Chapter 3. Forces, Momentum & Stress. 3.1 Newtonian mechanics: a very brief résumé

Chapter 3 Forces, Momentum & Stress 3.1 Newtonian mechanics: a very brief résumé In classical Newtonian particle mechanics, particles lumps of matter) only experience acceleration when acted on by external influences which are known as forces 1. The resistance to acceleration is described by a constant of proportionality known as the mass 2, a fundamental property of matter, that is positive or zero) and invariant under motion. Newton s second law, as originally formulated for a particle of mass m can be interpreted as a definition of force 3 F, ma = F m R = F, where a is the Lagrangian) acceleration of the particle; R is the current position; F is the resultant force applied to the particle; and the overdots represent material) differentiation with respect to time. Consider now a collection of particles, each one having mass m i) and position R i) in the fixed global coordinate system. The index i) is a Lagrangian marker to keep track of individual particles. We assume that there is mutual symmetric interaction between the particles that acts along the line between the particles 4 so that the particle i experiences a force F ij) due to the j-th particle and that F ij = F ji). Applying Newton s second law to each particle gives m i) Ri) = F ext i) + j i F ij), 3.1) where F ext i) is the resultant external force on the i-th particle. If we consider a subset of N particles, which we can define so that the index i I = {1,...N}, then we can define the total mass M, linear momentum P, and angular momentum or moment of momentum) about the origin H 0, of the ensemble by N N N M = m i), P = m i) Ṙ i) and H 0 = R i) m i) Ṙ i). 3.2) Using equation 3.1) and the fact that the mass of each particle in invariant, we can write N [ D P = ) N N m i) Ṙ i) = m i) Ri) = F ext i) t + ] F ij), j 1 1 This is essentially Newton s first law. 2 Strictly speaking this is the inertial mass. 3 Although Truesdell dislikes this interpretation, arguing that we typically have a priori knowledge about forces. 4 This is Newton s third law. 53

but because F ij) = F ji) the internal forces between particles in the subset cancel and we have N D P = F ext i) + j / I F ij) F, 3.3) Equation 3.3) states states that the material rate of change of the total linear momentum is given by the sum of the external forces and interactions from other particles not in the subset, which can be defined as the resultant external force, F. Equation 3.3) expresses the balance of linear momentum and also its conservation in the absence of external forces, the material rate of change of linear momentum is zero. Note that although momentum is not conserved when external forces are present, the equation 3.3) is nonetheless often termed the consveration of momentum equation, or even just the momentum equation. In taking this sum, we are assuming that momenta can be added even though they correspond to different spatial locations. The standard resolution of this concern is to recognise that N D P = m i N m i R i N m i = M R, where R = N R i))/m is the centre of mass and so equation 3.3) can be interpreted as an expression for the motion of the centre of mass. Similarly, we can write D H 0 = N Ṙ i) m i) Ṙ i) + N R i) m i) Ri) = [ N R i) F ext i) + ] F ij), j i because Ṙi) Ṙi) = 0. The mutual interaction between two particles i and j in the set is given by the two terms R i) F ij) + R j) F ji) = R i) F ij) R j) F ij) = R i) R j) ) F ij) = 0, because the line of action of the interaction is parallel to the line between the particles. Hence, the expression for the rate of change of the total angular momentum about the origin is D N H 0 = R i) F ext i) + F ij) L 0, 3.4) j / I which is equal to the resultant external torque 5 or moment) about the origin on the system, L 0. The equation 3.4) expresses the balance of angular momentum, and its conservation when the external torque is zero. 5 In classical mechanics the torque, or turning moment, is given by the cross product of the applied force and the distance from the point of application of the force to a pivot point, L 0 = R F. Here, the pivot point is taken to be the origin and F is the applied force. The torque vector points along the axis of rotation and its sense is such that in a right-handed coordinate system, positive magnitudes correspond to anticlockwise rotations.

