Vous êtes sur la page 1sur 7

CARTESIAN TENSORS

Tensor analysis plays a key role in many branches of Physics: Mechanics, Fluid dynamics, Electromagnetism theory, General theory of relativity, and so on. It is a powerful mathematical tool that can express physical laws in invariant forms. Hence, learning of tensor manipulation skills must be a priority goal for science and engineering students of undergraduate courses. In fact, as one will see in higher physics, practically all mathematical descriptions of physical quantities are tensors of some kind. Thus, in order to understand how to use mathematics to describe physics, it is important to understand what tensors are and how to work with them. In turn, this requires an understanding of some basic concepts, such as the concepts of Cartesian and general tensors, symmetric and anti-symmetric tensors, Kronecker delta and permutation tensors, metric tensors, covariance and contra-variance of tensors, covariant derivative of covariant and contra-variant tensors, Riemannian manifold, Christoffel symbols and the corresponding index manipulation, Riemann-Christoffel curvature tensor, Bianchi identities, Einstein tensor, and so on. We shall try to explain these concepts in this chapter. In the tensor analysis, the demonstrations are based on the indicial notation. Therefore, these are very elegant and compact. Although they require to handle complex symbolic expressions, without perhaps any physical insight into the problem at hand to begin with. Our focus in this chapter will also be to impart this precious element by establishing connection with the Euler-Lagrange equations, Hamiltons principle, etc. discussed in Chapter 12. The physical quantities should be independent of the choice of the coordinate system. So should be the mathematical quantities which are required to model them. As already mentioned above, in the tensor analysis we seek the coordinate system independent quantities for application in physics, that is, we seek those quantities that have component transformation properties that render the quantities independent of the observers coordinate system. An act of this kind imparts a kind of objective existence to the physical quantities. This is the reason tensors are ultimately defined strictly in terms of the transformation properties. We do this exercise, right at the outset, in the section 14.1. Some conventions may be spelt out here. When a small Latin index, either superscript or subscript, such as T, T, and T occur unrepeated in a tensor expression, it is understood to take all the values 1, 2, ..., N, where N is the number of dimensions of the space. However, when a small Latin index is repeated in a term, summation with respect to that index is understood, the range of summation being 1, 2, ..., N. The position of a tensor in the hierarchy of these objects is designated by an index n called rank. The number of components that a tensor possesses is given by 3n for the three dimensional space. The table given below summarizes this hierarchy. There are, in fact, nine dyads (e1 e1, e1 e2, e1 e3, ,e3 e3) which form a basis for the rank two tensor in Cartesian coordinates. Similarly, triads ei ej ek form a basis for a rank three tensor. Here ei , i = 1,2,3, are the unit vectors. Rank Number of components

Scalar Vectors Dyads Triads . Poly-ads

0 1 2 3 .. N

1 3 9 27 . 3n = (Dimension of space)Rank

We shall clearly see in subsections 14.3.2 and 14.3.3 that metric tensor is of fundamental importance in the tensor analysis. In three dimensions, the vector dr = du e1 + dv e2 + dw e3 which is the vector sum of the coordinate (u,v,w) differentials, in general, has got nothing to do with the physical distance unless we are in a flat Euclidean space, i.e. du = dx., dv = dy, dw = dz, e1= i, e2= j, and e3= k. In fact, the quantity dr. dr = (dr)2 represents only the square of the magnitude of a coordinate distance dr in a curved space. If now the provision is made in terms of the association of scalars ,, and , respectively, with the coordinate differentials (du,dv,dw) which carry the necessary physical distance information, then the vector d = du e1 + dv e2 + dw e3 qualifies to correspond to physical length. To find the physical length ds of d we form the inner product ds2 = d. d = 2 (du )2 + 2(dv)2 + 2(dw)2. The square root of the righthand-side, in principle, is supposed to provide the necessary physical distance. Now the question is can it be related to the coordinate distance dr. The answer is in affirmative. We write ds2 = d. d = 2 (du ) (du )+ 2(dv) (dv ) + 2(dw) (dw ) and note that ds2 = ((2 e1 e1+ 2 e2 e2 + 2 e3 e3). dr). dr = (. dr). dr. The components of the dyad are, in fact, components of a rank two tensor called metric tensor. The components could be arranged in a 3 by 3 diagonal matrix as shown:

