b2 ω M + r0 ε(cos ϕω J1 − ω sin ϕ J1
= −α sin 2ϕ (23c)
4π ε2
+ sin ϕω J2 + ω cos ϕ J2 ) = 0, (26c)
E ∗ b2
α M = +α α M + r0 ε(cos ϕα J1 + sin ϕα J2 ) = 0. (26d)
2π ε2
ε4 On introducing (20)–(23) into (25) and (26) it is
× + sin ϕ .
2
(23d)
(ε2 − 1)2 observed that the material equilibrium conditions are
satisfied identically.
Also the modified material quantities have to satisfy
Let us now turn our attention to reciprocity. As a first
material equilibrium conditions. Especially for the vec-
application we consider two material displacements λ1
torial and the scalar moments the change of the lever
and λ2 of the circular hole. The reciprocity theorem
arms due to the material displacements have to be ob-
states that the work of the change of J2 due to the mate-
served. As an example consider the material rotation
rial translation λ1 in the material translation λ2 is equal
depicted in Fig. 9
to the work of the change of the material force J1 due
Equilibrium of moments yields
to a material translation λ2 in the material translation
L + ω L + (ξ1 − ξ2 ω)(J2 + ω J2 ) λ1 . Thus
−(ξ2 + ξ1 ω)(J1 + ω J1 ) = 0. (24) λ2 λ1 J2 = λ1 λ2 J1 . (27a)
10 R. Kienzler
and calculate the change in J1 each time. Instead, by Erdogan F (1962) On the stress distribution in plates with collin-
using (27c), we translate the crack tip by an amount λ1 ear cuts under arbitrary loads. In: Proceedings of the Fourth
U.S. National Congress of Applied Mechanics. ASME
and calculate the change in M of void i, λ1 Mi , due to 1:547–553
this translation and we have Gross D (1982) Spannungsintensitätsfaktoren von Rißsystemen.
αi Ing Arch 51:301–310
αi J1 = λ1 Mi . (35)
λ1 Günther W (1962) Über einige Randintegrale der Elastomecha-
nik. Abh Braunschw Wiss Ges 14:53–72
Note that the change of J1 due to the growth of any Gurson AL (1977) Continuum theory of ductile rupture by void
void is obtainable from only one remesh. In this way nucleation and growth. Part I—yield criteria and flow rules
it is straight forward to construct influence surfaces for for porous ductile media. J Eng Mater Tech 99:2–15
Gurtin ME (2000) Configurational forces as basic concepts of
the J integral to assess the risk of voids in the neigh- continuum physics. Springer, New York
bourhood of the crack. Of course, the voids could have Herrmann G, Kienzler R (2007a) Reciprocity relations in Eshel-
different forms, they could also be cracks. bian mechanics. Mech Res Commun 34:338–343
Herrmann G, Kienzler R (2007b) A reciprocity relation couples
Newtonian and Eshelbian Mechanics. ASME J Appl Mech
(in press)
4 Conclusion Kienzler R, Herrmann G (2000) Mechanics in material space.
Springer, Berlin
Reciprocity theorems in Material Space have been Kienzler R, Herrmann G (2007) Nonlinear and linearized rec-
iprocity relations in structural configurational mechanics.
applied to problems of crack and defect interactions.
Acta Mech, doi:10.1007/s00707-007-0457-5
Whereas one part of the paper in devoted to the valida- Kienzler R, Kordisch H (1990) Calculation of J1 and J2 , using
tion of various versions of the theorems, another part the L and M integral. Int J Fract 43:213–225
and forthcoming research is concerned with the explo- Knowles JK, Sternberg E (1972) On a class of conservation laws
in linearized and finite elastostatics. Arch Rat Mech Anal
ration of useful applications of reciprocity relations.
44:187–211
One aspect is the construction of influence surfaces Marguerre K (1962) Elasticity, basic concepts. In: Flügge W
to assess the effect of damage (in the widest sense) Handbook of engineering mechanics. McGraw-Hill, New
on a main crack or defect. As in Physical Space, rec- York,
Maugin GA (1993) Material inhomogeneities in elasticity.
iprocity relations might be the basis for the establish-
Chapman & Hall, London
ment of a boundary-integral method in Material Space. Panasyuk MP, Savruk MP, Datsyshyn AP (1977) A general
The generalization to continuously distributed defects, method of solution of two-dimensional problems in the the-
material forces per unit of volume and material transla- ory of cracks. Eng Fract Mech 9:481–497
Rice JR (1968) A path independent integral and the approxi-
tion fields as λi = λi (x j ) (i, j = 1, 2, 3) would result
mate analysis of strain concentration by notches and cracks.
in reciprocity relations involving surface- and volume- ASME J Appl Mech 27:379–386
integral expressions with probably useful and far- Rohde L, Kienzler R (2005) Numerical computation and ana-
reaching applications. lytical estimation of stress-intensity factors for strips with
multiple edge cracks. Arch Appl Mech 74:846–852
Acknowledgements The support of G. Herrmann and Rohde L, Kienzler R, Herrmann G (2005) On a new method for
R. Kienzler by the Research in Pairs Programme of the Mathe- calculating stress-intensity factors of multiple edge cracks.
matisches Forschungsinstitut Oberwolfach is gratefully acknowl- Phil Mag 85:4231–4244
edged. Thanks are also due to Dipl.-Ing. R. Schröder for carefully Schweins G (1849) Fliehmomente oder die Summe (x X + yY )
drawing the figures. bei Kräften in der Ebene und (x X + yY + z Z ) bei Kräften
im Raume. Crelles J Reine Angew Math 38:77–88
Timoshenko SP, Goodier JN (1970) Theory of elasticity, 3rd edn.
McGraw-Hill, New York
References
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 13
doi: 10.1007/978-1-4020-6929-1_3, © Springer Science+Business Media B.V. 2007
14 C. Trimarco
unavoidably coupled with the magnetic induction The gauge transformations for the Lagrangian
through the vector potential. Due to the indetermi- electromagnetic potentials are also explored and
nacy of the scalar and vector potentials, additional by invariance arguments some interesting results
conditions, which are known as gauge conditions, can be shown in evidence. As an example of the
need to be properly established. role played by the vector potential in the configu-
As mentioned, a preliminary issue in deformable rational framework, the case of a superconducting
solids is to write the whole set of the mechan- material is discussed and a link between the super-
ical and electromagnetic equations. These equa- current and material momentum of configurational
tions can be shown to lead to new equations, which mechanics is pointed out.
are instructive and enlightening in the presence
of material defects or in crack propagation prob-
lems. In fact, the new equations, known as con- 2 The Lagrangian electromagnetic potentials
figurational balance laws, though mere identities
along the motion, can be viewed as additional bal- The Lagrangian scalar and vector electromagnetic
ance laws that govern the behaviour of defects or potentials φ and A, respectively, are defined
any other material inhomogeneity (Maugin and through the equations
Trimarco 1992a, b, Gurtin 2000, Kienzler and Curl A = B (1)
•
Herrmann 2000, Trimarco and Maugin 2001). The −A − Grad φ = E, (2)
notion of Eshelby stress tensor and that of material
where B and E are the magnetic induction and
momentum arises in a natural way in this context.
the electric field, respectively, in the Lagrangian
However, the configurational balance laws can
form. Curl and Grad denote the differential oper-
be alternatively perceived as primary laws when it
ators of curl and gradient, respectively, in the refer-
is possible to appeal to a nonclassical variational
ence configuration V of an electromagnetic body,
approach, which is based on the inverse motion
whose current configuration is denoted by V. V and
and inverse deformation (Maugin and Trimarco
V are here assumed to be simply connected open
1992a, Maugin 1993, Trimarco and Maugin 2001).
subsets of the Euclidean space E3 . If x ∈ V and
This approach can be extended to electromagnetic
X ∈ V, x = χ (X, t) represents the motion, t ∈ R
bodies. In this context, the lack of uniqueness of
being the time. F denotes the deformation gradient
the vector potential affects the unique definition of
and v = x• ≡; dtd
[χ (X, t)] the velocity. As usual, det
the Eshelby stress and that of material momentum,
F ≡; J > 0 (Truesdell and Noll 1965). With ref-
which are derived as canonical quantities from
erence to the Eq. 2, A• ≡ dA dt ≡ ∂[A(X, t)]/∂t|X
a suitable expression of the Lagrangian density.
denotes the material time derivative of A. The
Differently, the physical electromagnetic stress (a
potentials φ and A are related to the classical Euler-
Cauchy-like stress) and momentum do not depend
ian electromagnetic potentials φ and a in V, as fol-
explicitly upon the electromagnetic potentials and
lows:
are thus uniquely defined (Trimarco 2007a).
The set of transformations (the gauge transfor- A = FT a (3)
mations) for the electromagnetic potentials, under φ = ϕ − v • a. (4)
which the electromagnetic Maxwell fields are FT denotes the transpose of the deformation gra-
invariant, are re-proposed here in the Lagrangian dient. It is worth recalling that
form and their effect on the physical quantities of
curl a = b (5)
interest is examined. The treatment parallels the
classical approach to some extent. However, the −∂a/∂t ≡ −a,t = e + grad ϕ, (6)
gauge conditions for the material potentials can- b and e being the classical Eulerian magnetic induc-
not be a formal replica of the classical Coulomb tion and electric field, respectively. (∂a/∂t) denotes
or Lorenz-Lorentz gauge conditions of electro- the Eulerian time derivative of a (Becker 1964,
dynamics (Jackson 2002, Trimarco 2007a, b). The Jackson 1962). Note that the space and time differ-
problem has not an obvious solution and seems to ential operators that appear in the Eqs. 5 and 6 act
represent an open issue in deformable solids. in the current configuration V.
Configurational forces and gauge conditions 15
represent the material momentum and the mater- differently form the momentum (21). In the expres-
ial configurational stress, respectively. C ≡ FT F is sion (22) ρ is the mass density in V. It is worth
the right Cauchy-Green deformation tensor. E, B remarking that the momentum (21) can be recov-
and D are the electric field, the magnetic and the ered from the momentum density (22) by volume
electric induction, respectively, in the Lagrangian integration, in the linear approximation of the sec-
form. Do ≡ D − P ≡ Jεo C−1 (E + V × B), εo being ond order term (e × b), having assumed that b
the dielectric constant of a vacuum. coincides with a given external magnetic induction
With reference to the Eqs. 17 and 18, the canon- and that a is a divergence-free field (transverse or
ical quantities (∂L/∂V) and (∂L/∂F−1 ) depend on Coulomb gauge condition).
the (Grad A) and on A• explicitly. It is not difficult A second instructive example is provided by
to see that these quantities are not gauge invariant the London equation (1950) for superconductors.
with none of the introduced transformations. By This equation is the simplest equation that gov-
contrast, it can be shown that the Eq. 16 does enjoy erns the electric current in a superconductor and
the gauge invariance property with respect to the that accounts for the Meixner effect (b identically
transformations (9)–(10) and (13)–(14). A unique vanishing in the bulk of a superconductor), one
definition of these canonical quantities addresses of the most peculiar features of superconducting
gauge conditions for the Lagrangian electromag- materials (Kittel 1986). The London equation in
netic potentials (Trimarco 2007b). S.I. units reads
js = −[(µo λ2 )−1 ]a. (23)
4 Gauge dependence and gauge invariance js is the supercurrent density, µo the magnetic per-
meability of a vacuum and λ a phenomenological
Most of the quantities and of the equations of inter- positive constant known as the London penetra-
est in electromagnetic bodies are invariant with tion depth. In order to uniquely define the super-
respect to gauge transformations, which have been current, a restriction on a is needed. Similarly to
introduced in the previous section. For instance, the first example, this restriction is provided by the
the Maxwell equations in the Lagrangian form aforementioned gauge condition
are invariant under the transformations (9) and
div a = 0. (24)
(10). So are the Lagrangian electromagnetic fields,
the electromagnetic stress and momentum (Nelson Note that the quantity (curl js ) is gauge invari-
1979, Trimarco 2001). However, there are quanti- ant, differently from js in the Eq. 23. In fact, by
ties that clearly depend on the choice of the electro- taking into account the Eq. 5, the Eq. 23 can be
magnetic potentials as in the case of the canonical also written in gauge invariant form as
quantities (17) and (18).
curl js = −[(µo λ2 )−1 ]b. (25)
A simple example of gauge dependent quanti-
ties in classical electromagnetism is the momentum Superconductivity is a genuine quantum mech-
of a charged particle, which reads (Goldstein 1980, anical effect and a gauge invariant form of js can be
Becker 1966, Landau and Lifschitz 1974) found in a natural way in this context (Kittel 1986,
Blatt 1964, von Lau 1952, Fröhlich 1966, Jones and
p = mv + qa, (21)
March 1973). However, the phenomenon can be
where m and q represent the mass and the charge satisfactorily described in the framework of contin-
of the particle of interest, respectively. It is worth uum mechanics. In this framework and in the sim-
noting that the momentum density for an elec- ple case of type 1 superconductors, the following
tromagnetic body can be written as (Nelson 1979, simple model can be proposed. The supercurrent
Schoeller and Tellung 1992) is assumed to be related to the motion of electric
supercharges, in analogy with the idea that the elec-
pd = ρv + εo e × b. (22)
tric current is related with the velocity of the free
This momentum density is clearly invariant with electrons in a metal. Specific assumptions on the
respect to the gauge transformations (7) and (8), density ρs of the supercharges (which are not single
Configurational forces and gauge conditions 17
electrons, but rather group of electrons: in fact the Under conditions that are specified below, the
‘electron-pairs’ of the Bardeen–Cooper–Schrieffer density ρs , the velocity v and the supercurrent js
theory) and on their velocity v can be proposed. turn out to satisfy the following equations:
With the additional assumption that the supercur-
ρs (∂v/∂t) + ρs (grad v)v = −grad p + f, (29)
rent density js is proportional to the momentum
density and specifically to both, velocity and mass div js + (∂κρs /∂t) = 0. (30)
density, one is lead to gauge invariant equations The Eq. 29 represents the balance of momentum
that are reminder of the behaviour of Eulerian flu- of an Eulerian fluid in the presence of a pressure
ids (Fröhlich 1966). term p and a force f of electromagnetic origin. The
The treatment is consistent with the Ginzburg– Eq. 30 represents the conservation of the supercur-
Landau theory, according to which a free energy rent, which in terms of ψ and a reads
density of a superconductor depends upon a macro-
scopic complex-valued function ψ, this playing the js = (u/2i)[ψ̄(grad ψ) − ψ(grad ψ̄)] − κψ ψ̄a. (31)
role of an order parameter in the transition This expression can be evaluated by taking into
between the thermodynamical phases of a super- account the Eqs. 26, 27 and 28. Note that js is
conductor. This assumption is consistent with the invariant with respect to the gauge transformations
microscopic statistical behaviour of the super- (11) and (12), differently from the case expounded
charges, which behave as indistinguishable parti- in the previous section. It is also invariant with
cles that act as ‘bosons’, contrary to ordinary elec- respect to the gauge transformations (7) and (8),
trons that act as ‘fermions’ (Fröhlich 1966, Kittel provided ψ transforms as follows (Aharonov and
1986, Jones March 1973). As a result, a single func- Bohm 1959, Jackson 2002):
tion ψ describes the behaviour of all supercharges
(Fröhlich 1966). Accordingly, the density ρs is asso- ψ ∗ = ψ exp(ig κ/u). (32)
ciated with ψ as follows: It is worth remarking that the transformation
ψ(x, t) = (ρs ) 1/2
exp[iθ/u], (26) (32) only affects the phase of ψand thus does not
affect the density ρs , in accordance with the for-
so that ψ ψ̄ = ρs , ψ̄ being the complex conjugate mula (26). The gauge invariant Eqs. 29 and 30 hold
of ψ. Hereafter, it is understood that ρs = ρs (x, t) true provided ψ satisfies the equation
and that θ = θ (x, t), whereas u is a constant whose
(1/i)(∂ψ/∂t) = (1/2)u[(grad) − i(κ/u)a]2 ψ
physical dimensions are defined by the following
assumption on the velocity v: +cψ − d(ψ ψ̄)ψ, (33)
leads to the stationary case of Eq. 33. Invariance electromagnetic field is disregarded in the formula
with respect to a leads to the equation (19), one can write p = ρ0 CV ≡ −ρ0 FT v. Consis-
tently with this remark, the operator (∇ A ) should
curl (curl a) = µo (∂F/∂a). (35)
be related to the material momentum rather than
In this equation the quantity (∂F/∂a) plays the to the momentum or to the velocity. It is worth
role of a current density, in fact the supercurrent recalling that the dependence through (∇ A
) is
such as defined by the formula (31). typically quadratic as it accounts for the contribu-
tion of the kinetic energy density of particles in the
microscopic framework (Becker and Sauter 1964,
5 The Lagrangian electromagnetic potentials Landau and Lifschitz 1974).
in deformable superconductors By taking into account the Eqs. 28 and 38, the
Lagrangian supercurrent reads
In a deformable electromagnetic body the
J s = κJρs FT v, (39)
Lagrangian potentials are more appropriate than
the Eulerian ones, although they are related to one or, equivalently,
another through the formulas (3) and (4). Accord-
J s = JFT js . (40)
ingly, the gauge transformations that are inherent
deformable solids are those expressed by the for- Due to the explicit dependence of J s upon the
mulas (9)–(10) and (13)–(14). Also, according to deformation gradient, it is unlikely that the super-
the principle of material indifference (Truesdell current can be treated in this case like an Eulerian
and Noll 1965), the free energy density F that fluid as it was shown in the previous section for
generalises the Ginzburg-Landau expression (34) undeformed bodies, unless the body undergoes
should depend on (∇ A
) and on (Curl A), rather specific time-independent homogeneous deforma-
than on (∇ a ψ) and on (curl a), where tions. With reference to the Eq. 39 or 40, it is
worth remarking that supercurrent transforms in
= (J)1/2 ψ (36)
the material frame differently than ordinary elec-
and
¯ = Jρs . With reference to the formulas tric free-current i. In fact, the Lagrangian electric
(3), (11) and (13), the following relationship can current density reads
be recovered
j = JF−1 (i − ρe v), (41)
T
∇A = F ∇a , (37)
where ρe denotes the ordinary electric free
provided that β = iα. charge. The electric current j enters the inherent
Invariance arguments like those expounded in Maxwell equation in the Lagrangian form, which
Sect. 4 can be re-proposed for the free energy reads
density in the reference configuration F = F(
,
Curl H − (d/dt)D = j, (42)
(∇ A
), (Curl A), . . .) in order to derive the equa-
tion that govern
and the Lagrangian supercur- where H represents the Lagrangian magnetic field.
rent J s (Maugin 1992). The whole set of the Maxwell equations in the
Re-examination of the formulas (27) and (28) Lagrangian form can be found in Nelson (1979)
in the light of the transformation (37) suggests the and in Maugin and Trimarco (1991), though with
following transformation for v through the defor- different notation.
mation gradient:
FT v = Grad − κA, (38)
6 Final comments
being the Lagrangian velocity potential. The
transformation (38) is enlightening as the quan- Variational methods in configurational mechanics
tity FT v represents the mechanical part of material introduce the notion of material stress (the Eshelby
momentum such as introduced through the Eq. 19. stress) and material momentum. As is known, vari-
In fact, if V ≡ −F−1 v and if the contribution of the ational methods also address the notion of
Configurational forces and gauge conditions 19
energy-release rate of fracture mechanics (Maugin Kittel C (1986) Introduction to solid state physics, 6th edn.
and Trimarco 1992b, Gurtin 2000, Kienzler and John Wiley & Sons, New York
Landau L, Lifschitz E (1974) Méchanique quantique, 3rd
Herrmann 2000). The material stress and momen- edn. MIR, Moscou
tum, if derived as canonical quantities from the Laue von M (1952) Theory of superconductivity. Academic
Lagrangian (15), are affected by the indeterminacy Press, New York
that is introduced by the Lagrangian electromag- London F (1950) Superfluids: macroscopic theory of super-
conductivity, vol 1. Wiley, New York
netic potentials. However, this indeterminacy can Maugin GA (1992) Irreversible thermodynamics of
be solved by removing the gauge dependent terms deformable superconductors. CR Acad Sci Paris t 314
that appear in the formulas (17) and (18), as these Série II:889–894
terms balance each other identically in the Euler- Maugin GA (1993) Material inhomogeneities in elasticity
series applied mathematics and mathematical compu-
Lagrange equation (Eq. 16). In the most general tation, vol 3. Chapman and Hall, London
case, gauge invariant quantities are desirable in Maugin GA, Trimarco C (1991) Pseudomomentum and
order to circumvent the problem of introducing material forces in electromagnetic solids. Int J Electro-
gauge conditions. Here, the interest for supercon- magnet Mater 2:207–216
Maugin GA, Trimarco C (1992a) Note on a mixed varia-
ductors is twofold: first, for the crucial role that the tional principle in finite elasticity. Rend Mat Acc Naz
vector potential plays in these materials. Second, dei Lincei, Serie IX, v. III:69–74
for that a mechanical model for the electric cur- Maugin GA, Trimarco C (1992b) Pseudomomentum and
rent in deformable superconductors addresses the material forces in nonlinear elasticity. Acta Mechanica
94:1–28
material momentum rather than the momentum Maugin GA, Trimarco C (2001) Elements of field theory
or the velocity and thus addresses configurational in inhomogeneous and defective materials. Configura-
mechanics. tional mechanics of materials, CISM courses and lec-
tures, n. 427. Springer Verlag, Wien, pp 55–128
Nelson DF (1979) Electric, optic and acoustic interactions
in dielectrics. John Wiley, New York
References Schoeller H, Thellung A (1992) Lagrangian formalism and
conservation law for electrodynamics in nonlinear elas-
Aharonov Y, Bohm D (1959) Significance of electromag- tic dielectrics. Ann Phys 220:18–39
netic potentials in the quantum theory. Phys Rev 2d Trimarco C (2001) A Lagrangian approach to electromag-
Series 115(3):485–491 netic bodies. Technische Mechanik 22(3):175–180
Becker R, Sauter F (1964) Electromagnetic interactions, Trimarco C (2005a) On the material energy–momentum
vol 2. Blackie & Sons, London tensor in electrostatics and in magnetostatics.In: Stein-
Blatt JM (1964) Theory of superconductivity. Academic man P, Maugin GA (eds) Advances in mechanics and
Press, London mathematics, 11, mechanics of material forces. Springer
Fröhlich H (1966) Macroscopic wave functions in super- New York, Cptr 16, 161–171
conductors. Proc Phys Soc 87:330–332 Trimarco C (2005b) The total kinetic energy of an electro-
Goldstein H (1980) Classical mechanics, 12th edn. magnetic body. Phil Mag 85(33–35):4277–4287
Addison-Wesley, Reading Mass Trimarco C (2007a) Material electromagnetic fields and
Gurtin ME (2000) Configurational forces as basic concepts material forces. Arch Appl Mech 77:177–184 [Online
of continuum physics. Springer Verlag, New York http://maecourses.ucsd.edu/symi/other-papers.html]
Jackson JD (1962) Classical electrodynamics. J Wiley & Trimarco C (2007b) Configurational forces in dynamics
Sons, New York and electrodynamics. Proc Estonian Acad Phys Math
Jackson JD (2002) From Lorenz to Coulomb and other 56(2):116–125
explicit gauge transformations. Am J Phys 70(9):917– Trimarco C, Maugin GA (2001) Material mechanics of elec-
928 tromagnetic bodies. Configurational mechanics of mate-
Jones W, March NH (1973) Theoretical solid state physics, rials, CISM courses and lectures, n. 427, Springer Verlag,
vol 2. Wiley-Interscience, Wiley and Sons Ltd., New Wien, pp 129–172
York. [(1985) unabridged re-publication by Dover Pub- Truesdell CA, Noll W (1965) The nonlinear field the-
lications Inc., New York] ory of mechanics. Handbuch der Physik, Bd.III/3,
Kienzler R, Herrmann G (2000) Mechanics in material S. Flügge(ed). Springer Verlag, Berlin
space. Springer-Verlag, Berlin
The anti-symmetry principle for quasi-static crack
propagation in Mode III
Gerardo E. Oleaga
Abstract In this note we study a basic propagation lem is intimately related to that of finding a suitable
criterion for quasi-static crack evolution in Mode III. propagation criterion. There is a huge amount of liter-
Using classical techniques of complex analysis, the ature about this subject, and we shortly review a few
assumption of stable growth is expressed in terms of previous contributions to establish a reference for the
the parameters defining the elastic field around the tip. forthcoming discussion.
We explore the consequences of the local condition To our knowledge, one of the most widely accepted
obtained and analyse its role as a crack propagation law. criteria is the so-called principle of local symmetry.
In particular, we herein extend to bounded domains a It was proposed by Goldstein and Salganik in (1974)
number of results previously obtained for the whole and later analysed by Cotterell and Rice in (1980). For
plane. quasi-static propagation in a two dimensional in-plane
elastic field, it can be formulated as follows: the crack
Keywords Anti-symmetry principle · Quasistatic grows along the path that cancels the Mode II stress
crack growth · Linear elasticity · Brittle material · intensity factor, so that:
Anti-plane fields · Configurational forces · Crack
propagation law K II = 0. (1)
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 21
doi: 10.1007/978-1-4020-6929-1_4, © Springer Science+Business Media B.V. 2007
22 G. E. Oleaga
mathematical formulation of the free boundary prob- crack, that is Q(t; α) = κl(t). Therefore, the limit in
lem, since we have two scalar equations to find the (5) is written as:
evolution of a plane curve. Nevertheless, it is not phys-
E(t; α) 1
ically complete, because we are imposing the artificial L(α) = lim − =: G(α), (6)
t→0 κl(t) κ
constraint of path smoothness.
Another well-known local criterion is the maximum where G(α) is the energy release rate. In order to check
energy release rate principle. In our view, it is the most that the critical growth condition is satisfied, we must
straightforward way to approach the problem from ensure first that there are no possible unstable direc-
basic physical principles. Consider a family of vir- tions along other paths. This leads us to the following
tual extensions γα (t) of the actual crack configuration. inequality:
In this setting, t ≥ 0 is a parametrisation of the ex- G(α) ≤ κ. (7)
tended curve, α defines the specific curve of the family
and γα (t) represents the tip of the extended crack at Then, if a crack propagates under a critical or quasi-
“time” t. Let E(t; α) denote the amount of mechan- static regime, the growth directions are defined as fol-
ical energy released along this virtual extension and lows:
Q(t; α) the amount of dissipation involved (i.e., mi- α ∗ = crack growth direction ⇒ G(α ∗ ) = max G(α).
α
nus the work that must be done to break the atom bonds
on crack faces). We can say that the evolution along the (8)
path α := {γα (t), t ≥ 0} is possible if the following This is the so-called maximum energy release rate cri-
inequality holds for some small enough δ > 0: terion.
−E(t; α) ≥ Q(t; α) ∀ t ≤ δ. (3) In this note we start by reviewing the role of the
configurational force (6) as a basic object to determine
If we have a strict inequality in (3), the crack evolu- the path of a quasi-static growing crack in Mode III. In
tion turns to be unstable, and other physical ingredients Sect. 2 we show that this concept alone is not enough to
should enter the picture, for instance the kinetic energy define the shape evolution in an out of plane setting and
flux. On the other hand, we say that a crack configura- a more precise study of E should be done. In Sects. 3
tion is stable if the following condition holds: For all and 4 we resume the picture presented in our previous
virtual extensions α , there exists some δ > 0 such work (Oleaga 2004, 2006), where a basic condition for
that: crack stability called the anti-symmetry principle was
−E(t; α) ≤ Q(t; α) for all t < δ. (4) obtained for an unbounded domain, and give some fur-
Consider now two cases of this inequality. For a finite ther insight about the kind of crack propagation law
extension, Q(t; α) is always greater than zero (we obtained. In Sect. 5 we analyse the case of a bounded
have to do negative mechanical work to break the bonds domain, not considered in Oleaga (2004, 2006), and
on crack faces). On the other hand, E is always neg- the main differences with the former one are pointed
ative (elastic energy decreases with crack growth). We out.
consider the following limit for a fixed α:
E(t; α) 2 The role of the driving force in crack direction
L(α) := lim − . (5)
t→0 Q(t; α)
If L(α) is strictly greater than 1, according to (3) unsta- The discussion in the previous Section indicates that
ble growth along this path will take place. On the other the function G(α) in (6) has to be computed in the
hand, if it is strictly lower than 1 we say that crack first place. This task was performed by several authors
growth along this way is not possible. According to and it is worth to review some of these works here.
this, we say that the curve evolves in a quasi-static Let us parametrise the crack growing direction by the
or critical regime if it satisfies the stability condition number α := ϕ/π , where ϕ is the kinking angle with
(4) and at each configuration there exists some α with respect to the initial crack. Consider first some the-
L(α) = 1 (critical growth). oretical formulae for this quantity. In the analysis of
When Griffith’s model enters the picture, the dissi- the energy released for a virtual crack path, we have
pative term is proportional to the length of the extended the well known Eshelby-Rice-Cherepanov J integral
Anti-symmetry principle for quasi-static crack propagation 23
1
for straight extensions (cf. Rice 1968). Explicitly, for
α = 0 we have that: 0.8
0.6
G(0) = J1 (9) 0.4
0.2
where J1 (formula (68) in Rice 1968) is the compo-
G(α)
0
nent, in the direction parallel to the existing crack, of
−0.2
the vector given by (summation on repeated indices is
−0.4 Sih’s result
assumed):
−0.6 J integral
−0.8
Jk := U n k − σi j u i,k n j ds, (10)
C −1
−1 −0.5 0 0.5 1
α
where |n | = 1 is normal to C (a Jordan curve surround-
ing the crack tip), U is the elastic energy density, σi j Fig. 1 The energy release rate for different growing directions
are the components of the stress tensor and u i is the
displacement field. In pure Mode III, (10) reduces to
K being the stress intensity factor for the initial con-
the expression:
figuration, and µ the shear modulus. In Fig. 1 we see
Jk = U n k − σ3 j u 3,k n j ds ( j, k = 1, 2). a comparison between the true energy release rate and
C the one obtained using J as a generalised force in (11).
Thus, J1 gives the energy release rate for a straight It is to be noted that, even if J cannot be used as a
extension of the preexisting crack. Nevertheless, (9) good reference for the work done by the elastic forces, it
tells us nothing for other virtual directions of propa- shows correctly that the best direction to release energy
gation. It should be pointed out that in Rice’s article, is the same as the one defined by the preexisting crack
the quantity J is not considered as a vectorial force. (for Mode III). We conclude that the “crack driving
More recently, Gurtin introduced the vector (10) in force”, or more precisely the first order variation of the
Gurtin (formula (4.6) in Gurtin 1979) with a different energy, is not enough to determine the crack geome-
notation. This may be considered as a natural general- try: any smooth curve is compatible with the condition
ization of (9) when the coordinate axes are not parallel of maximum G imposed in (8). On the other hand, in
to the crack direction. The formula for G turns to be: addition to the well-known Griffith’s balance during
G = J · e, (11) critical growth
where e is a unit vector in the direction of (quasi-static) G(0) = κ (13)
motion. This expression suggests that J is a mechan- it is certainly useful to establish a necessary condition
ical force yielding the energy released for arbitrary for the stability of the crack configuration.
directions of motion, while freezing the actual config-
uration. But this is not the case, since (11) is only valid
for smooth extensions of the initial crack. Then it is not 3 The energy functional for kinks and straight
possible to give J the status of a force as a “gradient” extensions
of some free energy.
Let us see that (11) gives a wrong answer when e is In the previous Section we showed that in Mode III we
not pointing in the direction of the current crack con- should study the energy released functional in more
figuration. For this kind of “kinked” paths in Mode III detail to find a complete crack evolution law from first
the stress intensity factor was obtained by Sih (1965) principles. We will now summarize the results obtained
applying uniform loads at infinity. By the well-known in Oleaga (2004) where the energy released for a finite
Griffith–Irwin relationship, we can apply this result to extension of the crack is expanded in terms of crack
compute the energy release rate for different directions: length and more information about crack direction is
K2 1 − α α obtained. We refer to that article for any technical
G(α) = α := ϕ/π, (12) detailed not included here.
2µ 1 + α
24 G. E. Oleaga
f0
u0 ul z
u0 ζ =−z2 Upper half plane
ϕ=απ
o’ o
3.1 The boundary value problem Taking into account the assumed Neumann homoge-
neous condition, it follows that η0 is such that:
The energy release rate for different growing directions, ∂u 0 ∂u 0
η0 (ζ ) = −i ⇒ η0 real on 0 .
shown in (12), was obtained in Sih (1965) for an ini- ∂x ∂y
tially rectilinear crack in the plane, subjected to uniform The initial configuration may be carried to the upper
loading at infinity. This boundary conditions define a half plane by the elementary map
local field that is probably not rich enough to capture
the growing directional preference. To overcome this f 0 (z) = −z 2 (18)
(possible) drawback, we considered the semi-infinite Consequently, the complex function given by (Fig. 3)
straight crack as initial configuration with an arbitrary h 0 (z) := η0 ( f 0 (z))
displacement field u 0 around it, satisfying the equilib-
rium equation with no tractions on 0 : may be extended analytically by symmetry to the whole
plane, and then:
u 0 = 0 in R2\0 , (14) ∞
∂n u 0 = 0 on 0 . (15) h 0 (z) = cn z n cn ∈ R.
n=0
We also add the finite energy condition around the tip, Therefore, u 0 admits the following convergent expan-
for some r > 0: sion:
u 0 = Re h 0 f 0−1 (ζ )
|∇u 0 |2 < ∞. (16)
Br (tip) = c0 − c1r 1/2 sin (θ/2) − c2 r cos (θ ) + · · · (19)
where
Once the crack is extended, similar conditions hold for ζ = r eiθ f 0−1 (ζ ) = −ζ
u l together with the “matching condition” (Fig. 2):
We thus take (19) as the most general initial local field.
|u l − u 0 | → 0 uniformly at infinity (17)
Notice that in this case we are using t ≡ l = length 3.2 Kinked configurations: conformal mapping
of the crack extension to parametrise the displacements.
In this way, the elastic field at infinity is fixed dur- For a given angle ϕ we compute an expansion of the
ing crack advance and there is no mechanical work of Energy released:
the applied loading. The function u l satisfies a mixed fl (z) := − (z − a (l))1−α (z − b (l))1+α ,
boundary value problem and it is well defined for a
α := ϕ/π (−1 ≤ α ≤ 1) (20)
given initial field and crack configuration l (cf.
Oleaga 2004 for the details). where:
1+α
Using basic properties of harmonic functions we √ 1−α 2
write the field as the real part of an analytic function a (l) := − l ,
1+α
η0 : 1−α
√ 1+α 2
u 0 = Re [η0 (ζ )] η0 analytic in C\0 . b (l) := l . (21)
1−α
Anti-symmetry principle for quasi-static crack propagation 25
fl
Fl (z) = f 0−1 ◦ fl (z) = (z −a (l))1−α (z −b (l))1+α
z
ul ζ ∆Γ o’ b1 (l) b2 (l)
a’ ϕ
Upper half plane
= z + b0 (l) + + 2 · · · z → ∞.
b’ a o b
z z
(24)
Kinked Configuration
Notice that for α = 0 we have that a(l) = −b(l) and
then b0 (l) = 0.
−1
Fl On the other hand, Fl has the following scale invari-
f0
ance:
o’’ √ √ √ (n+1)
Fl (z) = l F1 z/ l ⇒ bn (l) = l bn (1)
a’’ b’’
and may be extended to the lower half plane, being a
sectionally holomorphic function in C\[a, b].
We conclude that u l admits the following expansion in where C encloses the interval [a(1), b(1)].
terms of some real coefficients cn (l): The first term of (26) is given by
∞
n µπ 2 1 − α α
u l = Re −1
cn (l) fl (ζ ) , ζ ∈ C\l . (22) −E = l c1 + O l 3/2 .
4 1+α
n=0
This provides a formula for the energy release rate in
terms of the kinking angle:
3.3 Relationship between cn and cn (l)
E µπ 2
G(α) = lim − = c A(α).
Using analyticity, Neumann boundary conditions and l→0 l 4 1
the asymptotic matching condition at infinity we can This is the same behavior as that in (12), already ob-
show that tained by Sih in (1965) for uniform stresses at infinity.
We can see that even if we take a general equilibrium
cn (l) = cn + (n + 1) cn+1 b0 (l) + O (l) , (23)
field around the tip, it is not possible to obtain the crack
where b0 (l) = − 21
((1 + α)b(l) + (1 − α)a(l)) is ob- deviation from the initial configuration by using the first
tained from the expansion at infinity of the conformal order expansion for the energy released. Therefore, we
map: must look at the energy functional in more detail.
26 G. E. Oleaga
0 0 0
tip tip
0 0 0
c2= 0 c2= 1
Recent results of Kager et al. (2004) provide explicit 4.3 The energy released functional and
solutions of the Loewner equation for some behaviours the anti-symmetry principle
of ξt . For instance, taking ξt = λt, (30) takes the form
2∂z Ft (z) It is possible to write the following expansion for the
∂t Ft (z) = − , F0 (z) = z. (32) energy released (µ = 1):
z − λt
28 G. E. Oleaga
1 then violating the stability condition (4).
E(t) = c12 Ft (z) + c1 c2
4i
C
Therefore, since c1 = 0, a second necessary condi-
× Ft (z)2 + 2Ft (z) (z + b0 (t)) dz tion for a stable configuration is the following:
c2 = 0. (38)
+ O |I t |4
π 2 This is equivalent to impose k2 = 0 in the typical
= c b1 (t) + 4c1 c2 (b0 (t) b1 (t) expansion of the initial displacement field around the
2 1 tip:
+ b2 (t))) + O |I t |4 , as |I t | → 0. (35)
where |It | is the length of the interval It := [a(t), b(t)] u 0 = k0 + k1r 1/2 sin (θ/2) + k2 r cos (θ ) + · · ·
(see Fig. 7 above ). Thus, taking into account (31), the This condition cancels the symmetric contribution to
asymptotic expression for the energy is given by: the displacement near the tip, keeping the first (anti-
π
E(t) = −2tc12 + 4c1 c2 b2 (t) symmetric) term of the expansion. We can state that
2 (38) must hold on every stage of the propagation pro-
+ O |I t |4 , as |I t | → 0. (36) cess, thus providing the second scalar condition to com-
We have now all the ingredients to compute the plete the free boundary problem. On the other hand,
energy released by the slit generated by (32). Insert- it has to be satisfied by other models imposing global
ing (34) in (36) we find that an extension evolving up minimization of the total energy (see for instance
to “time” t will release an amount of energy given by Buliga 1999; Francfort and Marigo 1998).
E = −π c12 t − 2π λ c1 c2 t 2 + O(t 3 ).
According to (4), in order to check the stability we
5 The case of a finite domain
must take into account the dissipation that is required
to open this path. For this purpose we need the evolu-
The stability of the configuration was studied by impos-
tion of the length of the extended curve. The trial path
ing a Dirichlet condition at infinity (see (17)). It is natu-
in the physical plane is given by
ral to ask if the same outcome can be obtained on a finite
8 3 2
− (γλ (t))2 = 4t − iλt 2 − λ2 t 2 + O(t 5/2 ). domain. In other words, what happens if we freeze the
3 3 displacements at a finite distance of the tip instead of
The length of this crack extension is given by the expan-
fixing them at infinity? Bearing this question in mind,
sion
we will explain which parts of the previous discussion
2λ2 2
l(t) = 4t + t + O(t 3 ) t → 0. should be revised. This part of the work was not previ-
3
We use the last computations in the following equa- ously considered in the reviewed articles (Oleaga 2004,
tion: 2006).
2κλ2
E + Q = (4κ − π c1 )t + 2
− 2π λ c1 c2
3
5.1 A simple setting for the boundary value problem
×t 2 + O t 3
The term of order t contains the balance between We start with an initially straight crack configuration
energy release rate G(0) and κ. It cancels out due to with the tip at the center of a circle of unit radius (the
the critical results are in fact independent of this particular geom-
growth equation (13) and we then have that:
etry). The selected field satisfies the equilibrium equa-
|c1 | = 4κ
π . To satisfy the stable growth condition (3)
tion (14), the Neumann boundary condition (15) and
the term of order t 2 must be positive or zero. Comput- the finite energy condition near the tip (16). We now
ing the minimum with respect to the parameter λ we can apply a one-to-one conformal map g0 transforming
3π c1 c2
obtain λ = , and for this value we have: the upper half unit disc onto the initial domain. Take
2κ
u 0 (x, y), the real part of an analytic function η0 (x +i y)
3π 2 2 2
2
E + Q = − c c t + O(t 3 ) < 0 for t → 0, in D\0 , where
2 κ 1 2
(37) D := {z : |z| < 1} .
Anti-symmetry principle for quasi-static crack propagation 29
g0(z)
z gt (z) z
ξ(t) ’
Γ0
o’ o
Γt a’
b’
[
a(t)
.
ξ(t)
]
b(t)
u0 ut
The analytic function h 0 := η0 ◦ g0 (defined on u t gt eiθ = u 0 gt eiθ
the right side of Fig. 8) is extended by symmetry to
the lower half disk due to the Neumman homogeneous gt eiθ = 1cf. Fig. 9 ,
condition. It then has a Taylor expansion with real coef-
u 0 gt eiθ = u 0 g0 g0−1 ◦ gt eiθ
ficients as follows:
n
u 0 g0 G t eiθ = Re h 0 G t eiθ ,
u 0 (g0 (z)) = Re (h 0 (z)) = Re cn (0) z n
n=0
where G t is defined as
cn ∈ R. (39) G t := g0−1 ◦ gt .
For simplicity, we will assume that (39) is convergent This map is analytic on a circular ring whose interior
on some open set containing D. circle contains the interval [a(t), b(t)] (cf. Fig. 10). It
We define the field u t as the one satisfying the equi- admits a Laurent expansion with real coefficients that
librium equation u t = 0 on D \ t , the Neumann we write as follows:
boundary condition ∂u/∂n = 0 on t , and the Dirich- G t (z) = G + −
t (z) + G t (z), (41)
let boundary condition on the unit circle: with
∞
∞
u t (x, y) = u 0 (x, y) for |x + i y| = 1.
G+
t (z) := ak (t) z k G−
t (z) := bk (t) z −k .
There is an analytic function ηt in D \ t such that k=0 k=1
u t (x, y) = Re ηt (x + i y) and the function (42)
h t (z) := ηt (gt (z)) Notice that G t (z) → z and bk+1 (t) bk (t) for t → 0.
This last assertion is a consequence of the Area The-
is analytic in the upper half unit disk and can be ex- orem (see the Appendix). Notice that G − t is analytic
tended to the whole disk by the Neumann homogeneous outside of a vanishing circle containing [a(t), b(t)].
condition. The map gt (z) sends the upper half unit disk To simplify a bit the notation we define the boundary
to D\t as shown in Fig. 9. We have that h t admits the
of
values real part of h t as Ut (θ ) := Re
the
following expansion: h 0 G t eiθ . Using Schwartz integral in the unit
∞
disk, we have the following explicit formula
h t (z) = cn (t) z n cn (t) ∈ R. (40) π iθ
1 e +z 1
n=0 h t (z) = Ut (θ ) dθ =
2π −π eiθ − z ζ :=e iθ 2πi
The main difference with the case of an unbounded
ζ +z dζ
domain corresponds to the behaviour of the coefficients × Re {h 0 (G t (ζ ))} (43)
|ζ |=1 ζ − z ζ
cn (t) for t → 0. The Dirichlet condition is applied on a
curve at a finite distance of the tip, while in the former We can then write (cf. (40)):
∞
case it was imposed on a unique point at infinity.
h t (z) = cn (t)z n
Assuming for a moment that we know gt (z), we can
n=0
write Schwartz integral representation of h t (z) for the ∞ π
1
unit disk. The boundary values of the real part of h t (z) = (1 + z) Ut (θ ) e−inθ dθ z n
are obtained from the following chain of equalities: 2π −π
n=0
30 G. E. Oleaga
We want now to relate the coefficients cn (t) with the We can now apply the complex version of the energy
ones of the initial field cn (0). Using the expansion for release formula (cf. 25):
h 0 (z) we have that:
1
E = h (z) h 0 (G t (z)) dz
4i C t
Ut (θ ) = Re h 0 G t eiθ ∞
∞ 1
k = k ck (t) z k−1
4i C
= Re ck (0) G t e iθ
⎛
k=1
⎞
k=0 ∞
×⎝ c j (0) (G t (z)) j ⎠ dz,
The expression for the first coefficient (notice that j=0
Ut (−θ ) = Ut (θ )) should read (we drop for a while the
where C is a simple curve enclosing the interval
dependence on t of the ai ’s and bi ’s):
[a(t), b(t)] and contained in D. Taking into account
∞
π j (45, 46) we can approximate the expression for E as
1
c0 (t) = Re ck (0) G t eiθ dθ follows:
π 0 π
j=0
E ≈ {c1 (t) (c1 (0)b1 (t) + 2c2 (0) (a0 (t)b1 (t)
= c0 (0) + c1 (0) a0 + c2 (0) 2
+ a1 (t) b2 (t))) + 2c2 (t) (c1 (0)b2 (t)
× a02 + 2a1 b1 + 2a2 b2 + · · · (44)
+ 2c2 (0) (a0 (t)b2 (t) + a1 (t)b3 (t)) + · · · )}
π 2
Similarly, we obtain that for c1 (t), ≈ c (0)a1 (t) b1 (t) + 2c1 (0)c2 (0)
2 1
∞
π j
× a0 (t)a1 (t)b1 (t) + a12 (t)b2 (t) + a0 (t)b12 (t)
2
c1 (t) = c j (0) Re G t eiθ cos(θ )dθ
π
j=0 0 · · · + 2c1 (0)c2 (0)a12 (t)b2 (t) + · · ·
= c1 (0) (a1 + b1 ) + 2c2 (0) (a0 (a1 + b1 )
Summing up:
+ a1 b2 + a2 (b1 + b3 )) + · · · (45) π 2
E ≈ c (0)a1 (t) b1 (t) + 2c1 (0)c2 (0)(a0 (t)a1 (t)
2 1
For c2 (t) we have:
× b1 (t) + 2a12 (t)b2 (t) + a0 (t)b12 (t) (47)
∞
π j
2 Notice that for a0 = 0 and a1 = 1 we obtain the same
c2 (t) = c j (0) Re G t eiθ cos(2θ ) dθ
π 0 asymptotic expression as that derived for the
j=0
= c1 (0) (a2 + b2 ) + c2 (0) unbounded domain (cf. (35)).
× a12 + 2a0 b2 + 2a1 b3 + · · · (46)
5.2 The construction of the conformal map
the main contribution being (notice that a1 (t) → 1 for
t → 0): To justify the anti-symmetry principle for a finite
domain it only remains to show that a suitable confor-
c2 (t) = c2 (0)a12 (t) + o(1). mal map can be constructed, and that its coefficients
Anti-symmetry principle for quasi-static crack propagation 31
π iθ
Ft t e +z
ωt (z) = z 1 − ρ(θ ) dθ + o(t) .
2π −π eiθ − z
(48)
z ξ+’(t)
We may write this as follows:
[
a(t)
ξ+(t)
. ] ’
a(t) ’
b(t)
Ft (C)
∞
ξ −( t) b(t)
ωt (z) = z (1 − tψ(z) + o(t)) , ψ(z) = pn z n .
ξ ’(t) n=0
−
This is consistent with the asymptotics of the stress Appendix A: The area theorem for
intensity factor obtained by Leblond in (1989), but a the Laurent expansion
thorough discussion of this subject would carry us far
from our main point here. We mention by pass that the Consider a family of univalent functions indexed by t,
physical length of the crack is (up to order t) l ∼ 4t, on a circular ring whose interior circle is of radius ρ(t),
and then the energy release rate is consistent with the with ρ(t) → 0 as t → 0. Moreover, assume that each
expression obtained before: function admits a Laurent expansion with real coeffi-
E π cients written as (41). It is possible to show that the
lim − = c12 (0). area inside the image circle Cr of radius r inside the
t→0 4t 4
circular ring, is given by:
This is also in accordance to Leblond’s paper, since
we should not have an explicit dependence of the body 1
Area = Im G t (z) G t (z) dz.
geometry in this order (of course, there is an implicit 2 Cr
information encoded in c1 ). Taking into account the Laurent expansion for G t in
To conclude, it remains to find the expression for the (41), we can write:
length in terms of t, up to order t 2 . We have to take into
∞ ∞
account that ωt (z) ≈ (1 − t p0 )z. The path in the (non 1
Area = Im an z n + bn z −n
physical) lower domain in Fig. 11 is given by (cf. 33): 2 Cr n=0
∞
n=1
2 i 2 3/2 ∞
γλ,t = (1−t p0 ) 2i t + λt − λ t
1/2
+ O(t 2 ). −n−1
3 18 × nan z n−1
− nbn z dz
n=1 n=1
In the physical domain we should have, recalling that ∞ ∞
G t (z) = g0−1 ◦ gt (z) and g0 (z) = −z 2 :
=π nr n an2 − nr −n bn2
8 3 2 2 2
n=1 n=1
g0 (γλ,t ) = (1−t p0 ) 4t − iλt − λ t + O t 5/2
2 2
3 3 where we avoided the dependence on t for the coeffi-
Therefore, the length of the path is given by: cients. Since Area is non-negative, and taking r → ρ(t)
we finally obtain the inequality:
2λ2 2
l(t) = (1 − t p0 )2 4t + t + O(t 3 ) t → 0 ∞
∞
3
nρ(t)−n bn2 ≤ nρ(t)n an2 . (50)
We conclude that the dissipation term obtained in the n=1 n=1
unbounded case is to be multiplied by the positive non-
The right hand side is convergent by construction; in
universal factor (1−t p0 )2 , the same as that multiplying
fact, it goes to zero for t → 0. Notice that
the relevant contribution to the released energy in (49).
3π c1 c2 ∞
Therefore, selecting λ = 1 +
2κ
as in the unbounded Im G+
t (z) G t (z) dz = π nρ(t)n an2 .
case, the constructed virtual path shows again the unsta- 2 Cr n=1
ble character of an initial field with c2 (0) = 0 (cf. (37)): We can then take the limit for r → ρ(t) since G +
t is
2 3π 2 2 2
2
analytic inside the outer circle of the ring (cf. (42)).
E + Q = − (1 − t p0 ) c c t
2 κ 1 2 This shows that the coefficients of the left hand series
+ O(t 3 ) < 0 for t → 0. in (50) must be, at least, bounded:
M
This path violates the stability assumption (4) and pro- nρ(t)−n bn2 (t) ≤ M ⇒ bn2 (t) ≤ ρ(t)n , for n ≥ 1,
n
vides a necessary condition for the initial field, namely
c2 = 0. for a suitable positive constant M. This provides the
asymptotic relative magnitude of the different bn ’s for
Acknowledgements This research was supported by Spanish t → 0.
research project BFM 2004-05634. Part of this work was pre-
sented at the International Symposium on Defects and Mate- References
rial Mechanics held in Aussois, 25-29 March 2007. I wish to
acknowledge the organizers C. Dascalu and G. Maugin for the Buliga M (1999) Energy minimizing brittle crack propagation. J
kind invitation to participate in this meeting. Elast 52:201–238
Anti-symmetry principle for quasi-static crack propagation 33
Cotterell B, Rice JR (1980) Slightly curved or kinked cracks. Int Marshall DE, Rohde S (2001) The Loewner differential equation
J Frac 16(2):155–169 and slit mappings. J Am Math Soc 18:763–768
Duren PL (1983) Univalent functions, a series of comprehensive Mumford D, Shah J (1989) Optimal Approximations by Piece-
studies in mathematics. Springer-Verlag, New York wise Smooth Functions and Associated Variational Prob-
Francfort GA, Marigo JJ (1998) Revisiting brittle fracture as an lems. Commun Pure Appl Math XLII:577–685
energy minimization problem. J Mech Phys Solids 46(8): Nehari Z (1975) Conformal mapping. Dover Publications,
1319–1342 New York
Goldstein RV, Salganik RL (1974) Brittle fracture of solids with Oleaga GE (2004) On the path of a quasi-static crack in Mode
arbitrary cracks. Int J Frac 10:507 III. J Elast 76(2):163–189
Gurtin M (1979) On the energy release rate in quasi-static elastic Oleaga GE (2006) The classical theory of univalent functions and
crack propagation. J Elast 9(2):187–195 quasi-static crack propagation. Eur J Appl Math 17:233–
Irwin GR (1957) Analysis of stresses and strains near the end of 255
a crack transversing a plate. J Appl Mech 24:361–364 Rice JR (1968) Mathematical analysis in the mechanics of frac-
Kager W, Nienhuis B, Kadanoff L (2004) Exact Solutions for ture. In: Liebowitz H (ed) Fracture. Academic, New York,
Loewner Evolutions. J Stat Phys 115(3/4):805–822 pp 191–311
Leblond JB (1989) Crack paths in plane situations I: General Sih GC (1965) Stress distribution near internal crack tips for
form of the expansion of the stress intensity factors. Int J longitudinal shear problems. J Appl Mech 51–58
Solids Struct 25(11):1311–1325
Configurational balance and entropy sinks
Marcelo Epstein
Abstract For evolutionary processes of material the nature of the forces responsible for the evolution
remodelling and growth, a comparison is drawn bet- of material inhomogeneities (Eshelby 1951). In biolo-
ween a conventional formulation and one that post- gical applications, the term remodelling is applied to
ulates the existence of additional balance laws for the those situations in which the mass density of the body
configurational forces. remains unaffected, as opposed to processes of growth
or resorption. Following this clear biological distinc-
Keywords Growth · Remodelling · Plasticity · tion, we will extend the terminology to all processes,
Configurational forces · Eshelby stress · Material regardless of the physical context in which they appear.
evolution Thus, for example, metal plasticity will be regarded as
a remodelling phenomenon, whereas erosion as a pro-
cess of (negative) growth (or: resorption).
1 Introduction The equations that govern processes of remodelling
and growth can be regarded as the result of focusing
One of the recent trends in Continuum Mechanics is the attention on a single component of a chemically reac-
unified description of processes that alter not only the ting mixture of several substances. Mass is, therefore,
spatial distribution of the body particles but also their not necessarily conserved and transfers of momentum,
material arrangement. These changes taking place in angular momentum, energy and entropy appear in the
the material manifold are known as processes of mate- equations as extra contributors to the total balance. In
rial evolution. They encompass such diverse pheno- a process of growth, these extra terms may be, at least
mena as crack propagation, plasticity, viscoelasticity, in part, attributed to the very addition or subtraction
and biological growth and remodelling. Their descrip- of mass. At best, we may have a case of “compliant”
tion is the task of a branch of Continuum Mechanics growth, whereby the new mass happens to enter at
that can alternatively be called Material Mechanics or the same velocity, specific energy and specific entropy
Configurational Mechanics. The terminology Eshelbian as the local substratum. At worst, not only will the
Mechanics is also used in deference to John D. Eshelby entering quantities be at a different state than the sub-
(1916–1981), one of the first scientists to clearly identify stratum, but the process of growth (or even just remo-
delling) itself may entail other sources of discrepancy.
M. Epstein (B)
Department of Mechanical and Manufacturing
Engineering, The University of Calgary, Calgary,
Remark 1.1 In (Cowin and Hegedus 1976) the “non-
Alberta, Canada T2N 1N4 compliant” source of internal energy is lumped as a
e-mail: mepstein@ucalgary.ca clearly indicated extra term in the balance equation,
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 35
doi: 10.1007/978-1-4020-6929-1_5, © Springer Science+Business Media B.V. 2007
36 M. Epstein
so that the compliant case can be easily recovered by of the difficulties involved in modelling these processes,
eliminating this extra term. This practice, which will see Segev and Epstein (1996).
be followed here, was also adopted for the totality of
balance laws in (Epstein and Maugin 2000), where the
compliant contributions were called “reversible”. To 2.1 Balance equations
avoid unnecessary confusion with the use of this ter-
minology in the strictest thermodynamical sense, we While at some conceivable level of analysis (such as
prefer to call them “compliant”. that of chemically reacting mixtures) the appearance or
disappearance of mass of one species may be accounted
A physical interpretation of the non-compliant sour- for by concomitant losses or gains in other species, in a
ces can be inferred from the fact that certain evolution bulk growth model we accept the existence of sources
processes cannot take place spontaneously, but need the and sinks of mass as part of the theory.1 On this basis,
participation of some external agents. From the mix- we obtain the following referential form for the balance
ture point of view alluded to above, these agents can of mass:
be perhaps traced back to the excluded components of ∂ρ R
= + DivM. (2.1)
the mixture. In certain cases (Epstein 2005), one may ∂t
imagine the presence of some microactuators which, where t denotes time, ρ R is the possibly time-varying
after detecting the present state of affairs as far as the mass density in the reference configuration, is the
presence of certain material defects, are programmed (smooth) volumetric source of mass, Div is the refe-
to make the defect pattern evolve in a prescribed way. rential divergence operator and M is a vector of possible
Since to the unaware observer these mechanisms may mass flux through the boundary with exterior unit nor-
appear to act against the natural tendency of an isolated mal N. The mass flowing through the boundary per unit
system to increase its entropy, we refer to the corres- time is given by:
ponding non-compliant terms as entropy sinks.
An alternative point of view advocates the postula- M =M·N (2.2)
tion of additional balance laws to be satisfied by the where N is the unit normal. In what follows, to simplify
material or configurational forces, thus rendering the the analysis, we will assume the mass flux to vanish
entropy sinks unnecessary. The question as to whether identically, so that only the volumetric source of mass
these extra balance laws are to be admitted into the remains.
analysis of remodelling processes is still the subject of By the standard procedure, we obtain the following
some controversy. The purpose of this paper is to show equation for the balance of linear momentum:
that, at least in the context of the class of problems dis- ∂v
cussed, there exists a precise relation between the two ρR = f R + p̄ R + DivT. (2.3)
∂t
approaches and that they are in fact equivalent.
In this equation v is the velocity, f R is the body force
per unit referential volume and T is the Piola stress.
The term p̄ R represents the non-compliant source of
2 The field equations of remodelling and bulk momentum.
growth In the absence of body and surface couples, the
balance of angular momentum yields the classical
By bulk, or volumetric, growth we understand a process symmetry of the Piola stress, namely:
of addition or removal of mass while the body par-
ticles retain their identity. It is only the mass density TFT = FTT (2.4)
that changes with time. In contradistinction with this We note that if mass flux had not been neglected, this
situation it is possible, and certainly meaningful and equation would no longer hold.
practical, to consider growth by addition of mass at the
1 The author wishes to thank an anonymous reviewer for brin-
boundaries of the body. Thus, for example, holes may
ging to his attention two recent papers by Garikipati et al. (2004,
be closed or created which change the topology of the 2006) where some of these issues are discussed in greater depth.
original body. These more complicated processes are Our purpose here, however, is to deal with the simplest possible
excluded from the present analysis. For some indication setting in which the comparison of approaches is meaningful.
Configurational balance and entropy sinks 37
Denoting by u ρ the internal energy per unit mass, of non-compliant entropy, which should be specified
we can write the local form of the energy balance as: constitutively. Notice the subtle distinctions between
∂u ρ this form of the Clausius-Duhem inequality, which does
ρR = ρ R rρ + Ū + tr [T(∇v)T ] − DivQ, (2.5) not presume mass conservation, and its usual counter-
∂t
where ∇ is the referential gradient operator, rρ and part, namely:
−Q.N represent, respectively, the rate of non- 1
mechanical energy supply per unit mass (“radiation”) ψ̇ R + s R θ̇ − tr [T(∇v)T ] + Q∇θ ≤ 0, (2.8)
θ
and per unit area (“conduction”) and where Ū is the where ψ R and s R are measured per unit referential
non-compliant volumetric contribution to the internal volume. Notice also that the entropy sink H̄ may be
energy. In principle, this term could have been absorbed present even if growth does not occur. It may, indeed,
into rρ , with the understanding that it may eventually be responsible for the process of remodelling.
be specified constitutively, rather than just externally.
What does the Second Law of Thermodynamics have to 3.1 The material archetype, its implants
say when one has come to terms with the assumption and evolution
that, at least for modelling purposes, matter is conti-
nuous? A possible answer to this question is the one A theory of growth and remodelling is inextricably
embodied in the Clausius-Duhem inequality. It is not intertwined with the concept of internal state variables.
our intention to either contest or defend the validity of From the point of view of a putative mixture theory,
this, or any other, particular form of the Second Law. this fact can be seen as a consequence of not including
We simply do not know how this fundamental law of all the components of the mixture. Be that as it may,
nature can be rendered compatible with such a bold, what are the appropriate internal variables to be used?
albeit manifestly useful, assumption. To further com- Clearly, this is a matter of definition of the model being
plicate matters, there is the issue of the use of inter- used. Here, we will be following the anelastic model,
nal state variables (Coleman and Gurtin 1967), which without claiming that it is the only possible one.2 In
can be regarded as implicated in any model of material an anelastic-like theory, the internal variables can be
evolution. Is the formulation of the Second Law for a motivated as follows (Epstein and Maugin 1990):
continuous medium to be modified, perhaps augmen- Let us introduce the notion of a hyperelastic material
ted, in the presence of internal variables? Or must we archetype as a material point endowed with a particular
recalcitrantly cling to the original form and live with (orthonormal) frame in which its material response is
the consequences? We choose to proceed to the formu- expressible in terms of a single scalar function:
lation of the Clausius-Duhem inequality adding terms ψ̄ = ψ̄(F), (3.1)
that are consistent with the philosophy used for the
balance equations formulated so far. representing a free energy per unit volume in terms of
Defining the entropy content per unit mass, sρ , and a linear map F. This map can be interpreted physically
introducing the Helmholtz free energy per unit mass as the deformation of the standard unit cube into an
as: arbitrary parallelepiped.
The role of this archetype is double. In the first ins-
ψρ ≡ u ρ − θ sρ , (2.6)
tance, it can be used to model the material response of
θ being the absolute temperature, the Clausius-Duhem a point X belonging to a given material body B. Let
inequality states that: the body be in some global reference configuration, in
1 which we adopt a Cartesian coordinate system. Then,
ρ R ψ̇ρ + ρ R θ̇ sρ − H̄ − tr [T(∇v)T ] + Q∇θ ≤ 0.
θ
2
(2.7) In fact, in some theories of bone remodelling, it is customary to
use a different model, whereby the change of density is accom-
We remark that, following Cowin and Hegedus panied by a change in elastic properties, rather than by a mere
(1976), we have added a further volumetric sink H̄ re-accommodation of the “relaxed configuration”.
38 M. Epstein
if X is actually made of the archetypal material, there This particular detail makes the Eshelby stress emerge
must exist a linear map P(X) from the archetype to the as the natural consequence of calculating the forces
tangent space TX B, called an implant of the archetype associated with the internal variables. Moreover, the
into X, such that the response of this point in the given Eshelby stress acquires precisely the meaning attribu-
reference configuration is given precisely by: ted to it by Eshelby himself, namely, the energy expen-
diture associated with the change in the inhomogeneity
ψ R (F, X) = J P−1 ψ̄(FP(X)). (3.2) pattern of a material. More graphically, the Eshelby
Otherwise, what right would we have to claim that both stress is the energy expended per “unit” remodelling.
the archetype and the body point are made of the same To see that this is indeed the case, let us calculate first
material? Here, J denotes the determinant of the sub- the derivative of Eq. 3.5 with respect to the deformation
scripted tensor. The only difference that may exist bet- gradient, thus obtaining the Piola stress as:
ween the responses is one of “scaling” (via some linear ∂ψ R ∂ ψ̄ T
T= = J P−1 P . (3.6)
map) and, possibly, of zero energy level (via an addi- ∂F ∂(FP)
tive constant). In the case of a solid material this last The remodelling counterpart of this calculation should
degree of freedom can be easily disposed of (as we have be obtained by taking the derivative with respect to the
assumed in the above equation) by assigning a value of implant, namely:
zero energy to the natural states. Given a material body,
∂ψ R ∂ ψ̄
it may so happen that each and every point has a res- = −J P−1 P−T ψ̄ + J P−1 FT . (3.7)
ponse given by the above equation. In that case we say ∂P ∂(FP)
that the body is materially uniform (according to Noll’s Combining the last three equations yields the following
terminology (Noll 1967)). result:
The second role that an archetype may play, whe- ∂ψ R
ther or not the body is materially uniform, is nothing = − ψ R I − FT T (P−T ), (3.8)
∂P
but the temporal counterpart of the spatial notion of where I is the identity tensor. The quantity within the
uniformity just introduced. What we mean by this is square brackets, namely:
that, although a material point, as time goes on, may
retain the “chemical identity” (as it were) provided by b = ψ R I − FT T, (3.9)
the archetype, it may so happen that the implant P(X) is precisely the classical expression of the Eshelby
evolves in time, thus becoming: stress.
If we denote the constant density of the archetype by
P = P(X, t). (3.3)
ρ̄ R , the density at the reference configuration is given
If this is the case, we say that the material point X by:
exhibits an anelastic behaviour, or that its constitution ρ R = ρ̄ R J P−1 , (3.10)
undergoes an anelastic time evolution.
Assuming now, for the sake of simplicity, that the whence:
body is both uniform and anelastic, we may say that ρ̇ R = −ρ R tr (P−1 Ṗ), (3.11)
such a body is characterized by a constitutive law that
includes, in addition to the deformation gradient, a col- or:
lection of internal variables encapsulated in the ρ̇ R
tr (L̄ P ) = − , (3.12)
matrix P: ρR
where we have denoted:
ψ R = ψ R (F, P, X, t). (3.4)
L̄ P = P−1 Ṗ. (3.13)
But these internal variables are not completely arbi-
Comparing this result with Eq. 2.1, we conclude that:
trary, as one might think by a cursory glance at the
preceding equation. Indeed, they enter the constitutive = − ρ R tr L̄ P + DivM , (3.14)
law in a right-multiplicative way only, namely:
which in the present analysis reduces to:
ψ R = ψ R (F, P, X, t) = J P−1 ψ̄(FP(X, t)). (3.5) = −ρ R tr L̄ P . (3.15)
Configurational balance and entropy sinks 39
stress. The evaluation of the linear operator b̃ on a usual “strong” formulation and, in fact, generalizes the
virtual implant velocity, namely: latter for weaker conditions on the space of admissible
functions.
tr (b̃Ṗ T ) = b̃αI ṖαI , (4.2)
While the consistency between the weak and strong
is called the virtual power of the configurational force points of view in the classical case is well grounded on
on the given virtual implant velocity. a tradition that goes as far back as d’Alembert, there
So far, we have just formulated definitions. Consider is no a-priori reason to suppose that the principle of
for a moment the classical case of a non-evolving mate- virtual power can be extended meaningfully to include
rial. We know that the mechanical field equations can remodelling phenomena or, for that matter, any case
be derived from the postulation of a principle of virtual where internal state variables enter the physical pic-
power. Recall that to achieve this aim one introduces ture. But it certainly can be done formally, as the fol-
the virtual power (EVP) of the external forces as: lowing treatment shows. The first thing to be done is
straightforward: the internal virtual power is to be sup-
EVP ≡ f R vd + t R vd S, (4.3) plemented with the integral of the local virtual power
∂ of the configurational force. Instead of Eq. 4.4, we have
where t R is the surface traction per unit referential area now the augmented expression:
of the unsupported part of the boundary. We are not
including inertia effects, for the sake of simplicity. We IVP ≡ tr (TḞT ) + tr (b̃Ṗ T ) d. (4.9)
now define the internal virtual power (IVP) as the inte-
gral of the expression (4.1), viz: The second modification entails, as one would expect,
a generalization of the external virtual power expres-
IVP ≡ tr (TḞT )d. (4.4) sion. It is here that an important difference between
the classical case and the new formulation arises. For,
The principle of virtual power stipulates the satisfaction whereas in the classical case we have at our disposal an
of the identity: ordinary velocity field (whose referential gradient deli-
vers the velocity of F), in the augmented formulation
IVP ≡ EVP, (4.5) the implant field “velocity” Ṗ is not necessarily inte-
for all virtual velocities. Now, in this classical case there grable. In other words, there doesn’t in general exist a
is an understood extra compatibility condition between vector field whose referential gradient is Ṗ. One way to
the virtual ordinary velocities (v) and the virtual velo- come out of this analogical impasse is to postulate the
cities of the deformation gradient (Ḟ). This condition existence of an external material body force B which
establishes that the deformation gradient velocities must performs virtual power on the same virtual field Ṗ as
be derived from the (globally smooth) ordinary velo- the internal material force b̃. The nature of this new
city field. That is, the identity (4.5) is enforced only external force is left to be specified as part of each
under the condition that: particular theory. It may very well vanish altogether or
it may be meaningfully stipulated from physical consi-
Ḟ = ∇v. (4.6)
derations. At any rate, as an external body force, B is
In this case, integration by parts and the arbitrariness supposed to be given directly, rather than determined
of the virtual velocity field v are immediately seen to by any constitutive equation. The augmented external
imply the field equation: virtual power becomes:
DivT + f R = 0, (4.7) EVP ≡ f R v + tr (BṖ ) d +
T
t R vd S.
and the natural boundary condition: ∂
(4.10)
TN = t R , (4.8)
Assuming the independence of the spatial and material
where N is the unit exterior normal to the (unsupported
virtual velocity fields, we obtain now the extra balance
part of the) boundary in the reference configuration.
equation:
Thus, the “weak formulation” (4.5) delivers the same
differential equations and boundary conditions as the b̃ − B = 0, (4.11)
42 M. Epstein
which doesn’t seem much of a balance equation, but formula for the Piola stress in terms of the derivative
which can be intelligently exploited, as we shall pre- of the referential free-energy density), but that the for-
sently see. mer does not. We see, then, that in this approach, the
To complete the theory, a dissipation principle is quantity b̃ R is not quite the Eshelby stress, but rather
postulated in the form of the following inequality: the sum of the Eshelby stress and the extra energetic
term b̂. To satisfy the inequality (4.18), an evolution
D
ψ R d ≤ tr (TḞT ) + tr (b̃Ṗ T ) d, equation will have to be given in terms of the pull-back
Dt
of b̂ (rather than that of b̃ R ) to the archetype.
(4.12) Returning now to the extra balance Eq. 4.11, we see
where we have omitted the non-mechanical (heating) that, even if the external material body force B were to
terms. Assuming the body to be uniform with a consti- vanish, the repercussion will not be the vanishing of the
tutive law given by: Eshelby stress, but the vanishing of b̃ R . In that case, the
extra energetic term b̂ will boil down to the classical
ψ R (F, X) = J P−1 ψ̄(FP(X)), (4.13) Eshelby stress, and the evolution equation will coincide
we obtain the local form of (4.12) as: with that of the previous formulation, whereby no new
balance law was postulated.
∂ ψ̄
tr −J P−1 P−T ψ̄ + J P−1 FT − b̃ Ṗ T At the other extreme, the dissipation inequality
∂(FP) tr [b̂LTP ] ≤ 0 can always be trivially satisfied by set-
−1 ∂ ψ̄ ting b̂ = 0 identically, thus rendering all remodelling
+ tr JP P − T ḞT ≤ 0,
T
(4.14)
∂(FP) processes apparently “reversible”. In this case, it fol-
which, in view of the identity: lows from Eq. 4.19 that the material force b̃ R coincides
with the classical Eshelby stress. The extra balance
∂ψ R ∂ ψ̄
= J P−1 PT , (4.15) Eq. 4.11, however, requires now that the external mate-
∂F ∂(FP) rial body force B be also equal to the Eshelby stress. In
can be written as: this way, one can specify any evolution law whatsoever
T ∂ψ R (for example, one of those we call of the “self-driven”
−tr ψ R I − F − b̃ R LTP
∂F type (Epstein 2005)) and always satisfy the dissipation
∂ψ R inequality at the price of an external agent (B) doing
+ tr − T ḞT ≤ 0, (4.16) the job of carrying out the prescription of the evolution
∂F
law by means of external sources of power. More to the
with
point, if we substitute the extra balance law (4.11) into
b̃ R ≡ −b̃ P T . (4.17) the dissipation principle (4.16), we obtain in general:
Rather than hastily interpreting this as the identical T ∂ψ R
−tr ψ R I − F − B R LTP
vanishing of the quantities within the square brackets, ∂F
one now introduces (in the terminology of Di Carlo ∂ψ R
2005) the extra energetic responses T̂ and b̂ satisfying + tr − T ḞT ≤ 0, (4.21)
∂F
the residual inequality:
with
− tr [T̂ḞT ] + tr [b̂LTP ] ≤ 0. (4.18)
B R ≡ −BP T . (4.22)
while making the identifications:
Focusing attention on the evolutionary part only,
∂ψ R
ψ R I − FT − b̃ R = −b̂, (4.19) namely:
∂F
and ∂ψ R
− tr ψ R I − FT − B R LTP ≤ 0, (4.23)
∂ψ R ∂F
− T = −T̂. (4.20)
∂F and comparing with the standard expression (3.23), we
We are still left with the task of specifying the extra conclude that, from the point of view of the standard
energetic quantities b̂ and T̂. Assume, for simplicity, formulation (without the extra balance law), what we
that the latter vanishes (so that we recover the usual have here is an entropy sink:
Configurational balance and entropy sinks 43
H̄ = tr (ψ R I − B R ) LTP (4.24) Epstein M (2005) Self-driven continuous dislocations and
growth. In: Steinmann P, Maugin GA (eds) Mechanics of
regulated from outside the system. In conclusion, whe- material forces. Springer, pp 129–139
ther one opts for the interpretation of these extraneous Epstein M, Maugin GA (1990) The energy-momentum tensor
and material uniformity in finite elasticity. Acta Mech
agents as external material body forces or as entropy 83:127–133
sinks in the standard dissipation inequality (3.23), there Epstein M, Maugin GA (2000) Thermomechanics of volumetric
doesn’t seem to be (at least in the situations just consi- growth in uniform bodies. Int J Plast 16:951–978
dered) a great deal of practical difference in the final Epstein M, Śniatycki J (2005) Non-local inhomogeneity and
Eshelby entities. Phil Mag 85(33–35):3939–3955
equations, since the extra configurational balance ulti- Eshelby JD (1951) The force on an elastic singularity. Phil Trans
mately delivers entropy sinks of a particular form. Roy Soc London A 244:87–112
Garikipati K, Arruda EM, Grosh K, Narayanan H, Calve
Acknowledgements This work has been partially supported S (2004) A continuum treatment of growth in biological
by the Natural Sciences and Engineering Research Council of tissue: the coupling of mass transport and mechanics. J
Canada. Mech Phys Solids 52(7):1595–1625
Garikipati K, Olberding JE, Narayanan H, Arruda EM, Grosh K,
Calve S (2006) Biological remodelling: stationary energy,
configurational change, internal variables and dissipation.
References J Mech Phys Solids 54(7):1493–1515
Gurtin ME (2000) Configurational forces as basic concepts of
Coleman BD, Gurtin ME (1967) Thermodynamics with internal continuum physics. Springer-Verlag
state variables. J Chem Phys 47(2):597–613 Noll W (1967) Materially uniform simple bodies with inhomo-
Cowin SC, Hegedus DH (1976) Bone remodeling I: theory of geneities. Arch Rational Mech Anal 27:1–32
adaptive elasticity. J Elast 6:313–326 Podio-Guidugli P (2002) Configurational forces: are they nee-
Di Carlo A (2005) Surface and bulk growth unified. In: Stein- ded? Mech Res Commun 29:513–519
mann P, Maugin GA (eds) Mechanics of material forces. Segev R, Epstein M (1996) On theories of growing bodies.
Advances in mechanics and mathematics, vol 11. Springer, In: Batra RC, Beatty MF (eds) Contemporary research
pp 53–64 in the mechanics and mathematics of materials. CIMNE,
Di Carlo A, Quiligotti S (2002) Growth and balance. Mech Res Barcelona, pp 119–130
Commun 29:449–456
Application of invariant integrals to the problems of defect
identification
Robert V. Goldstein · Efim I. Shifrin ·
Pavel S. Shushpannikov
Abstract A problem of parameters identification for Keywords Defect identification · Invariant integrals ·
embedded defects in a linear elastic body using results Spherical inclusion · Explicit formulae
of static tests is considered. A method, based on the use
of invariant integrals is developed for solving this prob-
lem. A problem on identification the spherical inclusion 1 Introduction
parameters is considered as an example of the proposed
approach application. It is shown that the radius, elastic The problems of defect, mainly cracks and cavities,
moduli and coordinates of a spherical inclusion center identification were considered in a number of publi-
are determined from one uniaxial tension (compres- cations. Most methods of defect identification use the
sion) test. The explicit formulae expressing the spheri- surface measurements for bodies subjected to dynamic
cal inclusion parameters by means of the values of cor- forces (see, for example, Bostrom and Wirdelius (1995);
responding invariant integrals are obtained for the case Alves and Ha-Duong (1999); Glushkov et al. (2002);
when a spherical defect is located in an infinite elastic Guzina et al. (2003); Vatuliyan (2004)). The data of sta-
solid. If the defect is located in a bounded elastic body, tic tests are also often used for the defect detection (see,
the formulae can be considered as approximate ones. Keat et al. (1998); Andrieux et al. (1999); Ammari et al.
The values of the invariant integrals can be calculated (2002); Engelhardt et al. (2006)). A review of different
from the experimental data if both applied loads and approaches for solving elastostatic and elastodynamic
displacements are measured on the surface of the body inverse problems is presented by Bonnet and Constan-
in the static test. A numerical analysis of the obtained tinescu (2005). Usually inverse problems are solved as
explicit formulae is fulfilled. It is shown that the formu- follows:
lae give a good approximation of the spherical inclu-
sion parameters even in the case when the inclusion is A defect and its location are described by some para-
located close enough to the surface of the body. meters;
A direct problem is solved by one of the numerical
R. V. Goldstein · E. I. Shifrin (B) · P. S. Shushpannikov
Institute for Problems in Mechanics RAS, Prosp.
methods for the prescribed parameters of the defect
Vernadskogo 101-1, Moscow, 119526 Russia and its location;
e-mail: shifrin@ipmnet.ru An error function, describing the difference between
R. V. Goldstein calculated and experimental data, is constructed;
e-mail: goldst@ipmnet.ru One of the optimization methods is used for the deter-
P. S. Shushpannikov mination of the unknown defect parameters, giving an
e-mail: shushpan@ipmnet.ru extremum for the error function.
