The determination of the elastic field of an ellipsoidal inclusion, and related problems By J.D. Esuenpy Department of Physical Metallurgy, University of Birmingham (Communicated by R. E, Peierls, F.R.S.—Received 1 March 1967) .pposed that « region within an isotropic clasti solid undergoes a spontancous change ‘which, ifthesurrounding material wore absent, would be same preseribed homogencous Aeformation. Because of the prosence of tho surrounding material stresses will bo proaent both inside and outside the region. The resulting elastie fll may be found very simply with tho help of a sequeneo of imaginary eutting, straining and wolding operations. Tn particular, ifthe ‘ogion is an ellipeoid the strain inside it is uniform and may be expressed in. terms of tabu lated olliptio intograls. In this caso a furthor problom may be solved. An ellipsoidal region im an infinite medium has elastio constants different from thosoof the rest of the material; hhow dooa the presence of this inhomogeneity disturb an appliod stross-ield uniform at large distances? Tt is shown that to answer several questions of physical or engineering interest it ia necessary to know only tho relatively simplo olaati field inside tho ellipsoid. 1. Tyrropuerion In the physics of solids a number of problems present themselves in which the uni- formity of an clastic medium is disturbed by a region within it which has changed its form or which has elastic constants differing from those of the remainder. Some of ‘these problems may be solved for a region of arbitrary shape. Others are intractable unless the region is some form of ellipsoid. Fortunately, the general ellipsoid is versatile enough to cover a wide variety of particular cases. It is the object of this, paper to develop a simple method of solving these problems. When « twin forms inside a crystal the material is left in a state of internal stress, since the natural change of shape of the twinned region is restrained by its surroundings. A similar state of strain arises if a region within the erystal alters its ‘unconstrained form beeause of thermal expansion, martensitic transformation, pre- cipitation of a new phase with a different unit cell, or for some other reason. These examples suggest the following general problem in the theory of elasticity. The transformation protilem A region (the ‘inclusion’) in an infinite homogeneous isotropic elastic medium undergoes a change of shape and size which, but for the constraint imposed by its surroundings (the matrix’), would bean arbitrary homogeneous strain. What is the elastic state of inclusion and matrix? We shalll solve this problem with the help of a simple set of imaginary cutting, straining and welding operations. Cut round the region which is to transform and remove it from the matrix. Allow the unconstrained transformation to take place. Apply surface tractions chosen so as to restore the region to its original form, put it ack in the holein the matrix and rejoin the material across the cut. The stress is now zero in the matrix and has a known constant value in the inclusion. The applied (376)The elastic field of am ellipsoidal inclusion 377 surface tractions have become built in as a layer of body force spread over the inter face between matrix and inclusion. To complete the solution this unwanted layer is removed by applying an equal and opposite layer of body force; the additional elastic field thus introduced is found by integration from the expression for the elastic field of a point force. So far nothing has been assumed about the shape of the inclusion, However, we shall find that if it is an ellipsoid the stress within the inclusion is uniform. This fact. enables us to use the solution of the transformation problem as a convenient stepping-stone in solving a second set of elastic problems. Superimpose on the whole solid a uniform stress which just annuls the stress in the inclusion. ‘The removal of the unstressed inclusion to leave a hole with a stress-free surface is then a mere formality, and we have solved the problem of the perturbation of a uniform stress field by an ellipsoidal cavity. More generally, suppose that the uniform applied stress does not annul the stress in the inclusion. Then the stress and strain in the inclusion are not related by the Hooke law of the material, since part of the strain arises from anon-elastic twinning or other transformation with which no stress is associated. The stress and strain are, however, related by the Hooke law of some hypothetical material, and the transformed ellipsoid may be replaced by an ellipsoid of the hypo- ‘thetical material which has suffered the same total strain, but purely