hr. J. Solids Sfrucrures, 1965, Vol. 1, pp. 417 to 438. Pergamon Press Ltd.

Printed in Great Britain

SECOND GRADIENT OF STRAIN AND SURFACE-TENSION IN

LINEAR ELASTICITY

R. D. MINDLIN
Department of Civil Engineering, Columbia University, New York, N.Y.

Abstract-In this paper there is formulated a linear theory of deformation of an elastic solid in which the
potential energy-density is a function of the strain and its first and second gradients. This is a theory in which
cohesive force and surface-tension are intrinsic. A solution is given for the strain and surface-tension, or
surface-energy per unit area, resulting from separation of a solid along a plane; and a comparison is made
with an analogous lattice model. Also presented are a general solution of the displacement-equation of equili-
brium in terms of stress functions and the particular solution for the concentrated force. The special case of a
liquid is considered and the solutions are given for the surface-tensions at plane and spherical surfaces.

1. INTRODUCIlON
IF THE potential energydensity of an elastic continuum is assumed to depend on the
rotation-gradient, in addition to the strain, there results a theory of elasticity with couple-
stresses, i.e. couples per unit area, in addition to the usual stresses. In a paper on elastic
materials with couple-stresses [l], R. A. Toupin observed that the rotation-gradient com-
prises only eight of the eighteen components of the strain-gradient and he proceeded to
establish the mathematical theory in which all the components are accounted for. The
additional components of the strain-gradient are accompanied by ten components of self-
equilibrating double forces per unit area, i.e. ,stresses contributing neither resultant force
nor couple, per unit area, across a surface in the material but, nevertheless, contributing to
the potential energy and to the boundary conditions. In a subsequent paper, Toupin and
Gazis [2] exhibited the correspondence between strain-gradient theory and atomic
lattices with nearest neighbor and next nearest neighbor interactions; and they explored
the consequences of an initial, homogeneous, self-equilibrating stress. They showed that
a free surface will draw in or push out in a surface-layer ; but only in non-centrosymmetric
materials. Toupin pointed out, in a private conversation, that this restriction could be
removed by taking into account, further, the second gradient of the strain. A start in
this direction was made in 1959 by Hart [3] for liquids. Also, Green and Rivlin [4] have
recently established the basis of a very general theory which includes strain-gradients
of any order. It seems worthwhile, however, to formulate completely the equations of the
simplest theory, incorporating the second gradient of strain, and to exhibit some typical
solutions. The cohesive force and surface-tension, or surface-energy per unit area, which
appear, may have some bearing on criteria of failure of solids and on a variety of surface-
phenomena observed in both solids and liquids.
In what follows, equations of equilibrium, boundary conditions and constitutive
equations are derived for a linear theory of elastic materials in which the potential
energy-density is a 2nd degree polynomial in the infinitesimal strain and its first and
second gradients. The case of the homogeneous, isotropic, centrosymmetric material is
417
418 R. D. MINDLIN

explored in some detail. Terms appear which describe the property of cohesion and a
general formula is given for surface-tension : one-half the product of a modulus of cohesion
and the normal gradient of the dilatation at the surface. After a formulation of the general
displacement-equation of equilibrium, a solution is given for the problem of separation of
a solid along a plane, including formulas for the surface-energy per unit area and for the
residual strain-which decays exponentially into the interior. A decay constant of the
order of magnitude of interatomic distances is estimated from electron diffraction data
obtained by Germer, MacRae and Hartman [5]. The theory admits either monotonic or
oscillatory decay: a property which is shown to be also the case in a solution of the analo-
gous problem for a monatomic lattice with first, second and third neighbor interactions.
Following this, a general solution of the displacement-equation of equilibrium is given in
terms of stress functions. Then the particular solution is found for the concentrated force.
Finally, a special theory for a liquid is formulated by restricting the strain and its first two
gradients to the dilatation and its first two gradients. Solutions are given for the surface-
tensions at plane, and spherical, liquid surfaces. It is found that, as the curvature of the
surface increases (decreases) from zero, the surface-tension becomes algebraically smaller
(larger).

2. STRESS-EQUATION OF EQUILIBBIUM AND BOUNDARY CONDITIONS

In the linear theory to be formulated, the potential energy-density, cl! is assumed to be
a function of three polyadics:
w = W(Q ;, E3), (1)
where
:: = +(Vu+uV), LW, &WV. (2)
In (2), V is the gradient operator and u is the displacement. The symmetric dyadic k is
the classical infinitesimal strain with six independent components. The triadic g, sym-
metric in the first two positions, has eighteen independent components which could be
replaced [6] by the strain-gradient Vk, with eighteen independent components which are
linear combinations of those of 2, or by the rotation-gradient and a symmetric triadic
with, respectively, eight and ten independent components which are linear combinations
of those of g or of Vk Similarly, 2, symmetric in the first three positions, has thirty inde-
pendent components which could be replaced by the second gradient of strain W8 or
by a variety of combinations of other polyadics. Thus, the energy-density could be
expressed as any one of a large number of functions of fifty-four independent variables.
All of these energy functions lead to the same displacement-equation of equilibrium.
The function chosen in (1) produces simpler forms of the boundary conditions and con-
stitutive equations than most and as simple forms as any.
The variation of the total potential energy, in a volume r/; with arbitrary variation of
u, is
6J;WdV = ~~(::V6u+:iWGu+:::VWG)dl! (3)

where
 Second gradient of strain and surface-tension in linear elasticity 419

and the rule for multiple scalar multiplication of dyads, triads, etc. is
ab:cd = (a-c)(b*d), abcidef = (a-d)(b-e)(c-f), etc.