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 45
doi: 10.1007/978-1-4020-6929-1_6, © Springer Science+Business Media B.V. 2007
46 R. V. Goldstein et al.
Since the error function can have several extrema, where n(X) = (n1 (X), n2 (X), n3 (X)) is a unit out-
the realization of optimization methods becomes a dif- ward normal to the boundary ∂V at the point X.
ficult problem. In this connection the methods, which The applied loads are self-equilibrated
help determine some defect parameters without using
(f )
the error function optimization, are of great interest ti (X)dS = 0, X ∧ t (f ) (X)dS = 0 (2.3)
(see, Andrieux et al. (1999); Ammari et al. (2002)). ∂V ∂V
In particular, a reciprocity gap principle was used in where ∧ is a vector product.
Andrieux et al. (1999) for a plane crack identification. If the defect G is a cavity or a crack we suppose that
The aims of the present publication are as follows: the boundary ∂G is unloaded
To supplement the reciprocity gap principle approach (f )
σij (X)Nj (X) = 0, X ∈ ∂G (2.4)
with other types of invariant integrals;
To develop an approach for obtaining explicit formu- where N (X) = (N1 (X), N2 (X), N3 (X)) is a unit nor-
lae for the defect parameters in the case when the mal to ∂G at the point X.
sizes of a defect are small as compared to the distance If the defect G is an inclusion, we suppose that G is
between the defect and the body boundary. an isotropic and linear elastic body with unknown shear
modulus µI and Poisson ratio νI . It is supposed also
complete bonding between the matrix and inclusion.
2 Statement of the problem Let us denote by σijI , eij
I and uI = uI , uI , uI the
1 2 3
stresses, strains and displacements of the inclusion. The
Let V be a simply connected domain in a three- mentioned suppositions lead to the following equations
dimensional space R 3 , G ⊂ V is an embedded subdo-
(I ) (I ) (I )
main, = V \G. Let us suppose that is an isotropic eij (X) = ui,j (X) + uj,i (X) /2,
linear elastic body with a shear modulus µM and Pois- X ∈ G, (i = 1, 2, 3; j = 1, 2, 3)
son ratio νM . The defect G can be a cavity, a crack
(I ) νI (I )
or an isotropic linear elastic inclusion. Let us introduce σij (X) = 2µI θ (I ) (X)δij + eij (X) ,
1 − 2νI
Cartesian coordinates OX1 X2 X3 . The stress-strain
3
state in the matrix we’ll mark with the superscript θ (I ) (X) = (I )
ekk (X) (2.5)
(f ) (f )
(f ) : σij is the stress tensor, eij is the strain tensor k=1
(f ) (f ) (f ) (I )
and u(f ) = u1 , u2 , u3 is the displacement vec- σij,j (X) = 0
tor. According to our suppositions the following equal- The bonding conditions have the following form
ities are valid
(f )
uI (X) = u(f ) (X), σijI (X)Nj (X) = σij (X)Nj (X),
(f ) (f ) (f )
eij (X) = ui,j (X) + uj,i (X) /2, X ∈ ∂G (2.6)
νM occupies the domain V . A regular elastic field in the the superscript (f ) + (r). Because the invariant inte-
(r) (r)
body we’ll mark by a superscript (r)(σij , eij , u(r) = grals for the regular elastic fields are equal to zero the
(r) (r) (r) following equalities are valid
(u1 , u2 , u3 )). Consider an integral
(f )+(r) (f ) (f )
Ji = Ji + Jiint (r)
(f ) (r) (r) (f )
RG(f ) (r) = ti ui − ti ui dS (3.1) (f )+(r) (f ) (f )
Li = Li + Liint (r) (3.3)
S (f )
(f )+(r) (f )
M =M + Mint (r)
(r) (r)
where S ⊂ is a closed surface, ti (X) = σij where the integrals describing the interaction between
(X)nj (X), n(X) = (n1 (X), n2 (X), n3 (X)) is a unit the applied and regular elastic fields have the following
outward normal to S. form
If the surface S doesn’t contain the domain G inside
(f ) (f ) (r) (f ) (r) (r) (f )
then RG(f ) (r) = 0, otherwise the values RG(f ) (r) can Jiint (r) = σkl ekl ni − tj uj,i − tj uj,i dS
differ from zero and give some information about the S
defect G. In the case when the loads t (f ) and displace- (f ) (f ) (r) (f ) (r)
Liint (r) = εij k σmn emn Xk nj + tj uk
ments u(f ) are available on the surface of the body ∂V ,
S
it is possible to take S = ∂V and for all known regu-
(r) (f ) (f ) (r) (f )
lar fields the values RG(f ) (r) can be calculated. It was +tj uk − tp up,j Xk − tp(r) up,j Xk dS
shown in Andrieux et al. (1999) that it is possible to (3.4)
reconstruct a plane crack using the appropriate regular
(f ) (f ) (r) (f ) (r)
fields. Mint (r) = σkl ekl Xi ni − tj uj,i Xi
It is well-known that for isotropic linear elastic solids S
the following invariant integrals are valid (see, Knowles
µM − µI
B= (4.5)
x1 = r sin θ cos ϕ, x2 = r sin θ sin ϕ, x3 = r cos θ 8µM [(7 − 5νM ) µM + (8 − 10νM ) µI ]
(4.2) (1−2νI) (6−5νM) 2µM +(3+19νI −20νI νM ) µI
A = −B +D
2 (1−2νI ) µM +(1+νI ) µI
(4.6)
Solution of the problem for a spherical inclusion in (1 − νM − 2νI νM ) µI − (1 − 2νI ) (1 + νM ) µM
D=
an infinite elastic solid under uniaxial tension (com- 4µM (1 + νM ) [2 (1 − 2νI ) µM + (1 + νI ) µI ]
pression) in the direction of the axis x3 was obtained
by Goodier (1933). According to Goodier (1933) the For the spherical inclusion identification we’ll use
solution of the problem outside the inclusion has the the following regular elastic fields with constant, linear
following form and quadratic stresses. Below the stress tensors and dis-
placements vectors are presented in the initial Cartesian
coordinates OX1 X2 X3
σ a3 3a 2 B 9a 2 ⎛ ⎞
u(3) = 2 − A − 2 + 5 (5 − 4νM ) − 2 σ 00
r
r r r σ (C1) = ⎝ 0 0 0 ⎠ , u(C1)
σr 0 00
cos 2θ · B + [(1 − νM ) ⎛ ⎞
4µM (1 + νM ) X1
σ ⎝ −νM X2 ⎠
+ (1 + νM ) cos 2θ ] (4.3) =
2µM (1 + νM )
−νM X3
⎛ ⎞
00 0
σ (C2) = ⎝ 0 σ 0 ⎠ , u(C2)
00 0
−2Bσ a 3 sin 2θ 3a 2 ⎛ ⎞
u(3)
θ = 5 (1 − 2νM ) + −νM X1
r2 r2 σ
σr = ⎝ X2 ⎠ (4.7)
− sin 2θ, u(3)
ϕ =0
2µM (1 + νM )
4µM −νM X3
⎛ ⎞
(3) 2µM σ a 3 12a 2 000
σrr = 2A + −10νM + 2 B σ (C3) = ⎝ 0 0 0 ⎠ , u(C3)
r3 r
00σ
36a 2 ⎛ ⎞
+ 10 (νM − 5) + 2 B cos 2θ −νM X1
r σ
σ = ⎝ −νM X2 ⎠
+ (1 + cos 2θ ) 2µM (1 + νM )
2 X3
Application of invariant integrals to the problems of defect identification 49
⎛ σ X1 −σ X2 ⎞ (f )
L L 0 r, M 0 and Mint r, M 0
σ (L1) = ⎝ −σ X2
0 ⎠ , u(L1)
L 0
0 0 0 (f ) (f ) (r) (f ) (r) (r)
Liint r, M 0 = εij k σmn emn xk nj +tj uk + tj
σ
= S
4LµM (1 + νM )
⎛ 2 ⎞ (f ) (f ) (r) (f )
× uk −tp up,j xk−tp(r) up,j xk dS
X1 − (2 + νM ) X22 + νM X32
⎝ −2νM X1 X2 ⎠ (4.10)
−2νM X1 X3
⎛ −σ X1 ⎞
0 L 0 (f ) (f ) (r) (f ) (r) (r)
Mint r, M =
0
σkl ekl xi ni − tj uj,i xi − tj
σ (L2) = ⎝ −σLX1 σ LX2 0 ⎠ , u(L2)
S
0 0 0
σ (f ) 1 (f ) (r) 1 (r) (f )
= uj,i xi − ti ui − ti ui dS
4LµM (1 + νM ) 2 2
⎛ ⎞
−2νM X1 X2 There is a simple connection between the integrals in
⎝ − (2 + νM ) X2 + X2 + νM X2 ⎠ (4.8) different coordinates. From (3.4) and (4.10) one has
1 2 3
−2νM X2 X3
⎛ ⎞ (f ) (f ) (f )
Liint (r, O) = Liint r, M 0 + εij k xk0 Jj int (r) (4.11)
0 0 −σLX1
σ (L3) = ⎝0 00 ⎠ , u(L3)
−σ X1
L 0 σ LX3
(f ) (f ) (f )
σ Mint (r, O) = Mint r, M 0 + xi0 Jiint (r) (4.12)
=
4LµM (1 + νM )
⎛ ⎞ All the interaction integrals (3.1), (3.4) and (4.10) for
−2νM X1 X3
⎝ ⎠ the applied elastic field (3) and regular elastic fields
−2νM X2 X3
(Ci), (Li) and (Q) can be calculated analytically. For
− (2 + νM ) X1 + νM X2 + X3
2 2 2
the calculation of the integrals they are written in the
coordinates M 0 x1 x2 x3 and the sphere ∂G is taken as a
surface S. In the expressions (4.7)–(4.9) the coordinates
Xi are replaced by xi + xi0 according to (4.1) and the
⎛ σ X2 ⎞
3
0 0 applied elastic field (4.3), (4.4) is transformed from
⎜ L2 ⎟ (Q) the spherical to Cartesian coordinates. The analytical
σ (Q) = ⎜ νM σ X32 −X1 ⎟,u
2
⎝0 L2
0 ⎠ expressions for some of interaction integrals (3.1), (3.4)
−σ X12
0 0 and (4.10) are as follows
L2
⎛ ⎞
(1 − νM ) X1 X32 + νM X13 /3 (3)
Jiint (Ck) = 0
σ
⎝ ⎠
(4.13)
= 0
2µM L2
− (1 − νM ) X1 X3 + νM X3 /3
2 3
It follows from (4.11)–(4.13)
(4.9)
L(3) (3)
iint (Ck, O) = Liint Ck, M
0 (3)
, Mint (Ck, O)
(3)
whereL is a typical linear size of the domain V . = Mint Ck, M 0 (4.14)
(f ) (f )
It is necessary to note that the integrals Ji , Jiint (r)
and RG(f ) (r) do not depend on the location of the
4π (1 − νM ) a 3 σ 2 (A + 15B)
origin of coordinates. At the same time the values of
(f ) (f ) (f )
RG(3) (C1) = ,
Li , Liint (r), M (f ) and Mint (r) depend on the loca- 1 + νM
tion of the origin of coordinates. In this connection RG(3) (C2) = RG(3) (C1) ,
we’ll denote below the integrals (3.4) in coordinates 4π (1−νM ) a 3 σ 2 [A−5 (1+4νM ) B]
(f ) (f )
OX1 X2 X3 by Liint (r, O) and Mint (r, O). The inte- RG(3) (C3) =
1+νM
(f )
grals in coordinates M 0 x1 x2 x3 we’ll denote by Liint (4.15)
50 R. V. Goldstein et al.
(3) 4π (1 − νM ) a 3 σ 2 (A + 15B) x10 Let us also note that
RG (L1) =
1 + νM L (3)
Mint
3
Lk, M 0 = − RG(3) (Lk) ; k = 1, 2, 3
2
(3) 4π (1 − νM ) a 3 σ 2 (A + 15B) x20 (4.22)
RG (L2) =
1 + νM L
To determine a radius and elastic moduli of the inclu-
(4.16) sion we’ll use the regular elastic field (4.9) with
4π (1 − νM ) a σ [A − 5 (1 + 4νM ) B]
3 2
quadratic stresses. Consider some interaction integrals
RG(3) (L3) =
1 + νM for the applied elastic field (3) and the regular field (Q)
x30 given by Eq. (4.9)
2 (1 − ν ) π a 3 σ 2
L (3) M
Mint Q, M 0 =
L 2
It follows from (4.15) and (4.16) that in the case when 2
A + 15B = 0 and A − 5 (1 + 4νM ) B = 0 the coordi- × −28Ba 2 + 3 (A − 5B) x10
nates of the center M 0 of the ball G can be expressed 2
by means of invariant integrals −3 (A + 15B) x30 (4.23)
x10 RG(3) (L1) x20 RG(3) (L2)
= , = ,
L RG(3) (C1) L RG(3) (C2) −4 (1 − νM ) π a 3 σ 2
RG(3) (Q) = − 4Ba 2
x30 RG(3) (L3) L2
= (4.17) 2
L RG(3) (C3)
+ (A − 5B) x10
Let us note that displacements in the regular fields
2
(4.7)–(4.9) are homogeneous vector-functions in the − (A + 15B) x30 (4.24)
coordinates OX1 X2 X3 . It is possible to prove the fol-
lowing proposition. It follows from (4.23) and (4.24)
u(r) (X) is a homogeneous
If (r) vector-function of order (3)
2Mint Q, M 0 + 3RG(3) (Q)
m u (kX) = k m u(r) (X) then the following equality
is valid
−64 (1 − νM ) π a 5 σ 2 B
1 = (4.25)
(f )
Mint (r, O) = − m + RG(f ) (r) (4.18) L2
2 From (4.15) and (4.19) one has
In particular (3) (3)
Mint (C3, O) − Mint (C1, O) = 120
(3) 1 (1 − νM ) π a 3 σ 2 B
Mint (r, O) = − m + RG(3) (r) (4.19) (4.26)
2
It is interesting to note that the sign of the expres-
From (4.17) and (4.19) it follows that coordinates of (3) (3)
sion Mint (C3, O)−Mint (C1, O) given by Eq. (4.26),
the defect center can be expressed also in terms of M-
coinciding with the sign of the constant B (Eq. (4.5)),
integrals
indicates the matrix or the inclusion is more stiff. If
(3) (3) (3) (3)
x10 3Mint (L1, O) x20 3Mint (L2, O) Mint (C3, O) − Mint (C1, O) > 0 then µM > µI , if
= (3)
, = (3)
, (3) (3)
L 5Mint (C1, O) L 5Mint (C2, O) Mint (C3, O) − Mint (C1, O) < 0 then µM < µI .
(3) (3)
(3) Let us suppose that Mint (C3, O)−Mint (C1, O) =
x30 3Mint (L3, O)
= (3)
(4.20) 0. In this case from (4.23) and (4.24) one has
L 5Mint (C3, O)
(3)
0 + 3RG(3) (Q)
a 2 15 2M int Q, M
The coordinates can be also calculated by other types =− (4.27)
of integrals, for example L 2 8 M (C3, O) − M (3) (C1, O)
(3)
int int
−RG(3) (L1) −RG(3) (L2)
x10 = , x20 = , It follows from (4.12), (4.19) and (4.27)
(3)
J1int (L1)
(3)
J2int (L2)
2RG (3) (Q) + x 0 J (3) (Q)
−RG(3) (L3) a 2 15 i iint
= (4.28)
x30 = (3)
(4.21) L2 4 M (3) (C3, O) − M (3) (C1, O)
J3int (L3) int int
Application of invariant integrals to the problems of defect identification 51
Using (4.28) in (4.26) one obtains an expression for the It follows from (4.35) and (4.36) that if a = 0 then
(3)
constant B Mint (Ci, O) = 0, i = 1, 2, 3. In this case the radius
(3) (3) of the defect can be calculated from (4.35)
Mint (C3, O) − Mint (C1, O)
B= (4.29)
120 (1 − νM ) π a 3 σ 2 (3)
After the calculation of the value B one can calculate 2µM (1 + νM ) (7 − 5νM ) Mint (C3, O)
a3 = (4.37)
the shear modulus µI using (4.5) 3π (1 − νM ) (9 + 5νM ) σ 2
µI 1 − 8µM (7 − 5νM ) B
= (4.30) The coordinates of the cavity center are calculated by
µM 16µM (4 − 5νM ) B + 1 means of the formulae (4.17), (4.20) or (4.21) and the
It follows from (4.15) and (4.19) problem of the spherical cavity identification is com-
(3) (3)
(1 + 4νM ) Mint (C1, O) + 3Mint pletely solved.
(C3, O) = −24π (1 − νM ) a 3 σ 2 A (4.31) In the case of a rigid inclusion (µI → +∞) the val-
ues A and B according to (4.5) and (4.6) have the form
From (4.29) and (4.31) one has
5B (1 + 4νM ) Mint (3) (3)
(C1, O) + 3Mint (C3, O) −1
B= ,
A= 16µM (4 − 5νM )
(3) (3)
Mint (C1, O) − Mint (C3, O) 19 − 33νM + 20νM 2
(4.32) A= (4.38)
16µM (4 − 5νM ) (1 + νM )
Finally one obtains an expression for the Poisson ratio
νI of the inclusion It follows from (4.15), (4.18) and (4.38)
νI = R/S (4.33) (3) (3)
Mint (C1, O) = Mint (C2, O)
where 2
−3π (1 − νM ) 5νM − 12νM + 1 a 3 σ 2
R = 4µM (1 + νM ) {(2µM + µI ) A = (4.39)
2µM (4 − 5νM ) (1 + νM )2
+ [2µM (6 − 5νM ) + 3µI ] B}
+ (1 + νM ) µM − (1 − νM ) µI
2
S = −4µM (1 + νM ) {(µI − 4µM ) A (3) −3π (1−νM ) 5νM −νM +3 a 3 σ 2
Mint (C3, O) =
+ [−4µM (6 − 5νM ) + µI (19 − 20νM )] B} µM (4−5νM ) (1+νM )2
+2 [(1 + νM ) µM − νM µI ] (4.40)
(3)
Consider some special cases of the considered problem. −µM (4 − 5νM ) (1 + νM )2 Mint (C3, O)
Let us suppose that G is a spherical cavity (µI → 0). In a3 = 2
3π (1 − νM ) 5νM − νM + 3 σ 2
this case the formulae (4.5) and (4.6) have the following
(4.41)
form
1 10νM − 13 (3)
B= , A= (4.34) Let us note that according to (4.39) the values Mint
8µM (7 − 5νM ) 8µM (7 − 5νM ) (3)
(C1, O) and Mint (C2, O) can become zero when νM
Using (4.34), (4.15) and (4.18) one has 2 − 12ν + 1 = 0
is a root of the quadratic equation 5νM M
(3) 3π (1 − νM ) (9 + 5νM ) a 3 σ 2 in the interval (0, 1/2), more precisely νM = νM 0 ≈
Mint (C3, O) = (4.35)
2µM (1 + νM ) (7 − 5νM ) 0.0864. So if νM = νM then the coordinates of the
0
fields with linear stresses. Let us take for an example how close to the boundary of the body can be a defect
⎛ σ X1 −σ X2 ⎞ so that the formulae will still be applicable.
0
L L Let OX1 X2 X3 is the Cartesian coordinate system.
σ (L4) = ⎝ −σLX2 0 0 ⎠ , u(L4)
νM σ X1
As an example consider the cube domain V : |Xi | ≤
0 0 L 10, i = 1, 2, 3. The Poisson ratio of the matrix is
⎛ ⎞
(1 − νM ) X12 + (νM − 2) X22 νM = 0.25. We consider below two types of the defects
σ ⎝ ⎠
= −2νM X1 X2 G: (1) G is a spherical cavity of the radius 1; (2) G is a
4µM L
0 rigid inclusion of the radius 1. The applied loads corre-
(4.42) spond to the uniform uniaxial tension in the direction
⎛ −σ X1 ⎞ of the axis X3 : t (3) (X1 , X2 , 10) = (0, 0, σ ) , t (3)
0 L 0
σ (L5) = ⎝ −σLX1 σ LX2 0 ⎠ , u(L5) (X1 , X2 , −10) = (0, 0, −σ ) , t (3) (±10, X2 , X3 ) =
νM σ X2 t (3) (X1 , ±10, X3 ) = (0, 0, 0) . For different locations
0 0 L
⎛ ⎞ of the defect center the direct problem was solved by
−2νM X1 X2 the FEM and the elastic field on the surface ∂V was
σ ⎝
= (νM − 2) X12 + (1 − νM ) X22 ⎠ calculated. After that the invariant integrals were cal-
4µM L
0 culated and the defect parameters were determined by
The calculations lead to the following expressions the formulae (4.17) and (4.37) for a cavity and by the
formulae (4.17) and (4.41) for a rigid inclusion. We
π (1 − νM ) (1 − 2νM ) (1 − 5νM ) a 3 σ 2 took L = 10 for the linear regular elastic fields.
RG(3) (L4) =
µ (1 + νM ) (4 − 5νM )
M
The results of the calculations are presented in the
x10 Tables 1 and 2, respectively. The numerical results show
L that obtained explicit formulae give a good approxima-
tion to the inverse problem solution for a spherical cav-
π (1 − νM ) (1 − 2νM ) (1 − 5νM ) a 3 σ 2
RG(3) (L5) = ity and a rigid inclusion even when the defect is located
µ (1 + νM ) (4 − 5νM ) close enough to the boundary of the body.
M
x20 Besides the limiting cases of a cavity and a rigid
(4.43) inclusion we considered also the case of elastic spheri-
L
cal inclusion. The calculations show that formulae
It follows from (4.15), (4.38) and (4.43) that in the case (4.17) also give very accurate results for the coordinates
when νM is equal or close to νM 0 the coordinates x 0 and
1 of a defect center in the case of elastic inclusion. At the
x2 can be calculated from the following formulae
0
same time formulae (4.28)–(4.33) used for the calcu-
2 lation of the radius and elastic moduli of the inclusion
x10 2 5νM − νM + 3 RG(3) (L4)
= are much more sensitive to the errors in the boundary
L (1 − 2νM ) (1 − 5νM ) (1 + νM ) RG(3) (C3) data and accuracy of the invariant integrals calcula-
(4.44) tions. To improve the accuracy of the calculations one
2
x20 2 5νM − νM + 3 (3)
RG (L5) can use other formulae containing other types of invari-
=
L (1 − 2νM ) (1 − 5νM ) (1 + νM ) RG(3) (C3) ant integrals and regular elastic fields. In this connec-
tion the problem arises to find the formulae which are
less sensitive to the boundary data errors and accuracy
5 Numerical analysis of the obtained formulae of invariant integrals calculations.
the results of one uniaxial tension (compression) sta- boundary data and accuracy of the invariant integrals
tic test are obtained. These formulae are exact for the calculation.
infinite elastic solids and approximate for the bounded
elastic bodies. Acknowledgements E. Sh and P. Sh. gratefully acknowledge
the support of RFBR grant 07-01-00448.
Numerical analysis of the formulae is fulfilled. The
results of the analysis show that the formulae give a
good approximation of the spherical cavity and spher-
ical rigid inclusion parameters even in the case when References
a defect is close enough to the body boundary. If a
defect is a spherical elastic inclusion then its location Alves CJS, Ha-Duong T (1999) Inverse scattering for elastic
is also determined very accurately. It is required to find plane cracks. Inverse Probl 15:91–97
Ammari H, Kang H, Nakamura G, Tanuma K (2002) Complete
the formulae for determination of the radius and elastic asymptotic expansions of solutions of the system of elasto-
moduli of the inclusion less sensitive to the errors in the statics in the presence of an inclusion of small diameter and
detection of an inclusion. J Elast 67:97–129
54 R. V. Goldstein et al.
Andrieux S, Ben Abda A, Bui H (1999) Reciprocity principle Guzina B, Nintcu Fata S, Bonnet M (2003) On the stress-wave
and crack identification. Inverse Probl 15:59–65 imaging of cavities in a semi-infinite solid. Int J Solids Struct
Bonnet M, Constantinescu A (2005) Inverse problems in elastic- 40:1505–1523
ity. Inverse Probl 21:R1–R50 Keat W, Larson M, Verges M (1998) Inverse method of identi-
Bostrom A, Wirdelius H (1995) Ultrasonic probe modeling and fication for three-dimensional subsurface cracks in a half-
nondestructive crack detection. J Acoust Soc Am 97(5 space. Int J Fract 92:253–270
Pt.1):2836–2848 Knowles JK, Sternberg E (1972) On a class of conservation laws
Engelhardt M, Schanz M, Stavroulakis G, Antes H (2006) Defect in linearized and finite elastostatics. Arch Rational Mech
identification in 3-D elastostatics using a genetic algorithm. Anal 44(3):187–211
Optim Eng 7:63–79 Vatuliyan AO (2004) On the determination of crack configuration
Glushkov EV, Glushkova NV, Ehlakov AV (2002) Mathematical in anisotropic medium. Appl Math Mech (PMM) 68(1):
model of the ultrasonic defectoscopy of spatial cracks. Appl 180-188 (in Russian)
Math Mech (PMM) 66(1):147-156 (in Russian)
Goodier JN (1933) Concentration of stress around spherical and
cylindrical inclusions and flaws. J Appl Mech APM-55-7
39–44
On application of classical Eshelby approach
to calculating effective elastic moduli of dispersed composites
K. B. Ustinov · R. V. Goldstein
Abstract The problem of finding effective elastic to be distributed uniformly and isotropic in space. The
moduli of media with spheroid inclusions in case of main goal of the paper is to demonstrate the influence
small concentration of these inclusions is addressed. A of two parameters—ratio of shear moduli of the inclu-
number of particular solutions, both known and new, sions and aspect ratio (ratio of the major and minor
were obtained as limit transitions and asymptotical axes)—on the effective elastic properties of the com-
expansion of the general solution, based on Eshelby’s posite. A special attention is devoted to limiting cases,
approach. A special attention was paid to determining when both parameters became simultaneously extre-
the ranges of applicability of the obtained asymptoti- mely small (large).
cal solutions. It was shown that for spheroid inclusions The general algorithm of calculating the effective
the areas of applicability of the asymptotic solutions elastic characteristics of media with inclusions of ano-
are determined by two parameters: the ratio of elastic ther material is known (Eshelby 1957; Mura 1982). The
moduli of the inclusion and the matrix and aspect ratio modern state of the problem may be found e.g., in the
of the inclusions. following works: Tucker and Liang (1999), Odegard et
al. (2002), Ustinov (2003a). Below the brief account is
Keywords Effective properties · Eshelby’s tensor · given. Let the Hooke law is satisfied for each point, r ,
Inclusion of the media
σ (r ) = (r ) ε (r ) (1)
1 Introduction: general relations
Where σ and ε are stress and strain tensors, respecti-
The problem of finding the effective elastic characteris- vely. is the 4th rank tensor of elasticity, which can
tics of heterogeneous media was addressed in numerous be expressed as follows
papers. The current work is restricted to considering
(r ) = 0 + (∗ − 0 ) θ (r ), (2)
the case of dilute suspension of one phase (inclusions)
within the material of the other phase (matrix), both where θ (r ) = 1 inside the inclusions, and θ (r ) = 0
phases supposed to be isotropic, the inclusions in the outside the inclusions. All the values corresponding to
forms of spheroid or three-axial flat ellipsoid supposed the matrix are marked with index 0, the values corres-
ponding to the inclusions are marked with asterisk. The
K. B. Ustinov (B) · R. V. Goldstein
Institute for Problems in Mechanics, Prospect Vernadskogo
direct (indexless) notation is used whenever it is pos-
101 Moscow RF, Moscow, 119526 Russian Federation sible; the averaged values are denoted with the bar over
e-mail: ustinov@ipmnet.ru the corresponding symbol.
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 55
doi: 10.1007/978-1-4020-6929-1_7, © Springer Science+Business Media B.V. 2007
56 K. B. Ustinov, R. V. Goldstein
Substituting (2) into (1) and averaging over the that expression (9) remains valid for arbitrary aniso-
volume, one has tropic matrix and aligned inclusions (also anisotropic),
___ ___ ___ the expression for Eshelby’s tensor in case of gene-
σ = 0 ε + (∗ − 0 ) εincl , (3)
_____ ral anisotropy, however, has to be obtained. In case of
where εincl are the stains, averaged over the volume anisotropic inclusions in isotropic matrix no additional
of inclusions; is the volumetric concentration of the work is required.
inclusions. On the other hand In case of arbitrary distribution of the inclusions over
___ ___
σ = e f f ε , (4) their orientation in space, their influence remains addi-
tive due to linearity of the case of small concentration,
where e f f is the tensor of effective moduli to be
_____ _____ and the effective properties may be calculated by inte-
found. Therefore, on expressing εincl in terms of ε
gration of (9) over all possible orientations with the
the problem may be regarded as solved.
appropriate weight functions. However more elegant
That may be done for ellipsoid inclusions. Thus,
way may be used. Consider the isotropic space distri-
equate the stresses within a single inclusion in the
bution of the inclusions in isotropic matrix; the effec-
matrix subjected to uniform field, ε0 , with the stresses
tive medium has to remain isotropic with its tensor of
within an equivalent inclusion (the same as in the mate-
elasticity expressed in terms of Lame constants by well-
rial of matrix) subjected to eigenstrains, ε∗ , (Eshelby
known formula
1957, 1961; see also Mura 1982),
i jkl = λδi j δkl + µ δik δ jl + δil δ jk (10)
∗ εincl = 0 (εincl − ε∗ ) , εincl = ε0 + εadd (5)
On the other hand, since each symmetric 4th rank tensor
where εadd is the additional, or constrained, strain in
has 2 linear invariants, the elastic constants may be
the equivalent inclusion, which is related to Eshelby’s
expressed in terms of these invariants
tensor as follows
2 1 1
εadd = Sε∗ (6) k = λ + µ = iii j j µ = 3i ji j − iii j j
3 9 30
The expression for the components of Eshelby’s ten- (11)
sor in case of isotropic matrix is known (Eshelby 1957,
Thus for the case under consideration
1961; see also Mura 1982). For the case of spheroid
inclusions they are given in Appendix I. Solving the ke f f = k0 + Diii j j
system of (5), (6) yields 9
−1
µe f f = µ0 + 3Di ji j − Diii j j (12)
εincl = I + S0 −1 (∗ − 0 ) ε0 (7) 30
The idea of using such invariants for calculating the
Here I stands for the unit 4th rank tensor Ii jkl = effective characteristics has been known at least since
2 (δik δ jl + δil δ jk ). The inverse of any 4th rank tensor
1
Kroner’s work (1958).
A is understood in the sense that A−1 A = A A−1 = I
(e.g. Walpole 1969). Remark 1
In case of a small concentration of the aligned inclu- As it has been noted, due to linear approximation, that
sions, externally applied strain, ε0 , may be considered appears naturally in case of small concentration of the
___
to be equal to ε , and accounting this, substituting (7) inclusions, the influence of each inclusion is additive.
into (3) gives Obviously, the influence of each inclusion on the inva-
___ ___
−1 ___ riants of the tensor of elasticity is independent of its
σ = 0 ε + (∗ − 0 )−1 + S−1 0 ε (8) orientation in space. Hence, the invariants of elastic
The comparison of (8) and (4) finally yields tensor remain the same for an arbitrary distribution of
−1 the inclusions in space.
e f f = 0 + D, D = (∗ −0 )−1 + S−1 0
Remark 2
(9) The considered case of a small concentration has both
In case of spheroid inclusion with semi-axes a1 , the inherent significance and may serve as a basis for
a2 = a1 , a3 , = 43 πa12 a3 N , where N is the num- more advanced approaches, e.g. differential self-
ber of inclusions per unit of volume. It is worth to note consistent method (Salganik 1973; Roscoe 1973).
On application of classical Eshelby approach 57
therefore, solutions for particular cases are of impor- The other elastic coefficients may be found by means
tance. A number of such solutions are known, which of well-known formulae of the theory of elasticity. This
describe the following cases: result coincides with the known results by Wu (1966).
- thin disk-like inclusions of arbitrary rigidity However, it is seen that the above formulae are no
(Wu 1966); longer valid for the unlimited increase or decrease of
- inclusions in the form of sphere of arbitrary rigidity elastic moduli of the inclusions. On performing limit
(Krivoglas and Cherevko 1959); transition a3 /a1 → 0 one may not be allowed to consi-
- thin needle-like inclusions of arbitrary rigidity (Wu der any parameters to be smaller then this ratio. Hence
1966); these formulae do not obey to (but still may) cover the
- thin disk-like inclusions of zero rigidity (cracks) range of parameters exceeding the following
(Walsh 1965); a3 µ∗ a1 a3
<< << ; << 1 (15)
- thin disk-like inclusions of infinite (very high) rigi- a1 µ0 a3 a1
dity (Kovalenko and Salganik 1977).
2.2.2 Long spheroids
However, some of these solutions contradict each other,
and neither of them can be obtained from another by Substituting (A1), (A2) into (9), and then result into
limit transition. Below, these solutions will be obtained (12), and performing limit transition a3 /a1 → ∞ yields
as the limit transitions and asymptotic expansions of the
(k∗ − k0 ) (3k0 + µ∗ + 3µ0 )
general solution, because all the cases may be treated ke f f = k0 1 + (16)
as the media with spheroid inclusions. The problem is k0 (3k∗ + µ∗ + 3µ0 )
3k0 (µ∗ − µ0 )
µe f f = µ0 1 + ×
5µ0 (µ∗ + µ0 ) (3k∗ + µ∗ + 3µ0 ) 3k0 µ∗ + 3k0 µ0 + 7µ∗ µ0 + µ20
× (µ∗ + µ0 ) (4µ0 (3µ∗ + 7µ0 ) + 3k∗ (µ∗ + 9µ0 )) (17)
+ µ0 8µ0 9µ2∗ + 23µ∗ µ0 + 8µ20 + 3k∗ 7µ2∗ + 52µ∗ µ0 + 21µ20
58 K. B. Ustinov, R. V. Goldstein
This result coincides with the known results by Wu 2.3.2 Penny-shaped extremely rigid inclusions
(1966). However, similar to the case of disk-like
inclusions, it should be kept in mind that the area of Similar to the case of crack, there exists a solution for
applicability of this solution is also restricted, although flat rigid inclusions that can not be obtained by limit
remaining wider transition µµ0 → ∞ in (13), (14), but may be obtained
∗
a1
2.4 Non-uniform limit transitions and areas lg
a3
of applicability of the asymptotics a) Disks
4
In the previous sections asymptotic formulae for calcu-
lating effective elastic moduli of media with inclusions
2
were obtained for some particular important cases.
Rigid
However, the question arises: what are the numeri- Soft
cal values of the parameters aa31 , and µ ∗
µ0 , for which -4 -2 2 4 lg
k*
the considered asymptotics are valid with acceptable k0
0
theoretical estimations.
-2
Needles
µ0 a3
∼ << 1 (30) 2.5.1 Flat soft inclusions
µ∗ a1
corresponding to the rigid disk-like inclusions, or Let us introduce variable, d, interrelating the introdu-
n ced small parameters so that
µ0 a1
∼ << 1 (31) µ∗ a3
µ∗ a3 =d (32)
µ0 a1
corresponding to the rigid needle-like inclusions, the Substituting (A1), (A2) into (9), and then result into
obtained asymptotics fail to produce acceptable results. (12), accounting for (32) and performing limit transi-
Consider these cases in more details. tion a3 /a1 = ρ → 0 one, after the back substitution
60 K. B. Ustinov, R. V. Goldstein
of d by means of (32) in terms of the introduced small The formulae for extremely rigid flat inclusions (23),
parameters, obtains the following expressions (24) may be obtained from here by limit transition
⎡ ⎤
4k0 1 − ν02 (1 + ν∗ )
k e f f = k 0 ⎣1 − ⎦ (33)
3 4k∗ 1 − ν02 (1 − ν ∗ ) + π aa31 k0 (1 − 2ν0 ) (1 + ν∗ )
⎡ ⎤
16 (1 − ν0 ) 4µ∗ (1 − ν0 ) (4 − 5ν∗ ) + µ0 (5 − ν0 ) (1 − 2ν∗ ) π aa31
µe f f = µ0 ⎣1 − ⎦ (34)
30 4µ∗ (1 − ν0 ) + µ0 (2 − ν0 ) π aa31 µ∗ (1 − ν0 ) (1 − ν∗ ) + µ0 (1 − 2ν0 ) π aa31
other hand, the result of limit transition aa31 → 0 of 2.5.3 Needle-like rigid inclusions
these formulae coincides with the result of limit transi-
tion µ
µ0 → 0 of formulae for flat spheroids (13), (14).
∗
As it is seen from Fig. 1, the range of applicability of
This means that the corresponding limit transitions are the asimptotics for intermediate and small rigidity of
uniform. needle-like inclusions is wider then in the other cases.
This suggests another parameter, d, interrelating the
main small parameters, namely
n
2.5.2 Flat rigid inclusions µ0 a3
=d (38)
µ∗ a1
For this case variable, d, interrelating the main small where n is the power to be determined from the condi-
parameters are introduced as follows tion of the uniform limit transition of the formulae to
µ0 a3 be obtained into (27), (28) for d = 0, and into (16),
=d (35) (17) for d → ∞.
µ∗ a1
Substituting (A1), (A2) into (9), and then result into
Substituting (A1), (A2) into (9), and then result into (12), accounting for (38) and performing limit transi-
(12), accounting for (35) and performing limit transi- tion aa31 → ∞ one, after the back substitution of d by
tion a3 /a1 = ρ → 0 one, after the back substitution means of (38) in terms of the introduced small parame-
of d by means of (35) in terms of the introduced small ters, obtains the following expressions (for n = 2)
parameters, obtains the following expressions
⎡ ⎤
8µ∗ (1−ν0 ) 16 (1−ν0 ) (7−5ν∗ ) +π µ ∗ a3
µ0 a1 (43−56ν 0 ) (1+ν∗ )
µe f f = µ0 ⎣1+ ⎦ (37)
15 16 (1−ν0 ) +π µ0 a1 (7−8ν0 ) 8 (1−ν0 ) (1−ν∗ ) +π µ
µ∗ a3 ∗ a3
µ0 a1 (3−4ν 0 ) (1+ν∗ )
On application of classical Eshelby approach 61
⎡ ⎤
⎢ 2k∗ (1 − ν0 ) (1 − 2ν0 ) (1 − 2ν∗ ) ⎥
k e f f = k 0 ⎣1 + 2 ⎦ (39)
6k0 (1 − ν0 ) (1 − 2ν0 ) + 3k∗ aa13 (1 − 2ν∗ ) (1 + ν0 ) 4 (1 − ν0 ) ln 2a3
a1 −1 −1
⎡ ⎤
⎢ 4µ∗ (1 − ν0 ) (1 + ν∗ ) ⎥
µe f f = µ0 ⎣1 + 2 ⎦ (40)
30µ0 (1 − ν0 ) + 15µ∗ aa13 (1 + ν∗ ) 4 (1 − ν0 ) ln 2a3
a1 −1 −1
Besides the above considered cases one may outline 3.1 Inclusions of intermediate rigidity
cases of rigid (absolutely rigid) and soft (voids) inclu-
sions. Such solutions may be obtained by limit transi- The result of substitution of (A9) into (9) and then into
tions µ µ∗
µ0 → ∞ and µ0 → 0, respectively, in the result
∗
(12) on limit transition a3 → 0 coincides with (13),
of substitution of (A1), (A2) into (9), and then into (14). This case corresponds obviously to the range of
(12). The final formulae are not presented here because
of its awkwardness. Distinguishing these cases appears a1
to have theoretical value only, because for the numeri- lg
a3
cal calculation one can use successfully the exact
solution (A1), (A2), (9), (12), and for analytical mani- Disks
pulation the simplifications yielding by them are not
significant to make the results more transparent com-
paring to the general case.
conditions (41) were used essentially while deriving the Fig. 3 Influence of the ratio of major and intermediate axes
formulae. of the cracks on the relative change of effective elastic moduli.
Solid line corresponds to volumetric modulus, dashed line does
to shear modulus, dotted-dashed line does to Young modulus
3.2 Extremely soft inclusions
(20). Using the well-known formulae of the theory of
Let us generalize the case of extremely soft inclusions elasticity the expression for the Young modulus may
(22) for the elliptical in plane inclusions. The effec- be written
tive moduli for this case may be obtained by substi-
a2 2(1 − ν02 )
tuting (A9) into (9) and then into (12) followed by E e f f = E 0 − E 0
a1 15E (t)
limit transition µ∗ → 0 and then a3 → 0. The result
3 a12 −a22
coincides with the result obtained by Hashin (1988). × +
Taking into account that the volume and the maximal E (t) (1−ν0 )a12 −a22 E (t) +ν0 a22 K (t)
area of the cross-section of the considered ellipsoid are a12 − a22
Ve = 43 πa1 a2 a3 , S = πa1 a2 , the expressions for the + 2
a1 − (1 − ν0 )a22 E (t) − ν0 a22 K (t)
effective elastic moduli may be written as
(46)
a2 2(1 − ν02 ) Figure 3 represents the influence of the shape of
k ef f
=k 1−
0
(42)
3 a1 (1 − 2ν0 )E(t) the ellipsoids in plane on the changes in the effective
moduli. The abscise axis corresponds to the ratio of
a
0 2 major and intermediate axes of the ellipsoid (the minor
µef f
=µ −µ 0
(1 − ν0 )
5 a1 axis corresponds to the thickness and does not affect the
result). The ordinate corresponds to the proportional to
4 a12 − a22
× − 2 concentration change in elastic moduli, divided by the
3E(t) (a2 − a12 (1 − ν0 ))E(t) − a22 ν0 K (t)
corresponding changes due to the circular cracks. The
a12 − a22 lines for shear and Young moduli correspond ν = 1/2.
+ 2 (43)
(a1 − a22 (1 − ν0 ))E(t) − a22 ν0 K (t) It is seen from (42), (43), (46) that the influence of the
Poisson’s ratio is weak. Thus for ν = 0 all three depen-
where E(t), and K (t) are elliptical integrals (A6) and
dences coincide (the absolute value varies in coeffi-
a22 cients). It is seen that reduction of ratio a2 /a1 from
t =1− (44) 1 to 0.6 leads only to 5% reduce of the moduli.
a12
√ 3.3 Extremely rigid inclusions
a1 a2 4 4
≡ = π (a1 a2 )3/2 N = √ S 3/2 N
a3 3 3 π
Similarly, for the extremely rigid elliptical in plane flat
(45)
inclusion, substituting (A9) into (9) and then into (12),
This quantity is reduced to (21) for the circular in plane and then performing limit transition µ∗ → ∞ and then
inclusions a1 = a2 . Formulae (42), (43) reduce to (19), a3 → 0, one obtains (Ustinov 2003b).
On application of classical Eshelby approach 63
1.8 Kell / Kcrcl Similar to what was done for spheroid inclusion,
ell / crcl the combined asymptotics may be obtained for simul-
1.6
Eell / Ecrcl taneous approach of both relative rigidity and aspect
ratio to their extreme values, however the correspon-
ding formulae are not represented here due to their
1.4
awkwardness.
1.2
a2 /a1 4 Conclusions
a2 1 (a12 − a22 )2
µ ef f
= µ +µ
0 0
(1 − ν0 ) 4
a1 5 a1 + a2 (1 − ν0 ) + 2a12 a22 ν0 E(t) − a22 (a12 + a22 )K (t)
4
4 a14 + a24 (3 − 4ν0 ) − 8a12 a22 (1 − ν0 ) E(t) + 4a22 (a12 + a22 )K (t)
− (48)
15a22 (3 − 4ν0 ) a12 (5 − 4ν0 )E 2 (t) − 4(a12 + a22 )(1 − ν0 )E(t)K (t) + a22 (3 − 4ν0 )K 2 (t)
Similar to the case of flat voids (cracks) the effective was considered in the frame of dilute concentration of
moduli are determined by their shape and concentration the inclusions, for which the influence of each inclu-
and are independent of their thickness. sion into effective moduli may be treated additively.
The Fig. 4 represents the influence of the ratio of The parameters determining the additive, proportio-
major and intermediate axes of the rigid inclusions on nal to the inclusions concentration, part of moduli are
the relative change of effective elastic moduli. The abs- Poisson’s ratios of matrix and inclusions, the ratio of
cise axis corresponds to the ratio of major and interme- elastic (compressive, or shear) moduli of the inclusions
diate axes of the ellipsoid (the minor axis corresponds and matrix and the aspect ratio of the inclusions, the last
to the thickness and does not affect the result). The ordi- two being of primary importance. In case of the large
nate corresponds to the proportional to concentration deviation of these parameters from unity, seven over-
changes in elastic moduli, divided by the corresponding lapping regions were distinguished, for which asymp-
changes due to circular rigid inclusions. All the lines totics are valid corresponding to approaching to zero
correspond to ν = 1/2. It is seen that the influence of (infinity) of various combinations of these two para-
the rigid particle increases with the divergence of the meters. These regions correspond to
inclusions shape from the circular one. 1. Rigid inclusions.
For circular inclusions formulae (47), (48) are redu- 2. Thin rigid inclusions. This is a combined asympto-
ced to (23), (24). However formulae for needle-like tics, the effective elastic moduli are determined by
inclusions (27), (28) may not be obtained from (47), (36), (37), the range of applicability does for (30).
(48) by a limit transition, because formulae (47), (48) 3. Thin inclusions of moderate rigidity. The effective
were obtained in the frame of the assumption (25) that elastic moduli are determined by (13), (14), the range
contradict assumption (26) used in deriving (27), (28). of applicability does for (15).
64 K. B. Ustinov, R. V. Goldstein
4π
4. Soft thin inclusions. This is a combined asymptotics, S1133 = S2233 = Pρ 2 − Pρ 2 + R I
the effective elastic moduli are determined by (33), 3
(34), the range of applicability does for (29). P
S3311 = S3322 = 4π − R − (P − 2R) I
5. Soft inclusions. 3
6. Needle-like inclusions of low and moderate rigidity. P
S3333 = 4π (1 − 3ρ 2 ) + R + 2 Pρ 2 − R I
The effective elastic moduli are determined by (16), 3
(17), the range of applicability does for (18). 2S2323 = 2S1313 = 2S3232 = 2S3131 =
7. Rigid needle-like inclusions. This is a combined
P
asymptotics, the effective elastic moduli are deter- 4π (1+ρ 2 ) + R − P(1 + ρ 2 )+R I
3
mined by (39), (40), the range of applicability does
2S1212 = 2S2121 = S1111 − S1122 (A1)
for (31).
where
Let us mention once more that the case of soft needle-
3 1
like inclusion is absent since the ranges described above P=
as regions 5 and 6 are overlapping (Fig. 2). 8π (1 − ν0 ) 1 − ρ 2
Each zone of intersection of the distinguished ranges 1 1 − 2ν0
R =
corresponds to succeeding limit transition (to zero, or 8π (1 − ν0 )
infinity) of the parameters aa31 , and µ∗
µ0 . =
2πρ
ρ−ρ − ρ 2 f or ρ < 1
The case of the inclusions in the form of flat three-
I 3/2 arccos 1
1 − ρ2
axial ellipsoids is also considered. Depending on the
2πρ
−1
values of the same parameters (the ratio of elastic moduli I = 3/2 ρ ρ 2 − 1− cosh ρ f or ρ > 1
of the inclusions and matrix and the ratio of the highest ρ2 − 1
and lowest axes of the inclusions) three particular cases a3
ρ = (A2)
were distinguished and their areas of applicability were a1
estimated. These cases are: extremely soft (crack-like)
inclusions, inclusions of intermediate rigidity, and extre-
mely rigid inclusions. Appendix 2
Note in conclusion that the obtained results enable
to calculate local stress-deformation field near the non- For the extremely flat inclusions with semi-axes a1 ≥
homogeneities and, hence, to search for the conditions a2 >> a3 the components of Eshelby’s tensor are pre-
of local strength violation according to the appropriate sented in (Mura 1982) as follows
criteria. The examples will be considered separately. 3a12 I11 (1 − 2ν)I1
The study was performed within the framework of S1111 = +
8π(1 − ν) 8π(1 − ν)
the Program for Fundamental Research of the Depart- 2
a2 I12 (1 − 2ν)I1
ment of Energetics, Machinery, Mechanics and Control S1122 = −
Processes, No. 12 and project 05-02-17542-a of 8π(1 − ν) 8π(1 − ν)
RFFI.
2
a3 I13 (1 − 2ν)I1
S1133 = −
8π(1 − ν) 8π(1 − ν)
(a12 + a22 )I12 (1 − 2ν)(I1 + I2 )
Appendix 1 S1212 = + (A3)
16π(1 − ν) 16π(1 − ν)
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 67
doi: 10.1007/978-1-4020-6929-1_8, © Springer Science+Business Media B.V. 2007
68 S. Cleja-Ţigoiu
formulated. No Cosserat type kinematics is assumed (1967) (within the constitutive framework of simple
in the model, and non-symmetric Cauchy stress T and materials), using the differential geometry concepts.
the stresses momentum µ (described by a third order The decomposition theorem of the connection with
tensorial field) are involved in the internal power Pint . metric property into a Levy-Civita connection and con-
The pair (T, µ) is power conjugated to the appropriate tortion was studied in Schouten (1954), Kondo and
rate of the elastic deformation measure. Yuki (1958), and applied in finite elasto-plasticity by
The principle of virtual power allows to formulate Steinmann (1994), Cleja-Tigoiu (2002a), see also Beju
the macroscopic balance laws of macromomentum and et al. (1983), Le and Stumpf (1996c)
angular momentum for (T and µ), which are similar to A Cosserat theory for elasto-(visco)plastic single
those derived in Fleck et al. (1994), and to derive the crystals, at finite deformations and based on the crystal-
micro-balance equations (i.e. the microstructural mate- lographic slip mechanism of plastic deformation, was
rial field equations) which are specific for the material elaborated in Naghdi and Srinivasa (1994), Le and
p p
force system (ϒ K , µK ), like Gurtin (2003). Stumpf (1996a), Steinmann (1994). In Forest et al.
To complete the constitutive equations within the (1997) the natural Cosserat strains are considered for
constitutive framework of finite elasto-plasticity, the the development of the constitutive equations and evo-
second law for isothermal processes is formalized simi- lution laws are proposed also for lattice torsion-
lar to Gurtin (2003, 2004). The second law leads to local curvature (second order) tensor.
free energy inequality or to free energy imbalance ψ̇ − Elasto-plastic model with dislocations was devel-
Pint ≤ 0, where ψ is the free energy density. We deter- oped in Le and Stumpf (1996b) based on an appropri-
mine the thermodynamic restrictions imposed by the ate principle of virtual work, and a thermodynamically
requirement to have the free energy imbalance satisfied consistent analysis of the anisotropic damage evolution
in any virtual process, for a given deformation state. was been performed by Stumpf and Hackle in (2003).
The admissible virtual process is defined based on Gurtin (2000) developed a gradient theory of single
the kinematical relationships, being consistent with crystal plasticity that accounts for geometrically nec-
them. The time derivatives of the appropriate distor- essary dislocations.
tions and connections put into evidence the presence Within the framework of finite elasto-plasticity the
of the velocity gradient L and of its gradient ∇χ L in evolution equations to describing the irreversible
the deformed configuration, as well as of the rate of behavior plays a fundamental role. Certain compatibil-
plastic distortion L p and of its gradient ∇K L p , written ity conditions concerning the evolution equation for the
in the configuration with torsion, respectively. measure of continuously distributed dislocation could
The constitutive hypotheses concerning microforces arrive, see for instance Acharia (2004), Cleja-Tigoiu
and micromomentum (called the material forces) are (2002a), Gupta et al (2006), Cleja-Tigoiu et al. (in
motivated by the dissipation inequality. The material press). The compatibility conditions are viewed in
forces are represented by a non-dissipative part ener- Cleja-Tigoiu et al. (in press), say for the given plas-
getic microforces, derived from the free energy and a tic metric tensor, as partial differential equations for
dissipative part, defined in a such way that the dissi- the torsion. We also mention here a second order the-
pation be positive. We emphasized the role played in ory, which allows for growth diffusion, developed in
p
the theory by the micro-stress ϒ K , which is force con- Epstein and Maugin (2000), especially for the evolu-
jugated to the rate of plastic distortion L p , and by the tion equation of the second order gradients that has
p
micro-stress momentum µK , which generates work in been considered.
conjunction with ∇K L p . The boundary conditions (appropriate to plasticity
We briefly recall different issues, involving continu- that account for the dislocation) are discussed for
ously distributed dislocations results, which are closely instance by Gurtin and Needlemen (2005) and Gupta
related to the model proposed here. et al. (2006).
A continuum theory for material with continuously Maugin in (1999), combining the energy and mom-
distributed dislocations has been developed by Kondo entum balance derives a so-called pseudo-momentum
and Yuki (1958), Bilby (1960), Kröner in (1963, 1992), equation to derive material forces. Well-known exam-
Kröner and Lagoudas in (1992) (for elastic models), ples of the material forces are driving forces on defects
and mathematically founded by Noll (1967) and Wang and the J-Integral in fracture mechanics.
Material forces in finite elasto-plasticity with continuously distributed dislocations 69
We mention the proposed model in our paper Cleja- The component representation of the curl is given in a
Ţigoiu (2002a), where the results from Cleja-Ţigoiu Cartesian basis by
(1990), Cleja-Ţigoiu and Soós (1990), and Cleja- ∂ A pk
(curl A) pi = i jk , while the third order tensor field
Ţigoiu (2001) have been extended to elasto-plastic ∂x j
∇A is characterized by
materials with continuous distributed dislocations, ∂ Ai j i
based on the hypotheses listed below in an heuristic ∇A = i ⊗ i j ⊗ ik . Thus we have the formulae
∂xk
manner: (i) The crystalline body is not homogeneous ∂ ∂vi i
and it has no relaxed (natural) global configuration. χ L ≡ k i ⊗ i j ⊗ ik .
∂x ∂x j
(ii) The local relaxed state is characterized by non- Lin(V, Lin) = {N : V −→ Lin linear}− defines the
Euclidean and non-Riemannian metric space. third order tensors and it is given by N = Ni jk ii ⊗i j ⊗ik .
(iii) Dynamical balance equations involve non- The scalar product of two third order tensors is given
symmetric Cauchy stress tensor and couple stresses. by N ·M = Ni jk Mi jk .
(iv) The crystalline body behaves as an elastic mater- Three configurations will be considered: k be a fixed
ial element, which means that the stress and the stress reference configuration of the body B, k(B) ⊂ E with
momentum are functions of the elastic distortion and the vector space Vk , χ (·, t) the deformed configura-
elastic connection. (v) The irreversible behavior of the tion at time t, for any motion of the body B, χ :
material is described by the evolution equations (of the B × R −→ E, there exists K, time dependent (non-
rate independent type) for plastic distortion as well as local) configuration with torsion, defined by the pair
for the gradient of plastic distortion. (p) (p)
In the present paper the developments are focused on (F p , k ), F p —plastic distortion and k —plastic con-
the balance equations (micro and macro), derived from nection.
an appropriate principle of the virtual power formulated F− the deformation gradient is defined by
within the constitutive framework of finite elasto- F(Z, t) = ∇(χ (·, t) ◦ k −1 )(Z), ∀ Z ∈ k(B). (2)
plasticity and on the restriction on the constitutive equa- We recall here the integrability theorem: Let F be a
tions imposed by the imbalanced free energy (i.e. function defined on a arcwise connected domain U. F
second law for isothermal processes). The presence is a gradient of χ , i.e. F(x) = ∇χ (x) for x ∈ U if and
of the material forces is a key point in the exposure, only if
and viscoplastic (generally rate dependent) constitutive (∇F(x)u)v − (∇F(x)v)u = 0
representation are derived. ⇐⇒ (curlF(x))(u × v) = 0, ∀ x ∈ U, (3)
List of notations. Further the following notations ∀u, v ∈ Vk .
will be used: Dislocation means (1) non-zero curl or (2) non-zero
E—the three dimensional Euclidean space, with the plastic torsion
vector space of translations V; (1) curl(F P )
= 0 or
Lin—the set of the linear mappings from V to V, i.e ∃ u, v such that (∇F p (u))v
the set of second order tensor, Skew ⊂ Lin the set of −(∇F p (v))u
= 0
all skew-symmetric second order tensors;
(2) ∃ u, v such that (S p u)v = ( p u)v − ( p v)u
u × v is the cross product, u ⊗ v and u · v denote the
= 0.
tensorial product and the scalar product of the vectors
u, v ∈ V. Let us introduce the following notation for a third order
As and Aa are the symmetrical and skew-symmetrical field generated by a connection, say , and by second
parts of the tensor A, here AT denotes the transpose order tensors, for instance F1 , F2 ,
of A; ([F1 , F2 ]u)v = ((F1 u)) F2 v, ∀u, v ∈ Vk . (4)
∂A φ(x) denotes the partial differential of the function
φ with respect to the field A.
Curl of a second order tensor field A is defined by the 2 Second order elastic and plastic pairs
second order tensor field of deformations
(curlA)(u × v) := (∇A(u))v − (∇A(v))u Let us consider the body B, k the initial configuration
(1)
∀u, v ∈ V. of the body (which will be not explicitly mentioned
70 S. Cleja-Ţigoiu
and the elastic connection is introduced in terms of Ax.4 The plastic distortion and the plastic connec-
the motion and plastic connections, both of them being tion are compatible each other, in the sense that the
related to the initial configuration, through the formula Frobenius integrability condition is satisfied
(p)
(e) (p)
p −1 p −1 = (F p )−1 ∇ F p . (15)
K ũ = F ((− )(F )
p
ũ)(F ) , ∀ ũ ∈ VK .
(7)
2.1 Relationships between the connections attached
Here VK := F p (Vk ).
to the plastic and elastic distortions
The pull back to the reference configuration k leads
to the following relationship between connections:
When we pass from one configuration, say from the
(e) (e) (p) initial configuration to another one, say K, as a conse-
back := (F p )−1 K [F p , F p ] = − , (8)
quence of rule of calculus formulae (13), from (15) the
derived from (7). relationship between the plastic connections follows
(p) (p)
Definition The differential of any smooth tensor field = −(F p )−1 K [F p , F p ], (16)
F̄, defined on k(B), with respect to the configuration
where the plastic connection with respect to K is intro-
with torsion K is given by
duced by
(∇K F̄)ũ = (∇ F̄)(F p )−1 ũ, ∀ ũ ∈ VK (9) (p)
K = F p (∇K ( F p )−1 ). (17)
Proposition 1
(p)
1. The multiplicative decomposition of the deforma- Remark Consequently we defined two pairs (F p , )
(p)
tion gradient F into the elastic and plastic distor- and ((F p )−1 , K ) of the appropriate plastic distor-
tions Fe , F p follows (p) (p)
tions, F p and (F p )−1 , and the connections and K ,
F = Fe F p . (10) which are compatible.
Material forces in finite elasto-plasticity with continuously distributed dislocations 71
The appropriate torsion (14), for the plastic connec- 2.2 Time-derivatives of the elastic connection with
tion defined in (15), becomes respect to relaxed configuration
(S p v)u = (F p )−1 [((∇ F p )v)u − ((∇ F p )u)v]. (18)
When we take the time derivative of the motion con-
Let us introduce the elastic connections associated nection (5), the rate of the total connection is expressed
with respect to the specified configurations in term of the second order velocity gradient
(e) (e) d
K = (Fe )−1 ∇K Fe , χ= Fe ∇χ (Fe )−1 . (19) () = F−1 (∇χ L)[F, F] (26)
dt
From the above definitions together with (9) the fol-
lowing relationship can be put into evidence where the velocity gradient in the actual configuration
is characterized by
(e) (e)
χ = −Fe K [(Fe )−1 , (Fe )−1 ]. (20)
L := ∇χ v(x, t), L = Ḟ(F)−1 , (27)
On the other hand for the elastic connection, say
(e)
where v is the vector field in the actual configuration.
K , the torsion can be similarly defined as the skew-
As a consequence of the multiplicative decomposi-
symmetric part of the connection
tion (10) the kinematics relationships
(e) (e)
(SeK v)u ≡ ( K v)u − ( K u)v. (21)
L = Ḟe (Fe )−1 + Fe L p (Fe )−1 ,
(28)
As a consequence of the multiplicative decomposi- Le = Ḟe (Fe )−1 , L p = Ḟ p (F p )−1
tion (10) and of the adopted definition for the connec-
tions, the relationships between appropriate elastic and follow. Le , L p are the rates of elastic and plastic dis-
plastic connections follows. tortions in the deformed configuration and the config-
Due to the symmetry of the motion connection intro- uration with torsion, respectively.
duced in (5), provided by the fact that at any time t F The rate of the plastic connection, relative to the
is the gradient of an appropriate application (2), reference configuration, can be derived from (15) under
(v)u − (u)v = 0. (22) the form similar to (26)
When we take the time derivative in (12) we get Consequently, taking into account (34) and (36) the
virtual time-derivative of the elastic connection with
d (e) d (p)
( K )(ũ) = F p (− ) ((F p )−1 ũ)(F p )−1 respect to the plastically deformed configuration can
dt dt be introduced by the Definition:
(p)
+ L p F p ((− )(F p )−1 ũ)(F p )−1 d (e)
virt ( K )(ũ) = Fe −1 ((∇χ
L)[Fe , Fe ])ũ−(∇K
L p )ũ
(p) dt
−F p ((− )(F p )−1 L p ũ)(F p )−1 (e) (e)
(p)
+
L p ( K ũ) − ( K (
L p ũ))
−F p ((− )(F p )−1 ũ)((F p )−1 L p ) (e)
−( ũ)
L p , ∀ ũ ∈ VK . (38)
∀ ũ ∈ VK . (32)
Using again (12) and the multiplicative decomposi-
tion (10) in relationship (32), the time derivative of the 3 The macro and micro balance equations
appropriate connections are related by
The principle of the virtual power at any arbitrary fixed
d (e) (e) (e) (e)
moment of the time t is built starting from the prin-
( K )(ũ)−L p ( K ũ)+ K (L p ũ)+( K ũ)L p
dt ciple of the virtual power derived from Fleck et al.
d (p) (33)
≡F p (− ) ((F p )−1 ũ)(F p )−1 . (1994) and using the result already proved by Cleja-
dt Tigoiu in (2002a), relative to the expressions of the
Let us introduce a linear operator applied to the elastic power expanded by an elasto-plastic material (without
(e) any relation with a principle of the virtual power).
connection K , dependent on the rate of plastic dis- First we recall the definitions for Piola-Kirchhoff
tortion L p , by stress tensor and the stress momentum as pulled back
(e) d (e) (e) to the configuration with torsion, and Mandel’s non-
(LL p [ K ])ũ := ( K )ũ − L p ( K ũ) symmetric stress measure, all of them being expressed
dt (34)
(e) (e) relative to K
+ K (L p ũ) + ( K ũ)L p ρK
K ≡ = det (Fe )(Fe )−1 Ts (Fe )−T , detFe =
for all ũ. The expression of the above operator (34), ρ
introduced in the left hand side of (33), leads to the µK = (det Fe )(Fe )T µ[(Fe )−T , (Fe )−T ], (39)
following equivalent formula 1 e T T e −T
K = (F ) (F ) , C = (F ) F ,
e e T e
ρK ρ
(e) d d (p)
(F p )−1 ((LL p [ K ])[F p , F p ]) = () − ( ). associated to the non-symmetric Cauchy stress T.
dt dt
The virtual power at a fixed moment of time is writ-
(35)
ten for any part P ⊂ B and we accept that a surface
Using (26 ), (29) and the multiplicative decomposition element in the actual configuration, with unit normal n,
(10), (35) becomes may transmit both force vector and couple vector. We
(e) start from the virtual power principle, VPP-I, formu-
(LL p [ K ]) = (Fe )−1 (∇χ L)[Fe , Fe ] − ∇K L p . (36) lated in Continuum Mechanics with couple stresses
Definition Starting from the kinematic relationships (see Fleck et al. (1994)).
derived above, for a given deformation state PVP-I. In the deformed configuration, ∀ P ⊂ B
(e) (p) bounded by a smooth surface ∂P, the virtual power at
(i.e. F, Fe , F p , K , are considered to be given), we a fixed moment of time
characterize a virtual process by ×
v—the virtual velocity, L—the virtual velocity gra- ρa · wdx + T · ∇wdx + 2 T ·θ dx
Pt Pt Pt
dient,
+ M · ∇θ dx = + Tn · wdσ
Le and L p the virtual rate of the elastic and plastic Pt ∂ Pt (40)
distortion, compatible with the kinematical relation-
+ Mn · θ dσ + ρb f · wdx
ships (27), (28), which means ∂ Pt Pt
v, and
L := ∇χ Le + F e
L= L p (Fe )−1 . (37) + ρbm · θ dx
Pt
Material forces in finite elasto-plasticity with continuously distributed dislocations 73
of the mechanical internal power as it follows from 2. The micro balance equation is expressed by
Cleja-Ţigoiu (2002a).
p p p
ϒ K − div µK = ρ̃Bm , in K(P, t),
Proposition 5
p
µK n = M p n on ∂K(P, t), (53)
1. The density of the internal mechanical power pro-
duced by the non-symmetric Cauchy stress tensor micro-traction condition.
can be written under the form
1 1 1 We remark that (51), (53)1 together with the bound-
T · L = T · Le + K · L p . (49)
ρ ρ ρK ary conditions (52) and (53)2 follow directly from the
2. The density of the internal mechanical power pro- principle of the virtual power, stipulated in (48), with-
duced by the couple stresses can be expressed with out any additional assumptions.
the aid of the third order tensor of stress momen- Finally we conclude that the theory can be based
tum defined in the actual configuration by µ or in either on the appropriate postulate of the variational
terms of the stress momentum pulled back to the principle for physical and material space or on the pos-
configuration with torsion µK tulates of the physical and material balance laws.
1 1
µ · ∇χ L = µ · (Fe )−1 ∇χ L[Fe , Fe ]
ρ ρK K
1 (e)
= µK · (LL p [ K ])
ρK 4 Free energy imbalance
1
+ µ · ∇K L p . (50)
ρK K Ax.6 There exists a free energy density function ψ,
The last equality, that put in evidence the elastic and invariant with respect to a change of frame in the actual
plastic part of the appropriate internal power, has been configuration
reformulated as a consequence of (36). (e) (p)
ψ = ψK (Ce , K , (F p )−1 , K ), (54)
Remark We use Gurtin’s argument, since there are no
motivation to suppose that the terms which enter the
represented in the configuration with torsion K by a
previous formulae (49) and (50) refer to the same stress (e)
and the same momentum, and we replace in (49) K function of the second order elastic pair (Ce , K ),
p
by the microstress ϒ K and in (50) µK , which is power and dependent on the plastic measure of deformation
p (p)
conjugated with ∇K L p , by µK .
((F p )−1 , K ).
The macro-balance equation at any time t can be Ax.7 The elasto-plastic behavior of the material is
L p = 0 and ∇K
derived from (48) if we take L p = 0. restricted to satisfy in K the free energy imbalance
For any v, a virtual velocity and for any second order
Le with curl(
tensor field Le )
= 0, i.e non-reducible 1
−ψ̇K + (Pint )K ≥ 0
to a gradient of an appropriate vector field, the macro- ρK (55)
for any virtual (isothermic) processes.
balance equations are derived. The micro-balance
equation at any time t can be derived from (48) if
L p and ∇K
L p are non-zero. Proposition 6 In (55) the internal power in the con-
figuration with torsion can be calculated starting from
Theorem 1 the expression
1. The impulse and momentum local balance equa-
tions can be written under the form 1
(T + T∗ ) · Le + ϒ K · L p
p
(Pint )K =
ρa = div T + ρb f ρ
(51) (e)
T∗ = div µ + ρBm , on Pt 1
+ µK · LL p [ K ]
with the appropriate boundary conditions on ∂Pt ρK
1 p
Tn = t(n), and µn = M(n). (52) + µ · ∇K L p , (56)
ρK K
Material forces in finite elasto-plasticity with continuously distributed dislocations 75
while the time derivative of the free energy is expressed 1 e −1
(F ) {T + T∗ }s (Fe )−T − 2∂Ce ψK
through ρ
·[(Fe )T {
L}s Fe − {Ce ·
L p }s ]
ψ̇K = ∂Ce ψK · Ċe − (F p )−T ∂(F p )−1 ψK · L p 1
+ {T + T∗ }a · ( L − Fe
L p (Fe )−1 )
d (e) ρ
+ ∂(e) ψK · K
dt 1 p
K
+ ϒ + (F ) ∂(F p )−1 ψK ·
p −T
Lp
ρK K
d (p)
+ ∂(p) ψK · K . (57) 1
K dt + µK · [(Fe )−1 (∇χ L)[Fe , Fe ] − ∇K Lp]
ρK
1 p
+ µK · ∇K Lp
The formula (56) is derived from macro and micro bal- ρK
ance equations, multiplied by Le and by L p , respec-
d e
tively. −∂ e K ψK · virt K − ∂(p) ψK
dt K
In order to pursuit the calculus
d (p)
·virt K ≥ 0 (61)
dt
1. We eliminate the rate of the elastic distortion,
which enters the expression (56) via formula (28), holds for any v, ∇χ
L ≡ ∇χ L, and for arbitrarily
as well the gradient of the rate of elastic ditorsion
given L , ∇K L .
p p
via the formula (36). Then only L and L p and their
appropriate differentials, ∇χ L and ∇K L p , enter The virtual kinematic processes are also character-
the internal power. ized by the virtual variations of the fields via the for-
2. In (57) the rate of the elastic strain is replaced by mulae
d (e) (e) (e) (e)
virt ( K) = Lp K − K L p − K [I, Lp]
Ċe = 2 (Fe )T {L}s Fe − 2 {Ce L p }s , (58) dt
+ (Fe )−1 (∇χ
L)[Fe , Fe ] − ∇K Lp
virt(Ċe ) = 2 (Fe )T {
L}s Fe − 2 {Ce ·
L p }s , 5 Thermodynamic restrictions
related to (58)
(e)
We provide the thermomechanic restrictions on the con-
virt(LL p [ K ]) = (Fe )−1 (∇χ
L)[Fe , Fe ] − ∇K
Lp, stitutive description of elasto-plastic material, based
related to (36). (60) on the imbalanced free energy condition. We require
the imbalanced condition written in (61) to be satis-
Everywhere we replace L, and ∇χ L, by the virtual L fied for any virtual process, defined by the formulae
L p and ∇χ
∇χ L̃, L p , and ∇K L p are replaced by Lp. (60)–(63), when L, ∇χ
L are arbitrary, and for the given
L , ∇K L .
p p
Proposition 7 The free energy imbalance is satisfied I. First step: we consider a virtual process in a such
for any virtual process, if the inequality written below way to have L p = 0, ∇KL p = 0.
76 S. Cleja-Ţigoiu
Thus (61) holds for any L, ∇χ L, if and only if the ∂C ψ̄ = (F p )−1 ∂Ce ψK (F p )−T
following constitutive restrictions ∂(p) ψ̄ = −(F p )T ∂(p) ψK [(F p )−T , (F p )−T ],
K
(Fe )−1 {T + T∗ }s (Fe )−T = 2ρ∂Ce ψK , ∂(e) ψK = (F ) p −T
∂(e) ψ̄[(F p )T , (F p )T ],
{T + T∗ }a = 0 K back
(64)
1 ∂F p ψ̄(F p )T ·
L p = −2C ∂Ce ψK ·
e
Lp
µ = ∂e K ψK ,
ρK K − (F p )−T ∂(F p )−1 ψK ·
Lp
(e) (e) (e)
are satisfied. + ∂(e) ψK · { L p − K [I,
Lp K − K L p ]}
II. Second step: we introduce the thermodynamic K
restriction (64) into imbalanced free energy con- (p) (p) (p)
dition (61) and we get the dissipation inequality + ∂(p) ψK · { L p − K [I,
Lp K − K L p ]} (68)
K
The constitutive form of ∂F p ψ̄ following from (68) can
1 p
ϒ + (F p )−T ∂(F p )−1 ψK · Lp be expressed through
ρK K
∂F p ψ̄ = −2Ce ∂Ce ψK − (F p )−T ∂(F p )−1 ψK
1 p
+ µ + ∂(p) ψK · ∇K L p − ∂(e) ψK
ρK K K K (e) (e) (p) (p)
+ DψK ( )[ ] + DψK ( )[ ]. (69)
(e) (e) (e)
p p p
·{L K − K L − K [I, L ]} − ∂(p) ψK (e) (e)
K Here the second order tensor denoted by DψK ( )[ ]
(p) (p) (p) (e)
Lp K − K
·{ L p − K [I,
L p ]} ≥ 0. (65) is defined for any third order tensor field ≡ X as it
follows
∂ψK ∂ψK
Let us introduce the free energy in the reference con- DψK (X )[X ] := Xs jk − Xi j p
∂X pjk ∂Xi js
figuration
∂ψK
− Xi pk i p ⊗ is . (70)
(e) (p) ∂Xisk
ψ = ψK (Ce , K , (F p )−1 , K ) Based on the dissipation inequality, we formulate
(e) (p)
(66)
≡ ψ(C, back , F p , ), the constitutive hypotheses:
Ax.8 The microforces contain:
taking into account the relationships (59), (8) and (16). (1) a dissipative part,
(2) a non-dissipative part, which are derived from the
Proposition 8 When we pass to the free energy den- free energy, the so-called energetic microforces,
sity in the reference configuration k, the dissipation and they are represented through
inequality becomes 1 p
ϒ = 2Ce (F p )∂C ψ̄(F p )T + ∂F p ψ̄(F p )T + Yγ Lp
ρK K
1 p 1 p
ϒ − 2Ce (F p )∂C ψ̄(F p )T − ∂F p ψ(F p )T ·
Lp µ = (F p )−T ∂(p) ψ̄[(F p )T , (F p )T ] + Yµ ∇K
Lp.
ρK K ρK K
1 p p −T
(71)
+ µ − (F ) ∂(p) ψ[(F ) , (F ) ]
p T p T
ρK K We remark that the non-dissipative parts of the micro-
·∇K L p , ∇K
L p ≥ 0, ∀ Lp. (67) forces were derived from the free energy through (71),
but only the last terms of the right sides of (71) are dis-
sipative, because the free energy cannot depend on the
In order to prove the above formulae we take into rates.
account the relationships between the partial deriva- Ax.9 The scalar constitutive functions Yγ , Yµ are
tives of the free energy expressed relative to the ini- defined in such a way to be compatible with the dissi-
tial configuration and to the configuration with torsion, pation inequality
derived from (66) together with (59), (8) and (16), under
the form Yµ ∇K L p · ∇KL p + Yγ
Lp · L p ≥ 0. (72)
Material forces in finite elasto-plasticity with continuously distributed dislocations 77
p p
Let us remark that if there exists a potential for plastic ∂ F31 ∂ F32
−
= 0, det F p = 1. (92)
deformation ∂x2 ∂x1
p ∂u 3 p ∂u 3 We kept the same notation for the tensor and its matrix
F32 = , F = with u 3 = u 3 (x 2 , x 3 ) representation, in a certain mentioned basis.
∂x2 33
∂x3
b The plastic metric tensor C p associated to the plastic
≡u· , (87) distortion (91) is given by
|b|
⎛ ⎞
which can be associated to a certain plane motion of 1 + (γ1 )2 γ1 γ2 γ1
dislocation, then in order to have non-zero curl it is C p = ⎝ γ1 γ2 1 + γ22 γ2 ⎠, (93)
necessary to accept a non-simply arc wise connected γ1 γ2 1
physical domain in a certain neighborhood of the given p p
with the notation F31 = γ1 , F32 = γ2 .
material point.
Let us introduce the function γ
Following again (Hirth and Lothe 1982) the screw
dislocation is characterized by a Burgers vector with γ = γ1 e1 + γ2 e2 with
(94)
the property listed below γ1 = γ1 (x 1 , x 2 ), γ2 = γ2 (x 1 , x 2 ).
b = (curl (F p )) N (areaA0 ) for From (91) together with (94) we get
(88)
curl (F p ) = a p (b ⊗ b),
C p = I + γ ⊗ γ + γ ⊗ e3 + e3 ⊗ γ . (95)
b
N = e3 , e3 = . Let us introduce a local basis (µ, ν, e3 )
|b|
Let us characterize the plastic curl in this case γ
ν= , for | γ |≡ γ12 + γ22 ,
p
∂ F31
p
∂ F32 |γ | (96)
curl (F ) =
p − (e3 ⊗ e3 ) (89) µ ∈ (e1 , e2 ), such that µ · ν = 0.
∂x2 ∂x1
From (91) together with (94) we get
If there exists a potential for plastic deformation, defin-
p p
ing the only two components F31 , F32 of the plastic C p = µ ⊗ µ + A p with
distortion supposed to be non-zero, then A p = (1+ | γ |2 )ν ⊗ ν+ | γ | (ν ⊗ e3
p ∂u 3 p ∂u 3 +e3 ⊗ ν) + e3 ⊗ e3 . (97)
F31 = , F = with
∂x1 32
∂x2 (90)
b The second order tensor A p has the matrix representa-
u 3 = u 3 (x 1 , x 2 ) ≡ u · ,
|b| tion in the basis {ν, e3 }
which means that an anti-plane motion of dislocation 1+ | γ |2 | γ |
Ap = . (98)
has been considered. Again, in order to have |γ | 1
non-vanishing curl, the physical space in a certain
neighborhood of the material point is locally a simply We define the positive square root tensor (A p )1/2 and
arc wise connected domain. the symmetric and positive definite tensor U p .