elastically. We have thus solved the following problem. The inhomogeneity problem An ellipsoidal region in a solid has clastic constants differing from those of the remainder (if, in particular, the constants are zero within the ellipsoid we have the case of a cavity). How is an applied stress, uniform at large distances, disturbed by this inhomogeneity? ‘The strain in the inclusion or inhomogeneity may be found explicitly in terms of tabulated elliptic integrals. The elastic field at large distances is also easy to deter- mine. The field at intermediate points is more complex, but for many purposes we do not need to know it. In fact, knowing only the uniform strain inside the ellipsoid wwe ean find the following items of physical or engineering interest: (i) The elastic field far from an inclusion. (ii) All the stress and strain components at a point immediately outside the inclusion. (iii) The total strain energy in matrix and inelusion. (iv) ‘The interaction energy of the elastic field of the inclusion with another elastic field. (v) The elastic field far from an inhomogeneity. (vi) All the stress and strain components at. a point immediately outside the inhomogeneity. (This solves the problem of stress concentration.) (vii) The interaction energy of the inhomogeneity with an elastic field, (viii) The change in the gross elastie constants of a material when a dilute dis- persion of ellipsoidal inhomogeneities is introduced into it. ‘Problems (i) to (iv) ean also be solved for an inclusion of arbitrary shape, (i) and (iv) trivially, (ii) and (iii) if we can evaluate the necessary integrals. They differ, of378 J.D. Rshelby course, from the problems considered by Nabarro (1940) and Krdner (1954) in whieh, the inclusion breaks away from the matrix, Problems (v) to (viii) can only be solved for the ellipsoid. They each have an analogue in the theory of slow viseous flow. ‘Many particular eases of these problems have heen discussed. Robinson (1951) gives references to earlier work; see also Shapiro (1947), Sternberg, Eubanks & Sadowsky (1951). Apart from some increase in generality (we consider shear trans- formations and the disturbance of an arbitrary shear stress by an ellipsoidal inhomo- geneity) our treatment is, perhaps, rather simpler and more direct than theorthodox method. Nowhere do we have to introduce ellipsoidal co-ordinates, search for suit- able stress functions or match stress and displacement at an interface. Indeed, we do not even use the equations of elastic equilibrium explicitly except in certain of the applications (i) to (viii). 2. ‘THE GENERAL INOLUSION We employ the usual suffix notation. A repeated suffix is summed over the values 1, 2, 3 and suffixes preceded by a comma denote differentiation: p= Bult 9, ue = PP 24,2, ‘The elastic displacement u,, strain ¢,, and stress p,, are related by 45 = Ha y+ tea), (21) Pas = Renn Beg + tess (2-2) in an isotropic medium with Lamé constants A, 1. When a particular set of elastic functions are distinguished by an affix (e.g. w?, ef, pG), it is to be understood that they are related by (2-1) and (2-2). It is often convenient to split a second-order tensor f,, into its scalar and so-called deviatorie parts: Fis = ts bie where Frm and ‘f, — 88s. Thus, for example, (2-2) may be written B= Bee, “Dey = Me'ey (R= A+ 4M), (2:3) and the inversion e ISK, "eis = "Pile is immediate, whereas to find ¢,; in terms of p,, from (2-2) is more difficult. For two tensors fz», 4 we also have the convenient relation fig; = 4/9 + Jiy'ujs there are no cross-terms between the scalar and deviatoric parts. The elastic energy density is thus il Bouts = URP+ Ip 'ey)'¢u) = 5 Pe agPu'Pa) (24) Following Robinson (1951) weshail givethename ‘stress-freestrain’ tothe uniform transformation strain ef; which the inclusion would undergo in the absence of the matrix. The main problem isto find the ‘constrained strain’ ef in the inclusion when it transforms while it is embedded in the matrix and also the strain set up in the matrix, which we shall also call ef. Let S be the surface separating matrix and inclusion, 2, its outward normal and dS, = n, dS the product of the normal and an clement of 8. We now earry out the steps outlined in the Introduction.