3.2 The continuum hypothesis In the continuum hypothesis we assume that we can replace the discrete particles by a continuous distribution of matter, so that kinematic quantities such as acceleration, velocity and displacement are the averages of the individual particulate molecular) motion. We introduce distributed masses and forces that are functions of position and the continuum hypothesis is essentially the assumption that such quantities exist. 3.2.1 Mass For a fixed point in space R, we define regions Ω ǫ such that R Ω ǫ and the volumes V Ωǫ of the regions tend to zero as ǫ 0. If the total mass of the continuum contained within the region Ω ǫ is M Ωǫ, then the mass density field ρr, t) is defined by the limit ρr, t) = lim ǫ 0 M Ωǫ /V Ωǫ. 3.5) Under the continuum hypothesis we assume that this limit exists and is well defined and does not depend on the precise choice of volumes Ω ǫ. Assuming the existence of a mass density then the total mass within a deformed region is given by Mt) = ρr, t) dv t, which generalises the expression for the total mass in equation 3.2). 3.2.2 Momentum Having asserted the existence of the mass density then the generalisation of the expression 3.2) for the total linear momentum in the deformed region is simply Pt) = ρr, t)v R, t) dv t ; and the total angular momentum of the deformed region about a point Z is given by H Z t) = R Z) [ρr, t)v R, t)] dv t. Note that these expressions may be further generalised to include the effects of additional point masses if required. 3.2.3 Balance of momenta We can now generalise the equations 3.3) and 3.4) to apply to continua by writing DP = D ρv dv t = F, 3.6) where F is the resultant force on the body; and DH Z = D R Z) ρv dv t = L Z, 3.7) where L Z is the resultant torque about the point Z. Although we know the governing equations, we do not yet know how to express the forces and torques that act on continua, rather than particles.

3.2.4 Loads: Forces and Torques The forces and torques that act on continua can be divided into three types: i) External body forces and torques that act on every mass within the region in question: Body forces include gravity, electromagnetic forces and the fictional centripetal/centrifugal forces that arise due to rotation if the governing equations are written in a rotating frame. A body force is usually expressed as a force density per unit mass, F R, t), or a force density per unit volume, F = ρf and under the continuum hypothesis we assume that such a densities exist. Hence, the total resultant body force acting on a volume is F V = ρf dv t, which is the generalisation of the first sum on the right-hand side of the equation of balance of linear momentum for systems of particles 3.3). Physical evidence suggests that body torques are not possible, but they can be included in the theory as a torque or moment) per unit mass or per unit volume. The argument used to rule out the existence of body torques is that if the couple is pictured as a pair of equal and opposite forces connected by a rod, then if the rod length tends to zero and the forces remain bounded the torque must be zero 6. Nonetheless, the body force can exert a net torque on the body about a point Z, which is given by L V Z = R Z) ρf dv t, the generalisation of the first sum on the right-hand side of the equation of balance of angular momentum for systems of particles 3.4). ii) External surface forces and torques that act on the exposed surfaces of the body: These forces typically arise due to contact between bodies and usually consist of a tangential friction and normal pressure. If R is a point on the external surface of the body, we define a set of surrounding surface regions S ǫ on the surface such that S ǫ, R S ǫ and the area of S ǫ is given by A ǫ. We assume that the areas A ǫ tend to zero as ǫ 0. If the resultant force acting on the surface S ǫ is given by F ǫ then the traction is defined by the limit TR, t) = lim ǫ 0 F ǫ /A ǫ, 3.8) and under the continuum hypothesis the limit exists. The traction field T is the force per unit area of the deformed surface. Hence the total resultant force on the volume due to surface loads is F S = T ds t, which can be identified with the second sum on the right-hand side of the equation of balance of linear momentum for systems of particles 3.3) if we further assume that it is only particles on the surface of the subset that can be affected by the particles outside the subset. In other words, all long-range influences must be treated as body forces. 6 There is always an exception of course, ferrofluids can be treated as suspensions of magnetic dipoles which do provide body torques, but the dipole does require unbounded forces.

