BIBLIOGRAPHY Adey, A. D., and Brebbia, C. A, “Basle Computational Methods for Engineers,” Wiley, New York, New York, 1983, lement Method for Practicing I, New York, 1977 JeeHall, Englewood Cliffs, N. Wiley, New York, 1976, Treatment of Patil Differential Equations,” Zienkiewicz, 0. C., “The Finite Element Method,” MeGraw-Hil, New York, 1977, APPENDIX CARTESIAN TENSORS 1.1 INTRODUCTION In this appendix we shall consider elements of Cartesian tensors. We shall er {tensor concepts in much of this text, and the accompanying notation where it is ceordingly not be used exclusively.) The material of this appendix has considerably greater application than simply for the study of solid mechani 1.2 VECTORS AND TENSORS mies courses we dealt primarily with two classes of quantities in phenomena, namely scalars and vectors. The scalar quantity, you will recall, required a single value for its specification and had no direction associated with it. On the other hand, we also worked with quantities that had both magnitude and direction. Accordingly, they could be represented as directed line segments. If we in accordance with the parallelogram law, we call them ly, you will recall that we could specily a vector by giving three scalar components of the vector for a given reference frame. In this appendix we shall view scalars and vectors in a somewhat different ‘manner so as to enable us to properly understand more complex quantities that we shall need for this text, namely tensors. First, note that a scalar quantity does not de- pend for its value on any particular reference. Stated another way, the scalar is in variant with respect t0 a rotation of the coordinate axes that may be associated with the problem. A vector, on the other hand, does require some reference for purposes of conveying its full meaning. Thus, for a velocity V we may refer to reference xyz (sce Fig. I-1a) in conveying its properties in the form: VaVuit Kyi t Vek 7090S EPPERES ES EEA DEE © ® FIGURE 1.1 Rectangular components of V for two sets of axes. for we may employ reference x’ tion in the form as shown in Fig. I.1b to convey the same informa- Va Veil + yi + VER Despite the need for a reference, certain aspects of vectors are independent of the reference, Thus, for example, the magnitude of the vector is the same for any refer ence that may be used. In other words, we can say that the sum of the squares of the vector components is invariant for a rotation of reference axes. Stated mathematically VI4V}+VESVE+VE+VEEVEY Also, when considering two vectors itis clear thatthe relative orientation of the vectors, as given by an angle @, is independent of the coordinate system. A number of Useful operations can accordingly be derived because of these invariances, These are the familiar operations of vector algebra. Vector equations involving such operations then can be stated independent of references. There are more complex quantities that have certain characteristics independent of coordinates. Furthermore, certain very useful operations can be developed for such quantities which do not depend on a reference and which can convey certain Physically important meanings, We will soon define tensor quantities as one such class of quantities. And, just as we have vector equations that represent certain physic laws, so shall we be able to present certain tensor equations that play a similar role. To investigate such quantities we shall first go back to scalars and vectors to establish ‘more useful definitions for these quantities which will permit us to proceed in an orderly way to the more complex quantities. As has been intimated in this discussion, the relation between a quantity and a reference is of critica importance. 13 INTRODUCTION TO TENSOR NOTATION We shall now introduce a notation that will permit us to express the transformation equations of second-order tensors in a very compact form. In doing so we shall be able Appendtx I: Cartesian Tensors = 711 to better understand the relation between scalars, vectors, and second-order tensors, and we shall at the same time be able to readily extrapolate our thinking to include so- called higher-order tensors. For this purpose, we introduce the concept of the free index. The free index is any letter that appears only once as a subscript in a group of terms. Thus, jis the free index in the following expressions: HAY When we have an expression with a free index such