‘The elastic field of an ellipsoidal inclusion 379 I. Remove the inclusion and allow it to undergo the stress-free strain ef, without altering its elastic constants. Let PU = Rey + Quel be the stress derived from ef by Hooke’s law. At this stage the stress in the inclusion and matrix is zero. IL. Apply surface tractions — pZn, to the inclusion. This brings it back to the shape and size it, had before transformation. Put it back in the matrix and reweld across S. The surface forces have now become a layer of body force spread over 5. TIL Let these body forces relax, or, what comes to the same thing, apply a further distribution +pJ)n, over 8. The body is now free of external force but in a state of self-stress because of the transformation of the inclusion, Since the displacement at r due to a point-force F, at r’ is (Love 1927) ieee 1 7_o , 2 Ule—1) = Sale =P] tempo) aaa, | Tb 5) the displacement impressed on the material in stage III is g(r) = f aSpRU lr), (28) 8 where o is Poisson's ratio. It will be convenient to take the state of the material at ‘the conclusion of stage II as a.state of zero displacement. This is reasonable, since the stress and strain in the matrix are then zero and the inclusion, though not stress-free, has just the geometrical form which it had before the transformation occurred. ‘uf is then the actual displacement in the matrix and inclusion. The strain in matrix or inclusion is G = Mel, + ug). ‘The stress in the matrix is derived from ef by Hooke’s law: PE = ACB y+ 208, ‘On the other hand, the inclusion had a stress — pf, even before stage TIT, so that the stressin itis phy = pij— Dh = Ae —0) bi, + 2u(04 ef), ee where, according to our general convention, pf is the stress derived by Hooke’s law from the strain of in the inclusion. By using Gauss’s theorem and the equivalenes of @{éz, and —2/éx%, when acting on |r—r" |, (27) may be made to read i 1 ae x : 8 = empha) PIR Fog Ph Pe (28) where o-f and ef |r-r’|ao are the ordinary Newtonian potential and the biharmonie potential of attracting matter of unit density filling the volume V bounded by S. Evidently var 26 (29) — Sr inside 8, and Voy = 2996 = Hm avap { O outside 8, a5 Vol. agn. A.380 J.D. Eshelby Generally we must know both y/and ¢. However, if we are interested only in the dilatation in the material, it is enough to know ¢: 1-20 a . =~ Sad aye (210) Again, if ef is a pure dilatation 4e7%,, then jap tbe 4 3(1—0) aresult due toCrum (Nabarro 1940). In this case the dilatationise?(1+0)/3(1—o)in ‘the inclusion and zero in the matrix. Thus, for example, with o = }, the constraintof the matrix reduces the free expansion of the inclusion by a factor . ‘The second derivatives of a potential function satisfying V?U = —4np undergo a jump AU, ,; = —47Apnyn; on crossing a surface (with normal n,) across which the density jumps by Ap (Poincaré 1899). (Thisis perhaps more familiar in the form: the jump in attraction across a double layer is equal to its moment. In our problem —§,; is the potential of a double layer over $ with unit moment directed along the 2, axis and ¢; is the corresponding force.) This gives for ¢ the expression 9,1 (out) ~ 9 «5 (in) = dorm; (21) for the difference at adjacent points just inside and outside 8. Applying the same argument in turn to y ;;, which is the potential derived from the density — 26 [47 ae Veal OUt) —Yeepalin) = Sangnyrne (212) ‘From (2-11), (2-12) and (2-8) we can find the stresses and strains just outside the inclusion from their values at an adjacent point just inside without having to solve the exterior problem at all. We easily find that e (out) = and ‘eG (out) = 'ein) +5 "eRenymynem—ehongm een g i lite . o eF (mem $8y)- (2-13) ‘The C quantities are related by (2-2) and so are the 7’ quantities. Thus either or both sides of these equations may be expressed in terms of stress without trouble. This solves problem (ii). We can find a convenient alternative form for uf by noting that (2-5) may be written as onl —0)U, = Ft [8 Jo) 855+ feeble]. (214) Inserting this in (2-6) and using Gauss's theorem to convert to a volume integral we find wf, sent, ts)The elastic field of an ellipsoidal inclusion 381 where rand 1 = (1,,%,5/,) are the length and direction of the line drawn from the volume element dy to the point of observation x and Sige = (1-20) Bisle+ bunds) — Split Sh lyley (2-16) Gis = (1—2e) Bile + uly — Subd) + Biller (2-17) For points x remote from the inclusion we may take everything except dv outside the sign of integration to obtain UP (x) = Vf fignl Lomp(l—o) 1? = Vefagin| 81-0), (2:18) where r and 1 are now the distance and direction of x from the inclusion. This solves problem (i). ‘The strain energy density in the inclusion is }p/,¢¥,, where ef, is the strain derived from p{, by Hooke’s law. By (2-7) the elastie energy in the inclusion is thus {piles —efyav. (219) al ‘The elastic energy in the matrix is 1 o. 1 1 — af eimeass=—3f vhuras,=—5 1 phefae. 