Finally, we prove the formula is accepted in Gurtin (2003), within the framework
Up ≡ (C p )1/2 =µ⊗µ of small deformation theory. The power expended
1 in terms of an appropriate force system is given
+ (2+ | γ |2 )ν ⊗ ν
4+ | γ |2 under the form (T· Ḣe +T p · Ḣ p +S p ·∇ Ḣ p )dx.
|γ | P
+ (ν ⊗ e3 + e3 ⊗ ν) Here T p a second-order microstress and S p a polar
4+ | γ |2 (102)
(third-order) microstress that together perform
2
+ e3 ⊗ e3 work in the evolution of the defects through this
4+ | γ |2 structure, for any part P of the body.
γ
where ν = , µ · ν = 0, µ · e3 = 0. b. In Gurtin (2004) thepower expended within any
|γ |
Proof We write the Hamilton-Caley theorem for the part P has the form (T · Ėe + T p · Ḣ p + S p ·
P
tensor fields (A p )1/2 curlḢ p )dx. Here Ee = {He } S , and the micro-
√ scopic stress performs work in conjunction with
A p − (A p )1/2 (tr(A p )1/2 ) + (detA p )Î2 = 0 (103)
the rate of Burgers vector, which is characterized
When we apply the trace operator in (103) we get by G = curl(H p )
√
(tr(A p )1/2 )2 = (trA p ) + 2 (detA p ) (104) c. Within the constitutive framework of Crystal plas-
ticity, based on the multiplicative decomposition
and consequently
of the deformation gradient, the tensor field
1
(A p )1/2 = A p
+ (detA p )Î
2 . (105) G = 1/(detF p )F p curl(F p ) (i.e just Noll’s dislo-
tr(A p )1/2 cation density, in Noll (1967), here in terms of
Using the explicit expression of the trace, the formula the plastic distortion F p ) is considered by Gurtin
(99) follows at once. (2000) to be a measure of geometrically neces-
sary dislocations. In this
case the internal power
Proposition 10
α α
is written in the form (T · L +
e
(π v +
1. The torsion of the plastic connection, attached to P α
the plastic connection in the reference configura- ξ α · gradv α ))dx, where the sum is espanded to the
tion , can be expressed in term of ω, by slip systems and for each α, π α —internal micro-
S = ωe3 ⊗ [e2 ⊗ e1 − e1 ⊗ e2 ]. (106) forces and ξ α —microstresses are introduced as
forces conjugated to slip and they produce work
2. The second order torsion tensor is derived under
by slip and by slip gradient respectively.
the form
∂γ2 ∂γ1 We conclude
N = ωe3 ⊗ e3 , with ω = − . (107)
∂ x1 ∂ x2 1. As a peculiar aspect of the models proposed by
Proof By direct calculus we derive the expression of Gurtin in the mentioned papers, the microbalance
the plastic connection equation generates the viscoplastic yield condi-
tions or the viscoplastic flow rules. On the other
p ≡ (F p )−1 (∇F p ) = e3 ⊗ e1 ⊗ (∇γ1 )
hand no yield criteria has been introduced, and
+e3 ⊗ e2 ⊗ (∇γ2 ) (108) the irreversible behavior can develop at the very
A full characterization of the edge dislocation as well beginning.
as the appropriate compatibility condition can be found 2. Let us remark that the micro balance equation
in Cleja-Ţigoiu et al. (2007, in press). together with the viscopalstic constitutive equa-
Concluding remarks. tion for the microforces and microstress momen-
In order to compare the presented here results with tum generate an appropriate flow rule (see also the
the existing in the literature results in the field, we put non-local yield condition in Gurtin (2000))
into evidence different internal power expressions, pro- 2Ce (F p )∂C ψ̄(F p )T + ∂F p ψ̄(F p )T + Y (d p )L p
posed by Gurtin.
−divK ((F p )−T ∂(p) ψ̄[(F p )T , (F p )T ]
a. The additive decomposition of the displacement
+h Y (d )∇K
p
vector ∇u = He +H p into elastic and plastic parts 2 p
L p ) = ρ̃Bm , (109)
80 S. Cleja-Ţigoiu
with d p := L p · L p + h 2 ∇K L p · ∇K L p , for the yield) function defined in the physical force system
plastic incompressible case ρK = ρ0 . (T, µ), such that the plastic (viscoplastic) behav-
3. Following the methodology developed in Cleja- ior can develope if and only if during the deforma-
Ţigoiu (2002b), we can derive from the proposed tion process the physical force system lays on the
here model the behavior of the elasto-plastic mater- yield surface or it is situated outward the surface
ial, in the case of small elastic strains but large elas- f K (T, µ) ≥ 0. Consequently the behavior would
tic rotation Re . In this case Ce I + 2 with the be elastic if the physical force system remains
elastic strain | |<< 1, and the connection is gen- inside the yield surface.
erated by the gradient of the elastic rotation as well
as the ∇ . Moreover, when we restrict ourselves Acknowledgments The author acknowledges support from the
Romanian Ministry of Education and Research through
to small elastic and plastic deformations, i.e. the CEEX programm (Contract No. CERES-11-12/25.07.2006).
small rotation and small strains, model within the
constitutive framework adopted by Gurtin (2000)
follows, but the microstress momentum are still
presented.
4. The plasticity and the damage defects localize References
over the narrow region of the material. The plas-
tic and the damage evolution process are inho- 1. Acharya A (2004) Constitutive analysis of finite defor-
mation field dislocation mechanics. J Mech Phys Solid
mogeneous at the macroscale. The macroscopic
52:301–316
inelastic deformation are sensitive to the structural 2. Bilby BA (1960) Continuous distribution of dislocations.
defects within the volume element. Hence, nonlo- In: Sneddon IN, Hill R (eds) Progress in solid mechanics.
cal theories are necessary to adequately take into North-Holland, Amsterdam, pp 329–398
3. Beju I, Soós E, Teodorescu PP (1983) Spinor and non-
account mechanisms which take place in the neigh- Euclidean tensor calculus with applications. Ed. Tehnica,
borhood of the considered material points (see the Bucureşti Romania, Abacus Press, Tunbridge Wells, Kent,
comments by Brünig and Ricci (2005)). A nonlo- England (romanian version 1979)
cal theory of an isotropically damaged materials 4. Brüning M, Ricci S (2005) Nonlocal continuum theory of
an isotropically damaged metals. Int J Plast 21:1346–1382
can be developed based on our proposed model, 5. Cleja-Ţigoiu S (1990) Large elasto-plastic deformations
in order to characterize the damage state config- of materials with relaxed configurations – I. Constitutive
uration (identified with a configuration with tor- assumptions, II. Role of the complementary plastic factor.
sion), independently of the current elastic defor- Int J Eng Sci 28:171–180, 273–284
6. Cleja-Ţigoiu S (2001) A model of crystalline materials
mation, i.e. for Ce = I. The damage produced by with dislocations. In: Cleja-Ţigoiu S, Ţigoiu V (eds) Pro-
the microdefects can be identified with the pres- ceedings of 5th international Seminar geometry continua
ence of screw and/or edge dislocations. and microstructures. Sinaia, Romania, pp 25–36
5. The proposed description of the edge and screw 7. Cleja-Ţigoiu S (2002a) Couple stresses and non-
Riemannian plastic connection in finite elasto-plasticity.
dislocations (corresponding to deformation fields, ZAMP 53:996–1013
which remain incompatible due to the non-zero 8. Cleja-Ţigoiu S (2002b) Small elastic strains in finite
curlF p ) can be utilized in order to describe the elasto-plastic materials with continuously distributed dis-
appropriate fracture mode, by plane deformation locations. Theor Appl Mech 28(29):93–112
9. Cleja-Ţigoiu S, Maugin GA (2000) Eshelby’s stress ten-
and anti-plane deformation. The non-zero Burgers sors in finite elastoplasticity. Acta Mechanica 139:231–249
vector can be utilized in order to describe the cracks 10. Cleja-Ţigoiu S, Soós E (1990) Elastoplastic models with
motion. relaxed configurations and internal state variables. Appl
6. The quantities with the dimension of length Mech Rev 43:131–151
11. Cleja-Ţigoiu S, Fortunée D, Vallée C (2007) Torsion
appears as additional material parameters, in the equation in anisotropic elasto-plastic materials with con-
expression of the free energy density, in the inten- tinuously distributed dislocations. Math Mech Solids
sity for the accumulated effect of the rate of plastic doi:10.1177/1081286507079157
distortion L p and of the gradient of L p . 12. Epstein M, Maugin GA (2000) Thermomechanics of vol-
umetric growth in uniform bodies. Int J Plast 16:951–978
7. In the model (although at the general framework 13. Fleck NA, Muller GM, Ashby MF, Hutchinson
presented here there is not necessary) we can JW (1994) Strain gradient plasticity: theory and experi-
assume the existence of the viscoplastic (or an ment. Acta Metall Mater 42:475–487
Material forces in finite elasto-plasticity with continuously distributed dislocations 81
14. Forest S, Cailletand G, Sievert R (1997) A Cosserat theory 27. Le KC, Stumpf H (1996b) Nonlinear continuum with dis-
for elastoviscoplastic single crystals at finite deformation. locations. Int J Eng Sci 34:339–358
Arch Mech 49(4):705–736 28. Le KC, Stumpf H (1996c) On the determination of the
15. Gupta A, Steigmann D, Stölken JS (2006) On the evolution crystal reference in nonlinear continuum theory of disloca-
of plasticity and incompatibility. Math Mech Solids. online: tion. Proc Roy Soc London A 452:359–371
doi:10.1177/1081286506064721 29. Maugin GA (1999) The thermomechanics of nonlinear irre-
16. Gurtin ME (2000) On the plasticity of single crystal: free versible behaviors. World Scientific
energy, microforces, plastic-strain gradients. J Mech Phys 30. Naghdi PM, Srinivasa AR (1994) Characterization of dis-
Solids 48:989–1036 locations and their influence on plastic deformation in sin-
17. Gurtin ME (2002) A gradient theory of single-crystal vis- gle crystal. Int J Eng Sci 32(7):1157–1182
coplasticity that accounts for geometrically necessary dis- 31. Noll W (1967) Materially uniform simple bodies with
locations. J Mech Phys Solids 50:5–32 inhomogeneities. Arch Rat Mech Anal 27:1–32
18. Gurtin ME (2003) On a framework for small-deformation 32. Schouten JA (1954) Ricci-Calculus. Springer-Verlag,
viscoplasticity: free energy, microforces, strain gradients. Berlin
Int J Plast 19:47–90 33. Steinmann P (1994) A micropolar theory of finite deforma-
19. Gurtin ME (2004) A gradient theory of small-deformation tion and finite rotation multiplicative elastoplasticity. Int J
isotropic plasticity that accounts for the Burgers and for dis- Solids Struct 31:1063–1084
sipation due to plastic spin. J Mech Phys Solids 52:2545– 34. Steinmann P (1997) Continuum theory of dislocations:
2568 impact to single cristal plasticity. In: Owen DRJ, Onãte E,
20. Gurtin ME, Needleman A (2005) Boundary conditions in Hinton EComputational plasticity, fundamental and appli-
small-deformation, single-crystal plasticity that account cations. CIME, Barcelona,
for the Burgers. J Mech Phys Solids 53:1–31 35. Steinmann P (2002) On spatial and material settings of
21. Hirth J, Lothe JP (1982) Theory of dislocations. Krieger hyperelastostatic crystal defects. J Mech Phys Solids
Publishing, Malabar, Florida 50:1743–1766
22. Kondo K, Yuki M (1958) On the current viewpoints of 36. Stumpf H, Hackle K (2003) Micromechanical concept for
non-Riemannian plasticity theory. In: RAAG memoirs analysis of damage evolution in thermo-viscoplastic and
of the unifying study of basic problems in engng and quasi-brittle materials. Int J Solids Struct 40:1567–1584
physical sciences by means of geometry II (D), Tokyo, 37. Teodosiu C (1970) A dynamic theory of dislocations and its
pp 202–226 applications to the theory of the elastic-plastic continuum.
23. Kröner E (1963) On the physical reality of torque stresses In: Simmons JA, de Witt R, Bullough R (eds) Fundamental
in continuum mechanics. Gauge theory with disclinations. aspects of dislocation theory, Nat Bur Stand (U.S.), Spec.
Int J Eng Sci 1:261–278 Publ. 317, II, pp 837–876
24. Kröner E (1992) The internal mechanical state of solids 38. Wang CC (1967) On the geometric structure of simple bod-
with defects. Int J Solids Struct 29:1849–1857 ies, a Mathematical foundation for the theory of continuous
25. Kröner E, Lagoudas DC (1992) Gauge theory with discli- distributions of dislocations. Arch Rat Mech Anal 27:33–
nations. Int J Eng Sci 30:1849–1857 94
26. Le KC, Stumpf H (1996a) A model of elastoplastic bod-
ies with continuously distributed dislocations. Int J Plast
12((5):611–627
Distributed dislocation approach for cracks in couple-stress
elasticity: shear modes
P. A. Gourgiotis · H. G. Georgiadis
Abstract The distributed dislocation technique pro- the near-tip stress show significant departure from the
ved to be in the past an effective approach in studying predictions of classical fracture mechanics.
crack problems within classical elasticity. The present
work aims at extending this technique in studying crack Keywords Distributed dislocations · Cracks ·
problems within couple-stress elasticity, i.e. within a Couple-stress elasticity · Integral equations
theory accounting for effects of microstructure. As a
first step, the technique is introduced to study finite-
length cracks under remotely applied shear loadings 1 Introduction
(mode II and mode III cases). The mode II and mode III
cracks are modeled by a continuous distribution of glide The present work is concerned with the study of mode
and screw dislocations, respectively, that create both II and mode III finite-length cracks in a material with
standard stresses and couple stresses in the body. In microstructure. We assume that the response of the
particular, it is shown that the mode II case is governed material is governed by couple-stress elasticity. This
by a singular integral equation with a more complica- theory falls into the category of generalized continuum
ted kernel than that in classical elasticity. The numerical theories and is a particular case of the general approa-
solution of this equation shows that a cracked material ches of Toupin (1962), Mindlin (1964), and Green and
governed by couple-stress elasticity behaves in a more Rivlin (1964). As is well-known, ideas underlying
rigid way (having increased stiffness) as compared to couple-stress elasticity were advanced first by Voigt
a material governed by classical elasticity. Also, the (1887) and the Cosserat brothers (1909), but the sub-
stress level at the crack-tip region is appreciably higher ject was generalized and reached maturity only with the
than the one predicted by classical elasticity. Finally, works of Toupin (1962), Mindlin and Tiersten (1962),
in the mode III case the corresponding governing inte- Mindlin (1964), and Koiter (1964).
gral equation is hypersingular with a cubic singularity. Earlier application of the couple-stress elasticity,
A new mechanical quadrature is introduced here for mainly on stress-concentration problems, met with
the numerical solution of this equation. The results in some success providing solutions physically more ade-
the mode III case for the crack-face displacement and quate than solutions based on classical elasticity (see
e.g. Mindlin and Tiersten 1962; Weitsman 1965; Bogy
and Sternberg 1967a, b). Work employing couple-stress
P. A. Gourgiotis · H. G. Georgiadis (B)
Mechanics Division, National Technical University
theories on elasticity and plasticity problems is also
of Athens, Zographou Campus, Zographou, 15773, Greece continued in recent years (see e.g. Vardoulakis and
e-mail: georgiad@central.ntua.gr Sulem 1995; Huang et al. 1997; Chen et al. 1998;
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 83
doi: 10.1007/978-1-4020-6929-1_9, © Springer Science+Business Media B.V. 2007
84 P. A. Gourgiotis, H. G. Georgiadis
Anthoine 2000; Lubarda and Markenscoff 2000; that there is no work at all in modeling cracks with
Bardet and Vardoulakis 2001; Georgiadis and Velgaki distribution of dislocations in materials with micro-
2003; Grentzelou and Georgiadis 2005). structure. Therefore, the present work aims at exten-
Nevertheless, there is only a limited number of stu- ding the technique in couple-stress elasticity. In another
dies concerning the effects of couple-stresses in crack recent work by the present authors (Gourgiotis and
problems. One of the earlier works in this subject is Georgiadis 2007) the mode I crack problem was also
that of Sternberg and Muki (1967) who considered the considered within the same framework. A comparison
mode I finite-length crack by employing the method of between the mode II case studied here and the mode
dual integral equations. They provided only asympto- I case leads to the conclusion that the opening mode
tic results and showed that both the stress and couple- is mathematically more involved than the shear mode.
stress fields exhibit a square-root singularity while the This is in some contrast with situations of classical elas-
rotation field is bounded at the crack-tip. The same ticity where the two plane-strain crack modes involve
method was adopted by Ejike (1969) for a circular equivalent mathematical effort.
(penny-shaped) crack in couple-stress elasticity and As in analogous situations of classical elasticity,
by Paul and Sridharan (1980, 1981) for a finite-length a superposition scheme will be followed. Thus, the
crack in micropolar elasticity. Using the Wiener-Hopf solution to the basic problem (body with a traction-
technique, Atkinson and Leppington (1977) studied free crack under remote shear field) will be obtained
the problem of a semi-infinite crack with exponen- by the superposition of the stress field arising in the
tially decayed normal tractions on the crack faces. More un-cracked body (of the same geometry) to the ‘correc-
recently, Huang et al. (1997) provided near-tip asymp- tive’ stresses and couple-stresses induced by a conti-
totic fields for the mode I and mode II crack problems, nuous distribution of dislocations chosen so that the
in couple-stress elasticity, by using the method of eigen- crack-faces become traction-free. The stress field for
function expansions. Also, Zhang et al. (1998) by a discrete glide and screw dislocation in couple-stress
employing the Wiener-Hopf technique investigated the elasticity will serve, respectively, as the Green’s func-
mode III semi-infinite crack in couple-stress elasticity tion for the mode II and mode III problem. However,
in the special case where the second couple-stress we note that deriving the stress field of a discrete dis-
moduli is set equal to zero. Moreover, using a simi- location within generalized continua is by no means a
lar approach, Huang et al. (1999) obtained full-field straightforward task. Within the framework of couple-
solutions for semi-infinite cracks under mode I and stress elasticity a lot of research has been devoted to dis-
mode II loadings in elastic-plastic materials with strain- locations. Representative references include work by
gradient effects. Kroner (1963), Misicu (1965), Teodosiu (1965), Cohen
Here, we aim at providing full-field solutions to (1966), Anthony (1970), Knesl and Semela (1972) and
the mode II and mode III finite-length crack problems Nowacki (1974). Finally, it is shown that due to the
within couple-stress elasticity by introducing an nature of the above Green’s functions and the boundary
approach based on distributed dislocations. Since the conditions that arise in couple-stress elasticity, the afo-
pioneering work of Bilby et al. (1963), Bilby and Eshe- rementioned procedure results for the mode II case in a
lby (1968) the distributed-dislocation technique has singular integral equation (SIE), whereas for the mode
been employed to analyze various crack problems in III case in a hypersingular integral equation (IE) with
classical elasticity. A thorough exposition of the tech- a cubic singularity. In order to solve this hypersingular
nique can be found in the treatise by Hills et al. (1996). IE, a new mechanical quadrature is constructed.
The strength of this analytical/numerical technique lies
in the fact that it gives detailed full-field solutions for
crack problems at the expense of relatively little analy- 2 Basic concepts and equations of couple-stress
tical demands as compared to the elaborate technique elasticity
of dual integral equations and, also, of relatively little
computational demands as compared to the Finite Ele- In this Section, we briefly present the basic ideas and
ment and Boundary Element methods. Although the equations of couple-stress elasticity. The theory
technique has proven to be very successful in studying employed here is a particular case of form III in the
crack problems within classical elasticity, it appears general Mindlin’s (1964) approach. Nevertheless, we
Distributed dislocation approach for cracks 85
expressed in dimensions of [length]−1 . Notice also that directly be derived from the equilibrium of the mate-
Eq. 16 can alternatively be written as rial tetrahedron, i.e. the relations between tractions and
1 stresses are given by (3) and (4).
κij = ej kl ∂i ∂k ul = ej kl ∂k εil . (17) Introducing the constitutive equations of the theory
2
is now in order. We assume a linear and isotropic mate-
Equation 17 expresses compatibility for curvature rial response, in which case the potential-energy den-
and strain fields. In addition, there is an identity, i.e. sity takes the form
∂k κij = ∂k ∂i ωj = ∂i ∂k ωj = ∂i κkj , which expresses
1
compatibility for the curvature components. The com- W ≡ W εij , κij = λεii εjj + µεij εij + 2ηκij κij
patibility equations for the strain components are the 2
usual Saint Venant’s compatibility equations. We notice + 2η κij κj i , (20)
also that κii = 0 because κii = ∂i ωi = (1/2) eij k uk,j i
where λ, µ, η, η are material constants. Then, Eq.
= 0 and, therefore, that κij has only eight independent 20 leads, through the standard variational manner, to
components. The tensor κij is obviously an asymmetric the following constitutive equations
tensor.
Now, regarding the traction boundary conditions, we ∂W
τij ≡ σ(ij ) = = λδij εkk + 2µεij , (21)
note that at first sight, it might seem plausible that the ∂εij
surface tractions (i.e. the force-traction and the couple-
traction) can be prescribed arbitrarily on the external ∂W
mij = = 4ηκij + 4η κj i . (22)
surface of the body through relations (3) and (4), which ∂κij
stem from the equilibrium of the material tetrahedron. In view of (21) and (22), the moduli (λ, µ) have the
However, as Koiter (1964) pointed out, the resulting same meaning as the Lamé constants
number of six traction boundary conditions (three of classical
elas-
ticity theory, whereas the moduli η, η account for
force-tractions and three couple-tractions) would be in couple-stress effects.
contrast with the five geometric boundary conditions Finally, the following
that can be imposed. Indeed, since the rotation vec- points
are of notice: (i) The
couple-stress moduli η, η are expressed in dimen-
tor ωi in couple-stress elasticity is not independent of sions of [force]. (ii) Since κii = 0, mii = 0 is also valid
the displacement vector ui (cf. (15)), the normal com- and therefore the tensor mij has only eight independent
ponent of the rotation is fully specified by the distri- components. (iii) The scalar (1/3) µkk of the couple-
bution of tangential displacements over the boundary. stress tensor does not appear in the final equation of
Therefore, only the three displacement and the two tan- equilibrium, nor in the reduced boundary conditions
gential rotation components can be prescribed indepen- and the constitutive equations. Consequently, (1/3) µkk
dently. As a consequence, only five surface tractions is left indeterminate within the couple-stress theory.
(i.e. the work conjugates of the above five independent (iv) The following restrictions for the material constants
kinematical quantities) can be specified at a point of should prevail on the basis of a positive definite poten-
the bounding surface of the body. These are three redu- tial-energy density (Mindlin and Tiersten 1962)
ced force-tractions and two tangential couple-tractions
(Mindlin and Tiersten 1962; Koiter 1964)
η
(n) 1 3λ + 2µ > 0, µ > 0, η > 0, −1 < < 1.
Pi = σj i nj − eij k nj ∂k m(nn) , (18) η
2
(23a,b,c,d)
(n)
Ri = mj i nj − m(nn) ni , (19)
where m(nn) = ni nj mij is the normal component of the 3 Plane problems of couple-stress elasticity
deviatoric couple-stress tensor mij . Finally, it is worth
noting that in the micropolar (Cosserat) theory of elas- The cases of plane strain and anti-plane strain are exa-
ticity (see e.g. Nowacki 1972), the traction boundary mined here and the basic equations are given. In what
conditions are six since the rotation is fully independent follows, vanishing body forces and body couples are
of the displacement vector. In this case the tractions can assumed.
Distributed dislocation approach for cracks 87
Next, the non-vanishing components of the anti- For a body occupying a domain in the (x, y)-plane
symmetric part of the force-stress tensor are obtained under conditions of anti-plane strain, the displacement
from (8) as field takes the general form
1 ∂mxz ∂myz ux ≡ 0, uy ≡ 0, uz = w (x, y) = 0. (38a,b,c)
αxy = − αyx = − + = − 2η∇ 2 ωz .
2 ∂x ∂y Again, by virtue of (13)–(16), the non-vanishing com-
(31) ponents of strain, rotation and curvature are given as
It should be noticed that the independence upon the 1 ∂w 1 ∂w
εxz = εzx = , εyz = εzy = , (39a,b)
coordinate z of all components of the force-stress and 2 ∂x 2 ∂y
couple-stress tensors, under the assumption (24c), was
proved by Muki and Sternberg (1965). Indeed, it is 1 ∂w 1 ∂w
ωx = ωyz = , ωy = ωxz = − , (40a,b)
noteworthy that, contrary to the respective plane-strain 2 ∂y 2 ∂x
88 P. A. Gourgiotis, H. G. Georgiadis
µ
2 (1 + β) b
µb 2
2 µb myy = −mxx =
mxz = − K2 (r/
) cos 2θ − K0 (r/
) , πr2
cos 4θ
π r2 π
(54) 3µb
2 (1 + β)2 2
2
− − K2 (r/
) cos 4θ
πr2 r2
µb 2
2 µb (1 + β)2
myz = − K2 (r/
) sin 2θ. (55) + K2 (r/
) cos 4θ
π r2 2π
Examining now the asymptotic behavior of the above µb (1 + β)2
− K0 (r/
) (3 cos 4θ + 1) ,
stress field (to determine the possibility of singulari- 8π
ties), we note that as r → 0 the following asymptotic (60a,b)
relations hold (see e.g. Abramowitz and Stegun 1964)
1 2
2 3µb
2 (1 + β)2 2
2
− K (r/
) = O r −1 , myx = − K2 (r/
) sin 4θ
r r2
2 πr2 r2
r [K2 (r/
) − K0 (r/
)] = O r −1 , µb (1 + β)2
− K2 (r/
) (2 sin 4θ + sin 2θ )
4π
K0 (r/
) = O (ln r) . (56a,b,c)
3µb (1 + β)2
+ K0 (r/
) sin 4θ
In view of (56), as the dislocation core (r → 0) is 8π
2
approached,
the components
of the force-stress ten- µb (1 + β) 2
− − K2 (r/
) sin 2θ,
sor σxx , σyy , σxy , σyx exhibit a Cauchy singularity π r2
(just as in classical elasticity), the couple-stress mxz (61)
becomes logarithmically unbounded, while myz
remains bounded. Finally, when
→ 0 the stress field mxy = myx − 2µ
2 (1 − β) ∇ 2 w, (62)
of classical elasticity for a discrete glide dislocation is The following points are of notice now: (i) Using the
recovered. well known asymptotic properties of the modified Bes-
sel functions, we conclude that as r → 0 the asymme-
tric and the symmetric shear stresses behave as ∼ r −3
4.2 Screw dislocation and ∼ r −1 , respectively, whereas the couple-stresses
behave as ∼ r −2 . (ii) When β = −1 (i.e. when η =
For a screw dislocation with strength b the displace- −η ), the above stress field degenerates into the res-
ment field in couple-stress elasticity is given as (see pective one in classical elasticity for a screw
our derivation in Appendix A) dislocation.
2
b b 2
w= θ− (1 + β) − K 2 (r/
) sin 2θ,
2π 4π r2 5 Formulation of crack problems by a distribution
(57) of dislocations
where the ratio β ≡ η /η should satisfy the following
inequality −1 < β <1. The stress and couple-stress 5.1 Mode II crack
fields corresponding to (57) are obtained from Eqs. 42–
46 as Consider a straight crack of length 2a embedded in
the xy-plane of infinite extend in a field of pure shear
µb µb
τxz = − sin θ, τyz = cos θ, (58a,b) (Fig.1). The crack faces are traction free and the body is
2π r 2π r
considered to be in plane-strain conditions. The crack
faces are defined by n = (0, ±1). Then, according to
µb µb
2 (1 + β)
σxz = − sin θ + sin 3θ, (18) and (19), the boundary conditions along the crack
2π r πr3
faces are written as
µb µb
2 (1 + β)
σyz = cos θ − cos 3θ, σyx = 0, σyy = 0, myz = 0 for |x| < a,
2π r πr3
(59a,b) (63a,b,c)
90 P. A. Gourgiotis, H. G. Georgiadis
whereas the regularity conditions at infinity are where − signifies Cauchy principal value integration
∞ ∞ ∞ ∞ ∞ and B (ξ ) = db/dξ is the dislocation density at a point
σyx = σxy = τxy → σ0 , σyy , σxx → 0,
ξ (|ξ | < a), this density being defined in the same way
m∞ ∞
xz , myz → 0 as r → ∞, (64a,b,c) as in classical elasticity (see e.g. Hills et al. 1996).
2 1/2
where r = x + y 2 now is the distance from the The kernel k (x, ξ ) is defined as
origin and the constant σ0 denotes the remotely applied
2 2
2 1
shear loading. k(x, ξ ) = − − K2 (|x − ξ |/
) −
x − ξ (x − ξ )2 2
Then, the crack problem is decomposed into the fol-
(x − ξ ) 2
2
lowing two auxiliary problems. − − K2 (|x − ξ |/
)
The un-cracked body
2 (x − ξ )2
It can readily be verified that the appropriate Mind-
+ K0 (|x − ξ |/
) . (69)
lin’s stress functions for the un-cracked body of infinite
extent subjected to boundary conditions (64a,b,c) are To show that k (x, ξ ) is regular, we expand the latter
as follows in series as x → ξ (see e.g. Abramowitz and Stegun
= − σ0 xy, = 0. (65a,b) 1964) and obtain
The stress field that corresponds to the above stress k (x, ξ ) = (a1 + a2 ln |x − ξ |) (x − ξ )
functions can be found from (32)–(34) as
+O (x − ξ )3 ln |x − ξ | , (70)
σyx (x, y) = σxy (x, y) = σ0 ,
σxx = σyy = 0, mxz = myz = 0. (66a,b,c) where ai are constants depending on the characteristic
Notice, that there are no couple-stresses induced in the material length
. Since lim (x − ξ )n ln |x − ξ | = 0
x→ξ
un-cracked body, the body being in a state of pure shear. for n > 0, we conclude that k (x, ξ ) is regular in the
The corrective solution closed domain −a ≤ (x, ξ ) ≤ a.
Consider a body geometrically identical to the initial The solution B (ξ ) in (68) is determined in the class
cracked body (Fig. 1) but with no remote loading now. of Hoelder continuous functions and may be written as
The only loading applied is along the crack faces. This a product of a regular bounded function and a funda-
consists of equal and opposite tractions to those gene- mental solution. Asymptotic analysis, within the fra-
rated in the un-cracked body. The boundary conditions mework of the couple-stress elasticity, showed that the
along the faces of the crack are written as displacement ux behaves as ∼ r 1/2 in the crack tip
Distributed dislocation approach for cracks 91
0.7 7
a 20 a 10
0.6 a 10 6
0 .3
0.5 a 5 5
µu x 0.4 1 2 4
2 yx
0 0.3 clas .
K II 3
0.2
classical elasticity 2 classical elasticity
0.1 couple-stress elasticity
couple-stress elasticity
1
0
-1 -0.5 0 0.5 1
x 0
0 1 2 3 4 5 6
x
Fig. 2 Normalized upper-half tangential crack displacement
profile (ν = 0.3) Fig. 4 Distribution of the shear stress ahead of the cracktip
1.8
1.7 0
and ν = 0 the increase is 62% and 73%, respectively.
0 .2
1.6
It should be noted that when
/a = 0 (no couple-
1.5 0 .4 stress effects) the above ratio becomes evidently
K
1.4 KI I /KIclas.
I = 1. Therefore, the ratio plotted in Fig. 3
K clas .
exhibits a finite jump discontinuity at
/a = 0; the ratio
1.3
at the tip of the crack rises abruptly as
/a departures
1.2 from zero. The same discontinuity was observed by
1.1 Sternberg and Muki (1967), who attributed that kind of
1
behavior to the severe boundary-layer effects predicted
0 0.2 0.4 0.6 0.8 1 by the couple-stress elasticity in stress-concentration
a
problems. Finally, it can be shown that the ratio
Fig. 3 Variation of the ratio of stress intensity factors in couple- decreases monotonically with increasing values of
/a
stress elasticity and in classical elasticity and tends to unity as
/a → ∞. The case
/a → ∞ is
rather impractical since generally the relation between
lengths in a usual crack problem will be
<< a, i.e.
x → a + the following asymptotic relations hold
the crack length will be much greater than the mate-
a
B (ξ ) rial length. However, in an attempt to explain the lat-
dξ = O (x − a)−1/2 ,
−a x − ξ ter finding, we note that the case
→ ∞, with a =0,
α resembles a situation where, in a sense, there is no
B (ξ ) k (x, ξ ) dξ = O (1) , x > a, (79a,b) microstructure in the body, since the ‘building blocks’
−α
of the material are of infinite size. Of course, this case
where the dislocation density is defined in (71). Thus, has an obscure physical meaning, but, as far as stresses
we conclude that σyx exhibits a square root singula- are concerned, the solution shows that the material
rity at the crack tip. In light of the above, we define exhibits a behavior similar to the one for a material
the stress intensity factor in couple-stress elasticity as governed by the classical theory.
KI I = lim [2π (x − a)]1/2 σyx (x, y = 0) for the Further, the distribution of the shear stress σyx ahead
x→a +
right crack tip (x > a) The dependence of the ratio of the of the crack tip (see Fig. 4) shows that the couple-
stress intensity factor in couple-stress elasticity KI I to stress effects are dominant for x <
, whereas outside
the one in classical elasticity upon
/a is given in Fig. 3. this zone σyx gradually approaches the distribution of
It is observed that for
/a → 0 and Poisson’s ratio the classical solution. For convenience, a new variable
ν = 0.4, there is a 50% increase in KI I when couple- x̄ = x − a is introduced measuring now distance from
stress effects are taken into account, while for ν = 0.2 the crack tip in the RHS of Fig. 1.
Distributed dislocation approach for cracks 93
0.2 y
0 σ0
a = 10
-0.2 ν = 0 .3
(2 π )1 2 m xz
1 2 clas . -0.4
K II
-0.6
-0.8
x
a a
-1
0 2 4 6 8 10
x
equal and opposite tractions to those generated in the crack tip region, where r denotes the polar distance
un-cracked body, i.e. from the crack tip. Thus, the dislocation density B (ξ )
1 can be expressed as
σyz + ∂x myy = −σ0 , myx = 0 for |x| < a
2 1/2
and y = 0. B (ξ ) = f (ξ ) a 2 − ξ 2 , (88)
(84a,b)
where f (ξ ) is a continuous bounded function in ξ ≤ |a|.
The corrective stresses in (84a,b) may be generated by Finally, to ensure uniqueness the dislocation density
a continuous distribution of discrete screw dislocations must satisfy the following auxiliary condition stem-
along the crack faces. The stresses induced by the conti- ming from the requirement of single-valuedness of the
nuous distribution of dislocations are obtained as inte- displacement along a closed loop around the crack
grals of Eqs. 59–62. Note that (84b) is automatically a
satisfied since a discrete screw dislocation does not give B (x)dx = 0. (89)
rise to couple-stresses myx along the crack line. Then, −a
satisfaction of the boundary condition (84a) leads, after
Now, the near-tip behavior of the stress and couple-
lengthy calculations, to the governing hypersingular
stress field for the mode III problem can be determi-
IE of the mode III problem in couple-stress elasticity
ned from the singular nature of the respective stress
(|x| < a)
and couple-stress field of a discrete screw dislocation.
a Again, confining our attention to the RHS crack tip and
c1 c2
2 taking into account the following result (Chan et al.
− σ0 = = + +c3 k (x, ξ ) B (ξ ) dξ
−a x−ξ (x−ξ )3 2003)
(85) a
∂n B (ξ )
− dξ = O (x − a) 1/2−n
, for n ≥ 0
where = signifies Hadamard’s finite-part integration ∂x n −a x − ξ
(see e.g. Kutt 1975; Paget 1981), B (ξ ) is the dislo- as x → a + , x > a (90)
cation density function at the point ξ (|ξ | < a), and
with the dislocation density being given by (88), we
µ β 2 + 2β + 9 µ (1 + β) (β − 3)
c1 = , c2 = , conclude that τyz , σyz given as the integrals of (58b)
16π 2π
and (59b) behave as ∼ x̄ −3/2 and ∼ x̄ , respecti-
1/2
µ (1 + β)2
c3 = . (86) vely, whereas the couple-stresses mxx , myy given by
π the integration of (60a,b) exhibit a square root singula-
Further, the kernel k (x, ξ ) is defined as rity at the crack tip. Again, x̄ = x − a is the distance
2 2
2 from the RHS crack tip along the crack line. Finally,
k(x, ξ ) = − 6 K2 (|x − ξ |/
) − in light of the above, the total shear stress defined as
(x − ξ )3 (x − ξ )2
tyz = σyz + (1/2) ∂x myy has the following asymptotic
1 1
+ + 3K0 (|x − ξ |/
) behavior tyz ∼ x̄ −3/2 near the crack tip. Such a beha-
2 4(x − ξ )
vior was detected before in the mode III crack problem
1
−5K2 (|x − ξ |/
) − . (87) of gradient elasticity (Georgiadis 2003). The two pro-
4 blems present similarities in their mathematical ana-
Expanding k (x, ξ ) in series as x → ξ and using the lysis. Finally, as we show in Appendix C, despite the
asymptotic properties of the modified Bessel functions, hypersingular nature of the above stress field, the strain-
it can be readily shown that k (x, ξ ) is regular in the energy density is integrable in the vicinity of crack tips
closed domain −a ≤ (x, ξ ) ≤ a. We also note that when and, also, the J -integral takes a bounded value.
β = −1 (i.e. when η = −η ), Eq. 85 degenerates For the numerical solution of the hypersingular inte-
into the SIE that governs the counterpart problem in gral equation in 85, the appropriate quadrature is
classical elasticity. constructed here by taking into account the cubic singu-
In addition, Zhang et al. (1998) showed, by using larity of the integral equation and the endpoint behavior
the Williams eigenfunction asymptotic analysis, that of the dislocation density (details are given in Appen-
the crack face displacement behaves as ∼ r 3/2 in the dix B). Equation 85 after the appropriate normalization
Distributed dislocation approach for cracks 95
n
0.8
π c2 t k 1
− σ0 = − 2 (−1)i+k − 0.6
p 2 1 − tk2 tk − si a 50
i=1 0.4
n a 10
f (si ) 1 − si2 π w 0.2
× +
a 5
1/2
(1 + n) 0
0
(tk − si ) 1 − tk2
i=1
-0.2
c1 c2
× + + c3 k (t ,
k is ) -0.4
(tk − si ) p2 (tk − si )3
-0.6
× f (si ) 1 − si2 , (91) -0.8
-1
where -1 -0.5 0 0.5 1
1 x
k(tk , si ) = − 2
p (t − si )3
k Fig. 7 Normalized upper and lower crack displacement profiles
2 1 under remote mode III loading (β = 0)
6 K2 (p|tk − si |) − 2 +
p (tk − si )2 2
1
+ 3K0 (p|tk − si |)
4(tk − si )
3
1
− 5K2 (p|tk − si |) − , (92) a 500
4 2
0 .9
with p = a/
, and the set of the n discrete integration 0
1 0 .9
points are given by 2
1 2
t yz
clas . 0
si : Un (si ) = 0, si = cos(iπ /(n + 1)), i = 1, . . . , n, K III
(93a) -1
microstructure are pronounced (i.e. for x >
) the total Acknowledgements This paper is a partial result of the Pro-
shear stress tends to the classical KIclas.
I I shear stress.
ject PYTHAGORAS II / EPEAEK II [Title of the individual
program: “Application of gradient theory for the solution of
boundary value problems by the use of analytical methods and
mixed adaptive finite elements” (# 68/8213)]. This Project is
6 Concluding remarks co-funded by the European Social Fund (75%) of the European
Union and by National Resources (25%) of the Greek Ministry
of Education.
In this paper, the technique of the distributed dislo-
cations was used in order to solve finite-length shear
crack problems in couple-stress elasticity. The tech-
Appendix A: The screw dislocation in couple-stress
nique provides an alternative approach to the elabo-
elasticity
rate analytical method of dual integral equations used
before to attack asymptotically the mode I crack pro-
Let the direct Fourier transform and its inverse be
blem. Moreover, the present approach is capable to pro-
defined as
vide a full-field solution. In fact, we have obtained here ∞
1
the stress distribution ahead of the crack tips and the w ∗ (ξ, y) = w (x, y) eixξ dx, (A1a)
crack-face displacements (i.e. our results are not res- (2π )1/2 −∞
tricted to the crack-tip region). Also, our solution to ∞
1
the finite-length crack in mode III is quite novel in the w (x, y) = w ∗ (ξ, y) e−ixξ dξ, (A1b)
literature. (2π )1/2 −∞
The governing integral equations are derived using where i ≡ (−1)1/2 . Transforming the field equation (47)
the discrete-dislocation stress fields in couple-stress with (A1a) gives the following ODE
elasticity, as the Green’s functions of crack problems.
d 4 w∗ 2 2 d 2 w∗
In particular, it is shown that the mode II problem is
2 − 2
ξ + 1 +
2 ξ 4 + ξ 2 w ∗ = 0,
governed by a single singular integral equation. In the dy 4 dy 2
where δ+ (ξ ) = 1/2 (δ (ξ ) + i/π ξ ) is the Heisenberg arises naturally in generalized continuum theories
delta function (Roos 1969) and δ (ξ ) is the Dirac distri- where the field equations and the boundary conditions
bution. The constants A1 (ξ ) and A2 (ξ ) are now com- are of higher order than the respective ones in classical
puted using the transformed boundary conditions i.e. elasticity. Although a lot of work has been done in the
literature for Hadamard type integrals (a = 2) (see e.g.
A1 (ξ ) = − (π/2)1/2 b 1 +
2 (1 + β) ξ 2 δ+ (ξ ) ,
Kutt 1975; Paget 1981; Ioakimidis 1983, 1995; Kaya
A2 (ξ ) = (π/2)1/2 b
2 (1 + β) ξ 2 δ+ (ξ ) , (A6a,b) and Erdogan 1987; Monegato 1987, 1994; Tsamas-
where β = η /η.
With the aid of the inversion formula phyros and Dimou 1990; Korsunsky 1998; Kabir et
in (A1b), we obtain the integral representation for the al. 1998; Hui and Shia 1999), only a few papers have
displacement field due to a screw dislocation been published concerning integrals with a > 2. In a
recent work by Chan et al. (2003), a systematic treat-
b ∞ −y|ξ | ment of hypersingular integrals was presented based
w (x, y) = − e − (ξ
)2 (1 + β)
2 −∞ on the Kaya/Erdogan approach. This approach leads to
very good results, with the only caveat that when the
y (1+
2 ξ 2 )
1/2
−1
(B.1) can be rewritten as
1 Un (t) = 1 − t 2 − (n + 1) Tn+1 (t) + tUn (t) ,
ds 1 d2
S (t) = = f (s) w (s) = n ≥ 0, (B.10)
−1 (s − t)3 2 dt 2
1 1/2 ds we write (B.8), after some lengthy algebra, under the
× − f (s) 1 − s 2 form
−1 (s − t) tk
p 1 1/2 − Tj+1 (tk ) + 2 (1 + n) Uj (tk ) − (1 + n)
1 d 2 1 − tk2
= B j − U j (s) 1 − s 2
n
2 dt 2 −1 1 − si2
j =0
× Uj (tk ) = 2 Uj (si ).
p (1 + n) (si − tk )3
ds π i=1
× =− Bj Tj+1 (t) , (B.4) (B.11)
(s − t) 2
j =0
π
where the prime denotes differentiation with respect to
Using (B.2), multiplying (B.9) by 2 Bj
and summing
over j from 0 to p, we then get
t. We note that the interchange of the order of diffe-
rentiation and integration in (B.4) is valid in view of ∼ −π (1 + n) f (tk ) + π (1 + n) tk f (tk )
S (tk ) =
results by Monegato (1994). 2 1 − tk2
n
Next, we establish the following identity 1 − si2
Tj +1 (t) Un (t) − Tn+1 (t) Uj (t) +π f (si ). (B.12)
− i=1
(1 + n) (si − tk )3
Un (t)
n
One further step is needed now that would lead to the
ai
= for j < n, si : Un (si ) = 0, (B.5) evaluation of the right hand side of (B.12) only at n
si − t points si : Un (si ) = 0. This can be done with the aid of
i=1
where the partial-fraction expansion above is possible the Lagrange interpolation formula, which will be exact
because the degree of the numerator in the left hand within the class of polynomials chosen to represent
side of (B.5) is less than that of the denominator. It can f (t)
n
easily be found (Korsunsky 1998) that the coefficients Un (t)
ai in (B.5) are given by the relation f (t) =
f (si ). (B.13)
Un (si ) (t − si )
i=1
1 − si2
ai = Uj (si ) . (B.6) Differentiating (B.13) with respect to t and then sub-
1+n stituting t with t = tk : Tn+1 (t) = 0, we get
Equation (B.5) takes now the form n
t k 1
Tn+1 (t) Uj (t) f (tk ) = Un (tk ) −
− Tj +1 (t) + 1 − tk2 (tk − si )
Un (t) i=1
n
1 − si2 f (si )
= Uj (si ) . × . (B.14)
(1 + n) (si − t)
(B.7) Un (si ) (tk − si )
i=1
In light of the above analysis, (B.12) can be written as
Differentiating (B.7) twice with respect to t and selec- n
ting a discrete set of points tk , k = 1, . . . , n + 1 such 1 tk
∼
S (tk ) = π −Un (tk ) +
that Tn+1 (t) = 0, we obtain (si − tk ) 2 1 − tk2
i=1
Uj (tk )
− Tj+1 (tk ) + Tn+1
(tk ) ×
1
Un (tk ) (si − tk ) Tn+1 (si )
U (t ) U (t ) − U (t ) U (t )
j k n k j k n k 1
+2Tn+1 (tk ) + f (si ) 1 − si2 ,
Un2 (tk ) (1 + n) (si − tk ) 3
n
1 − si2 (B.15)
=2 Uj (si ) . (B.8)
i=1
(1 + n) (si − t)3
where
Further, employing the well known identities about the (2k − 1) π
derivatives of Chebyshev polynomials tk : Tn+1 (tk ) = 0, tk = cos ,
2 (1 + n)
Tn+1 (t) = (n + 1) Un (t) , n ≥ 0, (B.9) k = 1, . . . , n + 1,
Distributed dislocation approach for cracks 99
S t et S t sin t 2
0.3 0.08
Chan et al. (2003) Chan et al. (2003)
0.2 Proposed Quadrature 0.06 Proposed Quadrature
0.1 0.04
0 0.02
-0.1 0
f s
f s
-0.2 -0.02
-0.3 -0.04
-0.4 -0.06
-0.08
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
s s
Fig. B1 Solution of the hypersingular integral equation (B.1) using the proposed quadrature and comparison with the semi-analytical
method of Chan et al. (2003)
iπ where the remainder may consist of Cauchy type and
si : Un (si ) = 0, si = cos , i = 1, . . . , n.
1+n regular kernels. In that case, (B.17) takes the form
1 1/2
= K (s, t) f (s) 1 − s 2 ds
Finally, taking into account that −1
n
1 t k
∼
=π (−1)i+k +
1/2 (si − tk ) 2 1 − tk2
Un (tk )=(−1)k+1 1 − tk2 , Tn+1 (si ) = (−1)i , i=1
1 K (si , tk )
(B.16) × + f (s i ) 1−s 2
i .
(si −tk ) 1 − t 2
1/2 (1 + n)
k
(B.19)
we write the resulting formula under the form
To check the validity of the proposed quadrature, we
1 solve the hypersingular integral equation in (B.1) for
1/2 ds
= f (s) 1 − s 2 ∼ two cases, i.e. for the loading function S (t) being defi-
=
−1 (s − t)3 ned as: (i) S (t) = et , and (ii) S (t) = sin t 2 . For single-
n
1 tk valuedness, the following auxiliary condition should
i+k
π (−1) + also be taken into account
(si − tk ) 2 1 − tk2 1 1/2
i=1
f (s) 1 − s 2 ds = 0. (B.20)
1 1 −1
× 1/2 +
(si − tk ) 1 − tk2 (1 + n) (si − tk )3 Then, (B.17) and (B.20) form a system of n + 2 equa-
tions in n unknowns which is solved in the least-squares
× f (si ) 1 − si2 . (B.17) sense. It is shown (see Fig. B1) that our results are in
excellent agreement with the ones obtained by using
the semi-analytical method of Chan et al. (2003).
It is noteworthy, that formula (B.17) also holds in
precisely the same form for the more general case when
the integral kernel is split up into a hypersingular part Appendix C: Evaluation of the strain-energy
of order a = 3 and a remainder density at crack tips and the J-integral
body of the paper lead to an integrable strain-energy The cases of mode II and mode III cracks are exa-
density in the vicinity of crack tips and also lead to a mined in what follows.
bounded value of the J -integral. The procedure follo- Mode II
wed is analogous in many respects with the one adopted In the case of plane-strain, the strain-energy density
in the work by Georgiadis (2003). reads
The strain-energy density function in (20) reads, in 1
terms of stresses W = (1 − ν) τxx 2
+ τyy2
+ 2τxy
2
− 2ντxx τyy
4µ
W =
1
τij τij −
ν
τii τjj + 2
1
1 2
2µ 1+ν 4
1 − β 2 + m xz + m yz .
2
(C4)
8µ
2
mij mij − βmij mj i , (C1) As shown before, the couple-stresses mxz , myz are
bounded (non-singular) in the crack-tip vicinity in the
where β is the ratio of the couple-stress moduli defined mode II case, whereas both the asymmetric and symme-
as β = η /η. tric stresses exhibit a square root singularity (see also
Further, the path-independent J -integral within the Huang et al. 1997). Now, the term in square brackets in
couple-stress theory is given by (Atkinson and Lep- (C4) is the same as in classical elasticity and behaves in
pington 1974; Lubarda and Markenskoff 2000) exactly the same way, while the second term (the one
∂uq ∂ωq involving couple-stresses) is bounded in the crack-tip
J = W nx − Pq − Rq d
∂x ∂x vicinity. Therefore, by following the standard proce-
∂uq ∂ωq dure to check upon the integrability of the strain-energy
= W dy − Pq + Rq d , (C2) density around a singularity (see e.g. Barber 1992), we
∂x ∂x
where a Cartesian rectangular coordinate system is atta- conclude that the strain-energy density is integrable
ched to the RHS crack tip with the distance x measured indeed in the crack-tip vicinity.
now from the tip, is a piece-wise smooth simple two- Further, taking into account that in the mode II case
dimensional contour surrounding the crack-tip, W is both the normal stress σyy and the couple-stress
myz
the strain-energy density, uq is the displacement, ωq are zero along the crack line y = 0± and that the
is the rotation, Pq is the force-traction defined in (18), crack-faces are defined by n = (0, ±1), the J -integral
and Rq is the couple-traction defined in (19). in (C3) finally takes the form
ε
For the evaluation of the J -integral, we consider the ∂ux (x, y = 0)
J = lim 2 σyx (x, y = 0) ·
rectangular-shaped contour in Fig. C1 with vanishing ε→+0 −ε ∂x
“height” along the y- direction and with ε → +0. This
type of contour permits using solely the asymptotic × dx . (C5)
near-tip stress and displacement fields. It is
noted that
Now, in view of the asymptotic behavior of the fields
upon this choice of contour, the integral W dy in
entering (C5), we obtain
(C2) becomes zero if we allow the ‘height’ of the rec-
tangle to vanish. In this way, the expression for the A2I I ε −1/2 −1/2
J -integral becomes J = lim (x+ ) (x− ) dx
ε ε→+0 µ −ε
∂uq ∂ωq
J = lim 2 Pq + Rq dx . (C3) π A2I I
ε→+0 −ε ∂x ∂x = , (C6)
2µ
where the product of distributions inside the integral
y
was obtained by the use of Fisher’s theorem (see e.g.
Georgiadis 2003), i.e. the operational relation (x− )λ
(x+ )−1−λ = −π δ (x) [2 sin (π λ)]−1 with λ = −1,
−2, −3, ... and δ (x) being the Dirac delta distribu-
0 x tion. Finally, we note that the amplitude factor AI I
is connected with the asymptotic results of Huang et
al. (1997), in the mode II case, through the relation
Fig. C1 Rectangular-shaped contour surrounding the cracktip AI I = 21/2 [(3 − 2ν) (1 − ν)]1/2 BI I .
Distributed dislocation approach for cracks 101
Huang Y, Zhang L, Guo TF, Hwang KC (1997) Mixed mode Monegato G (1994) Numerical evaluation of hypersingular inte-
near tip fields for cracks in materials with strain-gradient grals. J Comp Appl Math 50:9–31
effects. J Mech Phys Solids 45:439–465 Muki R, Sternberg E (1965) The influence of couple-stresses
Huang Y, Chen JY, Guo TF, Zhang L, Hwang on singular stress concentrations in elastic solids. ZAMP
KC (1999) Analytic and numerical studies on mode 16:611–618
I and mode II fracture in elastic-plastic materials with Nowacki W (1972) Theory of micropolar elasticity. CISM Inter-
strain gradient effects. Int J Frac 100:1–27 national Centre for Mechanical Sciences No. 25. Springer-
Hui CY, Shia D (1999) Evaluations of hypesingular integrals Verlag
using Gaussian quadrature. Int J Numer Methods Eng Nowacki W (1974) On discrete dislocations in micropolar elas-
44:205–214 ticity. Arch Mech 26:3–11
Ioakimidis NI (1983) A direct method for the construction of Paget DF (1981) The numerical evaluation of Hadamard finite-
Gaussian quadrature rules for Cauchy type and finite-part part integrals. Numer Math 36:447–453
integrals. Anal Numer Theor Approx 12:131–141 Paul HS, Sridharan K (1980) The penny-shaped crack problem
Ioakimidis NI (1995) Remarks on the Gauss quadrature rule for a in micropolar elasticity. Int J Eng Sci 18:651–664
particular class of finite-part integrals. Int J Numer Methods Paul HS, Sridharan K (1981) The problem of a Griffith crack on
Eng 38:2433–2448 micropolar elasticity. Int J Eng Sci 19:563–579
Kabir H, Madenci E, Ortega A (1998) Numerical solution Palmov VA (1964) The plane problem of non-symmetrical
of integral equations with logarithmic-, Cauchy- and theory of elasticity. Appl Math Mech (PMM) 28:1117–1120
Hadamard-type singularities. Int J Numer Methods Eng Roos BW (1969) Analytic functions and distributions in physics
41:617–638 and engineering. Wiley
Kaya AC, Erdogan F (1987) On the solution of integral equations Sternberg E, Muki R (1967) The effect of couple-stresses on the
with strongly singular kernels. Q Appl Math XLV 1:105– stress concentration around a crack. Int J Solids Struct 3:69–
122 95
Knesl Z, Semela F (1972) The influence of couple-stresses on Teodosiu C (1965) The determination of stresses and couple-
the elastic properties of an edge dislocation. Int J Eng Sci stresses generated by dislocations in isotropic media. Rev
10:83–91 Roum Sci Techn, Ser Mec Appl 10:1462–1480
Koiter WT (1964) Couple-stresses in the theory of elasticity, Part Toupin RA (1962) Perfectly elastic materials with couple
I. Proc Ned Akad Wet B67:17–29, II:30–44 stresses. Arch Ration Mech Anal 11:385–414
Korsunsky AM (1998) Gauss-Chebyshev quadrature formulae Tsamasphyros G, Dimou G (1990) Gauss quadrature rules for
for strongly singular integrals. Q Appl Math LVI 3:461– finite part integrals. Int J Numer Methods Eng 30:13–26
472 Van Dyke M (1964) Perturbation Methods in Fluid Mechanics.
Kroener E (1963) On the physical reality of torque stresses in Academic Press, New York
continuum mechanics. Int J Eng Sci 1:261–278 Vardoulakis I, Sulem J (1995) Bifurcation analysis in geome-
Kutt HR (1975) The numerical evaluation of principal value inte- chanics. Blackie Academic and Professional, Chapman and
grals by finite-part integration. Numer Math 24:205–210 Hall, London
Lubarda VA, Markenskoff X (2000) Conservation integrals in Voigt W (1887) Theoritiscke Studien uber die Elastizitatsve-
couple stress elasticity. J Mech Phys Solids 48:553–564 rhaltnisse der Krystalle. Abhandl Ges Wiss Gottingen 34:3–
Mindlin RD (1964) Micro-structure in linear elasticity. Arch 51
Ration Mech Anal 16:51–78 Weitsman Y (1965) Couple-stress effects on stress concentration
Mindlin RD (1963) Influence of couple-stresses on stress around a cylindrical inclusion in a field of uniaxial tension.
concentrations. Exp Mech 3:1–7 ASME J Appl Mech 32:424–428
Mindlin RD, Tiersten HF (1962) Effects of couple-stresses in Zhang L, Huang Y, Chen JY, Hwang KC (1998) The mode-III
linear elasticity. Arch Ration Mech Anal 11:415–448 full-field solution in elastic materials with strain gradient
Misicu M (1965) On a general solution of the theory of singular effects. Int J Frac 92:325–348
dislocations of media with couple-stresses. Rev Roum Sci
Techn, Ser Mec Appl 10:35–46
Monegato G (1987) On the weights of certain quadratures for
the numerical evaluation of Cauchy principal value integrals
and their derivatives. Numer Math 50:273–281
Bifurcation of equilibrium solutions and defects nucleation
Claude Stolz
Abstract The purpose of this article is to revise some a material. Under a uniform radial tensile loading p,
concepts on defects nucleation based on bifurcation the solution is assumed to be radial. We investigate the
of equilibrium solutions. Equilibrium solutions are ob- possible branches of radially symmetric configurations
tained on a homogeneous body and on a body with an involving a traction free internal cavity bifurcation from
infinitesimal defect such as cavity under the same pre- the homogeneous sphere. At the bifurcation point pc ,
scribed dead load. First void formation and growth in the equilibrium solution for the porous sphere under the
non linear mechanics are examined. A branch of radial load pc tends to the equilibrium solution of the homo-
transformation bifurcates from the undeformed con- geneous sphere when the volume of the cavity tends to
figuration in presence of a small cavity. Two cases of zero.
behaviour are examined. One case is the growth of the Ball (1982) and Horgan and Pence (1989) have stud-
cavity by only the deformation of the shell. In another ied different classes of bifurcation problems for non lin-
modelling the cavity evolves like a damaged zone, the ear incompressible elasticity, which simulate the
transition between the sound part and the damaged one appearance of a cavity inside a homogeneous sphere.
is governed by a local criterium. Each configuration Example in finite incompressible elastoplasticity have
leads to the definition of a nucleation criterion based been also investigated by Chung et al. (1987). In both
on a presence of a bifurcation state, common state of cases for all values of p one possible solution is that
the homogeneous body and a body with an infinitesimal the sphere remains solid, this is due to incompressibil-
defect. ity. On the other hand, for a certain range of loading p
one has another configuration with an internal infini-
Keywords Bifurcation · Nucleation of defects · tesimal cavity. But as pointed out by Ball, only specific
Hyperelasticity · Local damage · Composite sphere classes of hyperelastic incompressible materials have
the possibility of such cavitation under radial dead load.
1 Introduction However if the core of the sphere is made by a damaged
material which does not suffer any tensile stress after
We consider a composite sphere with external radius a critical stretch, we obtain a critical pressure for all
Re composed by a cavity of radius Ri surrounded by classes of hyperelastic incompressible materials. This
point generalizes the point of view of Ball and gives
C. Stolz (B)
Laboratoire de Mécanique des Solides, Ecole new definition of nucleation of defects.
Polytechnique, CNRS, UMR7649, Palaiseau 91128, France In Sect. 3, the possibility of cavitation is generalized
e-mail: stolz@lms.polytechnique.fr to other classes of materials in small strains.
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 103
doi: 10.1007/978-1-4020-6929-1_10, © Springer Science+Business Media B.V. 2007
104 C. Stolz
We point out that this point of view gives for each Horgan CO, Pence TJ (1989) Void nuleation in tensile dead
configuration an upper bound for the critical value of loading of a composite incompressible non linearly elastic
sphere. J Elast 21:61–82
nucleation because the approach is based on an analysis Ogden RW (1997) Non linear elastic deformations. Dover
of bifurcation, then the value depends on the geometry Publications
and of the path of the loading. Stolz C, Pradeilles-Duval RM (1997) Thermomechanical ap-
proach of running discontinuities. In: Variations of domains
and free-boundary problems in solids mechanics. IUTAM,
Springer, pp 245–251
References Stolz C, Pradeilles-Duval RM (2004) Stability and bifurcation
with moving discontinuities. In: Mechanics of material
Ball JM (1982) Discontinuous equilibrium solutions and cavi- forces, AMMA11, Springer, Chap 26, 261–268
tation in non linear elastiticty. Phil Trans R Soc Lond A
306:557–610
Chung DT, Horgan CO, Abeyaratne R (1987) A note on a bifur-
cation problem in finite plasticity related to void nucleation.
Int J Solids Struct 23:983–988
Theoretical and numerical aspects of the material and spatial
settings in nonlinear electro-elastostatics
Duc Khoi Vu · Paul Steinmann
Abstract The formulation of the spatial and mate- motion form) and in an appropriate continuum mechan-
rial motion problem in nonlinear electro-elastostatics ical setting (the material setting). This material form of
is revisited in this work. A finite element discretiza- the balance equation of linear momentum has proved
tion is realized and a numerical example is presented to be very useful in fracture mechanics. Concerning
to demonstrate possible application of the formulation with material defects, the problem of material forces in
in studying the closing process of cracks. nonlinear electro-elasticity was also studied by many
other authors (Epstein and Maugin 1990; Epstein and
Keywords Nonlinear electro-elastostatics · Electro- Maugin 1991; Huang and Batra 1996; Kalpakides and
mechanical coupling · Material forces Agiasofitou 2002; Maugin and Epstein 1991; Pak
and Herrmann 1986a, b; Trimarco 2007). In a recent
study (Vu and Steinmann 2007), the material force
1 Introduction problem in nonlinear electro-elastostatics was revis-
ited. By using a variational approach, the striking sim-
The nonlinear electro-elastic behavior of electro- ilarity between the spatial and material motion stresses
sensitive materials that exhibit large displacements and as well as the similarity between the spatial and mate-
change their mechanical properties in response to the rial governing equations of the two problems of elas-
application of electric field was and still is the subject tostatics and electro-elastostatics were revealed. With
of many researches due to the interesting application the introduction of some material tractions and body
of these materials in developing artificial forces to capture the energetic changes that are associ-
muscles (Bar-Cohen 2002). Because of their great poten- ated with changes in material configuration and mater-
tial, the analysis of defects of electro-sensitive materi- ial motions of defects relative to the ambient material,
als is of high interest. Maugin (1993) considered the it was shown that the two problems of material and spa-
material defect problem (in form of material inhomo- tial motions are only equivalent for defect free bodies.
geneity) related to nonlinear electro-elasticity with the In this work, the formulations presented in the work of
help of the material forces, where the balance equa- Vu and Steinmann (2007) are discretized by the finite
tion of linear momentum of the spatial motion prob- element method for the case of a crack. A numerical
lem is transformed into an appropriate form (material example is presented to highlight the possible applica-
tion of the material force method in studying the closing
D. K. Vu (B) · P. Steinmann
Chair of Applied Mechanics, University of Kaiserslautern,
process of cracks in cracked structures made of electro-
P.O. Box 3049, 67653 Kaiserslautern, Germany sensitive materials undergoing large deformation. For
e-mail: vuduc@rhrk.uni-kl.de details about the material force method and its
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 109
d oi: 10.1007/978-1-4020-6929-1_11, © Springer Science+Business Media B.V. 2007
110 D. K. Vu, P. Steinmann
Fig. 1 Spatial motion problem in nonlinear elastostatics Fig. 2 Material motion problem in nonlinear elastostatics
applications, the readers are referred to, for example, partial derivative of the material position vector with
the works (Denzer 2006; Denzer et al. 2003; Kienzler respect to the spatial one. In the material motion prob-
and Herrmann 2000; Kuhl and Steinmann 2004; Liebe lem, material motion stresses are defined as material
et al. 2003; Steinmann 2000, 2002a, b, c, 2005; motion Cauchy (Eshelby) stress (in reference to the
Steinmann et al. 2001). material configuration) and material motion Piola stress
p (in reference to the spatial configuration). Again, cor-
responding to these two material motion stress tensors,
2 Material and spatial settings in nonlinear two governing systems can be established. The first one
electro-elastostatics is constructed in reference to the material configuration
and uses the material motion Cauchy stress tensor as
2.1 Spatial and material motion problem in nonlinear variable. The second one is constructed in reference to
elastostatics the spatial configuration and uses the material motion
Piola stress tensor as variable.
The two problems that are concerned with the mater- In nonlinear elasticity, it is well-known that by the
ial force method are the spatial and the material motion direct approach, the balance equations of linear momen-
problems. For nonlinear elastostatics, in the spatial tum of the material motion problem (in reference to
motion problem (Fig. 1), the position vector x of a point the spatial or material configuration) can be derived
in the spatial (or deformed) configuration is described directly from their counterparts in the spatial motion
by a nonlinear spatial motion map ϕ that maps a point problem. This transformation can be realized by mul-
X from the material (or undeformed) configuration B0 tiplying the spatial motion balance equations of linear
into the spatial configuration Bt . The deformation is momentum by the deformation gradient F. As an alter-
characterized by the spatial motion deformation gra- native to the direct approach, the variational approach
dient F defined as the partial derivative of the spatial can be used to derive the governing equations of the
position vector with respect to the material one. Spa- spatial and material motion problem. As a first step
tial motion stresses are represented by the two spatial toward a more complete discussion on the formula-
motion stress tensors Cauchy σ and Piola P. Corre- tions of spatial and material motion problem, in this
sponding to these two spatial motion stress tensors, two work the system under consideration is assumed to
governing systems, each consists of one balance equa- be conservative in the sense that the elastic material
tion and one boundary condition, can be established. response can be characterized by some internal poten-
The first one is constructed in reference to the spatial tial energy densities and the conservative loading can
configuration and uses the spatial motion Cauchy stress be characterized by some external potential energy den-
tensor as variable. The second one is constructed in ref- sities per unit volume. Furthermore, only zero spatial
erence to the material configuration and uses the spatial motion tractions are considered. The whole system is
motion Piola stress tensor as variable. then characterized by some total potential energy den-
In the material motion problem (Fig. 2), the posi- sities U0F (F, ϕ, X), U0 f ( f , x, ) in reference to the
tion vector X of a point in the material configuration material configuration, or Ut F (F, ϕ, X), Ut f ( f , x, )
is described by a nonlinear material motion map in reference to the spatial configuration such that
that maps a point x from the spatial into the material U0F | F,ϕ,X = U0 f f ,x, and that Ut F | F,ϕ,X = Ut f
configuration. The deformation is characterized by the | f ,x, . These total potential energy densities are defined
material motion deformation gradient f defined as the as the sum of the corresponding internal and external
Theoretical and numerical aspects of the material and spatial settings 111
potential energy densities. The assumption about the symmetric. Besides the non-symmetric property of the
conservative property of the system under considera- Cauchy stress, it is noted that difficulties also appear
tion is made in order to make it possible to derive all in dealing with the jump conditions for Cauchy stress
stresses and all body forces from internal and exter- at the boundary of the considered body or across a sur-
nal potential energy densities. For the derivation of face of discontinuity within the body. This is due to
the material motion formulations, the core argument the fact that on the one hand the Cauchy stress dif-
here comes from the work of Steinmann (2005), who ference across a surface must balance both electrical
noted that taking a variation of a total potential energy and mechanical surface tractions. On the other hand,
functional at fixed spatial placement x only leads to a any traction measured by mechanical means is related
stationary point for the case of configurational equilib- to the contribution of both mechanical and electrical
rium. In more general cases, configurational or rather effects, since no available experiment can separate the
material tractions acting on the material boundary and effects of the Cauchy and Maxwell stresses unambigu-
material forces, acting on defects such as vacancies, ously (McMeeking and Landis 2005). This leads to the
interfaces, dislocations, cracks and the like must be definition of the so-called spatial motion total stress
considered. These material forces capture the energetic tensor σ̂ , which is the combination of the Cauchy and
changes that go along with material motions of the Maxwell stresses.
defects relative to the ambient material. For the sake of simplicity, let us assume that we
only have homogeneous Neumann boundary condi-
tions such that: d · n = 0, σ̂ · n = 0, where d is
2.2 Spatial and material motion problem in nonlinear the electric displacement and n is the outward point-
electro-elastostatics ing unit normal at the boundary ∂Bt . In the spatial
motion problem (Fig. 3), the balance equations of lin-
For the nonlinear electro-elastostatic problem, a vari- ear momentum and the (homogeneous) boundary con-
ational formulation for the spatial motion problem is ditions can be derived from the stationary condition of
built from the basic equations of electrostatics. The the augmented total potential energy functionals, which
obtained variational formulation states that the govern- requires the first variations of these functionals at fixed
ing equations of the nonlinear electro-elastostatic prob- material placement X to vanish:
lem is equivalent to the stationary condition of some
augmented total potential energy functionals, wherein δx Û0F d V = 0 and δ x Ût F dv = 0 (1.1–2)
B0 Bt
the augmented total potential energy densities Û0F ,
Û0 f (in reference to the material configuration) or Ût F , where Û0F = Û0F (F, E, ϕ, X), Ût F = Ût F (F, E,
Ût f (in reference to the spatial configuration) are com- ϕ, X), e and E denote, respectively, the electric field
puted by taking into account the elastic energy stored vectors in reference to the spatial and material configu-
in the bulk of the material and the electric energy con- rations. In reference to the material configuration, Û0F
tributed from the free space. Based on these augmented can be computed as:
1
total potential energy functionals, the balance equations Û0F = W0E (F, E; X) − ε0 J C −1 : [E ⊗ E]
of linear momentum and boundary conditions of the 2
material motion problem can be derived. +V0F (ϕ; X) (2)
In electro-elastostatics, the electric field acting on where the first term W0E (F, E; X) is the elastic energy
material is governed by Faraday’s law, Gauss’ law for density stored in the bulk of the material, the second
electricity, and a relationship describing the link between term represents the contribution of the free space, the
the electric polarization, the electric displacement and third term V0F (ϕ; X) is the external potential energy
the electric field. Because of the electric polarization, density characterizing external loads, ε0 is the vacuum
the electric field exerts on matter a body force that permittivity, J = det (F) and C = F t · F. The condi-
can be considered as a function of the electric field tions (1.1–2) lead to the governing equations:
vector and the electric polarization. With this electric ∇ x · σ̂ + bt = 0 and σ̂ · n = 0 (3)
body force, the balance equation of linear momentum
is the same as that of a normal nonlinear elastic sys- in reference to the spatial configuration, or:
tem except the fact that the Cauchy stress tensor is not ∇ X · P̂ + b0 = 0 and P̂ · N = 0 (4)
112 D. K. Vu, P. Steinmann
Fig. 3 Spatial motion problem in nonlinear electro-elastostatics Fig. 4 Material motion problem in nonlinear electro-
elastostatics
material and spatial motions are only equivalent for
in reference to the material configuration, where P̂
defect free bodies, wherein: B̂ dt = B̂ d0 = 0. Note
denotes the counterpart of the spatial motion Piola stress
that the stresses in the above systems can be computed
tensor in elastostatics, bt and b0 denote the spatial body
through the corresponding potential energy densities
forces, and N is the outward pointing unit normal at the
as:
boundary ∂B0 .
In the material motion problem (Fig. 4), if only dis- P̂ = J σ̂ · F −t = ∂ F Û0F and
tributed defects are considered, the energetic changes ˆ · f −t = ∂ f Ût f
p̂ = J −1 (9.1–2)
can be captured by some distributed configurational Furthermore, the spatial motion stress σ̂ and the
forces B̂ dt and material tractions T̂ dt in reference to the ˆ (Eshelby stress) have a similar
material motion stress
spatial configuration or B̂ d0 and T̂ d0 in reference to the energy-momentum type:
material configuration. Correspondingly, the first varia-
tions of the augmented total potential energy function- σ̂ = Ût f I − f t · p̂ + e ⊗ d and
als at fixed spatial placement x become equal to the ˆ = Û0F I − F t · P̂ + E ⊗ D
(10.1–2)
energy changes associated with these material forces: where D denotes the electric displacement vector in ref-
erence to the material configuration. For more details
δX Ût f dv = : B̂ dt · δdv
Bt B about the material and spatial settings in nonlinear
t electro-elastostatics, the readers are referred to the work
+ T̂ dt · δds ≤ 0 (5) of Vu and Steinmann (2007).
∂ Bt
in reference to the spatial configuration, or: 3 Finite element discretization
δX Û0 f d V = : B̂ d0 · δd V 3.1 Material force method
B0 B0
+ T̂ d0 · δd S ≤ 0 (6) Application of the material motion formulations above
∂ B0
in fracture mechanics is directly related to the compu-
in reference to the material configuration, where Û0 f = tation of the material surface force fsur defined as:
Û0 f ( f , e, x, ), Ût f = Ût f ( f , e, x, ), Û0 f ˆ · N dS
fsur = (11)
| f,e,x, = Û0F | F,E,ϕ,X , Ût f | f ,e,x, = Ût F | F,E,ϕ,X . ∂ B0
The conditions (5) and (6) lead to the balance equa- In order to compute this material force by the finite ele-
tions of linear momentum and boundary conditions of ment method, let us consider the balance equation of
the material motion problem written as: linear momentum (8.1) in the case B̂ 0 = 0 and B̂ d0 = 0.
∇ x · p̂ + B̂ t =: − B̂ dt and p̂ · n =: T̂ dt (7.1–2) By multiplying this equation with a test function (mate-
rial virtual displacement) W , under the necessary
in reference to the spatial configuration, or: smoothness and appropriate boundary assumptions, we
ˆ + B̂ 0 =: − B̂ d0 and
∇X · ˆ · N =: T̂ d0 (8.1–2) arrive at the virtual work equation:
in reference to the material configuration, where p̂ and W sur = W int (12)
ˆ denote the material motion total stress tensors, B̂ t where W sur denotes the material variation of the poten-
and B̂ 0 denote the material body forces. Based on these tial energy due to its complete dependence on the mate-
formulations, it can be shown that the two problems of rial position and W int denotes the material variation of
Theoretical and numerical aspects of the material and spatial settings 113
the potential energy due to its implicit dependence on in fracture mechanics and the integration of the nor-
the material position: mal projection of the material motion Cauchy stress
(Eshelby stress) over a surface (in 3D) or a line (in 2D)
W sur = W ·ˆ · N d S and enclosing a crack tip:
∂ B0
W int
= ∇X W : ˆ dV (13.1–2) J = lim ˆ · Nd S (19)
∂ V0 →0 ∂ V r
r
B0 0
By using the finite element method, the virtual dis- where ∂V0r is the regular part of the boundary ∂V0 .
placement field W can be discretized over the element The singular part of the boundary, ∂V0s , denotes the
e occupying the element domain B0e as: crack tip. The regular and singular parts of the boundary
underconsideration ∂V0 are defined such that: ∂V0 =
n en
Wh = NnWn (14) ∂V0r ∂V0s and ∅ = ∂V0r ∂V0s .
B0e It should be noted that the concept of J-integral (19)
n=1
is an extension of the J-integral in elasticity to elec-
where W n is the nodal value of W , N n is the shape
troelasticity. This extension using the theory of mater-
function at node n and n en is the number of nodes of
ial forces was given by, for example, Pak and Herrmann
element e. To evaluate the gradient of W with respect
(1986a, b), Epstein and Maugin (1990, 1991), Maugin
to X, the following formula is used:
and Epstein (1991) and Dascalu and Maugin (1994). In
n en
order to maintain the similitude with the formulation
∇X W h = W n ⊗ ∇X N n (15)
B0e of the J-integral in pure mechanics, or in other words,
n=1
for J to be path-independent, some special conditions
Over the element e, the left- and right-hand sides of must be specified, namely on the crack faces traction
(12) can be discretized as: and charge must be zero: σ̂ · n = 0, d · n = 0, which
n en are assumed above. For more details, see, for example,
sur,h
W e= Wn · Nn ˆ · N dS (16) the work of Dascalu and Maugin (1994) or
B0 ∂ B0e
n=1 Abendroth et al. (2002). The electric boundary con-
and: dition on the crack faces increases the complication
n en in analyzing crack problems in electroelasticity. With
ˆ · ∇X N n d V
W int,h = Wn · (17) respect to the electric boundary conditions, two types
B0e B0e
n=1 of crack are usually considered: impermeable (where
Because of the arbitrariness of the material virtual nodal electric displacements are zero on the crack surfaces)
displacement W n , the global discrete material nodal and permeable (where electric displacements are con-
force characterizing the material surface loads at node tinuous through the crack surfaces) crack. It is well-
n can be eventually computed as: known that electrically impermeable conditions lead
to overestimating, while electrically permeable con-
fsur,h
n = A n el ˆ · ∇X N n d V
(18) ditions lead to underestimating the influence of the
e=1
B0e electric field on crack propagation. In order to prop-
where n el is the number of elements used to discretize erly model cracks in electroelasticity, special attention
the domain B0 . Note that in (18) the material motion should be paid to the effect of the medium filling the
stress tensor can be computed by (10) when the spatial crack and a transition model bewteen permeable and
motion problem is solved (Vu et al. 2007). impermeable crack models with increasing crack open-
ing may be used, see for example (Qi et al. 2001; Wang
and Jiang 2003). In this work, for the sake of simplicity
3.2 Relation to J-integral we consider only the electrically impermeable condi-
tion d · n = 0.