The physical evidence for surface torques is not clear, but in certain theories of granular materials sands, powders, etc), it is assumed that different elements of the continuum can apply surface torques on each other due to irregularities of the particles. The body of work that includes surface torques is called Cosserat theory or the theory of micro-polar materials, but we shall not include them here. As for body forces, a surface traction can exert a net torque on the body about a point Z which is given by L S Z = R Z) T ds t, the generalisation of the second term on the right-hand side of equation 3.4) under the same assumptions about long-range interactions. iii) Internal forces that act between mass elements within the body: By Newton s third law mutual interaction between all internal masses is zero. Nonetheless, we may examine the state of the internal forces by considering an internal surface that divides a volume, see Figure 3.1. The force per unit area exerted on the material on one side of the surface by the material on the other can be represented by a stress vector or internal traction vector that is again defined by the limit of the resultant force acting on an area of of the internal surface as the area tends to zero. 3.3 The stress tensor 3.3.1 Cauchy s Postulate We can characterise a surface by a unit normal N taken in a certain sense. If the surface is external to the body then we can define the normal to be directed out of the volume. If it is internal, there is a certain ambiguity which can be resolved by treating the surface as one dividing the body into two regions, see Figure 3.1, and taking the outer unit normal in each region. Naturally we have to assume that the unit normal to the surface is well-defined. Under Cauchy s postulate we assume that the surface traction depends only on the unit normal, N, to the surface, i.e. TR) = TNR), R). In particular, we are not assuming any dependence of the traction on higher derivatives of the surface, such as curvature. In fact, the hypothesis can be proved directly from the momentum equation in the absence of any surface couples, a result known as Noll s theorem. If we assume that the resultant force due to surface loads on a region tends to zero as the volume of the region tends to zero, it can be shown by considering a small cylindrical volume a pillbox) that TNR), R) = T NR, R). Hence, for an internal surface, the surface tractions exerted by one half of the material on the other are equal and opposite, see Figure 3.1 3.3.2 The existence of the stress tensor We now consider the force balance on a small tetrahedral region in the deformed body with volume dv, see Figure 3.2 The tetrahedron is chosen so that three edges are parallel to the coordinate

Γ Ω t T Ω + t N N + T + Figure 3.1: A region of a continuum is divided by a nominal internal surface Γ into the region Ω + t and Ω t. A outer unit normal can be defined for Ω + t and Ω t, say N + and N respectively. On Γ, N + = N and the associated stress vectors, T + and T are also equal and opposite. ξ 3 T 1) N T 2) G 3 ξ 2 T T 3) G 1 G 2 ξ 1 Figure 3.2: An infinitesimal tetrahedron in the deformed body aligned with the Eulerian coordinates. Three edges of the tetrahedron are directed along the covariant base vectors G i, which means that normals to corresponding three faces are given by G j. directions ξ i which means that three of its faces are aligned with planes of constant ξ i. The plane of constant ξ i is spanned by the two covariant base vectors G j where j i. Each tetrahedral face ξ i is constant can be represented by a vector normal to the face directed out of the volume with magnitude equal to the area of the face, S i) = N i) ds i) = Gi ds i) G ii, no summation)

where G i / G ii ) is the unit vector normal to the face of constant ξ i and the area of the face is ds i). We have used the contravariant upper index) vector to ensure orthogonality to the face. The final face of the tetrahedron is not aligned with any coordinate direction and has the vector representation, N ds. In the infinitesimal limit, the curves bounding each face become triangles and the edges of the final face can be written as differences between the covaraint base vectors G i. By taking cross-products of the bounding vectors to determine the vector representations of each face, it follows that the vector sum of the surfaces is zero NdS i G i ds i) G ii ) = 0. 3.9) We decompose the normal into components in the contravariant basis N = N i G i, and then taking the dot product of equation 3.9) with G j gives N j ds = ds j) G jj ) N j G jj ) ds = ds j) no summation). In general, three stress vectors T i) will act on the faces of constant ξ i and a stress vector T will act on the final face of the tetrahedron. If the tetrahedron of material obeys the balance of momentum 3.6) and, if the tetrahedron is sufficiently small, the total resultant force is given by ρf dv + TdS i T i) ds i) = ρ V dv, where F is the resultant body force acting on the volume and ρ V its acceleration. Assuming that the body force and the acceleration remains bounded then these terms are negligible because if a typical edge length is given by dǫ, then dv dǫ) 3, whereas ds dǫ) 2. As dǫ 0, dv is an order of magnitude smaller than ds and so the surface tractions are the only terms that remain TdS i T i) ds i) = 0, TdS i T i) N i G ii )ds = 0, T = i N i G ii ) T i), 3.10) From Cauchy s postulate the traction vector T must remain invariant 7 provided the normal N does not change. By construction N i are components of the normal vector that transform covariantly, so T i) G ii ) must transform contravariantly in order for T to remain invariant. Hence, we can write which means that T i) G ii ) = T ij G j, T = T ij N i G j T G k = T ij N i G j G k T k = T ij N i δ k j = T ik N i. 7 We are actually assuming here that the traction is a vector that is invariant under coordinate transformation.