20) ‘The first member exhibits it as the work done in setting up the clastie field by applying suitable forces to the surface S; the sign is correct if the normal points from inelusion to matrix, The second follows because displacement and normal traction. are continuous across 8. The third follows from Gauss’s theorem, the equilibrium equation pj, , = 0 and the symmetry condition pf, = pf. The total strain energy in matrix and inclusion is thus Eye =H wea, (221) In the special case where ef is a uniform expansion we have at once from (2:10) and (221) By, = 2u(e?"V(1+0)/9(1—c), whatever be the shape of the cavity, as pointed out by Crum (Nabarro 1940). ‘The interaction energy of the elastic field wf with another field uf is (Eshelby 1951, 1956) Bi = | lobut rut) as, 222) taken over any surface ¥' enclosing the inclusion, Let us take ¥ to be a surface just outside S. Onceagain, since uf and the normal stress are continuous across 5, (2-22) can be converted into an integral over a surface just inside S, and hence intoa volume integral over the inclusion: Ess = [obes—agesnae. ‘The second term in the integrand is equal to — p{e4, so that Ese, -f phefdv= f Ptehao=—[ paul as, (28) r v 2 252382 J.D. Eshelby ‘This solves problem (iv). The same result is reached by evaluating (2-22) over alarge sphere using the remote field (2-18). It is fortunate that we need only e% and not ef. In fact the last member of (2-23) has formally the appearance of being the work done against the external field in “blowing up’ the inelusion (regarded as rigid) to a final shape specified by e%. It is perhaps not obvious that this should be so, since the inclusion is notrigid and its final shape is described hy a displacement uf which may be quite complicated (e.g. it produces a barrel or pincushion distortion of the cubical inclusion which Cochardt, Schoek & Wiedersich (1955) consider). If we regard the inclusion as capable of moving through the matrix, as in the elastic model of a sub- stitutional atom, B= ~ Bye 0% (2-24) is the ‘force’ on the inclusion, where &; is a veetor specifying its position, Let Eisuns, be the change of internal energy when the inclusion transforms in the absence of the matrix. Consider the sum B= Basan, + Ba, + Bit Give their adiabatic values to A, «, x and suppose that the constrained transforma- tion occurs without any heat flow. Then F can be interpreted indifferently as the enthalpy change of the inclusion, the enthalpy change of the body (inclusion plus matrix) or the change of internal energy,of the body and loading mechanism regarded as a single thermodynamic system. There is a similar interpretation for an isothermal process if we read ‘Helmholtz. free energy’ for ‘internal energy’ and ‘Gibbs free energy’ for ‘enthalpy’ and give A, 1, « their isothermal values. Since problems (v) to (viii) ean only be solved for an ellipsoid their discussion is deferred to § 4. As a simple example of the use of (2-18), suppose that we need the field at large distances from a dislocation loop of area A in the 2, plane with its Burgers vector along the positive x, axis. We have to insert a sheet of material of area A and thick- ness }. One way to do this is to cut out a disk of area A and height h, give it a per- manent strain ¢% = b/h to increase its height by b and then foreo it back into the cavity. In(2-18) we have toput V = Ah, eX, = b/hand the other e?, equal to zero, Thus 14, = 6A gag) 8m(1—o) 2, ‘Suppose next that the Burgers vector lies in the plane of the loop and, say, along the zr, axis. We now give the disk a permanent shearef, = 30/1, which gives its upper and lower surfaces a relative offset b, and re-insert it in the matrix. In the limit k->0 we have a displacement discontinuity & across the loop. Putting V = Ah, ef, = eh = Ybjh and the other ef zero in (2-18) we get = bAgas/4m(1—o) 24, (2-25) reproducing a result of Nabarro’s (1951). Tt is perhaps not quite clear that the restraint of the matrix will not reduce the offset to something less than 5. Actually this is not true in the limit h—-0. But we can see that (2-26) is correct by inserting Vek = }Ab in (2-23); this gives Bj, = —bApd, which is the correct interaction energy for such a loop (Nabarro 1952). Indeed by the same argument we can find theThe elastic field of an ellipsoidal inclusion. 