In the definitions (4) it is to be understood that i, +, z have the same symmetries as

& E, P, respectively.
By application of the chain rule of differentiation and the divergence theorem, the
right-hand side of (3) is converted into the sum of

- ~JV(:--V:+W:&Sudl/+ S,n*(;-V:+W::)GudS (5)

and

Jsn.(:-V:):V&dS+ {/r.:i W&d& (6)

where S is the boundary of V and n is the unit outward normal to S. Now, V&r is not
independent of 6u on S because, if 6u is known on S, so is the surface-gradient of 6u.
This is analogous to the situation in couple-stress theory [7] where, if the displacement
is known on S, so is the normal component of rotation-just as, in the classical theory of
flexure of plates, if the deflection at the edge is known, so is the component of rotation
about an axis normal to the edge. To prepare for the formulation of a variational principle,
we resolve V&I into a surface-gradient and a normal gradient:

V&I = VSu + nD6u ; c = (I-lm)~v, D=n*V, (7)

where I is the unit dyadic. The term, in the integrand, with ~SU as coefficient can then be
reduced to one with &I as coefficient by employing the chain rule and a surface integra-
tion-leaving only the independent variations 6u and D&I as coefficients. A similar
procedure, with some additional steps, can be used to convert the term with W&I as
coefficient to terms with the independent variations au, D&I and D26u as coefficients.
Thus, if @ is a dyadic and v is a vector,

Q:Vv= Q,:~v+n.rp.Dv=~.(~.v)-~.~).v+n-g.Dv.
.
By the surface divergence theorem [8] for a smooth, closed surface (which S is assumed
to be),

Hence,
{/CVvdS = Ss(L.O)*vdS+ Ssn*cP.DvdS, (8)
where
L = n6 n-6 (9)
Similarly, if Y is a triadic,
j,IiVcDdS = S,(L+):OdS+ ~~n*P:DQ,dS. (10)
420 R. D. MINDLIN

Thus, by (8) the first integral in (6) can be written as

j~L*[n*(:-V.:)]+dS+ j~nn:(:-V+DbudS (11)

and, by (lo), the second integral in (6) is

jsL.(n.:):VgndS+j~nn:::W6udS. (12)

Again by (8), the first integral in (12) can be written as

jsL - [L - (n - $)] - 6u dS + js n * [L - (n - $1. D&I dS. (13)

As for the second integral in (12), we suppose that S is a coordinate surface of an ortho-
gonal, curvilinear, coordinate system and interpret n as one of the orthogonal unit
vectors in that system. Then
DVv = n* Wv = V(n*Vv)--(Vn)*Vv.

But V(n - Vv) = VDv = 6Dv + nD2v and @n) - Vv = &n) - h. Hence :

DVv = ~Dv + nD2v - (\jn) * 6~.

Accordingly,

@ : DVv = 0 : [~Dv - (6) * cv + nD2v],

= ~.[e.Dv-(~~).cp.v]-(~.6).Dv-+~.[(~n).8].v+n-~.D2v,

= ~.[@*Dv-(~n)*@*v]-(V*Q).Dv-L*[&).9]*v+n.cD*D2v,

where the last step is a consequence of n *&I) = 0 and (9). Hence, using the surface
divergence theorem and noting again that n - (Vn) = 0, we can write the second integral
in (12) as

- sL*[(&)*(nn::)].6dS+ j sL+n::)D6udS+ j~nnnI&D6udS. (14)

j

Altogether, the right-hand side of (3) is the sum of (5), (ll), (13) and (14); so that (3)
becomes

SI WdV = - j[V.(:-V*:+W::)l.ijudV+ j n*(i-V. <+VV:$).&dS

V s

+ j~{L.[n.(:-V.:)+L.(n. :)-(k)*(nn:$])-*duds

nn:(~-V.~)+n.[L.(n.~)]+L.(M:~)}.D6udS
+ j&

+ js (MII i ?) * D26u dS.

 Second gradient of strain and surface-tension in linear elasticity 421

We now assume the following principle of stationary potential energy:

61 WdV= J;fGudV+ Ss(~.6u+Z.D6u+~.DzSu)dS, (16)

V

where f is the body force per unit volume and i, 1 t are generalized surface tractions.
From (15) and (16) there ensues the stress-equation of equilibrium

v+-vG+w S)+f = 0 (17)

and the traction boundary conditions

; = -(:-V~:+W::)+L~[n-(;-V*:)+L+l+(hl)+l::)], (184

i = nn:(~-V~:)+*[L+l~~)]+L*(nn:32), W)

&lnni:. (184
Thus, there are three scalar equations of equilibrium and only nine scalar boundary
conditions instead of what appeared at first, in (5) and (6), to be 3+9+ 18 = 30 boundary
conditions. This reduction is analogous to the reduction from three to two boundary
conditions in the classical theory of flexure of plates.
Other admissible sets of nine boundary conditions, in terms of u, Du and Du or
products of appropriate components of t and u or i and Du or t and Du, are apparent
from the form of (16) and the uniqueness theorem which would follow the steps from (1)
to (17) in reverse.