For a general curved space, with orthogonal curvilinear coordinates {x}, the physical distance between two neighboring points will be written as ds2 = g dx dx where g are the components of the metric above. In the penultimate and the ultimate subsections,viz. 14.3.2 and 14.3.3 respectively, we wish to address one of the basic problems of tensor calculus in Riemannian manifold (A differentiable manifold is Riemannian if there exists within it a covariant rank two symmetric metric tensor g > 0 at every point. For example, the 3-D Euclidean space.). This is to try and find an object which is function of the metric tensor and some of its derivatives. A solution to this problem is the Riemann Christoffel tensor R to be developed towards the end of the sub-section 14.3.3. If index gymnastics were a physical sport, this subsection would be a training session for the fit and athletic. In it, the covariant derivative is established, the Riemann curvature tensor is found, properties are analyzed, and the Einstein curvature tensor is found and shown to obey the contracted Bianchi identities, which has importance in Einsteins law of gravitation. From a neutral perspective, the important aspect of these mathematical manipulations is that, though they introduce no new physical principles, they

express certain fundamental relationships in a compact and elegant form which necessarily hold in our universe obeying the general principle of relativity at macroscopic level. Cartesian tensor of rank two and physical properties of crystalline solids The crystalline solid properties, such as the electrical and thermal conductivities, susceptibility, dielectric permittivity and magnetic permeability can be described by Cartesian tensors of rank two. For example, Ohms law for anisotropic media, with choice of an orthogonal coordinate system, may be written as j = E,where j stands for current density, E for electric field, and for electrical conductivity. The summation above is carried out over the repeated subscript. In contrast, for isotropic media j = E. Thus,the number of quantities necessary to define electrical conductivity in an anisotropic media is not one but nine. The specific values of the coefficients 11, 12, 13, 21, 22,..depend on the choice of the coordinate system fixed to the crystalline solid. In a transformation from the system {x} to the system {x}, the coefficients are transformed according to = T T as shown in Eq.(8), where T ,T are the orientation of the axes {x} with respect to the axes {x}. One may also write the inverse transformation = T T . If one requires to find the resistivity tensor from above, one may write E = j. In the general case, the individual components of the resistivity tensor are not the reciprocals of the corresponding components of the conductivity tensor, but are found from the relation = (-1)+ / where = det( ). The object is the minor obtained from the above determinant by crossing out the the th row and th column. The tensors of rank two are geometrically interpreted as surfaces of second order. The components of a symmetric rank two tensor (S = S ) are used as coefficients in the general equation of this second-order surface: S11x12+ S22x22+ S33x32+ S12x1 x2+ S13x1 x3+ S21x2 x1+ S23x2 x3+ S31x3 x1+ S32x3 x2 =1. The surface thus obtained is referred to as the representative quadratic of a symmetric rank two tensor. The second order surfaces for which S = S possess the socalled principal axes that is three mutually perpendicular directions such that if they are chosen as the coordinate axes, the general equation of the surface given above transforms to the simple form S11x12+ S22x22+ S33x32 = 1. Clearly, if (S11, S22,S33) > 0, the surface under consideration is an ellipsoid. If two coefficients are positive and one negative, the surface is a hyperboloid of one sheet. If two coefficients are negative and one positive, the surface is a hyperboloid of two sheets. Furthermore, the magnitude of the radius vector r of the representation quadratic in a chosen direction is related to a quantity characterizing the corresponding property in this direction, say the electrical conductivity, by the relation = 1/r2. How to determine the directions of the mutually orthogonal principal axes? To answer this we proceed as follows: At the points of intersection of the principal axes with the representation surface the normal to the surface is parallel to the radius vector. If P denotes a point on the representation surface given by the relation above, and {x} denotes the components of the vector OP(and forms a vector S x) , then this vector is parallel to the surface normal at the point P in accordance with the property of the radius vector. The radius vector and the normal are parallel to each other if their corresponding components are proportional to each other, i.e. S x = x where is a constant. This system has a non-zero solutions only when det(S - ) = 0. We, in fact, get a cubic called secular equation which when solved for yields the direction in which the radius vector of a quadratic surface is parallel to the normal of the surface. That is, the values of determine the directions of the mutually orthogonal principal axes.