Now consider the simple case of a structure with a sin- By taking the decomposition of the boundary ∂V0
gle crack lying in a subdomain V0 of B0 . A direct appli- into singular ∂V0s and regular parts ∂V0r into account,
cation of the material force method is based on the link the resulting material force acting on the whole bound-
between the vectorial generalization J of the J-integral ary ∂V0 is given by:
114 D. K. Vu, P. Steinmann
fsur = fsur,s + fsur,r = 0 (20) However, due to the lack of experimental results, mate-
rial properties are inadequate at this point. Therefore
where:
for the testing purpose, in this example we assume that
fsur,s = ˆ · Nd S and
the plate’s material has the following properties: bulk
∂ V0s modulus κ = 10 MPa, shear modulus µ = 5 MPa,
c1 = 10−3 Pa · m2 /V 2 and c2 = 6 × 10−5 Pa · m2 /V 2 .
fsur,r = ˆ · Nd S
(21.1-2)
∂ V0r The parameter λ is computed by using the relation-
ship: λ = κ − 2µ 3. By using these values, numerical
From (20), the vector J can be computed as: results confirmed that (24) is actually a good approxi-
ˆ · Nd S mation. Nevertheless, it should be noted that for other
J = −fsur,s = − (22)
∂ V0s material properties, the contribution of the free elec-
tric field may become important to the crack closing
The discretized version Jh of J can be therefore com- and should be counted for. Because the contribution
puted by the global discrete material nodal force at the of the free electric field to the energy release rate (or
node characterizing the crack tip. For details, see for J-integral) may be negative, as noted by Pak (1990)
example Steinmann et al. (2001). and Dascalu and Maugin (1994), the neglect of the free
electric field contribution may lead to slowing down of
the crack closing. However, with insufficient numerical
4 Numerical example evidence, we refrain from discussing further the effect
of the free electric field.
As numerical example, a 2-D plain strain rectangular Due to symmetry only half of the plate is modeled
plate (length a = 60 mm, width b = 40 mm) with a by 672 four-node quadrangular and 24 three-node tri-
crack at the center of the plate (crack length c = 20 mm) angular elements, in which the triangular elements are
is considered. The plate is assumed to be made of a used to model the crack tip. For simplicity, it is assumed
compressible neo-Hookean-like material, of which the that when close, the two crack surfaces have no elec-
elastic energy density W0E is defined as: trical contact. The loading process is divided into 36
µ λ steps. In the first 10 steps, increasing extension forces
W0 E = [C : I − 3] − µ ln J + [ln J ]2 are applied near the two ends of the plate. In the next 26
2 2
+c1 I : [E ⊗ E] + c2 C : [E ⊗ E] (23) steps, an increasing electric potential difference ϕ is
applied between the two ends of the plate while keep-
where the last term could be thought of as having the ing the extension forces constant. Under the application
nature of electrostriction coupling. For the purpose of of the extension forces, the crack is opened. With the
investigating quantitatively the closing of the crack, electric potential loading, the electrostatic forces cause
let us, for simplicity, ignore the contribution of the the plate to shrink and with a large enough ϕ, the
free space term ( 21 ε0 J C −1 : [E ⊗ E]) in the augmented electrostatic forces lead to closing of the crack. Fig-
total potential energy density Û0F (Eq. 2) and use in ure 5 presents the closing process of the crack from
the calculation: maximum opening (step 10) to complete closing (step
36). Zoom-in pictures of the crack opening are shown
Û0F ≈ W0E (24)
in Fig. 6. Material forces at the crack tip at different
This energy density means that in the absence of an steps are computed and normalized by the absolute
electric stimulation the material behaves exactly as a value obtained at step 10 (where maximum crack tip
material of the compressible neo-Hookean type in non- opening displacement or CTOD occurs). The normal-
linear elasticity. The elastic behavior of this material is ized material forces at the crack tip presented in Figs.
controlled by two parameters µ and λ, which are actu- 7 and 8 show clear changes in magnitude and direction
ally two Lamé coefficients. Under the application of of these forces. In the opening process of the crack, it
an electric field, the energy density (24) means that is observed that the material force vector at the crack
our material will exhibit a nonlinear coupling behavior tip lies on the crack axis and points toward the ambi-
through the term C : [E ⊗ E]. The material constants ent space (steps 1 to 10). The magnitude of this vector
c1 and c2 would have to be determined experimentally. increases with increasing CTOD. Under the application
Theoretical and numerical aspects of the material and spatial settings 115
Fig. 5 Crack closing under electric loading (from left to right: ϕ = 0, 2000, 4000, 6000, 8000 and 10400 V): deformed configuration
100 1.5
Normalized material force
Normalized material force
80 1
60 0.5
40 0
20 -0.5
0
-1
5 10 15 20 25 30 35 2 4 6 8 10 12 14
Loading step Loading step
Fig. 7 Normalized material force at crack tip: loading steps 1–36 Fig. 8 Normalized material force at crack tip: zoom of loading
steps 1–15
puted material forces at the tip of the crack offer useful
of the electric potential loading, this vector becomes
information and can be used in prediction and analysis
smaller and eventually changes direction (in Figs. 7
of crack closing.
and 8, positive values represent material forces point-
ing toward the ambient material). This observation sug- Acknowledgements The work is funded by German Research
gests a useful application of the material force method Foundation (DFG) within the Research Training Groups 814,
in the study of crack closing. Engineering Materials at Multiple Scales: Experiments, Model-
ing and Simulation; Grant Number: GRK 814.
5 Conclusion References
The material force problem in nonlinear electro- Abendroth M, Groh U, Kuna M, Ricoeur A (2002) Finite
elastostatics is considered in this work. By using the element-computation of the electromechanical J-integral
for 2-D and 3-D crack analysis. Int J Fract 114:359–378
finite element method, material surface forces are dis- Bar-Cohen Y (2002) Electro-active polymers: current capabili-
cretized and computed for the case of a single crack ties and challenges. In: Proceedings of SPIE – electroactive
structure made of electro-sensitive materials. The com- polymer actuators and devices
116 D. K. Vu, P. Steinmann
Dascalu C, Maugin GA (1994) Energy-release rates and path- Pak YE, Herrmann G (1986a) Conservation laws and the mater-
independent integrals in electroelastic crack propagation. ial momentum tensor for the elastic dielectric. Int J Eng Sci
Int J Eng Sci 32(5):755–765 24:1365–1374
Denzer R (2006) Computational configurational mechanics. Dis- Pak YE, Herrmann G (1986b) Crack extension force in a dielec-
sertation, Technische Universität Kaiserslautern, Germany tric medium. Int J Eng Sci 24:1375–1388
Denzer R, Barth FJ, Steinmann P (2003) Studies in elastic frac- Qi H, Fang D, Yao Z (2001) Analysis of electric boundary condi-
ture mechanics based on the material force method. Int J tion effects on crack propagation in piezoelectric ceramics.
Numer Methods Eng 58:1817–1835 Acta Mech Sinica 17(1):59–70
Epstein M, Maugin GA (1990) Inhomogeneities, Eshelby’s ten- Steinmann P (2000) Application of material forces to hyperelas-
sor and J-integral in electroelasticity. In: Hsieh RKT (ed) tostatic fracture mechanics. I. Continuum mechanical set-
Mechanical modelling of new electromagnetic materials. ting. Int J Solids Struct 37:7371–7391
Elsevier, Amsterdam pp 253–258 Steinmann P (2002a) On spatial and material setting of hypere-
Epstein M, Maugin GA (1991) Energy-momentum tensor and lastodynamics. Acta Mech 156:193–218
J-integral in electrodeformable bodies. Int J Appl Electro- Steinmann P (2002b) On spatial and material setting of thermo-
magn Mater 2:141–145 hyperelastodynamics. J Elasticity 66:109–157
Huang Y-N, Batra RC (1996) Energy-momentum tensors in non- Steinmann P (2002c) On spatial and material settings of hyper-
simple elastic dielectrics. J Elasticity 42:275–281 elastostatic crystal defects. J Mech Phys Solids 50:1743–
Kalpakides VK, Agiasofitou EK (2002) On material equations in 1766
second gradient electroelasticity. J Elasticity 67:205–227 Steinmann P (2005) On potential energy shifts in hyperelastic
Kienzler R, Herrmann G (2000) Mechanics in material space energy-momentum tensors. Technische Mechanik 25(3–
with applications to defect and fracture mechanics. 4):174–181
Springer, Berlin Heidelberg New York Steinmann P, Ackermann D, Barth FJ (2001) Application of
Kuhl E, Steinmann P (2004) Computational modeling of heal- material forces to hyperelastostatic fracture mechanics. II.
ing: an application of the material force method. Biomech Computational setting. Int J Solids Struct 38:5509–5526
Model 2:187–203 Trimarco C (2007) Material electromagnetic fields and material
Liebe T, Denzer R, Steinmann P (2003) Application of the mate- forces. Arch Appl Mech 77(2–3):177–184
rial force method to isotropic continuum damage. Comput Vu DK, Steinmann P (2007) Nonlinear electro- and magneto-
Mech 30:171–184 elastostatics: material and spatial settings. Int J Solids Struct
Maugin GA (1993) Material inhomogeneities in elasticity. Chap- (accepted)
man and Hall Vu DK, Steinmann P, Possart G (2007) Numerical modelling
Maugin GA, Epstein M (1991) The electroelastic energy- of non-linear electroelasticity. Int J Numer Methods Eng
momentum tensor. Proc Roy Soc London A 433:299–312 70:685–704
McMeeking RM, Landis CM (2005) Electrostatic forces and Wang XD, Jiang LY (2003) The effective electroelastic property
stored energy for deformable dielectric materials. ASME of piezoelectric media with parallel dielectric cracks. Int J
J Appl Mech 72:581–590 Solids Struct 40:5287–5303
Pak YE (1990) Crack extension force in a piezoelectric material.
ASME J Appl Mech 57:647–653
Energy-based r-adaptivity: a solution strategy
and applications to fracture mechanics
Michael Scherer · Ralf Denzer · Paul Steinmann
Abstract This paper deals with energy based Keywords Mesh optimization · Hyperelasticity ·
r-adaptivity in finite hyperelastostatics. The focus lies Constrained energyminimization · Material forces ·
on the development of a numerical solution strategy. Barrier method
Although the concept of improving the accuracy of
a finite element solution by minimizing the discrete
potential energy with respect to the material node point 1 Introduction
positions is well-known, the numerical implementation
of the underlying minimization problem is difficult. In The finite element solution of a boundary-value
this paper, energy based r-adaptivity is defined as a problem in solid mechanics is based on a variational rep-
minimization problem with inequality constraints. The resentation of the problem. In this paper, the focus lies on
constraints are introduced to restrict the maximum dis- problems that are associated with the principle of mini-
tortion of the finite element mesh. As a solution strategy mum potential energy. In a standard finite element
for the constrained problem, we use a classical barrier method the deformation is interpolated by a linear com-
method. Beside the theoretical aspects and the imple- bination of fixed shape functions that are predetermined
mentation, a numerical experiment is presented. We by the material node point positions and the connectiv-
illustrate the performance of the proposed r-adaptivity ity of the mesh. Following a Ritz method, the unknown
in the case of a cracked specimen. nodal deformations, the spatial node point positions, are
determined by minimizing the potential energy of the
discrete mechanical system. Since the necessary condi-
tion for a minimum of the discrete potential energy with
respect to the spatial node point positions requires the
M. Scherer (B) · R. Denzer
Department of Mechanical Engineering, equilibrium of internal and external spatial node point
Applied Mechanics, University of Kaiserslautern, forces, it is denoted as spatial equilibrium condition.
P.O. Box 3049, 67653 Kaiserslautern, Germany The discrete potential energy in the state of equi-
e-mail: mscherer@rhrk.uni-kl.de
librium solely depends on the finite element approxi-
R. Denzer mation, i.e. the element connectivity and the material
e-mail: denzer@rhrk.uni-kl.de
node point positions that determine the shape functions
P. Steinmann of the finite element approximation. In linear elasticity
Department of Mechanical Engineering,
Applied Mechanics, University of Erlangen–Nürnberg,
it is a proven fact, that the reduction of the potential
Egerlandstr. 5, 91058 Erlangen, Germany energy corresponds to a reduction of the discretization
e-mail: steinmann@ltm.uni-erlangen.de error measured by the energy norm. The concept of
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 117
doi: 10.1007/978-1-4020-6929-1_12, © Springer Science+Business Media B.V. 2007
118 M. Scherer et al.
energy based mesh optimization has been transferred ated with the nodes as discrete configurational forces.
to finite elasticity by several authors, see Askes et al. Mueller et al. (2002, 2004) and Mueller and Maugin
(2004), Kuhl et al. (2004), Mosler and Ortiz (2006), (2002) investigated the application of configurational
Thoutireddy (2003) and Thoutireddy and Ortiz (2004). node point forces in the context of r- and h-adaptivity,
A discrete deformation ϕ h1 is regarded as more accu- fracture mechanics and inhomogeneities. As a solu-
rate than ϕ h2 , if I(ϕ h1 ) < I(ϕ h2 ), where I denote tion strategy for the r-adaptivity, they used a staggered
the potential energy. Numerical experiments justify this steepest descent method.
reasonable approach, but a strict mathematical proof In the context of finite elasticity, energy-based mesh
is not given in the cited literature. Throughout this optimization was investigated by Thoutireddy (2003)
paper, we deal with a pure r-adaptive scheme, i.e. the and Thoutireddy and Ortiz (2004). They denoted their
discrete potential energy is minimized with respect to scheme as a variational arbitrary Lagrangian Eulerian
the material node point positions, whereas the element method (VALE). To solve the underlying minimization
connectivity is fixed. An advantage of the considered problem, they used a staggered cg method and incorpo-
r-adaptivity is that the computation of the gradient of rated connectivity changes of the mesh to improve the
the discrete potential energy with respect to the material performance of their scheme. Mosler and Ortiz (2006)
node point positions is straightforward. focused on the numerical implementation of the VALE
As mentioned before, the concept of energy-based r- method, they proposed a simultaneous solution strat-
adaptivity is well-known. However, the numerical egy based on the regularization of the configurational
implementation represents an ongoing research area in forces. Moreover, they developed a strategy to mini-
computational mechanics, see Askes et al. (2004), Kuhl mize the potential energy by changing the mesh con-
et al. (2004), Mosler and Ortiz (2006) and Thoutired- nectivity. In the two-part publication of Askes et al.
dy and Ortiz (2004). The solution strategy presented (2004) and Kuhl et al. (2004), the considered r-adap-
in this paper follows the concept of Felippa (1976b). tivity is embedded into an ALE formulation for non-
We define energy-based r-adaptivity as a minimization linear elasticity. Therein, the variational formulation
problem with inequality constraints which are intro- is derived from the principle of stationary potential
duced to restrict the maximum distortion of the finite energy. The finite element discretization of the var-
element mesh. As a solution strategy for the constrained iational formulation renders two coupled systems of
problem, we use a classical barrier method. Similar to equations which correspond to the derivatives of the
a penalty method, the barrier method transforms the discrete potential energy with respect to the spatial and
constrained problem into a sequence of unconstrained material node point positions. To solve the resulting
problems. Each problem of the resulting sequence is equations, the authors proposed staggered and simul-
solved by a staggered Newton method that alternately taneous Newton schemes.
shifts the material node points and recovers the state of This paper is structured as follows. Section 2 outlines
equilibrium. the considered class of boundary value problems and
Earlier publications dealing with linear elasticity the variational formulation that is used for the numerical
already present several solution strategies. Carrol and analysis. Section 3 deals with the finite element approach,
Barker (1973), McNeice and Marcal (1973), Felippa i.e. the design of a finite element approximation for the
(1976a, b), Bathe and Sussman (1983) and Sussman spatial deformation and its usage in the framework of a
and Bathe (1985) used staggered schemes, steepest Ritz method. Section 4 is concerned with the main topic:
descent, conjugate gradient (cg) or derivative free meth- Energy-based r-adaptivity. In Sect. 5, we present a
ods. Carpenter and Zendegui (1982) proposed a cg numerical example, a cracked specimen, that illustrates
method that determines the displacements and nodal the performance of the solution strategy proposed in
positions simultaneously. Sect. 4. A discussion in Sect. 6 completes the paper.
Braun (1997) emphasized the relation between con-
figurational mechanics and energy-based r-adaptivity. 2 Boundary value problem and variational
He showed that the gradient of the discrete potential formulation
energy with respect to the material node point positions
can be expressed in terms of the Eshelby stress tensor Mainly to introduce terminology and the notation, we
and denoted the components of the gradient associ- start with a brief reiteration of the considered class
Energy-based r-adaptivity 119
r-adaptive scheme. The material node point positions by Felippa (1976b). We assume that for all admissi-
of the finite element mesh are introduced as additional ble material node point positions X a unique solution
variables in order to further minimize the discrete x∗ of the minimization problem (12) exists. The spatial
potential energy, but the connectivity of the mesh is node point positions x∗ are considered to be an implicit
fixed. For notational convenience, we store the coor- differentiable function of the material node point posi-
dinates of variable material node point positions in a tions
column vector
T x∗ = x∗ (X). (22)
X = [X 1 ], . . . , [X n n X ] . (19)
Substituting Eq. 22 into Eq. 21 we obtain the discrete
Note that the number of variable material node point potential energy in the state of spatial equilibrium
positions n n X is smaller than the total number of nodes.
To ensure that the discrete material domain B0h is an I h∗ (X) = I h (x∗ (X), X). (23)
approximation of the continuous counterpart B0 the
This new function I h∗ depending solely on X is consid-
motion of the nodes is restricted to certain constraints,
ered to be the objective function of energy based mesh
namely: Nodes on the boundaries are allowed to change
optimization. According to our experience, the Hessian
their material positions along the boundary and nodes
of I h∗ is nonsingular, but suffers from negative eigen-
situated on vertices are fixed. For simplicity, positions
values, i.e. I h∗ is also nonconvex. The derivative of
of nodes on Dirichlet boundaries and nodes on Neu-
I h∗ with respect to the material node point positions
mann boundaries with nonvanishing tractions are also
equals the explicit derivative of I h with respect to X
fixed.
g evaluated in the state of equilibrium1
Since each global shape function Ni depends on all
material node points that determine the shape of the ∗ T ∂I h∗ ∂I h ∂x∗ ∂I h
R = = · +
∂x x∗ ,X ∂X ∂X x∗ ,X
g
support of Ni , the approximation of the spatial defor- ∂X
mation depends on the material and spatial node point
T
∂I h
position = = Rx∗ ,X . (24)
∂X x∗ ,X
ϕ h = ϕ h (X, x, X). (20)
The gradient R∗ consists of components Ri∗ that are
Thus, the discrete potential energy can be regarded as associated with the nodes, i.e.
a function of both sets of variables T
R∗ = [R∗1 ], . . . , [R∗n n X ] . (25)
I h (x, X) = I(ϕ h (X, x, X)). (21)
This interpretation was proposed by several authors Since the components Ri∗ can be expressed in terms of
in the case of linear (Carpenter and Zendegui 1982) the Eshelby stress tensor , they are denoted as discrete
and nonlinear elasticity (Braun 1997; Mosler and Ortiz configurational forces or material node point forces, see
2006; Thoutireddy and Ortiz 2004) and suggests a Braun (1997), Mueller et al. (2002) and Steinmann et al.
simultaneous minimization with respect to the material (2001). If body forces and surface tractions are absent,
and spatial node point positions. In general, I h (x, X) is an explicit expression for the material node point forces
nonconvex since the Hessian matrix has negative eigen- reads as
n el
values, see Mosler and Ortiz (2006). This complicates ∗
Ri = A · ∇ X Nie dV. (26)
the numerical treatment of the simultaneous minimiza- e=1 B0e
tion. For instance, if a Newton method it used, it is not
where
guaranteed that the Newton direction is a descent direc-
tion. Another problem arising with the simultaneous = W0 I − F t · P. (27)
approach using a Newton method is that the Hessian
The material residual R∗ can be used to formulate a
matrix of I h can be singular. Moreover, if the simulta-
steepest descent method for minimizing I h∗ . By an
neous minimization of I h fails, then a numerical solu-
iteration step of the form
tion of the variational problem is not at hand, not to
mention a mesh improvement. Xn+1 = Xn − αn R∗ (28)
The staggered solution strategy that is used in this
∂ Ih
paper is based on an interpretation that was proposed 1 In Eq. 24, we exploit that ∂x x∗ ,X =0
122 M. Scherer et al.
the material nodes are shifted in the opposite direction with Lagrange multipliers
of the material node point forces. Thereby, the stepsize
λl ≥ 0, l = 1, . . . , n g . (34)
αn > 0 is chosen such that
The constraints which are not yet specified define a
I h∗ (Xn+1 ) < I h∗ (Xn ). (29)
feasible domain restricting the mobility of the material
Note that this steepest descent method is a staggered node points. The constraints
scheme: After shifting the node points the equilibrium
has to be recovered. The outlined scheme and varia- – have to be defined properly such that an adequate
tions thereof, e.g. cg methods, were proposed in sev- optimization of the mesh is possible, but an overly
eral references as solution strategies for energy-based distortion of the elements is avoided.
mesh optimization in linear (Bathe and Sussman 1983; – have to be adapted to the different element types,
Carroll and Barker 1973; Felippa 1976a; McNeice and since different element geometries require individ-
Marcal 1973; Muller et al. 2002; Sussman and Bathe ual strategies to avoid an inadmissible shape.
1985) and nonlinear elasticity (Mueller and Maugin Concerning the applicability of various algorithms of
2002; Rajagopal et al. 2006; Thoutireddy and Ortiz nonlinear programming, it is preferable that the func-
2004). tions gl characterizing the constraints are twice contin-
Assuming that the minimization of I h∗ with respect uously differentiable.
to the material node point positions is an unconstrained
problem, the first order necessary condition reads as 4.3.1 Inequality constraints
∗
R = 0. (30)
Results of numerical experiments presented in Sect. 5 In the following, we define a set of constraints that
suggest that the existence of a material mesh that ful- is especially designed for linear triangular elements.
fills Eq. 30 is questionable for some problems/meshes, In contrast to Felippa (1976b), our constraints are not
since the mobility of the material nodes is restricted directly based on the Jacobian. We restrict the volume-
by the mesh. Elements with negative Jacobians at the preserving (distortional) deformation of each material
Gauss points are inadmissible and lead to an abortion element. Following a standard finite element procedure
of a finite element computation. Therefore, we explic- the deformation of a triangular element is described
itly take into account the naturally given restrictions of via linear isoparametric shape functions and node point
the finite element mesh by introducing inequality con- coordinates. As a measure for the distortional deforma-
straints which restrict the distortion of each element. tion we use a scalar invariant of the deformation gra-
dient. Since the deformation gradient and the derived
strains are constant, one constraint g e ≤ 0 per ele-
4.3 Minimization problem with constraints ment is sufficient to avoid an inadmissible change of
the shape and, consequently, a negative volume.
A general formulation of the constrained minimization The deformation of each linear triangular element e
problem reads as is described by the map
I h∗ (X) → Min (31) Φ re : B0r
h
→ B0h , (35)
subjected to where B0r
e and B0e
denote the material domains that
gl (X) ≤ 0, l = 1, . . . , n g . (32) the element e occupies in the original material mesh
The first order necessary condition of the constrained and the current adapted material mesh, see Fig. 3. The
problem, the so called Kuhn-Tucker condition, reads deformation map
as ⎡ Φ re = X e ◦ X re,−1 , (36)
ng l
T ⎤
⎢R +
∗ λ ∂g
l
∂X ⎥ is based on two bijective mappings
⎢ l=1 ⎥
⎢ ⎥
n en
⎢ ⎥ X re : Bξe → B0r
e
, X re (ξ ) = Ni (ξ )X r,i , (37)
RKT = ⎢ λ1 g 1 ⎥=0 (33)
⎢ ⎥ i=1
⎢ .. ⎥
⎣ . ⎦
n en
X e : Bξe → B0e , X e (ξ ) = Ni (ξ )X i , (38)
λ g
n g n g
i=1
Energy-based r-adaptivity 123
2. The Newton method that is used to recover the to calculate dnk . Moreover, it can be used to accomplish
spatial equilibrium fails to converge. the predictor step
3. The norm of the gradient of I k in the (n + 1)-th
∗,k ∂x∗
step is higher than in the n-th step. xn+1 ≈ xn, pr = xn +
k k
· Xn+1
k
− Xnk (60)
∂X Xnk
If one of the errors listed above occurs, the k-th iter-
∗,k
ation is stopped, α k is increased, the predictor Xpr
k is which provides a good estimator for xn+1 .
recalculated and the k-th iteration is restarted.
After specifying the fundamental flow of the algo-
rithm, we focus on the computation of the Newton 4.4.3 Sequence α k
direction d. First, we derive the gradient and the Hes-
sian of I k with respect to X. Considering the Eqs. 24 The calculation of the scaling parameter α k is based on
and 48, the gradient reads as α k = ηκ α k−1 , (61)
k T h∗ + T
∂I ∂I k ∂I where 0 < η < 1 and κ > 0. The parameter η is con-
R =
k
= +α
∂X ∂X ∂X stant during the overall numerical procedure, whereas
= R∗ + α k R+ (55) κ is adapted according to the following rules. If the k-th
iteration has to be stopped (since one of the three listed
By differentiating the gradient, we obtain the Hessian
errors occurs), κ is reduced by multiplication with a
∂ 2I k ∂Rk constant factor 0 < cred < 1
=
∂X 2 ∂X κ → cred κ, (62)
∂R ∂x∗ ∂R k ∂R
+
= · + + α . (56)
∂x x∗ ,X ∂X ∂X x∗ ,X ∂X and α k is recalculated until the k-th iteration terminates
successfully. If the k-th iteration succeeds at the first try,
To calculate ∂x∗ /∂X, we differentiate the spatial equi-
κ is increased according to
librium equation (13) with respect to X
∂r ∂x∗ ∂r κ → cinc κ, (63)
· + = 0, (57)
∂x x∗ ,X ∂X ∂X x∗ ,X where cinc > 1 is a constant factor. Then, the new scal-
and obtain ing parameter α k+1 of the next minimization problem
is determined. Note that 0 < ηκ < 1 holds.
∂x∗ ∂r −1 ∂r
= − · . (58)
∂X ∂x x∗ ,X ∂X x∗ ,X Remark 3 A serious problem often arising with the
Explicit expressions for the derivatives of the spatial barrier method is that the first subproblem is difficult
and material residual with respect to the material and to solve numerically. A feature of the constraints pre-
spatial node point positions in Eqs. 58 and 56 are given sented in Sect. 4.3.1 is that the inverse barrier function
in the references (Kuhl et al. 2004; Thoutireddy 2003; has a minimum, if the original and the adapted mesh
Thoutireddy and Ortiz 2004). Finally, we obtain the coincide This is an advantage, since the material node
Newton direction dnk by solving the linear system of point positions of the original are a good estimator for
equations the solution of the first subproblem X0 , if the value of
α 0 is chosen high enough. The barrier function regular-
∂Rk
· dnk = −Rk Xk . (59) izes the problem. If the constraints are based on a ref-
∂X Xnk n
erence triangle as described in Remark 1 of Sect. 4.3.1,
The presented solution strategy is computationally the barrier function seem to have the character of an
expensive, since each step of the proposed Newton energy for mesh smoothing. (For more detailled infor-
scheme involves an additional Newton iteration to mations about energies designed for mesh smoothing
recover the spatial equilibrium. But this additional we refer to Knupp (2000a, b).) Consequently, the mesh
Newton iteration can be accelerated by means of the that minimizes the barrier function does not coincide
matrix ∂x∗ /∂X. In the n-th step of the Newton scheme with the original mesh. Therefore, the set constraints
of the mesh optimization, ∂x∗ /∂X is necessary in order based on the the original mesh is preferred here.
126 M. Scherer et al.
a) b)
see Fig. 5a. Note that the computation does not take
any advantage of the symmetry of the problem. On the
upper and lower boundary of the specimen constant
vertical displacements u = 0.5 are prescribed which
Fig. 4 Cracked specimen corresponds to a 5% elongation. Only the middle node
on the right boundary is fixed in horizontal direction.
The specimen consists of a compressible hyperelastic
4.4.4 Predictor for Xk+1
material with a strain energy density
The predictor Xprk+1 for the solution of the k + 1-th λ 2 µ
W0 = ln (J ) + [I : C − 3] − µ ln(J ) (66)
minimization problem is based on the assumption that 2 2
the solutions Xk are discrete points of a differentiable of neo–Hookean type, where C = F t · F is the right
curve X∗ (α). Given the tangent tk = dX∗ / dα of this Cauchy-Green strain tensor and λ, µ denote the Lamé
curve at the point Xk = X∗ (α k ), the predictor reads as parameters. The material properties are λ = 1.15×104
k+1
Xpr = [α k+1 − α k ] tk + Rk . (64) and µ = 7.69 × 103 which corresponds to a Young’s
modulus E = 2.0 × 104 and a Poisson’s ratio ν = 0.3.
To compute tk , we solve the linear system of equations The parameters of the algorithm are set to: tol1 =
∂Rk 1 × 10−6 , tol2 = 1 × 10−8 , η = 0.5, cred = 0.6, cinc =
· tk = −R+ Xk . (65)
∂X k X
1/cred . With the exception of the meshes depicted in
Fig. 6, the results presented in this section refer to a
set of constraints that is characterized by δ = 0.2.
5 Numerical experiment Concerning the r-adaptive procedure, the motion of
interior nodes is unconstrained, nodes on the upper
Figure 4 illustrates the example, the stretching of a cra- and lower boundary and nodes on vertices are fixed
cked specimen of dimensions B = 10 × H = 20, and nodes on the left and right boundary as well as
L a = 5. The finite element mesh used for the numer- nodes on the edges of the crack are allowed to change
ical analysis consists of 784 elements and 442 nodes, their positions along the boundary. Therefore, only the
Energy-based r-adaptivity 127
a) b) c)
Fig. 6 Cracked specimen—adapted material meshes for varying g e > −5 × 10−5 ) are gray colored. (a) δ = 0.2, I h = 3526,
δ, termination criterion of the barrier method: RKT < 1×10−5 , R∗ = 1.5; (b) δ = 0.1, I h = 3523.9, R∗ = 0.683;
elements with active constraints (numerically characterized by (c) δ = 0.05, I h = 3523.5, R∗ = 0.690
Fig. 8 Cracked
4
specimen—evolution of the x 10
spatial mesh and the Cauchy
stresses σ22 in the vicinity 2.5
of the crack tip
2
1.5
0.5
4
x 10
2.5
1.5
0.5
mesh, the adapted mesh is also symmetric. The evo- u = 0.25 are prescribed on the upper and lower bound-
lution of the tensile stress σ22 and the spatial mesh in ary, a further reduction up to δ = 0.0031 is possible
the vicinity of the crack tip is shown in Fig. 8. The (a further reduction has not been investigated). Once
maximal tensile stress occurring at the crack tip of the more, the r-adaptivity in the vicinity of the crack tip
adapted mesh (k = 27) is σ22 = 2.96 × 104 which seems to be terminated but the number and position of
corresponds to an 800% increase with respect to the elements with active constraints approximately coin-
original mesh. Note that the crack tip is a singularity cides with mesh depicted in Fig. 6c. Badly shaped,
where the stresses tend to infinity. Despite the singu- acicular elements occur having a length-to-height ratio
larity, the presented r-adaptivity stops, since the barrier of approximately 300. Again, the level of the mate-
method converges to a Kuhn-Tucker point. Figure 6 rial residual and the potential energy stagnates. Simi-
shows two additional meshes obtained for δ = 0.1, lar results have been obtained for an example without
δ = 0.05. Note that both meshes represent the final a singularity, a plate with a hole subjected to tension
state of the iteration, whereas RKT < 10−5 . The which is not further discussed in this paper. As men-
evolution of the active constraints (represented by gray tioned in Sect. 4.2, this suggests that the existence
colored elements) indicate that for δ = 0.05 the r-adap- of a material mesh with completely vanishing mate-
tive process in the vicinity of the crack tip is termi- rial node point forces is questionable at least for some
nated but not on the flanks of the crack. The level of problems/meshes. Known exceptions are meshes with a
the discrete potential energy and the norm of mate- very small number of nodes or problems that are exactly
rial residual stagnates. In the present example, the bar- solved by the finite element method. In the latter case
rier method does not converge to a Kuhn-Tucker point the node point positions can usually be chosen arbi-
if δ is further reduced. But if smaller displacement trarily.
Energy-based r-adaptivity 129
Fig. 9 Cracked
specimen—evolution of the
material node point forces
near the crack
Figure 9 illustrates the evolution of the configura- the adapted mesh always occur at the node points of
tional node point forces in the vicinity of the crack the darkest elements which are associated with active
tip. Note that the arrows representing these forces are constraints. Only elements with active constraints pro-
equally scaled in all figures and that material node point vide “reaction forces” that are in equilibrium with the
forces assigned to fixed nodes, e.g. the crack tip node, material node point forces. Material node point forces
are not depicted. The norm of the material residual vec- of nodes that are not associated with active elements
tor of the adapted mesh (k = 27) is R∗ = 1.5 which tend to vanish completely.
corresponds to a 98% reduction with respect to the orig- In Fig. 11 the discrete potential energy I h and the
inal mesh. Since Lagrange multipliers λe > 0 exist, additional energy I + is plotted versus α. The graphs
the material residual does not vanish completely. The show that the discrete potential energy I h decreases
active constraints provide “material reaction forces” and the additional energy I + increases monotonically
such that the sum of the material node point forces as α decreases. The energy obtained for k = 27 is
and the reaction forces vanishes, see Eq. 33. Figure I h = 3526. This corresponds to a relative decrease
10 shows the entire original and adapted material mesh
I h,27 − I h,ref
(k = 27), the material node point forces that are assigned = 79% (68)
to variable material node points and the values of the I h,orig − I h,ref
functions g e characterizing the inequality constraints. where I h,ref is the discrete potential energy of a much
Note that nonvanishing material node point forces of finer so called reference mesh. The reference mesh is
130 M. Scherer et al.
Fig. 10 Cracked a) b)
specimen—material forces
and values of the functions
g e . (a) Original mesh; (b)
adapted mesh, k = 27,
α = 1.5 × 10−12 −0.05 −0.05
−0.1 −0.1
−0.15 −0.15
−0.2 −0.2
−0.25 −0.25
refined near the crack tip of the specimen and consists point positions. The inequality constraints are intro-
of 9,648 triangular elements and 4,915 nodes. In the duced to restrict the distortion of the material elements
left diagram of Fig. 12 the distance during the r-adaptive process. We have presented a set
of constraints designed for linear triangular elements
ϕ k − ϕ ref 2 =
h h
(ϕ kh − ϕ ref
h )2 dV that restrict the distortional deformation of the ele-
B0 ments. As a solution strategy for the constrained prob-
≈ L -error
2
(69) lem, we use a classical barrier method with an inverse
between the spatial deformation of the adapted mesh ϕ kh barrier function.
and the spatial deformation of the reference mesh ϕ ref
h is Due to the complexity of the underlying minimi-
plotted versus α. The distance between the two functions zation problem, the scheme is computationally expen-
measured by the L 2 -norm is an approximation for the sive, but the results of the presented numerical example
L 2 -error, i.e the distance between the numerical and the are promising. The proposed barrier method is suit-
exact solution. For k = 27, the approximated L 2 -error able to optimize meshes with a high number of nodes.
h − ϕ h = 5.5 × 10−2 which corresponds to
is ϕ 27 Moreover, an appreciable improvement of the finite ele-
ref 2
a 71% reduction with respect to the original mesh. Note ment solution has been achieved. Unfortunately, the
that within k = 7 steps a 68% reduction is achieved. barrier method does not always succeed to get as close
to a Kuhn-Tucker point as in the presented example.
Below a given point the scaling parameter of the bar-
6 Conclusions rier method cannot be further reduced. Therefore, the
performance of the Newton method that is used to solve
This paper describes a solution strategy for energy- the unconstrained subproblems of the barrier method
based r-adaptivity in finite elasticity. Energy-based should be improved. For instance, by using special
r-adaptivity is defined as a problem of nonlinear pro- line search schemes that take advantage of the known
gramming, a minimization problem with inequality form of the singularity in a barrier function. Due to the
constraints. The objective function of the problem is nonconvexity of the discrete potential energy (I h∗ ),
the discrete potential energy in the state of equilibrium it is reasonable to apply a modified Newton method
which is considered to be a function of the material node that works with a modified, positive definite Hessian
Energy-based r-adaptivity 131
3620
3610 0.15
3600
3590
≈ L2-error
3580
Ih
3570
0.1
3560
3550
3540
3530
∂X ∂ X1 ∂ XnX
1 where
⎧
∂g e ⎨γ −2 ∂γ · ∇ X N e if X i ∈ ∂B0e
0.5
= ∂ f re r r,i
, (71)
∂ Xi ⎩
0 else
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 with
α
∂γ e −t 2
Fig. 11 Cracked specimen—discrete potential energy I h and e = −γ f r + e f re . (72)
∂ fr jr
additional energy I + versus α
And the second derivative is given by
⎡
⎤
∂ 2 ge ∂ 2 ge
matrix. Moreover, we want to investigate, if alterna- ⎢ ∂ X 1∂ X 1 · · · ∂ X nn X ∂ X 1 ⎥
tive algorithms of nonlinear programming, e.g. primal- ⎢ ⎥
⎢ ⎥
dual interior point method, penalty barrier methods or ∂ g
2 e ⎢ . .. . ⎥
⎢ . . . ⎥
penalty methods, are more suitable to solve constrained =⎢ . . ⎥ (73)
∂X 2 ⎢ ⎥
minimization problem that is associated with the pre- ⎢ ⎥
⎢ ⎥
sented r-adaptivity. To enhance the applicability of the ⎣ ∂ 2 ge ∂ 2 ge ⎦
∂X ∂X
··· ∂X ∂X
method, we want to develop constraints for other types 1 nn X nn X nn X
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 133
doi: 10.1007/978-1-4020-6929-1_13, © Springer Science+Business Media B.V. 2007
134 D. Materna, F.-J. Barthold
an overview. In these disciplines we consider varia- material residual problem in one step. In Sect. 4,
tions of the material configuration and we are interested sensitivity relations for the physical and material prob-
in the change of the state variables and the objective lem are investigated and we derive explicit formula-
functional due to these variations. The variations are tions for the variations of the physical and material
required in order to solve the corresponding Lagrang- residuals. The finite element approximation of the sen-
ian equation using nonlinear programming algorithms. sitivity relations is discussed in Sect. 5. Finally, we
In many engineering applications, the energy func- propose different solution algorithms for the optimiza-
tional of the problem is used as the objective functional tion problem and summarize the theoretical as well as
of the optimization problem. The relations of configu- computational treatment by means of selected exam-
rational mechanics and its application to mesh opti- ples from mesh optimization and shape optimization.
mization can be obtained by using techniques from
variational design sensitivity analysis applied to the
energy functional (Barthold 2003, 2005; Materna and 2 Preliminaries
Barthold 2006a, b, 2007a). The energy depends on
the state function and the design function. The design This section is concerned with some settings and defini-
function specifies the shape and/or the topology of the tions. We consider an open bounded material body with
material body. The first variation of the energy func- an undeformed reference placement R ⊂ E3 with a
tional with respect to changes in the design leads to the piecewise smooth, polyhedral and Lipschitz-continu-
material residual or weak form of the material or con- ous boundary = ∂ R such that = D ∪ N and
figurational force equilibrium. In the context of sen- D ∩ N = ∅, where D denotes the Dirichlet bound-
sitivity analysis, we can interpret the configurational ary and N the Neumann boundary, respectively. The
forces as the sensitivity of the energy with respect to corresponding deformed current placement is denoted
variations in the design. The classical physical problem by t ⊂ E3 . The deformation of the material body
and the material problem are coupled by the pseudo from R into t is given by the nonlinear mapping
load operator, which is used for the solution of struc- ϕ : R → t , x = ϕ(X). Here, ϕ maps the mate-
tural optimization problems. A simple application is rial particle X from the reference placement R to the
the above mentioned mesh optimization problem. For spatial coordinates x in the deformed placement t for
mesh optimization, we choose simply the nodal coordi- any fixed time t ∈ It . The corresponding deformation
nates as design variables. The optimal nodal positions gradient, i.e. the tangent map of ϕ from the material
(the design) of a given mesh (the reference configu- tangent space TX R to the spatial tangent space Tx t ,
ration) can be computed using standard optimization as well as its Jacobian J are given by F = ∇ X ϕ :
algorithms. Furthermore, the material residual is TX R → Tx t and J = det F. Furthermore, we
directly related to the J -integral or energy release rate assume that the deformation is injective, sufficiently
and we can perform a sensitivity analysis for these smooth and that J > 0, such that there exist the inverse
quantities. deformation mapping φ : t → R , X = φ(x). The
The minimization of the energy is directly related corresponding deformation gradient f and its Jacobian
to the minimization of the compliance of the system or j are given by f = ∇x φ and j = det f .
equivalently to the maximization of the stiffness and is The above introduced kinematical settings in the
used as objective in shape and topology optimization, referential placement R and the current placement
see e.g. Dems and Mróz (1978) and Bendsøe and Sig- t could be enhanced by using the intrinsic formula-
mund (2003). In this context, the configurational forces tion by Noll (1972). This is based on the concept of a
on the design boundary are indicators in which direc- differentiable manifold, see for instance Truesdell and
tion the boundary has to move in order to minimize the Noll (2004), Bertram (1989) and Marsden and Hughes
compliance or to maximize the structural stiffness. (1994) for details. An enhancement of the above men-
The paper is organized as follows. After some pre- tioned approach by Noll in the context of variational
liminaries about the notation, we begin with an abstract design sensitivity analysis was proposed by Barthold
setting of a classical structural optimization problem and Stein (1996) and Barthold (2002, 2007). Following
and formulate an energy minimization problem. We the intrinsic concept, a given manifold can be described
obtain the classic physical residual problem and the locally using an intrinsic coordinate system defined on
Variational design sensitivity analysis 135
an independent continuous parameter space P with Therefore, the material particle X as well as the spatial
local coordinates . Without going into detail, this particle x depend on the design variable, i.e. X = X(s)
leads to two fundamental mappings, a design depen- and x = x(X(s)). The explicit coupling between X and
dent local reference placement mapping κ : P × Is → s, i.e. the explicit structure of f (s) depend on the par-
R , X = κ(, s) and a time dependent local current ticular problem. The local current placement mapping
placement mapping µ : P × It → t , x = µ(, t) x = µ(, t) can be expressed by the local displace-
for any fixed time t ∈ It and any design s ∈ Is 1 . ment mapping u = υ(, s, t). With these, a quantity
The corresponding tangent maps and its Jacobians are (·), which depends on the state variable u and the design
given by K = ∇ κ : T P → TX R and JK = variable s is denoted by (·)(u, s).
det K as well as M = ∇ µ : T P → Tx t and
J M = det M, respectively. With these mappings, the The variation of a quantity (·)(ϕ; X, t) with respect to x
deformation map ϕ and its tangent map can be written at fixed X and t is denoted by δx (·) =: (·) x and the vari-
in the form ϕ = µ ◦ κ −1 and F = ∇ X ϕ = M K −1 ation of (·)(φ; x, t) with respect to X at fixed x and t by
and for the inverse mapping φ follow φ = κ ◦ µ−1 δ X (·) =: (·) X . In order to avoid confusion between the
and f = ∇x φ = K M −1 , respectively. The difference small x and the capital X and due to the fact that the ref-
vector between the reference and current placements is erence configuration of the material body is prescribed
the displacement u = x − X. The corresponding local by a design function s, we prefer to use δu (·) instead of
displacement mapping υ written in terms of the local δx (·) and δs (·) instead of δ X (·), respectively. Further-
mappings κ and µ is given with u = υ(, s, t) = more, the short notation (·) denote the total variation
µ(, t) − κ(, s). of a quantity (·) and (·) u as well as (·) s the partial var-
Different gradient operators can be defined, i.e. iation with respect to u and s, i.e. (·) u := (·) x = δx (·)
grad := ∇x , Grad := ∇ X and GRAD := ∇ corre- and (·) s := (·) X = δ X (·). With this notation, the total
sponding to the variables x, X and of the considered variation of a quantity (·)(u, s), which depends on the
domains R , t and P , respectively. Overall, K , M deformation and the design, is given by the partial var-
and F are used to perform pull back and push forward iation with respect to u and a fixed design ŝ as well
transformations between current placement, reference as the partial variation with respect to s and a fixed
placement and parameter space. deformation û, i.e.
For problems with changes in the material configura-
tion, e.g. shape optimization or problems from config- (·) (u, s) = (·) u (u, ŝ) + (·) s (û, s) (1)
urational mechanics, the intrinsic formulation in local In the same way we define the second variations (·) uu
coordinates has advantages because we can deal with := δx2x (·) and (·) ss := δ 2X X (·) as well as the mixed
two independent placement mappings, see Barthold variations (·) us := δx2 X (·) and (·) su := δ 2X x (·), respec-
and Stein(1996) and Barthold (2002, 2007) for more tively.
details.
The partial variation of J with respect to u leads to the The contributions of the physical residual R(u, s; η)
physical residual (5) R : V → R in terms of the reference configuration are given by
R(u, s; η) := a(u, s; η) − F(s; η). (12) a(u, s; η) = P : Grad η d (18)
R
In the same manner, variation with respect to changes
in the design s leads to the material residual G : D → R F(s; η) = b R · η d (19)
in the form R
where P is the first Piola-Kirchhoff stress tensor and
G(u, s; ψ) := b(u, s; ψ) − L(s; ψ). (13)
b R = −∂x V R are the body forces per unit volume in
The semilinear forms a : V ×V → R and b : D×D → the reference configuration derived from the potential
R contain the parts of the partial variations with respect VR .
to deformation gradients, i.e. For the material residual G(u, s; ψ) follows after a
pull back to the reference configuration
d
a(u, s; η) := U (; µ, κ, M
dε P b(u, s; ψ) = (
+ V R 1) : Grad ψ d (20)
R
+ε ∇ η̃, K ) d
ε=0
L(s; ψ) = R · ψ d
binh (21)
d
R
= U R (X; ϕ, F + ε ∇η) d
dε R ε=0 where
= W R 1 − F T P is the well-known energy
(14) momentum or Eshelby tensor (Eshelby 1951, 1975).
R = −∂ X W R |ex pl. denotes the inho-
Furthermore, binh
d mogeneity force given by the explicit derivative of W R
b(u, s; ψ) := U (; µ, κ, M, K
dε P with respect to X.
+ε ∇ ψ̃) d
= Ut (x; φ, f + ε ∇ψ) d
d
weak form can be obtained by a re-parametrization of
= U R (X; ϕ + ε η, F) d
(16)
dε R ε=0 the direct deformation problem in terms of the inverse
deformation φ (Govindjee and Mihalic 1996). For a
d
detailed discussion about the direct and inverse defor-
L(s; ψ) := U (; µ, κ + ε ψ̃, M, K ) d
dε P ε=0 mation problem and its duality see for instance (Shield
d
1967; Chadwick 1975; Steinmann 2000; Kuhl et al.
= Ut (x; φ + ε ψ, f ) d
Remark 5 The semilinear forms a(u, s; η) and constraints due to variations in the design. In the con-
b(u, s; ψ) could be expressed in terms of the symmet- text of structural optimization this is termed as design
ric second Piola-Kirchhoff
stress tensor S in the form sensitivity analysis.
The total partial derivative Ds f of a function f (s, u)
a(u, s; η) = S : E u (u, η) d (22)
R = f (s, u(s)) with respect to s and the explicit partial
derivative ∂s f are connected by the relation
b(u, s; ψ) = S : E s (u, ψ) d ∂ f ∂u
R Ds f (s, u) = ∂s f + . (26)
∂u ∂s
+ (W R + V R )1 : Grad ψ d (23) With this, we introduce the total partial variation of a
R functional J (u, s) = J (u(s), s) with respect to s in
where the relations the form
∂J ∂ u
P : Grad η = F S : Grad η Ds J (u, s) · δs = ∂s J · δs + · δs. (27)
∂u ∂s
= S : sym{F T Grad η} = S : E u (u, η) The above sensitivity relation depends on the partial
(24) variation ∂s J · δs = Js (u, s, δs) and a second part
and with the derivatives ∂u J and the changes in the state
∂s u, which is a result of the functional dependencies.
− F T P : Grad ψ = −F T F S : Grad ψ The derivatives ∂s J and ∂u J are given explicitly, but
= −S : sym{F T F Grad ψ} the derivative of the deformation ∂s u with respect to s is
= S : E s (u, ψ) (25) given implicitly, because they are the results of an anal-
ysis for the deformation, see Sect. 4.2. This is a clas-
have been used. The terms E u (u, ·) and E s (u, ·) denote
sical well-known expression in structural optimization
the partial variations of the Green-Lagrange strain ten-
for the design sensitivity of the objective functional.
sor E with respect to u and s, respectively, see Appen-
In the case of J (u, s) = E(u, s), i.e. we choose the
dix A for the explicit form of these quantities. The terms
energy as objective functional, we have
a(·, ·) and b(·, ·) have the same structure with different
variations of the Green-Lagrange strain tensor E, but Ju = E u = ∂u J · δu = R(u, s; δu) = 0 (28)
and Js = ∂s J · δs = G(u, s, δs). With this, the sec-
b(·, ·) has an additional term as a result of the variation
of the domain R . We will see in the following that it is ond part of the above sensitivity relation vanishes and
very useful to use this expressions because we utilize it remains only the material residual, i.e.
some symmetry properties and get similar structures Ds J (u(s); s) · δs = Js (u, s; δs) = G(u, s; δs). (29)
for the variations of the physical and material prob- Therefore, in the context of structural optimization, we
lem. Furthermore, for the computational treatment it is can interpret the material residual or configurational
helpful to deal with symmetric quantities. forces as the sensitivity of the energy with respect to
Remark 6 In this paper we consider a pure variational variations in the design, see Remark 4.
setting for elasticity. The theory of configurational
forces is not restricted to elastic problems and varia- 4.2 Sensitivity of the physical residual
tional approaches. Inelastic problems play an impor-
tant role in many fields, e.g. elastic-plastic fracture The solution of the Lagrangian requires the partial vari-
mechanics. For an overview of different applications ations of the physical residual with respect to u and s,
and approaches see e.g. Steinmann and Maugin (2005) respectively. The total variation of the physical residual
and the references therein. R = a(u, s; η) − F(s; η) = 0 reads
R = R u (u, s; η, δu) + R s (u, s; η, δs) = 0 (30)
4 Variational design sensitivity analysis where the partial variations are given by
R u (u, s; η, δu) := au (u, s; η, δu) − Fu (s; η, δu)
4.1 Sensitivity of the objective functional
(31)
The solution of structural optimization problems R s (u, s; η, δs) := as (u, s; η, δs) − Fs (s; η, δs).
require the variations of the objective functional and the (32)
Variational design sensitivity analysis 139
call t (·; ·) the tangent pseudo or fictitious load oper- Finally, using Eq. 23, we obtain the material pseudo
ator for the material problem, compare with Eq. 34. load operator (44) and the tangent material stiffness
Both operators are bilinear forms d : D × D → R and (43) in the form
t : D × V → R.
As a result of the permutableness of variations, i.e. t (u, s; ψ, δu) = S : E su (u, ψ, δu) d
R
Gu = Jsu
= Jus = R s (45)
+ E s (u, ψ) : C : E u (u, δu) d
and the symmetry of the bilinear forms, i.e. as (u, s; ·, η) R
= as (u, s; η, ·), we obtain for the variation of G with + S : E u (u, δu) Div ψ d
respect to u R
Gu (u, s; ·, δu) = R s (u, s; δu, ·) = p(u, s; δu, ·) − b R · δu Div ψ d. (51)
R
= as (u, s; δu, ·) − Fs (s; δu, ·). (46)
and
Thus, due to symmetry, the partial variation Gu leads to
the tangent pseudo load operator of the physical prob- d(u, s; ψ, δs) = S : E ss (u, ψ, δs) d
R
lem (34), i.e.
+ E s (u, ψ) : C : E s (u, δs) d
t (u, s; ψ, δu) = p(u, s; δu, ψ). (47) R
Therefore, we have additional to specify only the mate- + S : E s (u, ψ) Div δs d
rial tangent operator d(u, s; ψ, δs) = Gs .
R
With these notations the total variation yields the + S : E s (u, δs) Div ψ d
form R
G = p(u, s; δu, ψ) + d(u, s; ψ, δs) = 0. (48) + W R Div ψ Div δs d
R
After rearranging the above terms we can formulate the
following sensitivity equation for the material problem. − W R 1 : Grad ψ Grad δs d.
R
Problem 5 Let δ û ∈ V be a given fixed variation in (52)
the state. Find δs ∈ D such that Here, E ss (u, ·, ·) denote the second variation of E with
d(u, s; ψ, δs) = −Q(u, s; δ û, ψ) ∀ ψ ∈ D, (49) respect to s. See Appendices A and B.2 for the explicit
formulations of the above terms. As a result of symme-
where try follow t (u, s; ψ, δu) = p(u, s; δu, ψ), see (39).
Furthermore, d(·, ·) has a similar structure like k(·, ·),
Q(u, s; δ û, ψ) := p(u, s; δ û, ψ) = Gu (u, s; ψ, δ û)
but additional terms occur as a result of the second var-
(50) iation of R , see (38).
is the pseudo load of the material problem for the var-
iation δ û.
5 The finite element approximation
This is a variational equation for the sensitivity of the
design due to changes in the deformation. For a given 5.1 The discrete optimization problem
variation in the state δ û, we can calculate the variation
in the design δs. The finite element formulation is based on a conform-
ing Galerkin method defined on meshes Th = {K } with
Remark 8 It is interesting to note, that due to symmetry a mesh parameter h consisting of closed cells K which
both the sensitivity of the deformation and the sensi- are either triangles or quadrilaterals. The boundary ∂ K
tivity of the design depend on the pseudo load opera- of each element K is assumed to be Lipschitz-con-
tor p(·, ·). Therefore, this operator plays an important tinuous. On the mesh Th we define finite dimensional
role for the solution of the minimization problem for element spaces V h ⊂ V and D h ⊂ D consisting of
u and s. cellwise polynomial functions.
Variational design sensitivity analysis 141
The corresponding discrete state uh ∈ V h ⊂ V and The optimality condition (11) for the material prob-
discrete design s h ∈ D h ⊂ D are determined by the lem holds only for stationary points u∗h and s∗h of the
following discrete version of Problem 2. Lagrangian, i.e. L s = G(u∗h , s∗h ; ψ h ) = 0.
This means, that the material residual is not fulfilled
Problem 6 Find {uh , s h } ∈ V h × D h of the objective for every uh
= u∗h = uh (s∗h ), i.e.
function J : V h × D h → R such that
G(uh , s h ; ψ h )
= 0 ∀ uh
= u∗h = uh (s∗h ) ∈ V h .
J (uh , s h ) = E(uh , s h ) → min (53) (58)
{uh , sh }∈V h ×D h
Hence, we obtain a material residual as a result of the
subject to the constraint
non-optimal solution uh , which is a result of the non-
< A(uh , s h ), ηh > = R(uh , s h ; ηh ) = 0 optimal design s h in the sense of the minimization of
∀ ηh ∈ V h . (54) Problem 6. With the definition of the errors (56) we
have s∗h = s h + es,h and therefore
For the optimal solutions u∗h ∈ V h and s∗h ∈ D h in the G(u∗h , s∗h ; ψ h ) = G(uh (s h +es,h ), s h +es,h ; ψ h ) = 0.
chosen approximation spaces V h and D h we require
(59)
that
From this, a suitable approximation for the error could
J (u∗h , s∗h ) ≤ J (uh , s h ) ∀ uh , s h ∈ V h × D h . (55) be obtained from the linearization
The error of the state eu,h and the design es,h in the G(ûh , ŝ h ; ψ h ) + Ds G(ûh , ŝ h ; ψ h ) · es,h + O = 0.
chosen approximation spaces are introduced by
(60)
eu,h := u∗h − uh and es,h := s∗h − sh , (56) Finally, with the total partial derivative
respectively. ∂G ∂G ∂ u
Ds G(u(s), s) = + (61)
For a matrix description of the derived sensitivity ∂s ∂u ∂s
operators we introduce the discrete approximations for follow the tangent operator of the material problem
displacements and design, i.e. the nodal displacement
bT (ûh , ŝ h ; ψ h , es,h )
vector u ∈ Rn and the vector of design variables s ∈
Rm . Here, n and m are the dimensions of the introduced = Ds G(ûh , ŝ h ; ψ h ) · es,h
= G(ûh , s h + εes,h ; ψ h )
. (62)
dε ε=0
gent operators lead to the usual tangent stiffness matrix
We obtain an equation for the error in the design or
K ∈ Rn×n , the tangent pseudo load operator matrix
rather for the error in the material residual in the form
P ∈ Rn×m as well as the tangent material stiffness
matrix D ∈ Rm×m associated to the bilinear forms bT (ûh , ŝ h ; ψ h , es,h ) = −G(ûh , ŝ h ; ψ h ) ∀ ψ h ∈ D h .
k(·, ·), p(·, ·) and d(·, ·). Furthermore, R ∈ Rn denotes (63)
the physical residual vector and G ∈ Rm denotes the
This is used for a staggered solution algorithm, see Sect.
material residual vector associated to the functionals
6.2. Furthermore, it could be used for a mesh optimi-
R(uh , s h ; ·) and G(uh , s h ; ·), respectively.
zation algorithm, see Sect. 7.1.
5.2 The error in the material residual 5.3 The discrete sensitivity equations
The fulfillment of the physical residual for every admis- With the above definitions, the discrete versions of the
sible design is an important constraint of the optimiza- sensitivity equations for the physical (35) and material
tion problem, i.e. (48) residuals become
R(uh , s h ; ηh ) = 0 ∀ sh ∈ Dh . (57) K δu + Pδs = 0 or δu = −K −1 Pδs (64)
142 D. Materna, F.-J. Barthold
If we substitute the first equation of (79) into the second R, we have to solve G(û, s; ψ) = 0 within a Newton
and vice versa, we can eliminate u or s by using the algorithm. This requires the total partial derivative (26)
Schur complement in order to solve the problem one of the functional G(u, s) = G(u(s), s) with respect to
after the other and to minimize the size of the system. s, which is given by Eq. 61. With this, the linearization
We obtain a formulation for the state variables of G in the direction s yields
[ K − P D−1 P T ]u = P D−1 G − R (80) G(û, ŝ, ψ) + Ds G(û, ŝ, ψ) · s + O = 0 (82)
or for the design variables with the tangent operator
−1
[D− P K T
P ]s = PT K −1 R − G. (81)
Ds G(u, s, ψ) · s = bT (u(s), s; ψ, s)
This requires well-conditioned matrices K and D in d
order to compute the corresponding inverse matrices = G(u, s + εs; ψ)
dε
+G(u(s + εs), s; ψ)
. (83)
Remark 11 It seems, that the naturally best way to find ε=0
a solution of the optimization problem (9) is a full The remainder O can usually be neglected and we
Newton method, i.e. the simultaneous solution of the obtain a linear equation for s in the form
physical and material problem. We expect the best con-
vergence speed and hence, the lowest computational bT (û, ŝ; ψ, s) = −G(û, ŝ, ψ) ∀ ψ ∈ D. (84)
cost. This optimization problem is non-convex in gen-
The finite element approximation reads
eral and for real problems with a large number of design
variables such an algorithm is very sensitive and not bT (ûh , ŝ h ; ψ h , s h ) = −G(ûh , ŝ h , ψ h ) ∀ ψ h ∈ D h .
stable, see also Sect. 7.1. Furthermore, the system (79) (85)
has a typical saddle point structure and we have to be
careful by solving this system. Often a preconditioner This is equivalent to Eq. 63, in which we have replaced
is needed in order to obtain a well-conditioned system the error in the design es,h by the design increment
matrix. See for instance Benzi and Golub (2004) and the s h .
references therein for preconditioners for this type of
equations. Therefore, other solution algorithm should Remark 12 In Sect. 5.2 we have mentioned, that the
be considered, see e.g. Nocedal and Wright (1999) for material residual is not fulfilled for every uh
= u∗h =
algorithms used in classical structural optimization. uh (s∗h ), i.e. G(uh , s h ; ψ h )
= 0, ∀ uh
= u∗h ∈ Vh and
hence, for a given solution uh of the physical problem,
we have to solve the material problem with a Newton
6.2 Staggered solution method algorithm in order to find a s∗h such that uh (s∗h ) = u∗h
and hence G(u∗h , s∗h ; ψ h ) = 0.
Alternatively, we propose a staggered solution method,
i.e. we solve the physical problem with a fixed design ŝ In the discrete case follows from Eq. 44 ∂u G = P T
and thereafter the material problem with a fixed defor- and with Eq. 64 ∂s u = −K −1 P and hence a matrix
mation û. For a given solution û of the physical residual description
144 D. Materna, F.-J. Barthold
∂G ∂G ∂u 6.3 Steepest descent method
bT (ûh , ŝ h ; ψ h , s h ) = ψ T + s
∂s ∂u ∂s
= ψT D − P T K −1 P s. The simplest way to solve the optimization problem is
the well-known steepest descent method. We rewrite
(86) the objective function to J (s) = J (u(s), s).
Finally, we have to solve the discrete version of (84), For a given starting point s0 and u(s0 ) we calculate
i.e. a search direction v := −Js and perform a line search
with ε > 0 such that J (s + εv) < J (s). This leads to
D − P T K −1 P s = −G (87) the solution algorithm
in every Newton step. With the definition of the sen- si+1 = si + εi v i = si − εi Js (u, si ). (91)
sitivity operator S p of the physical problem (68) this
If we choose the energy of the system (8) as the objec-
relation could be expressed as
tive functional, i.e. J = E, we have in the discrete
D + P T S p s = −G. (88) case
si+1 = si + s = si − εi G(uh , s h,i ) (92)
Remark 13 Additional we have to fulfil in every step
the constraint A(uh , s h ; ηh ) = 0, which means that where G(uh , s h,i ) ∈ R is the material residual vector
m
we have to solve the equilibrium equation for the associated to the functional G(uh , s h ; ·) at the current
primal problem. Moreover, the computation of the state uh (s h,i ). This is a pure gradient based method
accurate material residual G(ûh , ŝ h ; ψ h ) for the next and it is not necessary to calculate the second deriva-
iteration step requires the solution of the state variable tives of the objective functional, but a large number of
uh for the updated current design, i.e. we have to solve iterations and functional evaluations are required.
R(uh , s h ; ηh ) = 0. For a simple linear physical prob- In order to update the state variables uh during the
lem, only the solution of one linear equation is required. iteration, we can use the same procedure as mentioned
In the general nonlinear case, we have to perform some in Remark 13.
Newton iterations for the physical problem in order to Remark 15 The steepest descent directions at two con-
find the new state of equilibrium for the current design, secutive steps are orthogonal to each other if the line
i.e. we have to solve the tangent version of the primal search is based on the minimization problem
problem
J (si + ε v i ) = J (si − ε Js (si )) → min (93)
ε
k(ûh , ŝ h ; ηh , uh ) = −R(ûh , ŝ h ; ηh ) (89)
in order to find a step size parameter εi . Furthermore,
in every Newton step. The best performance of the algo- often the steepest descent exhibit ‘zigzag’ behavior and
rithm could be achieved, if we transfer beforehand the the solution oscillates, see e.g. Nocedal and Wright
data of the state uh to the new design. To do this, we (1999) for details. This tends to slow down steepest
can use the sensitivity relation for the physical problem descent method although it is convergent and hence
(68) and obtain with a sufficient small s an approxi- a large number of iterations are required in order to
mation for the change in the state from find the solution. A simple and efficient improvement
is the conjugate gradient method. This method is also
u = −K −1 P s = S p s. (90)
based only on gradient information of the objective
After this, the system is usually still unbalanced, but functional. The new design is given by si+1 = si +εi v i
only few Newton iterations are required in order to find with the conjugate direction
the new state of equilibrium.
v i := − G(si ) − βi v i−1 . (94)
Remark 14 In a classical staggered solution algorithm, The scale factor βi is determined for instance by the
the coupling terms P in (79) are usually simply Fletcher-Reeves or Polak-Ribiere method, see Table 2.
neglected. As a result of the functional dependencies The conjugate gradient method should be preferred
G(u, s) = G(u(s), s) of the material residual, we have over the steepest descent method, because the rate of
nevertheless the coupling terms in the tangent operator convergence could be improved without appreciable
(83) for the solution of the material motion problem. cost.
Variational design sensitivity analysis 145
Table 2 Summary of
Algorithm Discrete formulation
discrete solution algorithms
K P u R
Full Newton =−
T
P D s G
Staggered Newton D − P T K −1 P s = −G
G(si )T G(si )
βi = (Fletcher-Reeves method)
G(si−1 )T G(si−1 )
G(si )T G(si ) − G(si−1 )
βi = (Polak-Ribiere method)
G(si−1 )T G(si−1 )
In Table 2 we have summarized the discrete solution Remark 16 The mesh optimization problem can be
algorithms for the optimization problem (53). These interpreted as an inverse problem. It requires the solu-
algorithms could be used in this pure form or could be tion of the material residual problem (96) or the cou-
enhanced by different modifications, see for instance pled system (95). Different numerical difficulties arise
Nocedal and Wright (1999). due to the solution of this problem. In many cases,
the Hessian matrix of the system is ill-conditioned and
becomes singular or close to singular during the itera-
7 Applications tions and therefore the Newton algorithm is not stable
and fails. The optimization problem is non-convex in
7.1 Mesh optimization general and hence, the solution needs not to be unique.
Therefore, the problem could be termed ill-posed in the
An application of the proposed framework is the mesh sense of Hadamard and reasonable regularization meth-
optimization problem. In the case of mesh optimization ods should be used in order to regularize the problem,
the nodal coordinates X are chosen as design variables, see for instance Engl et al. (2000) and the references
i.e. s = X. All inner nodes or a subset of nodes are cho- therein. The numerical difficulties are also mentioned
sen to build up the design space D h . Additional we can by Askes et al. (2004) and Mosler and Ortiz (2006).
include nodal coordinates in tangential direction on the The authors have proposed different strategies in order
boundary. Finally, it follows a r -adaption algorithm for to overcome these problems.
mesh optimization.
The optimal nodal positions (the design) of a given Remark 17 There exist some nodal coordinates (design
mesh (the reference configuration) can be computed by variables), which do not contribute to a change in the
using the solution algorithms which are listed in Table energy during the mesh optimization process, even
2. For the full Newton method, Eq. 79, follows though the corresponding material residual is not zero.
Hence, these nodal coordinates should not be consid-
K P u R
=− . (95) ered. The sensitivity of the material residual
PT D X G
The proposed staggered solution algorithm (87) yields δG = D + P T S p δ X (98)
the
form could be used in order to find out the most relevant
D − P T K −1 P X = −G (96) nodal coordinates (design variables).
and the steepest descent method (92) for mesh optimi-
The regularization and the identification of the most
zation follows in the form
relevant design variables are discussed in more detail
X i+1 = X i − ε G(uh , X i ). (97) in Materna and Barthold (2007a, b).
146 D. Materna, F.-J. Barthold
structure. The minimization of the compliance is equiv- modeled using Bézier curves each with three internal
alent to the maximization of the stiffness. Let control points. We consider only the lower and right
boundary of the L-shape as design boundary and keep
C := W R d (100)
R the loaded boundary fixed. Including the corner points,
be the strain energy of the structure. We are looking overall 14 coordinates of the control points are design
for the optimal design for which the structure attains variables si . The L-shape is modeled with the plane
a minimum of internal elastic energy among the struc- strain condition with E = 105 and ν = 0.3 and loaded
tures of constant volume V0 or material cost. We use by line loads tx = t y = 5.
a staggered solution algorithm and we formulate the The optimum was attained within 37 iterations. The
optimization problem only in terms of the design. This initial and final solutions as well as a selected itera-
ends in the following problem. tion are shown in Fig. 4. The optimal shape leads to a
uniform distribution of the material residual G(X) on
Problem 7 Find s ∈ D of the objective functional J : the design boundary. The norm of the material residual
D → R such that the strain energy G(s) on the design variables decreases from 0.4009 to
J (s) = J (u(s), s) = C → min (101) 0.1911, i.e. a reduction of 52 %. The material residual
s∈D
on the boundary does not vanish, because there is still
subject to the constraints an ambition to find a state with lower internal energy,
R(u, s; η) = 0 (102) i.e. a more stiffer structure. The side constraints for the
design variables and the volume constraint avoid this
V − V0 = 0. (103)
movement.
The design variables s2 and s3 connect two geome-
Example: L-shape
try patches, respectively. Therefore, they have a large
influence domain, i.e. small variations result in large
We consider as an example from shape optimization
changes in the geometry and hence in the internal
a L-shaped cantilever problem, see Fig. 3. The opti-
energy. This is indicated by large material residuals
mization task is to generate the most efficient material
G(si ) on these design variables, see Fig. 4.
distribution with respect to the overall stiffness of the
The internal energy decreases from 0.9015 to 0.5409,
structure. The nonlinear programming problem under
i.e. a reduction of 40 %. The distribution of the internal
consideration consists of the objective function (inter-
energy for the initial and optimized design is shown in
nal elastic energy), the constraint function (constant
Fig. 5. The energy distribution of the optimized shape
volume) and the geometrical design variables, i.e. we
is more smooth in comparison with the initial shape.
solve Problem 7.
The L-shape consists of three geometry patches as Remark 18 It is interesting to note, that the material
indicated in Fig. 3. The boundaries of the patches are residual G on the design boundary is the indicator in
148 D. Materna, F.-J. Barthold
(a)
(b)
(c)
Variational design sensitivity analysis 149
3.5
8
3
6
2.5
4 2
1.5
2
1
0.5
(a) (b)
which direction the boundary has to move in order as s is the increment of the design variables obtained
to minimize the internal energy. The optimal shape is from the solution at the current iteration.
obtained, if the material residual is uniformly distrib- Furthermore, the sensitivity of the energy with
uted over the design boundary, see Fig. 4c. respect to a variation of the design variable si is given
by
The mesh nodal coordinates X are connected with the
∂X
geometrical design variables si by the so-called design E s i = G · δ X i = G · δsi = G · V i δsi . (107)
∂si
velocity fields V i , see e.g. Choi and Kim (2005a, b).
The design velocity field corresponding to the design The design velocity field connects the variation in the
variable si is given by geometrical design variable si with the variation in the
energy.
∂X
V i (X) = . (104)
∂si
The design velocity field V i (X) characterize the 7.3 Sensitivity of the energy release rate
changes of the finite element nodal point coordinates X
with respect to the changes of arbitrary design param- For a homogeneous elastic body, the material residual
eter si . or configurational forces are directly related to the well-
For example, a variation of the design variables s2 known J -integral, which is in linear fracture mechanics
and s11 generate the fields shown in Fig. 6a and b, equal to the energy release rate G, i.e.
respectively. Hence, these fields reflect the influence J = G = −G(u, s; ψ). (108)
domains of the design variations.
The velocity fields are also important and funda- Therefore, the sensitivity of the material residual (48)
mental in the context of mesh updating and smoothing. can be used to calculate the sensitivity of the J -integral
Let X be the vector of nodal coordinates, than the new or energy release rate G, which can be derived as
shape is obtained from G = −G = −(d(u, s; ψ, δs) + p(u, s; δu, ψ)).
X(s) = X(s) + εX(s) (105) (109)
Hence, for a given variation in the design δs we can cal- The calculation of the sensitivities require the varia-
culate the variation in the energy release rate or rather tions of the field quantities with respect to u ∈ V and
the variation in the material or configurational residual s ∈ D. Let η, ν ∈ V be admissible variations for the
G. deformation and ψ, χ ∈ D be admissible design vari-
An application for the sensitivity of the energy ations. Some important relations for the variations in
release rate of a problem from fracture mechanics is the physical and material space are listed below.
given in Barthold and Mesecke (1999).
A.1 Variations of strains
1
E s (u, ψ) = − (F Grad ψ)T F + F T F Grad ψ A.2 Variations of stresses
2
1
We consider a hyperelastic material, i.e. there exist a
=− Grad ψ T F T F + F T F Grad ψ
2 free energy function
such that
= − sym{F T F Grad ψ}. (118) ∂
∂
P = ρR and S = ρ R . (124)
Furthermore, the second and mixed variations follow ∂F ∂E
in a straightforward manner for variations ν ∈ V and The variations of the first Piola-Kirchhoff stress tensor
χ ∈D P read
1
P u (u, η) = A : F u = A : Grad η
E uu (η, ν) = Grad η T Grad ν + Grad ν T Grad η (125)
2
P s (u, ψ) =A: F s = −A : F Grad ψ (126)
= sym{Grad ν T Grad η} (119)
where
1 ∂P ∂ 2
E us (u, η, ψ) = − (Grad η Grad ψ)T F A= = ρR (127)
2 ∂F ∂F∂F
+ Grad η T F Grad ψ is called the first elasticity tensor (Marsden and Hughes
+(F Grad ψ)T Grad η 1994). The variations of the second Piola-Kirchhoff
stress tensor S read
+F Grad η Grad ψ
T
S u (u, η) = C : E u = C : sym{F T Grad η} (128)
= − sym{Grad ψ F Grad η S s (u, ψ) E s
T T
=C: = −C : sym{F F Grad ψ} (129)
T
A.3 Variation of line, surface and volume The explicit forms of the tangent operators are given
by
The variations of line, surface and volume elements
with respect to s are given by k(u, s; η, δu) = S : Grad δu T Grad η d
R
(d X) s = Grad ψ d X, (134)
+ F T Grad η : C : F T Grad δu d
(d A) s = [ Div ψ 1 − Grad ψ T ] d A, (135) R
B.1 Variations of the physical residual B.2 Variations of the material residual
The physical residual (12) written in terms of the ref- The material residual (13) written in terms of the ref-
erence configuration is given by erence configuration by using of Eq. 25 is given by
R(u, s; η) = P : Grad η d − b R · η d G(u, s; ψ) = (
+ V R 1) : Grad ψ d
R R R
= S : E u (u, η) d − b R · η d. −
R R R · ψ d
binh
R
(139)
= S : E s (u, ψ) d
Under the assumption that (b R ) u = 0 follow for (33) R
and (34) + (W R + V R )1 : Grad ψ d
R
k(u, s; η, δu) = S : E uu (η, δu) d
R − R · ψ d
binh (144)
R
+ E u (u, η) : C : E u (u, δu) d.
R where
= W R 1− F T P = W R 1− F T F S. The pseudo
(140) load operator (44) for the material problem follow in
the form
p(u, s; η, δs) = S : E us (u, η, δs) d t (u, s; ψ, δu) = S : E su (u, ψ, δu) d
R R
+ E u (u, η) : C : E s (u, δs) d + E s (u, ψ) : C : E u (u, δu) d
R
R
+ S : E u (u, η) Div δs d + S : E u (u, δu) Div ψ d
R
R
− b R · η Div δs d (141) − b R · δu Div ψ d. (145)
R R
Variational design sensitivity analysis 153
As a result of the symmetry of the pseudo load operator d(u, s; ψ, δs) = S : Grad δs T F T F Grad ψ d
for the physical (34) and material problem (44) follows R
t (u, s; ψ, δu) = p(u, s; δu, ψ).
+ S : F T F Grad δs Grad ψ d
If we consider a homogeneous elastic body with- R
out any material inhomogeneity, the variation of L(·)
vanishes identically, i.e. L s = 0. Therefore, only the + S : F T F Grad ψ Grad δs d
R
variation bs (u, s; ψ, δs) contribute to the material tan-
gent operator d(u, s; ψ, δs). Following the chain rule, + F T FGrad ψ:C:F T FGrad δsd
the variation is given by R
− S : F T F Grad ψ Div δs d
R
d(u, s; ψ, δs) =
s : Grad ψ +
:(Grad ψ) s d
R
− S : F T F Grad δs Div ψ d
R
+
: Grad ψ Div δs d. (146)
R + W R Div ψ Div δs d
R
A straightforward calculation by using the relations − W R 1 : Grad ψ Grad δs d.
from Appendix A leads to R
(149)
d(u, s; ψ, δs) = S : E ss (u, ψ, δs) d
R
Remark 19 All variations are obtained with respect to
+ E s (u, ψ) :C: E s (u, δs) d the deformation map ϕ, i.e. all operators have to be eval-
R uated at the current deformation ϕ. It is also possible to
+ S : E s (u, ψ) Div δs d express all the variations in terms of the actual displace-
R ment field u. The variation of the deformation gradient
F = Grad ϕ = 1 + H is then given by F s (u, ψ) =
+ S : E s (u, δs) Div ψ d H s (u, ψ) = −H Grad ψ. The variations of the Green-
R
Lagrange strain tensor yield
+ W R Div ψ Div δs d
R E s (u, ψ) = − sym{F T H Grad ψ} (150)
− W R 1 : Grad ψ Grad δs d. E us (u, η, ψ) = − sym{Grad ψ T H T Grad η
R
(147) +F T Grad η Grad ψ} (151)
dϕ du
= 1+ = −K −1 P(ϕ) = S(ϕ) (153) Dems K, Mróz Z (1978) Multiparameter structural shape optimi-
ds ds zation by the finite element method. Int J Numer Methods
Eng 13(2):247–263
and hence Engl HW, Hanke M, Neubauer A (2000) Regularization of
du
= S(ϕ) − 1 = S(u) = −K −1 P(u).
inverse problems. Kluwer Academic Publishers
(154) Eshelby JD (1951) The force on an elastic singularity. Philos
ds
Trans R Soc Lond 244:87–112
Eshelby JD (1975) The elastic energy-momentum tensor. J Elast
5:321–335
Govindjee S, Mihalic PA (1996) Computational methods for
References inverse finite elastostatics. Comput Methods Appl Mech
Eng 136(1–2):47–57
Askes H, Kuhl E, Steinmann P (2004) An ALE formulation Gurtin ME (2000) Configurational forces as basic concepts of
based on spatial and material settings of continuum mechan- continuum physics. Springer-Verlag, New York
ics. Part 2: classification and applications. Comput Methods Kalpakides VK, Balassas KG (2005) The inverse deformation
Appl Mech Eng 193:4223–4245 mapping in the finite element method. Philos Mag 85(33–
Barthold FJ (2002) Zur Kontinuumsmechanik inverser Geo- 35):4257–4275
metrieprobleme. Habilitationsschrift, Braunschweiger Sch- Kamat MP (1993) Structural optimization: status and promise.
riften zur Mechanik 44-2002, TU Braunschweig, URL American Institute of Aeronautics and Astronautics, Wash-
http://hdl.handle.net/2003/23095 ington
Barthold FJ (2003) A structural optimization viewpoint on con- Kienzler R, Herrmann G (2000) Mechanics in material space.
figurational mechanics. Proc Appl Math Mech 3:246–247 Springer-Verlag, Berlin Heidelberg New York
Barthold FJ (2005) On structural optimisation and configu- Kuhl E, Askes H, Steinmann P (2004) An ALE formulation based
rational mechanics. In: Steinmann P, Maugin GA (eds) on spatial and material settings of continuum mechanics.