3. CONSTITUTIVE EQUATIONS
We shall consider, here, only homogeneous, centrosymmetric, isotropic materials.
Then, in the 2nd degree polynomial W, with constant coefficients, the variables 8, i, 3 can
appear in the linear terms only as scalars and in the quadratic terms only as products
that are scalars. This eliminates the terms linear in 2 and the products of i with i and 2.
Also, there are only two quadratic products of k, five of & seven of 2 and three of Q with
i that are scalars. Furthermore, we may omit the constant term and the term linear in i
because, just as in classical elasticity, the material configuration, to which W is referred,
can be chosen so that these terms do not appear. What remains, then, is
W = ~~E~~Ejj+~&~j&~jf~~&~j~~~~~+~~&~~~E~jj+~~E~~~Ejj~

+ a4EijkEijk+ aSEijkEkji + blEiij,zkkll + b2EijkkEijll

+ b3EiijkEjkll + b4EiijkEllkj+ b5siijkslljk + b6EijklEijkl

+ b7Eijkl&jkli + Cl&ii&jjkk + C2&ij&ijkk + c3&i$kkij + b&iijj, (19)

where sij, Eijk, &ijkl (i,j, k, 1 = 1,2,3) are the components of & E, g, respectively, in any
orthogonal coordinate system, and the summation convention for repeated indices is
employed.
The constants I and p are the usual Lame constants and the five a, are the additional
constants which appear in Toupins strain-gradient theory [ 1,6].
422 R. D. MINDLIN

The choice of the polynomial (19) implies that terms of higher degree are small in
comparison with those retained. This, in turn, implies that certain assumptions as to
order of magnitude have been made in addition to [Vul << 1 of the classical linear theory.
While k is dimensionless, $ has the dimension of reciprocal of length and d has the
dimension of reciprocal of the square of length. Thus, the ratios bJp have the dimension
of the fourth power of length and the ratios an/p, b,/p, c,,/p have the dimension of square
of length. Suppose that l2 is the ratio of one of the constants a,, to CL;i.e. 1 is a material
property with the dimension of length. Then we assume that a,,J,u12,b,,/@, cn/p12,b,/p12
are of order of magnitude unity and
[Vu/ -+ 1, lpvul 4 1, 12(wvuJ 4 1,

where 111 signifies the absolute values of the components of Y. With these assumptions,
all of the terms in (19) are of the same order of magnitude and terms of higher degree
are negligibly small.
From (19) and (4) follow the constitutive equations:

ZP9 = biSpq + 2PEpq+ CIEiij,dpq + C2Epqii+ tC3(Eiipq+ Eiiqp), (204

Tppr = a,(&,iia,, + EqiiSpr)+ ha,(&iipJqr + 2Eriisqp+ %t@pr)

+ 2Widpq + 2wpqr + a&,,,+ Erpq), Vb)

=pqrs = SblEiijjbpqrs + %2&jdjjkpqrs + bb3[(Eiijk + EiikjPjkpqrs+2&jsdjpqrl

+ !%hEiisjajpqr+ 3 bsEiij$jpqr + 2bbEpqrs+ &(Eqrsp + Erspq+ ~spqr)

+ SC1Edpqrs + k2Eijaijpqrs + h3Eisbipqr + @Obpqrs cw

where zpqyrpqry7pqrs are the components of :, I, :, respectively; 6ij is the Kronecker delta
and
dijkl = 6ijbkl + dik6jl + 8jksil, Bijklmn = 6ik6j16fn,+ 6i/$jm6h + 6i16jm6kW

The zpq are like the components of the usual stress with the dimensions of force per
unit area; but here they depend on the second gradient of strain in addition to the strain.
The rpqr have the character of double forces per unit area. The combinations &rpq,-rprq)
are the eight components of the deviator of the couple-stress, i.e. couples per unit area,
while the remaining ten independent combinations are self-equilibrating. The thirty rpqrs
have the dimensions of force and the character of triple forces per unit area-all self-
equilibrating.

4. SURFACETENSION
Of special interest, in the energydensity (19), is the term linear in 2 which produces
the homogeneous, self-equilibrating components of cohesive force, &b+Spqrs,in (2Oc). The
coefficient bO,with the dimensions of force, is a modulus ofcohesion. It gives rise to surface-
tension; or, equivalently, to the energy, per unit area, associated with the formation of a
new surface.
In classical hydrostatics, surface-tension is introduced by postulating the existence of
a vanishingly thin surface-membrane under uniform tension. If the liquid is under no
external forces, the surface-tension is equal to the total potential energy divided by the
 Second gradient of strain and surface-tension in linear elasticity 423

surface-area. In a solid, the surface-tension in a body under no external forces need not
be uniform; but we may adopt the same definition for the average surface-tension:

T,,, = WI-49 (21)

where A is the area of the surface and

or, from (19) and (20),

%- = tSY(::Vu+:iWu+.:::VWn)dV+~b,SVV.dV. (22)

The second integral in (22) enters because the cohesive force is constant whereas the
remaining parts of the stresses vary in proportion to the strain and its gradients.
Now, if there are no external forces, i.e. if &2,; and f are zero, the first integral in (22)
vanishes ; as can be shown by the procedure employed in Section 2. Hence,

K.,==$ V2VudV=$ n*W-udS.

sV ss
With this result, it seems appropriate to take
T= $b,n- W *uIs (23)
as the definition of the punctual surface-tension, or surface-energy per unit area, in a
centrosymmetric, isotropic, elastic solid; i.e. one-half the product of the modulus of
cohesion and the normal gradient of the dilatation at the surface.