To illustrate, we consider the following problem: Suppose the polarizability p of a crystal introduced in section 14.1 has the following components with respect to a coordinate system in appropriate units: p
=

Indicate the directions with respect to this coordinate system in which the polarization vector has the same direction as the applied electric field. In this case, the secular equation yields three real roots for , viz. 1= 4, 2 = 16, and 3= 25. We thus find that the values of the diagonal components of the polarizability tensor referring to the principal axes are equal to p1= 4, p2 = 16, and p3= 25 in appropriate units. Suppose now the components of the vectors along the principal axes , whose orientation relative to the given coordinate system is to be determined, are p(1), p(2) , and p(3). Also, suppose the components of the vector p(1) are to be denoted by p1(1), p2(1) , and p3(1). Similarly, the components of the vector p(2) are to be denoted by p1(2), p2(2) , and p3(2).,and so on. These are the so-called eigenvectors corresponding to the eigenvalues 1= 4, 2 = 16, and 3= 25. A full matrix V whose columns are these eigenvectors is V= 0 0 0.8660 0.5000 0.5000 -0.8660 1.0000 0 0

The table of direction cosines fixing the orientation of the principal axes {x} with respect to the initial axes {x}then takes the following form: AXES x1 x2 x3 x1 0 0 1 x2 0.8600 0.5000 0 x3 0.5000 -0.8600 0

We notice that the determination of the direction cosines of the mutually orthogonal principal axes is basically an eigenvalue-eigenvector problem. It must be noted that the eigenvalues of a real symmetric matrix are real and the corresponding eigenvectors are orthogonal. If the eigenvalues
are degenerate, the eigenvector subspace corresponding to the degenerate eigenvalues is orthogonal to the other eigenvectors; within this subspace, the eigenvectors can be chosen to be orthogonal by the GramSchmidt procedure discussed in Chapter 13.

Having explained these preliminaries, we consider a rigid body of arbitrary shape rotating about a fixed axis with angular velocity = n. Consider a small mass element dm at a point , with position vector r relative to an arbitrarily chosen origin. The velocity of the mass element dm = dV, where = (r) is the (mass) density, is v = r . The angular momentum dL of the elemental dm at r is dL = ((r) dV r v ) giving L = dV . This in turn yields L = ijk rj klm l rm dV = (iljm imjl) rj l rm dV .

Thus Li = Iijj , where Iij = (r2ij rirj)dV. Here Iij is symmetric tensor of rank two and independent of the n axis chosen. Now it is always possible to choose a coordinate system such that the inertia tensor is a diagonal matrix. This is a direct result of a mathematical theorem stating that any real symmetric matrix is diagonalizable. For the kinetic energy, T, we have dT = 1/2 (dV )( r)2 or, T = ijk j rk ilm l rm dV = (jlkm jmkl) j rk l rmdV. Thus we have T = (2r2 (r )2)dV = Iijij . z- axis

As an example, we consider a cube of side a of constant density with mass M = a3, The moment of inertia tensor is Iij = (r2ij rirj)dV. This gives I11 = I12 = ((x2 + y2 + z2) x2) = [(1/3)y3xz + (1/3)z3xy] 0a = (2/3)a5 = (2/3)Ma2 , ( xy) = (/4) [x2y2z] 0a = (1/4) a5 = (1/4)Ma2 .