383 remote field of an arbitrary loop of area A, normal n, and Burgers vector by. The interaction energy is —b;p§jn;A for any pf. Equation (2-28) then shows that VeF, = 2(bynj-+byn,) and (218) gives Ab nyu! 0) 7, uy ‘There is, in fact, a more general connexion with dislocation theory. ‘The stress-free strain in the inclusion may always be imagined to be (or may actually he) the result of plastic deformation. A set of dislocation loops (with equal Burgers vectors) expanding from zero sizeon a close set of equally-spaced planes will give a shear if their Burgers vectors lie in the planes, or an extension perpendicular to the planes if their Burgers vectors are at right angles to the planes. (In the latter case their movement is non-conservative.) If the deformation oceurs in the absence of the matrix these loops will, so to speak, disappear into free space. But if the inclusion is ‘embedded in the matrix the dislocations will lodge in the surface 8 separating matrix from inclusion. $ then beeomes the discontinuity surface of a Somigliana (x9x4, 1915) dislocation. In this generalized type of dislocation there is a variable discon- tinuityd,of displacement across 8. Inour model, d;makes itself felt through the gaps and interpenetrations of matter which we should find if we tried to re-insert the transformed inclusion into the hole in the matrix without pulling it back to its original shape. It is easy tosee that d, has the value Je2.2} at a point 2’ of S and henee to verify, after some manipulation, that our expression for uf agrees with Somigliana’s. 3. ‘THE ELLIPSOIDAL INCLUSION In discussing the elastic field inside an inclusion itis eonvenient to redefine the fin (2:16), (2+17) to be the direction cosines of a line drawn from the point of observation (ry, ,%,) = (,,2) to the volume element de. This involves changing the sign of the integrals in (2+15). Let us first integrate over an elementary eone dol) centred on the direction 1 = (1,,,/,) = (lm, 7) with its vertex at x. It gives a contribution 1(l)do to fave. ‘Thus Smy(1— 7) u(x) = ~ il (1) deo(l) gisx(1), (3-1) which gives the displacement at x in terms of an angular integration over the polar diagram r = r(1,m,n) of the surface S as viewed from x. More briefly we could go directly from (2-6) to (3-1) by writing} = 4V%rin (2-14) applying Stokes’s theorem in the form nous, and noting that do = n,r,dS/r° For the ellipsoid XYa2+ YY+ Be 7(1) is the positive root of (z+ rlyPa?+ (y+ rm)?/b2+ (2+ npc? = 1,384 J.D, Eshelby that is, e(l) = —flg+ (fl +elg)t, (32) where 9 = Pat + mio + nile? (3) and f= lala?-+my[?+nele®, ¢ = 1—atat—y2|oe—24]c, The sign of the square root is evidently correct, since ¢ is positive if x is within the ellipsoid. In any case, we may omit this term when (3-2) is inserted in (8+1) since it is evenin1, whilst gig, is odd. ‘To retain the advantages of suffix notation we introduce the ‘vector’ Ay= Hat, Ay= mb, Ay = njot. ey, A ‘Then ft = gon [| Anita (4) watessene x)= thf, Mint ina a =o) Jee are uniform and depend only on the shape of the ellipsoid. ‘The same is also true for an anisotropic medium. (‘This verifies a hypothesis of Frank’s (private communica- tion).) Forit can easily be shown (sce, for example, Eshelby 1951) that (2-5) has then to be replaced by Use) = F.D,l)Ir, where the functions of direction D, cannot generally be found in finite form. A repetition of the argument will evidently lead to an expression like (3-4), but with the gj, no longer given by (2-17) It is convenient to write the relation between the constrained and stress-free strains in the inclusion in the form 8 = Samm Cn (3°6) From the symmetry of the problem it is clear that the Siz, have some of the ‘properties of the elastic coefficients of an orthorhombic crystal with its axes parallel to the axes of the ellipsoid, though relations of the form Sy» = Sggyy ate not valid. Coefficients coupling an extension and a shear (S112, Sina, Sgax«+-) of one shear to another (Sjass---) are zero. In fact, (3:5) vanishes if any one of J, m,n appears raised to an odd power in the integrand. The reduction of surface integrals of the type J 1%nin** dwg to simple integrals has been given by Routh (1892). We find Si = On +R, Suse = QB*Ly, — Rly, (87) Siar = @ Ha? +6) Lin + RAL +b), 3 1-20 where Q= sna) = gals)’ 4Q+R Pdo ~~ dw acu mabe f° wwe aie i “I (38) Edo du. I= [ig 20 | esas dw Ne -\er 9 trate) DOA with A= (P+ ut (+ wh(e+uyh‘The elastic field of an. ellipsoidal inclusion 385 ‘The remaining coefficients are found by simultaneous eyclie interchange of (1, 2,3), (a,5,¢), (I,m). Ip, Jy, F, ovcur as coefficients in the expression b= He —2) L+H -y) ht HOA) for the Newtonian potential within an ellipsoid of unit density. We havet (Kellogg 1929) drrabe —————-— (F' — B) Baap 4nabe (ese yt Pey@ail ae ™P tae @ where F = F(6,k) and E = E(6,k) are elliptic integrals of the first and second kinds of amplitude and modulus = sin (1 eta), b= (a? 02) (a? —c2)} and it is assumed that a>b>e. The relations L+h+l= 4, (3:10) Toa Fay + Ine = 471/304, (311) PL yy + 0% yy + Oye = Ly (3:12) follow from the w integrals when we use the definition (3:3) of g and the relation E+ m24-n2 = 1, Again, if we split the factor (a2-+u) (b+ u)~1 in the w integral for J,, into partial fractions we have 