The quantity T ij is called the Cauchy stress tensor and represents the j-th component in the deformed covariant Eulerian basis) of the traction vector acting on a surface with normal component in the i-th direction also in the deformed covariant Eulerian basis). The above argument demonstrates that if TN) then it is a linear function of N, which means that T = T N), where T is a tensor 8. The covariant base vectors are not necessarily of unit length and so the physical components of the stress tensor are obtained by expressing the stress vectors in terms of unit tangent vectors T i) = j σ ij) G j / G jj ) σ ij) = G jj )/G ii ) T ij no summation). The components σ ij) are not, in general, a representation of a tensor unless the coordinate system ξ i is orthonormal, in which case G ii = G jj = 1 no summation) and so σ ij) = T ij. In the global Cartesian basis in the deformed configuration we can write σ IJ = σ IJ = X I ξ i T ij X J ξ j. If we only use Cartesian coordinates, then ξ i = x i and ξ j = X j, so as claimed. σ IJ = δ I i T ij δ J j = T IJ, 3.3.3 Alternative forms of the stress tensor We derived the components of the Cauchy stress tensor T ij by considering the force per unit area in the deformed configuration in which the traction vector and the normal to the surface are both represented in base vectors associated with the Eulerian coordinates, ξ i. Alternatively we can represent the components of this tensor in the base vectors associated with the Lagrangian coordinates, ξ i, in the deformed position. Exactly the same tetrahedral argument can be applied to show that the traction is given by T = T ij N i G j, where T ij is the required tensor after a suitable coordinate transformation T ij = ξi kl ξj T k ξ ξ ; 3.11) l this is the stress tensor used by Green & Zerna and is sometimes called the body stress. Although it is given a different name the body stress represents the same quantity as the Cauchy stress, but with components in a different coordinate system. 8 The word tensor is derived from the Latin verb tendere meaning to tighten or or to stretch. The demonstration that the stress vector acting on a surface is a linear mapping of the normal vector is due to Cauchy and establishes the existence of an object that we now call a tensor.

A completely different family of stress tensors is formed by considering the force per unit area in the undeformed configuration, which essentially introduces multiplication by the Jacobian of the deformation gradient tensor J. The total force on an infinitesimal area is given by and using Nanson s relation 2.40), we obtain F = T ds = T ij N i G j ds = T ij G j da i, F = T ij G j J ξk ξ i da k = JT ij ξk nk ξ i G j ds = P jk n k G j ds = t ds, where P jk = JT ij ξ k / ξ i, are the components of a tensor often known as the first Piola Kirchhoff 9, or nominal, stress tensor and t is the force per unit undeformed area. In the global Cartesian basis, the first Piola Kirchhoff stress tensor is often written as P = Jσ T F T. 3.12) The first Piola Kirchhoff stress tensor can be interpreted as the force in the current configuration per unit area of the deformed configuration. It is a two-point tensor because the normal is now the normal to the undeformed area which has now been decomposed into components in the undeformed Lagrangian base vectors, g i, but the traction itself is still decomposed into the deformed Eulerian base vectors, G j. If instead we decompose the traction into components in the deformed Lagrangian base vectors, G j, then we obtain the second Piola Kirchhoff stress tensor t = P jk n k ξ l ξ j G l = JT ij ξk ξ i ξl ξ j n kg l = s kl n k G l = JT kl n k G l. The components s kl are the components of the second Piola Kirchhoff stress tensor which is simply J multiplied by the body stress tensor, although the k component does refer to a different basis. In the global Cartesian basis, it is often written as S = JF 1 σf T. 3.13) The second Piola Kirchhoff stress tensor can be interpreted as the force per unit area of the undeformed configuration decomposed into the Lagrangian base vectors in the deformed configuration. Alternatively, in the Cartesian basis, it is often interpreted as a fictional force the force mapped back via the inverse deformation gradient tensor) per unit undeformed area. The interpretations are the same because mapping the components of a vector from the deformed to the undeformed configuration has the same effect as the same as mapping the undeformed base vectors into the deformed ones. Decomposing the force into the two different coordinate systems gives F = F i G i = F j G j, then F j = ξj ξ i F i, but converting to from curvilinear to Cartesian coordinates in the deformed and undeformed positions, we obtain F j = ξj F J = ξj i ξ x J ξ i F I = ξj F I F J = x J F I, X I X I X I which expresses the transformation from the components of the real force, F I, to those of the fictional force, F I in the global Cartesian basis. 9 Note that in some texts the first Piola Kirchhoff stress tensor is defined to be the transpose of the definition given here.