Mechanics of material forces, vol 11. Springer-Verlag, pp Part 1: Generic hyperelastic formulation. Comput Methods
219–228 Appl Mech Eng 193:4207–4222
Barthold FJ (2007) Remarks on variational shape sensitivity anal- Marsden JE, Hughes TJR (1994) Mathematical foundations of
ysis based on local coordinates. Eng Anal Boundary Elem. elasticity. Dover, New York
Accepted for publication Materna D, Barthold FJ (2006a) Coherence of structural opti-
Barthold FJ, Mesecke S (1999) Remarks on computing the mization and configurational mechanics. Proc Appl Math
energy release rate and its sensitivities. In: Mota Soares Mech 6(1):245–246. DOI 10.1002/pamm.200610103
C, Mota Soares C, Freitas M (eds) Mechanics of compos- Materna D, Barthold FJ (2006b) Relations between structural
ite materials and structures, NATO science series, Serie E: optimization and configurational mechanics with appli-
applied science, vol 361. Kluwer Academic Publishers, pp cations to mesh optimization. In: Sienz J, Querin OM,
341–350 Toropov VV, Gosling PD (eds) Proceedings of the 6th
Barthold FJ, Stein E (1996) A continuum mechanical based for- ASMO-UK/ISSMO conference on engineering design opti-
mulation of the variational sensitivity analysis in structural mization, St Edmund Hall, Oxford, UK, University of
optimization. Part I: analysis. Struct Optim 11(1/2):29–42 Leeds, UK, pp 173–181
Bendsøe MP, Sigmund O (2003) Topology optimization—the- Materna D, Barthold FJ (2007a) On variational sensitivity analy-
ory, methods and applications. Springer-Verlag sis and configurational mechanics. Comput Mech. Accepted
Benzi M, Golub GH (2004) A preconditioner for generalized for publication
saddle point problems. Siam J Matrix Anal Appl 26(1):20– Materna D, Barthold FJ (2007b) The use of singular value decom-
41 position in sensitivity analysis and mesh optimization. To
Bertram A (1989) Axiomatische Einführung in die Kontinuums- be submitted
mechanik. BI-Wissenschaftsverlag, Mannheim Maugin GA (1993) Material inhomogeneities in elasticity.
Braun M (1997) Configurational forces induced by finite-ele- Chapman & Hall, London
ment discretisation. Proc Estonian Acad Sci Phys Math McNeice GM, Marcal PE (1973) Optimization of finite element
46:24–31 grids based on minimum potential energy. ASME J Eng
Carpenter WC, Zendegui S (1982) Optimum nodal locations for Indust 95(1):186–190
a finite element idealization. Eng Optim 5:215–221 Mosler J, Ortiz M (2006) On the numerical modeling of varia-
Carroll WE, Barker RM (1973) A theorem for optimum finite- tional arbitrary lagrangian-eulerian (vale) formulations. Int
element idealizations. Int J Solids Struct 9:883–895 J Numer Methods Eng 67:1272–1289
Chadwick P (1975) Applications of an energy-momentum ten- Mueller R, Maugin GA (2002) On material forces and finite ele-
sor in non-linear elastostatics. J Elast 5(3–4):249–258 ment discretizations. Comput Mech 29:52–60
Choi KK, Kim NH (2005a) Structural sensitivity analysis and Nocedal J, Wright SJ (1999) Numerical optimization. Springer-
optimization 1—linear systems. Mechanical engineering Verlag
series. Springer-Verlag, Berlin Noll W (1972) A new mathematical theory of simple materials.
Choi KK, Kim NH (2005b) Structural sensitivity analysis Arch Rational Mech Anal 48(1):1–50
and optimization 2—nonlinear systems and applications. Shield RT (1967) Inverse deformation results in finite elasticity.
Mechanical engineering series. Springer-Verlag, Berlin Z Angew Math Phys 18:490–500
Variational design sensitivity analysis 155
Steinmann P (2000) Application of material forces to hyperelas- Truesdell C, Noll W (2004) The nonlinear field theories of
tic fracture mechanics. I. Continuum mechanical settings. mechanics, 3rd edn. Springer-Verlag, Berlin
Int J Solids Struct 37:7371–7391 Zhang S, Belegundu A (1993) Mesh distortion control in shape
Steinmann P, Maugin GA (eds) (2005) Mechanics of material optimization. AIAA J 31(7):1360–1362
forces. Springer-Verlag
Thoutireddy P (2003) Variational arbitrary lagrangian-euleri-
an method. Phd thesis, California Institute of Technology,
Passadena, California
An anisotropic elastic formulation for configurational forces
in stress space
Anurag Gupta · Xanthippi Markenscoff
Abstract A new variational principle for an aniso- 1993; Gupta and Markenscoff 2007 and references
tropic elastic formulation in stress space is constructed, therein). A systematic way of obtaining expressions for
the Euler–Lagrange equations of which are the equa- these forces follows from Noether’s theorem
tions of compatibility (in terms of stress), the equi- (Noether 1918; Gelfand and Fomin 2000), where given
librium equations and the traction boundary condition. a variational principle, conserved integrals (path inde-
Such a principle can be used to extend recently obtained pendent in 2D) are obtained as necessary and sufficient
configurational balance laws in stress space to the case conditions for satisfying respective symmetries of the
of anisotropy. variational principle. Configurational forces are then
interpreted as these conserved integrals which vanish
Keywords Configurational forces · Stress in case of a non-dissipative flow, but fail to vanish if dis-
formulation · Anisotropy sipative mechanisms are active (i.e. when symmetries
are broken) (Eshelby 1975).
In (Li et al. 2005), the authors use a variational prin-
1 Introduction ciple of Pobedrya (Pobedrja 1980; Pobedria and
Holmatov 1982) to obtain a class of conservation laws.
Configurational forces (in solid mechanics) provide us The Euler–Lagrange equations of this principle were
with the forces necessary for driving the dissipative the Beltrami–Michell compatibility equations in the
mechanisms which are responsible for the kinetics of domain and stress equilibrium with traction boundary
defect flow. In case of a thermodynamic equilibrium, condition on the boundary. It was shown by Pobedrya
vanishing of configurational forces are additional rela- (1980) and Kucher et al. (2004) that such a boundary
tions required for a complete determination of the sys- value problem in terms of stress is well defined and it
tem. Popular examples are dislocation motion, crack is sufficient to satisfy equilibrium on the boundary to
propagation, delamination etc. (Eshelby 1956; Maugin satisfy it in the domain. The conservation laws were
A. Gupta (B) obtained assuming the case of linear, homogeneous
Department of Mechanical Engineering, University and isotropic elasticity with vanishing body forces and
of California, Berkeley, CA 94720, USA incompatibility. In a sequel paper (Markenscoff and
e-mail: agupta@berkeley.edu
Gupta 2007) these were extended for non-vanishing
X. Markenscoff incompatibility and body force distribution and appli-
Department of Mechanical and Aerospace Engineering,
University of California, La Jolla, San Diego,
cability of such quantities was demonstrated using
CA 92093, USA examples from dislocation theory and heat flow in a
e-mail: xmarkens@ucsd.edu domain with a spherical cavity. More recently these
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 157
doi: 10.1007/978-1-4020-6929-1_14, © Springer Science+Business Media B.V. 2007
158 A. Gupta, X. Markenscoff
conserved quantities (or configurational forces) were with pi (xk ) being the prescribed traction and nj (xk ),
interpreted as the necessary and sufficient dissipative the unit normal to ∂. The system of Eqs. 1 and 4 is
mechanisms so as to maintain compatibility during over-determined as there are nine equations for six un-
the propagation of a inhomogeneity (or a defect) (Gupta knowns. On the boundary, there is an under-
and Markenscoff 2007). determinacy by three conditions (Georgievskii and
The present paper aims at formulating a variational Pobedrya 2004). Pobedrya (1994) has introduced a set
principle which would extend the earlier principle of of equations to deal with this problem. His formulation
Pobedrya (1980) to the case of anisotropy. Such a var- transfers the equilibrium Eq. 1 to the boundary and
iational principle can then be used to obtain the con- therefore leaving only Eq. 4 to be solved in the bulk.
figurations balance laws in stress space (Li et al. 2005; The resulting system of equations is well defined over
Markenscoff and Gupta 2007) for the case of aniso- the whole domain (Pobedrja 1980; Kucher et al. 2004).
tropic elasticity. Before constructing a variational prin- An outline of his theory for anisotropic elasticity is now
ciple in Sect. 3, we first discuss Pobedrya’s formulation discussed.
(α)
of the boundary value problem of anisotropic linear Let aij be a set of tensors (α = 1, . . . , N ) con-
elasticity in stress space. structed from the invariant basis tensors of the symme-
try group G. These invariant basis tensors were
originally obtained for the representation of anisotropic
2 Formulation tensors and there are systematic procedures for their
evaluation corresponding to the symmetry group G (see
The classical problem of linear and homogeneous elas- (Weyl 1946; Smith and Rivlin 1957; Smith 1970) for
ticity in terms of stress involves equilibrium and com- foundations and (Markenscoff 1976) for a possible
(α)
patibility equations in the bulk (bulk is denoted by application). The tensors aij are constructed such that
∈ R3 ) and a traction boundary condition (boundary they are pairwise orthogonal and add up to the unit
is denoted by ∂ ≡ ∩ R3 / ). Assuming absence of tensor,
body forces and inertial terms, the equilibrium equation (β)
is given as follows, aij(α) aij N
= δαβ , aij(α) = δij , (6)
a(α) a(β)
σij,j = 0, ∀xk ∈ (1) α=1
with σij (xk ) denoting the stress. The compatibility rela- where a(α) = aij(α) aij(α) (no summation implied un-
tions are written in terms of strain eij (xk ) as, der α). A set of linear invariants of the incompatibility
ηij ≡ −ikl j mn eln,km = 0, ∀xk ∈ (2) tensor can then be constructed as,
(α)
where ikl is the alternating tensor and ηij is the η(α) = ηij aij . (7)
Kröner’s incompatibility tensor (Kröner 1981) (vanish-
Define a tensor Hij as,
ing of which ensures compatibility). The compatibility
relations in terms of stresses can be obtained by us- N
ing an appropriate constitutive law in the framework of Hij = ηij + ξ (α) η(α) aij(α) , (8)
linear elasticity, α=1
eij = Dij kl σkl , (3) where ξ (α) are arbitrary scalars. The condition Hij = 0
is equivalent to the compatibility condition (2), ηij = 0
where Dij kl is the (constant) elastic compliance tensor.
if ξ (α) = −(a(α) )−2 for all α. Indeed if ηij = 0 then by
Denote η̄ij as the incompatibility tensor thus obtained
definitions (7) and (8), Hij = 0. If Hij = 0, then by Eq.
as a function of the stress tensor. Therefore the com- (β)
patibility relations in terms of stresses are, 8, Hij aij = η(β) + ξ (β) η(β) (a(β) )2 = 0, where rela-
tions (6) and (7) have been used. If ξ (α) = −(a(α) )−2 ,
η̄ij ≡ −Dlnpq ikl j mn σpq,km = 0, ∀xk ∈ . (4) then η(α) = 0 and consequently, using Hij = 0 in (8)
Equations 1 and 4 are to be supplemented by prescrib- we obtain ηij = 0.
ing traction on the boundary, Define a vector valued function,
σij nj = pi , ∀xk ∈ ∂ (5) Ai = Rij σj k,k , (9)
An Anisotropic Elastic Formulation 159
where Rij is a positive definite operator. Therefore, and thereafter combine Eqs. 16 and 17 to write,
condition Ai = 0 is equivalent to equilibrium relation
N
(1). Construct a tensor Aij = Ai,j + Aj,i whose lin- 1 (α)
H̄ij,j =
Ai + δij + ξ (α) aij
ear invariants are denoted by A(α) = Aij aij(α) . Another 2
α=1
tensor can then be defined as,
×(η̄(α) + A(α) ),j . (18)
N
(α)
Āij = Aij + ξ (α) A(α) aij . (10) Substitution of relations (12) and (14) into (18) then
α=1
results into
Ai = 0, i.e. Ai is harmonic. The vector
Reasoning along the lines of the paragraph following
Ai vanishes on the boundary (by (13)2 ) and therefore
Eq. 8, we note that Aij = 0, if and only if Āij = 0
it vanishes inside (since Ai is harmonic). Therefore
given that ξ (α) = −(a(α) )−2 .
equilibrium relation (1) is satisfied in the bulk. As a
Form a tensor,
consequence of this result and Eqs. 11 and 12, obtain
H̄ij = Hijσ + Āij , (11) Hijσ = 0. The compatibility relation (4) then follows
where superscript σ in Hijσ
indicates that Hij is ex- from this equality. This is exactly the generalization to
pressed in terms of stress rather than strain (using (6)). anisotropy of Pobedrya’s method.
(α) η̄ a (α) where η̄
Therefore Hijσ = η̄ij + N α=1 ξ (α) ij ij
is as given in (4) and η̄(α) = η̄ij aij(α) . The formula-
tion of Pobedrya (Pobedrya 1994) involves satisfying
equation, 3 A New Variational Principle
H̄ij = 0, ∀xk ∈ (12)
Consider the following functional,
in the bulk and conditions,
σij nj = pi , σij,j = 0, ∀xk ∈ ∂ (13)
(σij , σij,k ) = Eij k σij,k dV
on the boundary. The solution to Eqs. 12 and 13 satisfies
equations of equilibrium and of compatibility as given 1
(β) − Eij k σij nk − σij,j σik,k
in (1) and (4). From Eqs. 11 and 12 obtain, H̄ij aij = ∂ 2
(η̄(β) + A(β) )(1 + ξ (β) (a(β) )2 ) = 0, and therefore if 1
− σij nj σik nk + pi σij nj dA, (19)
ξ (α) = −(a(α) )−2 is satisfied, we obtain, 2
η̄(β) + A(β) = 0. (14)
where
Differentiate Eq. 11 to get,
σ
H̄ij,j = Hij,j + Āij,j (15) Eij k = − Dlnpq ikl j mn σpq,m + Ai δj k + Aj δik
N
which will now be evaluated term by term. Let η̄ de-
note the trace of η̄ij . Therefore η̄ = η̄ij δij and from + ξ (α) (−Dlnpq rkl smn σpq,m
α=1
(4), η̄ = 2(Diij k σj k,pp − Dmnj k σj k,mn ). It is easy to
(α) (α)
see that η̄ij,j = 21 (η̄),i . Also, note that η̄ = η̄ij δij = + Ar δsk + As δrk )ars aij (20)
(α) N
η̄ij N α=1 aij = α=1 η̄(α) , where the first relation in
σ can then be and Ai is as given in (9). Define L = Eij k σij,k , then
(6) has been used. Using these results Hij,j
to obtain the Euler–Lagrange equations corresponding
evaluated as,
N to the above functional, evaluate ∂σ∂Luv,w
using following
1 (α) formulae,
σ
Hij,j = δij + ξ (α) aij η̄(α),j . (16)
2
α=1 ∂(Dlnpq ikl j mn σpq,m σij,k )
For calculating Āij,j , start from (10) and note that (i)
∂σuv,w
Aij,j =
Ai + Aj,j i and Aj,j = 21 Aij δij = 21 Aij = Symuv [Dlnpq ikl j mn (δup δvq δwm σij,k
N (α) 1 N
α=1 aij = 2 α=1 A(α) to obtain, +δui δvj δwk σpq,m )]
1
N
(α)
= Dlnuv ikl j wn σij,k
Āij,j =
Ai + δij + ξ (α) aij A(α),j (17)
α=1
2 +Symuv (Dlnpq uwl vmn σpq,m )
160 A. Gupta, X. Markenscoff
∂(Dlnpq rkl smn σpq,m σij,k ) These equations give conditions on the arbitrary vari-
(ii)
∂σuv,w ables ξ (α) and Rij which make the variational principle
= Symuv [Dlnpq rkl smn (δup δvq δwm σij,k suitable. Under such conditions Eq. 22 simplifies to,
+δui δvj δwk σpq,m )] δ = − Euvw,w δσuv dV + σij,j δσik,k dA
= Dlnuv rkl swn σij,k ∂
+Symuv (Dlnpq rwl smn δui δvj σpq,m ) + (σik nk − pi )δσij nj dA. (24)
∂
∂(Rip δqr δj k σpq,r σij,k )
(iii) = Symuv (Riu δvw σij,j ) The equation Euvw,w = 0 is equivalent to (12). Finally,
∂σuv,w note that Eq. 23 should be satisfied for arbitrary σij,k
+Symuv (Rup δvw σpq,q ) and therefore we obtain necessary and sufficient con-
∂(Rrp δqt δsk σpq,t σij,k ) ditions,
(iv) = Symuv (Rru δvw σij,s )
∂σuv,w −Dlnuv ikl j wn + 2Riu δvw δj k
+Symuv (Rrp δiu δj v δsw σpq,q ), N
where the notation Symuv (·)uv = + ξ (α) (−Dlnuv rkl swn ars
(α)
2 [(·)uv
+ (·)vu ] has
1
α=1
been employed. Also note, that for a symmetric ten-
(α) (α)
∂σij
sor σij , ∂σuv = Symuv (δiu δj v ). Using these relations +2Symrk (Rru δvw ark ))aij = 0 (25)
obtain, which are used as restrictions on the arbitrary variables
∂L ξ (α) and Rij .
= Symuv (Euvw ) − Dlnuv ikl j wn σij,k Functional (19) can be used to derive conservation
∂σuv,w
laws using the formalism of Noether’s theorem (Noe-
+2Symuv (Riu δvw σij,j )
ther 1918; Gelfand and Fomin 2000). Noether’s
N
theorem provides a systematic procedure to obtain dive-
+ ξ (α) (−Dlnuv rkl swn σij,k rgence-free quantities from given symmetries of a var-
α=1
iational problem such as one formulated in (19). The
(α) (α)
+2Symuv Symrs (Rru δvw σij,s ))ars aij . divergence-free quantities in an integral form provide
(21) conservation laws which in the presence of singularities
and inhomogeneities will result into configurational
Taking a variation of the functional (19) with respect forces acting on these defects. An extension of Noe-
to σuv , we evaluate, ther’s theorem to tensorial fields was already achieved
d ∂L (Li et al. 2005; Markenscoff and Gupta 2007) and con-
δ = − δσuv dV − Euvw servation laws were obtained for translation, rotation,
dxw ∂σuv,w ∂
pre-stress and scaling symmetries, which are new and
∂L
− δσuv nw dA + σij,j δσik,k dA distinct (Gupta and Markenscoff 2007). Application of
∂σuv,w ∂
these conservation laws allowed for the determination
+ (σik nk − pi )δσij nj dA, (22) of the incompatibility in the interior of the domain by
∂ surface data. The variational principle obtained in this
where the ‘frozen condition’ (Pobedrja 1980): paper extends previous work to the case of anisotropy.
δEij k nk = 0 on ∂ has been assumed. The frozen
Acknowledgements This research was supported by a SEGRF
condition ensures that variation δσuv does not intro-
fellowship from Lawrence Livermore National Laboratory (AG)
duce a flux of incompatibility across the boundary into and by NSF Grant No. CMS-0555280 (XM).
the bulk. Noting expression 21, Eqs. 12 and 13 are then
recovered as the Euler–Lagrange equations if,
References
− Dlnuv ikl j wn σij,k + 2(Riu δvw σij,j )
N
Eshelby JD (1975) The elastic energy-momentum tensor. J
+ ξ (α) − Dlnuv rkl swn σij,k Elasticity 5:321–335
Eshelby JD (1956) The continuum theory of lattice defects. In:
α=1
(α) (α) Seitz F, Turnbull D (eds) Solid state physics. Academic
+ 2Symrs (Rru δvw σij,s ) ars aij = 0. (23) Press, New York pp 79–144
An Anisotropic Elastic Formulation 161
Gelfand IM, Fomin SV (2000) Calculus of variations, Dover Maugin GA (1993) Material inhomogeneities in elasticity.
Georgievskii DV, Pobedrya BYe (2004) The number of indepen- Chapman & Hall, London
dent compatibility equations in the mechanics of deform- Noether E (1918) Invariante variationsprobleme, Göttinger
able solids. Prikl Mat Mekh 68(6):1043–1048 (trans: J Appl Nachrichten, (Mathematisch-physikalische Klasse) 2:235–
Math Mech 68:941–946 (2004)) 257 (trans: Transport Theory and Statistical Physics 1:186–
Gupta A, Markenscoff X (2007) Configurational forces as dissi- 207 (1971))
pative mechanisms: a revisit, Comptes Rendus Mecanique Pobedrja BE (1980) A new formulation of the problem in
(In press) mechanics of a deformable solid body under stress. Soviet
Kröner E (1981) Continuum theory of defects. In: Balian R et al Math Doklady 22(1):88–91
(eds) Les Houches, Session 35, 1980 – Physique des defa- Pobedria BE, Holmatov T (1982) On the existence and unique-
uts. North-Holland, New York, pp 215–315 ness of solutions in the elasticity theory problem with
Kucher V, Markenscoff X, Paukshto M (2004) Some properties stresses, Vestnik Moskovskogo Universiteta, Seriya 1 (Ma-
of the boundary value problem of linear elasticity in terms tematika Mekhanika), 1, pp 50–51
of stresses. J Elasticity 74(2):135–145 Pobedrya BYe (1994) The problem in terms of a stress tensor
Li S, Gupta A, Markenscoff X (2005) Conservation laws of lin- for an anisotropic medium. Prikl Mat Mekh 58(1):77–85.
ear elasticity in stress formulations. Proc Roy Soc Lon A (trans: J Appl Math Mech 58(1):81–89 (1994))
461:99–116 Smith GF, Rivlin RS (1957) The anisotropic tensors. Quart Appl
Markenscoff X (1976) Independent 4th-order elastic coefficients Math 15(3):308–314
for quartz. Appl Phys Lett 29(12):768–770 Smith GF (1970) The crystallographic property tensors of orders
Markenscoff X, Gupta A (2007) Configurational balance laws 1 to 8. Ann N Y Acad Sci 172(5):57–106
for incompatibility in stress space. Proc R Soc Lond A, Weyl H (1946) The classical groups, their invariants and repre-
463:1379–1392 sentations. Princeton Univ Press, Princeton, NJ
Conservation laws, duality and symmetry loss in solid
mechanics
Huy Duong Bui
Abstract The paper deals with conservation laws one, at Symi 2005, belong to a series of meetings orig-
which are not of the pure divergence type and thus do inated from G. Herrmann‘s idea to gather many peo-
not provide a path-independent integral for use in Frac- ple involved in Conservation laws and Configurational
ture Mechanics. It is shown that Duality is the right tool Forces in Solid Mechanics.
to re-establish the symmetry between equations and to My first contact with George Herrmann began in
provide conservation laws of the pure divergence type. the middle of 70s, when he encouraged me to investi-
The loss of symmetry of some energetic expressions gate on Conservation laws in thermoelasticity. At this
is exploited to derive a new method for solving some time, most works concentrated on computational prob-
inverse problems. In particular, the earthquake inverse lems, for example on how to derive stress intensity
problem is solved analytically. factors from path-independent integrals, like Rice’s
J-Integral. I realized that George Herrmann’s question
Keywords Conservation laws · Duality · Symmetry · for thermo-elasticity is rather difficult, because of the
Symmetry loss · Inverse problem presence of an entropical “source” term in the cor-
responding conservation law which prevents us from
deriving a path-independent integral. I was haunted by
1 Introduction his question for a while, until the solution found for a
true path-independent integral in linear thermoelastic-
This review paper summarizes two informal presenta- ity (Bui and Proix 1984). The method of solution to
tions at the Conference on “Problems in Solids Mechan- derive a path-independent integral from the one which
ics” 2006 at Symi, which is a paradisaical island of is not a true path-independent integral, appears to be a
Greece, and at a workshop at Aussois 2007, which is a quite general one. It is based on the notion of Duality.
nice place of the French Alps, held in the memory of The notion of duality and symmetry is closely linked
George Herrmann. Both conferences and the previous to the concept of “Virtual Power”. This was introduced
in France in the Mechanics of continuous media by Paul
Dedicated to George Herrmann. Germain for an adequate representation of the action
(forces, stress. . .) on a body (Germain 1973, 1978). In
H. D. Bui (B)
Department of Mechanics, Laboratory of Solid Mechanics,
one of his paper, he wrote “This concept is very seldom
Ecole Polytechnique, Palaiseau, France considered in the English scientific community, which
e-mail: hdbui37@yahoo.fr directly made use of equations, for example the classi-
H. D. Bui cal Newton law (f = ma) or Cauchy law (divσ = ρa)”.
Lamsid/CNRS Electricite de France, Clamart, France It originates from the mathematical concept of spaces
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 163
doi: 10.1007/978-1-4020-6929-1_15, © Springer Science+Business Media B.V. 2007
164 H. D. Bui
elasticity, rod, strings, etc.). When I generalized Tonti’s tostatics is generalized to elastodynamics, using oper-
diagram of elasto-statics to dynamic elasticity (Bui ators C, S∗ in the kinematic chart which are adjoint to
1992), I saw the beautiful structure of the equations C∗ and S of the dynamic chart. Operators D, D∗ are
using dual variables. I discovered that the links between D = −div, D∗ = 21 (∇ + ∇ t ). R is the left curl for
dual variables are always governed by differential oper- symmetrical second order tensor, R∗ is the right curl
ators and their adjoint ones (∗) (Fig. 1). for symmetrical second order tensor.
Conservation laws, duality and symmetry loss in solid mechanics 165
3 The DN and ND maps ear form G(u, u∗ ) = 2(1 − ν 2 )(KI KI∗ + KII KII∗ )/E. By
choosing a symmetric adjoint field (KI∗ = 1, KII∗ = 0)
When I was still a research student, I wondered how we extracted the SIF in mode I by G(u, u∗ ) =
the Dirichlet boundary value problem in elasticity, with 2(1 − ν 2 )KI /E and similarly by considering an
the prescribed datum uid (x), x ∈ ∂ on the bound- anti-symmetric adjoint field (KI∗ = 0, KII∗ = 1) I
ary can be replaced by the corresponding Neumann obtained the stress-intensity factor in mode II. The vir-
boundary value problem with the prescribed datum tual crack propagation is richer than the derivative of
Tid (x), x ∈ ∂. I found the solution to my problem for a energy since it contains the classical result 1/2G(u, u) ≡
half plane but ignored at this time that the map uid (x) → J1 -integral. There is a profound difference between the
Tid (x), called the Dirichlet-to-Neumann map (DN) or its virtual crack propagation method and the J2 -method
inverse the Neumann-to-Dirichlet map (ND), will have which involves a crack translation out of its plane with
very important applications in the solution of some the same loading. The virtual method G(u, u∗ ) is based
inverse problems in elasticity investigated in the 90s on a crack propagation in its direction, but under a
by mathematicians. This paper published in the six- virtual load giving rise to u∗ , KI∗ and K∗II .
ties (Bui 1967), indicates that the concept of duality In the 70s, I observed some intriguing results on the
has been the driving concept of my works for many energy release rate in elastodynamics. As a student, I
decades. The DNs are the key tools to solve the “Crack always learned that a formula describing physical phe-
detection problem by a geometry approach” (Bui 1993). nomena must be independent of the motion of the frame
reference in which measurements are made. This is the
objectivity principle in Physics. The energy release rate
4 Duality in fracture mechanics formula in elasto-dynamics for a moving crack with
the velocity V does not satisfy this principle since the
In 1973, I found that Rice’s J-integral is only one velocity is explicitly present in its expression, in plane
description of the energy release rate G, by a path- strain mode I (Achenbach and Bazant 1972)
independent integral and that the dual I-integral is
another possible one. This offers a great advantage in 1 − ν2 2
considering both descriptions with dual variables and G= KI fI (V ) (1)
E
spaces, conservation laws and dual laws since the min-
imum theorems for the potential energy W(ε) and the
where fI (V ) = β1 (1 − β2 )/{(1 − ν)(4β1 β2 − (1 +
2
complementary potential U(σ ), under certain condi-
tions, provide us exact bounds of the J-integral (Bui β22 )2 }, βi = 1 − (V 2 /ci2 ), c1 for P-wave, c2 for
1974). Another important applications of the virtual S-wave. How to restore the objective formula for the
power principle in Fracture Mechanics is provided by dynamic G? My response to this question was duality.
the notion of the virtual crack propagation. Classically, Let me introduce for the mode I the same local defin-
one deals with the energy release rate G as the deriv- itions of stress-intensity factor and crack displacement
ative of the energy with respect to the crack length. intensity factor as known in quasi-statics, respectively
Therefore, G in mixed modes I + II is well known as √
the quadratic form G = (1 − ν 2 )(KI2 + KII2 )/E. The KIσ = lim σ22 2π r,
r→0
question was raised on how to separately extract the
u µ 2π
stress-intensity factors. Many methods were proposed KI = lim [[u2 ]] (2)
r→0 4(1 − ν) r
consisting in calculating the derivatives of the energy
in the Ox1 direction J1 = (1 − ν 2 )(KI2 + KII2 )/E (crack In quasi-statics, both definitions provide the same SIF.
propagation along Ox1 ) and in the Ox2 direction J2 = In elastodynamics, I found a symmetrical formula for
−(1 − ν 2 )KI KII /E (crack translation out of its plane). the energy release rate which is nothing but the duality
Such an unphysical method (for an actual derivative) between stress and strain rate near the moving crack tip
was criticized by many authors. I tried to look at the (Bui 1977)
virtual power method, with arbitrary adjoint fields u∗
and discovered that the virtual power of the energy of a 1 − ν2 σ u
cracked body, in two-dimensions, is equal to the bilin- G= K I KI (3)
E
166 H. D. Bui
This objective formula agrees with the traditional one adjoint field u∗ is taken as the asymptotic field for a
since it can be proved that KIu = KIσ fI (V). semi-infinite crack, with stress-intensity factors KI∗ ,
Another beautiful application of duality is about a KI∗I and the other adjoint field is w ∗ = (2KI∗ /µ)
conservation law in linear thermo-elasticity, which is (r/2π )1/2 (1 − ν) cos(ϕ/2). For a cracked body having
George Herrmann’s question to me. It is well known a line crack along the negative part of Ox1 , the condi-
that classical conservation laws in thermo-elasticity tion ∂(w∗ − u1∗ )/∂n = 0 is satisfied by such explicit
involves a source term, namely in the form divA(u, τ ) asymptotic solutions u∗ , w∗ so that there is no need
= Bu, τ ). Precisely because of the source term B that for a finite element computation of the adjoint fields
the thermo-elastic J-integral is not a purely path- which would equally introduce further difficulties in
independent integral, since it involves an area integral the estimation of near tip fields.
too. The symmetry is lost when we consider the pair (u, Does a symmetry exist in conservation laws in elas-
τ ) lone. We restore the symmetry by considering the todynamics? It is clear that the conservation law
dual pair ({u, τ }, {u∗ , w∗ }) and obtained a conservation divσ [u] = ρ ü or the ones derived by Fletcher (1976)
law in the form are not symmetric. To restore the symmetry, it is nec-
divA(u, τ ; u∗ , w∗ ) = 0 using dual variables, with- essary to introduce adjoint fields v(x, t; τ ) satisfying
out a source term. The conservation law in linear ther- the elastodynamic wave equations divσ [v] = ρ v̈ such
moelasticity of the pure divergence form, for a line that v(x, t; τ ) ≡ 0 for t > τ where τ is an arbitrary con-
crack problem (along negative Ox1 ) is given by stant. We obtained the symmetric conservation law in
∂ 1 ∗ 1 elastodynamics given in (Bui and Maigre 1988) as
ui σij,1 − u∗i,1 σij − γ τ (u∗1,j − w,j
∗
)
∂xj 2 2 ⎧ τ ⎫
⎨ ⎬
∗ ∗ (n · σ [u] · v − n · σ [v] · u)dt =0
+γ τ,j (u1 − w ) = 0 (4) div (6)
⎩ ⎭
0
where γ = −αµ(3λ + 2µ)/(λ + µ), α is the thermal
coefficient, λ and µ are Lamé’s coefficients. The actual This conservation law has been exploited in (Bui
temperature field τ as well as the scalar adjoint field w∗ et al. 1992) to extract the stress-intensity factors in
are harmonic, while the field u∗ is an elastic one. The dynamic modes I and II by choosing appropriately the
stress free condition is assumed on the crack σ · n = adjoint dynamical fields.
0 as well as the zero normal heat flux ∂τ/∂n = 0. (∗ ) The loss of physical significance of adjoint problems
The adjoint fields (u∗ , w∗ ) is not a thermoelastic one. is often encountered. For example consider the heat
There is a small coupling between adjoint fields, by diffusion equation div(kgradT) − ∂t T = 0. Its adjoint
imposing the following condition on the crack faces equation div(kgradT∗ )+∂t T∗ = 0 is not physical, since
∂(w∗ − u1∗ )/∂n = 0, which has no precise physical it describes backwards diffusion, violating the second
meaning (∗) . This boundary condition on the crack faces principle of thermodynamics.
has only been introduced as a mathematical condition
which enables the following path independent integral
1 ∗ 1 5 Duality in plasticity
T = ui σij,1 nj − u∗i,1 σij nj
2 2
Dual variables are crucial in the thermodynamics of
−γ τ (u∗1,n − w,n ∗
) + γ (u∗1 − w ∗ )τ,n ds (5) irreversible processes. The contributions of J.J. Moreau,
Q.S. Nguyen, P. Germain, P. Suquet, A. Ehrlacher, C.
The path independent T -integral in thermoelasticity Stolz and others in France, during the period 1960–
(5) was presented at the Eshelby Symposium (Funda- 1990 are very important to clarify the nature of dissi-
mentals of deformation and Fracture, Sheffield April, pation in Plasticity and Fracture. Internal rate variables
1984), in honour of a great scientist who impinged α̇, including the plastic strain rate ε̇p , α̇ = (ε̇p , β̇)
on many works in Fracture Mechanics. Among appli- describe the evolution of materials. The variable α̇ is
cations of Eq. 5 to Fracture Mechanics, we recall the the dual to the generalized force A, so that A · α̇ ≥ 0
most important one, namely T = (1 − ν 2 )(KI KI∗ + represents the dissipation rate. If one introduces the
KI I KI∗I )/E in mixed mode plane strain, whenever the free energy per unit volume W(ε, α) so that
Conservation laws, duality and symmetry loss in solid mechanics 167
Therefore, R is a defect indicator (also called a reci- a non-linear equation for determining the crack plane
procity gap): (containing ) as well as the displacement disconti-
R = 0 ⇔ no defect inside ∂, nuity [[u]]. The non-linearity come from the fact that
R = 0 ⇔ existence of a defect. (u) is unknown. It is impossible to solve the non-
Therefore, defects detection reduces to the search linear inverse problem with classical methods based
of the zeros of a functional. A series of recent papers on the fields equations, because the crack support (u)
of my two research teams showed that the reciprocity depends on the unknown u. Now, the variational form
gap functional provided a closed form solution to some makes it possible to solve the inverse problem in a
inverse crack detection problems, for electrostatics closed form. The method of solution consists of two
(Andrieux and Ben Abda 1992) static elasticity steps:
(Andrieux et al. 1999; Andrieux and Ben Abda 1996), 1. Find the normal n to the fault plane by adequate
diffusion equation (Ben Abda and Bui 2003), transient choice of u∗ , and the fault plane (u).
acoustics (Bui et al. 1999), and elastodynamics with 2. Once the fault plane having determined, find the
the exact solution to an earthquake inverse problem discontinuity [[u]].
(Bui et al. 2005a, b). Closed form solutions can only be
obtained in particular case as for a planar crack in lin- Remark that step 2 is a linear inverse problem which
ear material. The method of exact solutions does apply is incomparably simpler than the original problem.
neither to inhomogeneities nor to non linear materi- An adequate choice of the adjoint function allows
als. In the particular case of small inhomogeneities, the invertibility of the above equation. We exploit here
i.e. small perturbations of the homogeneous medium, the arbitrariness of the choice of functions u∗ to obtain
closed form solutions can be obtained for the linearized the desired results. Traditional methods deal with the
inverse problems: fields equations (elastic equilibrium equation, bound-
ary conditions, with an unknown geometry). Therefore,
a. For the scalar equation div{(1+h(x)gradϕ}=0 with the only possible method consists in finding the best fit-
unknown h(x), with data ϕ and ∂ϕ/∂n on the bound- ting of measurements with predicted data correspond-
ary, see the papers by Bui (1994) and Calderon ing to some guess geometry S
(1980).
b. For a small isotropic perturbation δL of homoge- 2 2
d d
neous isotropic elastic medium, we can cite Ikehata { and [[u]]} = arg min v − u +
T(v) − T
S, [[v]]
(1998).
(14)
7 Solution to the earthquake inverse problem where v is the solution of the boundary value problem
with the geometry /S and with one of the bound-
As shown above, the reciprocity gap R is defined as the ary condition, either ud or Td (two possible numerical
external boundary functional over Sext which is known solutions!). This classical method of solution is essen-
from the data u1 = ud and T(u1 ) = Td and from tially a numerical one. It is well known that the above
the chosen adjoint functions {u∗ , T(u∗ )}, denoted by optimization procedure is mathematically an ill-posed
R(ud , Td ; u∗ , T(u∗ )). In planar crack detection prob- problem (Tikhonov and Arsenine 1977). Undoubtedly,
lems in 3D quasi-static elasticity, we can prove the fol- the reciprocity gap functional, based on the symmetry
lowing variational equation: Find (u), [[u]] such that loss in Fracture Mechanics is the right tool to solve
these inverse problems in closed form.
[[u]] · T(u∗ )ds = ud · T(u∗ ) − u∗ · T d ds In practice, the earthquake inverse problem consid-
(u) Sext
ered here is illustrated in Fig. 2, with measured acceler-
. ation data on the ground G and estimated far field data
= R(ud , Td ; u∗ , T(u∗ )), ∀u∗(13)
on the half-sphere H using a classical model of point
In the homogeneous body case (no crack), the left hand source (Aki and Richards 1980). The near fields mea-
side of the above equation equals zero. By R = 0, surements, together with the far-fields and adequate
we recover the symmetry between fields u and u∗ . In adjoint fields are inserted in the elastodynamic reci-
the symmetry lost case, the above equation provides procity gap functional
Conservation laws, duality and symmetry loss in solid mechanics 169
Fig. 2 The unknown moving fault (t) in its plane near the
free surface G. Acceleration data are measured on the ground G; wave front at t>0
Estimated data on the half-sphere H are provided by a model of
a point source near the origin with measured seismic moments Fig. 3 Back propagation in direction p of shear adjoint wave u∗
parallel to k (k⊥p)
∞
[[u]] · T (u∗ )dsdt
0 (u)
∞
= ud · T(u∗ ) − u∗ · Td dsdt
0 Sext
.
= R(ud , Td ; u∗ , T(u∗ )) (15)
with adequate adjoint fields u∗ ,
one adjoint field to
determine first the fault plane, and another one to deter-
mine the fault geometry.
Step 1 can be solved by different choice of adjoint
fields. One interesting method is provided by the Fig. 4 Convex hull containing the crack defined by the positions
“instantaneous reciprocity gap functional” (IRGF) of plane shock wave when |IRGF| = 0(ε)
which makes use of plane shear impulse waves k of
propagation vectors p. The adjoint displacement is
If the initial wave front is still in the right of the right
given by
crack tip, the adjoint stress vector is equal to zero on
u∗ (x, t; τ ) = akY(t − x · p/cs − τ ), (16) the crack faces, as u∗ is constant. The IRGF is equal to
where τ is a parameter defining the initial wave front zero up to the moment when the wave front reaches the
as x · p/cs + τ = 0 (t = 0), and Y(y) is the down crack tip. By varying both the directions of propagation
step function Y(y < 0) = 1, Y(y > 0) = 0 defined in vectors p and the time parameters τ , one obtains a set of
terms of the usual Heaviside function H(y) by Y(y) = adjoint waves encompassing the crack. The condition
1 − H(y) (Y is understood as the limit for h → 0 of IRGF = 0 of the set of adjoint fields defines a convex
the regularized step function Yh (y) which represents a hull containing the crack. The smallest convex hull is
smooth transition from 0 to 1 in a narrow band of width obtained when all waves “arrive” at the crack tips (or
h, at y = 0, Yh (y > h/2) = 0, Yh (y < −h/2) = 1). the crack front in 3D). The numerical criterion for the
Suppose that parameter τ s chosen so that the initial arrival times of waves is IRGF = O(ε), where ε is a given
wave front is outside the crack, in the left as displayed small number.
in Fig. 3. As the adjoint wave propagates in direction Figure 4 shows the computed wave front when the
p, towards the region where u∗ = 0, the adjoint field magnitude of IRGF is equal to a given small quantity
vanishes u∗ = 0 on the crack for any t > 0. Therefore, O(ε). In this case the adjoint stress is the travelling
IRGF vanishes for any time t > 0. If the initial wave Dirac delta. The simulated earthquake is provided by a
front intersects the crack, then IRGF = 0. The adjoint sudden release of geological stresses on the crack.
field vanishes again when the wave front “arrives” at This is analogous to X-rays tomography. When the
the left crack tip. ray is outside a tumour, one does not have any attenua-
170 H. D. Bui
tion of the X-ray signal A = 0. But an attenuation signal a real number, and
A = 0 is detected when the ray intersects the tumour.
This is why the terminology of “generalized tomog- ψ(x, t; s, q) = exp(iqt − ηt) exp(is · x) exp[x3 {|s2 |
raphy” is sometimes used in the literature of inverse +(iq − η)2 /c22 }1/2 ] (18)
problems (Bui 2006).
where η is a real vanishing positive number (η → 0+ )
The IRGF provides both the normal and the fault which has been introduced for convergence reasons (in
plane. It does not determine the crack geometry when the space of tempered
distributions). Let us introduce
the crack is concave. To solve step 2, we make use of the vector u⊥ = ([[u2 ]] , − [[u1 ]] , 0) orthogonal to
the following adjoint field ⊥ [[u]] in the fault plane (x3= ⊥
vector 0).Vectors [[u]] and
u as well as the field div u have the same
spatial support in the
fault plane. From Eq. 15, it can be
u∗ ≡ w(s,q) (x, t) = curl{ψ(x, t; s, q)e3 } (17)
proved that div u⊥ is explicitly determined by the
data R(ud , Td ; w(s,q) , T(w(s,q) )), via an inverse spa-
where x =(x1 , x2 , x3 ), with Ox3 along the normal tial Fourier transform (Fx )−1 with respect to variables
n = e3 to the fault plane x3 = 0. The adjoint field (s1 ,s2 ) images of (x1 , x2 ) and an inverse time Fourier
is parameterized by two variables s = (s1 , s2 , 0) and q transform, (Ft )−1 with respect to q, with Ft defined as
Conservation laws, duality and symmetry loss in solid mechanics 171
. ∞
Ft [[u]](x, q) = 0 [[u(x, t)]] exp(iqt)dt Bui HD (1984) A path-independent integral for mixed modes
of fracture in linear thermo-elasticity. In: IUTAM sympo-
⊥ 1
(Ft )−1 (Fx )−1
sium on fundamental of deformation and fracture. Sheffield,
div u = p 597, April 1984
2µ
Bui HD (1992) On the variational boundary integral equations in
×R ud , Td ; w(s,q) , T w(s,q) elastodynamics with the use of conjugate functions. J Elast
−1/2 28:247
× |s|2 − (q + iη)2 /c21 (η → 0+ ) Bui HD (1993) Detection de fissure par une méthode
géométrique, In: Horowitch J, Lions JL (eds) A propos
(19) des grands Systèmes des Sciences et de la Technologie.
Masson, Paris
Equation 19 explicitly solves the earthquake inverse Bui HD (1994) Inverse problems in the mechanics of materials:
problem for determining the history of the fault, since an introduction. CRC Press, Boca Raton
(t) = supp{div([[u⊥ ]])}. Bui HD (2006) Fracture mechanics: inverse problems and solu-
tions. Springer
Bui HD, Maigre H (1988) Extraction of stress intensity factors
8 Concluding remarks from global mechanical quantities. C R Acad Sci Paris
306(II):1213
Bui HD, Proix JM (1984) Lois de conservation en thermoélas-
I would like to mention first that duality is a very old ticité linéaire. C R Acad Sci Paris 298(II):325
philosophical principle in Asia. Duality is synonym of Bui HD, Ehrlacher A, Nguyen QS (1980) Crack propagation in
parallelism, or complementary things, sometimes an coupled thermo-elasticity. J Meca 19:697. Gauthier-Villars,
Paris
opposition between things: Yin and Yang in China, Bui HD, Maigre H, Rittel D (1992) A new approach to the experi-
Am and Duong (the Vietnamese words for Female and mental determination of the dynamic stress intensity factors.
Male, respectively), Positive and Negative, the Sky and Int J Solids Struct 29:2881–2895
the Earth, Water and Fire etc. In Sciences, duality is Bui HD, Constantinescu A, Maigre H (1999) Inverse scattering
of a planar crack in 3D acoustics: closed form solution for
found in Solid Mechanics, X-rays Tomography and a bounded solid. C R Acad Sci Paris 327(II):971–976
Mathematics (Schwartz 1978) etc., see Table 1. Bui HD, Constantinescu A, Maigre H (2005a) An exact inver-
sion formula for determining a planar fault from boundary
measurements. Inv Ill-posed Probl 13(6):553–565
Bui HD, Constantinescu A, Maigre H (2005b) The reciprocity
References
gap functional for identifying defects and cracks. In: Mroz
Z, Stavroulakis GE (eds) In Parameter identification of
Achenbach JD, Bazant ZP (1972) Elastodynamic near-tip stress materials and structures. CISM course and lecture, vol 469.
and displacement fields for rapidly propagating cracks in Springer, Wien, New York
orthotropic materials. J Appl Mech 97:183 Calderon AP (1980) On an inverse boundary problem. In: Meyer
Aki K, Richards PG (1980) Quantitative seismology, theory & WH, Raupp MA (eds) Seminar on numerical and applica-
methods, vol 1. W.H. Freeman and Cie, New York tions in continuum physics. Brazilian Mathematical Soci-
Andrieux S, Ben Abda A (1992) Identification de fissures planes ety, Rio de Janeiro, pp 65–73
par une donnée de bord unique: un procédé direct de locali- Fletcher DC (1976) Conservation laws in linear elastodynamics.
sation et d’identification. C R Acad Sci Paris 315(I):1323– Arch Rat Mech Anal 60:329
1328 Germain P (1973) The method of virtual power in continuum
Andrieux S, Ben Abda A (1996) Identification of planar cracks mechanics. Part II. In: Domingos JJD et al (eds) Applica-
by complete overdetermined data inversion formulae. tions to continuum thermodynamics. J. Wiley, New York,
Inverse Probl 12:553–563 pp 317–333
Andrieux S, Bui HD, Ben Abda A (1999) Reciprocity and crack Germain P (1978) Duality and convection in continuum mechan-
identification. Inverse Probl 15:59–65 ics. In: Fichera G (ed) Trends in applications to mechanics.
Ben Abda A, Bui HD (2003) Planar cracks identification for the Pitman, London, pp 107–128
transient heat equation. Inverse Ill-posed Probl 11(1):67–86 Halphen B, Nguyen QS (1975) Sur les matériaux standards
Bui HD (1965) Dissipation of energy in plasticity. Cahier généralisés. J Méca 14:39
Groupe Français de Rhéologie 1:15 Ikehata M (1998) Inversion for the linearized problem for an
Bui HD (1967) Transformation des données aux limites relatives inverse boundary value problem in elastic prospection.
au demi-plan élastique. C R Acad Sci Paris 265:862–865 SIAM J Math Anal 50(6):1635
Bui HD (1974) Dual path-independent integrals in the boundary- Lubliner J (1974) A simple theory of plasticity. Int J Solids Struct
value problems of cracks. Eng Fract Mech 6:287–296 10:310
Bui HD (1977) Stress and crack displacement intensity factors Mandel J (1965) Energie élastique et travail dissipé dans les
in elastodynamics. In: Proceedings of 4th international con- modèles rhéologiques. Cahier Groupe Français de Rhéolo-
ference fracture, vol 3. Waterloo, p 91 gie 1:1
172 H. D. Bui
Moreau JJ (1963) Fonctionnelles sous différentiables, C R Acad Rockafellar RT (1966) Characterization of the subdifferentials
Sci Paris 257:4117–4119 of convex functions. Pac J Math 17:497–510
Moreau JJ (1966) Fonctionnelles convexes. Séminaire Equations Schwartz L (1978) Théorie des distributions. Hermann, Paris
aux Dérivées Partielles. Collège de France Tikhonov A, Arsenine V (1977) Méthode de résolution de prob-
Nguyen QS, Bui HD (1974) Sur les matéraux à écrouissage posi- lèmes mal posés. Editions Mir, Moscou
tif ou négatif. J Meca 13(2):321–342 Tonti E (1975) On the formal structure of physical theories.
Rice JR (1968) Mathematical analysis in the Mechanics of Frac- Cooperative Library Instituto di Polytechnico di Milano,
ture. In: Liebowitz H (ed) Fracture. Academic Press, p 191 Milano
Phase field simulation of domain structures in ferroelectric
materials within the context of inhomogeneity evolution
Ralf Müller · Dietmar Gross · David Schrade ·
B. X. Xu
Abstract A phase field model for simulating the vice versa. The working principle is based on the atomic
domain structures in ferroelectric materials is proposed. structure of these materials. Many piezoelectric mate-
It takes mechanical and electric fields into account, rials have a so-called Perovskite structure. At a criti-
thus allowing for switching processes due to mechan- cal temperature, the Curie temperature, a spontaneous
ical and/or electrical loads. The central idea of the break of the symmetric distribution of charge carriers
model is to take the spontaneous polarisation as an in unit cells occurs. This leads to a spontaneous po-
order parameter and to provide an evolution law for this larisation, which is responsible for the multifunctional
parameter. The concept of evolving inhomogeneities material behavior.
(configuratioanl forces) can be used in this context, On the microscopic level the spontaneous polar-
as the spatial distribution of the spontaneous isation arranges in areas in which it is almost con-
polarisation describes the inhomogeneity of the sys- stant. These areas are termed domains. In between these
tem. The evolution is found to be in agreement with domains interfaces form, which are called domain walls.
the second law of thermodynamics and to resemble the Due to the crystal structure a single crystal contains
(classical) Ginzburg-Landau equation. Numerical sim- 90 and 180◦ domain walls, according to the change of
ulations show the features of the model and the inter- the polarisation across a domain wall. Under external
action of domain structures with defects. loads of mechanical or electric character these domain
structures change, and cause a rather complex micro-
Keywords Phase field · Configurational mechanics · structural evolution, which influences the meso- and
Micromechanics · Ferroelectrics · Finite element macroscopic material behaviour. The macroscopic
method material properties are often modeled by ferroelectric
material models, which resemble plasticity models. The
main shortcoming of these models is that they take the
1 Introduction microstructure only in an averaged sense into account,
neglecting the fine scale structures of the domains. The
Ferroelectric materials are widely used in sensor and intention of this investigation is to contribute to the
actuator applications. These materials allow for a con- simulation of these ferroelectric materials on the mes-
version of electric signals into mechanical output and oscopic level of domains. In the present approach a
phase field model based on continuum physics is pro-
R. Müller (B) · D. Gross · D. Schrade · B. X. Xu
Civil Engineering and Geodesy, TU Darmstadt,
posed and developed. The phase field model, also some-
Hochschulstr.1, 64289 Darmstadt, Germany times termed order parameter model, is derived and
e-mail: r.mueller@mechanik.tu-darmstadt.de analysed using thermodynamic arguments within a
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 173
doi: 10.1007/978-1-4020-6929-1_16, © Springer Science+Business Media B.V. 2007
174 R. Müller et al.
configurational mechanics approach of evolving inho- To close the system constitutive relations are needed.
mogeneities. Similar approaches can be found in mod- For a piezoelectric material usually a linear relation
elling of martensitic phase transitions, see for example between σ and E, ε as well as D and E, ε are pos-
McCormack et al. (1992), Müller (1998), Wang et al. tulated. These material laws neglect the spontaneous
(1993). In the context of ferroelectric materials phase polarisation P, which must to be considered on the
field models can be found in Wang et al. (2004), Wang mesoscopic level of domains.
and Zhang (2006a, b), Zhang and Bhattacharya (2005a,
b), Bhattacharya and Ravichandran (2003), Soh et al.
2.2 Configurational force balance
(2006). In other models the domain walls are treated
as sharp interfaces, at which jump conditions have to
Assuming that in the volume the following equation
be satisfied. Within the context of ferroelectrics and in
a variational setting of configurational forces this can ˜ + g̃ = 0
div (5)
be found in Mueller et al. (2005). A thermodynamic holds true, with appropriate boundary conditions, we
approach based on configurational forces on a point will formulate the second law of thermodynamics in
defect or a domain wall is given in Goy et al. (2006), the following way
Schrade et al. (2007). For the general theory of configu-
rational forces the reader is referred to Maugin (1993), (t ∗ · u̇ − Q∗ ϕ̇) dA + ( f · u̇ − q ϕ̇) dV
∂B
Gurtin (1996). The model presented here is solved by B
using finite element techniques. For a general treatment + (n) ˜ · Ṗ dA− Ḣ(ε, E, P, grad P) dV ≥0,
∂B B
of configurational forces in a finite element context, we
(6)
refer the reader to Mueller et al. (2002), Mueller and
Maugin (2002) and the works cited in there. Within where the work done by the forces and charges on the
the context of ferroelectrics stationary solutions are surface and in the volume are taken into account. Using
given in Su and Landis (2006, 2007). Various exam- the balance law (5) the second law can be put in the
format
ples demonstrate the main features and possible appli-
cations of the model. ∂H ∂H
σ− : ε̇ − D + · Ė
B ∂ε ∂E
˜ ∂H ˙ ∂H
2 Phase field model + − : grad P− g̃+ · Ṗ dV
∂grad P ∂P
≥ 0. (7)
2.1 Standard balance law
From this we can deduce the following constitutive
We will assume quasi-stationary conditions for both the relations
mechanical and electric fields, thus the stress σ satisfies ∂H ∂H ˜ = ∂H .
σ = , D=− , (8)
an equilibrium condition and the electric displacement ∂ε ∂E ∂grad P
D obeys Gauß’ law It is worth mentioning, that besides the classical fluxes,
i.e. stress and electric displacement, the polarisation
divσ + f = 0, div D = q, (1)
forces appear as work conjugated quantities to the order
where f represents a volume force and q a charge den- parameter P and thus become ‘constitutively’ related to
sity. In addition there might be boundary conditions by P. The derivation presented here relies on ideas given
tractions and surface charges, given by in Gurtin (1996), which are extended to a vector valued
σ n = t∗ on ∂Bt , D · n = −Q∗ on ∂Bq . (2) order parameter P, see also Su and Landis (2007).
The strains ε and the electric field ϕ are defined by In order to ensure a non-negative dissipation, the fol-
1 lowing relation has to be fulfilled in all processes,
ε= gradu + gradT u , E = −gradϕ, (3) ∂H ∂H
2 − + g̃ Ṗ = − ˜
− div Ṗ
where u is the displacement and ϕ the electric potential. ∂P ∂P
These fields are associated with appropriate boundary ∂H ∂H
conditions, as =− − div Ṗ
∂P ∂grad P
u = u∗ on ∂Bu , ϕ = ϕ ∗ on ∂Bϕ . (4) ≥ 0, (9)
Phase field simulation of domain structures 175
4
3
2.3 Phase field potential 2
1
The potential H of the presented phase field model 0
0.5
consists of three contributions: 0.5
0
H =H ent
+H sep
+H int
. (10) 0
-0.5 -0.5
Here the ‘classical’ electric enthalpy is given by
1 Fig. 1 Domain separation part in phase field potential with
H ent = (ε − ε 0 ) : [C(ε − ε 0 )] − (ε − ε 0 ) minima at P = (±P0 , 0) and P = (0, ±P0 )
2
1
: T E − E · AE − P · E, (11)
2 and takes variations in the polarisation into account.
where C is the elastic tensor, is the piezo-electric Due to the thermodynamic analysis in the previous sub-
and A the dielectric tensor. The spontaneous strain and section, the following constitutive relations now can be
polarisation are denoted by ε0 and P. It is crucial and derived
physically senseful to let the spontaneous strain and ∂H
the piezo-electric tensor depend on the spontaneous σ = = C(ε − ε 0 ) − T E.
∂ε
polarisation. This is done by ∂H
D=− = (ε − ε 0 ) + AE + P. (16)
3 | P| 1 ∂E
ε ( P) = ε0
0
e⊗e− 1 , (12)
2 P0 3 The evolution equation for the polarisation P is chosen
where e is the direction of P, i.e. e = P/| P|. For in agreement with the thermodynamic consideration
the piezo-electric coupling tensor the dependency is using configurational forces.
given in index notation by
δH ∂H ∂H
| P| Ṗ = −M = −M − div
bkij = b ei ej ek + b⊥ (δij − ei ej )ek δP ∂P ∂grad P
P0
∂H ent ∂H sep
1 = −M + − λ P , (17)
+ b= [(δki −ek ei )ej +(δkj −ek ej )ei ] . (13) ∂P ∂P
2
where M is the mobility parameter and δH /δ P the
Details on this can also be found in Kamlah (2001). This variational derivative. This equation can be understood
introduces the material parameters ε 0 , P0 , b , b⊥ , b= . as the Ginzburg-Landau equation in order parameter
The separation energy forces the system to assume models. The three terms in (17) describe the evolution
poled states in the absence of external loads. It is mod- of the order parameter P due to the electro-mechani-
elled by a forth order polynomial of the form cal energy, due to the separation energy and due to a
H sep = a1 + a2 (P12 + P22 ) + a3 (P14 + P22 ) smoothing term. The parameter λ introduces the energy
associated with areas of varying order parameter, i.e.
+ a4 P12 P22 , (14)
domain walls. It can also be thought of as an internal
which is applicable to a system with 90 and 180◦ length scale, which is introduced in the model. Eq. (17)
domains in the x1 − x2 plane. A sketch of H sep for is assumed in many theories with non-local internal
this 2d setting is depicted in Fig. 1. variables or order parameters. Within the context of
The ‘domain wall energy’ H int is given by ferroelectric materials (17) can be understood as a spe-
1 cial realisation of the theories proposed in Maugin and
H int = λ grad P 2 , (15) Pouget (1980), Maugin (1988).
2
123
176 R. Müller et al.
The numerical implementation is done using a finite In all simulations we assume that there is no flux of
element formulation. For simplicity and to keep the polarisation across the boundary of the simulation, i.e.
numerical effort limited we restrict ourselves to 2d homogeneous Neumann boundary conditions are
problems. The discretisation is done by 4-node bi-lin- assumed for the phase field parameter P. The bound-
ear plane elements. Details on this can be found in any ary conditions on P arise as a mathematical neces-
standard text book on finite elements, without claim sity from the introduction of higher gradients on P to
of completeness we cite Hughes (2000). The degrees model domain walls. Using Voigt notation the material
of freedom at each node I are the two displacements data are chosen as, which approximates a ferroelectric
uI1 , uI2 , the electric potential ϕ I and the spontaneous material.
⎡ ⎤
polarisation (order parameter) P1I , P2I . Thus each node 12.0 7.5 0.0
has five degrees of freedom, C = ⎣ 7.5 12.0 0.0⎦ · 1010 N/m2 ,
T 0.0 0.0 2.6
dI = uI1 uI2 ϕ I P1I P2I . (18)
0.0 0.0 13.0
= C/m2 ,
−5.2 15.0 0.0
The discretisation leads to a set of non-linear equations
(residual) which depend on the nodal values and the 6.0 0.0
A= · 10−9 C/Vm, P0 = 0.4C/m2 ,
rates of the nodal values, i.e. 0.0 6.0
J ε0 = 0.003
RI (dJ , ḋ ) = 0. (19)
M = 10A/Vm , λ = 10−6 Vm3 C. (23)
Discretising with a first order difference formula The coefficients a1 , . . . a4 in the separation part are
J 1 J chosen as
ḋ ≈ dn+1 − dJn , (20)
t a1 = 107 N/m2 , a2 = −12.5 · 107 Nm2 /C2 ,
where n+1 and n denote the values at time tn+1 and tn in a3 = 39.0625 · 107 Nm2 /C4 ,
conjunction with an evaluation of the residual at time a4 = 234.375 · 107 Nm2 /C6 . (24)
tn+1 yields an implicit time integration scheme, where
This ensures a potential minimum at P = (0, ±P0 )
dJn+1 − dJn and at P = (±P0 , 0).
I I J
Rn+1 = R dn+1 , = 0. (21)
t
4.1 Microstructure evolution—self organisation
Due to the non-linear character of the phase field model,
this establishes a non-lienear system of equations for
Starting with a random distribution of initial polarisa-
the unknown nodal values dJn+1 . A Newton iteration
tions the development of subsequent microstructures is
is used to solve this system. For the numerical perfor-
calculated. As we assume charge free boundaries, i.e.
mance a consistent tangent matrix has to be derived. It
D · n = 0, it is energetically very unfavourable to have
has the following structure:
polarisations with a normal component to the boundary.
1 IJ These components would require very strong electric
SIJ = KIJ + D where
t fields to satisfy the boundary conditions. Therefore the
∂RI ∂RI system starts to arrange in such a manner, that it forms
KIJ = , D IJ
= . (22) two vortices, with eight 90◦ domain walls. In Fig. 2 the
∂dJ ∂ ḋ
J
evolution and self organisation of the system is shown.
This establishes a robust and fast algorithm, which Due to non aligned polarisation vectors, strong internal
allows for sufficiently large time steps. The iteration fields develop and a rapid self organisation to structures
matrix SIJ turns out to be symmetric, which allows with larger domains is the consequence. From a numer-
for an efficient storage and solution. The symmetry of ical point of view this initial period is very difficult and
SI J is due to the fact that all constitutive laws and the crucial with respect to the stability of the integration
evolution law are derived from a potential. scheme.
Phase field simulation of domain structures 177
Fig. 2 Evolution of
polarisation in absence of
external loads or fields (a)
random initial
configuration, going
through states (b) and (c),
ending at (d) equilibrium
state
Fig. 4 Domain wall velocity as a function of the applied electric compressive load, respectively. These are the states the
field E
system would switch back and fourth under continued
cyclic loading. If D · n = 0 on the left and right bound-
Fig. 6. In the final configuration the polarisation has ary, the system behaves a lot stiffer when compressive
switched completely. In Schrade et al. (2007) a sharp loading is applied. This situation is illustrated in Fig. 8b,
interface model was used to study this kind of defect d, f where multiple horizontal domains form.
for gadolinium molybdate. There, domain wall pinning
was observed for low fields, whereas for high fields the
domain wall could be moved over the defect.
5 Summary
4.4 Mechanical poling A continuum phase field model has been established.
Using the concept of an order parameter in conjunction
Besides electrically induced domain switching the with the evolving inhomogeneities (configurational
model is also capable of reproducing ferroelastic forces), a thermodynamically consistent evolution law
switching. Figure 7a shows the random initial distribu- was proposed. The polarisation balance is used to pro-
tion of the polarisation. The boundary conditions are pose an evolution law for the order parameter of the
ϕ = 0 on all edges, and bipolar normal stress load- systen, i.e. an evolution law for the spontaneous po-
ing is applied in the vertical direction. The series of larisation P, which is in agreement with the Ginz-
pictures in Fig. 7 shows the polarisation at alternating burg-Landau approach. The model takes electrical and
compressive (b, d, f) and tensile (c, e) stresses. Pic- mechanical fields into account to model the microstruc-
tures (e and f) reflect the states at maximum tensile and ture evolution in ferroelectric materials.
Central part in the numerical derivation is a finite ele- reproducing the microstructures on the domain level.
ment scheme, which is used to solve the non-linear par- Crucial for the numerical Implementation is a robust
tial differential equations. In extension to finite element time integration, which is achieved by an implicit time
formulations for piezoelectric materials, which use the integration scheme together with a Newton iteration in
mechanical displacement and the electric potential as each time step.
nodal variable, the spontaneous polarisation P is intro- The microstructures predicted by the model agree
duced. This is also different from ferroelectric mod- with simple analytical models. Effects like domain wall
els, which use the remanent polarisation as an internal pining, which are experimentally observed, can be
(Gaußpoint) variable. The present model is capable of reproduced. The effect of possible defect structures,
180 R. Müller et al.
such as point defects, can be included, but this is outside McCormack M, Khachaturyan AG, Morris JW (1992) A two-
of the scope of this paper. Future works will include the dimensional analysis of the evolution of coherent precipi-
tate in elastic media. Acta metall mater 40(2):325–336
simulation of mesoscopic material responses, i.e. the Mueller R, Maugin GA (2002) On material forces and finite ele-
overall properties of single crystals with domain struc- ment discretizations. Comp Mech 29(1):52–60
ture evolution during poling, as well as poly crystalline Mueller R, Kolling St, Gross D (2002) On configurational forces
structures. The incorporation of defects (point defects) in the context of the finite element method. Int J Numer
Methods Eng 53:1557–1574
is necessary to describe effects like fatigue and ageing Mueller R, Gross D, Lupascu D (2005) Driving forces on domain
and will be pursued in future works. walls in ferroelectric materials and interaction with defects.
Comp Mat Sci 35:42–52
Müller WH (1998) Zur Simulation des Mikroverhaltens thermo-
References mechanisch fehlgepasster Verbundwerkstoffe. Fortschritt-
Berichte VDI, Reihe 18, Nr. 234, Düsseldorf
Schrade D, Mueller R, Gross D, Utschig T, Shur V, Lupascu
Ahluwalja R, Cao W (2000) Influence of dipolar defects on D (2007) Interaction of domain walls with defects in ferro-
switching behavior in ferroelectrics. Phys Rev B 63:012103 electric materials. Mech Mater 39:161–174
Ahluwalja R, Cao W (2001) Size dependence of domain pat- Soh A, Song Y, Ni Y (2006) Phase field simualtions of hystere-
terns in a constrained ferroelectric system. J Appl Phys sis and butterfly loops in ferroelectrics subjected to electro-
89(12):8105–8109 mechanical coupled loading. J Am Ceram Soc 89:652–661
Bhattacharya K, Ravichandran G (2003) Ferroelectric perovsk- Su Y, Landis C (2006) A non-equlibrium thermodynamics frame-
ites for electromechanical actuation. Acta Mater 51:5941– work for domain evolution; phase field models and finite
5960 element implementation. Proceeding to the SPIE
Cao W, Cross L (1991) Theory of tetragonal twin structures in Su Y, Landis C (2007) Continuum thermodynamics of ferroelec-
ferroelectric perovskites with a first-order phase transition. tric domain evolution: Theory, finite element implementa-
Phys Rev B 44(1):5–12 tion, and application to domain wall pinning. J Mech Phys
Flippen R (1975) Domain wall dynamics in ferroelectric/ferro- Sol 55:280–305
elastic molybdates. J Appl Phys 46(3):1068–1071 Wang J, Zhang T-Y (2006a) Effect of long-range elastic interac-
Gross D, Kolling S, Mueller R, Schmidt I (2003) Configurational tions on the toroidal moment of polarisation in a ferroelec-
forces and their application in solid mechanics. Eur J Mech tric nanoparticle. Appl Phys Lett 88:182904
A/Solids 22:669–692 Wang J, Zhang T-Y (2006b) Size effects in epitaxial ferroelectric
Goy O, Mueller R, Gross D (2006) Interaction of point defects in islands and thin films. Phys Rev B 73:144107
piezoelectric materials—numerical simulations in the con- Wang Y, Chen L, Khachaturyan AG (1993) Kinetics of strain-
text of electric fatigue. J Thero Appl Mech 44(4):819–836 induced morphological transformation in cubic alloys with
Gurtin ME (1996) Generalized Ginzburg-Landau and Cahn- a miscibility gap. Acta metall mater 41(1):279–296
Hillirad equations based on a microforce balance. Physica Wang J, Shi S-Q, Chen L-Q, Li Y, Zhang T-Y (2004) Phase field
D 92:178–192 simulations of ferroelectric/ferroelastic polarisation switch-
Hughes T (2000) The finite element method. Dover, Mineola, ing. Acta mater 52:749–764
New York Xiao Y, Shenoy V, Bhattacharya K (2005) Depletion layers and
Kamlah M (2001) Ferroelectric and ferroelastic piezoceram- domain walls in semiconducting ferroelectric thin films.
ics—modeling of electromechanical hysteresis phenomena. Phys Rev Lett 95:247603
Continuum Mech Thermodyn 13:219–268 Zhang W, Bhattacharya K (2005a) A computational model of
Maugin GA (1988) Continuum mechanics of electromagnetic ferroelectric domains. Part I: model formulation and domain
solids. North-Holland, Amsterdam switching. Acta mater 53:185–198
Maugin GA (1993) Material inhomogeneities in elasticity. Zhang W, Bhattacharya K (2005b) A computational model of
Chapman & Hall, London, Glasgow, New York, Tokyo, ferroelectric domains. Part II: grain boundaries and defect
Melbourne, Madras pinning. Acta mater 53:199–209
Maugin GA, Pouget J (1980) Electroacoustic equations for one-
domain ferroelectric bodies. J Acoust Soc Am 68(2):575–
587
An adaptive singular finite element in nonlinear fracture
mechanics
Ralf Denzer · Michael Scherer · Paul Steinmann
Abstract In the case of nonlinear fracture mechanics mulation in the vicinity of the singular points. This
the type of singularity induced by the crack tip is com- is caused by the fact that in finite element formula-
monly not known. This results in a poor approximation tions typically Lagrangian or serendipity polynomi-
of the near crack tip fields in a finite element setting als are used as shape function N n (ξ ). It is obvious
and induces so called spurious—or residual—discrete that polynomials are not adequate to represent singular
material forces in the vicinity of the crack tip. Thus behaviours. To overcome this problem different spe-
the numerical calculation of the crack driving mate- cial finite elements with singular behaviour were intro-
rial force in nonlinear fracture is often not that precise duced in the literature. Some of them are restricted
√
as in linear elasticity where we can use special crack to typical stress singularities like 1/ r or 1/r which
tip elements and/or path independency. To overcome occur in the case of cracks in linear elastic or perfect
this problem we propose an adaptive singular element, plastic materials, respectively, see e.g. Barsoum (1977).
which adapts automatically to the type of singularity. Other special finite elements have a variable, i.e. a user
The adaption is based on an optimisation procedure given, type of singularity, like Akin (1976), Tracey and
using a variational principle. Cook (1977), Stern (1978), Hughes and Akin (1980),
Staab (1983) and Lim and Kim (1994). One of the
Keywords Singular finite element · Singularity essential problems while using these elements is the
computation · Nonlinear fracture determination of the correct type of singularity of the
given problem, which usually needs at least an asymp-
totic analytical solution of the problem at the position
1 Introduction of the singularity. Especially in the case of nonlinear
material behaviour and/or large strain analysis there
Problems with singular behaviour in the stress and/or is only a rare number of analytical solutions available
strain fields at certain points, e.g. re-entrant corners or in the literature, see e.g. Rice (1968), Knowles and
cracks, lead often to large inaccuracies of the numeri- Sternberg (1973, 1983), Herrmann (1989, 1992) and Le
cal field approximations based on a finite element for- and Stumpf (1993). We therefore propose in this work
an optimisation procedure based on a variational prin-
R. Denzer (B) · M. Scherer · P. Steinmann ciple to determine the type of singularity numerically
Applied Mechanis, University of Kaiserslautern,
67663 Kaiserslautern, Germany which minimises the (discrete) total potential energy
e-mail: denzer@rhrk.uni-kl.de of a system.
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 181
doi: 10.1007/978-1-4020-6929-1_17, © Springer Science+Business Media B.V. 2007
182 R. Denzer et al.
2 Spatial motion problem spatial motion Piola and Cauchy stresses, have been
introduced here. For the present case of a conserva-
2.1 Kinematics and kinetics tive mechanical system, they follow from the potential
energy density as
To set the stage and in order to introduce terminology
P = ∂ F U0 ⇒ σ = j P · F t = Ut I − f t · ∂ f Ut
and notation, we briefly reiterate some key issues per-
taining to the geometrically nonlinear kinematics and (5)
kinetics of the quasi-static spatial motion problem. The second expression in Eq. 5 thereby denotes the
Thereby, in order to introduce the relevant concepts, energy-momentum format of the spatial motion Cau-
we merely consider a conservative mechanical system. chy stress. For the sake of conciseness and without dan-
In this case, the internal potential energy density Wτ per ger of confusion, we omit the explicit indication of the
unit volume in B τ with τ = 0, t characterises the hyper- spatial or material parametrisation.
elastic material response and is commonly denoted as Moreover, distributed volume forces bτ per unit vol-
stored energy density. Moreover, an external potential ume follow from the explicit spatial gradient of the total
energy density Vτ characterises the conservative load- potential energy density
ing. Then, the conservative mechanical system is essen-
tially characterised by the total potential energy density b0 = −∂ϕ U0 ⇒ bt = j b0 (6)
per unit volume Uτ = Wτ + Vτ .
In the spatial motion problem in Fig. 1, the place-
2.2 Virtual work and discretisation
ment x of a ‘physical particle’ in the spatial configu-
ration Bt is described by the nonlinear spatial motion
The pointwise statement in Eq. 4 for the solution of the
deformation map
spatial motion problem is multiplied by a test function
x = ϕ(X) (1) (spatial virtual displacement) w under the necessary
in terms of the placement X of the same ‘physical parti- smoothness and boundary assumptions to render the
cle’ in the material configuration B0 . The spatial motion virtual work expression
deformation gradient, i.e. the linear tangent map associ-
ated with the spatial motion deformation map, together w · σ · n da = ∇x w : σ dv
∂ Bt Bt
with its determinant are then given by
wsur wint
F = ∇X ϕ(X) and J = det F (2)
− w · bt dv ∀ w (7)
and its inverse by Bt
f = F −1 and j = det f (3) wvol
which will be discussed more precisely later. whereby wsur
denotes the spatial variation of the total
The quasi-static balance of momentum for the spa- bulk potential energy due to its complete dependence
tial motion problem reads on the spatial position, whereas the contributions wint
− Div P = b0 ⇒ − div σ = bt (4) and wvol denote the spatial variations of the total bulk
potential energy due to its implicit and explicit depen-
The two-point description stress P and the spatial
dence on the spatial position, respectively.
description stress σ , see Fig. 1, which are called the
Obviously, the quasi-static equilibrium of spatial
forces is recovered, if arbitrary uniform spatial virtual
displacements w = θ = const. are selected for the
evaluation of Eq. 7.
θ· σ · n da + bt dv = 0 ∀ θ (8)
∂ Bt Bt
The domain is next discretised in n el elements with
B0h = ∪ne=1
el
B0e and Bth = ∪ne=1
el
Bte . The geometry in Bt
Fig. 1 Kinematics and kinetics of the spatial motion problem and B0 is interpolated from the positions ϕ n and X n of
An adaptive singular finite element in nonlinear fracture mechanics 183
n el
the n en nodes by shape functions N n (ξ ) on each ele- where by
ment, with n ∈ [1, n en ] denoting the local node num-
A denotes the assembly operator for all
e=1
bering n el finite elements. In conclusion of these consider-
ations, the discrete spatial node point (surface) forces
n en
ϕ h |B2e = N n (ξ )ϕ n and are thus energetically conjugated to variations of the
n=1 spatial node point positions.
n en
X h |B2e = N n (ξ )X n (9)
n=1 3 Variable power singular element
whereby we introduced the isoparametric domain
2 = [−1, 1]n dim with n dim the dimension of the prob- In the sequel we will follow the generation of a 2D tri-
lem. angular element introduced by Hughes and Akin (1980)
Thus, the elementwise discretisation of the virtual with a variable type of singularity. In the next section
spatial displacement field w into nodal values w n which we then formulate a numerical scheme based on a vari-
are interpolated as well by the shape functions N n in ational principle to compute the type of singularity for
the spirit of an isoparametric expansion, renders the a discretised problem.
representation We start with a one-dimensional 3-node singular ele-
ment, as depicted in Fig. 2a, with the following shape
n en
functions:
w |
h
B2
e = N n (ξ )wn (10)
κ
n=1 r κ − 2 21 r
Corresponding gradients of the virtual spatial dis- B 1 (r, κ) = 1 − 2r +
κ
placement field are given in each element by 1 − 2 21
⎡
κ ⎤
n en
r κ − 2 21 r
∇X w h |B2e = w n ⊗ ∇X N n (ξ ) (11) ⎢ ⎥
B 2 (r, κ) = 2r − 2 ⎣
κ ⎦ (17)
n=1 1 − 2 21
n en
κ
∇x w h |B2e = w n ⊗ ∇x N n (ξ ) (12) r κ − 2 21 r
n=1 B 3 (r, κ) =
κ
Lastly, based on the above discretisations, the cor- 1 − 2 21
responding deformation gradient F takes the element-
whereby r ∈ [0, 1]. Please note that these shape func-
wise format
tions are capable to represent a point singularity of
n en
order r κ . Even though this one-dimensional element
F |B2 =
h e
ϕ n ⊗ ∇X N n (ξ ) (13)
n
has little practical interest, it is possible to construct a
two-dimensional 9-node element, see Fig. 2b, by a ten-
The elementwise expansions for the internal and the
sorial expansion Ñ i (r, s, κ) = B j C k with a standard
volume contributions therefore read
3-node element C i (s) = B i (r = s, κ = 2), i.e. a 2nd
n en
order Lagrange polynomial. The corresponding shape
wint
e = w n · σ · ∇x N n dv (14)
Bte functions
n=1
n en
wvol Ñ 1 = B 1 C 1 Ñ 4 = B 1 C 3 Ñ 7 = B 2 C 3
e = wn · bt N n dv (15)
Bte
n=1 Ñ 2 = B 3 C 1 Ñ 5 = B 2 C 1 Ñ 8 = B 1 C 2 (18)
Finally, considering the arbitrariness of the spatial Ñ = B C
3 3 3
Ñ = B C
6 3 2
Ñ = B C
9 2 2
virtual node point displacements w n , the global dis-
crete spatial node point forces characterising external are capable of representing a line singularity of the
spatial surface loads are computed as order r κ exactly. By degenerating this 9-node quad-
n en
n el rilateral element with
fsur
h
= A σ · ∇x N n − bt N n dv
e=1 n=1 Bt
e
(16)
N 1 (r, s, κ) = Ñ 1 + Ñ 4 + Ñ 8 (19)
184 R. Denzer et al.
Fig. 2 Generation of a s
7-node triangular singular
finite element. (a) 1D 4 7 3
3-noded element; (b) 2D 3
9-noded element; (c) 6
7-noded triangular element
8 9 6
7 5
1
4
1 2 3 r 1 5 2 r 2
a b c
and renumbering the rest of the nodes appropriately, Hereby the admissible variations Dδ ϕ are given by
ϕ
we end up with a 7-node triangular element, Fig. 2c, ∇X Dδ ϕ = Dδ F in B0 and Dδ ϕ = 0 on ∂B0
which captures a point singularity of the order r κ and
is therefore an appropriate finite element for e.g. crack
problems. 4.2 Discrete Dirichlet functional
If we use these elements in the vicinity of a crack
tip, then the obvious numerical representation of the If we reformulate the Dirichlet functional with the dis-
displacement field and e.g. a linearised strain is crete global spatial deformation map
7 n np
u(r, s, κ) = N i (r, s, κ)ui (20) ϕ (X, ϕ n , κ) =
h
N n (X, κ)ϕ n (κ) (24)
i=1 n=1
i.e. u ∝ r κ and ε ∝ r [κ−1] . (21) whereby n np is the total number of node points in the
Now the problem arises how to determine κ for non- discretised system, we get the discrete version of the
linear problems, where we do not know the value of κ Dirichlet functional
in advance by e.g. an analytical solution. A solution
h = h (ϕ n (κ), κ) → min (25)
will be proposed in the next section.