5. DISPLACEMENT-EQUATION OF EQUILIBRIUM
Upon substituting the straindisplacement relations (2) into the stress-strain relations
(20) and the resulting expressions for the stresses into the stress-equation of equilibrium
(17), we find the displacement-equation of equilibrium
[~+2p-(ii-2qV~+m4]WU-[[Ct-(ii-c3)V2+6V4]VxVxu+f = 0, (24)
where
5 = 2(a1+a2+a,+u4+U~), 6 = 2(b,+b,+b,+b,+b,+b,+b,),
z = c,+c,+c,, ii = 2(u, +a,), 6 = 2(b, + bs).

Alternatively, (24) may be written in the form

(~+2p)D:,D:2W~u-pD;1D;2VXVXU+f = 0, (25)
where
D?.
11= 1 -l?.V2
IJ
i = 1,2; j = 1,2; (26)
and
2(~+2~)I:j = a-2~f[(a-22)2-46(~+2~)]~, j = 1,2; (27a)
2~1~3= a-C,+[(ii-C,)2-46~]*, j = 1,2. (27b)
424 R. D. MINDLIN

Thus, in general, solutions of the equation of equilibrium will contain Lames

constants, 1 and p, and four additional material constants, I,,, having the dimension of
length. Application of boundary conditions may introduce some or all of the remaining
material constants which appear in the energy-density (19).
The conditions for positive W do not include relations between 5 (or zi) and 5, ?
(or 6, c3) and, hence, supply no indication of the character, real or complex, of the lij
In what follows, the lij will be treated as if they were real and positive; but complex lij
are equally admissible.

6. STRAIN AND SURFACETENSION AT A PLANE SURFACE OF A SOLID

In a rectangular coordinate system x, y, z, consider the half-space x > 0 without
body force and with its plane surface traction-free:
~=t=&) on x = 0. (28)
Assume
u, = u,(x), u, = u__= 0. (29)
Then, with primes designating differentiations with respect to x, the stress-equation of
equilibrium (17) reduces to
T~,-T~,,+T~x::,, = 0, (30)
while the boundary conditions (18) become

(G,--Z:,x+T:,xx)x=o = (r,,x-r&),=0 = (rxxxx)x=o = 0. (31)

Also, with the assumed form of displacement (29), the constitutive equations (20) yield
Txx = (A + 2&s + FE, TX,, = Ze, r xxxx = b, + CE+6&I, (32)
where E = du.Jdx. Further, the displacement-equation of equilibrium (25) becomes

(33)

One integration of the stress-equation of equilibrium (30) makes T,,-T!_+T~,,,

equal to a constant which, by the first of the boundary conditions (31) must be zero.
This disposes of one of the three boundary conditions on x = 0. The remaining two,
expressed in terms of E through (32), are
[(ii - F)&- ti&lX
= 0 = 0, (E&+ti&),=O= -b,. (34)
The solution of (33), vanishing at infinity, is
uX = ~le-x/l+A,e-/. (35)
Inserting (35) into (34) and making use of the definitions (27a), we find the following
equations for determining the constants A, and A, :
A,1:,(1:.,+I~o)+A,I:,(l:2+z:o) = 0, (364

A1112(1:,+l:o)+Azlll(l:,+I:,) = w,,42/u-+&4 (36b)

where r:, = ?/(,?.+2~). From (36), the values of A1 and A, follow immediately.
 Second gradient of strain and surface-tension in linear elasticity 425

The formula (23), for the surface-tension yields

or

bi#:, -4,)
(37)
?-= 2(~+2~)~~ll(!:2+1:o)2-112(~:1+I:~)21
This is also, for each surface, the energy per unit area required to separate the body along
a plane.
As may be seen from (35), the displacement and strain decay exponentially with
distance from the surface. It is known that such an effect is confined to an extremely thin
surface-layer in physical materials; so that I,, and Irz are probably very small lengths.
An estimate of order of magnitude can be obtained from measurements of electron
diffraction at nickel surfaces by Germer, MacRae and Hartman 152 They conclude that the
displacement of the superficial layer of atoms toward the interior is five times as large as
that of the next layer. The assumption of a simple exponential decay, proportional to edX,
would lead to an I of about five eighths of the distance between adjacent planes of atoms.
This is a ratio of the same order of magnitude as the analogous one of about three eighths
of the diameter of a sphere, in a simple cubic array of contiguous elastic spheres, computed
for the case when only the rotation-gradient (leading to couple-stresses) is taken into
account, in the energy-density, in addition to the strain [9], If the 1, are complex, the
solution (35) would take the form of products of negative exponentials and trigonometric
functions. Then the decay of displacement and strain would be oscillatory and the
estimate of order of magnitude would apply to the real part.