3.3.4 Equations of equilibrium If a body is in equilibrium then there is no change in momentum or angular momentum, in other words all acceleration terms, both linear and angular are zero. An alternative interpretation is that all forces and torques balance so that there is no net force or torque acting on the body. Linear momentum In the absence of any acceleration terms, the conservation of momentum 3.6) applied to a region in the deformed configuration yields the equation T ds t + ρf dv t = 0, 3.14) and using equation 3.10) to replace the traction vector, we have N i G ii )T i) ds t + ρf dv t = 0, i N i T i ds t + ρf dv t = 0, 3.15) G where T i = T i) GG ii ) is a vector that transforms contravariantly. The purpose of this rearrangement is so that we can use the divergence theorem 1.56) in the Eulerian curvilinear coordinates, A i 1 N i ds t = G ξ iai G) dv t to transform equation 3.15) into T i G 1 ξ i + ρfdv t = 0. 3.16) The equation must be valid for any volume, which means that the integrand must be zero 1 T i G ξ i + ρf = 0, 3.17) which is the governing differential equation of motion known as Cauchy s equation. If we express the stress vector, T i in terms of the Cauchy stress tensor, we have 1 ) G G ξ i T ij G j + ρf = 0, which can be written as T ij i + ρf j) G j = 0, where, as usual, the symbol refers to covariant differentiation in the Eulerian coordinates in the deformed configuration. This step could be interpreted as the definition of the divergence of a

second-rank tensor 10. The equation is often written as R T + ρf = 0, which looks nice and compact but requires knowledge of how to take the divergence of a tensor and, in particular, over which index the contraction is to be taken. In components in the basis G j, we have the governing equation T ij i + ρf j = 0, and in Cartesian coordinates the equation reduces to which starts to look a lot less intimidating. Angular momentum T IJ,I + ρf J = 0, In the absence of any angular acceleration, conservation of angular momentum applied to a region in the deformed configuration yields R Z) TdS t + ρr Z) F dv t = 0. The vector Z is a constant, so it can be taken outside the integrals, to give [ ] R TdS t + ρr F dv t Z TdS t + ρf dv t = 0, R T ds t + ρr F dv t = 0, from the equations of equilibrium 3.14). Introducing the vector T i, such that T = N i T i / G as above, the equation becomes R 1 T i N i ds t + ρr F dv t = 0, G and using the divergence theorem gives [ ] 1 G ξ i R T i) + ρr F dv t = 0. 3.18) Expanding the derivative of the cross product gives [ ] 1 G i T i 1 T i + R G G ξ i + ρf dv t = 0, 10 It is consistent with the covariant derivative of a tensor, as it should be. Using the definition of the Christoffel symbol 1.46) and the derivative of the square-root of the determinant 1.54), we have 1 ) GT ij G ξ i G j = T ij,i G j + T ij Γ k ijg k + Γ k kit ij G j = T ij,i + Γ j ikt ik + Γ k kit ij) G j = T ij i G j.

but the term in square brackets is equal to zero by Cauchy s equation 3.17) and so G i T i dv t = 0. 3.19) G 1 Once again the equation must be valid for any volume, which is only possible if the integrand is zero 1 G i T i = G i T ij G j ) = ǫ ijk T ij G k = 0. G Each component of the vector equation must be zero, ǫ ijk T ij = 0 1 2 [ ǫ ijk T ij + ǫ jik T ji] = 0, and using the antisymmetry property of the Levi-Civita symbol, we have 1 [ 2 ǫ ijk T ij T ji] = 0. It follows that the Cauchy stress tensor must be symmetric T ij = T ji. From their definitions, it follows that the body stress 3.11) and second Piola Kirchhoff 3.13) stress tensors are also symmetric, but the first Piola Kirchhoff stress tensor 3.12) is not, which is because it is a two-point tensor. 3.4 Axiom of Objectivity In general, the motion of an observer will affect the values of measured physical quantities. The underlying behaviour of material should not depend on the motion of the observer, so we wish to understand how the apparent material properties change due to observer motion. A quantity that is independent of the observer is said to be objective or material frame indifferent. An inertial frame of reference is a frame in which Newton s laws apply and so the results derived above apply only in inertial frames. Non-inertial frames introduce fictitious forces into the governing equations due to frame acceleration, e.g. Coriolis or centripetal/centrifugal forces, which is because the acceleration is not objective. We shall neglect relativistic effects which means that spatial distances and the time intervals between events measured by all observers must remain the same. We suppose that one observer measures spatial position and time by the pair R, t) and another by the pair R, t ). If the distances between points and time between events must remain the same for both observers then the most general possible transformation is R t ) = Qt)Rt) + Ct), t = t a, 3.20) where Qt) is an orthogonal matrix, Ct) is an arbitrary point in space or translation vector) and a is a constant time shift, which means that the material derivatives D/ and D/ coincide. For a vector to be objective then its length must not change under the observer transformation, i.e. A A = A A.