Hereby, we use in the vicinity of a singular point,
i.e. a crack tip, the previously described singular ele-
4 Adaptive singular elements ments with variable power r κ . This functional depends
explicitly and implicitly, via ϕ i on the unknown singu-
4.1 Dirichlet principle larity type r κ . We now propose, that the type of singu-
larity adjusts in such a way, that it minimises the total
By means of the total potential energy of a continuous potential energy of the system. Therefore we are able
system to formulate the necessary condition to minimise h
(ϕ) = W0 − ϕ · b0 dV with respect to κ by
B0
∂h ∂h ∂ϕ n ∂h
p = · + =0 (26)
− ϕ · t 0 dA → min (22) ∂κ impl ∂ϕ n expl ∂κ ∂κ expl
∂ B0t
ϕ If we use a staggered scheme for the minimisation,
and Dirichlet boundary condition ϕ = ϕ p on ∂B0 the
we can incorporate the spatial equilibrium condition
state of static equilibrium could be expressed as the
minimum of this Dirichlet functional. If we restrict our- ∂h
rn = =0 (27)
selves to convex problems, a necessary condition for ∂ϕ n expl
the minimum
is
and end up with a minimisation problem on the spatial
Dδ = Dδ F : ∂ F W0 − Dδ ϕ · b0 dV equilibrium surface
B0
∂h ∂h
−
p
Dδ ϕ · t 0 dA = 0 (23) rκ = = =0 (28)
∂ B0t
∂κ impl ∂κ expl
An adaptive singular finite element in nonlinear fracture mechanics 185
For simplicity we restrict ourselves to problems without For a compressible neo-Hookean material we assume
body forces, i.e. b0 = 0, and traction free boundaries, a stored energy density as
p
i.e. t 0 = 0. Therefore the only loads applied on the
ϕ
1 h
system are Dirichlet boundary conditions ϕ p on ∂B0 . W0h (F h (κ), κ) = µ I b(κ) − 3 − µ ln J h (κ)
2
This reduces our discrete Dirichlet functional to
1
2
+ λ ln J h (κ) (35)
h (ϕ h ) = W0 (F h (κ)) dV → min (29) 2
B0 Here µ and λ are the Lame parameters, I b is the first
Incorporating the discrete global deformation map invariant, i.e. the trace, of the Finger tensor b = F · F t
from Eq. 24 leads to and J the determinant of the deformation gradient F.
h (ϕ n , κ) In Eq. 34 we need the Piola stress tensor which reads
n qp
n el
in this case
= A
e=1 q=1
W0 (ϕ n , κ) det J h (ϕ n , κ)wq (30)
P = [λ ln J − µ]F −t + µF (36)
whereby we approximate the integral over the body B0 as well as the derivative of the (discrete) deformation
as a numerical integration scheme with n qp quadrature gradient
points in each element with weights wq and the Jaco-
n en
bian F h (κ) = ϕ n ⊗ ∇X N n (κ) (37)
∂ X nh
n en n=1
J h (κ) = = X n ⊗ ∇ξ N n (κ) (31) which is calculated as
∂ξ
n=1
∂ Fh n en ∂ ∇ξ N n
∂h ∂ W0h
∂κ
=
e=1
A q=1
∂κ
det J h (κ) wq (32) 5.2 Linear elasticity
by Eq. 32, of the singular elements. This leads to indef- = un ⊗ [J h ]−t (κ) · (41)
∂κ ∂κ
inite integrals with a singularity at one end point, e.g. n=1
the crack tip. Therefore we apply the so called “double Thus the format of the linearised case is very similar
exponential integration scheme” by Muhammad and to the geometrically nonlinear case which is an advan-
Mori (2003). tage for the numerical implementation.
186 R. Denzer et al.
5.3 Numerical minimisation the numerical solution coincides very good with the
analytical solution. The difference between both solu-
At this point we have all necessary information together tion is of the order 10−4 . In the case of the large strain
to minimise the discrete total potential energy of the problem, where we use the above given neo-Hookean
system, Eq. 30, reformulated as a minimisation prob- material for the V-shape notch problem, the calculated
lem on the spatial equilibrium surface, Eq. 28, with types of singularity differs from the linear elastic case.
respect to the unknown singularity type represented by But in this case an analytical solution is not known
the power r κ . This is a nonlinear one-dimensional equa- in the literature. We should mention here, that in the
tion, so as a minimisation algorithm we use a golden geometrically nonlinear case the deformation gradient
section search with parabolic interpolation as described F ∝ R [κ−1] within the complete singular finite element,
in Forsythe et al. (1977) which converges very robust whereby R is the radius in the material configuration.
but sometimes—depending on the example—slow. As Therefore the presented solution may be not directly
an alternative method we have also implemented a comparable with an asymptotic analytic solution of the
quasi-Newton BFGS method as described in Bertsekas problem, in the case it is known.
(1995) which converges often much faster.
Fig. 4 Kinematics and kinetics of the material motion problem 9 Weak form and discretisation
For the material motion problem the free energy density Fsur =
h
A Σ · ∇X N n − N n B 0 dV
e=1 n=1 B0
e
(46)
Wt = j W0 with spatial reference is expressed in terms
of the material motion deformation gradient f (or its Based on these results we advocate the Material Force
inverse F). The explicit dependence on the material Method with the notion of global discrete material node
placement is captured by the field X = Φ(x) point (surface) forces, that (in the sense of Eshelby) are
generated by variations relative to the ambient material
Wt = Wt ( f , Φ(x)) (44)
at fixed spatial positions. Such forces corresponding to
Then the familiar constitutive equations for the so- the material motion problem are trivially computable
called Eshelby stress in B0 are given as Σ = j p · f t = once the spatial motion problem has been solved.
188 R. Denzer et al.
case of the standard P2 element has a value of only (asymptotic) analytical solutions. In the case of linear
cracktip
Fsur = 0.8346. elasticity, the proposed method shows a very high accu-
The second example, shown in Fig. 5c and d, a racy in comparison to known analytic solutions and the
straight crack in a compressible neo-Hookean material precision of calculation of material forces, e.g. crack
is discussed in the sequel. The geometry of the speci- driving forces, are significantly improved. Although
men is the same as in the linear elastic case, but in this the proposed method improves also the precision of
case we applied a load of Jappl = 16. This results in numerically calculated material forces in the geomet-
much larger deformations in the vicinity of the crack rically nonlinear case, it turns out, that the order of
tip, but the overall deformations remains still small, so singularity depends on the element length and the load
that we can approximately use the linear elastic relation level.
Jappl = K I2 /E . Also in this case the adaptive singular
element performs better than the standard P2 element.
In the lower row of fig. 5c and d, where the discrete References
material forces are scaled by a factor of 20 in com-
Akin JE (1976) The generation of elements with singularities.
parison to the upper row, the numerical “noise”, the so Int J Numer Methods Eng 10:1249–1259
called spurious or residual material forces, in the case Barsoum RS (1977) Triangular quarter-point elements as elastic
of the adaptive singular element are lower in compar- and perfectly-plastic crack tip elements. Int J Numer Meth-
ison to the P2 element. The resulting (discrete) crack ods Eng 11:85–98
cracktip Bertsekas DP (1995) Nonlinear programming. Athena Scientific
driving forces are Fsur /Jappl = 0.8393 for the P2 Denzer R, Barth FJ, Steinmann P (2003) Studies in elastic frac-
cracktip
element and Fsur /Jappl = 0.8878 for the proposed ture mechanics based on the material force method. Int J
element. In this example we calculate κ = 0.6949 as Numer Methods Eng 58:1817–1835
Forsythe GE, Malcolm MA, Moler CB (1977) Computer meth-
the order of the singularity. It should be noted, that the ods for mathematical computations. Prentice-Hall
calculation of the crack driving force by a domain inte- Herrmann JM (1989) An asymptotic analysis of finite deforma-
gral method, which is in the sense of discrete material tions near the tip of an interface-crack. J Elasticity 21:227–
forces simply the sum of all discrete material forces in 269
Herrmann JM (1992) An asymptotic analysis of finite deforma-
a given subdomain including the crack tip, see Denzer tions near the tip of an interface-crack: part i. J Elasticity
cracktip
et al. (2003), the “exact” value of Fsur /Jappl = 1.0 29:203–241
is regained. We should state here once more, that the Hughes TJR, Akin JE (1980) Techniques for developing ‘spe-
cial’ finite element shape functions with particular reference
calculated order of singularity is not directly compa- to singularities. Int J Numer Methods Eng 15:733–751
rable to an asymptotic analytic solution, because κ is Knowles JK, Sternberg E (1973) An asymptotic finite-deforma-
calculated for a finite length łel of the adaptive finite tion analysis of the elastostatic field near the tip of a crack.
element. This is indicated by further computational J Elasticity 3:67–107
Knowles JK, Sternberg E (1983) Large deformations near a tip
experiments, where we observe that the numerically of an interface-crack between two neo-Hookean sheets. J
calculated order of singularity κ in the geometrically Elasticity 13:257–293
nonlinear case depends on the element length łel and Le KCh, Stumpf H (1993) The singular elastostatic field due to
the applied load level. Here, further numerical studies a crack in rubberlike materials. J Elasticity 32:183–222
Lim W-K, Kim S-C (1994) Further study to obtain a variable
are necessary. Additionally, it may be possible, that the power singularity using quadratic isoparametric elements.
singularity is not of a power type and may be calcu- Eng Fract Mech 47(2):223–228
lated by the fact that the Eshelby stress ∝ 1/R, see Maugin GA (1995) Material forces: concepts and applications.
Maugin (1995), using W0 for a specific material. Appl Mech Rev 48:213–245
Muhammad M, Mori M (2003) Double exponential formulas
for numerical indefinite integration. J Comput Appl Math
161:431–448
11 Summary Rice JR (1968) A path independent integral and the approximate
analysis of strain concentration by notches and cracks. J
We introduced an optimisation procedure based on a Appl Mech 35:379–386
variational principle to determine adaptively the order Rice JR (1974) Limitations to the small scale yielding approxi-
mation for crack tip plasticity. J Mech Phys Solids 22:17–26
of singularity in problems even in geometrically non- Staab GH (1983) A variable power singular element for analysis
linear cases, where we normally do not know the type of of fracture mechanics problems. Comput Struct 17(3):449–
singularity at re-entrant corners or cracks in advance by 457
190 R. Denzer et al.
Steinmann P (2000) Application of material forces to hyperelas- Tracey DM, Cook TS (1977) Analysis of power type singu-
tostatic fracture mechanics. i. Continuum mechanical set- larities using finite elements. Int J Numer Methods Eng
ting. Int J Solids Struct 37:7371–7391 11:1225–1233
Steinmann P, Ackermann D, Barth FJ (2001) Application of Williams ML (1952) Stress singularities resulting from various
material forces to hyperelastostatic fracture mechanics. ii. boundary conditions in angular corners of plates in exten-
Computational setting. Int J Solids Struct 38:5509–5526 sion. J Appl Mech 19:526–528
Stern M (1978) Families of consistent conforming elements with
singular derivative fields. Int J Numer Methods Eng 14:409–
421
Moving singularities in thermoelastic solids
Arkadi Berezovski · Gérard A. Maugin
Abstract The solution of the evolution problem of the example of a phase-transition front propagation in
a discontinuity requires the formulation of a kinetic a shape-memory alloy bar.
law of the progress relating the driving force and the
velocity of the singularity. In the case of a crack, the Keywords Moving discontinuity · Phase-transition
energy-release rate can be computed (in quasi-statics front · Jump relations · Kinetic relation
and in the absence of thermal and intrinsic dissipations)
by means of the celebrated J -integral of fracture that is
known to be path-independent and, therefore, provides 1 Introduction
a very convenient estimation of the driving force once
the field solution is known. However, the velocity at Phase-transition front and crack propagation are most
the crack tip remains undetermined. A similar situation known examples of moving singularities in solids. Both
holds for a displacive phase-transition front propaga- the phase-transition front and the crack represent quasi-
tion. The driving force acting on the phase boundary inhomogeneities, and, therefore, these phenomena can
can be determined, but not the velocity of the displacive be considered under one umbrella in the framework of
phase-transition front. From the thermodynamic point the canonical thermomechanical theory (Maugin 2000).
of view, both the phase transition and the crack propa- At the same time, it is well known that the material (dri-
gation are non-equilibrium processes; entropy is pro- ving) force in each case is computed differently. The
duced at the evolving discontinuity. Therefore, stress main problem in both cases is the establishment of the
jumps are determined by means of non-equilibrium kinetic relation between the driving force and the velo-
jump relations at the discontinuity. Then the kinetic city of the singularity (Abeyaratne and Knowles 2006).
relations can be obtained depending on the choice of The product of this velocity and the corresponding dri-
excess stress behavior. The procedure is illustrated on ving force determines the entropy production at the
evolving discontinuity (Maugin 2000).
Recently (Berezovski and Maugin 2007), a kine-
A. Berezovski (B) tic relation for the straight brittle crack was discussed.
Institute of Cybernetics, Tallinn University of Technology, It was shown that the equation of motion for a crack
Akadeemia tee 21, 12618 Tallinn, Estonia
following from the energy balance at the crack tip in
e-mail: Arkadi.Berezovski@cs.ioc.ee
the framework of the linear elasticity theory (Freund
G. A. Maugin
Institut Jean Le Rond d’Alembert, Université Pierre et
1990; Fineberg and Marder 1999) corresponds to the
Marie Curie, UMR 7190, 4 Place Jussieu, 75252 Paris local equilibrium approximation. A more general non-
Cedex 05, France equilibrium description allows to achieve a better
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 191
doi: 10.1007/978-1-4020-6929-1_18, © Springer Science+Business Media B.V. 2007
192 A. Berezovski, G. A. Maugin
800
In the Piola-Kirchhoff formulation (Maugin 1993),
700 the balance of mass, linear momentum, and energy with
600
no external supply of energy and body force can be
Crack velocity (m/s)
interpretation of experimental data. A typical example where t is time, ρ0 (X) is the matter density in the refe-
of the comparison of the theory (Berezovski and rence configuration, v is the physical velocity, T is the
Maugin 2007) with experimental data for the crack pro- first Piola-Kirchhoff stress, p = ρ0 (X)v(X, t) is the
pagation in Polyester/TiO2 nanocomposite (Evora et al. linear momentum, S is the entropy per unit volume,
2005) is shown in Fig. 1, where the theoretical curve H = K + E, K(v, X) = ρ0 v2 /2 is the kinetic energy
is calculated using the measured value of the limiting per unit volume in the reference configuration,
velocity VT = 724 m/s. E(F, θ ; X) is the corresponding internal energy, Q is
It should be noted that the theoretical prediction the material heat flux.
is applied only for the values of the stress intensity The second law of thermodynamics reads
factor KI > KI c , and its critical value is KI c =
0.85 MPa m1/2 (Evora et al. 2005). ∂S
+ ∇R · (Q/θ ) ≥ 0, (4)
Here a similar description is kept in the derivation ∂t X
of the kinetic relation for the phase-transition front.
We start with the material setting of thermoelasticity. where θ is the absolute temperature.
Then we discuss the jump relations at a discontinuity. The above set of Eqs. 1–4 is valid in any conti-
Due to the irreversibility of the considered process, we nuously inhomogeneous material. It is clear that the
apply non-equilibrium jump relations at the singularity existence of a discontinuity surface S breaks the sym-
(Berezovski and Maugin 2004). These jump relations metry of the problem and that, in the case of phase tran-
allow us to determine the stress jump in terms of the sition, it may be viewed as breaking the translational
driving force. Having the value of the stress jump, we invariance of the whole physical system on the mate-
are able to derive the kinetic relation in the case of a dis- rial manifold. Therefore, the presence of S manifests a
placive phase-transition front. This is demonstrated on lack of the material homogeneity for the whole system
the one-dimensional example of phase-transition front under study. Accordingly, the equation associated with
propagation in a bar. this lack of invariance must play a prominent role in
further considerations concerning S. This equation is
the balance of pseudo-momentum (Maugin 1993)
2 Material setting
∂P
− ∇R · b = f int + f inh , (5)
We express the governing equations on the material ∂t X
manifold. This gives rise to the notion of material
forces, which are made apparent through a canonical where P = −p · F is the pseudo-momentum,
projection of the thermomechanical problem onto the b = −(L1R + T · F) is the dynamical Eshelby stress
material manifold. tensor, L = K − W is the Lagrangian density, F is the
Moving singularities in thermoelastic solids 193
where square brackets denote jumps, V is the material fS = −[W ] + N · T · [F · N] , (17)
velocity, and V̄N is its normal component at the points
of the discontinuity surface S. where W is the free energy per unit reference volume
Both Eqs. 5 and 8 are nonconservative. Therefore, and . . . denotes the mean value across a discontinuity
the corresponding jump relations across S should exhi- surface.
bit source terms to be jointly determined by the ther- Let us consider the mechanical equilibrium condi-
modynamic study: tion (16)2 in more detail. It is equivalent, for fluid-like
system, to the uniformity of pressures, which can be
V̄N [P] + N · [b] = −f, (13) represented in terms of thermodynamic derivatives by
∂U
[p] = = 0, (18)
V̄N [θ S] − N · [Q] = qS , (14) ∂V S,M
194 A. Berezovski, G. A. Maugin
where p is pressure, U is the internal energy, V is point X belonging to an oriented surface of unit
volume, and M denotes mass. normal N and with an operation such as that of taking
It should be emphasized that the thermodynamic the discontinuity applied to it.
derivative in Eq. 18 is computed for a fixed entropy Then the derivative of entropy with respect to defor-
value. However, an irreversible progress of crack front mation gradient can be computed by
or phase boundary is accompanied by entropy produc-
tion. Therefore, we have proposed to exploit a more ∂S 1 ∂f f ∂θ
= − , (24)
convenient thermodynamic derivative (Berezovski and ∂F T θ ∂F T θ2 ∂F T
Maugin 2004), namely
and one can determine the stress jump at the discon-
∂U tinuity in terms of the driving force as follows
= 0, (19) (Berezovski and Maugin 2005a)
∂V p,M
fS ∂θ ∂f
In the thermomechanical case, the latter is equivalent [T] · N = − · N. (25)
θS ∂F T ∂F T
to (Berezovski and Maugin 2004)
This relation gives us the possibility to determine the
∂S velocity of the phase boundary.
θ + T · N = 0. (20)
∂F T
This means that we can determine the value of the jump 5 Phase-transition fronts in a bar
of stress tensor if we know the behavior of entropy in
the vicinity of a discontinuity. In the case of phase-transition front propagation, we
consider the simplest possible one-dimensional for-
mulation. This is the boundary value problem of the
4 Stress jumps tensile loading of a shape memory alloy bar that has
uniform cross-sectional area A0 and temperature θ0
As shown (Abeyaratne et al. 2001), in the case of phase (Abeyaratne et al. 2001). The bar occupies the inter-
transformations there are two sources of entropy pro- val 0 < x < L in a reference configuration and the
duction: heat conduction and phase transformation. In boundary x = 0 is subjected to the tensile loading
the isothermal case (θ = const), the entropy produc-
tion rate due to heat conduction vanishes. This means σ (0, t) = σ̂ (t) for t > 0. (26)
that the entropy jump is determined only by the driving
force and the temperature at the phase boundary, The bar is assumed to be long compared to its diameter
so it is under a uniaxial stress state and the stress σ (x, t)
fS depends only on the axial position and time. Supposing
[S] = . (21)
θS the temperature is constant during the process, it is cha-
racterized by the displacement field u(x, t), where x
In the case of phase-transition front, we may expect denotes the location of a particle in the reference confi-
that the entropy behaves like guration and t is time. Linearized strain is further assu-
med so the axial component of the strain ε(x, t) and the
f
S= , (22) particle velocity v(x, t) are related to the displacement
θ by
where the function f is defined similarly to Eq. 17 ∂u ∂u
ε= , v= . (27)
∂x ∂t
f = −W + N · T · F · N. (23)
The density of the material ρ0 is assumed constant. All
This formally defined quantity—a generator function— field variables are averaged over the cross-section of the
acquires a physical meaning only when evaluated at a bar. The geometry of the problem is shown in Fig. 2.
Moving singularities in thermoelastic solids 195
where
ρ0 VS [v] + [σ ] = 0, (30)
f = −W + σ ε. (37)
VS [ε] + [v] = 0, (31) It follows from Eq. 37 that the jump of the first term in
the right hand side of Eq. 36 is
VS θ [S] = fS VS , (32) 1 ∂f 1
= − [σ ] , (38)
θ ∂ε σ 2
where VS is the velocity of the phase boundary and
the driving traction fS (t) at the discontinuity is defined
and the relation (35) can be represented as
by (cf. Truskinovsky 1987; Abeyaratne and Knowles
1990) 1 f ∂θ 1 ∂θ
[σ ] = = fS
2 θ ∂ε σ θ ∂ε σ
fS = −[W ] + σ [ε], (33) 1 ∂θ
+ f . (39)
θ ∂ε σ
where W is the free energy per unit volume. In general, the function f is determined up to an arbi-
The second law of thermodynamics requires that trary constant, which in the isothermal case can be cho-
sen such that f = 0 (Berezovski and Maugin 2005a).
fS VS ≥ 0 (34) In the latter case, we obtain the value of the stress
196 A. Berezovski, G. A. Maugin
jump at the phase boundary in the form (Berezovski where σ̄A is the averaged stress in austenite, we can
and Maugin 2005b) rewrite the expression for the stress jump at the phase
boundary as follows
1 ∂θ
[σ ] = 2fS . (40)
θ ∂ε σ [ε̄] = A[σ̄ ] − B, (48)
0.6 600
Experimental
0.55
Phase boundary speed (normalized)
Numerical
0.5 500
0.45
400
0.4
Stress, MPa
0.35
300
0.3
0.25 200
0.2
0.15 100
0.1
0
0.05 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
1 2 3 4 5 6 7 8 9 10
Driving force (MPa) Strain
for kinetic relation has the form plotted in Fig. 4 together with experimental data for the
quasi-static loading (McKelvey and Ritchie 2000).
D(fS − fcr )
ρ0 VS2 = . (56)
AD(fS − fcr ) + ADfcr − B
7 Conclusions
The obtained kinetic behavior is illustrated in Fig. 3,
A kinetic relation for phase-transition front propagation
where the coefficients are calculated for the particular
is obtained by means of the procedure similar to that for
case of a Ni–Ti shape memory alloy.
moving straight-brittle crack (Berezovski and Maugin
Material properties for the Ni–Ti shape memory
2007), exploiting the driving force concept and non-
alloy were extracted from the paper by McKelvey and
equilibrium jump relations at the discontinuity. The
Ritchie (2000), where all the details of experimental
driving force is determined in the framework of the
procedure are well explained. The Young’s moduli are
material setting of thermoelasticity. Stress jump across
62 and 22 GPa for austenite and martensite phases, res-
the discontinuity is related to the driving force due to
pectively, Poisson’s ratio is the same for both phases
the non-equilibrium jump relation. The value of the
and is equal to 0.33, the density for both phases is
stress jump is used then in the jump relation for linear
6450 kg/m3 .
momentum, resulting in an explicit form of the kine-
The corresponding stress in austenite is 393 MPa
tic relation. Considered example represents the moving
(McKelvey and Ritchie 2000), and the critical value of
phase boundary in a simplified form, but it opens a pos-
the driving force is 2.25 MPa. The values of coefficients
sibility to generalizations and more detailed numerical
are the following: D = 414, AD = 0.0127 MPa−1 ,
simulations.
B = 0.0124. The values of the velocity of the phase
boundary are normalized by the longitudinal wave velo- Acknowledgements Support of the Estonian Science Founda-
city in austenite (3774 m/s). tion (A.B.) is gratefully acknowledged. G.A.M. benefits from a
To validate the obtained results, the local stress– Max Planck Award for International Cooperation (2001–2005).
strain relation at a fixed point which was initially in
austenitic state was calculated numerically. The best
fitting of experimental observations (McKelvey and References
Ritchie 2000) corresponds to the value for the transfor-
mation strain 3.3%. Just this value is used for the com- Abeyaratne R, Knowles JK (1990) On the driving traction acting
parison with the experimental data. After unloading, on a surface of strain discontinuity in a continuum. J Mech
Phys Solids 38:345–360
the reverse phase transformation occurs. It is supposed Abeyaratne R, Knowles JK (2006) Evolution of phase transi-
that the reverse transformation begins immediately after tions: a continuum theory. Cambridge University Press,
unloading. The calculated local stress–strain relation is Cambridge
198 A. Berezovski, G. A. Maugin
Abeyaratne R, Bhattacharya K, Knowles JK (2001) Strain- Freund LB (1990) Dynamic fracture mechanics. Cambridge
energy functions with local minima: modeling phase trans- University Press, Cambridge
formations using finite thermoelasticity. In: Fu Y, Ogden Maugin GA (1993) Material inhomogeneities in elasticity.
RW (eds) Nonlinear elasticity: theory and application. Cam- Chapman and Hall, London
bridge University Press, Cambridge, pp 433–490 Maugin GA (1997) Thermomechanics of inhomogeneous –
Berezovski A, Maugin GA (2004) On the thermodynamic condi- heterogeneous systems: application to the irreversible pro-
tions at moving phase-transition fronts in thermoelastic gress of two- and three-dimensional defects. ARI – Int J
solids. J Non-Equilib Thermodyn 29:37–51 Phys Eng Sci 50:41–56
Berezovski A, Maugin GA (2005a) On the velocity of moving Maugin GA (2000) On the universality of the thermomechanics
phase boundary in solids. Acta Mech 179:187–196 of forces driving singular sets. Arch Appl Mech 70:31–45
Berezovski A, Maugin GA (2005b) Stress-induced phase- Maugin GA, Trimarco C (1995) The dynamics of configuratio-
transition front propagation in thermoelastic solids. Eur J nal forces at phase-transition fronts. Meccanica 30:605–619
Mech – A/Solids 24:1–21 McKelvey AL, Ritchie RO (2000) On the temperature depen-
Berezovski A, Maugin GA (2007) On the propagation velocity dence of the superelastic strength and the prediction of the
of a straight brittle crack. Int J Fract 143:135–142 theoretical uniaxial transformation strain in Nitinol. Philos
Cermelli P, Sellers S (2000) Multi-phase equilibrium of crystal- Mag A 80:1759–1768
line solids. J Mech Phys Solids 48:765–796 Truskinovsky L (1987) Dynamics of nonequilibrium phase
Evora VMF, Jain N, Shukla A (2005) Stress intensity factor and boundaries in a heat conducting nonlinear elastic medium.
crack velocity relationship for polyester/TiO2 nanocompo- J Appl Math Mech (PMM) 51:777–784
sites. Exp Mech 45:153–159
Fineberg J, Marder M (1999) Instability in dynamic fracture.
Phys Rep 313:1–108
Dislocation tri-material solution in the analysis of bridged
crack in anisotropic bimaterial half-space
T. Profant · O. Ševeček · M. Kotoul ·
T. Vysloužil
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 199
doi: 10.1007/978-1-4020-6929-1_19, © Springer Science+Business Media B.V. 2007
200 T. Profant et al.
transformations, thin film growth, etc. (Eshelby 1970; transfer matrix for each wedge, it is easy to arrive at
Maugin 1995; Gurtin 2000; Kienzler and Herrmann an expression for the whole multimaterial corner, as it
2000; Simha 2000), and hence provides a comprehen- relates the variables between its external faces (ω0 and
sive framework for studying the influence of these ω N ):
processes on fracture. Second, based on the configura- (1) (2)
tional balance laws computational errors due to u N (r, ω N ) KN KN u1 (r, ω0 )
= (3) (4) , (3)
discretization and approximate numerical solution N (r, ω N ) KN KN 1 (r, ω0 )
methods can be quantified (Braun 1997; Mueller et al.
where K N is obtained by the product of the sequence
2002). Consequently, the configurational forces
of the successive transfer matrices Ei of all the wedges
approach allows evaluating the crack shielding/anti-
in the corner:
shielding due to sharp interface.
K N = E N · E N −1 · · · E2 · E1 ,
−1
2 Determination of eigenvalues and eigenvectors
Ei = XZ1−λ (ωi ) Z1−λ (ωi−1 ) X−1 . (4)
in Williams-like asymptotic expansion
This is worthy of note that Ting’s procedure directly
The procedure originally developed by Ting (1997) is yields a linear system whose size is 3 × 3 or 6 × 6,
an efficient tool for the singular characterization of irrespective of the number of materials N , contrary to
non-degenerate anisotropic multimaterial corners, of traditional analytical procedures leading a linear sys-
which a crack impinging bimaterial interface is a spe- tem of (6N × 6N ).
cial case. The ith material wedge occupies the polar The eigenpairs λ, q can also be evaluated using the
sector ωi−1 < θ < ωi , i = 1, . . . , N . Perfect bond- method developed by Papadakis and Babuska (1995).
ing is considered between material wedges. Fixed or Their method can be used with multi-material wedges,
free boundary conditions are considered at the exter- with anisotropic materials and general boundary con-
nal faces. The solution can be written in the condensed ditions under the assumption of plane strain. Along the
form using the complex variable z α = x1 + pα x2 = interfaces at θ = ωi , the following continuity condi-
r (cosθ + pα sinθ ) = r ζα (θ ): tions are assumed
w (r, θ ) = r 1−λ XZ1−λ (θ ) q, (1) [u]i = 0, [t]i = 0, (5)
where w T (r, θ )
= [u(r, θ ), (r, θ )]T ,
u stands for the
where u is the displacement vector, t is the traction
displacement vector and is the stress function vector,
vector and the brackets denote a jump along θ = ωi .
see Appendix A. pα are three distinct complex numbers
The problem of finding the characteristic exponent λ
with positive imaginary parts, which are obtained as the
can be viewed as the following eigenvalue problem:
roots of the characteristic equation (A1). The matrices
Find the characteristic exponent λ such that there exists
X and Z are defined as
1−λ F = 0 such that
A Ā ζ∗ 0
1−λ
X= , Z1−λ = , (2) dF (θ )
L L̄ 0 ζ̄∗ = H (λ; θ ) F (θ ) in ωi ≤ θ ≤ ωi+1 ,
dθ
where the matrices A and L are defined in Appendix A, O1 (λ) [F (ωi )] = 0 for θ = ωi ,
the overbar denotes the complex conjugate. ζ∗1−λ =
Singularity exponent λ 0.6716824 Computed using the dislocation approach Eq. B16
0.671825 Computed using the Shoot method
Auxiliary singularity exponent s=2−λ 1.3283176 Computed using the dislocation approach Eq. B16
1.328175 Computed using the Shoot method
Choi and Earmme (2002) used the so-called alternating and for the case in Fig. 4, it gives the following relations
technique that generalizes the formulas (B5) and (B6), for potentials
204 T. Profant et al.
⎧ ⎡ ⎤
⎪
⎪
∞
⎪
⎪ ⎣nα (z) + MII L̄II αβ nβ z − pαII h + pβII h ⎦, z∈2
⎪
⎨
β
α (z) = n=1 (14)
⎪
⎪
∞
⎪
⎪ ¯
(z) + II II
n
− II
+ II
, z ∈ 1,
⎪
⎩ C αβ βo CM αβ
L̄ βγ β z p β h p γ h
β γ n=1
ditions, the Burgers vector component b y is equal to Distance from the material interface [mm]
zero. The integral equation then reads Fig. 5 Calculations of σx x along the plane of dislocation with
1 Burgers vector b = (0.01,0) [mm] using either FEM and the alter-
appl
σx x (y) + σ̂br (δ (y)) + nating technique for several locations yo of dislocation core
2π
Dislocation tri-material solution in the analysis of bridged crack 205
and σ̂br (δ(y)) is the bridging stress from Eq. 11. The g(s)
b∗x (s) = (*)
integral equation (16) must be complemented with the (1 − s)λ
condition
A quadrature method has to be adapted to the singu-
0 (−λ,0)
larities of b∗x (s)using Jacobi polynomials Pn (s) at
δ (y) = bx (y0 ) dy0 , (17) ∗
all. The form of the density function bx (s) disables to
y receive the closed form solution of the regular kernel
which relates the crack opening displacement δ to the K x x (s, t), which has to be approximated (Profant et al.
(−λ,0)
dislocation density bx . The regular kernel K x x (y, y0 ) 2007). To avoid the application of Pn (s) polyno-
was obtained from the truncated semi-analytical solu- mials, the following approximation of the density func-
tion of Choi and Earmme (2002). The regular kernel tion can be used (Dewynne et al. 1992)
describes the interaction of a dislocation with the bima-
1+s λ
terial interface and with the free surface as well. K x x b∗∗
x (s) = b g
−1 −1 (s) + g(s)
(y, y0 ) possesses a complicated structure and 1−s
depends on elastic constants of both materials and on the where g−1 (s) is some known, bounded function on
layer thickness. It can expressed as the truncated series [−1, 1] such that g−1 (s) = −1 and b−1 is an unknown
NK
k1,n constant, which equals b∗∗ x (−1). This equality serves
K x x (y, y0 ) = y − k y − k . (18) as an additional consistency condition. This form of
k2,n 0 3,n 4,n
n=1 the density function corrects the crack opening at the
are the constants developed from the alternat-
The ki,n crack mouth without the influence on the stress inten-
ing technique discussed above. The substitutions sity factor. Because the objective of the paper is to find
y0 y the stress intensity factor, the crack mouth opening cor-
s = 2 + 1, t = 2 + 1 (19)
h h rection will be omitted below. Observe that while the
allow to reduce the integral equation (16) to the form value of the function g(s) at the crack tip obtained by
appl
σx x (y) + σ̂br (δ (y)) using the density function bx (s) is g(1) = 0.021, the
⎧ ⎫ 1 density function b∗x (s) leads to the result g(1) = 0.020.
1 ⎨ 2 2
⎬ b (s)
−1 x The last value corresponds to the approximation of the
+ Re L II M II
βα B II
α1 ds
2π ⎩ 1β ⎭ s−t regular kernel N f = 20, the approximation of the den-
β=1 α=1 −1
sity of the Burgers vector N B = 10 and the chosen
1 collocation points
+ bx (s) K x x (t, s) ds = 0, (20)
π 2i + 1
−1 ti = 2 cos − 1.
4 NB
where
NK The integral equation may be solved using the
k1,n
K x x (t, s) = . (21) Gauss-Jacobi quadrature. The function g(s) is sought in
k2,n s − k3,n t − k4,n the form of linear combination of Jacobi polynomials
n=1
The procedure involves the reduction of the integral Pn(−λ,λ) (s)
equation and constraints to a system of algebraic equa- ∞
NB
tions using the collocation technique. g (s) = cn Pn(−λ,λ) (s) ∼
= cn Pn(−λ,λ) (s). (22)
The dislocation density is sought in the form bx (s)= n=0 n=0
(1 − s)−λ (1 + s)λ g(s), where g(s) is a bounded func-
This allows to express the integral containing the reg-
tion. As mentioned elsewhere (Hills et al. 1996), this
ular kernel K x x (t, s) in the closed form by integrat-
choice means that bx (−1) must vanish, i.e. that crack
ing each component of the truncated series in Eq. 18.
faces at the mouth are forced to be parallel and the
It is useful to apply the theory of the curve complex
solution is over-constrained. Nevertheless, this incor-
integrals developed by Muschelishvili (1953) because
rect end-point behaviour at the crack mouth had a neg-
of the ambiguous behaviour of the (1 − s)−λ (1 + s)λ
ligible effect on the calculated stress intensity factor.
around the points s = ±1. Hence, after the integra-
Remark A more acceptable form of the density func- tion, the regular part of the integral equation (20) can
tion can be found as (Dewynne et al. 1992) be written as
206 T. Profant et al.
√
1 NK
1
k1,n bx (s)ds Hti p = lim 2πr λ σx x (r, θ = π/2)
r →0
bx (s)K x x (t, s)ds = ⎧ ⎡ ⎛
k2,n s − k3,n t − k4,n √ ⎨ ieλiπ
II
n=1 −1
−1
= 2π h λ g(1)Re ⎣ L 11
I ⎝
C11 M11
1 ⎩1 − e 2πiλ
NK
k1,n 1+s λ k1,n g(s)ds
= λ−1
k2,n 1−s s − (k3,n t + k4,n )/k2,n
II −1
II −1 p1II
n=1 −1 × B11 + M12
II
B21 + C12
p1I
NK
k1,n 2πieiλπ (k3,n t + k4,n ) + k2,n λ
⎞
=
p II λ−1
k2,n 1 − eiλ2π (k3,n t + k4,n ) − k2,n −1 −1
⎠+L 12
n=1 × M21 II
B11II
+M22 II II
B21 2 I
NB ship. There are a great number of methods available for
× cn cot (−π λ) (1 − ti )−λ (1+ti )λ Pn(−λ,λ) (ti ) the determination of stress intensity factors such as e.g.
n=0 the finite element method with contact elements, the
boundary element method, the boundary collocation
(−λ) (n+λ+1) 1−ti
− F n+1, −n; 1+λ; method, or the weight function method. High efficiency
(n+1) 2
of the weight function method consists in that once
N k
k1,m 2πieiλπ (k3,m ti + k4,m ) + k2,m λ the weight function(s) are known the bridging intensity
+
k2,m 1 − e iλ2π (k3,m ti + k4,m ) − k2,m factor can be easily calculated for any bridging stress
m=1
distribution by evaluating the integral of the form of
NB
Eq. 40. Moreover, it allows setting up a bridging stress-
× cn Pn(−λ,λ) ((k3,m ti + k4,m )/k2,m )
n=0
crack opening displacement relationship by analysing
the experimental crack opening displacement data and
− (αm s + · · · + α0 )
m
s=(k3,m ti +k4,m )/k2,m
= 0, solving an integral equation. The weight function
method has been extensively used to the modelling
(25) of bridged crack problems (Fett and Munz 1997). For
where F(n 1 , n 2 ; n 3 ; x) stands for the hypergeometric a complicated domain, the weight function has to be
function, (n) is the Gamma function and i = 0, 1, . . . , obtained numerically, e.g. from FEM calculations (Sar-
N B − 1. The strength of the singularity in stress may be rafi et al. 1998). As to a crack impinging on the bimate-
quantified in the usual way by defining the local gener- rial interface, such calculations have not been reported
alized SIF Hti p . Using the function-theoretic methods yet. The weight function is obtained numerically by
(Erdogan et al. 1974; Hills et al. 1996) one obtains performing a number of calculations of the generalised
Dislocation tri-material solution in the analysis of bridged crack 207
stress intensity factor due to unit line load applied to the Remark For the crack making a right angle with the
crack face at arbitrary points. To this end, an application interface between two different orthotropic materials
of the reciprocal theorem seems to be very efficient. whose axes of material symmetry are parallel with or
perpendicular to the interface, the most singular stress
5.1 Application of the -integral fields at the tip can be written as
In the absence of body forces the reciprocal theorem σi j = H I r −λ1 σiIj (θ ) + HII r −λ1 σiIIj (θ ). (30)
states that the following integral is path independent
In this case, the eigenvalue problem for the exponent
(Stern and Soni 1976)
λ1 has a double root yielding two linearly independent
(u, v) = σi j (u) n i v j − σi j (v) n i u j ds, (27) fields σiIj and σiIIj which can be taken to be symmet-
ric and anti-symmetric relative to the crack plane. In
where is any contour surrounding the crack tip and u, the case where λ1 is a multiple eigenvalue, then so is
v are two admissible displacement fields. If the follow- 2−λ1 , and one has to solve a system of equations for the
ing displacement fields are considered u = Ui (x) = corresponding generalized intensity factors. For exam-
r 1−λi ūi (θ ), v = U j (x) = r 1−λ j ū j (θ ), one can show ple, consider the case where λ1 is a double eigenvalue,
that the contour integral is equal to zero for corresponding to two eigenfunctions ū1I (θ ) and ū1II (θ ).
−(1 − λi ) = 1 − λ j and non-zero if −(1 − λi ) = The displacement field u has the following expansion
1 − λ j . Thus, for the set of basis functions {Ui (x) = u = H I r 1−λ1 ū1I (θ ) +HII r 1−λ1 ū1II (θ ) +h.o.t. . . .
i
r|i| (1−λ|i| )
ūi , i = ±1, ±2, . . .}, there is a kind of orth- (31)
ogonality with respect to the “scalar” product (·, ·). The orthogonality condition (28) then leads to the fol-
Since the basis function corresponding to coefficient lowing system to solve for the generalized intensity
f 1 = H (the generalized stress intensity factor) in the factors H I and HII
asymptotic expansion for u is r 1−λ1 ū1 (θ ), due to the
1−λ I −(1−λ I
1−λ I −(1−λ II
r 1 ū1 , r 1) ū 1) ū
−1 r
1 ū , r
former orthogonality conditions it holds 1 −1
1−λ II −(1−λ ) I
1−λ II −(1−λ ) II
∞ r 1 ū1 , r 1 ū−1 r 1 ū1 , r 1 ū−1
−(1−λ1)
u, r ū−1 = f i r 1−λi ūi , r −(1−λ1) ū−1
−(1−λ )
HI
u, r 1 I
ū−1
i=1 = . (32)
HII u, r −(1−λ1 ) ū−1 II
= f 1 r 1−λ1 ū1 , r −(1−λ1) ū−1 .
For the crack impinging interface at an oblique angle
(28) and/or for an arbitrary orientation of the axes of mate-
Thus, the generalized stress intensity factor H = f 1 rial symmetry with respect to the interface, the eigen-
can be computed as follows: value problem no longer has double roots. Instead of
(32), the two most singular fields of the interest are
u, r −(1−λ1 ) ū−1
H =
1−λ . (29) (1) (2)
r 1 ū1 , r −(1−λ1 ) ū−1 σi j = H1r −λ1 σi j (θ ) + H2 r −λ2 σi j (θ ) (33)
On the other hand, since the exact solution u is not Corresponding to each eigenvalue is only one eigen-
known, a finite element solution uh can be used as an function instead of two. Thus, the generalized inten-
approximation for u so to obtain an approximation for sity factors H1 and H2 in (33) are calculated using the
H , (Profant et al. 2007). Due to the path independence, formula (29).
the -integral standing in the denominator of Eq. 29 is The dual fields can be easily calculated using the dis-
evaluated along an infinitesimal path that shrinks to the location approach described in the Appendix B assum-
crack tip. The solution for stresses and displacements ing the distribution of dislocations in the form
corresponding to the basis function r −(1−λ1 ) ū−1 are wk
fˆk (r ) = , ∞ < r < 0, (34)
sometimes referred to as auxiliary (dual) fields which (−r )2−λ
satisfy the same equilibrium and constitutive equations where wk is the eigenvector of matrix D corresponding
as the actual fields σi j and u i and the dual tractions to the eigenvalue s = 2 − λ. After a tedious algebra, the
vanish on crack faces. These dual fields are more sin- -integral along the infinitesimal contour follows as
gular than the actual fields and they are not physically
meaningful. r 1−λ1 ū1 , r −(1−λ1 ) ū−1 = c1 − c2 . (35)
208 T. Profant et al.
where (−1)λ
⎧ × (cot (π λ) − i) wk × Re Aiα
⎪ ταII
(θ ) 2−λ
⎨ 0 α
1
λ−1
c1 = Re II
L iα sin θ − pαII cos θ × II
p̄βII + δik pαII
λ−1
4 (λ−1) ⎪
⎩ α
G αβ M̄βk
− π2 β
λ−1
π
II λ−1 (−1)λ
× II
G αβ M̄βk p̄βII + δik pα × (cot (π λ) − i) vk dθ + Re
β ταII (θ )λ
0
(−1)λ ⎧
× (cot (π λ) − i) vk × Re Aiα ⎨
λ
ταII (θ ) α × I
L iα sin θ − pαI cos θ Cαβ Mβk II
⎩
λ−1
λ−1
α β
× II
G αβ M̄βk p̄βII + δik pαII
λ−1 (−1)λ
β × pβ II
(cot (π λ) − i) wk × Re
π ταI (θ )2−λ
(−1)λ ⎧
× (cot (π λ) − i) wk dθ + Re ⎨ λ−1 (−1)λ
ταII (θ )1−λ
0 × Aiα Cαβ MβkII
pβII
⎧ ⎩ α
β
ταI (θ )λ
⎨ ⎧
× I
sin θ − pαI cos θ II 3π ⎪
⎩
L iα Cαβ Mβk
2
⎨ 0
α β × (cot (π λ) − i) vk dθ + Re II
L iα
⎪
⎩ π
λ−1 (−1)λ π α
−2
× pβ II
(cot (π λ) − i) vk × Re
ταI (θ )λ
λ−1
⎧ × sin θ − pαII cos θ G αβ M̄βkII
p̄βII + δik
⎨ λ−1 (−1)λ β
× Aiα Cαβ MβkII
pβII
⎩ α
β
ταI (θ )1−λ
λ−1 (−1) λ
× pαII (cot (π λ) − i) wk × Re
3π
ταII (θ )2−λ
2 ⎧
× (cot (π λ) − i) wk dθ + Re II
L iα ⎨ λ−1
IIλ−1
α × Aiα II
G αβ M̄βk p̄βII
+ δik pα
π ⎩
⎡ &λ−1 % α β
p̄βII
⎣ II 1 (−1)λ
× sin θ − pα cosθ
II
G αβ M̄βk II × (cot (π λ) − i) vk dθ . (37)
pα pαII
β ταII (θ )λ
δik (−1)λ
+ II λ
vk × Re Aiα
pα τα (θ )
II
α Now assume that a pair of line forces acts on the crack
λ−1
II λ−1
faces at a point yb , see Fig. 6. Other loading is absent.
× II
G αβ M̄βk p̄βII
+ δik pα Equation 27 modifies with help of Eq. 28 as
β
(−1)λ
× (cot (π λ) − i) wk dθ , (36)
ταII (θ )
1−λ σi j (u) n i r −(1−λ1) ū −1 j −σi j r −(1−λ1) ū−1 n i u j
3
⎧ ds+2Fr −(1−λ1) ū−1 =H r 1−λ1 ū1 , r −(1−λ1) ū−1 .
⎪
1 ⎨ 0
(38)
c2 = Re II
L iα sin θ − pαII cos θ
4 (λ−1) ⎪
⎩ α
− π2
3 is an arbitrary contour enclosing a domain contain-
λ−1
λ−1
× II
G αβ M̄βk p̄βII + δik pαII ing both the crack tip and the pair of line forces. By
β definition, the weight function W (yb , h) follows as
Dislocation tri-material solution in the analysis of bridged crack 209
Fig. 6 A pair of line forces acting on the crack faces and the 5.2 Determination of the generalized bridging stress
integration path intensity factor
Fig. 9 Applied, closure and total crack opening displacement for several values of the Weibull modulus m, σ0 f = 2300 MPa. The
applied tensile loading (a) σ0 = 90 MPa, (b) σ0 = 140 MPa
Dislocation tri-material solution in the analysis of bridged crack 211
Fig. 10 Applied, closure and total crack opening displacement for several values of the Weibull modulus m, σ0 f = 2300 MPa. The
applied tensile loading (a) σ0 = 90 MPa, (b) σ0 = 140 MPa
ld=1
Ωin
M2 M1
x2 ld
7 Discussion y=x/ε
x1
z1
As it was already mentioned, a detailed quantification
of the strength of singularity in stress and displace- Fig. 12 Zoomed-in view of crack neighbourhood perturbated
ment field around the crack terminating at the bima- by a small crack extension
terial interface is a prerequisite for an assessment of
fracture behaviour of many modern composites. The
computational model developed in this paper provides which provides a zoomed-in view into the region sur-
a basis for an application of the criterion of fracture rounding the crack. The displacement uε of the per-
at an interface between anisotropic materials when a turbed elasticity problem due to the presence of the
competition between deflection and penetration at the crack can now be expressed in terms of the regular
interface has to be analysed by seeking the path, which coordinate x and the scaled-up coordinate z as follows:
maximises the additional energy released by the frac- uε (x) = uε (εz) = Vε (z), (43)
ture process for a finite and small crack extension. If
crack deflection occurs preferentially to penetration at where the definition of the function Vε has been intro-
the interface, the following condition must be satisfied: duced, simply by a change of variable from x to z. Note
Wd = δWd − G ci ld > W p = δW p − G c1 l p , where that Vε (z) is viewed here as exactly the same displace-
G ci is the interface toughness, G c1 is the toughness of ment as denoted by uε (x), i.e., the solution of the per-
the material M1 and δW is a change of the potential turbed elasticity problem, but in terms of the scaled-up
energy between the original and new crack position. variable y, instead of the regular variable x. The change
Matched asymptotic procedure can be used to derive in the potential energy δW between the solutions in the
the change of potential energy. unperturbed u0 (x1 , x2 ) and perturbed uε (x1 , x2 ) situ-
Consider a perturbation of the domain with crack ations for unchanged boundary conditions (unchanged
impinging on the interface between materials M2 and applied loads) is by use of Betti’s theorem as follows:
M1 as shown in Fig. 12; the perturbation is a deflected 1
ε
(double) crack extension of length ld or penetrating δW = σi j u n i u 0j − σi j u0 n i u εj ds
2
crack extension of length ld with the small perturbation
parameter ε defined as = K 1 H 2 ε2(1−λ1 ) + · · ·
2(1−λ1 )
l l
δ=
1, l = l p , ld , (41) = K1 H 2 + · · · , ε
1, (44)
L L
where L is the characteristic length of . A second where is any contour surrounding the crack tip and
scale to the problem can be introduced, represented by n i its normal pointing toward the origin, see also the
the scaled-up coordinates Sect. 5.1. The inner domain asymptotics approximates
x x x δW with K 1 depending on the perturbation due to a
1 2
z = , or (z 1 , z 2 ) = , , (42) small crack extension l through inner expansion func-
ε ε ε
Dislocation tri-material solution in the analysis of bridged crack 213
−
ε
Cinh = − n · ê ci jkl εi+j uε εkl u ds, (47) rect physical interpretation of Jbr is following—it is the
2 total force on the dipole distribution (per unit length
of crack) and the quantity −Jbr tδa is the Gibbs free
where n is the interface unit normal, ci jkl is the elas-
energy change when the bridging traction are shifted
tic moduli tensor, ε+ and ε − are the strain values on
along the crack faces by δa in the crack direction keep-
the right side and left side of the interface respectively,
ing the crack tip fixed in position (t is the thickness of
[[ci jkl ]] = ci+jkl − ci−jkl . Cinh quantifies the crack tip
the material).
shielding or antishielding and originates from material
inhomogeneity in the direction of crack growth. Equat-
8 Summary
ing the crack tip J -integral in Eq. 46 to the release rate
G in Eq. 45 allows scaling of the far field J -integral
• Dislocation trimaterial solution was found to hold
with the generalized stress intensity factor H . Such
also for the considered finite specimen.
scaling could be helpful in experimental study of crack
• GSIF was calculated both by CDD technique and
growth across the bimaterial interface. Observe that
FEM combined with the -integral. The results
Cinh equals to zero when crack propagates along the
obtained were found to be in good agreement.
interface. In the case of a bridged crack terminating at
• The weight functions for a crack in the surface layer
the bimaterial interface, Eq. 46 modifies as (see Fig. 13)
terminating at the layer/substrate interface were cal-
1−2λ1
1 l culated using FEM combined with the -integral.
K 1 Hti2 p 2 (1 − λ1 ) + Jbr = J f ar + Cinh ,
L L The weight functions were further used in the calcu-
(48) lation of the bridging GSIF.
• The influence of the bridging fibres was analyzed
where Eq. 45 was used. Jbr represents the resisting using the advanced model, involving a statistical dis-
force arising from the bridging tractions and in the tribution of fibre strength and pull out of fibres from
present geometry is given by the matrix.
δ ∗ • The relationship with the configurational mechanics
∂δ
Jbr = σbr (y) dy = σbr (δ) dδ, (49) approach was discussed.
∂y
0
Acknowledgements The support through the grants GACR
where δ ∗ is the crack opening at the end of the bridged 101/05/0320, 106/05/H008, and 101/05/P290 is gratefully
region. acknowledged.
214 T. Profant et al.
Appendix A and
1
qα = Mαk dk , (B3)
Following Lekhnitskii (1963), for plane deformation, 4π
the elastic field can be represented in terms of complex where the vector dk is related to the Burgers vector bi
potential functions 1 (z 1 ), 2 (z 2 ), 3 (z 3 ), each of through the equation
which is holomorphic in its arguments z α = x + pα y. i
Here, pα are three distinct complex numbers with pos- bi = Bik dk , with Bik = Aiα Mαk − Āiα M̄αk
2 α
itive imaginary parts, which are obtained as the roots % &
of the characteristic equation
= −Im Aiα Mαk , (B4)
det ci1k1 + p (ci1k2 + ci2k1 ) + p 2 ci2k2 = 0, (A1) α
where the matrix Mαk is defined as the inverse of L iα ,
where ci jkl is the tensor of elastic constants, i.e. σi j = Mαk L kβ = δαβ . The quantities pα , Aiα , L iα are given
ci jkl u k,l , which satisfies the symmetry conditions in Appendix A.
cijkl = ci jlk = c jikl = ckli j . (A2) The potentials for the interaction of an edge disloca-
tion with the interface of two anisotropic materials can
With these holomorphic functions, the representation be obtained in terms of o (z) by invoking the standard
for the displacements u i and stresses σi j is analytical continuation arguments along the interface,
3 as described by Suo (1990). The solution for the two
u i = 2Re Aiα α (z α ) , media can be written as
α=1 (z) = I (z), z ∈ 1,
3
(z) = II (z) + o (z), z ∈ 2, (B5)
σ2i = 2Re L iα α (z α ) ,
α=1 I (z) = Co (z),
3
C = iM I H−1 AII MII − ĀII M̄II LII , z ∈ 1,
σ1i = −2Re L iα pα α (z α ) , (A3)
α=1 H = i A I M I − ĀII M̄II ,
Here, ()’ designates the derivative with respect to the ¯ o (z),
II (z) = G
associated arguments, and A and L are matrices given
by G = −iMII H̄−1 ĀII M̄II − Ā I M̄ I L̄II , z ∈ 2.
1 I I dk where λ is the stress singularity exponent, which is
σ1i (x, y) = − L p II
Cαβ Mβk
4π α iα α z α − ςβ yet unknown, vk are the components of corresponding
β
eigenvector, and H is the generalized stress intensity
+ C.C., z ∈ 1, (B9)
factor (GSIF). Substitute Eq. B13 in Eq. B12, integrate
1 I dk and apply the traction-free condition on the plane of
σ2i (x, y) = L II
Cαβ Mβk
4π α iα z α − ςβ the crack to obtain
β ⎧⎡ % &λ−1
+ C.C., z ∈ 1, (B10) ⎨ p̄βII
Re ⎣ L iα G αβ M̄βk − II
II II
where C.C. denotes the complex conjugate of the pre- ⎩ α pα
β
ceding expression, superscript I and I I refers to the
material 1 and 2 respectively, and the convention of csc (π λ) − δik cot (π λ) vk = 0. (B14)
summing over repeated Latin indices is used. Introduce
a function f k at a point on the crack (x = 0, y) which Eq. B14 can be briefly written as
relates to the elemental Burgers vector δbi between yo
and y0 + δy0 as D (λ) v = 0, where
⎧⎡ % &
⎨ II λ−1
δbi = Bik δdk = Bik f k (yo ) δyo , (B11) p̄ β
Dik (λ) = Re ⎣ II
L iα II
G αβ M̄βk − II
and integrate (B7) along the whole crack. The tractions ⎩ α pα
β
produced at a point (0, y) by the density function f k
can be expressed as csc (π λ) − δik cot (π λ) . (B15)
⎧ ⎡ ⎛
II
0
1 ⎨ p
II ⎣ ⎝G αβ M̄βk α The parameter λ is calculated from the characteristic
σ1i (y) = − L II
4π ⎩ α iα p̄βII equation
β −∞
⎞ ⎤⎫
0 ⎪ Det [D (λ)] = 0 (B16)
f k (yo ) dyo ⎟ f k (yo ) dyo ⎥⎬
⎠ + M II
αk ⎦ and the eigenvector v is determined from Eq. B15 up to
pαII
y − yo y − yo ⎪
⎭
p̄βII −∞ a multiplicative constant.
⎡ ⎛ ⎞
⎜ 0
II ⎢
II
II p̄α f k (yo ) dyo ⎟
+ L̄ iα ⎣ ⎝Ḡ αβ Mβk II ⎠ Appendix C
pβ p̄αII
α β −∞ pβ II y − y o
⎤⎫
0 Recurrent calculations of bridging stress as a function
f k (yo ) dyo ⎦⎬ of position along the crack face consist of the following
+ M̄αk II
. (B12)
y − yo ⎭ steps:
−∞
The asymptotic stress field near the crack tip is mod- (1) In the first step, the displacement of the upper crack
elled as a continuous distribution of dislocations with face for unbridged crack in a bimaterial body due
density function to tensile loading σ1 is calculated using FEM:
f k (yo ) = H vk (−yo )−λ , yo < 0, (B13) vappl(1) (y) = vappl0 (y) σ1 (C1)
216 T. Profant et al.
Fig. C1 Displacement of the upper crack face caused by the unit Fig. C2 Displacement of the upper crack face for a number of
loading σ0 positions of the applied line load
where vappl0 is the displacement of the upper crack In next sub-steps the total crack face displacement
face caused by the unit loading σ0 , see Fig. C1. is refined as follows:
(2) In the second step, the total displacement of the vappl(n) (y)
vn,m n +1 (y) = vn,m n , n=1
upper crack face is recurrently refined in several vn,m n (y)−vbr (n,m n ) (y)
sub-steps. In the first sub-step of the recurrent cal- (C4)
culation the total displacement, vn,1 , n = 1, is set where vbr (n,m n ) < 0 is the crack face displace-
equal to vappl(n) , n = 1. Then the bridging stress ment due to the bridging stress σ̂br (n,m n ) (y) =
is computed via σ̂br (n,m n ) [vn,m n (y)]. Note that for the exact value
of vbr (x) the ratio in Eq. C4 equals to one. The
σbr (n,1) m+1
σ̂br (n,1) = σbr (n,1) exp − recurrent calculation stops when
cf
, -. / vn,m n +1 (x) − vn,m n (x) 2
< T O1 (C5)
fraction of intact fibres
vn,m n +1 (x)
σbr (n,1) m+1 where TO1 is a prescribed tolerance.
+σ P(n,1) 1− exp − ,
cf
, -. /
fraction of broken fibres References
(C2)
0 Braun M (1997) Configurational forces induced by finite-
where σbr (n,1) (y) = β vn,1 (y), n = 1, σ P(n,1) element discretization. Proc Estonian Acad Sci Phys Math
2c τ
(x) = Rf (h − vn,1 (y)), n = 1. This makes 46:24–31
Budiansky B, Amazigo J (1989) Toughening by aligned, fric-
possible to estimate the corresponding crack face tionally constrained fibers. J Mech Phys Solids 37:93–109
displacement vbr (n,1) using the FEM solution for Budiansky B, Hutchinson JW, Evans AG (1986) Matrix fracture
the crack face displacement due to the unit line in fiber-reinforced ceramics. J Mech Phys Solids 34:167–
force from the relation 189
Choi ST, Earmme YY (2002) Elastic study on singularities inter-
acting with interfaces using alternating technique part. Part
vbr (n,1) (y) = vbr 0 (y, yi ) σ̂br (n,1) (yi ) S (yi )
I. Anisotropic trimaterial. Int J Solid Struct 39:943–957
i Curtin WA (2000) Stress-strain behaviour of brittle matrix com-
(C3) posites. In: Kelly A, Zweben C (eds) Comprehensive com-
posite materials, vol 4. Pergamon Press, pp 47–76
where vbr 0 (y, yi ) is the crack face displacement Dewynne JN, Hills DA, Nowell D (1992) Calculation of the
due to the unit line force acting at the point yi , see opening displacement of surface-breaking plane cracks.
Comp Methods Appl Mech Eng 97:321–331
Fig. C2. S(yi ) is the area per node at the point yi Erdogan F, Gupta GD, Ratwani M (1974) Interaction between
and the summation is performed over all node rows a circular inclusion and an arbitrary oriented crack. J Appl
behind the crack tip. Mech 12:1007–1013
Dislocation tri-material solution in the analysis of bridged crack 217
Eshelby JD (1970) Energy relations and the energy-momentum Papadakis PJ, Babuska I (1995) A numerical procedure for the
tensor in continuum mechanics. In: Kanninen M et al determination of certain quantities related to the stress inten-
(eds) Inelastic behavior of solids. McGraw Hill, New York sity factors in two-dimensional elasticity. Comp Methods
pp 77–115 Appl Mech Eng 122:69–92
Fett T, Munz D (1997) Stress intensity factors and weight func- Profant T, Ševeček O, Kotoul M (2007) Calculation of K -factor
tions. Computational Mechanics, Inc. and T -stress for cracks in anisotropic bimaterials. accepted
Gurtin ME (2000) Configurational forces as basic concepts of in Eng Fract Mech
continuum physics. Springer, New York Sarrafi-Nour GR, Coyle TW, Fett T (1998) A weight function for
He MY, Hutchinson JW (1989) Crack deflection at an inter- the crack surface tractions in chevron-notched specimens.
face between dissimilar elastic materials. Int J Solids Struct Eng Fract Mech 59:439–445
25:1053–1067 Simha NK (2000) Toughening by phase boundary propagation.
Hills DA, Kelly PA, Dai DN, Korsunsky AM (1996) Solution of J Elast 59:195–211
crack problems. Kluwer Academic Publishers, Dordrecht Simha NK, Fischer FD, Kolednik O, Predan J, Shan
Kienzler R, Herrmann G (2000) Mechanics in material space. GX (2005) Crack tip shielding or anti-shielding due to
Springer, Berlin smooth and discontinuous material inhomogeneities. Int J
Lekhnitskii SG (1963) Theory of elasticity of an anisotropic Fract 135:73–93
body. Holden-Day, San Francisco Stern M, Soni ML (1976) On the computation of stress intensi-
Marshall DB, Cox BN (1988) J-integral method for calculating ties at fixed-free corners. Int J Solid Struct 12:331–337
steady-state matrix cracking in composites. Mech Mater Suo Z (1990) Singularities, interfaces and cracks in dissimilar
7:127–133 anisotr. media. Proc R Soc Lond A 427:331–358
Martin E, Leguillon D, Lacroix C (2001) A revisited criterion Thouless MD, Evans AG (1988) Effects of pull-out on the
for crack deflection at an interface in a brittle bimaterial. mechanical properties of ceramic matrix composites. Acta
Compos Sci Technol 61:1671–1679 Metall 36:517–522
Martinez D, Gupta V (1994) Energy criterion for crack deflec- Ting TCT (1997) Stress singularities at the tip of interfaces
tion at an interface between two orthotropic media. J Mech in polycrystals. In: Rossmanith HP (ed) Proceedings of
Phys Solids 42:1247–1271 first international conference on damage and failure of
Maugin GA (1995) Material forces: concepts and applications. interfaces. Balkema Publishers, Rotterdam, Netherlands,
ASME Appl Mech Rev 48:213–245 pp. 45–82
Mueller R, Kolling S, Gross D (2002) On configurational forces Tullock DL, Reimanis IE, Graham AL, Petrovic
in the context of the finite element method. Int J Numer JJ (1994) Deflection and penetration of cracks at an
Meth Eng 53:1557–1574 interface between two dissimilar materials. Acta Metall
Muschelishvili NI (1953) Some basic problems in the mathemat- Mater 42:3245–3252
ical theory of elasticity. Noordhoff, Holland
Study of the simple extension tear test sample for rubber
with Configurational Mechanics
Erwan Verron
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 219
doi: 10.1007/978-1-4020-6929-1_20, © Springer Science+Business Media B.V. 2007
220 E. Verron
C
(a) C
A F
A
B (b)
F l
c
(C0) a
e1 a c
h dc
e3 d
C' F
(c) h0
C'
b
e2
2a
A'
(C)
-ah 0 e 1
D'
f
2ah 0 e 1
B'
(C0 ')
Obviously, f is not exactly equal to F−1 because the – the regions c and the region d in which strain and
crack has grown and (C0 ) differs from (C0 ). stress distributions are complicated, the former cor-
respond to the neighbourhood of the points in which
forces apply and the latter corresponds to the neigh-
bourhood of the tip.
3 Previous derivations
When the crack length increases from c to c + dc,
3.1 Derivation of Rivlin and Thomas (1953) authors consider that:
– moreover, after crack growth the region a is still Eshelby (1975b) proposed a new derivation of the
in uniaxial extension with the same stretch ratio λ tearing energy for the trousers sample.
because F does not change, the region b is still In this way, following Knowles and Sternberg (1972)
undeformed, and the strain and stress distributions he calculated the following surface integral defined in
in regions c and d remain unchanged. the reference configuration (C0 ) shown in Fig. 2a
So, the length between legs extremities increases by γ1 = W − Pi1 u i,1 d S (11)
dl = 2λ dc, and then, S
where S is a surface embracing the tip of the crack,
∂l
= 2λ. (5)
∂c F
P is the first Piola-Kirchhoff stress tensor and u is the
displacement vector. In fact, this scalar quantity cor-
Considering that the strain energy of the sample is a responds to the first component of the configurational
function of the two quantities l and c, its change dw force
is related to the changes of crack length dc and overall
length dl by γ = n d S (12)
S
∂w ∂w in which n is the unit vector normal to the surface d S
dw = dl + dc. (6)
∂l c ∂c l and is the energy momentum (configurational stress)
Recalling that crack growth is due to the applied force tensor proposed by Eshelby (1951, 1975a)
F which remains constant, the previous equation can = W I − ∇Xt u P, (13)
be written as
where I is the 3×3 identity tensor and the superscript ·t
∂w ∂w ∂l ∂w
= + . (7) denotes the transposition. γ1 being path-independent,
∂c F ∂l ∂c
c F ∂c l the author considers a surface S made up of parts of the
With respect to the previous statement about the specimen surface with normal e2 and e3 in regions A,
increase in volume of the regions a and the decrease D and B; a cross-section of the region B (which is not
in volume of the region b, the change in energy can be deformed); and cross-sections of both legs (in regions
determined in the reference configuration (C0 ), A). The two first contributions are equal to zero; then
∂w γ1 reduces to
= W 2a h 0 , (8)
∂c F ∂u 2
γ1 = 2 W − P21 a h0. (14)
where W stands for the strain energy density per unit ∂ X1
of undeformed volume and 2a h 0 is the cross-section A given point with initial coordinates (X 1 , X 2 = 0)
of the undeformed sample. Moreover, the force F is being first swung round then stretched, the displace-
related to the change in energy by ment in the e2 -direction is
∂w u 2 = ±λ X 1 − X 2 , (15)
= F. (9)
∂l c
where the sign depends on the leg: plus sign for the
Finally, recalling Eqs. 5, 8 and 9, Eq. 7 becomes right-hand side leg and minus sign for the left-hand side
∂w leg (see Fig. 2). The stress is given by P21 = ±F/a h 0 ,
= W 2a h 0 − F 2λ, (10) with the same remark for the signs. Thus, γ1 reduces to
∂c l
the tearing energy being obviously deduced from Eq. 4. γ1 = 2 (W a h 0 − F λ) . (16)
Recalling that the total energy release rate (tearing
energy) is related to γ1 by
3.2 Derivation of Eshelby (1975b)
1
T =− γ1 , (17)
In order to highlight the connection between the confi- h0
gurational stress tensor (also called the elastic energy in which the minus sign merely indicates that the tear
momentum tensor or the Eshelby stress tensor) and will run to the bottom, the result of Rivlin and Thomas
the path-independent integral for energy release rate, given by Eqs. 4 and 10 is recovered.
Simple extension tear test sample for rubber with Configurational Mechanics 223
4 A new derivation this way, only the material transformation, i.e. defined
in M3 , between (C0 ) and (C0 ) has to be considered, the
Our method to determine the tearing energy of the trou- forces applied to the sample being known (Fig. 2b).
sers test sample is based on the general theory of Confi- The analysis of the problem proposed by Rivlin and
gurational Mechanics. Before presenting our approach, Thomas (1953) and recalled in Sect. 3.1 is adopted;
some basic results concerning the configurational stress regions A, B, C and D of (C0 ) are transformed into
tensor are recalled. Only results necessary for our deri- A , B , C and D of (C0 ) in the following manner:
vation are given, for more details the reader can refer
to Maugin (1993, 1995). – the two regions A are transformed into the two
regions A by the material translation −dc e1 of
the material surface a h 0 (−e1 ) (grey cross-section
4.1 Basic definitions in Fig. 2c),
– the region B is transformed into the region B by the
The configurational stress tensor permits to quantify material translation −dc e1 of the material surface
the evolution of the reference configuration of a given 2a h 0 e1 (grey cross-section in Fig. 2c),
body subject to deformation. This tensor, denoted – the regions C and D are transformed into regions
through the rest of the paper, is defined as C et D by a ‘rigid body material motion’.
5 Conclusion References
In this paper, the derivation of the energy release rate Andriyana A, Verron E (2007) Prediction of fatigue life impro-
vement in natural rubber using configurational stress. Int J
of the simple extension tear sample was revisited in the Solids Struct 44:2079–2092
framework of Configurational Mechanics. Obviously, Eshelby JD (1951) The force on an elastic singularity. Phil Trans
such a derivation is possible because of the simpli- R Soc Lond A 244:87–112
city of the mechanical fields in the different regions Eshelby JD (1975a) The elastic energy-momentum tensor. J
Elast 5(3-4):321–335
of the sample: the tearing energy reduces to the value
Eshelby JD (1975b) The calculation of energy release rates. In:
of one of the component of the tensor and does not Sih GC, van Elst HC, Broek D (eds) Prospects of fracture
involve a surface integral. In more general cases, the mechanics. Noordhoff, Leyden pp 69–84
use of path-integrals, i.e. J -type integrals, is manda- Greensmith HW, Thomas AG (1956) Rupture of rubber - III -
Determination of tear properties. Rubber Chem Technol
tory because the configurational stress tensor, i.e. the
29:372–381
integrant, is not uniform. Nevertheless, such a study is Griffith AA (1920) The phenomena of rupture and flow in solids.
useful to illustrate the relevance of the configurational Phil Trans R Soc Lond A 221:163–198
mechanics quantities to determine changes in energy Kienzler R, Herrmann G (1997) On the properties of the Eshelby
tensor. Acta Mech 125:73–91
between different reference configurations of a given Knowles JK, Sternberg E (1972) On a class of conservation laws
body. Moreover, this type of approach can be conside- in linearized and finite elastostatics. Arch Rat Mech Anal
red to design new test samples for fracture mechanics: 44:187–211
similarly to the simple tests for the mechanical res- Maugin GA (1993) Material inhomogeneities in elasticity.
Chapman and Hall, London
ponse of materials (uniaxial extension, simple shear, Maugin GA (1995) Material forces: concepts and applications.
equibiaxial extension . . .), new tear samples can be Appl Mech Rev 48:213–245
proposed by studying the configurational stress field Rivlin RS, Thomas AG (1953) Rupture of rubber. I. Characteris-
and the possibility to calculate it analytically. Moreo- tic energy for tearing. J Polym Sci 10:291–318
Thomas AG (1994) The development of fracture mechanics for
ver, it has been shown here that the components of elastomers. Rubber Chem Technol 67:G50–G60
are appropriate quantities for the study of macroscopic Verron E, Andriyana A (In Press) Definition of a new predictor
cracks in rubberlike materials. Motivated by this small for multiaxial fatigue crack nucleation in rubber. J Mech
study, the tensor can be considered as an efficient tool Phys Solids. doi:10.1016/j.jmps.2007.05.019
Verron E, Le Cam J-B, Gornet L (2006) A multiaxial criterion for
for the study of the evolution of microscopic defects crack nucleation in rubber. Mech Res Commun 33:493–498
in elastomers; such an approach was recently proposed
by Verron et al., 2006, Verron and Andriyana, In Press,
Andriyana and Verron, 2007) to predict fatigue crack
initiation in rubber under multiaxial loading conditions.
Stress-driven diffusion in a deforming and evolving elastic circular
tube of single component solid with vacancies
Chien H. Wu
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 227
doi: 10.1007/978-1-4020-6929-1_21, © Springer Science+Business Media B.V. 2007
228 C. H. Wu
strain energy in dX S F → dx. The derivative of this V x1
J = 1 = 1 + 3ε∗ , ε∗ = η1 x1 − xave
∗ 1
free energy with respect to the eigentransformation is a V xave
generalized configurational stress, which, in the limit as
the eigentransformation tends to the identity transfor- = η1 c1 − cave
1
/c (2.2)
mation, tends to the classical energy momentum tensor where 1 1 1
of Eshelby (1970). Moreover, the generalized config- V xave V̄ 1 xave − V̄ 0 xave
3η =
1 = ,
urational stress may be directly linked to the chemical V xave1 1
V xave
potential, a result that was first reported by Wu (2001). ∂
The eigentransformation in a single-component V xave
1
= 1 V xave 1
. (2.3)
∂x
solid with vacancies as well as the three-way transfor- 1 0
Since V̄ and V̄ are the partial molar volumes, the
mations among the three sets of coordinates are explic- above identity indicates that η1 is the linear strain in
itly defined in Sect. 2 in an axially symmetric setting. the lattice due to replacing a vacancy with an atom.
The implication of a time dependent material surface The principal eigentranasformation in cylindrical
is presented in Sect. 3 where the desired boundary con- coordinates
dition is obtained. This condition requires the simulta- ⎡ ∗ now becomes ⎤ ⎡ ⎤
r 0 0 1 + εr∗ 0 0
neous satisfaction of the surface and bulk dissipation F∗ = ⎣ 0 ∗θ 0 ⎦ = ⎣ 0 1 + εθ∗ 0 ⎦,
inequalities. It involves the linear combination of the 0 0 z ∗ 0 0 1 + εz∗
chemical potential and its gradient and is, therefore, of
a radiation type. r∗ = ∗θ = ∗z = ∗
. (2.4)
2 Elasticity and diffusion in the bulk εr∗ = εθ∗ = εz∗ = ε∗
The solid body, which occupies a cylindrical region
2.1 The eigentransformation, elastic transformation V in X, is deformed into a new cylindrical region v
and deformation gradient in x : (r, θ, z). The associated principal deformation
gradient in cylindrical coordinates is
⎡ ⎤
Let a single-component solid be defined by a molar r 0 0
concentration (mol/m3 ) c1 and a vacancy concentration F = ⎣ 0 θ 0 ⎦
c0 , so that the total molar concentration c, which is the 0 0 z
number of available lattice sites per unit volume, and ∂r
the associated mole fractions x1 = c1 /c and x0 = c0 /c r = = 1 + εr = 1 + ∂u/∂r
∂R
satisfy r
and θ = = 1 + εθ = 1 + u/r (2.5)
R
∂z
c = c 1 + c 0 , 1 = x1 + x0 . (2.1) z = = 1 + εz = 1 + constant
∂Z
where u = u (r ) is the radial displacement and the
Since the two mole fractions are not independent, we
condition of a generalized plane strain is enforced. As a
use x1 to define the composition.
matter of practice the spatial radial coordinate r , instead
Let V (xave
1 ) be the molar volume (m3 /mol) of the
of the reference R, will be used as the independent var-
solid at zero pressure, and mole fraction x1 = xave
1 . Iso-
iable in all linear descriptions of the problem. Finally,
thermal conditions are presumed throughout this paper.
the combination of F and F∗ is termed the elastic trans-
We associate V (xave1 ) with a uniform state of a solid
formation Fe = FF∗−1 , which has the principal form
body occupying a region V in a chosen reference cylin- ⎡ e ⎤ ⎡ ⎤
r 0 0 1 + er 0 0
drical coordinate system X : (R, , Z ). Axial symme-
Fe = ⎣ 0 eθ 0 ⎦ = ⎣ 0 1 + eθ 0 ⎦
try is implied throughout the paper. If x1 = x1 (R, t)
0 0 z e 0 0 1 + ez
is radially nonuniform, the associated V (x1 ) may be
used to compute the associated eigentransformation er = εr − ε∗
F∗ defined in terms of the Jacobian of transformation and eθ = εθ − ε∗ (2.6)
J ∗ (R, t) = V (x1 (R, t))/V (xave
1 ). In terms of a linear ez = εz − ε ∗
elastic theory to be used in the sequel, this Jacobian is This completes the kinematical description for axial
approximately symmetric problems.