as V;, we can consider that we are presenting any one component of the array of terms formed by having i become anx,ay,oraz, This array accordingly is Alternatively, V; can be interpreted to represent the entire ser of components given above, Clearly, V; may then be considered a vector since it includes the three rec tangular components of the vector. The particular meaning of the free index depends ‘on the context of the discussion, Similarly, the expression Ay has two free indices and ccan represent any one component of the array of terms formed from all possible permutations of the subscripts with i and j taking on the values x, y, and z, This array ‘of terms must then be Alternatively, the expression Ay can represent the entire set of the above components. ‘Next we shall introduce the concept of dummy indices by prescribing that when letters f, j,k, f, and m are repeated in an expression we sum terms formed by letting the repeated indices take on successively the values x, y, and 2. Thus, the expression Ay¥; used earlier may represent a set of three terms each of which is a sum of three expressions. Thus, retaining the free index i we have Ag¥)=AgVg + AyVy + Ai ‘And if we wish to express the set for the free index we get (AxVx + AxyVy + AgV2) AWW) =| Aya Ayyly + Aye Asx Ve + AzyVy + AssVi,fle Appendix i Cartesian Tensors: If we have a double set of repeated indices such as Ay By, we let take on the value x and sum terms with j ranging over x, y, and z; then we add to this three terms found by letting # become y while letting j range again over x, y, and 2;ete, We thus get a sum of nine terms. It should be clear that, since a pair of dummy indices always sums out, the particular letters of the set ijklm used i A,B, = AB = Ay By 16. These ate called i 14 SCALARS AND VECTORS We shall now reconsider scalar and vector quant in terms of the notation that we have just presented, The new perspective that we achieve will help in setting forth tensor concepts later. First, note that since a scalar has only a single value we have no need for sub- scripts, And since there is no change in a scalar when there is a rotation of the refer- ence axes we have es in terms of rotation of axes and n M=M (i) as the relation between primed and unprimed axes. Now consider a vector V (or V;) shown in Fig. 1.2 as a accord with the pat seted segment. In lopram law, we have for the rectangular components of V: Ve = V cos (V,x) Vy = Veos (V.9) V, = V cos (V, 2) When the reference is rotated to a primed position as shown in the diagram, we wi hhave a new set of components that again can be computed from the parallelogram law as follows: = Vo 008 (2', x) + Vy cos (x,y) + Vz 0s (x', 2) = Vx 08 9", x) + Vy 08 9", ») + V; 008 6", 2) Vz = Ve 008 (2, x) + Vy 008 (2, ») + V, 005 (¢,2) We will now introduce forthe cosine terms above the letter a with two subscripts, The first subseript identifies « primed axis and the second subscript an umprimed axis. Furthermore, ay’y then represents the cosine of the angle between the x' and y axes and is thus identical in meaning to cos (x’,y). The above equation can then be writ Ver = Vege + Vyte'y + Vedy's Vy = Vadyie t+ Vyay'y + Vety's Ver = Vieds'g + Vydary + Vedy's Appendix I: Cartesian Tensors 713 FIGURE 1.2 Vector V and its rectangular components. Using index notation, we can express these equations as Ve =a; ay This is a transformation equation for rectangular components of V from un- primed to primed axes, This is often written as V5 (3) where the first subscripts for the a’s are understood to represent primed axes without the need for a prime. Also, the prime has been moved to change Vito Vj, We can con. clude that a necessary condition for a quantity to be a vector is that the components transform according to the above equations under a rotation of axes. We can also show that this is a sufficient condition. Thus, we can define a vector asa set of ponents that satisfies the above transformation eqt 1.5. TENSORS: SYMMETRY AND SKEW-SYMMETRY In the previous section we redefined vectors and scalars in terms of certain transforma: equations for the components of these q 's, considering a primed and iprimed reference rotated arbitrarily relative to cach other about a common ng defining transformation equations; origin, we have the follo SCALARS: M=M" vectors: Aj=ayAy We shall now generalize these transfor nore complex quantities, Thus, we will define a