Now if we define a vector to be the difference between to points in space, At) = Rt) St), then A t) = R t) S t) = Qt)Rt) + Ct) Qt)St) Ct) = Qt)[Rt) St)] = QA. Hence, writing out the components in Cartesian coordinates A A = A I A I = Q IJA J Q IK A K = Q T KI Q IJA J A K = δ KJ A J A K = A J A J = A A, because Q is orthogonal. Thus the vector is objective if A = QA, which is equivalent to saying that the vector satisfies the appropriate tensor transformation laws in order to remain invariant under the observer transformation. Hence, for a tensor of order two, M, to remain invariant under observer transformation it must obey the transformation rule M = QMQ T. Taking the material derivative of equation 3.20), and suppressing the explicit dependence on time, gives the transformations for the velocity and acceleration and V = DR = D QR + C) = QR + QṘ + Ċ = QV + QR + Ċ, 3.21a) A = DV = QA + 2 QV + QR + C, 3.21b) where the overdot indicates the material derivative. Hence, as we already know, the velocity and acceleration are not objective and depend on the motion of the observer. We can give physical interpretation to the non-objective terms in equation 3.21a,b) by introducing the observer spin tensor, Ω = QQ T, associated with the observer transformation. Taking the material derivative of the identity QQ T = I gives QQ T + Q Q T = Ω + Ω T = 0, which shows that Ω is antisymmetric, so Ω = QQ T ) T = Q Q T. Inverting equation 3.20) we obtain R = Q T R C), which can be used in equation 3.21a) to write V QV = QQ T R C) + Ċ = Ċ + ΩR C). If the terms on the right-hand side were zero, the velocity would be objective. The first term reflects the relative velocity of the observers and the second is their relative rotation 11. Inverting 11 The product of an antisymmetric tensor and a vector can be written as the cross product of two vectors: Ω IK A K = e IMK ω M A K Ω IK = e IMK ω M e JKI Ω IK = e JKI e IMK ω M, e JKI Ω IK = [δ JM δ KK δ JK δ MK ] ω M = 3ω J ω J = 2ω J. ω J = 1 2 e JKIΩ IK. If the two observers are not translating relative to one another C = 0 then for a fixed vector R, we can write DR = ΩR = 1 2 ω R, which expresses a rigid-body rotation about the axis ω with angular velocity ω /2.

] the relationship for velocity, we obtain V = Q [V T Ċ ΩR C) inverse relationship for position in equation 3.21b) we obtain A QA = C + 2 QQ ] [V T Ċ ΩR C) + QQ T R C), = C + 2ΩV Ċ) + Ω Ω 2 )R C), and using this and the after using the fact that Ω = QQ T Q Q T = QQ T QQ T Q Q T = QQ T + Ω 2. The acceleration would be objective if the terms on the right-hand side were zero. The first term is the translational acceleration; the second is the Coriolis acceleration; the third combines the rotational and centripetal accelerations. The acceleration and velocity are only objective if the transformation is Galilean: C and Q are both constants, which means that the observers are not rotating relative to each other. 3.4.1 Objectivity of deformation measures The deformation gradient tensor transforms via or F IJ = X I x J = x J {Q IK t)x K t) + C I t)} = Q IK F KJ, F = Q F, so the deformation gradient tensor is not objective because it is a tensor of second rank, but transforms like an objective vector 12. The right) Cauchy-Green tensor is invariant under change of observer because c = F T F = QF) T QF = F T Q T QF = F T F = c; and hence so is the Green Lagrange strain tensor e = 1 2 c I ) = 1 c I) = e. 2 Thus, these tensors are not objective in the Eulerian sense described above because they are based on the Lagrangian, rather than the Eulerian, coordinates. Such quantities are sometimes called objective in a Lagrangian sense or observer invariant, rather than observer independent. The Eulerian Almansi) strain tensor is based on the Eulerian coordinates and does transform objectively E = 1 2 I C ) = 1 2 QQ T F T F 1) = 1 2 QQ T QF T F 1 Q T) = QEQ T. 3.4.2 Objectivity of rate of deformation measures The transformation of the Eulerian velocity gradient tensor, L, can be obtained by using the relationship 2.48) Ḟ = LF L = ḞF 1 ; and so L = Ḟ F 1 = D ) QF) F 1 Q T = QF + Q Ḟ F 1 Q T = QLQ T + QQ T = QLQ T + Ω. 12 The reason is because the deformation gradient tensor is two-point in character; the index associated with the Eulerian coordinate transforms objectively, but the index associated with the Lagrangian coordinate does not change under an Eulerian observer transformation.