Stress-driven diffusion in a deforming and evolving elastic circular tube 229
We note that the symbol τ has been reserved for lin- is emphasized that the coupling between concentration
ear elasticity stresses to avoid confusion. The Piola and elastic deformation is completely captured in Ŵε .
counterpart of (2.10) is The principal Piola stresses are now giving by
∂ Pr − Pθ ∂ψ ∂ Ŵε J ∗ ∂ ŴεS F J∗ e
Pr + = 0, Pr = = = ∗ = P ,
∂R R
(2.21) ∂r ∂r r ∂re r∗ r
∂ ŴεS F
where the use of the reference coordinate R as the Pre = , (2.29)
∂re
independent variable is introduced for clarity. Since the
molar concentrations are related by (2.1), (2.19) may ∂ψ ∂ Ŵε J ∗ ∂ ŴεS F J∗
be written Pθ = = = ∗ = ∗ Pθe ,
∂θ ∂θ θ ∂θ e θ
˙ r + Pθ
˙ θ + Pz
˙ z + µ10 ċ1 , µ10 = µ1 − µ0 , ∂ ŴεS F
ψ̇ = Pr Pθe = , (2.30)
∂eθ
(2.22)
∂ψ ∂ Ŵε J ∗ ∂ ŴεS F J∗ e
where µ10 is a relative chemical potential between µ1 Pz = = = ∗ = P ,
∂z ∂z z ∂ez ∗z z
and µ0 in that the process merely indicates the replac-
ing of a vacancy with an atom. Let g be the Gibbs ∂ ŴεS F
Pze = . (2.31)
energy and ω the grand potential. Then, for isothermal ∂ez
conditions,
Differentiating Ŵε with respect to r∗ , ∗θ and ∗z , we
g = ψ − Pr r − Pθ θ − Pz z , get the principal components of the energy-momentum
ġ = −r Ṗr − θ Ṗθ − z Ṗz + µ10 ċ1 , (2.23) tensor r , θ and z
∂ Ŵε J ∗ SF
Pr∗ =
r
= W − P r r =
e e
,
∂r∗ r∗ r∗
˙ r + Pθ
ω = ψ − µ10 c1 , ω̇ = Pr ˙ θ + Pz
˙ z − c1 µ̇10 . r ≡ r∗ Pr∗ = W − Pr r , (2.32)
(2.24)
∂ Ŵε J ∗ SF
θ
Pθ∗ = = W − Pθ θ =
e e
,
The desired free energy is constructed as follows: ∂∗θ ∗θ ∗θ
θ ≡ ∗θ Pθ∗ = W − Pθ θ , (2.33)
ψ r , θ , z , c1 , c0 = G c1 , c0
+Ŵε r , θ , z , r∗ x1 , ∂ Ŵε J ∗ SF
Pz∗ =
z
= W − Pz z =
e e
,
∗θ x1 , ∗z x1 (2.25) ∂∗z ∗z ∗z
in which z ≡ ∗z Pz∗ = W − Pz z , (2.34)
G = cG x1 , G x1 in which W and W S F are, respectively, Ŵε and ŴεS F
evaluated at R. The chemical potentials µ1 and µ0 may
= RT x1 ln x1 + 1 − x1 ln 1 − x1 now be calculated from (2.25)–(2.27) and (2.19). They
are
+int x 1 1 − x1 , (2.26)
∂ψ dŴε 1 − x1
µ1 = = G + x0 G + ,
∂c1 dx1 c
Ŵε r , θ , z , r∗ x1 , ∗θ x1 , ∗z x1
∂ψ dŴε −x1
µ0 = 0 = G − x1 G + (2.35)
= J ∗ ŴεS F re , eθ , ez , (2.27) ∂c dx1 c
where
r θ z
re = , eθ = ∗ , ez = ∗ . (2.28) dŴε ∂ Ŵε dr∗ ∂ Ŵε d∗θ ∂ Ŵε d∗z
r∗ θ z = + +
dx1 ∂r∗ dx1 ∂∗θ dx1 ∂∗z dx1
In (2.26), int is the regular solution interaction energy, d ∗
R the gas constant and T the absolute temperature. It = Pr∗ + Pθ∗ + Pz∗ (2.36)
dx1
Stress-driven diffusion in a deforming and evolving elastic circular tube 231
and ∂ Cr − C θ ∂µ10
Cr + = −c1 , (2.47)
∂R R ∂R
d∗ dε∗
= = η1 (2.37)
dx1 dx1 which is the equilibrium condition for the configura-
tional force system. Once again, the use of R as the
by (2.2). Making use of (2.26) and (2.32)–(2.37), we independent variable is introduced for clarity. It is noted
finally get that the configurational body force on the right-hand
side is proportional to the driving force for diffusion. It
1 − x1
µ1 = G + 1 − x1 G + 3P ∗ η1 , (2.38) is related to the energy dissipation rate per unit volume
c and unit time, ḋdissipation , or the entropy production rate,
η̇production , by
x1
µ0 = G − x1 G − 3P ∗ η1 , (2.39)
c ∂µ10
ḋdissipation = T η̇production = −Jr1 > 0. (2.48)
∂R
η1
µ10 = G + 3P ∗ . (2.40)
c By way of comparison, the energy-momentum tensor
where, after applying (2.4), (2.32)–(2.34) and (2.20), satisfies
1 ∗ 1 ∂ r − θ 1 dŴε ∂c1 1
P∗ ≡ Pr + Pθ∗ + Pz∗ = ∗ (W − J σ ) . (2.41) ∂c
3 r + = = µ10
− G ,
∂R R c dx1 ∂ R ∂R
The above is an exact expression in a nonlinear elastic- (2.49)
ity formulation. In the linear elasticity setting of (2.7)–
(2.18), this expression becomes, for plane strain, where the “body force” cannot be directly identified as
1 a material driving force.
P ∗ = ∗ [W − (1 + εr + εθ ) τ ] W − ετ − τ, In connection with our earlier linear elastic result,
ε = εr + εθ . (2.42) the chemical potential, (2.40), in conjunction with
(2.26) and (2.42), is further simplified to become
The grand potential ω of (2.24) is a functional of the
principal stretches and the relative chemical potential µ10 = RT ln x1 − ln 1 − x1 + int 1 − 2x1
µ10 , i.e., ω = ω̂(r , θ , z , µ10 ). The principal con-
η1 3η1
figurational stresses Cr , Cθ and C z are now defined − 3τ + (W − ετ ) (2.50)
in terms of ω, following the definition of the energy c c
momentum tensor (2.32)–(2.34), viz. or, after the substitution of the general elasticity solu-
∂ ω̂ tion (2.14),
Cr = ω − Pr r = ψ −µ10 c1 − Pr r , Pr = ,
∂r µ10 = RT ln x1 − ln 1 − x1 + int 1 − 2x1
1 2
(2.43) 2E η1 2E η
1
− 1−2ν c C 1 + 1−ν c x − xave
1 (2.51)
∂ ω̂ 1
Cθ = ω−Pθ θ = ψ−µ10 c1 − Pθ θ , Pθ = , + 3ηc (W − ετ )
∂θ
(2.44) where the constant C1 depends on the boundary con-
∂ ω̂ ditions of the elasticity problem, c.f. (2.16). It is also
C z = ω − Pz z = ψ − µ10 c1 − Pz z , Pz = .
∂z noted that the last term of (2.51), which involves the
(2.45) energy density, has rarely been included as a part of the
full chemical potential. Applying (2.14) and (2.2), we
It follows from the identities
obtain from the above
∂ω ∂ 2
= ω̂(r (R, t), θ (R, t), z (R, t), µ10 (R, t)), 2E η1
∂R ∂R ∂ 10 1 RT ∂c1
µ = − 2int +
∂ ω̂ ∂R c x1 1 − x1 (1 − ν) c ∂ R
c1 = − 10 , (2.46)
∂µ 3η1 ∂
and the equilibrium condition, (2.21), that + (W − ετ ) . (2.52)
c ∂R
232 C. H. Wu
The radial flux of the massive species Jr1 (R, t) is own vapor/liquid, boundary conditions for the diffusion
problem must be reexamined.
∂µ10 D 1 ∂µ10 We shall concentrate from now on situations where
Jr1 = c1 vr1 = −c1 M =− c (2.53)
∂R RT ∂R x0 = 1 − x1 1. The potentials µ1 , µ0 and µ10 may
be approximated by
where vr1 is the radial drift velocity of the massive
component, M the mobility and D the diffusivity. 3η1
µ10 −µ0 − RT ln x0 + int − τ
Combining (2.52) and (2.53), we get c
3η1
∂c1 D3η1 1 ∂ + (W − ετ ) , (2.58)
Jr1 = −Deff − x (W − ετ ) , (2.54) c
∂R RT ∂R
3η1
µ x −RT + int x0 −
1 0
τ
1 2int c
Deff = D x1 −
1−x
x1 1 RT 3η1
1 2 + (W − ετ ) µ1 µ0 . (2.59)
2E η c
+ , (2.55)
(1 − ν) RT c Ignoring the products of the derivatives of µ10 and x1 ,
we get from (2.56)
where Deff is the effective diffusion coefficient, a func-
tion of R. The effect of self-stress on Deff in an axi- ∂x1 D 1 ∂ ∂µ10
ally symmetric setting has been studied by Wang et al. = R or
∂t RT R ∂ R ∂R
(2002). Once again, the appearance of the energy-
∂x0 D 1 ∂ ∂µ0
density term, which acts like a source, appears to be = R , (2.60)
new. The diffusion equation governing the process is ∂t RT R ∂ R ∂R
and from (2.57)
∂c1 1 ∂ 1 D 1 ∂ ∂µ10 ∂c1 ∂c0 Dave ∂
∂c1
=− R Jr = Rc1 . =− = R
∂t R ∂R RT R ∂ R ∂R ∂t ∂t R ∂R ∂R
(2.56)
D3η 1 ∂
1 ∂
+ R (W − ετ ) (2.61)
Combining the above with (2.54), we obtain RT R ∂ R ∂R
where
∂c1 1 ∂ ∂c1
= Deff R
∂t R ∂R ∂R 2
1 2int 2E η1
D3η 1 ∂
1 ∂ Dave = D − + . (2.62)
+ 1
x R (W − ετ ) , (2.57) 0
xave RT (1 − ν) RT c
RT R ∂ R ∂R
which is the governing equation for the concentration.
Without the inclusion of the energy-density term, this The boundary conditions for the diffusion equations
equation is completely uncoupled from the elastic field will now be examined in conjunction with the evolu-
even though the presence of Young’s modulus in Deff tion of a phase boundary, a dividing cylindrical surface
remains. If, in addition, the solid is also in equilibrium between the solid cylinder containing vacancies and the
with its vapor/liquid then the boundary condition for vacuum surrounding taken to be a vacancy reservoir.
diffusion becomes unrelated to the underlying mechan-
ical equilibrium problem. We have thus ended up with
a chemical version of the thermal stress problem where 3 A time dependent isotropic surface
temperature is unaffected by mechanical stresses. Most
of the published results fall into this category (Larche The inner cylindrical surface R = A and outer surface
and Cahn 1978a,b, 1982). Thus, to examine the effect R = B will be considered while the end surfaces will
of applied mechanical tractions on diffusion when the not be treated. Both are assumed to be time dependent,
elastic body is not in contact and equilibrium with its i.e. A(t) and B(t). We shall first concentrate on the
Stress-driven diffusion in a deforming and evolving elastic circular tube 233
outer surface B(t). For a generic function F(R, t), the which indicates that Ḃ is just the radial drift velocity
following notation will be used: vr1 (B, t).
◦ d
F= F (B (t) , t) = Ḟ (B (t) , t) + F (B (t) , t) Ḃ. 3.2 Superficial dissipation inequality
dt
(3.1)
The governing condition for a surface can be derived
The quantity Ḃ is the radial velocity of the bounding from the second law in the form of a dissipation inequal-
surface R = B. ity (Gurtin 2000). In terms of a radial thermodynamic
force Gr B , it is
ance law is
1
d = (ω − r Pr ) + ˜
ω̃ − θ P̃θ , (3.8)
c (R, t) 2π RdR + 2π B (t) c̃ (B (t) , t)
1 1 B R=B
dt Bε
where the second term is the surface counterpart of the
= 2π B − (ε) Jr1 B − , t , (3.2) ˜ θ a sur-
first. In particular, P̃θ is a surface stress and
where Jr1 (B − , t) is an inward flux to the region Bε . face principal stretch. Simplifying the above, we get
Since B + (ε) is in the vacuum, Jr1 (B + , t) = 0. Taking
Gr B = ψ − µ10 c1 − r Pr
the time derivative and then letting ε → 0, we get
1
◦ = ψ − µ10 c1 − r Pr + (3.11)
Ḃ 0 B R=B
Ḃc0 (B, t) − Ḃc + c̃ (B, t) + c̃0 (B, t) = Jr0 (B, t) .
B which, after applying (2.25)–(2.42), may be written
(3.5)
Gr B = cµ0 − J σr + W +
Unless the thickness of the physical surface layer is B B
comparable to the surface radius, all the c̃ terms may
= cµ0B − (1 + ε B ) τr B + W B + . (3.12)
be ignored from (3.3) and (3.5). We thus have B
where a subscript B indicates evaluation at R = B.
Ḃc1 (B, t) = Jr1 (B, t) = c1 vr1 (B, t)
It is now necessary to find a way to satisfy the dis-
⇔ Ḃ c0 (B, t) − c = Jr0 (B, t) (3.6) sipation inequality (3.7). In view of (3.6), Ḃ is just the
234 C. H. Wu
radial drift velocity vr1B = vr1 (B, t) which satisfies the 4 Concluding remarks
interior dissipation inequality (2.48) via (2.53). Thus,
to satisfy both (3.7) and (2.48) in a consistent manner, Diffusion affected shape changes are usually small, for
we simply enforce the identities at R = B: most practical problems, and the involved processes
are slow. It is perhaps for these reasons that this class
D ∂µ10 D 1 of problems is not widely studied. One exception is
− = vr1B = Ḃ = − Gr B (3.13)
RT ∂ R B RT cδ impression creep, which has been persistently pursued
after by Li since 1977 (Li 2002; Chu and Li 1977). Even
where δ is the thickness of the surface, which can at there the effect of shape change on the elastic response
most be of a few atomic layer thick. For x0 = 1−x1 is deemed to be negligible. If such changes are to be rig-
1 and using (2.58) and (3.12), we deduce from the above orously followed the needed calculations can only be
incremental, as the underlying elasticity solution has to
∂µ10 ∂µ0 be continuously updated. Moreover, to the best of our
δc = Gr B or − δc = Gr B . (3.14)
∂ R B ∂ R B knowledge, none of the known results have made use
of the configurational equilibrium equations to follow
While there is nothing new with the proposition behind the process of energy dissipation. It just seems that the
(3.13), the actual form of (3.14) appears to be novel in axial symmetric equations do provide us with a simpler
the study of nonequilibrium phase-boundary motion. setting to explore and/or understand some of the open
Since Gr B is a function of µ0B , the boundary condi- questions further.
tion is of the radiation type. The appearance of δ in the
above indicates that the left-hand side is on the order of Acknowledgements The research support of the National
Science Foundation, CMS-0010077, is gratefully acknowledged.
δ/B relative to the right-hand side. Thus, the equilib-
rium condition Gr B = 0 is merely a first order approx-
imation of the actual nonequilibrium requirement. The
References
explicit form of this condition, written now in terms of
the independent variable r for a linear-theory descrip- Chu SNG, Li JCM (1977) Impression creep; a new creep test.
tion, is J Mater Sci 12:2200–2208
Abstract A crack is steadily running in an elastic contrast, for intersonic crack propagation without a
isotropic fluid-saturated porous solid at an intersonic process zone, the asymptotic stress and pore pressure
constant speed c. The crack tip speeds of interest are might display a discontinuity across two or four sym-
bounded below by the slower between the slow longitu- metric rays emanating from the moving crack tip. Under
dinal wave-speed and the shear wave-speed, and above Mode II loading condition, the singularity exponent
by the fast longitudinal wave-speed. Biot’s theory of for energetically admissible tip speeds turns out to be
poroelasticity with inertia forces governs the motion weaker than 1/2, except at a special point and along
of the mixture. The poroelastic moduli depend on the special curves of the (n, c)-plane. The introduction of
porosity, and the complete range of porosities n ∈ [0, 1] a finite length process zone is required so that 1. the
is investigated. Solids are obtained as the limit case energy release rate at the crack tip is strictly positive
n = 0, and the continuity of the energy release rate as and finite; 2. the relative sliding of the crack surfaces
the porosity vanishes is addressed. Three characteristic has the same direction as the applied loading. The pres-
regions in the plane (n, c) are delineated, depending ence of the process zone is shown to wipe out possible
on the relative order of the body wave-speeds. Mode first order discontinuities.
II loading conditions are considered, with a permeable
crack surface. Cracks with and without process zones Keywords Dynamic fracture · First order
are envisaged. In each region, the analytical solution to discontinuity · Poroelastic material · Asymptotic
a Riemann–Hilbert problem provides the stress, pore analysis · Process zone
pressure and velocity fields near the tip of the crack. For
subsonic propagation, the asymptotic crack tip fields
are known to be continuous in the body [Loret and 1 Introduction
Radi (2001) J Mech Phys Solids 49(5):995–1020]. In
Super-Rayleigh fracture propagation, and a fortiori
E. Radi (B) intersonic and supersonic fracture propagation, have
Dipartimento di Scienze e Metodi dell’Ingegneria, long been questioned. Laboratory experiments on
Università di Modena e Reggio Emilia, via G. Amendola 2, metals, and in situ observations on the earth crust have
42100 Reggio Emilia, Italy
been reported, but they were too isolated for their vali-
e-mail: eradi@unimore.it
dity to be accepted. For example, super-shear fracture
B. Loret
Laboratoire Sols, Solides, Structures, Institut National
over a long portion of the 1979 Imperial Valley fault
Polytechnique de Grenoble, BP 53X, 38041 Grenoble has been advocated by Archuleta (1984) and Spudich
Cedex, France and Cranswick (1984). Mode I supersonic crack
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 235
doi: 10.1007/978-1-4020-6929-1_22, © Springer Science+Business Media B.V. 2007
236 E. Radi, B. Loret
propagation along weak crystallographic planes in tip. Indeed, experimental setups where the energy requi-
potassium chloride has been realized by Winkler et al. red for fracture to proceed is furnished by laser-pulsed
(1970) and Curran et al. (1970). Still, the argument that expanding plasma indicate that supersonic Mode I crack
led to doubt the reality of the phenomenon and the propagation can be realized, Winkler et al. (1970),
validity of the theoretical justifications was as follows. Curran et al. (1970). However, totally supersonic crack
Even if energy considerations would not exclude super- propagation is out of the realm of this analysis which is
Rayleigh fracture and intersonic fracture propagations, concerned with intersonic propagation only.
the phenomenon should be unstable to branching. Such Indeed, this work aims at extending available theore-
fractures were thought to branch quickly, unless they tical analyzes on intersonic crack propagation in solids
were guided by a plane of weakness. and addresses the phenomenon in fluid-infiltrated po-
In fact, the argument is subtle, as it turns out that, for rous materials. The existence of cracks dynamically pro-
isotropic solids, the range of fracture speeds (cR , cE ) pagating at high speeds in such materials is expected to
between the √ Rayleigh wave-speed cR , and the Eshelby be relevant in geophysical and environmental problems,
speed cE = 2 cS , with cS shear wave-speed, is indeed including the rupture of tectonic faults, as just pointed
unstable with respect to perturbations of the fracture out, as well as in engineering applications such as
speed under most loading conditions. This is why obser- hydraulic fracturing, and general mining operations.
vations of repetitive and controlled intersonic fracture
√ The analysis is expected to be more complex than
correspond in fact to crack tip speeds higher than 2 cS . for solids for at least two reasons, namely inherent
The first laboratory experiments that seem to clearly fluid-solid couplings and the existence of three body
witness the existence of intersonic fracturing are quite wave-speeds. Indeed, a prominent characteristics of the
recent, Rosakis et al. (1999). Two plates of polyester mechanical response of fluid-infiltrated geomaterials
resin in contact along a plane of weakness were sub- like soils, rocks, and biomaterials, like articular carti-
jected to Mode II shear loading, a projectile fired by lages and intervertebral discs, lies in the significant cou-
a gas gun impacting one of the plates. Events were pling of deformation with diffusion. This feature gives
recorded through a ultra high speed camera. The publi- rise to quite distinct short- and long-term responses.
cation of the laboratory experiments of the CalTech Crack propagation in fluid saturated poroelastic
group has triggered several in situ observations of the media has been initially analyzed under quasistatic
phenomenon to be reported, and helped in structuring conditions, e.g., for shear loading by Rice and Cleary
their interpretation. Bouchon and Vallée (2003) ana- (1976); Rice and Simons (1976) and Rice (1985) and
lyzed the Kunlunshan earthquake of magnitude 8.1. later for tensile loading by Atkinson and Craster (1991).
They report that fracture ran over 400 km with an ave- Under these circumstances, the region close to the crack
rage speed of 3.8 km/s, which is higher than the shear tip is practically drained, even for rapid crack propa-
wave-speed cS 3.5 km/s of the rocks at the sites. gation. Inclusion of inertia of the sole solid phase does
The fracture speed c was sub-Rayleigh in the first hun- not alter the above observation as shown in dynamic
dred kilometers, that is c ≤ cR 0.92 cS 3.2 km/s, thermoelasticity by Atkinson and Craster (1992).
and
√ it later became equal to 5 km/s, which is close to As for fluid-infiltrated elastic-plastic media, Radi
2 cS . These in situ observations agree with the succes- et al. (2002) performed an asymptotic analysis for the
sion of events captured by the CalTech group. Similar problem of quasi-static crack propagation. The beha-
conclusions were drawn from observations on earth- vior was shown to be asymptotically drained at the
quakes that took place in Turkey in 1999, Bouchon crack tip as well. Moreover, the coupling between plas-
et al. (2001). In the same vein, Dunham and Archuleta tic dilatancy and fluid diffusion was shown to influence
(2004) view the 2002 Denali earthquake as displaying strongly the distribution of pore water pressure: by
evidence of a transition from sub-Rayleigh to super- increasing the flux of water towards the crack tip, this
shear rupture: records show two simultaneous pulses, coupling contributes to dissipate the amount of sup-
parallel and normal to the fault (supershear rupture), plied energy, and thus reduces the energy available to
followed by a normal pulse (propagating at the Ray- fracture the material.
leigh wave-speed). The analysis of crack propagation in poroelastic
Mode I intersonic or supersonic speeds require the materials without a process zone has been extended
loading to be applied on the crack faces or at the crack to a dynamic context by Loret and Radi (2001). These
Mode II intersonic crack propagation in poroelastic media 237
authors obtained a closed-form asymptotic solution for longitudinal wave-speed cL under shear-dominated
the stress, pore pressure and displacement fields near loading (Mode II). They found that the stress and velo-
the tip of a crack, rapidly propagating in a poroelastic city fields suffer infinite jumps across two symmetric
material at a subsonic speed, that is, at a speed smal- rays emanating from the moving crack tip. The resul-
ler than any of the three elastic wave-speeds. Indeed, tant stress singularity at the crack tip and along the
three elastic body waves propagate in poroelastic mate- singular rays is smaller than 1/2, Georgiadis (1986),
rials, two longitudinal waves, affecting both solid and thus yielding a vanishing energy release rate,√except
fluid phases, and one shear wave affecting the sole solid for the special intermediate crack tip velocity 2 cS .
phase. The fast longitudinal wave-speed c1 is the A stress signal emitted to front by a sub-Rayleigh
largest, whereas the order of the slow longitudinal wave- crack tip provides a vehicle for the creation of ‘daugh-
speed c2 (also called Biot’s wave-speed, or second lon- ter cracks’ propagating at speeds larger than Eshelby
gitudinal wave-speed) and of the shear wave-speed c3 wave-speed. To alleviate this discontinuous transition,
depends on material parameters, and mainly on poro- Andrews (1976) developed a model where the presence
sity. The fully dynamic analysis of crack propagation of a slip-weakening zone allows energy to be absorbed
performed by Loret and Radi (2001) involves inertia by the rupture process and the secondary cracks to pro-
of both solid and fluid phases. It reveals a qualitative pagate temporarily at the shear wave-speed cS before
change with respect to the quasi-static problem, leading coalescing at speed cL . Along this line, Broberg (1989,
to square-root singularity for the partial, total and effec- 1995, 1999) showed that the introduction of a cohe-
tive stresses as well as for the pore pressure. Moreover, sive zone extends the energetically admissible range of
these authors show that super-Rayleigh, but subsonic, speeds to the entire intersonic region cS to cL , removing
crack propagation, i.e., cR < c < min{c1 , c2 , c3 }, is the problem of vanishing energy release rate. The expe-
forbidden, because it would occur with negative energy rimental investigations made by Rosakis et al. (1999,
flow to the crack tip, which implies crack face contact 2000) on intersonic crack growth confirmed the pre-
or compressive normal tractions ahead of the crack dictions of the above theoretical analysis and motivated
tip. Therefore, the propagating speed of a remotely further analytical and numerical studies, e.g., Broberg
loaded Mode I crack in poroelastic materials can not (1999); Gao et al. (1999); Huang and Gao (2001).
exceed the Rayleigh wave-speed cR . Thus, in that res- The stability with respect to perturbations of the
pect, poroelastic materials behave similarly to homoge- crack tip speed of shear cracks with cohesion running
neous linear elastic solids, according to Freund (1979); at intersonic speeds was addressed by Burridge et al.
Washabaugh and Knauss (1994); Broberg (1989); Bro- (1979). If, in a speed range, the force required to main-
berg (1996); Broberg (1999). Note, however, that the tain steady propagation is decreasing as the crack tip
existence and nature of Rayleigh waves in poroelastic speed increases, propagation is unstable. This conclu-
materials depend on the permeability properties of the sion has been refined recently by Obrezanova and Willis
free (fractured) surface. For non-dissipative materials, (2007) who point out that stability with respect to per-
the Rayleigh wave-speed for impermeable surface has turbations of the crack tip speed depends on the details
been shown to be smaller than any of the three elastic of the applied loading. For a Dugdale–Barenblatt cohe-
wave-speeds, i.e., cR < min{c1 , c2 , c3 }. For permeable sive model, they showed that stability may take place
surfaces, a dispersive Rayleigh wave can propagate at even for speeds smaller than Eshelby speed.
a speed cR between the slow Biot’s wave-speed and the As for Mode I, investigations on subsonic super-
shear wave-speed, and thus at intersonic speed, namely Rayleigh crack propagation in linear elastic materials
c2 < cR < c3 < c1 (work in progress). by Freund (1990); Washabaugh and Knauss (1994) and
Even if it does not address solids but fluid-infiltrated Geubelle and Kubair (2001) indicate that, under tensile
media, the approach adopted here capitalizes upon the loading conditions, energy is not absorbed by the crack
analyses of intersonic crack growth in linear elastic tip but emanates from it. This conclusion is not accep-
materials performed in the last three decades, since the table on physical grounds, unless the loading is applied
early investigations of Burridge (1973); Brock (1977); directly at the crack tip, Curran et al. (1970); Broberg
Burridge et al. (1979); Freund (1979). These authors (1989).
examined the possibility of crack propagation at tip The fact that intersonic Mode I propagation is ruled
speed between the shear wave-speed cS and the out for both cohesive and non cohesive cracks for
238 E. Radi, B. Loret
remotely applied loadings is sometimes advocated to terms of the various fields close to crack tip. In Sect. 3,
explain that intersonic shear crack propagation in a the field equations are formulated in terms of eigensolu-
plane of weakness is somehow insensitive to branching. tions and their relative amplitudes are imposed by the
Still, laboratory tests show that perturbations through boundary conditions. Only permeable crack surfaces
frictionless branches of the main crack line may slow are considered. Closed form solutions are obtained, via
down intersonic perturbation, and tensile wing cracks a complex variable approach, for the three intersonic
may also develop, Biegel et al. (2007). regions (i), (ii), (iii), in Sects. 5, 6 and 7 respectively.
The present investigation aims at extending the sub- The analysis targets in the first place cracks with a pro-
sonic analysis performed by Loret and Radi (2001). cess zone for which the primary unknown functions
Focus is on the occurrence of crack propagation at inter- are obtained as solutions of inhomogeneous Riemann–
sonic crack tip speed c, ranging between the largest and Hilbert problems. With the formalism of complex
the smallest of the three poroelastic wave-speeds. The variables in place, the solutions for cracks without a
body wave-speeds depend strongly on porosity n. Their process zone are more easily derived. The analyses
relative order delineates, in the plane porosity n versus in the three regions are made as systematic as pos-
crack tip speed c, three regions referred to as (i), (ii), sible. An overview of the generic tools is presented in
(iii). An analytical approach different from that develo- Sect. 4.
ped by Loret and Radi (2001) is required for intersonic Contact with first order discontinuities which are
crack propagation. Indeed, due to the modification in typical of cracks without a process zone is established
the character of the governing differential equations, in Sect. 8. Emphasis is laid on the representation of the
which from elliptic turn hyperbolic, strong disconti- stress and velocity fields in the two phases in presence
nuity rays (Mach fronts) emanating from the crack tip and in absence of first order discontinuities. While a
and pointing to rear may appear. Since the crack speed process zone smooth discontinuities across Mach rays,
is envisaged to be larger than one or two body wave- the associated domain of influence, or effective length
speeds, there may exist one or two Mach fronts on each scale, is found to depend strongly on both the length of
side of the fracture plane. Mode II loading conditions the process zone and the singularity exponent, which
only are envisaged, and Mode I is reported in Radi and take values in the interval ]0, 1/2].
Loret (2007). The zones where intersonic propagation is energe-
Another difference with respect to Loret and Radi tically admissible are sketched in the plane (n, c). The
(2001) is the introduction of porosity-dependent moduli, influences of the type of crack (with or without a pro-
according to a simple, but realistic, macroscopic rule. cess zone), and of the porosity, are summarized and
This approach makes it possible to investigate the whole discussed in Sect. 9.
range of porosities n ∈ [0, 1], and to establish a contact Notation. The convention of summation over repea-
with linear elastic solids for a vanishing porosity. ted mute indices applies. Applied to the symmetric
The paper is organized as follows. The crack tip is second order tensor A with components Aij in cartesian
assumed to dynamically propagate under Mode II loa- axes, the divergence operator div is defined component-
ding conditions along a rectilinear path in the isotropic wise as (divA)i = Aij,j where (),j denotes the partial
poroelastic medium, at a fixed intersonic speed c, such derivative with respect to the j -th spatial coordinate.
that min{c2 , c3 } ≤ c ≤ c1 . The field and constitutive ∇ denotes the gradient operator, namely (∇a)i = a,i
equations are recorded in Sect. 2: they adopt the for- for a scalar field a, and is the Laplacian operator, i.e.,
malism due to Bowen (1982) to describe the classical a = a,ii .
Biot’s model, Biot (1941, 1956a,b,c). The solid and I is the second order identity tensor, tr is the trace
fluid inertial contributions are taken into consideration operator, i.e., for a second order tensor A, trA = Aii .
but Biot’s inertial coupling is neglected from the start. The dot product between any pair of tensors A and B
On the other hand, the mechanical coupling associa- indicates double contraction, namely A · B = tr A BT .
ted to the volume changes of the solid and fluid phases The subscripts + and − indicate the two sides of a dis-
and the viscous coupling due momentum transfers via continuity line, and [[A]] = A+ − A− denotes the jump
Darcy’s law are accounted for. However, the viscous of an arbitrary scalar or tensor A.
coupling is found to be higher order, and thus neglec- The operators Re and Im define the real and imagi-
ted in the analysis performed to describe the lower order nary parts of a complex number.
Mode II intersonic crack propagation in poroelastic media 239
2 Field and constitutive equations Let us assume that the displacement in the solid
skeleton us vanishes as r a as r tends to zero (a >
Let uk , vk and ak denote, respectively, the vectors of 0). Then, velocity vs and strain s behave as r a−1 , at
displacement, velocity and acceleration in phase least for steady state crack propagation so that one can
k = s, w. With body forces and inertial coupling neglec- replace the time derivative with a spatial derivative.
ted, the balance of momentum of phase k expresses in Moreover, it follows that ∇divus , us and as behave
terms of the apparent stress σ k and of the momentum as r a−2 as r tends to zero.
transfer p̂k , Similarly, let us assume that the displacement in
the fluid skeleton uw behaves as r b as r tends to zero
div σ k + p̂k = ρ k ak , k = s, w. (2.1)
(b > 0). Then, vw and w behave as r b−1 and ∇divuw
Let n be the porosity, that is the current volume fraction and aw behave as r b−2 .
occupied by the fluid. The apparent stress in the fluid Now, if b < a, then the leading order term in equa-
phase is expressed in terms of the intrinsic pore pressure tion (2.3)1 is that containing ∇divuw which can not
p as σ w = −np I. The apparent mass densities of the vanish by itself and thus b ≥ a. Similarly, if a < b,
two phases ρ s and ρ w are linked to the intrinsic mass then the leading order term in equation (2.3)2 is that
densities ρs and ρw of the solid constituent and fluid containing ∇divus which can not vanish by itself and
by the relations ρ s = (1 − n)ρs , ρ w = nρw . thus a ≥ b. Therefore the sole possibility is a = b. It
The total stress of the mixture σ is defined as the follows that the dissipative terms containing the velo-
sum of the apparent stresses in the phases, namely cities in (2.3) are negligible with respect to the other
σ = σ s − n p I. The latter depend on the strains in the terms as r tends to zero.
solid and fluid phases, namely s = sym ∇us and w =
sym ∇uw respectively, through the isotropic linear elas-
tic constitutive equations, that express in terms of the 2.1 Porosity-dependent elastic moduli
shear modulus µ and of the three longitudinal moduli
λss , λww and λsw , As shown by Bowen (1982) and Loret and Harireche
(1991), the longitudinal elastic moduli can be related
σ s = 2 µ s + (λss tr s + λsw tr w ) I, to the drained bulk modulus K, to Biot’s coefficient κ
−np = λsw tr s + λww tr w . (2.2) and to Skempton’s coefficient B, namely:
The momentum transfers are taken in a format that K B n2 n2 Ks2
λww = = ,
makes the balance of momentum of the fluid phase κ(1 − κ B) 1−n n
+ Ks2 − K
equivalent to Darcy’s law of seepage, namely Ks Kw
p̂w = −p̂s = −ξ (vw − vs ) + p ∇n. The coefficient κ 2 λ2
ξ = n2 ρw g/Kh is proportional to the inverse of the λsw = − 1 λww , λss = K − µ + sw .
n 3 λww
hydraulic conductivity Kh (unit: m/s). Here g is the (2.4)
intensity of the gravity.
The porosity n is assumed to maintain a constant In turn, Biot’s coefficient and Skempton’s coefficient,
and uniform distribution over space. This assumption which range between 0 and 1, are defined in terms of the
may be reasonable out of the very near crack tip zone, drained bulk modulus and of the intrinsic bulk moduli
where the effects of cavitation, micro-inhomogeneities, of the solid and fluid constituents, Ks > 0 and Kw > 0
nonlinear mechanical behavior and finite deformation respectively, as
dominate. Then, the equations of balance of momentum K 1 n K K
κ =1− , =1+ − . (2.5)
(2.1) express in terms of displacements via the consti- Ks B κ Kw Ks
tutive equations (2.2), velocities and accelerations: The existence of a positive definite strain energy func-
⎧ tion restricts the range of variation of the poroelastic
⎪
⎪ µus + (µ + λss )∇divus + λsw ∇divuw
⎪
⎪ parameters requiring that:
⎪
⎨ +ξ (vw − vs ) = ρ s as ,
(2.3) µ > 0, λww > 0, K > 0. (2.6)
⎪
⎪ s +λ w
⎪
⎪ λ sw ∇divu ww ∇divu Although the analysis addresses primarily fluid-
⎪
⎩
−ξ (vw − vs ) = ρ w aw . saturated porous media, the underlying idea is to make
240 E. Radi, B. Loret
contact with elastic solids. For that purpose, the bulk Positive definiteness of the strain energy function (2.6)
modulus and shear modulus of the drained material are ensures the three body wave-speeds to be real and legi-
assumed to depend on the porosity n. A wide range of timates the notations (2.9)1,2 . The following inequali-
data displaying the dependence in porosity of both dry ties, which derive from (2.6) as well, are instrumental
moduli and dry wave-speeds of several cemented mate- throughout the analysis:
rials are reported by Knackstedt et al. (2005). These
max{c2 , c3 } ≤ c1 ,
data are approximated according to the simple, but rea-
listic rule, c2 ≤ min{cs , cw } ≤ max{cs , cw } ≤ c1 . (2.11)
Let c0 be the so-called frozen mixture wave-speed,
K µ 1 n n q
= = 1+ q −( ) , (2.7)
Ks µs 1−q nc nc min(cs2 + cms
2
, cw
2
+ cmw
2
) ≤ c02
where · denotes the positive part of its argument. λss + 2 µ + λww + 2λsw
=
Thus the moduli Ks and µs of the solid constituent ρs + ρw
are recovered as the porosity vanishes, and Poisson’s ≤ max(cs2 + cms
2
, cw
2
+ cmw
2
). (2.12)
ratio ν of the drained material remains equal to Pois-
son’s ratio νs of the solid constituent, namely ν = When the relation of dynamic compatibility is satisfied,
νs = (3 r − 2)/(6 r + 2), with r = Ks /µs = K/µ. Biot (1956c),
Moreover, the drained moduli vanish at the upper limit cs2 + cms
2
= cw
2
+ cmw
2
, (2.13)
n = nc ≤ 1. The slopes of the function that relate
then c12 = c02 = cs2 + cms
2 , while c2 = c2 − c2 = c2 −
2 s mw w
the moduli to the porosity are enforced to vanish at
cms . The terminology will appear appropriate when the
2
n = nc , whatever the value of the parameter q ∈]0, 1[.
spectral analysis is addressed.
The value of this upper limit porosity may be as low
The two longitudinal wave-speeds c1 and c2 are
as 0.5. The consequences for the body wave-speeds of
equal when λsw = 0 and cs = cw , and then c1 =
this dependence in porosity will be analyzed in the next
c2 = cs = cw . However, this very peculiar situation
section.
will not be considered. The variation of the poroelastic
wave-speeds, c1 , c2 and c3 , with the porosity n is repor-
ted in Fig. 1. The relative order of the three poroelastic
2.2 The three regions (i), (ii) and (iii) in the plane wave-speeds delineates, in the plane porosity n versus
(n, c) crack tip speed c, three intersonic regions denoted (i),
(ii) and (iii) in Fig. 1 and defined in explicit form in
The squares of the speeds of propagation of the two lon- Table 1.
gitudinal waves and of the single shear wave in poroe- As for the dependence of the moduli in porosity,
lastic solids are defined as, a parameter q = 1/2 and a limit porosity nc = 1 are
1/2
c12 2 assumed so that the relation (2.7) simplifies to K/Ks =
= 2 cs + cw ± 2
1 2 2 1
cs − cw + 4 csw
2 2 4
, √
µ/µs = (1− n)2 . This dependence implies the modu-
c22
lus of elastic coupling λsw ≥ 0 to be positive or zero,
µ
c32 = s , (2.8) as can be checked by inserting the above relation for
ρ K/Ks in (2.4) and (2.5). This sign justifies the two last
with the notations: definitions in (2.9).
2 µ + λss λww Irrespective of the values of q ∈]0, 1[ and nc , the
cs2 = s
, cw
2
= w , csw
4
= cms
2
cmw
2
, wave-speeds c1 and c3 tend, as the porosity vanishes,
ρ ρ
λsw λsw to their counterparts in the solid constituent,
cms
2
= s , cmw 2
= w. (2.9)
ρ ρ Ks 4 µs
c12 → cL
2
= + , c22 → 0,
The longitudinal speeds (2.8)1 satisfy the following ρs 3 ρs
µs
relations: c32 → cS2 = . (2.14)
ρs
c12 + c22 = cs2 + cw
2
, c12 c22 = cs2 cw
2
− csw
4
, At small porosity, the slow longitudinal wave-speed is
csw
4
= (c12 − ck2 )(ck2 − c22 ), k = s, w. (2.10) proportional to n(1−q)/2 , so that the initial slope dc2 /dn
Mode II intersonic crack propagation in poroelastic media 241
5 longitudinal wave-speed,
cL
1 − nc nc
cE +
crack tip speed c (km/sec)
4
ρs ρw
supersonic c12 → , c22 → 0, c32 → 0. (2.15)
1 − nc nc
+
3 c1 Ks Kw
cS
cR
c3 i)
2
iii) c* c2
3 Local crack tip fields
cR c3
ii)
1
c2 A plane crack propagates at constant speed c > 0 along
subsonic
a rectilinear path in an infinite medium. The problem
0 geometry is sketched in Fig. 2. Besides a fixed carte-
0 0.2 0.4 0.6 0.8 1.0
porosity n sian system (0, X1 , X2 , X3 ), a second cartesian coor-
dinate system (0, x1 , x2 , x3 ) centered at the crack tip
Fig. 1 Variations with porosity n of the three acceleration body and moving with it in the X1 direction is used, with the
wave-speeds, c1 fast longitudinal wave-speed, c2 slow longitu- out-of-plane x3 -axis along the straight crack front.
dinal wave-speed and c3 shear wave-speed, and corresponding
intersonic regions (i), (ii) and (iii). The dashed curve c∗ (n), defi-
The process zone has a finite length L much smaller
ned by the vanishing of the parameters , Eq. (4.1), or α, Eq. than the crack length. Under steady-state crack pro-
(4.2), will be shown to separate the region (iii) in two zones pagation, an arbitrary scalar field ϕ = ϕ(X1 , X2 , t)
of admissible and non admissible crack propagation. It prolon- depends on time t through the x-coordinates, and,
gates, in the subsonic region, into a Rayleigh wave-speed cR
associated to a zero pore pressure. The parameters used for the
although an abuse of notation, it will be denoted ϕ =
solid constituent are: shear modulus µs = 20 GPa, bulk modulus ϕ(x1 = X1 −c t, x2 = X2 ). Its material time derivative
Ks = 36 GPa (Poisson’s ratio νs = 0.265), mass density ρs = satisfies the relation,
2500 kg/m3 , porosity dependence nc = 1, q = 0.5 in Eq. (2.7);
and for the fluid constituent: bulk modulus Kw = 3.3 GPa, mass ϕ̇ = −c ϕ,1 . (3.1)
density ρw = 1000 kg/m3 . For these parameters, the wave-speed
c2 is equal to the shear wave-speed c3 for n = ni−ii ∼ 0.3115,
and to the Rayleigh wave-speed cR for n = n∗ ∼ 0.245
The analysis first addresses the field equations in terms
of eigensolutions. Boundary conditions provide the
relative amplitudes of the eigensolutions.
Table 1 Definitions of the regions
Region Range of tip speed c x1+a3 x2=0 x1+a2 x2=0
crack
X1 L x1 = X1 -ct
is quite large. On the other hand, in his description process zone
of ultrasonic wave-speeds, as opposed to acceleration
wave-speeds of interest here, Berryman (1995) obtains
a much smoother relation c2 (n), through the high-
frequency device of Biot (1956b). √ Fig. 2 Problem geometry for a crack tip propagating at constant
For linear elastic solids, Eshelby speed 2 cS is speed c, with a process zone of length L. Depending on the loca-
tion of the current point in the plane (n, c), two or four symmetric
below the longitudinal wave-speed
√ cL if Poisson’s ratio
rays emanating from the crack tip (dashed lines) may, either be
νs is positive: indeed cL / 2 cS = (1+νs /(1−2 νs ))1/2 . the loci of first order discontinuities associated to a crack without
At the upper limit n = nc , the situation becomes a process zone, or, strongly influence the stress and velocity fields
similar to that of a compressible fluid with a single in the case of a crack with a process zone
242 E. Radi, B. Loret
With help of (3.9), the longitudinal eigenmodes asso- As a consequence of the spectral analysis, the lon-
ciated to (3.7), AJ = [AsJ , Aw T
J ] for J = 1, 2, can be
gitudinal and rotational displacement potentials (3.6),
shown to satisfy the relations: being real valued functions, assume the respective form:
⎧
Aw cJ2 − cs2 c2 ⎪
⎪ ϕ s (x1 , x2 ) = Re [F1 (z1 )] + Re [F2 (z2 )],
J
s = = 2 mw 2 , J = 1, 2. (3.10) ⎪
⎨
AJ cms
2 cJ − cw
−2 (3.14)
⎪ ϕ w (x1 , x2 ) = cms (c12 − cs2 ) Re [F1 (z1 )]
As for the shear eigenmode, B s is arbitrary, and ⎪
⎪
⎩
−(cs2 − c22 ) Re [F2 (z2 )] ,
B w = 0, (3.11)
and
and thus the lower order term of the pore fluid displace- ⎧
ment field is irrotational. From (2.11), the ratio Aw s ⎨ψ s (x1 , x2 ) = Im [G(z3 )],
1 /A1
w s
is deduced to be positive, while A2 /A2 is negative: the (3.15)
⎩ψ w (x , x ) = 0.
1 2
eigen-coefficients of the fluid and solid phases associa-
ted to the fast longitudinal wave are of the same sign, Here, the analytic primary functions F1 (z1 ), F2 (z2 ) and
while they are of opposite signs for the slow wave, G(z3 ) embody the real constants As1 , As2 and B s , res-
which correlatively is much more diffusive. An extreme pectively.
situation takes place when the relation of dynamic com-
patibility (2.13) holds: the fast longitudinal wave is
then non dissipative, namely Aw s 4 Overview of the analysis
1 /A1 = 1, while the
slow longitudinal wave is highly dissipative, namely
Aw s s w 4.1 The steps of the analysis
2 /A2 = −ρ /ρ .
Biot’s inertial coupling has been neglected from the
start. Still, while it is difficult to quantify, its effects may The analysis aims at obtaining the primary unknown
be interesting to explore. Some preliminary elements functions F1
(z), F2
(z) and G
(z), of the complex
have been given in Loret and Rizzi (1998). Inertial cou- variable z = x1 + i x2 , that define the stress and dis-
pling does not alter the structures neither of the secular placement potentials by (3.14), (3.15). These analytical
polynomial yielding the wave-speeds. nor of the eigen- functions satisfy already the field equations. Still, they
value problem for the solid eigenmode. On the other are linked by symmetry conditions along the crack line
hand, the structure of the fluid eigenmode is modified x2 = 0, and loading conditions on the crack faces and
by inertial coupling as transverse motions in the solid process zone. The resulting constraints are stated in the
phase carry over to the fluid phase. next Sects. 4.4 and 4.5. Through algebraic manipula-
The eigenvalues j , j ∈ [1, 3], turn out to be real tions, the various equations to be satisfied are transfor-
or purely imaginary, depending on the region (i), (ii) med into inhomogeneous Riemann–Hilbert problems
and (iii). It is instrumental to define the real constants for the function F1
(z). The functions F2
(z) and G
(z)
are then deduced from F1
(z). Due to symmetry condi-
αj = |j | = |1 − c2 /cj2 |, j ∈ [1, 3], (3.12) tions, only the upper plane Im z ≥ 0 needs to be
listed in Table 2. considered.
The spatial dependence of the potentials expresses The inhomogeneity in the Riemann–Hilbert pro-
through the complex variables blems emanates from the cohesion along the process
zone. The length L of the process zone is finite. The
zj = x1 + j x2 , j ∈ [1, 3], (3.13) cohesion along the process zone is prescribed. For plot-
√
recorded in explicit form in Table 3, where i ≡ −1. ting purposes, it will be taken either constant or linearly
244 E. Radi, B. Loret
decreasing from a maximum at the crack tip x1 = 0 to region (ii) region (iii)
0 at the end of the process zone x1 = −L. A relation α (α32 − 1)2 c12 − c22 c12 − cs2
between the cohesion in the process zone and the ope- = > 0, α2 2 > 0 .(4.3)
4α3 cs2 − c22 cs − c22
ning, or slip, offset across the two faces of the crack, is
not introduced. More refined analyzes with a process While they are both positive in regions (i) and (ii), their
length varying according to specific constitutive equa- sign is an issue of physical significance in region (iii).
tions or a rate dependent cohesion along the process As indicated above, the primary function F1
(z) is
zone, as in Samudrala et al. (2002), are not envisaged obtained as the solution of a Riemann–Hilbert problem.
either at this stage. In the three regions (i), (ii) and (iii), the two other pri-
Finally, the energetic criterion defined in Sect. 4.6 is mary functions will be shown to be proportional to F1
,
used to discriminate solutions. F2
(z2 ) = R2 F1
(z2 ),
The analysis of the three regions (i), (ii) and (iii) can i α3
be made systematic, even if the developments lead to G
(z3 ) = R3 F1
(z3 ), (4.4)
3
split region (iii) in two sub-domains. To avoid repeti-
tion, it is instrumental to introduce first a few generic with the proportionality constants,
quantities. c12 − cs2 1 c2 c12 − c22
R2 ≡ 2 , R3 = −2 . (4.5)
cs − c2 2 2α3 c3 2 cs2 − c22
4.2 Generic quantities
In regions where z2 is real, Re [F1
(z2 )] is a real valued
function of the real variable z2 that will be denoted
The solutions are characterized by a singularity
f
(z2 ), and similarly for g
(z3 ) = Im [G
(z3 )]. Then,
exponent γ . This exponent is not affected by the pre-
(4.4) specializes, for any real z2 and z3 , to
sence of the process zone. It is defined in terms of the
crack tip speed c and body wave-speeds by a region- regions (i),(iii) regions (i),(ii)
dependent expression: f
(z2 ) = R2 Re [F1
(z2 )], g
(z3 ) = R3 Re [F1
(z3 )] .
(4.6)
⎧
⎪
⎪ 4α1 α3 (cs2 − c22 )
⎪
⎪ > 0, region (i),
⎪
⎪ 4α2 α3 (c1 − cs2 ) + (α32 − 1)2 (c12 − c22 )
2
⎪
⎪
⎪
⎨ 4α
3 α1 (cs2 − c22 ) + α2 (c12 − cs2 )
tan γ π = ≡ > 0, region (ii), (4.1)
⎪ (α32 − 1)2
⎪ c12 − c22
⎪
⎪
⎪
⎪
⎪
⎪ 4 α1 α3 (cs2 − c22 ) − (1 + α32 )2 (c12 − c22 )
⎪
⎩ , region (iii).
4 α2 α3 (c12 − cs2 )
A solution is physically admissible if it corresponds Moreover, the real functions f
(z2 ) and g
(z3 ) in the
to a positive and finite energy release rate G. These appropriate regions will be shown to vanish in front of
requirements restrict the singularity exponents in the the associate rays z2 = 0 and z3 = 0 respectively.
interval ]0, 1[. In each region, a coefficient α,
⎧
⎪ α1 > 0, region (i),
⎪
⎪ 4.3 Conditions on the process zone
⎪
⎪
⎪
⎪
⎨ c12 − cs2
α = α1 + c2 − c2 α2 > 0, region (ii), (4.2) A process zone, of finite length L, provides a transi-
⎪
⎪ s 2
⎪
⎪ tion between the sound material in front of the crack
⎪
⎪ (1 + α32 )2 c12 − c22
⎪
⎩ α1 − region (iii), tip, and the cracked material. The existence of a pro-
4 α3 cs2 − c22 cess zone is motivated by, or associated with, the fol-
will be shown to play a key role in defining the sign of lowing properties: 1. one component of the stress turns
G. α and have the same sign, and they are linked by out to be finite, while, in absence of process zone, all
the relations: stress components are singular at the crack tip; 2. it
Mode II intersonic crack propagation in poroelastic media 245
makes the energy release rate more positive; 3. it intro- Moreover, the Mode II symmetry conditions require:
duces a length scale in the field equations, and the-
σ22 (x1 , 0) = 0, for all x1 ;
refore smooths out the first order discontinuities that
might exist in absence of the process zone. The actual p(x1 , 0) = 0, for x1 > 0. (4.12)
efficiency of the process zone in achieving these goals In view of the symmetry condition (4.12)1 , the condi-
will be considered in the analysis. tion along the process zone may be recast in the follo-
For fluid-saturated porous media, the boundary wing format:
conditions along the process zone may be phrased in s
σ12 (x1 , 0) = σ12 (x1 , 0)
terms of an effective stress. Two prominent effective ⎧
stresses for a fluid-saturated porous medium are the ⎨ τ0 (x1 ) − tan φ κ̃ p(x1 , 0) for − L < x1 < 0,
intergranular, or Terzaghi’s, effective stress σ
, = (4.13)
⎩
0 for x1 < −L.
σ = σs − np I = σ
− p I
⇒ σ
= σ + p I = σ s + (1 − n) p I, (4.7)
4.5 Flow conditions along the crack surfaces
and the elastic, or Biot’s, effective stress σ
,
The pore pressure vanishes on the crack surfaces and
σ
= σ + κ p I = σ s + (κ − n) p I
process zone if these are permeable, namely:
2
⇒ σ = K − µ tr s I + 2 µ s .
(4.8) p(x1 , 0) = 0, for x1 < 0. (4.14)
3
These two effective stresses coincide only when the Alternatively, the crack faces may be considered to
solid constituent is incompressible, i.e., if κ = 1. For an be impermeable: the normal flux to the crack surface
anisotropic solid phase, the pore pressure would contri- vanishes. Darcy’s law is equivalent to the balance of
bute in a non isotropic way to Biot’s stress. In an iso- momentum of the fluid phase (2.1)2 , namely n (vw −
tropic context, the two effective stresses can be cast in vs ) = −Kh /(ρw g) (∇p+ρw aw ). The asymptotic ana-
a unified format through the scalar κ̃, with κ̃ = 1 for lysis of the field equations neglects the flux of water
Terzaghi’s effective stress and κ̃ = κ for Biot’s effec- with respect to the other terms in that equation, that is
tive stress. the rhs of that relation vanishes identically in the body.
The process zone is endowed with a given spatial dis- The impermeability condition consists then in enfor-
tribution of cohesive stress τ0 (x1 ), x1 ∈ [−L, 0], and cing the lower order term to vanish, namely
with a friction angle φ ≥ 0 as suggested by Rudnicki n (v2w − v2s )(x1 , 0) = 0, for x1 < 0. (4.15)
(2001). The stress components are constrained to
satisfy the condition Comments on a poroelastic model with a process zone:
1. The validity of the assumption of a permeable crack
σ12 (x1 , 0) + tan φ σ22 (x1 , 0) = τ0 (x1 ), for surface in an intersonic crack propagation context might
−L < x1 < 0, (4.9) be questioned. An alternative would have been to assume
an impermeable crack surface, if the material behavior
where is expected to be governed by its short term response.
s Rice and Simons (1976) consider a crack with a pro-
σ22 (x1 , 0) = σ22 (x1 , 0) + (κ̃ − n) p(x1 , 0)
cess zone, steadily moving, at speed well in the subso-
= σ22 (x1 , 0) + κ̃ p(x1 , 0). (4.10)
nic region: they assume permeable crack surfaces, i.e.,
a zero pore pressure on the crack faces and along the
process zone.
4.4 Mode II loading conditions
In fact, Rice and Simons (1976) find that, for a stea-
dily advancing crack tip, the behavior appears to be
Under Mode II loading conditions, the displacements
drained in a small region whose characteristic size is
us1 and uw
1 vanish ahead of the crack tip along x2 = 0 that of the diffusion length. This is because, in their
for x1 > 0, as well as their derivatives with respect to
analysis, the pore pressure, in contrast to the total or
x1 , namely:
apparent stresses, is regular and therefore much smal-
uk1,1 (x1 , 0) = 0, for x1 > 0, k = s, w. (4.11) ler than the stress. Outside this region, the undrained
246 E. Radi, B. Loret
behavior is rapidly recovered. Moreover, they note that a = c2 for the longitudinal Mach line and a = c3 for
the size of the drained region is inversely proportional the shear Mach line. The energy dissipation along these
to the crack tip speed, so that the validity of a drained rays should be estimated.
behavior along the crack surfaces becomes more and For that purpose, it is instrumental to generalize to
more questionable as the crack tip speed increases. poroelastic materials the notion of scalar driving trac-
2. For impermeable crack surfaces, the symmetry tion td = td (x1 , x2 ) on the singular ray, introduced for
condition on the crack line require the pore pressure solids by Abeyaratne and Knowles (1990), and satis-
to vanish ahead of the crack tip. However, behind the fying the condition a td ≥ 0. The generalization is
crack tip, the pore pressure may well not vanish and taken in the format,
be discontinuous across the crack. The possible res-
td = 1
2 [[σ s · s − np tr w ]]
trictions to which the pore pressure might have to be
submitted along the process zone represent another − 21 (σ s+ + σ s− ) · [[ s ]]
issue. + 21 (np+ + np− ) [[tr w ]]. (4.18)
3. The cohesion along the process zone and the crack
tip speed are assumed to be prescribed independently. Since the constitutive equations (2.2) linking the gene-
In fact, the actual speed of propagation is more proba- ralized stress (σ s , −np I) to the generalized strain
bly governed by energy considerations: for example, it ( s , w ) are linear, the driving traction on the singu-
might be expected to be the speed that corresponds to lar ray vanishes, and thus there is no dissipation when
a critical value of the energy release, which is proper the singular ray moves through the material. As a result,
to the material. Alternatively, a criterion in terms of the energy flux toward the crack tip is not affected by
the displacement offset at the end of the process zone the presence of a discontinuity ray, and the expressions
x1 = −L is used in Palmer and Rice (1973). of the energy release rate above hold for subsonic and
intersonic propagations.
singularity exponent
0.2
now expressed in terms of the displacement potentials
0.3 0.4
(5.1):
0.5
⎧ s
⎪
⎪ σ11 /µ = (1 + α32 + 2 α12 ) Re [F1
(z1 )] 0.2
⎪
⎪
⎪
⎪ +(1 + α32 − 2 α22 ) f
(z2 ) + 2 α3 g
(z3 ), 0.7
⎪
⎪
⎪
⎪
⎪
⎪ 0.1
⎪
⎪ s /µ = (α 2 − 1) (Re [F
(z )] + f
(z ))
σ22
⎪
⎪ 1 1 2
⎪
⎪
3 0.9
⎪
⎨
−2α3 g (z3 ),
0.0
(5.2) 0 0.2 0.4 0.6 0.8 1.0
⎪
⎪ s
⎪
⎪σ12 /µ = −2α1 Im [F1 (z1 )] + 2α2 f (z2 )
⎪ (c- c0)/(c1- c 0)
⎪
⎪
⎪
⎪
⎪ +(α32 − 1) g
(z3 ),
⎪ Fig. 3 Intersonic region (i). Singularity exponent γ of mode II,
⎪
⎪ as function of the relative crack tip speed (c −c0 )/(c1 −c0 ), with
⎪
⎪ 2
⎪
⎪ −np/µ = (1 + α32 ) cmw −2
(c1 − cs2 ) Re [F1
(z1 )] c0 = max(c2 , c3 ), for different porosity levels. Mode II√square-
⎪
⎪
⎩
root singularity, associated to γ = 1/2, occurs for c = 2 c3 at
−(cs − c2 ) f (z2 ) .
2 2
the limit of a vanishingly
√ small porosity, so that the special status
enjoyed by the speed 2 cS in linear elastic solids is retrieved
Similarly, the displacements (3.2) in the solid and fluid
phases become:
⎧ s are then obtained in terms of F1
(x1 ). The latter can in
⎨ u1 = Re [F1
(z1 )] + f
(z2 ) + α3 g
(z3 ), turn be shown to satisfy the set of conditions:
(5.3) ⎧
⎩ s
u2 = −α1 Im [F1
(z1 )] + α2 f
(z2 ) − g
(z3 ), ⎪
⎪ Re [F1
(x1 )] − Im [F1
(x1 )]
⎨ τ (x1 )
= , for x1 < 0, (5.6)
⎪ µ 2α
and ⎪
⎩
⎧ w 2 Re [F1
(x1 )] = 0, for x1 > 0,
−2
⎪ u1 = cms
⎪
⎪
(c1 − cs2 ) Re
[F1 (z1 )] where = tan γ π has been defined in (4.1), and α in
⎪
⎨ 2 2
−(cs − c2 ) f (z2 ) ,
(5.4) (4.2). Conditions (5.6) may be interpreted as an inho-
⎪ mogeneous Hilbert problem for the function F1
(z),
⎪
⎪ −2
uw = −cms α1 (c12 − cs2 )Im [F1
(z1 )]
⎪
⎩ 2
+α2 (cs − c22 )f
(z2 ) .
2 which, in the upper half-plane Im [z] ≥ 0, admits the
following solution (see Appendix A):
0
A sin γ π 1 1 |t|γ τ0 (t)
F1 (z) = γ + dt,
z 2 π µ α i zγ −L t − z
5.1 The Riemann–Hilbert problem (5.7)
The boundary conditions (4.11–4.14) along the crack defined up to the purely imaginary constant A.
line are expressed in terms of the primary functions via The variations of the exponent γ with the crack tip
the apparent stresses and pore pressure (5.2) and displa- speed are shown on Fig. 3. The singularity exponent
cements (5.3–5.4). The functions f
(x1 ) and g
(x1 ), under mode II loading conditions is smaller than 1/2,
for every value of c within the considered intersonic
c12 − cs2 range. Moreover, the exponent γ attains a maximum
f
(x1 ) = Re [F1
(x1 )], at an intermediate crack tip speed, which increases as
cs2 − c22
the porosity level n decreases. As n → 0, the maxi-
α32 − 1 c12 − c22
g
(x1 ) = Re [F1
(x1 )], for all x1 , mum approaches 1/2, so that the square-root singu-
2α3 cs2 − c22 larity of the special case observed for Mode II crack
(5.5) propagation in linear elastic materials at the particular
248 E. Radi, B. Loret
√
speed c = 2 cS (Burridge et al. 1973; Freund 1979) where Jτ (z, γ ) is defined by (5.15) below. With condi-
is retrieved. Along the boundary between the regions tions (5.5) and (5.8), Lemmas 2 and 3 in Appendix B
(i) and (iii), where α3 = 0, and porosities are smal- provide the remaining primary functions as indicated
ler than 0.3115, the exponent vanishes as indicated by in (4.4–4.6).
Eq. (4.1)1 . For larger porosities, along the boundary
between the regions (i) and (ii), where α2 = 0, the
variation of the exponent is mainly governed by the 5.3 Generic integrals
term α32 − 1. When this term vanishes, the exponent
√ event takes place for c2 = 0 and
is equal to 1/2. This It is instrumental to introduce two generic singular inte-
n = 1, and c2 = 2 c3 and n around 0.7 as displayed grals, Iτ (z, γ ) and Jτ (z, γ ) defined, for any complex
on Fig. 3. z = |z| ei θ, as
0
z τ0 (t)
Iτ (z, γ ) = | |1−γ dt, (5.14)
5.2 Matching the singular and finite solutions ahead −L t t −z
of the crack tip and
0 z1−γ τ0 (t)
Conditions (5.5) and (5.6)2 imply the relations: Jτ (z, γ ) = dt
−L |t|1−γ t − z
Re [F1
(x1 )] = f
(x1 ) = g
(x1 ) = 0, i(1−γ ) θ
=e Iτ (z, γ ). (5.15)
for x1 > 0. (5.8)
For a point z = x1 ∈] − L, 0[ of the process zone, the
Then, from (5.2)3 and (5.7), the shear stress ahead of
above integrals are contributed by a Cauchy principal
the crack tip is given by
value and an imaginary part,
s 2µ
σ12 (x1 > 0, 0) = γ α i A
x1 Iτ (x1 , γ ) = −
Iτ (x1 , γ ) + i π τ0 (x1 ), (5.16)
0
sin γ π |t|γ τ0 (t) and
+ dt . (5.9)
2 π µ −L t − x1
Jτ (x1 , γ ) = − cos(γ π ) −
Iτ (x1 , γ ) − sin(γ π ) π τ0 (x1 )
Due to the presence of a process zone, the shear stress
γ s + i (sin(γ π ) −
Iτ (x1 , γ ) − cos(γ π ) π τ0 (x1 )) .
must be bounded at the crack tip, namelyx1 σ12 (x1 , 0)→
0 as x1 →0+ , and this condition yields the constant A, (5.17)
sin γ π 1 0 τ0 (t) The double integral −
I−Iτ (γ ),
iA= dt. (5.10) 0 0
2π µ α −L |t|1−γ x τ0 (x) τ0 (t)
For a remotely loaded mode II crack, the shear stress −Iτ (γ ) = − − | |1−γ
I− dx dt ≥ 0,
−L −L t t −x
field far ahead of the crack tip, namely for x1 = D
(5.18)
L, must match the singular solution,
s iA KII is non negative under mild conditions on the spatial
σ12 (D, 0) = 2µ α γ = √ , (5.11) variation of the cohesion along the process zone.
D 2π D γ
Indeed, let us consider two points (x, t) and (t, x) sym-
where KII is the intersonic mode II stress intensity fac-
metric wrt the diagonal t = x. With X ≡ x/t ≥ 0, the
tor. The latter results from elimination of the constant
sum of the contributions of these two points is
A between (5.10) and (5.11):
0 x t 1
2 τ0 (t) | |1−γ − | |1−γ
KII = sin γ π dt. (5.12) t x (1 − x/t) t
π −L |t| 1−γ
If the cohesion along the process zone is constant, a positive. This is the case when in particular the cohesion
contour integration gives, for x ∈] − L, 0[, is constant along the process zone or when it varies
0 linearly, as τ0 (x) = (1 + x/L) τ0 .
x dt
Iτ (x, γ ) ≡ τ0 − | |1−γ
− Therefore, under these circumstances, crack propa-
−L t t − x
gation is energetically admissible.
π A polar representation of the stress and velocity
= + N (x, γ ) τ0 ,
tan γ π components is shown on Fig. 4 at a particular point
−x/L −γ
y of the region (i). The fields associated to the compa-
N(x, γ ) ≡ dy > 0. (5.20)
0 1 −y rison crack without a process zone and to the crack
Further integration by parts yields: with a process zone are quite similar in front of the
0 crack tip. They differ behind the crack tip, since the
−
I−
Iτ (γ ) ≡ τ0 −Iτ (x, γ ) dx two Mach lines are directed toward the rear. The crack
−L model with a process zone wipes out the jumps across
π 1 the two Mach lines associated to the comparison crack
= + L τ02 , (5.21)
tan γ π 1−γ without a process zone to be introduced in Sect. 8.
which is strictly positive for γ in the intervals
· · · [−2, −1.459[∪[−1, −0.430[∪[0, 1[∪[2, 2.43[· · · .
6 Crack propagation within the region (ii)
5.4 The energy release rate Within the intersonic region (ii), namely for c2 > c >
c3 as shown by Fig. 1, the sole space variable z3 is real,
Upon insertion of the expressions (5.13) and (4.4), or Table 3, so that Im [G(z3 )] is a real valued function of a
(5.5), into the displacement (5.3)1 , and use of (5.17), real variable and it will be denoted by g(z3 ). Then, the
the rate of sliding on the crack can be cast in the form: displacement potentials (3.14), (3.15) may be recast in
the following form:
1 c2 c12 − c22
us1,1 (x1 < 0, 0+ ) = Re [F1
(x1 )]
2 c32 cs2 − c22 ⎧ s
⎪
⎪ ϕ (x1 , x2 ) = Re [F1 (z1 )] + Re [F2 (z2 )],
(sin γ π )2 c2 c12 − c22 ⎪
⎪
=− (5.22) ⎪
⎪
4π µ c 2 c2 − c2 ⎨ϕ w (x , x ) = c−2 (c2 − c2 ) Re [F (z )]
⎪
3 s 2 1 2 ms 1 s 1 1
1 π τ0 (x1 ) (6.1)
−Iτ (x1 , γ ) − . ⎪
⎪ −(cs − c2 ) Re [F2 (z2 )] ,
2 2
α tan(γ π ) ⎪
⎪
⎪
⎪
⎪
⎩ s
The energy release rate results from (4.17) as, ψ (x1 , x2 ) = g(z3 ).
(sin γ π )2 c2 c12 − c22 1 The apparent stresses and pore pressure (3.5) can now
G= −
I−
Iτ (γ )
2π µ c32 cs2 − c22 α be expressed in terms of the unknown functions F1 (z1 ),
0
π F2 (z2 ) and g(z3 ):
− τ02 (x1 ) dx1 . (5.23) ⎧ s
tan(γ π ) −L ⎪ σ /µ = (1 + α32 + 2 α12 ) Re [F1
(z1 )]
⎪
⎪ 11
For 0 < γ < 1/2, the coefficient α, as defined by ⎪
⎪
⎪ +(1 + α3 + 2 α2 ) Re [F2 (z2 )] + 2 α3 g (z3 ),
2 2
⎪
Eqs. (4.2–4.3), is positive. Moreover, with N defined ⎪
⎪
⎪
⎪
by (5.20), the last term in (5.23) can be recast in the ⎪
⎪
⎪
⎪ s /µ = (α 2 − 1) (Re [F
(z )]
σ22
⎪
⎪ 3 1 1
following format, ⎪
⎪ +Re [F2
(z2 )]) − 2α3 g
(z3 ),
0 0 ⎨
x τ0 (t) − τ0 (x) (6.2)
τ0 (x) | |1−γ dt dx ⎪
⎪
−L −L t t −x ⎪
⎪
s
σ12 /µ = −2α1 Im [F1 (z1 )] − 2α2 Im [F2 (z2 )]
0 ⎪
⎪
⎪
⎪ + (α32 − 1)g
(z3 ),
+ N (x, γ ) τ02 (x) dx. ⎪
⎪
(5.24) ⎪
⎪
−L ⎪
⎪ 2
⎪
⎪ 2 −2
Under mild conditions on the variations of the cohesion ⎪
⎪−np/µ = (1 + α3 ) cmw (c1 − cs )Re [F1 (z1 )]
2
⎪
⎩
along the process zone, this term can be shown to be −(cs2 − c22 ) Re [F2
(z2 )] .
250 E. Radi, B. Loret
Similarly, the displacements (3.2) become in the solid 6.1 The Riemann–Hilbert problem
phase:
The introduction of the solid and fluid displace-
⎧
⎪ ments (6.3)1 and (6.4)1 into the Mode II symmetry con-
⎨us1 = Re [F1
(z1 )] + Re [F2
(z2 )] + α3 g
(z3 ),
(6.3) ditions (4.11) results in restrictions of the unknown
⎪
⎩us = −α Im [F
(z )] − α Im [F
(z )] − g
(z ), functions:
2 1 1 1 2 2 2 3
⎧
⎪
⎪Re [F1
(x1 )] + Re [F2
(x1 )] + α3 g
(x1 )
and in the fluid phase: ⎪
⎪
⎪
⎪
⎨ = 0, for x1 > 0,
⎧ 2
⎪ uw −2
(6.5)
1 = cms (c1 − cs ) Re [F 1 (z1 )]
2
⎪
⎪ ⎪
⎪
⎪ ⎪
⎪(c1 − cs ) Re [F1 (x1 )] − (cs − c2 ) Re [F2 (x1 )]
2 2 2 2
⎪
⎪ ⎪
⎨ −(cs2 − c22 ) Re [F2
(z2 )] , ⎪
⎩ = 0, for x > 0.
(6.4) 1
⎪
⎪ −2
⎪
⎪
⎪ uw
2 = −cms α1 (c1 − cs ) Im [F1 (z1 )]
2 2
⎪
⎩ −α (c2 − c2 ) Im [F
(z )] . The apparent stress components and pore pressure
2 s 2 2 2 being expressed by (6.2), the conditions on the tractions
Mode II intersonic crack propagation in poroelastic media 251
singularity exponent γ
⎪
⎪
0.7
⎪
⎪
⎪
⎪ −2α3 g (x1 ) = 0, for all x1 , 0.5
⎪
⎪
0.3
⎪
⎪ 0.4
⎪
⎨(c2 − c2 ) Re [F
(x )] − (c2 − c2 )
1 s 1 1 s 2
(6.6) 0.2
⎪ × Re [F2
(x1 )] = 0, for all x1 ,
⎪
n = 0.35
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ −2α1 Im [F1
(x1 )] − 2α2 Im [F2
(x1 )] 0.1
⎪
⎪
⎩
+(α32 − 1) g
(x1 ) = τ (x1 )/µ, for x1 < 0.
0.0
Manipulations of (6.6)1,2 show Re [F2
(x1 )] and g
(x1 ) 0 0.2 0.4 0.6 0.8 1.0
(c- c3)/(c2-c 3)
to be proportional to Re [F1
(x1 )]:
Fig. 5 Intersonic region (ii). Singularity exponent γ of Mode
c12 − cs2 II, as function of the crack tip speed c, for several porosities n.
Re [F2
(x1 )] = Re [F1
(x1 )],
cs2 − c22 For sufficiently large porosity n, Mode II square-root√singula-
rity, given by → ∞ or γ = 0.5, occurs for c = 2 c3 . At
the interface with the subsonic regime, the singularity exponent
c12 − c22 α32 − 1
g
(x1 ) = Re [F1
(x1 )], vanishes
cs2 − c22 2α3
for all x1 . (6.7)
A polar representation of the stress and velocity
Further substitution of these relations in (6.5) then components is shown on Fig. 6 at a particular point of
results in: the region (ii). The behavior is qualitatively similar to
that of region (i), to within the fact that there is now a
Re [F1
(x1 )] = Re [F2
(x1 )] = g
(x1 ) = 0, single Mach line in the upper plane.
⎧ s ⎧ s
⎪
⎪ σ11 /µ = (1 − α32 + 2α12 ) Re [F1
(z1 )] ⎨ u1 = Re [F1
(z1 )] + f
(z2 ) + α3 Re [G
(z3 )],
⎪
⎪
⎪
⎪ +(1 − α32 − 2α22 ) f
(z2 ) (7.3)
⎪
⎪ ⎩
⎪
⎪ +2α3 Re [G
(z3 )], us2 = −α1 Im [F1
(z1 )] + α2 f
(z2 ) − Im [G
(z3 )],
⎪
⎪
⎪
⎪
⎪
⎪ and in the fluid phase:
⎪
⎪ s /µ = − (1 + α 2 ) (Re [F
(z )] + f
(z )) ⎧ w 2
⎪
⎪ σ22 1 1 −2
(c1 − cs2 )Re [F1
(z1 )]
2
⎪
⎨
3
⎪
⎪ u1 = cms
+2α3 Re [G
(z3 )] , ⎪
⎪
(7.2) ⎨ −(cs2 − c22 ) f
(z2 ) ,
⎪
⎪
⎪
⎪ ⎪ (7.4)
⎪
⎪ s /µ = − 2α Im [F
(z )] − 2α f
(z )
σ12 ⎪
⎪ −2
uw = −cms α1 (c12 − cs2 ) Im [F1
(z1 )]
⎪
⎪
1 1 1 2 2 ⎪
⎩ 2
⎪
⎪ +(1 + α32 ) Im [G
(z3 )] , +α2 (cs − c22 )f
(z2 ) .
2
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ 2
⎪ 2 −2
⎪ −np/µ = (1 − α3 ) cmw (c1 − cs ) Re [F1 (z1 )]
2
⎪
⎩
7.1 The Riemann–Hilbert problem
−(cs − c2 ) f (z2 ) .
2 2
singularity exponent γ
0.15
(7.5)
⎪
⎪
⎪
⎪
⎪ (c2 − cs2 ) Re [F1
(x1 )] − (cs2 − c22 ) f
(x1 )
⎪ 1
⎩ 0.0
= 0, for x1 > 0.
Similarly, when expressed via the constitutive relations 0.20
0.22
(7.2), the conditions (4.11–4.13) on the tractions and
the vanishing of the pore pressure (4.14) on the crack 0.24
surfaces together with (7.5)2 yield the following system n = 0.25
of equations: -0.5
⎧ 0 0.2 0.4 0.6 0.8 1.0
⎪
⎪ (1 + α32 ) Re [F1
(x1 )] + f
(x1 )
⎪
⎪ (c- c2)/(c3-c 2) cR /cS
⎪
⎪ +2α3 Re [G
(x1 )] = 0, for all x1 ,
⎪
⎪
⎪
⎪ Fig. 7 Intersonic region (iii). Singularity exponent γ of Mode II,
⎪
⎪
⎨ as function of the crack tip speed c, for various porosities n. The
2α1 Im [F1
(x1 )] − 2α2 f
(x1 ) + (1 + α32 ) curve c = c∗ , for which = 0, is shown dashed on Fig. 1. For
(7.6)
⎪
⎪ ×Im [G
(x1 )] = −τ (x1 )/µ, for x1 < 0, sufficiently small porosity n, γ tends to 1/2 when c tends to c2 .