7. LATIICE MODEL

Consider a one-dimensional lattice of interacting particles distributed along the x-axis

at points x = nh, where n is a positive or negative integer and h is a constant which is
taken equal to unity. Suppose that forces between particles are limited to first, second
and third neighbors and include self-equilibrating initial forces. Now separate the lattice
between particles n = 0 and n = - 1 and consider the set-ignite lattice n 3 0. The
separation is effected by the addition of forces PO, P1, Pz, on particles 0,1,2, resj?ectively,
equal and opposite to the resultants of the initial forces exerted on those particles by the
particles - 1, -2, - 3. Since the initial forces are self-equilibrating,
P,+P,+P, = 0. (38)
Now, suppose that the change of force between two particles is proportional to their
relative displacement ; with constants of propo~ionaIity al, a2, a3 for first,. second and
third neighbors, respectively. Then, if u, is the displacement of the nth particle, the
equilibrium of that particle is expressed by
i-3
,zl @i(%ki-2%+%-i) = 09 n 3 3, (39)
i=2

i~~aI(U2+i-2U2+U2-i)+Uj(us-U2)= P2, n = 2, Wa)

426 R. D. MINDLIN

a1(uz-2u1+ug)+a2(u3-u1)+a3(uq-u1) = P,, n = 1, (4Ob)

a& - uO)+ a& - uO)+ a& - u,) = P,, n = 0. (4Oc)

If we adopt the notation Au, = u, + I - u,, we have

A2u, = u,+ 1-2u,+ u,_ I,
A4u,= u,+~-~u,+~+~u,-~u~_~+u~-~,
A%, = ~,+~-6u,+~+15u,+~ -2Ou,+15u,_,-6u,_,+u,_,.
Then, with
fll = a,+4a2+9a,, B2 = -a,-6a3, (41)
equation (39) can be written as
(pl -j?2A2+a3A4)A2u, = 0; (42)
or
(1 -L;A2)(1 -1;A2)A2u, = 0, (43)
where
i = 1,2. (44)

The solution of (43), vanishing as n -+ co, is

u, = A+-I+A;~-/z , (45)
where
cash /,L;1 = 1 +)&, i = 1,2. (46)
As for the boundary conditions, again there are three for the two constants A; and
A;. However, following Gazis and Wallis [lo], we sum the three equations (40) and obtain
(1 -i:A2)(1 -;1;A2)Au2 = 0, (47)

which is satisfied identically by the solution (45). Thus, as in Section 4 the general
condition of equilibrium disposes of one of the three boundary conditions. This leaves
two conditions which may be any two, independent, linear combinations of the three
conditions (40) other than their sum. The resulting formulas for A; and A; are lengthy
expressions of little present interest.
More important is a comparison of the solutions (45), for the lattice, and (35) for
the continuum. Both are comprised of two negative exponentials. In the case of the
continuum, no conclusion could be reached regarding the character, real or complex,
of the decay constants. The same is true for the lattice.
The quadratic part of the potential energy of the nth particle, n > 2, is
W, = *fi1(AuJ2 + ~~2(A2u,,)2+ ~a,(A3uJ2 + pI;AunA3u,;

i.e. an energy leading to the equilibrium equation (42):

-2 = (p1-p2A2i- c(,A4)A2u, = 0,
n
 Second gradient of strain and surface-tension in linear elasticity 427

where & = #&-28;. Th e conditions for positive wn supply no relation between 8;

(corresponding to in of the continuum) and the remaining constants; and, hence, no
indication of the character of the Ai and /.Q
For real, positive force constants, ui, standing in the reiations

(for example, going inversely as high powers of the distances between interacting particles)
the roots Ai in (44) would be complex and then (46) would require complex tci. However,
real, positive a,, though appealing, are not necessary for stability.

8. STRESS FUNCTIONS
In this section, it is proved that any solution, u, of the displacement-equation of
equilibrium, (24) or (25), in a region V bounded by a surface S, can be expressed as
u = B-(lf,+f;,- ~~~~~2Vz)W~B-~V(D~~D~2-k~1)(r*D~~D~~B~~o), (48)
where k = (A+2,u)/p, r is the position vector and
,uD;,D:J*B, = r D~,D~2f-4(1~,+1~2-~1~,1f2V2)V.f, (49)
pDZ,,D:2V2B = -f. (50)
First, by the usual proof of the Helmholtz resolution, functions cp and H can be
defined, in terms of u, so that
u = Q+VxH, V*H=O. (51)
Upon substituting (51) in (25) we find
~V*~kD~~D~*V~+D~~D~*VxH~+f = 0. (52)
Define
4aB = j-, ~(r,)[kD:,D_tzVV,+DZziD~,V xHIQdVQ, (53)

where
I&,) = r; 1(e-r1121-e-1122)/(1~1 -I;,), (54)
ri is the distance between a field point P(x, y, z) and a source point Q(<, tf, r) and
d$ = d< dq d[. Then, by a process similar to that for Poissons equation [ll),
D;,D;,B = kD:,D:,V~+D;1D;2V~H (55)
and, from (55) and (52),
~D$,D;,V2B+f = 0. (56)

The divergence of (55) produces

D;1D;2V*B = kD: 1D:J2q. (57)
Define

(58)
428 R. D. MINDLIN

and find
pD:,Df,VB, = rD:,D:2f-4(1:,+1:2-~1:,1:,V2)V.f (59)
by using (56) and (57). Now, define

B =.VxH-B+(I;,+l;, -l;11;2V2)W~B+kD:,D:2Vrp (60)

and note that, by (55) and (57),

D;1D;2B = 0, V*B=O. (61)
Then substitute (60) and (58) into (51) to get
u = B+B-(1;r+l:2- l;,1;2V2)W+-3V(D:1D:2-k-1)(r*D;1D;2B+B,,). (62)
Finally, define
B = B+B. (63)
In view of (61), we may write (62) and (56) in the form of (48) and (50); and (59) is
already in the form of (49). Thus, (48) is a complete solution of the equation of equilibrium,
(24) or (25), if the stress functions B, and B satisfy (49) and (50).