It follows that the symmetric part of the velocity gradient tensor, the Eulerian rate of deformation tensor transforms objectively D = 1 2 L + L T) = 1 2 QLQ T + Ω + QL T Q T + Ω T) = Q 1 2 but that the antisymmetric part, the spin tensor, does not W = 1 2 L L T) = 1 2 L + L T ) Q T = QDQ T, 3.22a) [ Q L L T ) Q T + Ω Ω T] = QWQ T + Ω. 3.22b) Naturally, this is completely consistent with our analysis of velocity and acceleration in 3.4, which indicated that any measure which involves relative rotation is non-objective. Equation 3.22b) reveals that the difference in instantaneous rotation rate measured by two different observers will differ by the relative rotation rate Ω. The equation can be written in the form W = Q T W Ω) Q, which demonstrates that the spin tensor can be made objective by subtracting the non-inertial observer rotation rate Ω. 3.4.3 Objective Rates We have already seen that the material derivative of an objective quantity a vector for example) is not objective because the velocity of the observer affects the perceived motion. If we work in Lagrangian framework the problem does not arise because the material derivative is objective in the Lagrangian sense, or observer invariant. Unfortunately, there are many circumstances in which we must work in the Eulerian framework, e.g. taking laboratory measurements or even applying boundary conditions. Thus, we are lead to consider rates of change that are objective in Eulerian formulations. Vectors The idea of objective rates is to construct derivatives that follow the motion in a more general sense than simple material derivatives, which follow the movement only of material points and therefore have no information about stretch or rotation of the continuum. Consider a vector field, bξ i, t), that is convected with the motion of the continuum, so that it obeys the same transformation rules as line elements. In the undeformed configuration we have bt = 0) = b k 0)g k and then in the deformed configuration Bt) = B k G k, where B k t) = ξk ξ i t) bi t), which follows from interpreting the motion as a coordinate transformation from Lagrangian ξ i to Eulerian ξ j coordinates and applying the standard contravariant transformation rule. Hence, Bt) = ξk ξ i bi G k = ξk ξ R = bi k ξ ξ = i bi G i = B i G i, i bi R so the convected vector is simply given by the components B i = b i t) in the base vectors associated with the Lagrangian coordinates in the deformed configuration, or convected base vectors.