second-order tensor as a set of nine ‘two fee subscripts, as, for example, Ay) under a rotation from an unprimed to a primed set of axes according714 Appendix #: Cartesian Tensors Aig ayxajAn (ay where the a's are the familiar direction cosines discussed earlier. Many important {quantities in the engineering sciences are second-order tensors. And as a result of the above transformation equations, second-order tensors have certain distinct and useful properties. Notice the way the de transformation builds up from scalars, with no free subscripts, to vectors, with one free subscript, to the second-order tensor with two free subscripts. For this reason, we call the scalar a zeroth-order tensor and the vector a first-order tensor. Continuing on in this process, we can define a pth order tensor as follows: (sy tty ash Meals phe Ay Ip wherein we have sets of quantities requiting p free indices. We shall say that a tensor such as Tyg, is symmetric with respect to eny two of its indices~say jk-if the values of the tensor components corresponding to these indices are equal to the tensor components corresponding to the reverse of these indices. That is Tyar= Tay (1s) In the case of a second-order tensor we Wve for symmetry the single possi Ay (7) In the represer of this set as an array, it means that the terms on one side of the ‘main diggonal may be considered as mirror image values of the terms on the other side oe < (18) ‘We leave it for you to show for for second-order tensors we have transformation from primed to unprimed coordinates Ay=exaeyAia Appendix I: Cartesian Tensors 715, Furthermore, a tensor such as Tyjgt is said to be skewsymmetric in any two of its indices, say jk, if the components of the tensors with indices reversed are negatives of each other. Thus we have as @ definition of skew symmetry: Tig =—Tyy skew-symmettic in (19a) Tyxt=—Tray skewsymmetsic in jl (49) Consider next those components ofa skew-symmetric tensor wherein the indices involved in the skew symmetry have identical numbers. For Ty in Eq. (1.94), for tance, we ate refering to Tiuy, Typ, and Tiaay, and for Tyas in Eq. (9b) we are referring 10 Tiss» Troka» and Taxa. It is clear that such terms must be zero. Thus, considering T;,,, We see from the skew-symmetry requirement Tau =~To Clearly for each {and £ we must have Tj = 0 in order to satisfy the above con. By =-B aay ‘And in the array representation we have 0 Ba By) Bx 0 By “By By 0, where the main diagonal must be composed only of zeros. Now suppose we have a tensor Tix, which is symmetric i is skew-symmetric in s, r. Itis useful to note that the tensors for indices # and s, and j and r, give a set all of whose terms are zero. Thus: jj and a tensor Burep ner product of these T yt Bijtp = Crttp =0 Also, if a tensor Ayes is an arbitrary skewsymmetric tensor in indices if and if AjrsByj = 0 for a nonzero tensor By, then we can consider that By must be symmetric in ij. It is also true that if Ages is an arbitrary symmetric tensor in ij for which AjrsBy = 0 for a nonzero tensor By, then By must be skewsymmetric in ij. Note ij can be any of the indices of a tensor. Finally, we will demonstrate that taking the partial deriva ve of an nth-order tensor dp, p, pq with respect tox, where Is a fre index, resuls in a tensor of rank (2 1). We'shall tind it convenient here to adopt the convention of denoting partial derivatives of a tensor by making use of commas and indices as follows: 20 = #0 = Bey OM Sep SOW‘716 Appendix I: Cartesian Tensors Accordingly, we wish to show at this time that A, ..,,i8an (n+ Ith-order tensor Suppose we start with a primed reference x). Then we may say =o, int 324 " ax; in) jg i8 an nth-order tensor, we use its transformed form in the unprimed the above equation: ee Aicena = 2s Ouetnny Mutagen t= Senta batty) ‘Thus we have kyl ql a Ajgeind = icky Ging it = Ary) = Oph, “gga, intitle (Arey a The above statement is the defining one for a tensor of order (n+ 1) and so we have demonstrated that A, ..j,, is an (m+ I)th-order tensor. Since we can perform the operation of taking partials ordinates xj, xp, etc., we can conclude that ith respect to other co- a Ane tult = Bagg Mov) is an (1 + 2)th-order tensor, while -_* lt” a Be Oooo) isan (n + 3)th-order tensor. 