9. CONCENTRATED FORCE
In an infinite region K let the body force be zero outside a finite region V, which
contains the origin and a non-vanishing field of undirectional forces f per unit volume.
A concentrated force is defined by
F = $imo [f,dV,. (64)
o-
To find B. and B for the concentrated force, we have to solve equations of the type
(49) and (50), viz. :
D:iDizV2$ = p, i = 1,2 (not summed). (65)
This may be done by constructing the pertinent Greens formula from Greens identity
(66)
5s n * ((PiVIL - $V(P& dS, = J, (qiV211/- +VrPJp d v,
in the usual way, as follows. In (66) replace pi and II/, successively, by
(l -1flv2)(1-lf2V2)((piv ICI),
1i11i2[1 -(Cl +1~L?)v21V2((pi~
$1,
[(li: + C2Y -21?11iz,l*v2(cPi~
$1,
[lfllfz- l~~lf2(lf~
+ l~2)]V4((pi9$)
and subtract the sum of the last three from the first of the resulting four equations.
Now choose
cpi(r,) = _L i = 1,2
71
 Second gradient of strain and surface-tension in linear elasticity 429

and note that

except at P. Exclude the point P by surrounding it with a spherical surface of radius E

and center P. Upon passing to the limit as E + 0, we find the Greens formula

4nJ = I, n it, Aj[~jViV~$- ~jbV~fP&dS,- I, PQ(Pi dVQ3 (67)

where
Zr = Diz,D;,, _Y2 = [l -(If, +li2,)V2]VZ, .Ys = v2, 94 = v4,
Al = 1, A, = -1s11l?123 A3 = 21i2,1~2-(li2,+1i2,)2, A, = l;r li,[(l;, + lh) - I$ I;,].

If B0 and B vanish at infkity at least as r - , the solution for the concentrated force
is, from (49), (50), (64) and (67),
4nj~B, = - lim j [r* D:,D:2f-4(1:i+1:2-~I:,I:2V2)V~flQcpl(tl)d~Q, (68)
Yo+O Y
4npB = yo_o
lim y fQ~2h)dvQ9 (69)
I

where r = (< + q2 + c2)*. Now 9

lim rr = r, lim r = 0.
vo+o vo+o

Hence, the solution for the concentrated force is

27r~B, = F * V(l:, + lf2- l~11~2V2)~,(r),
4apB = Fq,(r).

The limit for B follows directly from (69) while that for B. is obtained from (68) after
successive applications of the chain rule and the divergence theorem to separate f as a
factor, of the integrand of the volume integral, before passing to the limit. Finally, it is
apparent that B, and B have the required behavior at infmity.

10. ELASTIC LIQUID

The linear theory of deformation of an elastic liquid can be treated as a special case
of a solid in which the potential energy-density is a function of only the infinitesimal
dilatation and its gradients instead of the full infinitesimal strain and its gradients. Thus,

w, = W,(A, VA, WA), (70)

where A = V* II is the infinitesimal dilatation. For the centrosymmetric, isotropic liquid,
the form of W, is obtained from (19) by discarding all but the terms indicated in (70),
leaving
W, = #~A2+aIVA~VA+bI(V2A)2+b2WA:WA+c1AV2A+boV2A. (71)
This contains the term b2WA : WA which Hart omitted in a previous study [3].
430 R. D. MINDLIN

As noted by Hart, either the dilatation or the displacement may be subject to variation
in varying the potential energy:
6jy W,dV= - jV@A+pV6A+rr::W6A)dV, (72)

or
S j, W,dV = - jV(pV.Gu+~.W.du+n:WV.6u)dl: (73)

where, in either case,

p = -2 = -IA-c,V2A 3 (74a)

p= _aw,=
avL\.
-2a,VA,

aw,
- = -(b,+2b,V2A+c1A)I-2b,WA.
lr= -awA

We consider, first, 6A as the independent variation. Then, by applying the chain rule
and the divergence theorem, we transform the right-hand side of (72) to
- jy@-Vq,t+W:r)SAdV- jsn.(p-Brr)SAdS- jsn*xVSAdS.