⎪
⎪
⎪
⎪ Since c2 tends to 0 and c3 tends to cS as the porosity vanishes,
⎪
⎪ the maximum value of the x-axis, for which γ = 1/2 can be
⎪
⎪
c12 − cs2
⎪
⎪ Re [F1
(x1 )], for all x1 . reached, is cR /cS . At the interface with the subsonic regime, the
⎩f (x1 ) = 2
cs − c22 singularity exponent is equal to ±1/2, as indicated by (4.1)3 for
α2 = 0. Indeed, the sign of the numerator in (4.1)3 changes at
Elimination of f
upon combination of (7.6)1,3 c = c∗ . Moreover, the value of the porosity n∗ for which c∗ = c2
leads to is between 0.24 and 0.25
-1 -3
0 30 60 90 120 150 180 0 30 60 90 120 150 180
polar angle polar angle
0.1 10
longitudinal Mach line
0.05
5
np 0 vw
0
-0.05 longitudinal Mach line
-0.1 -5
0 30 60 90 120 150 180 0 30 60 90 120 150 180
polar angle polar angle
wave-speed of the poroelastic material in the intersonic 8 Comparison crack without a process zone
region (iii). and first order discontinuities
7.3 The energy release rate The existence of imaginary characteristics, or equiva-
lently of real eigenvalues 2 or 3 as indicated in Table
From now on, the analysis developed in region (i), Eqs. 3, may give rise to first order discontinuities across the
(5.6–5.24), can be shown to hold formally unaltered. associated rays z2 = 0 or z3 = 0. Indeed, the real
The associated energy release rate (5.23), is positive functions (4.6) that contribute to the solution, that is,
for 0 < γ < 0.5, that is for c2 < c < c∗ , as displayed to the stress and velocity fields, vanish in front these
on Fig. 7. rays. Their continuity across these rays is an issue that
A polar representation of the stress and velocity will be shown to be solved differently by the crack
components is shown on Fig. 8 at a particular point of with a process zone and by the crack without a process
the region (iii). Once again, the behavior is qualitati- zone.
vely similar to those of regions (i) and (ii), to within Infinite jumps extending all along the Mach lines
the fact that there is a single Mach line, which is now without attenuation will be shown to exist for the cracks
longitudinal. without a process zone. In contrast, for the crack
Mode II intersonic crack propagation in poroelastic media 255
Mach line
zone. For example in region (i),
x1 - x2 tan =0
KII sin γ π H(−z2 )
f
(z2 ) = −R2 √ ,
α |z2 |γ
n
d
he ion a 2 µ 2π
ac at ed
re orm pe
ne nf t s KII sin γ π H(−z3 )
zo he i d a a dt g
(z3 ) = −R3 √ , (8.2)
|z3 |γ
t te
b aga
y
p x2 polar angle
2 µ 2π α
o
pr
Mach angle > 90 where H denotes the unit step function.
c dt x1 The above solution defining the stress and displace-
crack surface crack tip
ment, e.g., via (5.2–5.4) in region (i), the energy release
Fig. 9 The crack tip propagates at speed c to the right along the rate (4.16)2 can be cast in the form,
axis x1 , while the mechanical information propagates at speed a.
KII
2
sin γ π c2 c12 − c22 1−2γ
A first order discontinuity exists if c > a. In the intersonic range, G= lim x1
the material wave-speed a can be c2 or c3 , and the direction of 8 µ2 α c32 cs2 − c22 x1 →0+
the discontinuity is defined either by the Mach angle ζ such that
sin ζ = c2 /c or sin ζ = c3 /c, or by the polar angle θlong = π/2+
= 0 for γ < 1/2,
× (8.3)
tan−1 α2 or θshear = π/2+tan−1 α3 . The first order discontinuity > 0 for γ = 1/2,
is characterized by the fact that only a part of the plane, between
the crack line and the Mach line, is reached by the shear, or the positive sign holding where α is positive.
longitudinal, information propagated by the wave Thus the energy release rate vanishes, since in gene-
ral γ < 1/2. It is however finite and positive for
γ = 1/2. From the definition (4.1), an infinite =
endowed with a process zone, the continuity of the tan γ π is seen to correspond
solution across the rays z2 = 0 and z3 = 0 is proved in √
Appendix C. – in region (i), to n = 0 and c = √2 cS < cL ;
The developments below thus concentrate on the – in region (ii), to the curve c = 2 c3 , which, given
crack without a process zone. Emphasis is laid on points the shape of this region in the plane (n, c), occurs
(n, c) where the crack with a process zone is energe- only for sufficiently large porosities n;
tically admissible, while the energy release rate of the – in region (iii), to the segment n = 0 and c < cR ;
crack without a process zone is vanishing. To pinpoint – to the interface between the subsonic regime and part
this situation, the crack without a process zone is ter- of the region (iii) below the curve c = c∗ (n) where
med “virtual crack of comparison”. The focus will be α is positive, namely for n ≤ n∗ .
to analyze qualitatively the efficiency of the model with Note that, as the porosity decreases to zero, the tran-
a process zone to spatially smooth out the jumps of the sition to the value γ = 1/2 observed for linear elastic
fields associated to the comparison crack. Conversely, solids in regions (i) and (iii) is smooth, Figs. 3, 7.
it is worth scrutinizing which quantities remain conti- In conclusion, for a given porosity, intersonic Mode
nuous and which do not, when the length of the process II propagation is admissible only for some special
zone is decreased to zero. speeds of the crack tip, much like for linear elastic
solids.
8.1 Sketch of the solution in absence of a process zone 8.2 Nature of the field equations
The basic primary function F1
(z) can be cast in the Along a standard terminology, a partial differential
region-invariant format, equation, involving second order derivatives in both
KII 1 1 time and space, e.g., ∂ 2 ϕ/∂x12 + ∂ 2 ϕ/∂x22 − a −2
F1
(z) = √ , (8.1) ∂ 2 ϕ/∂t 2 = 0 is said to be hyperbolic if there exist
2 µ 2π α izγ
real characteristics which allow for the information to
in terms of the Mode II stress intensity factor KII . The propagate at finite speed, e.g., for a = 0 real. Other-
coefficients γ and α are defined by (4.1) and (4.2), wise, for a purely imaginary, it is said elliptic. Here, the
and the two other primary functions are given by the time derivative is given by (3.1), so that the above equa-
relations (4.4–4.6) like for the crack with a process tion becomes (1 − c2 /a 2 ) ∂ 2 ϕ/∂x12 + ∂ 2 ϕ/∂x22 = 0.
256 E. Radi, B. Loret
Thus according to the above terminology, the subsonic 8.3.1 Shear Mach line in regions (i) and (ii)
propagation of the crack tip, c < a, corresponds to an
elliptic equation. In these regions, the issue concerns the continuity of the
Since the stress is obtained from the potentials ϕ real function g
(z3 ) across the ray z3 = x1 +α3 x2 = 0,
which obey second order partial differential equations, with z3 = |z3 | ahead of the Mach line and z3 = |z3 | eiπ
the existence of real characteristics for ϕ in the hyperbo- behind the Mach line. It is instrumental to introduce
lic case corresponds in fact to a first order discontinuity, polar coordinates (r, θ ) centered at the crack tip, and to
i.e., a stress discontinuity, and, according to Hadamard express the stress and velocity in these coordinates. The
relations, to an accompanying velocity discontinuity: ray z3 = 0 is defined by the polar angle θshear = π/2 +
there exists a vector λ = (λi ) such that [[∇u]] = λ ⊗ n, tan−1 α3 . Let [[g
]] be the discontinuity of g
across the
[[v]] = −a λ, with n the unit normal to the wave front. Mach line at a current point (r, θshear ) of the Mach line.
In the present context, a first order discontinuity, The contributions of [[g
]] to the stress components, and
corresponds to the existence of a ray emanating from to the velocity components, Eqs. (5.2–5.4) in region (i)
the crack tip that limits the zone of extension of the and Eqs. (6.2–6.4) in region (ii), are collected. The sole
information propagated by the wave, as indicated by shear stress is observed to undergo a discontinuity,
Fig. 9. c2
s s s
Fig. 10 displays the variations of Mach angles with [[σrr ]] = 0, [[σθθ ]] = 0, [[σrθ ]] = µ [[g ]],
c32
the crack tip speed for two porosities. The closer
[[n p]] = 0, (8.4)
the crack tip speed to the critical wave-speed, i.e., c3
for the shear Mach line, c2 for the longitudinal Mach and, similarly the sole discontinuous velocity
line, the closer the Mach angle is to 90◦ . Conversely, the component is the solid radial velocity,
farther the crack tip speed from the critical wave-speed, c2
the more oblique is the Mach line. [[vrs ]] = [[g ]], [[vθs ]] = 0, [[vrw ]] = 0,
c3
The field equations involve four equations for four [[vθw ]] = 0. (8.5)
potentials. There exist four wave-speeds, accounting
for the multiplicity of degree two of the shear wave. While the function g
is continuous across the ray
The nature of the field equations thus depends on the z3 = 0 for the crack with a process zone, Appendix
region of the plane (n, c): C, it undergoes an infinite jump for the crack without
a process zone according to (8.2). Consequently, the
stress and velocity fields are continuous for a crack
– region (i): the equation associated to the fastest lon-
with a process zone, while they undergo, across the
gitudinal wave is elliptic, while the three others are
shear Mach line, an infinite jump that is parallel to,
hyperbolic;
and extends all along the Mach line, Fig. 11.
– region (ii): the two equations associated to the longi-
tudinal waves are elliptic while those two associated 8.3.2 Longitudinal Mach line in regions (i) and (iii)
to the shear waves are hyperbolic;
– region (iii): the equation associated to the second lon- In these regions, the concern is the continuity of the real
gitudinal wave is hyperbolic while the three others function f
(z2 ) across the ray z2 = x1 + α2 x2 = 0,
are elliptic; defined by the polar angle θlong = π/2 + tan−1 α2 .
– subsonic region: the four equations are elliptic. Collecting the contributions of the discontinuity of f
in the expressions of stress and pore pressure, Eqs. (5.2)
for region (i) and (7.2) for region (iii), result in
c2 c2
[[σrr ]] = − 2 2 + 2 µ [[f
]],
s
8.3 Discontinuities across the Mach lines c2 c3
s c2
A priori both shear and longitudinal, or dilatational, [[σθθ ]] = µ [[f
]], [[σrθ
s
]] = 0,
Mach lines may exist in the region (i), while the sole c32
shear Mach line may take place in region (ii), and the cs2 − c22 c2
[[n p]] = µ [[f
]], (8.6)
sole longitudinal Mach line in region (iii). cmw
2 c32
Mode II intersonic crack propagation in poroelastic media 257
Similarly, the discontinuity of f
, in (5.4) for region (i) sole non zero discontinuous components of the strain
and (7.4) for region (iii), yields discontinuities of the in phase k are
solid and fluid velocities, namely [[vrk ]] [[v k ]]
k
[[rθ ]] = k
, [[θθ ]] = θ , k = s, w. (8.8)
c2 2a a
[[vrs ]] = 0, [[vθs ]] = [[f
]], [[vrw ]] = 0,
c2 The first order shear discontinuity is thus clearly iso-
c2 − c2 c2 choric. Across the first order longitudinal discontinuity
[[vθw ]] = − s 2 2 [[f
]]. (8.7) on the other hand, the solid and fluid phases undergo
cms c2
volume changes that are of opposite signs, but do not
For the crack with a process zone, the function f
compensate. In fact, if the solid and fluid constituents
is continuous across the ray z2 = 0, Appendix C, and were incompressible, the volume changes in the body
so are the stress and velocity. On the other hand, for a would satisfy the constraint (1 − n) tr s + n tr w = 0,
crack without a process zone, f
undergoes an infinite and thus the jumps across the discontinuity would be
jump across the longitudinal Mach line according to such that (1 − n) [[vθs ]] + n [[vθw ]] = 0. For second order
(8.2), and so do the stress and velocity, Fig. 11. The waves, similar relations hold in terms of strain rate and
velocity jump is normal to the Mach line. acceleration, e.g., Eqs. (2.17) and (6.6) in Loret and
Harireche (1991).
8.3.3 Jumps in strain and balance of momentum Across a discontinuity line moving at speed a in the
direction n, the equation of balance of momentum, that
The strain discontinuity is obtained from the compati- each phase of the porous medium obeys, takes the form
bility relation [[∇u]] = −a −1 [[v]] ⊗ n. Therefore the [[σ k ]] · n + [[p̂k ]] + (ρ k a) [[vk ]] = 0, k = s, w. (8.9)
Fig. 11 Components of the stress and velocity which undergo a longitudinal, of the discontinuity. The first order shear disconti-
discontinuity across the Mach lines for a crack without a process nuity affects only the solid phase, while the first order longitu-
zone. The Mach rays emanating from the crack tip are directed dinal discontinuity affects both the solid and fluid phases. The
toward the rear, with respect to the direction of propagation of sketch indicates a longitudinal Mach angle larger than the shear
the crack. The entities which undergo a non zero first order dis- Mach angle, and thus corresponds to a porosity n > ni−ii in
continuity across the Mach lines depend on the nature, shear or region (i), as shown in Fig. 10(b)
258 E. Radi, B. Loret
FIELDS NON
2<c=1600m/s<c1
Longitudinal Mach line Longitudinal Mach line
0.002 0.002
Shear Mach ligne
0.0015 0.0015
0.001 0.001
0.0005 0.0005
0 0
-0.002 -0.0015 -0.001 -0.0005 0 0.0005 0.001 0.0015 0.002 -0.002 -0.0015 -0.001 -0.0005 0 0.0005 0.001 0.0015 0.002
=1000m/s<c2<c1
Shear Mach line
0.002 0.002
0.0015 0.0015
0.001 0.001
0.0005 0.0005
0 0
-0.002 -0.0015 -0.001 -0.0005 0 0.0005 0.001 0.0015 0.002 -0.002 -0.0015 -0.001 -0.0005 0 0.0005 0.001 0.0015 0.002
=1600m/s<c3<c1
Longitudinal Mach line Longitudinal Mach line
0.002 0.002
0.0015 0.0015
0.001 0.001
0.0005 0.0005
0 0
-0.002 -0.0015 -0.001 -0.0005 0 0.0005 0.001 0.0015 0.002 -0.002 -0.0015 -0.001 -0.0005 0 0.0005 0.001 0.0015 0.002
Fig. 12 Velocity fields for the comparison crack under Mode highlight the spatial heterogeneity of the velocity. The first order
II loading conditions. The velocity fields are scaled by r γ , r shear discontinuity gives rise to velocity discontinuities in the
distance to the crack tip, γ singularity exponent. The size of solid phase only, while the first order longitudinal discontinuity
the arrows pertains to each plot, and has been chosen so as to affects both phases, as sketched in Fig. 11
For the first order shear discontinuity, the stress and velo- longitudinal discontinuity, the relation (8.9) is satisfied
city discontinuities (8.4) and (8.5) can indeed be checked as well by the stress and velocity discontinuities (8.6)
to satisfy this relation, with a = c3 and, as indicated on and (8.7) with a = c2 and with n equal to −eθ (θlong ).
Fig. 9, with n equal to −eθ (θshear ), eθ (θ ) being the ortho- Here the diffusion contribution p̂k is of higher order close
radial vector (− sin θ, cos θ ). For the first order to the crack tip and does not contribute.
Mode II intersonic crack propagation in poroelastic media 259
0.0015 0.0015
0.001 0.001
0.0005 0.0005
0 0
-0.002 -0.0015 -0.001 -0.0005 0 0.0005 0.001 0.0015 0.002 -0.002 -0.0015 -0.001 -0.0005 0 0.0005 0.001 0.0015 0.002
0.0015 0.0015
0.001 0.001
0.0005 0.0005
0 0
-0.002 -0.0015 -0.001 -0.0005 0 0.0005 0.001 0.0015 0.002 -0.002 -0.0015 -0.001 -0.0005 0 0.0005 0.001 0.0015 0.002
0.0015 0.0015
0.001 0.001
0.0005 0.0005
0 0
-0.002 -0.0015 -0.001 -0.0005 0 0.0005 0.001 0.0015 0.002 -0.002 -0.0015 -0.001 -0.0005 0 0.0005 0.001 0.0015 0.002
Fig. 13 Velocity fields for the crack with a process zone under the other hand, the velocity discontinuities that are observed in
Mode II loading conditions. The velocity fields are scaled by r γ , the comparison crack are smoothed out by the process zone. The
r distance to the crack tip, γ singularity exponent. The latter is (dashed) Mach lines of the comparison crack have been kept for
the same as for the comparison crack without a process zone. On reference
8.4 Stress and velocity fields lines, a priori of infinite strength. In order to
appreciate the smoothing effects of the crack with a
With respect to the crack without a process zone, the process zone, the velocity fields, corresponding to the
crack with a process zone first improves the energe- cracks in absence and presence of a process zone, are
tic admissibility, and second removes the discontinuity displayed on Figs. 12 and 13 respectively, at a point
260 E. Radi, B. Loret
NON COHESIVE CRACK in REGION (i): scaled STRESSES COHESIVE CRACK in REGIONi ): scaled STRESSES
longitudinal shock s_rr longitudinal Mach line s_rr
0.002 0.002
0.6 0.6
-0.2 -0.2
0.001 0.001
-0.6 -0.6
-1 -1
0.0005 0.0005
-1.4 -1.4
0 -1.8 0 -1.8
-0.002 -0.0015 -0.001 -0.0005 0 0.0005 0.001 0.0015 0.002 -0.002 -0.0015 -0.001 -0.0005 0 0.0005 0.001 0.0015 0.002
the process zone perturbs this representation behind the and c1 , and tends to vanish close to c1 together with
crack tip only, over a distance of the order of the length the exponent γ , Fig. 3. In fact, G vanishes all along the
of the process zone. curves c = c3 and c = c1 together with γ . However it
Notice that the fact that, even for the crack with a is strictly positive at the interface of regions (i) and (ii).
process zone, the polar components of the stress are not 2. Cracks with and without process zones:
all bounded at the crack tip as already alluded to. The For the virtual comparison crack under Mode II, the
sole bounded stress component is the shear stress σ12 . A velocity, stress and pore pressure fields are singular
closeup of the shear stress near the process zone, shown not only at the crack tip, but also on two (or four)
on Fig. 16, indicates that the cohesion along the process symmetric rays emanating from the crack tip, where
zone extends continuously to the body in a manner that the partial stress fields undergo an infinite jump. These
depends strongly on the region of interest. The actual first order discontinuities are absent from the subsonic
value of the singularity exponent plays a prominent region, Loret and Radi (2001), and they are therefore
role on the departure between the crack with a pro- characteristic of the intersonic regions.
cess zone and the comparison crack: for example, at In presence of the process zone, crack propagation is
the point chosen in region (i), the singularity exponent admissible with a singularity exponent in general smal-
γ is very low, so that the jumps, while theoretically ler than 1/2, yielding thus a vanishingly small energy
of infinite strength, effects a very narrow thickness. release rate for a crack without a process zone. The sole
The range of influence of the discontinuity is much exceptions are the point and curves of the (n, c) plane
larger at the points (n, c), which are chosen as repre- indicated in Sect. 8.1. Indeed, then value of the singula-
sentative of regions (ii) and (iii), where γ is close to rity exponent γ = 1/2 implies the energy release rate
0.45. (8.3) to be strictly positive.
The above velocity fields were scaled by r γ , so that 3. Continuity of the energy release rate in the limit
their actual spatial variations were not apparent. Of of vanishing process length?
course, the spatial decay of the stress and velocity fields A crack without a process zone can not be thought
depends essentially on the singularity exponents, as of as the limit of a crack with a vanishing length L of
illustrated for the solid velocities on Fig. 17. process zone, since a crack with a process zone ensures
a stress component to be finite at the crack tip. Still, let
us consider the following issue: can we take the limit of
9 Summary and outlook the energy release rate as the process length L tends to
0, to infer the energetic admissibility of cracks without
To summarize the analysis, the zones where the inter- a process zone? The double integral (5.21) is central
sonic propagation of cracks with a process zone has to the issue. It indicates that the energy release rate
been shown to be admissible are displayed on Fig. 18. would vanish in the limit, except when the singularity
The results may be considered along several points exponent is zero, along the curve c = c∗ (n). On the
of view: other hand, the energy release rate for the crack without
1. Differences between the regions (i), (ii) and (iii): process zone does not vanish only if the singularity
Region (iii) is separated into two subregions by the exponent is equal to 1/2.
porosity dependent speed c∗ = c∗ (n) shown on Fig. 18. 4. Continuity of the singularity exponent at the subsonic
Crack propagation is admissible in the subregion interface
c2 < c < c∗ . At the interface between the subsonic regime and
The energy release rate is continuous across the the region (ii), the singularity exponent γ vanishes.
boundaries of the regions, Fig. 19. This property holds It is equal to 1/2 at the interface between the subso-
true also along the boundary (i)–(ii), even if the varia- nic regime and the part of the region (iii) below the
tion becomes stronger as the porosity increases. curve c = c∗ (n), and to −1/2 above this curve. There-
In reference to Fig. 18, let us consider a vertical line fore bypassing the longitudinal wave-speed below this
at constant porosity and increasing crack tip speed. curve takes place with a smooth variation of the Mode
Across region (iii), the energy release rate is positive II singularity exponent γ , while bypassing the shear
between c2 and c∗− , unbounded at c∗ , negative between wave-speed induces a discontinuity of the singularity
c∗+ and c3 , it vanishes at c3 , it is maximum between c3 exponent.
262 E. Radi, B. Loret
Fig. 16 Crack with a COHESIVE CRACK: SHEAR STRESS close to crack tip
process zone under Mode II region (i), n=0.8, c=1600 m/s
loading conditions. Closeup
s /τ in
of the shear stress σ12 0 0.0004
the neighborhood of the
crack tip. The cohesion 0.0002
varies linearly along the
process zone from 0 to a 0
value τ0 at the crack tip. -0.0015 -0.001 -0.0005 0 0.0005 0.001
While the cohesion along
the process zone extends cohesive zone
continuously from the crack 1
line to the body, the region (ii), n=0.7, c=1000m/s 0.9
direction and range of 0.8
influence of the cohesion 0.0004 0.7
depend strongly on the 0.6
0.0002 0.5
region, and on the value of
the singularity exponent 0.4
0 0.3
-0.0015 -0.001 -0.0005 0 0.0005 0.001 0.2
0.1
cohesive zone 0
0.0004
0.0002
0
-0.0015 -0.001 -0.0005 0 0.0005 0.001
cohesive zone
crack tip speed
-1 0 0.2
-1.1 -0.2
0.1
-1.2
-0.4
-1.3 0
-10 -5 0 5 -10 -5 0 5 -10 -5 0 5
position wrt crack x1/L position wrt crack x1/L position wrt crack x1/L
Mode II intersonic crack propagation in poroelastic media 263
3 3 3
Mode II Mode II Mode II
scaled energy release rate n=0.40
1 1 1
c*
0 0 0
iii) i) ii) i) ii) i)
c2 c3 c1 c3 c2 c1 c3 c2 c1
-1 -1 -1
1000 2000 3000 4000 1200 1600 2000 2400 2800 400 800 1200 1600 2000
crack tip speed c (m/s) crack tip speed c (m/s) crack tip speed c (m/s)
Fig. 19 Scaled energy release rates G # = G × µ/(Lτ02 ) as func- stress along the process zone. Along the curve c = c∗ in region
tion of crack tip speed for Mode II loading conditions for three (iii), the singularity exponent γ vanishes and the energy release
porosities n = 0.05, 0.40, 0.80, and for a constant cohesive rate becomes unbounded
contribute to a more realistic description of the phe- can be shown, via Schwarz reflection principle, to
nomena taking place in the process zone. The resul- satisfy the two conditions on the real axis:
ting field equations would result to be non-linear, and ⎧
⎨Re [H (x1 )] − Im [H (x1 )] = 21 f (x1 ), for x1 < 0,
a different solution strategy should be devised;
(A1)
– the crack faces immediately behind the crack tip ⎩
Re [H (x1 )] = 0, for x1 > 0.
have been assumed to be permeable. This assump-
tion needs to be qualified; The solution to this Riemann–Hilbert problem, which
– the analysis has focused on defining the energetic is analytic on the upper half-plane Im [z] > 0 and
admissibility of crack propagation. Crack propaga- bounded at infinity, is defined up to the purely ima-
tion has to take place over sufficiently long periods ginary constant A, as required by (A1)2 , e.g., Muskhe-
of time, and long lengths, to be observable in the lishvili (1962),
laboratory or in the field. In other words, the possi- 0
A cos γ π 1 |t|γ f (t)
bility of a spontaneous bifurcation of the trajectory H (z) = γ + dt. (A2)
from a rectilinear path, and the assumption of steady z 2iπ zγ −L t − z
propagation need to be appreciated.
half-plane, across the positive real axis, of the function Appendix C: remarks on the singular integrals
Fi (z), i = 1, 2, by setting
Fi (z) = F i (z) ≡ Fi (z), for Im [z] < 0. (B2) Some remarks on the integral Iτ (z, γ ), defined by
(5.14) are provided here. This integral has to be evalua-
Let us define ted over the upper half-plane Im z ≥ 0 so as to define
Fi± (x1 ) = lim Fi (z), for i = 1, 2, (B3) the primary functions F1
, F2
, and G
.
x2 →0± Numerical evaluation of the fields:
and then The integral is singular for z = x1 located along the
Fi+ (z) = lim Fi (z) = lim Fi (z) process zone. In order to remove the apparent singula-
x2 →0+ x2 →0− rity at the origin, a change of variable is introduced:
−
= Fi (x1 ), for i = 1, 2. (B4)
t = −|z| (−T )1/γ ⇔ T = −(−t/|z|)γ . (C1)
Condition (B1)1 then implies, with help of (B4),
− Then the integral Iτ (z, γ ) can be expressed in the fol-
F1+ (x1 ) − F1+ (x1 ) = F1+ (x1 ) − F1 (x1 ) = lowing formats:
F2+ (x1 ) − F2+ (x1 ) =
−1 0 U + cos θ − i sin θ
− Iτ (z, γ ) =
F2+ (x1 ) − F2 (x1 ), for x1 < 0. (B5) γ −(L/|z|)γ U 2 + 2U cos θ + 1
Using the symmetry properties (B2), it follows from τ0 (−|z| U ) dT if z = |z| eiθ complex,
(B5) that
(F1 −F2 )+ (x1 ) = (F1 −F2 )− (x1 ), for x1 < 0. (B6) −1 0 τ0 (−|z| U ) dT
=
Since F1 and F2 are analytic on the complex plane γ −(L/|z|)γ U + z/|z|
except along the negative real axis, condition (B6) ifz=x1 < −L or > 0,
implies that F1 − F2 is an entire function, vanishing
as z → ∞. The result then follows from Liouville’s −1 0 ϕ(T )
= − τ0 (−|z| U ) dT
theorem. γ −(L/|z|)γ T + 1
+ i π τ0 (x1 ) if − L < z = x1 < 0, (C2)
Lemma 2 Let F1 (z) and F2 (z) be two functions with
the same regularity as in Lemma 1, but satisfying the
where the symbol − denotes the Cauchy principal value,
following conditions along the real axis: U = (−T )1/γ , and
⎧
⎨ Re [F1 (x1 )] = Re [F2 (x1 )] for x1 < 0, ⎧
(B7) ⎨ T + 1 if T = −1,
⎩ ϕ(T ) = U − 1 (C3)
Re [F1 (x1 )] = Re [F2 (x1 )] = 0 for x1 > 0. ⎩
−γ if T = −1.
then F1 (z) = F2 (z) for every value of the complex
variable z. Note that, in (C2)3 , the evaluation of the singular inte-
gral uses the section (Iτ )+ (z, γ ) defined for points z of
Proof Let Hj (z)=i Fj (z), j =1, 2. Since Im Hj (z) =
the upper plane, namely the points of the x1 -axis have
Re Fj (z), the result follows by Lemma 1.
angle θ = 0 for x1 > 0 and θ = π for x1 < 0.
Lemma 3 Let F1 (z) and F2 (z) be two functions with For z = x1 ∈] − L, 0[, the integrand is cast in a for-
the same regularity as in Lemma 1, but satisfying the mat that highlights the singular point and eases its treat-
following conditions along the real axis: ment in a routine that handles the principal value. For
⎧ regions (i) and (iii), the zone where the principal value
⎨ Re [F1 (x1 )] = Im [F2 (x1 )] for x1 < 0,
(B8) needs to be evaluated in view of (4.6) takes the form of
⎩ strip, behind the Mach line z2 = 0, of thickness equal
Re [F1 (x1 )] = Im [F2 (x1 )] = 0 for x1 > 0.
to the process length, Fig. 20. A similar zone exists in
then F1 (z) = −i F2 (z) for every value of the complex
regions (i) and (ii) behind the Mach line z3 = 0.
variable z.
Continuity across the rays z2 = 0 and z3 = 0:
Proof Let H2 (z) = −i F2 (z). Since Re Hj (z) = For z = x1 = 0+ , the integral can be evaluated using
Im F2 (z), the result follows by Lemma 2. e.g., Gradshteyn and Ryzhik (1980, p. 292),
266 E. Radi, B. Loret
Mach line
A similar conclusion applies across the ray z3 = 0
x 2y -L PV for f 2" (z 2 ) x 2y 0 f 2" (z 2 0) 0
in region (ii), and across the ray z2 = 0 in region (iii).
z2 L z2 L L z2 0 z2 0 z2 0
References
y
Burridge R, Conn G, Freund LB (1979) The stability of a plane Palmer AC, Rice JR (1973) The growth of slip surfaces in the
strain shear crack with finite cohesive force running at inter- progressive failure of over consolidated clay. Proc R Soc
sonic speeds. J Geophys Res 84:2210–2222 Lond A 332:527–548
Curran DA, Shockey DA, Winkler S (1970) Crack propagation Radi E, Bigoni D, Loret B (2002) Steady crack-growth in
at supersonic velocities - II. Theoretical model. Int J Frac elastic-plastic fluid-saturated porous media. Int J Plasticity
Mech 6(3):271–278 18(3):345–358
Dunham EM, Archuleta RJ (2004) Evidence for a supershear Radi E, Loret B (2007) Mode I intersonic crack propagation in
transient during the 2002 Denali fault earthquake. Bull Seis- poroelastic media. Mech Mater (in press)
mol Soc Am 94(6B):S256–2268 Rice JR (1985) Shear localization, faulting, and frictional slip:
Freund LB (1979) The mechanics of dynamic shear crack pro- Discusser’s report. In: Bazant ZP (ed) Mechanics of geo-
pagation. J Geophys Res 84(B5):2199–2209 materials. Wiley, New York pp 211–216
Freund LB (1990) Dynamic fracture mechanics. Cambridge Rice JR, Cleary MP (1976) Some basic stress-diffusion solutions
University Press, Cambridge, UK for fluid-saturated elastic porous media with compressible
Gao H, Huang Y, Gumbsch P, Rosakis AJ (1999) On radiation- constituents. Rev Geophys Space Phys 14(2):227–241
free transonic motion of cracks and dislocations. J Mech Rice JR, Simons DA (1976) The stabilization of spreading shear
Phys Solids 47(9):1941–1961 faults by coupled deformation-diffusion effects in fluid-
Geubelle PH, Kubair D (2001) Intersonic crack propagation in infiltrated porous materials. J Geophys Res 81(29):5322–
homogeneous media under shear dominated loading: nume- 5334
rical analysis. J Mech Phys Solids 49(3):571–587 Rosakis AJ, Samudrala O, Coker D (1999) Cracks faster than the
Georgiadis HG (1986) On the stress singularity in steady-state shear wave speed. Science 284(5418):1337–1340
transonic shear crack propagation. Int J Frac 30:175–180 Rosakis AJ, Samudrala O, Coker D (2000) Intersonic shear crack
Gradshteyn IS, Ryzhik IM (1980) Table of integrals, series and growth along weak planes. Mater Res Innova 3(4):236–243
products. Academic Press, San Diego, USA Rudnicki J (2001) Coupled deformation-diffusion effects in the
Huang Y, Gao H (2001) Intersonic crack propagation. Part mechanics of faulting and failure of geomaterials. Appl
I: the fundamental solution. J Appl Mech Trans ASME Mech Rev 54(6):483–502
68(2):169–175 Samudrala O, Huang Y, Rosakis AJ (2002) Subsonic and inter-
Knackstedt MA, Arns CH, Val Pinczewski W (2005) Velocity- sonic mode II crack propagation with a rate-dependent
porosity relationships. Geophys Prospect 53:349–372 cohesive zone. J Mech Phys Solids 50:1231–1268
Loret B, Harireche O (1991) Acceleration waves, flutter insta- Spudich P, Cranswick E (1984) Direct observation of rupture
bilities and stationary discontinuities in inelastic porous propagation during the 1979 Imperial Valley Earthquake
media. J Mech Phys Solids 39(5):569–606 using a short baseline accelerometer array. Bull Seismol
Loret B, Radi E (2001) On dynamic crack growth in poroelastic Soc Am 74(6):2083–2114
fluid-saturated media. J Mech Phys Solids 49(5):995–1020 Washabaugh PD, Knauss WG (1994) A reconciliation of dyna-
Loret B, Rizzi E (1998) On the effects of inertial coupling on the mic crack velocity and Rayleigh wave speed in isotropic
wave-speeds of elastic-plastic fluid-saturated porous media. brittle solids. Int J Frac 65(2):97–114
In: de Borst R, van der Giessen E (eds) Material instabilities Winkler S, Shockey DA, Curran DA (1970) Crack propagation
in solids. J Wiley and Sons, Chichester pp 41–53 at supersonic velocities – I. Theoretical model. Int J Frac
Muskhelishvili NI (1962) Some basic problems of the Mech 6(2):151–158
mathematical theory of elasticity. Noordhoff, Leyden,
The Netherlands
Obrezanova O, Willis JR (2007) Stability of an intersonic shear
crack to a perturbation of its edge. J Mech Phys Solids, in
press, doi: 10.1016/j.jmps.2007.04.009
Material forces for crack analysis of functionally graded materials
in adaptively refined FE-meshes
Rolf Mahnken
Abstract This work describes the computation of During the past two decades much research of FGMs
fracture parameters in functionally graded materials has been focused on manufacturing, material design
(FGMs) with stationary cracks. To this end the con- and property estimation as well as thermal and struc-
tinuum concept of material forces is employed, such tural analysis, Surush and Mortensen (1998). Appli-
that the corresponding balance equation can be discret- cations of FGMs are composites, ceramics, alloys and
ized with a standard Galerkin finite element procedure. coatings. These materials consists of two or more
A domain-type formulation is used for evaluation of a phases with spatially varying volume fractions thus
vectorial J -integral, where in the practical implemen- rendering non-uniform microstructures. Compared to
tation the material nodal forces of the finite element dis- conventional materials, FGMs have the possibility of
cretization are summed up in a finite region of the crack- tailoring its gradation in order to maximize its perfor-
tip. In this way the numerical calculation is completely mance. They have a better quality of thermal barrier
independent from the alignment of the finite element and anti-fatigue thus resulting into superior thermal and
mesh or any selected integration contour, which is most structural performance in high temperature environ-
attractive for adaptively refined finite element meshes. ments. Contrary to non-homogeneous materials with
For illustrative purpose the accuracy of the method is abrupt discontinuity, like laminated composite struc-
discussed for two examples based on comparison with tures, FGMs are not subjected to stress singularities at
available theoretical and numerical solutions. sharp bimaterial interfaces.
Due to the spatially distributed microstructure, from
Keywords Material forces · Finite elements · the viewpoint of continuum mechanics, FGMs possess
Stress intensity factor · J -integral the distinguishing feature of non-homogeneity. Conse-
quently mechanical material properties such as Young’s
modulus, yield strength, fracture toughness fatigue,
1 Introduction creep resistance are varying within the structure, see
e.g. Erdogan (1995). The same holds for thermal, mag-
Functionally graded materials (FGMs) are advanced netic and piezoelectric properties. Therefore parallel to
materials that possess continuously graded properties. developments in processing, research of FGMs must
be pursued on theoretical modeling, numerical imple-
R. Mahnken (B) mentation and experimental validation. In this way an
Department of Engineering Mechanics (LTM),
University of Paderborn, Warburger Str. 100,
experimental technique for evaluating elastic properties
33098 Paderborn, Germany of FGMs is investigated in Marur and Tippur (1998). A
e-mail: rolf.mahnken@ltm.uni-paderborn.de comprehensive review on fracture and failure of FGMs
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 269
doi: 10.1007/978-1-4020-6929-1_23, © Springer Science+Business Media B.V. 2008
270 R. Mahnken
including related references is presented in Paulino To give a conceptual motivation we recall, that spatial
et al. (2003). Micromechanically issues on effective (or physical) forces (in the sense of Newton) are gen-
properties of functionally graded composites are treated erated by variations relative to the ambient space at
in Zuiker and Dvorak (1994). One of the few references fixed material position. In 1951 Eshelby (1951, 1965)
considering inelastic behavior is Giannakopoulos et al. introduced the concept of a force acting on an elastic
(1995). singularity. He defined this force as the negative (mate-
Analytical investigations for the crack analysis of rial) gradient of the strain energy, and for this reason
FGMs have been performed in several publications. called it material (driving, configurational) force. Fol-
In this way Delale and Erdogan (1983) investigated lowing Steinmann (2001b) material forces (in the sense
the crack problem in an infinite plane, where the elas- of Eshelby) are generated by variations relative to the
tic properties varied exponentially in the direction of ambient material at fixed spatial position. The dual-
the crack. They showed that the asymptotic crack-tip ity between spatial and material forces carries over to
stress field possesses the same square root singular- the concept of stresses, in the sense that the Cauchy
ity as in homogeneous materials. Furthermore Eischen stress of the spatial problem has its counterpart in the
(1987a) used the traditional eigenfunction expansion Eshelby stress of the material problem, see e.g. Shield
technique of Williams (1957) to show, that the leading (1967), Chadwick (1975) and Steinmann (2001b). An
term in 2D crack-tip elastic fields remain square-root extensive exposition of related concepts in continuum
singular, which allows to express the stress fields in mechanics is given in Maugin (1993), Gurtin (2000)
terms of stress intensity factors (SIFs). In 1994 Konda and Kienzler and Herrmann (2000).
and Erdogan (1994) studied the behavior of an infi- Apart from valuable theoretical benefits, e.g. in defect
nite cracked plane with exponential property gradients mechanics, the continuum formulation in material space
in both in-plane directions. Further analytical results renders several computational advantages. As a conse-
for SIFs for some typical fracturing modes in FGMs quence finite element techniques well established for the
are presented in Erdogan (1995) and Erdogan and Wu spatial motion problem, such as the Galerkin weak for-
(1997). mulation and computation of the discrete node forces,
An important quantity for crack analysis is the carry over to the material motion problem with only
J -integral. For non-homogeneous materials the classi- minor modifications. In this way Braun (1997) intro-
cal formulation of Eshelby (1951), Cherepanov (1967, duced material forces in the context of the finite element
1968), Rice (1968) in general becomes path dependent. method in 1997, followed by various publications such
For certain classes of FGMs a path independent exten- as Nguyen et al. (2005), Müller et al. (2001), Müller and
sion of the J -integral is given in Honein and Herrmann Maugin (2002), Müller et al. (2004), Steinmann et al.
(1997). Here also a relation of the J -integral to SIFs is (2001), Steinmann (2001a,b), Rajagopal and Sivakumar
provided, thus obtaining the same mathematical struc- (2007), Denzer (2006), Denzer et al. (2003), amongst
ture as for homogeneous materials. Furthermore, for others.
general non-homogeneous materials a path indepen- Following Steinmann et al. (2001) and Steinmann
dent J ∗ -integral is established in Eischen (1987a,b) and (2001a,b), the discrete material nodal force acting on
Kim and Paulino (2002). a node representing the crack-tip can be interpreted as
Numerical methods for crack analysis of FGMs are a vectorial J -integral. However, the numerical results
presented in Gu et al. (1999), assigning different homo- indicate the necessity of a strong densification near the
geneous elastic properties to each finite element, Anlas crack-tip for the finite element mesh in order to get
et al. (2000), who numerically integrated an area region satisfying results. As explained by Denzer (2006), the
and Marur and Tippur (2000), considering a crack nor- accuracy of the numerical results depends strongly on
mal to the elastic gradient. A numerical procedure con- the accuracy of the Eshelby stress in the vicinity of the
sidering the J -integral in inhomogeneous materials is crack-tip, which due to the singularity of the Cauchy
also presented in Haddi and Weichert (1995). Further stress, in general, also has a singularity. This in turn
references are given e.g. in Kim and Paulino (2002). can render erroneous numerical results for the frac-
This work is concerned with the computation of ture parameters of interest. Therefore in this work we
fracture parameters in functionally graded materials by exploit a domain-type formulation for evaluation of a
exploiting the continuum concept of material forces. vectorial J -integral, where in the practical implemen-
Material forces for functionally graded materials 271
2.1 Equilibrium of physical forces where with respect to Cartesian coordinates xi and basis
vectors ei , i = 1, 2, 3, the nabla operator is defined as
To set the stage the boundary value problem (BVP) ∇(·) = ∂xi (·) ei , see e.g. Spiegel (1959).
for a functionally graded elastic material is introduced It is noteworthy, that the region with boundary
within a geometrically linear theory. For this purpose can be chosen arbitrarily within ∂B and B for eval-
in Fig. 1 we consider the configuration B ⊂ IEn dim uation of the Eqs. 2 and 4 and is not confined to the
occupied by a body B with position x ∈ IEn dim , and crack-tip region as indicated in Fig. 1. In this way it
where IEn dim denotes the Euclidean space with dimen- could be replaced by an arbitrary subdomain ∗ with
sion n dim = 2 or n dim = 3. Displacements are given by boundary ∗ . The only restriction is, that it must be free
the vector field u: B → IEn dim , and distributed body from singular stresses, so that the divergence theorem
forces per unit mass are given by the vector field b: (3) can be applied. Consequently Eq. 4 holds for any
B → IEn dim assumed to be independent from time t. point of the body with position x ∈ B except at a point
We concentrate on an arbitrary subdomain ⊂ B of singularity.
with boundary . This subdomain is loaded along A further fundamental ingredient of a geometrically
by surface tractions σ · n in terms of the symmetric linear theory is the strain–displacement relation
Cauchy stress tensor σ projected onto by the surface 1. ε = 21 h + hT ∀x ∈ B
(5)
normal n. Furthermore, within the region volume 2. h = u ⊗ ∇,
forces b are present. Then, upon defining the surface
where h is the displacement gradient. Let us assume
and body forces, respectively,
the functional relationship
fsur = σ · n dA and fvol = b dV (1) ψ = ψ[x, ε[x]] (6)
272 R. Mahnken
for the strain energy function of an FGM, then the stress 2.2 Equilibrium of material forces
tensor is obtained as
∂ψ This section is concerned with the derivation of a bal-
σ = . (7) ance law for material forces. In this context the so called
∂ε
Eshelby tensor is introduced, which will be used for
For an FGM with elastic isotropic behavior the strain
numerical evaluation of the J -integral in the ensuing
energy function is given as
section. The following derivation for a local balance
1. ψ[x, ε[x]] = 21 ε : C[x] : ε[x] law is adopted from Denzer (2006), where a tensorial
(8)
2. C[x] = 2G[x] Idev + K [x]1 ⊗ 1. notation is used. Note, that conceptually it is similar
to the derivation in Kim and Paulino (2002), where an
Here we define Idev = I − 1/31 ⊗ 1 with second and index notation is used.
fourth oder unit tensors I and 1, respectively. Further- As a starting point, we multiply the balance Eq. 4
more, in Eq. 8.2 we introduce the shear modulus G[x] with the negative displacement gradient −h of Eq. 5.2
and the bulk modulus K [x] of the FGM, related to and obtain by application of the chain rule the following
Young’s modulus E[x] and Poisson’s ratio ν[x] as identity
E[x]
1. G[x] = 2(1+ν[x]) − hT · (σ · ∇) − hT · b = (hT ⊗ ∇) : σ
(9)
E[x] −(hT · σ ) · ∇ − hT · b = 0. (12)
2. K [x] = 3(1−2ν[x]) .
Next, the spatial gradient of the strain energy function
Let us briefly comment on mathematical formula-
ψ in Eq. 6 is calculated by the chain rule as
tions representing the gradation of an FGM. This can
be achieved e.g. by a compositional distribution func- ∂ψ ∂ψ ∂ε ∂ψ
∇ψ = = · +
tion C[x]. An example for a two dimensional gradation ∂x ∂ε ∂x ∂x expl
is given in Hirano et al. (1990) as ∂ψ
= σ : (h ⊗ ∇) + , (13)
x − x1 n ∂x
C[x] = (C2 − C1 ) ,
expl
(10)
x2 − x1 where the constitutive relation (7) and the kinematic
where C1 and C2 denote composition values at the relations (5) have been used. The explicit derivative of
opposite ends of the FGM layer corresponding to the ψ is defined as
distances x1 and x2 , respectively. This function can be ∂ψ ∂ψ
= . (14)
multiplied e.g. with a reference value E 0 for Young’s ∂x expl ∂x ε=const
modulus, such that E[x] = E 0 C[x] reflects the grada-
tion of the elastic behavior. Upon using the identity (Denzer 2006)
An alternative approach is considered in the field of
σ : (h ⊗ ∇) = hT ⊗ ∇ : σ , (15)
topology optimization. It states, that at each point of the
domain the material property is related to the pseudo- Eq. 13 renders the relations
density ρ p describing the amount at each point of the
∂ψ
domain (Bendsoe and Sigmund 2003). In this way the σ : (h ⊗ ∇) = ∇ψ −
elasticity within the domain is expressed as ∂x expl
E H = ρ p E0 , (11) ⇒ hT ⊗ ∇ : σ
∂ψ
where E 0 is Young’s modulus of the basic material = ψ1 · ∇ − . (16)
(Silva and Paulino 2005). ∂x expl
In order to complete the boundary value problem, Inserting this result into Eq. 12 and rearranging we
boundary conditions must be specified: Upon subdivid- obtain
ing the boundary ∂B with outward normal n into disjoint
∂ψ
parts ∂u B ∪ ∂σ B = ∂B with ∂u B ∩ ∂σ B = ∅, Dirich- ψ1 · ∇ − (hT · σ ) · ∇ − − hT · b = 0, (17)
∂x
let boundary conditions u = ū on ∂Bu and Neumann expl
boundary conditions σ · n = t̄ on ∂Bσ are prescribed. which is written as a balance equation in local format
Material forces for functionally graded materials 273
2. For non-homogeneous materials the integral under On the segments c+ , c− the function q varies con-
the limit in Eq. 23 is not divergence free due to the tinuously from 1 to 0, which allows application of the
balance Eq. 18.1, and therefore corresponding line divergence theorem to the following line integral
integrals in a finite vicinity of the crack-tip become
path dependent. As a consequence the numerical · n dA = (q ) · n dA − (q ) · n dA
c+ ,c−
evaluation of the line integrals (23) and (24) can
become erroneous. = (q ) · ∇ dV
3. We assume, that derivatives of the elastic modulo
are bounded at the crack-tip, such that the volume − q ψ n − q hT · σ · n dA
integral for the material body force vector B in Eq. c+ ,c−
19 vanishes for vanishing integration line ε→0 .
Consequently, = · (∇q) dV + q ( · ∇) dV
J = lim · m dA − [[ψ]]q n+ dA
→0 c+
= lim · m dA + B dV + q hT · t̄ dA, (27)
→0 c+ ,c−
∗
=J , (25) and where the Neumann boundary condition σ · n = t̄
which is the so-called J∗ integral vector of Eischen on c+ , c− has been considered. The notation [[ψ]] =
(1987a). Based on the equilibrium Eq. 22 a deri- ψ + − ψ − reflects the discontinuity (or jump) in the
vation of Eq. 25 is also given in Kim and Paulino strain energy function across the crack opening. Upon
(2002). In Eischen (1987a) path independence of exploiting Eq. 27 and the balance Eq. 18.1 the integral
the integral is shown for general non-homogeneous vector in Eq. 23 is
materials.
J = lim · m dA = − lim · n dA
→ →0
2.4 Domain integrals = − lim · (∇q) dV
→0
The numerical evaluation of fracture parameters on the − q B dV − [[ψ]]q n+ dA
basis of the line integrals (24) and (25) has two dis- c+
advantages. First it is liable to numerically erroneous
results, also for the case of path independence for the + q h · t̄ dA .
T
(28)
c+ ,c−
J -integral. Second, it demands non-standard Finite
As noted in Kim and Paulino (2002), the third term on
Element data structures. In order to circumvent these
the right hand side including the jump [[ψ]] must be
difficulties we reformulate the line integrals as domain
accounted for, so that relatively large regions can be
integrals, similarly to the domain integral method of
used for evaluation of the J -integral by a domain inte-
Li et al. (1985). The resulting procedure can easily
gral approach. Note, that the term vanishes for evalua-
be implemented into existing finite element programs.
tion of J in Eq. 24 with e|| parallel to a straight crack.
Furthermore, it enables the computation of fracture
parameters using relatively large regions, thus improv-
ing the numerical accuracy. 3 Virtual work and discretization
We consider an arbitrary region around the crack-
tip, for the two-dimensional case shown in Fig. 1. 3.1 Spatial and material motion problem
According to the domain integral method of Li et al.
(1985) a smooth function q ∈ H1 (B) (H1 (B) being a In the following the quasi-static local balance
function-space, where functions and its first derivatives equations for the spatial motion equilibrium and the
are square-integrable) is introduced with the properties material motion equilibrium will be formulated as a
1 on weak (or variational) form, following closely the pro-
q(x) = (26)
0 on 0 . cedure in Steinmann et al. (2001), Steinmann (2001a,b)
Material forces for functionally graded materials 275
⊃
kin discretization is performed, thus rendering the finite
element physical nodal forces and the finite element n
xh Β e η
material nodal forces.
ξ
We multiply the local balance Eq. 4 with a test Je
crack Ω0
function (spatial virtual displacement) v under the nec- Γ0
Β
e
n en
1. v ·σ ·n dA = (v ⊗ ∇) : σ dV
uh ⊗ ∇x = un ⊗ ∇x
ne ,
∂B B Be
n=1
− v · b dV ∀ v
n en
B vh ⊗ ∇x = vne ⊗ ∇x
ne , (32)
Be
2. V · ·n dA = (V ⊗ ∇) : dV n=1
∂B B
n en
Vh ⊗ ∇x = Vne ⊗ ∇x
ne .
− V · B dV ∀ V. (29) Be
B n=1
Next, as indicated in Fig. 2, the domain B h ≈ B is dis- Then, the element wise contributions of the right hand
Nel
cretized into Nel elements, such that B h = ∪e=1 Be . sides in Eqs. 29.1 and 29.2 are
Let n en be the number of all nodes within Be and
n en
n en
n = 1, . . . , n en be the local numbering on each ele- 1. vn ·
e
σ · ∇x
e dV −
n
vn ·
e
b
ne dV
ment. Then the geometry, the nodal displacements, the n=1 Be n=1 Be
n en
n,e
polated by shape functions
ne as = vne · fint
n=1
n en
n en
n
n en
en
xh =
ne xn , uh =
ne un , 2. Vne · · ∇x
ne dV− Vn · B
ne dV
Be Be
n=1 n=1 n=1 Be n=1 Be
n en
n en
n en
vh =
ne vne , Vh =
ne Vne , (30) = n,e
Vne · Fint , (33)
Be Be
n=1 n=1 n=1
n en n,e
2. Fint = · ∇x
ne − B
ne dV (34)
Je = x h ⊗ ∇ξ = xn ⊗ ∇ξ
ne . (31) Be
Be
n=1
are nodal force vectors fint n,e n,e
, Fint ∈ IRn dim related to
Here ξ ∈ IRn dim represents the vector of natural coor- the nth node at element e. We let Nen be the number of
dinates within the reference element as indicated in nodes within B h . Then, upon assembling all element
Fig. 2. (The notation Je should not be confused with wise and node wise contributions of Eq. 34 to a node
the vectorial J -integral introduced in Eq. 23.) By use N = 1, . . . , Nen , internal parts for physical and mate-
of ∇x {•} = ∇ξ {•} · Je −1 the corresponding material rial nodal forces are obtained as
276 R. Mahnken
Nel
n,e
Remark 3.1
N
1. fint = fint , 1. A numerical procedure for evaluation of the third
e=1 n∈J Ne term on the right hand side in Eq. 37 including
Nel
the jump [[ψ]] is described in Kim and Paulino
n,e
N
2. Fint = fint , (2002).
e=1 n∈J Ne 2. For simplicity, we restrict ourselves to problems
J Ne = {n : Node n corresponds to N } . (35) with a symmetry of ψ with respect to the crack
front, such that [[ψ]] = 0. Furthermore, traction
Concerning dimension of the above nodal vectors we free crack surfaces are considered, such that t̄ = 0.
N , F N ∈ IRn dim .
observe fint int We also note, that the first two terms on the right
n,e
side of Eq. 37 represent the nodal force vector Fint
in Eq. 34.2. In the practical implementation this
3.2 Material force method for J -integral evaluation
expression is calculated by numerical integration
using Gaussian quadrature points, see e.g. Hughes
As outlined in Steinmann (2001b), for infinitely small
(1987). This avoids the use of singular stresses at
regions the vectorial J -integral in Eq. 25 collapses to
the crack-tip, such that we can refrain from the lim-
the material nodal force at the crack-tip. However, the N in Eq. 35,
symbol in Eq. 37 and, by use of Fint
numerical results indicate the necessity for a strong
approximate the J -integral simply by
densification of the finite element mesh near the crack-
tip in order to get satisfying results. As explained by
Nel
_L =4W
2
σ
(e) (f) (g)
Ω0 r/W = 0.02
(h)
Table 1 Single edge specimen in tension: comparison of normalized stress intensity factors
E 2 /E 1 Erdogan (1995) Kim and Paulino (2002) Mat.-Force (reg.) Mat.-Force (adapt.)
Table 2 Single edge specimen in tension: convergence of the normalized stress intensity factors for the first ten circular regions for a
regularly refined mesh in Fig. 4b
E 2 /E 1 Nel Radii r/W of circles
0.0 0.002 0.004 0.006 0.008 0.010 0.012 0.014 0.016 0.018 0.02
0.1 1216 3.371 3.489 3.496 3.499 3.493 3.492 3.493 3.493 3.492 3.493 3.492
0.2 1216 3.176 3.287 3.294 3.297 3.291 3.290 3.291 3.291 3.290 3.291 3.290
1 1216 2.724 2.819 2.825 2.828 2.822 2.821 2.822 2.822 2.822 2.822 2.822
5 1216 2.283 2.362 2.367 2.369 2.365 2.364 2.365 2.364 2.364 2.364 2.364
10 1216 2.098 2.171 2.175 2.177 2.173 2.172 2.173 2.173 2.173 2.173 2.173
Table 3 Single edge specimen in tension: convergence of the normalized stress intensity factors for the first ten circular regions for an
adaptively refined mesh in Fig. 4f
E 2 /E 1 Nel Radii r/W of circles
0.0 0.002 0.004 0.006 0.008 0.010 0.012 0.014 0.016 0.018 0.02
0.1 735 3.330 3.495 3.494 3.494 3.491 3.492 3.492 3.493 3.494 3.492 3.492
0.2 569 3.138 3.293 3.291 3.293 3.290 3.290 3.290 3.291 3.291 3.290 3.289
1 321 2.775 2.830 2.821 2.820 2.820 2.821 2.820 2.822 2.821 2.820 2.820
5 266 2.329 2.375 2.359 2.364 2.363 2.364 2.363 2.364 2.364 2.363 2.363
10 272 2.140 2.183 2.168 2.172 2.171 2.172 2.172 2.173 2.172 2.172 2.172
x2 Material 2 E2
2H=10
x1
4.5 2h
a=5
Material 1 E1
L=25 L=25 1
54
2a/H=0.06
region 0 . (c) material
forces in deformed crack-tip Ω 0
2a/W = 0.06
Ω0
2a/W = 0.06
√
Table 4 Three point bending specimen: comparison of normalized stress intensity factors K I H /P, a = 5.0
E 2 /E 1 Kim and Paulino (2002) MCC Kim and Paulino (2002) J∗ Mat.-Force (reg.) Mat.-Force (adap.)
√
Table 5 Three point bending specimen: convergence of the normalized stress intensity factors (K I H /P, a = 5.0) for the first ten
squares for a regular refined mesh in Fig. 6a
E 2 /E 1 Nel Length 2a/H of squares
0 0.006 0.012 0.018 0.024 0.030 0.036 0.042 0.048 0.054 0.06
0.05 1168 30.25 31.32 31.34 31.33 31.32 31.30 31.30 31.27 31.26 31.25 31.24
0.1 1168 23.11 23.92 23.94 23.93 23.93 23.92 23.93 23.92 23.92 23.92 23.92
0.2 1168 17.70 18.32 18.33 18.32 18.32 18.32 18.32 18.32 18.32 18.32 18.32
0.5 1168 12.15 12.58 12.58 12.58 12.58 12.58 12.58 12.58 12.58 12.58 12.58
1 1168 9.148 9.468 9.471 9.465 9.467 9.466 9.466 9.466 9.466 9.466 9.466
2 1168 7.069 7.316 7.318 7.313 7.315 7.314 7.314 7.314 7.314 7.314 7.314
5 1168 5.313 5.499 5.499 5.496 5.497 5.497 5.496 5.497 5.497 5.498 5.498
10 1168 4.437 4.593 4.592 4.590 4.593 4.596 4.593 4.600 4.601 4.602 4.603
20 1168 3.827 3.961 3.961 3.958 3.971 3.982 3.980 3.997 4.002 4.004 4.008
√
Table 6 Three point bending specimen: convergence of the normalized stress intensity factors (K I H /P, a = 5.0) for the first ten
squares for an adaptively refined mesh in Fig. 6d
E 2 /E 1 Nel Length 2a/H of squares
0 0.006 0.012 0.018 0.024 0.030 0.036 0.042 0.048 0.054 0.06
0.05 610 29.98 31.30 31.21 31.39 31.31 31.31 31.31 31.30 31.30 31.30 31.32
0.1 489 22.90 23.91 23.84 23.98 23.92 23.92 23.92 23.91 23.91 23.91 23.93
0.2 327 18.03 18.30 18.12 18.21 18.37 18.36 18.26 18.27 18.26 18.27 18.28
0.5 182 11.94 12.37 12.51 12.43 12.58 12.64 12.52 12.51 12.49 12.49 12.49
1 190 8.993 9.325 9.427 9.367 9.473 9.514 9.437 9.429 9.414 9.414 9.416
2 164 6.948 7.217 7.290 7.248 7.317 7.346 7.302 7.297 7.286 7.286 7.285
5 312 5.212 5.423 5.475 5.445 5.486 5.507 5.480 5.487 5.480 5.476 5.477
10 341 4.359 4.536 4.582 4.558 4.590 4.599 4.585 4.586 4.582 4.579 4.575
20 470 3.755 3.909 3.948 3.928 3.956 3.963 3.957 3.958 3.954 3.951 3.950
Braun M (1997) Configurational forces induced by finite- Li FZ, Shih CF, Needleman A (1985) A comparison of meth-
element discretization. Proc Estonian Acad Sci Phys Math ods for calculating energy release rates. Eng Frac Mech 21:
46:24–31 405–421
Chadwick P (1975) Applications of an energy-momentum Makowski J, Stumpf H, Hackl K (2006) The fundamental role of
tensor in non-linear elastostatics. J Elast 5:249–258 nonlocal and local balance laws of material forces in finite
Cherepanov GP (1967) Crack propagation in continuous media. elastoplasticity and damage mechanics. Int J Solids Struct
J Appl Math Mech 31:503–512 43:3940–3959
Cherepanov GP (1968) Cracks in solids. Int J Solids Struct Marur PR, Tippur H (1998) Evaluation of mechanical proper-
4:811–831 ties of functionally graded materials. J Test Eval JTEVA
Delale F, Erdogan F (1983) Fracture mechanics of functionally 26(6):539–545
graded materials. J Appl Mech 50:609–614 Marur PR, Tippur H (2000) Numerical analysis of crack-tip
Denzer R (2006) Computational configurational forces. Disser- fields in functionally graded materials with a crack normal
tation, Technical University of Kaiserslautern, Report No.: to the elastic gradient. Int J Solids Struct 37:5353–5370
UKL/LTM T 06-04, Lehrstuhl für Technische Mechanik Maugin GA (1993) Material inhomogeneities in elasticity.
Denzer R, Barth FJ, Steinmann P (2003) Studies in elastic frac- Chapman & Hall, London
ture mechanics based on the material force method. Int J Müller R, Maugin GA (2002) On material forces and finite ele-
Num Meth Eng 58:1817–1835 ment discretizations. Comput Mech 29(1):52–60
Eischen JW (1987a) Fracture of non-homogeneous materials. Int Müller R, Kolling S, Gross D (2001) On configurational forces
J Fract 34:3–22 in the context of the finite-element method. Int J Num Meth
Eischen JW (1987b) An improved method for computing the J2 Eng 53:1557–1574
integral. Eng Fract Mech 26:691–700 Müller R, Gross D, Maugin GA (2004) Use of material forces
Erdogan F (1995) Fracture mechanics of functionally graded in adaptive finite element methods. Comput Mech 33:
materials. Composit Eng 5:753–770 421–434
Erdogan F, Wu BH (1997) The surface crack problem for a plate Nguyen TD, Govindjee S, Klein PA (2005) A material force
with functionally graded materials. ASME J Appl Mech method for inelastic fracture mechanics. J Mech Phys Solids
64:449–456 53:91–121
Eshelby JD (1951) The force on an elastic singularity. Philos Parameswaran V, Shukla A (2002) Asymptotic stress fields for
Trans Roy Soc, Math Phys Sci A 244:87–112 dynamic stationary cracks along the gradient in functionally
Eshelby JD (1965) The continuum theory of lattice defects. Prog gradient materials. ASME J Appl Mech 69:240–243
Solid States Phys 3:79–114 Paulino GH, Jin ZH, Dodds RH (2003) Failure of function-
Giannakopoulos AE, Surush S, Finot M, Olsson M (1995) El- ally graded materials. In: Karihaloo B, Knauss WG (eds)
astoplastic analysis of thermal cycling: layered materials Comprehensive structural integrity, vol 2. Elsevier Science,
with compositional gradients. Acta Metall Mater 43(4): pp 607–644
1335–1354 Rajagopal A, Sivakumar SM (2007) A combined r-h adaptive
Gu P, Dao M, Asaro RJ (1999) A simplified method for calculat- strategy based on material forces and error assessment for
ing the crack-tip field of functionally graded materials using plane problems and bimaterial interfaces. Comput Mech
the domain integral. ASME J Appl Mech 66(1):101–108 41:49–72
Gurtin ME (2000) Configurational forces as basic concepts of Rice JR (1968) A path independent integral and the approxi-
continuum physics. Springer-Verlag mate analysis of strain concentration by notches and cracks.
Haddi A, Weichert D (1995) On the computation of the J-inte- J Appl Mech 35:379–386
gral for three-dimensional geometries in inhomogeneous Shield RT (1967) Inverse deformation results in finite elasticity.
materials. Comput Mat Sci 5:143–150 ZAMP 18:490–500
Hirano T, Teraki J, Yamada T (1990) On the design of function- Silva ECN, Paulino GH (2005) Topology optimization design
ally graded materials. In: Yamanouchi M, Koizumi M, Hirai of functionally graded structures. Proc of 6th World Con-
T, Shiota I (eds) Proceedings of the 1st International Sym- gress of Structural and Multidisciplinary Optimization, Ria
posium on Functionally Gradient Materials, Sendai, Japan, de Janeiro, 30 May-03 June 2005, Brazil
1990 Spiegel MR (1959) Vector analysis and an introduction to ten-
Honein T, Herrmann G (1997) Conservation laws in non- sor analysis. Schaum’s outline of theory and problems.
homogenous plane elastostatics. J Mech Phys Solids McGraw-Hill, New York
45:789–805 Steinmann P (2001a) Application of material forces to hyper-
Hughes TJR (1987) The Finite Element Method: linear static and elastic fracture mechanics, Part I: continuum mechanical
dynamic analysis. Prentice-Hall, Englewood Cliffs, NJ setting. Int J Solids Struct 37:7371–7391
Kienzler R, Herrmann G (2000) Mechanics in material space Steinmann P (2001b) A view on the theory and computation
with application to defect and fracture mechanics. Springer- of hyperelastic defect mechanics. In: Proc. of European
Verlag Conference on Computational Mechanics, ECCM, Crakow,
Kim JH, Paulino GH (2002) Finite element evaluation of mixed Poland
mode stress intensity factors in functionally graded materi- Steinmann P, Ackermann D, Barth FJ (2001) Application of
als. Int J Num Meth Eng 53:1903–1935 material forces to hyperelastic fracture mechanics, Part II:
Konda N, Erdogan F (1994) The mixed mode crack problem computational setting. Int J Solids Struct 38:5509–5526
in a non-homogeneous elastic plane. Eng Fract Mech 47: Surush S, Mortensen A (1998) Fundamentals of functionally
533–545 graded materials. Institute of Materials, London
Material forces for functionally graded materials 283
Williams ML (1957) On the stress distribution at the base of a Zuiker J, Dvorak G (1994) The effective properties of function-
stationary crack. ASME J Appl Mech 24:109–114 ally graded composites- I. Extension of the Mori-Tanaka
Zienkiewicz OC, Taylor RL (2005) The finite element method, method to linearly varying fields. Composit Eng 4:19–35
6th edn. vol 1. Mc Graw-Hill, London
Zienkiewicz OC, Zhu JZ (1987) A simple error estimator and
adaptive procedure for practical engineering analysis. Int J
Num Meth Eng 24:337–357
A multiscale approach to damage configurational forces
C. Dascalu · G. Bilbie
Abstract A two-scale homogenization method is 2005). In parallel with these developments, the mate-
used to construct a damage model in the framework rial or configurational mechanics (e.g. Maugin 1993;
of configurational mechanics. The upscaling procedure Gurtin 2000; Kienzler and Herrmann 2000) has been
allows for the identification of damage configurational established as a new framework for the modeling of
forces as the result of the microscopic fracture analysis. defects. The present contribution aims at providing a
The obtained damage equation incorporates stiffness micromechanical analysis for the formulation of brittle
degradation, material softening, unilaterality, induced damage in the frame of material mechanics.
anisotropy. The balance of configurational forces nat- Most of the local damage models are based on phe-
urally captures a microscopic length, leading to size nomenological assumptions and this makes difficult
effects in the overall damage response. The new the introduction of micro-structural lengths parameters,
approach is illustrated in the case of brittle damage, for with clear physical meaning. In a recent work (Dascalu
a three point bending test. Extended finite elements are et al. 2008), we have proposed a multiscale approach
used for the numerical modeling of macro-crack ini- to damage based on a full homogenization procedure,
tiation and growth. The influence of the microscopic without phenomenological assumptions, starting from
size on the failure initiation stress is analyzed and it is a microscopic energy analysis. The present paper
shown that this dependence follows a Hall–Petch type extends the previous results to damage configurational
rule. mechanics, by identifying the damage configurational
forces as the result of the micro-mechanical analysis. It
also proposes a framework for the complete description
1 Introduction of the failure process, in which macro-fracture initia-
tion is numerically introduced by using extended finite
In the last decades an important effort has been made to elements (Moes 1999). This allows us to quantify the
formulate damage models based on micromechanical influence of the size of the microstructure on the frac-
analysis (e.g. Nemat-Nasser and Hori 1999; Andrieux ture initiation process.
et al. 1986; Prat and Bazant 1997; Caiazzo and Con- We extend the homogenization formalism in Das-
stanzo 2000; Pensée et al. 2002; Lene 2004; Basista calu et al. (2008) to the framework of material mechan-
and Gross 1989; Li et al. 2004; Raghavan and Ghosh ics and deduce expressions of damage material forces
from the micro-mechanical description of the crack
C. Dascalu (B) · G. Bilbie
Laboratoire 3S-R, UJF, INPG, CNRS UMR 5221, BP 53,
propagation. In Agiasofitou and Dascalu (2007), a dif-
38041 Grenoble cedex 9, France ferent approach has been adopted to construct macro-
e-mail: cristian.dascalu@hmg.inpg.fr scopic material forces, by starting from the microscopic
Defect and Material Mechanics. C. Dascalu, G.A. Maugin & C. Stolz (eds.), 285
doi: 10.1007/978-1-4020-6929-1_24, © Springer Science+Business Media B.V. 2008
286 C. Dascalu, G. Bilbie
Bε
in which a microscopic energy balance is connected
to the macroscopic balance of material momentum in x1
while for cracks in contact we obtain at the order m = The function u(1) can be looked for in the form
−1, 0, 1 in ε:
u(1) (x, y, t) = ξ pq (y)ex pq (u(0) )(x, t) (22)
[σi(m)
j Nj] = 0, N j σi(m) < 0, T j σi(m) =0 (14)
j j where ξ (y) are the characteristic functions represent-
It can be shown that the function u(0) = u(0) (x, t) is ing elementary deformation modes of the unit cell
independent of the microscopic variable, representing (Bakhvalov and Panasenko 1989).
by this way the macroscopic displacement field. Relations (15)–(16), for every p and q and exkl (u(0) )
= δkp δlq ∈ R+ and with periodicity conditions on the
For given u(0) (x, t), for open traction-free cracks, cell boundary, are equivalent to the cell problem
we deduce the following boundary-value problem for ∂
the function u(1) : ai jkl e ykl (ξ pq ) = 0, in Ys , (23)
∂yj
∂
ai jkl e ykl (u(1) ) = 0, in Ys (15) ai jkl e ykl (ξ pq )N j = −ai j pq N j , on CY ± (24)
∂yj
ai jkl e ykl (u(1) )N j = −ai jkl exkl (u(0) )N j , on CY ± and for exkl (u(0) ) = −δkp δlq ∈ R−
(16) ∂
ai jkl e ykl (ξ pq ) = 0, in Ys (25)
∂yj
and with periodicity boundary conditions on the exter-
[ai jkl e ykl (ξ pq )N j ] = [ai j pq N j ], on CY (26)
nal boundary of the cell.
For closed cracks, the corresponding cell problem Ni ai jkl (e ykl (ξ pq ) − δkp δlq )N j < 0 on CY (27)
reads Ti ai jkl (e ykl (ξ pq ) − δkp δlq )N j = 0 on CY (28)
∂
ai jkl e ykl (u(1) ) = 0, in Ys (17) By introducing the mean value operator
∂yj
1
[ai jkl e ykl (u(1) )N j ] = −[ai jkl exkl (u(0) )N j ], on CY · = ·dy, (29)
|Y | Ys
(18)
(1) (0)
where |Y | is the measure of Y , we can prove that
Ni ai jkl (e ykl (u ) + exkl (u ))N j < 0, on CY
i(0) (0) (0)
j ≡ σi j = Ci jkl (d)exkl (u ) (30)
(19)
Ti ai jkl (e ykl (u(1) ) + exkl (u(0) ))N j = 0, on CY where
material manifold. In the presence of such inhomoge- introduces dependance of the global response on the
neities, the balance of material momentum is giving microstructural size.
rise to material forces. The objective of this section Consider now a propagation criterion: propagation
is to deduce the expression of damage material forces occurs when the microscopic driving force reaches a
from by micromechanical analysis, through homoge- threshold Fc , specific to the material:
nization. In a recent paper (Agiasofitou and Dascalu
Fε = Fc (41)
2007), we started with the microscopic balance of mate-
rial momentum and performed homogenization to With these, the macroscopic balance of material
deduce the macroscopic law of damage. Here, we start momentum can be expressed as
with the macroscopic balance of material momentum
1 dCi jkl (d) (0) (0) Fc ∂d
which will be linked with the microscopic energy anal- exkl (u )exi j (u ) + = 0.
2 dd ε ∂ xm
ysis of fracture, as performed in Dascalu et al. (2008), in
(42)
order to get the expression of the macroscopic damage
forces. This law express the inhomogeneity character of
From the equilibrium Eq. 33 we can deduce damage. For a spatially homogeneous state of damage
(0) the last term on the left vanishes, while when inhomo-
∂ Bm j
+ Fm = 0 (34) geneities in damage appear this term is not zero and the
∂x j new state o damage is found by equating the parenthesis
where with zero.
(0) (0) (0) We remark that we deduced (39) under the assump-
Bm j = W (0) δm j − i j u i,m (35) tion ḋ = 0, that is under the hypothesis of tempo-
is the Eshelby stress tensor rarily in-homogeneous damage. Actually the Eq. 39
1 could be multiplied by ḋ to get an equation similar to
W (0) = Ci jkl (d)exkl (u(0) )exi j (u(0) ) (36) (42), but coming from the energy balance, as (39) was
2
deduced from the balance of energy. In this way, one can
where
see that the trivial solutions of these equations, corre-
(0)
i j = Ci jkl (d)exkl (u(0) ) (37) sponding to damage homogeneity in time and, respec-
tively, in space, show the specific difference between
and
the energy and material momentum approaches. For a
1 dCi jkl (d) ∂d
Fm = − exkl (u(0) )exi j (u(0) ) (38) state of damage which is not homogeneous in space
2 dd ∂ xm and time, the two approaches lead to the same equation
is the damage material force. of damage.
In Dascalu et al. (2008), we proved that for an extend- The previous damage equation is coupled with the
ing micro-crack in a periodicity cell, i.e. for ḋ = 0 equilibrium Eq. 33. In what follows we will solve
where we denoted by ḋ the derivative with respect to numerically this system.
time, we have
1 dCi jkl (d) Fε
exkl (u(0) )exi j (u(0) ) + =0 (39) 5 Damage problem: Numerical implementation
2 dd ε
The scalar quantity Fε is defined as the projection of
We give in this section the details of the numerical
(7) on the propagation direction. The combination of
resolution of the two-sale damage problem deduced
the previous two relations yields
previously.
1 dCi jkl (d) ∂d
Fm = − exkl (u(0) )exi j (u(0) ) Summarizing the equations of the previous sections,
2 dd ∂ xm the equations of the macroscopic elasto-damage prob-
Fε ∂d lem are:
= (40)
ε ∂ xm
– homogenized equilibrium
That is, the macroscopic damage material force is
∂
expressed with the microscopic driving force on Ci jkl (d)exkl (u(0) ) = 0 (43)
micro-cracks and a material length parameter which ∂x j
290 C. Dascalu, G. Bilbie
6 Numerical results
Fig. 5 Effective
coefficients for crack
orientation (c). For the
orientation (d), we have the
same values, except C1112
which changes its sign
−C1112
292 C. Dascalu, G. Bilbie
(a)
(b)
(c)
(d)
dependence has been detected for the damage initiation Caiazzo AA, Constanzo F (2000) On the constitutive relations of
load values. This result is in accordance with our pre- materials with evolving microstructure due to microcrack-
ing. Int J Solids Struct 37:3375–3398
vious analyses in Dascalu et al. (2008). For a definite Dascalu C, Bilbie G, Agiasofitou E (2008) Damage and size
answer to the general question of mesh dependency effects in solids: a homogenization approach. Int J Solids
of the numerical solution, more detailed studies are Struct 45:409–430
necessary. Gurtin ME (2000) Configurational forces as basic concepts of
continuum physics. Springer-Verlag, New York
Kienzler R, Herrmann G (2000) Mechanics in material space
with applications to defect and fracture mechanics.
7 Conclusions Springer-Verlag, Berlin Heidelberg
Leguillon D, Sanchez-Palencia E (1982) On the behavior of a
The asymptotic homogenization technique was cracked elastic body with (or without) friction. J Mech
Theor Appl 1:195–209
employed to construct a brittle damage model in the Lene F (2004) Damage constitutive relations for composite
framework of configurational mechanics. The upscal- materials. Eng Fract Mech 25:713–728
ing procedure allowed for the identification of damage Li S, Linder C, Foulk W (2007) On configurational compatibil-
configurational forces as the consequence of the micro- ity and multiscale energy momentum tensors. J Mech Phys
Solids 55:980–1000
scopic fracture analysis. We showed that the balance of Li S, Wang G, Morgan E (2004) Effective elastic moduli of two
configurational forces naturally captures a microscopic dimensionl solids with distributed cohesive microcracks. Eu
length, leading to size effects in the overall damage J Mech A 23:925–933
response. Maugin GA (1993) Material inhomogeneities in elasticity.
Chapman and Hall, London
Finite element solutions for a three-point bending Moes N (1999) A finite element method for crack growth with-
test has been obtained and the capacity of the model to out remeshing. Int J Num Meth Solids 45:131–150
predict macroscopic fracture initiation and growth has Nemat-Nasser S, Hori M (1999) Micromechanics: overall prop-
been illustrated. Extended finite elements were used erties of heterogeneous materials. Elsevier, Amsterdam-
Lausann-New York
for the numerical modeling of macro-crack initiation Peerlings RHJ, Fleck NA (2004) Computational evaluation of
and evolution. The influence of the microscopic size strain gradient elasticity constants. Int J Multiscale Comput
on the failure initiation load has been proved to follow Eng 2:599–619
a Hall–Petch type rule. Pensée V, Kondo D, Dormieux L (2002) Micromechanical anal-
ysis of anisotropic damage in brittle materials. J Eng Mech
128:889–897
Petch NJ (1968). In: Liebowitz H (ed) Fracture, vol 1, chap 5.
References Academic Press, New York, p 351
Prat PC, Bazant ZP (1997) Tangential stiffness of slastic materi-
Agiasofitou E, Dascalu C (2007) Material forces in microfrac- als with systems of growing or closing cracks. J Mech Phys
tured bodies. Arch Appl Mech 77:75–84 Solids 45:611–636
Andrieux S, Bamberger Y, Marigo JJ (1986) Un modèle de Raghavan P, Ghosh S (2005) A continuum damage mechanics
matériau microfissuré pour les bétons et les roches. J Mec model for unidirectional composites undergoing interfacial
Theor Appl 5:471–513 debonding. Mech Mater 37:955–979
Bakhvalov N, Panasenko G (1989) Homogenisation: averaging Sanchez-Palencia, E (1980) Non-homogeneous media and vibra-
processes in periodic media. Kluwer Academic Publishers tion theory. Lecture notes in physics, vol 127. Springer, Ber-
Group, Dordrecht lin
Basista M, Gross D (1989) The sliding crack model of brittle Smyshlyaev VP, Cherednichenko KD (2000) On rigorous der-
deformation: an internal variable approach. Int J Solids ivation of strain gradient effects in the overall behav-
Struct 35:487–509 iour of periodic heterogeneous media. J Mech Phys Solids
Bazant ZP, Planas J (1997) Fracture and size effect in concrete 48:1325–1357
and other quasibrittle materials. CRC Press, Boca Raton,
FL
Benssousan A, Lions JL, Papanicolaou G (1978) Asymptotic
analysis for periodic structures. North-Holland, Amsterdam