1.6 VECTOR OPERATIONS USING TENSOR NOTATION: THE ALTERNATING TENSOR We have already shown that the dot product A+B can be expressed as A,B; in index notation. Another operation that we have used extensively is the cross product AX B=C, where ‘Append 1: Cartesian Tensors 117 Cy = 4yBy ~AgDy Cy =A3B, ~ A, Bs (12) Cy = A,B, — A2By In order to be able to get this result by using index notation, we introduce the alter nating tensor ey, defined a8 follows: eye =0 for those terms of the set for which i, j, k do not form some permutation of 1, 2, 3. (Example: if any two of the subscripts are equal, then such terms are 2010.) eyx = | for those terms of the s ‘can be arranged by an even number of permutatior wing indices that form the sequence 1, 2,3 or that to form this sequence. eye =I for those terms in the set that require an odd number of permutations to reach the sequence 1,2,3. Thus we have ena = ena = es == 0 13 = ern = ea12 = 1 ens = 601 = 1m =—1 Now if we go back to Eq. (1.12), we can readily demonstrate that the following expression represents the terms comprising the set of the cross product: C= cic AjBe (13) Thus, carrying out the double summation over dummy indices j and k, we get for #=1 C1 = eye AjBe = CA Br + €.A2Bi + €1s1AsBr + C1124Bs + €rad2Be FeymAaBy + e1194,By + e1na4aBy + ern yBy Now employing the defini alternating tensor we find C, = A.By—AsBy as you may verify. Now consider the triple sealar product (A X B)+C. Using index nota we get (AX B)-C= eye BRC (114) nore, you will recall from your mechanics courses that Ar Ay (AX B)-C=|B, By Bs a 6OE ORB ENOL EEE! TDS and so comparing Eqs. (1.15) and (1.14) we get the following expression for a deter- ‘inant: 4, Az Ay By By By] =eyAjByCy (1.16) Cr Cr Cs We may put the above relation in a more convenient form from an index point of view by noting that yj BeCr= en ABCy (.i7) where we have changed the dummy indices. Now egy can be made into éyx by two interchanges of indices, and so we have ens = Cie Accordingly, Eq. (1.17) can be written as e174) BRC = 6x AsBjCr We can then write Eq. (I.16) in the following manner: 41 Az Ay By By Bs |= ey, AsBjCy 1G C3 Let us next consider the field operators of vector analysis, using index notation. First, we have the gradient operator, which, acting on a function ¢, is given as, 29; 4 28, 8 v9 = 26; 4 9854 28 Oo art ay} * ae & To write V6 in index notation we can express the Cartesian unit vectors as follows: is iti k=i, Accordingly, we can then say Woo, (us) Also, the expression @ ;, with fas free index, is the index counterpart of V@. (A esis Smuemenscrgmee ®t: Teese: RES ae Appendix I: Cartesian Tensors = 719 ‘The divergence of a vector, you will recall, is given as follows in vector notation: AV 4 Vy 4 WV, vv = vey nhs 4 yy OMe dvVe vey ax ay ae Using index notation, we get av, wv = OM 1.19) divV ax, dis) Next we consider the cur! operator. Again from vector analysis, we have ijk a a a wivevxv=|2 9 9 . dx, ox dx Ve Vy Ve face Oenen Ax, Oe, Oxy MoM MN ‘Then by direct analogy to the determinant expression above, we can suy Ve f = yu LE = gn Vig 1.20) VV eye aot cue Vesh (120) Or considering the index 1 10 be a free index, we can express V X V indicially simply as Cun Vege Finally, we consider the Laplacian operator V2, From veetor analysis we have Hence, using index notation, we get VEY And so aay vie:FIGURE 1.3 Prismatic element in a body. 1.7 GAUSS'S THEOREM jon we considered the in tensor notation. Eq me we pt ‘Assume we have an nth-order tensor Ty. 's di and dx, as shown in the diay ing integral over the volume 5 of this, mm dx, dz dxs \g oUt integration with respect 10. we get [ff 10 0.0 trl v.F we can replace dra. ds mn the left end of the we generally write »,, vs, and Ps 38 der dyn Rd dg ith Appendix I: Cartesian Tensors 721 prism.t We thus have I Om Jdxy dxz des = [fo % dA)e offer. dA) Now, integrating over all the prisms comprising the volume V, we get I 2Or a as ao = ff On wae ff Cnn dd he F le of the above equation clearly covers the entire surface of volur is accordingly replaceable by a closed-surface integral. We then have the following statement: 2) for the x, dire y direction xj. Accordingl ment is given as follows: 1 J[e0- ff dna aay where / is a free index. This is Gauss's theorem in a generalized form.