Taking the material derivative of B gives DB = Bk t G k + B kdg k = Bl t G l + B k v l k G l = [ B l t + Bk v l k ] G l, after using equation 2.43). The first term is a vector that will transform objectively because the material partial) derivative of B l the Lagrangian coefficients) is invariant under observer transformation and the base vectors transform appropriately G i = QG i. The second term involves the covariant derivative with respect to the basis G i or G i depending on the observer. Hence it will not remain invariant under observer transformation and it is this term that renders the material derivative non objective. If we decompose the material derivative directly into components in the convected basis, DB/ = DB k /) G k, then we have that DB k = Bk t + B l v k l. The material and partial derivatives do not coincide in the convected basis because the convected base vectors change with time. Thus, an objective derivative can be formed by taking the vector B = Bk t G k = B k G k, where B k = DBk B l v k l. The equation is a tensor equation so we can simply transform to the Eulerian coordinates ξ i to obtain B k = DBk B l V k l = Bk + V i B k t i B l V k l, which can be written in vector form as B = DB LB, and is called the upper convected Oldroyd) derivative and represents a derivative that is taken following the motion of material lines. Note that if B = 0, the vector is invariant under the motion 13. Note also that V = V, because when the observer is moving with the material lines of t the continuum the acceleration is due only to the acceleration of material particles, which is indeed independent of the observer. The upper convected derivative can also be obtained by considering the material derivative of a vector and writing the vector as the product of the deformation gradient tensor and the undeformed vector in order to isolate an objective derivative DB = D DF Fb) = b + FDb. The final term transforms objectively because the material derivative of a Lagrangian vector is invariant under observer transformation and the deformation gradient transforms as an objective vector F = QF. Thus, the objective upper convected derivative is obtained by B = DB DF b = DB DF F 1 B = DB LFF 1 B = DB LB. 13 This is not entirely obvious, but can be easily derived because B = 0 implies that b i / t = 0 and so the vector is invariant in the Lagrangian viewpoint and the transformation to Eulerian coordinates B k = ξ k / ξ i )b i ensures that B remains invariant.

This derivation does not reveal the provenance of the name upper convected, which arises because the derivative represents the material derivative of the upper contravariant) components of a vector when convected with the motion. We confirm that the derivative is, in fact objective, from B = DB L B = DQB QLQ T + Ω)QB, B = QB + QḂ + QLB ΩQB = ΩQB ΩQB + Q [ DB LB ] = QB, where the definition Ω = QQ T has been used. If we apply the same considerations to B = B k G k, then we would obtain the lower convected derivative B k = DB k + B l V l k = B k + V i B k t i + B l V l k, which can be written in vector form as B = DB + LT B. The lower convected derivative has not found to be as useful in describing the behaviour of continua as the upper convected derivative because it describes a derivative taken following material areas in an incompressible medium, or the product of material areas and a mass density in a compressible medium. The upper and lower convective derivatives take the complete deformation into account, but we know that non-objectivity only arises through relative rotation of the observers. We can formulate the co-rotational or Jaumann rate by using the polar decomposition F = RU to consider the material derivative of a vector that rotates, but does not stretch with the deformation DB = D DR Rb) = b + RDb. As above, the second term transforms objectively because the rotation of the observer simply changes the rotation matrix 14 R = QR. Thus, we can form an objective derivative by taking B = DB DR b. We can find the material derivative of the rotation tensor by considering using the polar decomposition DF = DR U + RDU = LF = LRU, so that ) DR L = U + RDU U 1 R T = ṘRT + R UU 1 R T, and hence D = 1 2 L + L T ) = 1 2 ṘR T + RṘT ) + 1 ) 2 R UU 1 + U 1 U R T, 14 We know that F = R U = QRU and the polar decomposition is unique. QR is the product of two orthogonal matrices so it is orthogonal and by uniqueness of the polar decomposition R = QR.

but because R is orthogonal DR T R)/ = DI/ = 0 and so ṘRT + RṘT = 0, which yields D = 1 ) 2 R UU 1 + U 1 U R T and W = 1 ) L L T = 2 ṘRT + 1 2 R UU 1 U U) 1 R T. If we neglect the deformation of the motion then U = 0, and W = ṘRT, which means that Ṙ = WR. Finally,we obtain the standard expression for the Jaumann rate B = DB WRb = DB WB. Note that a sum of objective derivatives will also be objective and the Jaumann rate is a linear combination of the upper- and lower- convected derivatives B = 1 2 B + B ). Tensors Objective rates for tensors of higher rank can be constructed from the same ideas. If we consider the material derivative of a convected second rank tensor DA = D [ ) A ij A ij G i G j = t G i G j + A ij DGi G j + G i DG ] j. = Aij t G i G j + A [ [ ] A ij v l i G l G j + G i v l ij j G l = + A lj v i l + A il v j l ]G i G j, t we can define the upper convected derivative A ij = DAij A lj V i l A il V j l, and the lower convected derivative is found to be or A = DA LA ALT ; A ij = DA ij + A lj V l i + A il V l j, or A = DA + LT A + AL. The Jaumann rate for a second-rank tensor is formed from the appropriate linear combination of the upper and lower convected derivatives. A = 1 2 A + A ) = DA + 1 L T L ) A + A 1 ) L L T = DA WA + AW. 2 2