Thus by symmetry we have Iij = Ma2 2/3 1/4 1/4 1/4 2/3 1/4 1/4 1/4 2/3

As already stated above it is always possible to choose a coordinate system such that the inertia tensor is a diagonal matrix. To do this we consider the secular equation. In this case, the secular equation yields three real roots for , viz. 1= 0.1667, 2 = 0.9167, and 3= 0.9167. We thus find that the values of the diagonal components of the moment of inertia tensor referring to the principal axes are equal to I1= 0.1667 Ma2, I2 = 0.9167 Ma2, and I3= 0.9167 Ma2. A full matrix V whose columns are the corresponding eigenvectors is V= -0.5774 -0.1543 0.8018 -0.5774 0.7715 -0.2673

-0.5774 -0.6172 -0.5345 The table of direction cosines fixing the orientation of the principal axes {x} with respect to the initial axes {x}then takes the following form: AXES x1 x2 x3 x1 -0.5774 -0.1543 0.8018 x2 -0.5774 0.7715 -0.2673 x3 -0.5774 -0.6172 -0.5345

The discussion and example above clearly show that one can diagonalize the moment of inertia tensor. Referring to the principal axes we have, in general, I = A 0 0 0 B 0 0 0 C

where A,B, C are the principal moments of inertia. A geometric interpretation is given by the Inertia Ellipsoid, which is defined by Iijij = 1 (ie absorb 2T) into ). If we rotate to the principal axes, transforms as vectors, i.e. i = lijj where lij are direction cosines, and we have A 12+ B 22 + C 32 = 1 in the normal form. This is an ellipsoid as (A,B,C) > 0 (We have seen this in the example above). A natural question is Are there directions for such that L is parallel to independent of the choice of the coordinate system? .The answer to the question posed above could be obtained once we find the eigenvalues (i), i = 1, 2, 3 and the corresponding eigenvectors (i) where L(i) = (i)(i) (no sum). Now L= , where is a 3 3 diagonal matrix means that ( Iij - ij ) j = 0. For non-trivial solution det( Iij - ij ) = 0. First writing this as det( Iij - ij ) = (1/6) ijk lmn( Iil - il ) ( Ijm - jm ) ( Ikn - kn ) = 0 and then expanding this yields M N + P 2 3 = 0 where M = (1/6) ijk lmn Iil Ijm Ikn , N = det(I), N = (1/6) ijk lmn(il Ijm Ikn+ Iil jm Ikn + Iil Ijm kn) = (1/6) (jmkn jnkm)IjmIkn 3 = (1/2)[(Tr(I))2 Tr(I2) ], and

P = (1/6) ijk lmn(il jm Ikn+ il Ijm kn + Iil jm kn) = (1/6) (jmkn jnkm) jmIkn 3 = Tr(I). As detA, and TrA are invariants, the coefficients M, N, P are invariants of the tensor I. Thus the eigenvalues and the eigenvectors do not depend on choice of the coordinate system. We now summarize the physics relating to rigid body rotation. The angular velocity = u is a vector quantity. The magnitude of represents the rate of rotation about the axis defined by the unit vector u. The velocity of a point on a body rotating with angular velocity is given by v = r which leads to the equation relating the derivative of a vector A in an inertial reference frame(XYZ) fixed on the laboratory to the derivative in reference frame(xyz) rotating with angular velocity fixed on the body(say, on the centre of gravity G) : (d/dt)(Ainertial) = (d/dt)(Abody) + A We have seen above that every rigid body has an associated inertia tensor that is symmetric and real-valued. We have shown below the components in terms of integrals assuming the rigid body to be a continuous one:

Since the local xyz reference frame is defined as attached to the rigid body (and moving with it), the six inertia terms remain constant as the rigid body moves in space. This is the reason why we define the xyz frame as being attached to the rigid body. Otherwise, the inertia terms will change with time (usually). This would add unnecessary complexity to the problem. Now, if xyz is aligned with the principal directions of inertia of the rigid body, we have

We have seen above that the principal axes of a body (about an origin,say, G) are mutually orthogonal axes through G with the property that if points along the axes, then L = . In other words, the principal axes are the eigenvectors of the inertia tensor.

Vous aimerez peut-être aussi