* Using the notation presented for spatial partial derivative, we can also give this equation as follows: fi Tx: rua ff Thu dd (1.24) Iv ‘The appearance of the minus sign here isa result ofthe use tion for area vectors. Thus, », on the left side domain we have chosen to work (@», dA) will be the positive num! |. We could have proceeded i the generalization of the above state- uously varying tangent plane on each piece. Th This theorem can be extended to bodies whose axes cut the composite bodies pe is such sre than twice, provided such a body can be decomposed into contiguous have the property specified in the development.Suppose that Tig... is the zeroth-order tensor @ ({e., a scalar), We then have Mfoo-ffom If we wish to revert to vector notation, this form of Gauss law becomes I veer ff oan as) where dA =vdA. Suppose next that Tjy... isa ft We then get [ff ve ffrne = pref free indices 7 and j m, We get vector) Vj replace by a 27) ‘equation in terms of vectors to arrive at (1.28) ise of this form of Gauss's theorem, form it is called the ler next that the vector fleld V is given V=oit vj +Kk ‘Then Eq. (1.28) can be given in the Following way: tre divergence theorem has 4 simple p 56 4 28 4 1) ay = FB (gaye + Ytyy + Ka) dA 2 (4 24 42) a ffcon + yy ) (1.29) mn of Green's theorem, we get the fol I (i+ 2) aA =f ras (30) We now set Y= Oand x = 0 in Ea (129) and et be the pro wand w. We get on Wf se off oe fff et -ffcome as Rearranging, we reach the following fo [for fom f= Generalizing for any coordinate x, we get [forefoot 0 ‘The correspondi [ee formar t— ff ma 32) ‘We have much occasion in this text ¢ of two fu representing an integration by parts ration by parts formula for two dimensions is ize the above f 1.8 GREEN'S FORMULA Another formu special forms. For formula and its various lowing integral 1s theorem may be proved without vectorial considerations for any three functions iecewise continuous and differentiable. See Kaplan (1951).724 Appendix I: Cartesian Tensors rafff oss ay (133) v wherein, you will note, we have as an integrand the inner product of two gradients. ‘The functions ¢ and are assumed to be continuous in the domain V with continuous first and second partial de is domain. The domain V has all the restrie~ needed for the developm. theorem, Note now that: ‘Similar relations may be written for x2 and x5 so that Eq. (1.33) may be expressed as follows: =- 2 (pv [fesse Mou off 62)« 34) Wemay now use Gaus’ theorem forthe lst integral. Thats, a, av - ay But on expanding (8¥/€xj)0 we rea Yin the Isoxye-foere ‘We can then express Eq, (1.34) in the following manner: jonal derivative of en's formula which, using vector not lowing form: readily recognized to have the [for [ff eros ffo% ax (36 Now write Eq, (1.35) with @ and y interchanged. Subtracting equat the left sides cancel, leaving the following ms, we see that i i ‘ (37) ‘This is the well-known Green's formula. In [ff ors veo ff (eo) a 038) ; , |. while the closed 1 integral becomes surface egral becomes tes 1.9 KRONECKER DELTA ‘Asa fina item we present the useful Kronecker delta defined as y=) tims by=0 ifiey Thus, d;2 = 815 = Oand by, = 89 = 1 J#k, we get zero contribution. Only when j= do we get @ nonzer ‘and for that expression Kronecker delta is unity. We thus get for rj ‘Accordingly, we can say ie = Aix (1.40) reasoning Aubin = Any REFERENCE Kaplan, W., “Advanced Ci ‘Addison-Wesley, Reading, Mas, 1951, Chapters PROBLEMS LL Given the following information: Ag=10 Be=3 C= 16fee 2 110 Appendix & Cartesian Zenvors AB) GRRE 16) AaPiCiAR Given te following aus: As? A=3 0 Ayes Bu-0 9 Bun? By? By =} Bat B=} By=6 0 By=-} By =0 G=! G=3) Ges compute or dents the aay fo the flowing: ABE OBR) AiBanG Show that if \ isa constant and Ay is «second-order tensor, then My isa second-order tensor. Show that if Ay and By are second-order tensors, then (Ay + By) isa second-order tensor. Evaluate the following terms: © Men Oar Mey We index & does not equal the index p. Also Show that ejuejp =0 for those terms where index & equals the index p. Thus, we can show that ej ei = 2 for those terms where ‘ay for the representation of ey ip eibyp = Cap = 0 ° Thus, show that ein sup = ep Finally, show thatthe contraction (make into a pair of dummy indies) of the above over kp gives 6. Show that 8 xg4m = 0. IV = 3x +292) + 102*K, compute Ce in © Viay Won? + 2siny +224, compute: er Meu Ova om Write aay @WOAB —) LAXBCID © ve Given the ftlowingssondorde tensor ¢ tay: 0 oo oo. Appendix I: Cartesian Tensors 727 Figure Ls wrt Gus haem Ta fra dain shown in 18. 12 Ghee deus or oe Vextyit a forte donala hownis Het 13 tno by ts teeing tga et Lanter /, de » De otto tenet, 14 Ghent folowing sve eur 1000 0 S000) =[ 0 3000 4000 \-5000 4000-3009, find ry