Hart did not observe that V6A is not independent of 6A on S; but, by the same procedure
as in Section 2, we find
/,n-rVsAdS = j,[L*(n*n)]GAdS+ j nn:nDGAdS,
s
thereby converting to independent variations 6A and D6A. Hence
6j W,dV = - j,(p-V*p+W:n)GAdV- j n*(p-V-@AdS

- js [L - (n * n)]dl dS - js M : aD6A dS. (75)

The principle of stationary potential energy now assumed is

dj WA dV = j (p6A dV+ j &?A+ f,Dc?A) dS, (76)

V V S

where cp is the potential of a body force per unit volume and !&.,iA are generalized surface-
tractions. Equating the right-hand sides of (75) and (76), we find the stress-equation of
equilibrium
p-v.p+w:lc+q? = 0 (77)
and the traction boundary conditions

tla = -n*(p-V*rc)-L*(n.R), & = --::a. (78)

The equation of equilibrium on A is obtained by substituting (74) into (77), with the
result :
AA--_(a,-c,)V2A+2(b,+b,)V4A = cp, (79)
 Second gradient of strain and surface-tension in linear elasticity 431

or
,I(1 - I;V2)(1 - I;V2)A = cp, (80)
where
n1: = a,-c,+[(a,-c,)2-2~(b1+bz)]~, i = 1,2. (81)

We now turn to (73), in which &I is the variation, and find, by application of the chain
rule and the divergence theorem,

~jvW~dV=jv[V@-V~p+W:n)]~budl/-js(p-V~p+W:n)uv%rdS

- s[n.(p-V*x)]V*6udS-j n*rc*W*GudS. (82)

s s

Here, again, Hart did not notice the lack of independence of &I, V - &I and W - &I on S :
but, by the same procedure as in Section 2, we find

S{ W,dV= f [V(p-V*p+W:rc)]*&rdV- j (p-V*p+W:rc)n*6udS

V V

- s(L[n*(p-V*rt)+Ls(n-n)]-L~[(~n)M:rc]J-JudS
s
- s{[n*(p-V*n)+L*(n-n)]n+L(M:a)j*D6udS
5
- s (M : n)n *D26udS. (83)
J

In this case, the principle of stationary potential energy is assumed to be:

61 W,dV= J f*6udl+ j$v5u+&DSu+:*D2Gu)dS, (84)

V V

so that the stress-equation of equilibrium is

V(p-V~p+W:lc)-f = 0 (85)
and the traction boundary conditions are

i = -n(p-V.p+W:n)-L[n.(p-V*n)+L*(n*n)]+L*[@n)M:u], (86a)

z = -M*(p-V~u)-nL~(n~ff)-L(M:rc), (86b)

;= -_HRII:u. (86c)

The displacement-equation of equilibrium, obtained by substituting (74) in (85) and

employing (81), is
I(l-l:V~)(I-I:V2)W~u+f = 0. (87)
The formula for the surface-tension is obtained as in Section 4:
T = +b,n - VA(, (88)
and is the same for variation of A as for variation of u.
432 R. D. MINDLIN

11. DILATATION AND SURFACETENSION AT A PLANE SURFACE OF A

LIQUID
The problem of the liquid half-space x P 0, with traction-free boundary and no body
force, is formulated, first, in terms of the equations derived from variation of A. The
solution of (80), vanishing as x + co, with A a function of x only and cp = 0, is
A = A,e-+A,e-z; (89)
and the two boundary conditions (78) become
[(2a,-c,)A-2(b,+b,)A],=, = 0, [~,A+2(b,+b~)A]~=~ = -b,, (90)
where primes signify differentiations with respect to x. Substituting (89) in (90), we have
a pair of equations for A, and AZ-the solution of which, after the use of (81), is

where 1: = cl/L. Then (88) and (89) yield the formula for the surface-tension at a plane,
liquid surface :

(91)

In the alternate case of the formulation derived from varying u, the appropriate
solution of (87) may be written in the form
u, = Ao-llA;e-Xl- 12A;e-X2, (92)
There are now the three boundary conditions (86); but they lead to
A0 = 0, A; = AI, A; =A2
so that the two solutions are the same.

12. DILATATION AND SURFACETENSION AT A SPHERICAL SURFACE OF A

LIQUID
We consider a liquid sphere under no body force and with a traction-free surface of
radius ro. The appropriate solution of the equation of equilibrium (80), regular at the
origin t = 0, can be written in the form
A = A,l,r- sinh(r/l,)+A2/2r-1 sinh(r/l,). (93)
Then the boundary conditions (78) yield
i=2
iTI A;rJ2(bl+ b,) -(2a, - cJf ](r, cash ri - sinh rJ = 0, (94a)
--
i=2
c Ai{rJ2b2(1+2r;)+2bI+c,lf]sinhr,-4b2coshri} = -b,r& (94h)
i=l
 Second gradient of strain and surface-tension in linear elasticity 433

where ri = r&. Again, it may be verified that the same result is obtained from the
equations derived by varying the displacement.
To find a simple result for the tist order effect of the curvature of the surface, assume
that the radius of curvature is large in comparison with the Ii (i.e. ri is>1). Then the
solution of (94) is

2bJ r,,(l$ + li)e-

A
= - 11(122+I~)Z-/2(1:+I~)2+4b2~-1r01(1:-I~)

and the surface tension is, to the first order in li/ro,

T,
(95)
T x 1 + (8b2To/b$,)

where To is the surface-tension at a plane surface as given by (91). The condition of

positive definiteness of the quadratic part of the potential energy-density (71) requires
b2 > 0. Hence, as the radius of the surface diminishes, the surface-tension becomes
algebraically smaller.
A result of the classical theory of surface-tension is that the internal pressure in the
sphere is 2T/r,. In the present type of theory, a vatiety oflquantities could be assigned the
role of internal pressure. We could choose p or t B; but t, could take a variety of forms
for various choices of independent variables and independent variations. Internal pressure,
along with stress, appears to have lost its former significance owing to the consideration
of long range forces. However, dilatation is defined uniquely and can serve as a basis of
comparison of the two theories. In the classical case, the dilatation in the sphere would
be uniform and proportional to T/r,. In the present theory, the dilatation is non-uniform ;
but we may calculate the average dilatation:

3
A,,, = -47$ Adv
sy

= -$1$
A&(cosh ri - r; sinh ri).