3(a*—04) I) = J,—J,. Thus when I,, J, have been calculated from (3-9) we have for J, 1, =40-1,-I, and the remaining quantities are found from Fay = (hy — 1a) 3(a — 8°), (318) Ing = 47/302 — Ty, Tye (314) and their cyclic counterparts. For the oblate spheroid (a = b>) with ia aca forse Le () -2)4 (8:15) therelation (3-13) fails, though notits analogues for J,, or J, But from the w integrals for I,, and Ly, it is clear that J,, =3/,, and we may use (+14). For the prolate spheroid (6 = ¢b, k = (1—b%a%)}, k= Bla; “i Ti-ok ez = m(2—o)/4(1-o); a392 J.D. Eshelby Here E(k) and K(k) are complete elliptic integrals. It follows that for a very thin plate y approaches zero and there is complete accommodation. If the operative shear is ef instead of e%, tending to deform the plate in its own plane, the corre- sponding accommodation factor 1 —2S,.,. approaches unity as the thickness of the plate decreases and there is no accommodation. We may compare these results with the case where ef is a pure dilatation, Then, as we saw in §2, Bu, = pV eP (140/910) whatever is the shape of the inclusion. In @ rigid matrix the energy would be 4«V(e7)2and so the accommodation factor is always 2(1—20)/3(1—o)or }fora = 4. Asa rough illustration of how these results might be used, consider the formation of martensite in iron, Zener (1946) has shown that the thermodynamics of the process suggest that a strain energy of 200 cal is associated with each mole trans- formed. Suppose that the transformation involves a 5% volume expansion and 2 10°shear, so thate? = 0-05,¢% = 0-009. Itiseasily seen that the total strain energy is the sum of the values it would have if the dilatation or shear acted alone. With = 8x 10% dyn/em?, ¢ =}, and V one molar volume, the shape-independent dilatational contribution to the energy is 25¢al. This leaves 265 cal for the shear contribution. The quantity 2 (e)*has the value 1900cal. Ifthe transformed region issupposed to be an ellipsoid the accommodation factor is thus y = 265/1900 = 0-14, and this tells us something about its shape. For example, if it is assumed to be a circular disk, (5-3) shows that its thickness/diameter ratio must be 0-08. In the presence of an applied stress the free energy change associated with the transforma- tion becomes Fy, + Bs, instead of Fy, For the case we have been considering equation (3-22) gives Ejay, ~$8-10-*( —}p4)—3-10-pif cal/mole if the applied field pf is measured in dynes/em*, Fora cavity, the equations (4-3), (4-4) for the ellipsoidal inhomogeneity simplify to oh Sieh = ef (4) he ‘we shall only consider this case. Suppose that an ellipsoidal cavity is perturbing a simple shear ef, have at onee for the equivalent stress-free strain, putting jz, = 0 in (45), oh = esl, (65) with the notation of (5-2). The interaction energy is Bios, = —4V hel = — VS*/2uy. (56) If we let the ¢ axis of the ellipsoid become very small we have an elliptical crack. From (5-3) itis clear that Vy remains finite asc approaches zero and the interaction energy of the crack with the applied shear stress S is, Egy, = —2nad?S4/3ye. Consider next the displacement of the faces of the crack. If the c axisis still finite, the displacement of a point 2; at the surface of the cavity is (eG+w§)a;The elastic field of an ellipsoidal inclusion 393 plus the displacement uf due to theapplied field, We suppose that wf = in the plane of the crack. If we evaluate uf’ from (3:6) and (3-19) and pass thelimit > 0 and use (5:5) we find that in the plane of the crack. E uf = (bS|uy) (1 —a3/a?— 23/09) = Au, (57) uP = 0, uf = amr, a = m(1— 20) bef/4(1—o) a9; the + and — signs refer to the upper and lower faces of the erack. Thus the plane of the crack is tilted through an angle a, but it remains a plane. The relative displace- ment Ati, of the faces is everywhere parallel to the z, axis and has an ellipsoidal distribution. ‘There is a problem in dislocation theory closely related to the theory of the sheared erack, Under the influenceof a stress yi = S dislocation loops expand inthe 1, plane from a source at the origin and pile up against an elliptical barrier until thoir back-strossos annul pif at the source. What is their distribution if each loop is in equilibrium under the combined action of the other loops and the applied stress? In the limit of a large number of loops with very small Burgers veetors the crack and dislocation problems coincide (cf. Leibfried 1954). If the souree has its Burgers vector parallel to the 2, axis the dislocations are of pure edge type where they cross thea, axis and the number of them erossing a length day is dry 2At (2,22, = 0)/A 2. Where they cross the zy axis they are of pure serew type and their density is @Au,(x, = 0, 2,)/A bz, The interaction energy of the loops is given by (5-6). In diagrams the tip of an array of piled-up dislocations is often drawn curling up or down. The remarks following (5-7) do not support this. As a further example consider a spheroidal cavity (6 = c) ina material subject to simple tension stress 7". If the a axis coincides with the direction of T we need only know ef; to find the interaction energy. The non-zero components of ei are ef, cei. From (5-4) we have ened ch = ee, where (58) © (1=By. Seq) (T= Si1) = 281s Sn? with the abbreviated notation 8, = Siu, Sis = Siiss-- On the other hand, if the a axis is at right angles to the direction of T we only need to know ef. The non-vanishing applied strains are ef, eff =e = — ced and we find wpe Oy ee (Uo) 0-8il—aoh* EE TES By TES F818) 25,08 (60) or, say, @ = beg. ‘The interaction energies are respectively Eq(l) = —2VePYE (6-10) and Bin) = —3VETP/E, (1) whore # is Young’s modulus. Iethe direction of 7 remains unaltered and the cavity changes from the parallel to the perpendicular orientation, the interaction energy changes from (5-10) to (5:11).304 J.D. Eshelby ‘The parallel or transverse orientation is energetically favourable aceording as the spheroid is oblate or prolate. We may take the case of the prolate spheroid as illustrating the orienting effect of an applied stress on a di-vacaney in a metal (A. Seeger, private communication). For ajc = 2,0 = § we find ¢ = 224, ¢ = 588, ‘When approaches zero the numerator of ¢ becomes 1 — 2c and the denominator approaches zero as an(1—2c)/4e(1 —). The product eV in (5-10) remains finite and ‘we reproduce Sack's (1946) value —8¢*(1 —o2)7/2H for the interaction energy of a penny-shaped crack in tension. ‘The total strain at the surface of the cavity ef; + ef is, according to (5-4), given by the right-hand sidesof (2-13) with e©, ‘e% replaced by e”, ‘ef. The normal is given by (3:28) and the stress concentration can be found from (23). For the spheroid in tension we must supplement (5:8) by ch = && = ef\(6— 68, —1)/28 9 and (5-9) by (Sy) ef = Saleh + eh) — red. ‘These results actually apply to a quite general state of triaxial stress symmetrical about the polar axis of the spheroid, for in the applied strain haha och, ehh =0, we may take ¢ to be any number unconnected with Poisson’s ratio. (The ¢ implicit in the Si; must, of course, be put equal to Poisson’s ratio.) The stress concentration about an ellipsoid in shear is fond similarly from (5:5). ‘Two-dimensional problems involving an infinite elliptic cylinder ean be dealt with. similarly, using (3-17) and (3-18). 'The interaction energy per unit length is, Eine, = —BAvieh, where A is the cross-sectional area of the cylinder. The passage to the limit b-> 0 or ‘4-0 is in this case very easy and we ean derive well-known results for eracks in plane strain tension or shear. Another simple case is that of a crack joining the points = +a, y = 0 perturbing a uniform stress ps = 8, Both before and after the introduction of the crack there is a state of anti-plane strain in which the displace- ment is everywhere perpendicular to the ay plane. The interaction energy and relative shift of the faces of the crack are &, —na*st/2 and Aus = (2S/x) (42-29). (512) Several writers have derived approximate expressions for the reduction of energy by a crack (or an array of dislocations simulating a crack) by supposing that the applied stress is effectively relaxed to zero in a region about the crack whose dimen- sions are of the order of the width of the crack. (In the same sense we might say that in (5-10) or (611) the ‘energy drainage volume’ of the cavity was ¢ or £ times its geometrical volume.) This method gives correct results, but the logic behind it is not clear. If the applied stressis maintained by constant external loads, the elastie energy is increased by a certain amount when the crack is introduced (compare equation (4-8)). At the same time, the loading mechanism has to expend twice this amount of‘The elastic field of an ellipsoidal inclusion 395 work, Thus the decrease of total energy (—~ By.) is numerically equal to the inerease of clastie energy. On the other hand, if the applied stress is due to sources of internal strain, or if the body is strained by rigid clamps, the elastic energy (now the total energy) clearly decreases (compare equation (4-9)). But even here the decrease is in no simple sense located near the crack. This may easily be shown explicitly for the case of the crack in anti-plane strain (equation (5+12)). In terms of the elliptic co-ordinates &, 7 defined by x= acoshfcosy, y =asinhgsing, the displacement is ‘uy = (a8 |p) cosh E sin. (513) For, when £ is large enough for hyperbolic sine and cosine to be indistinguishable it reduces to the uniform state of shear tt, = Syl = (aS|)sinh sin 7, (514) whilst the traction on a curve £ = const. is proportional to ét9/28 and so vanishes on the limiting ellipse £ = 0 defining the crack. The energy in the small rectangle Edy at E77 is 4ye{ (Oty 9E)? + (Bua/0n)%} AE dy. Tf we evaluate this using the displace- ment without ((5-14)) and with ((5-13)) the crack, we find that the change of energy density AZ at any point due to the introduction of the erack is given by ABAEdy = 4S" /p1) 008 2ndEdy. (5-15) We may say that there is stress relaxation between the hyperbolas » = 47, jm and stress concentration outside them. The integral of (5-15) over any ellipse with the ends of the crack for foci is precisely zero. By judiciously deforming the ellipse we can find a curve within which the energy ‘relaxation’ is positive or negative Attempts to evaluate interaction energies in this way lead not only to errors of sign {which may be corrected by common sense), but also to incorrect numerical factors, ‘The problem of a rigid and incompressible ellipsoidal inhomogeneity is also relatively simple, since (4-3) and (4-4) reduce to Swath = — ‘From the solution Goodier's (1936) analogy enables us to find the perturbation of the slow motion of a viscous fluid when a solid ellipsoid is immersed in it. We have only to puto = fin the matrix and interpret , 1,, ¢;, and p,, as viscosity, velocity, rate of strain and stress. The energy density becomes half the rate of dissipation of energy per unit volume. Equation (4-11) or (4-12) enables us to find the viscosity of a dilute suspension of ellipsoids, Bix. is positive for a rigid inclusion and so the viscosity is inereased. Equation (4-11) now states that a viscometer working at constant load will produce a lower rate of deformation and so will dissipate less energy, whilst equation (4-12) states that a viscometer working at constant speed will have to work harder to maintain a prescribed rate of strain, For a single immersed ellipsoid the increase in the rate of energy dissipation is clearly twice H,., for the related elastic problem. ‘The calculation is much simplified by the fact that for o = 4, R = 0 in (3-7), whilst the dilatations e4, 7, e° are all zero. ‘We can easily verify, for example, Jeffery’s (1922) expression for the energy dissipa- 26 Vol 241. A.396 J.D. Eshelby tion by a prolate spheroid, as amended by Hisenschitz,(1933)."To find the viscosity of a dispersion of spheroids it is necessary to decide what orientation they will take up. ‘The elastic analogy suggests (though it does not prove) that they will ultimately orient themselves so as to minimize the energy dissipated. This agress with Jeffery’s hypothesis, verified experimentally by Taylor (1923). ‘REFERENCES Broggemann, D. A. G. 1937 Ann. Phyo, Lps., 29, 160. Byrd, P. F. & Friedman, M.D. 1054 Handbook of elliptio integrals, p. 4, Berlin: Springer- ‘Verlag. Cochardt, A. W., Schoek, @. & Wiedersich, H. 1953 Acta Met. 3, 533. Dirichlet, G. L. "1839 Verh. K. Preuss. Akad. Wiss. p. 18 (Werke, 1, 383). Bisenschile, Re 1933 Z. phys. Chem. A, 163, 138. Eshelby, J.-D. 1931 Phil. Trans, A, 244, 87. Buhelby, J.D. 1933 Acta Met. 3, 487 Eshelby, J.D. 1936 Tn Progress in solid stale physic (ed. Seitz & Turnbull), 8, 79. New York ‘and London? Academie Pres Goodior, J.N. 1936 Phil. Mag. [7], 22, 678. Hashin, 2. 1955 Bull. Ree. Coun. Toradl, C, 5, 46. Telfery, G.B. 1922 Proe. Roy. Soe. A, 102, 161 Kellogg, 0. D. ro29 Potential theory. Berlin: Springor-Verlag. Kroner, B. 1954 data Met. 2, 301 Kynob, G. J. 1956 Proo. Roy. Soc. A, 237, 00. Landau, L. D. & Lifshitz, E, M. 1954° Mekhanika Splosinykh Sred, p. 686, Moscow: Giostekhizdat. Leibftied, G. 1954. Z- angow. Phys. 6, 251. Love, A. EH. 1927 Theory of elasticity. Cambridge University Press. Mackeneio, JK. 19§0 Proe. Phys. Soe. B, 63, 2. Nabarro, FR. N. 1940 Proc, Roy. Soc. A, 175, 519. Nabarro, F.R.N. 1931 Phil. Mag. (7), 42, 1224. Nabarro, F.R.N. 1952 Adeanc. Phys. 1, 260. Osborn, J. A. 1945 Phys. Fev. 67, 351. Poincaré, H. 1899 Théorie du Potentiel Newtonien, p. 118. Paris: Carré et Neud. Robinson, K. 95x J. Appl. Phys. 22, 1045. Routh, D. J. 1892 Analytical states, 2, 121. Cambridge University Press. Sack, R. A. 1946 Proc. Phys, Soo. 88, 720. Shapiro, G.8, 1947 Dokl. Akad, Nauk SSSR, 58, 1800. Somigliana, C. xor4 R.C. Accad. Lineoi, (5), 23 (1), 463. Somigliana, C. xox3 H.C. Aecad. Lineoi, (5), 24 (1), 656. Stornberg, H., Eubanks, R. A. & Sadowsky, M. A. rosx J. Appl Phys, 22, 1121 Taylor, G1. 1923. Proc. Roy. Soc. Ay 103, 58. Zener, C. x46 ‘Trans, Amer. Inst. Min. (Metall.) Bugre, 167, 513.