Assuming, again, that ri % 1, we find, to the first order in li/ro,

6c, T
Aae x ---7
bol r.

where T is given by (95). Thus, as in the classical theory, the average dilatation in the
sphere is directly proportional to the surface-tension and inversely proportional to the
radius.
Now consider a spherical cavity of radius ro. For this we take the solution of (80),
with cp = 0, in the form
A = ~l~l~~1~~r~~+~2~2~~~~~~. (96)
434 R. D. MINDLIN

The conditions (78), for a free boundary, require

i=2

izI Ai[2(b, +bz)-(2a, -C,)l?](l +ri)lie-* = 0, (974

i=2

i~~Airi[2b,(l+2r,1+2r~2)+2b,+c,~~]e- = -b&. (9%)

Again we consider the case ri $ 1 and find, to the first order in li/rO,

T,
(98)
T = 1 - (8b,T,/b&,)

Thus, for both convex and concave spherical surfaces, the surface-tension, to the
first order in Zi/yO,is

To
T= (99)
1 + 8b,b,rcT,

where K is the curvature: positive for the sphere and negative for the cavity.

Acknowledgements-The idea that the extension of R. A. Toupins strain-gradient theory, to include the second
gradient of strain, would bring in the cohesive force and its effects in centrosymmetric materials was suggested
to me several years ago by Dr. Toupin. I am indebted to him for many valuable discussions of the associated
mathematical theory.
This investigation was supported by the Office of Naval Research under a contract with Columbia University.

REFERRNCES

[l] R. A. TOUPIN,Arch. ration. Mech. Analysis 11, 385 (1962).

[2] R. A. TOUP~Nand D. C. GAZIS,Proceedings ofthe International Conference on Lattice Dynamics, Copenhagen.
August, 1963 (Edited by R. F. WALLIS)pp. 597602. Pergamon-Press (1964).
[3] E. W. HART,.Phys. Rev. 113, 412 (1959); 114, 27 (1959); J. Chem. Phys. 39, 3075 (1963).
[4] A. E. GREEN and R. S. RIVLIN.Arch. ration. Mech. Analysis 16, 325 (1964).
[5] L. H. GERMER,A. U. MACRAEand C. D. HARTMAN,J. appl. Phvs. 32, 2432 ( 1961).
[6] R. D. MINDLIN.Arch. ration. Mech. Analysis 16, 51 (1964).
[7] R. D. MINDLINand H. F. TIERSTEN,Arch. ration. Mech. Analysis 11, 415 (1962).
[8] L. BRAND, Vector and Tensor Analysis, p. 222. John Wiley & Sons (1947).
[9] R. D. MINDLIN,Expf Mech. 3, 1 (1963).
[lo] D. C. GAZIS and R. F. WALLIS,Surface Science 3, 19 (1964).
[I I] H. and B. S. JEFFREYS,Methods of Mathematical Physics, 2nd edition, p. 210. Cambridge University Press
(1950).
 Second gradient of strain and surface-tension in linear elasticity 435

APPENDIX
Boundary conditions at edges and corners

In Sections 2 and 10, it was assumed that the surface S is smooth. Additional boundary
conditions are required if S has edges and corners.
Suppose the surface has an edge, C, formed by the intersection of two segments,
S, and Sz, of S. Then, for each segment, the surface divergence theorem [8] is

5s,~.fdS=~s,(~.n~.fdS+~cmi.fids, i = 1,2, (not summed)

where s is measured along C in the direction of its unit tangent si; and mi (= si x ni) is
the unit outward normal to C tangent to &. The contributions of the line integrals have
to be added to the previous results. Thus, to (8) and (lo), we have to add

[m*cD*v] ds, (8A)

fc

and

c[m*\Y :@I ds, (lOA)

f

where the boldfaced brackets Cl denote the value of the enclosed quantity on C when
approached over S1 minus its value when approached over S,; and ds is positive in the
direction of sl.
Considering, first, the case of the solid, we must add to (ll), (12) and (13):

c[nm:(~G*:)]*6uds. (1lA)
f

rnn:;:V&r]ds, (12A*)
f c[
[m*[L*(n*@*&rds, (13A)
Ic
respectively, while to (14) we have to add

- Qc[m.($n)*(nn:$]*&rds (14Al)

and
mnn i ; * IMu] ds. (14A*)
$c[

The integrands in (12A*) and (14A*) require resolution into terms with coefficients that
are independent variations. For (14A*) we write

$-c (nnm i ;)Jl.,6u ds - $c (nmn i ;),D,,~u ds, (14A2)

where D,,@I and (MIII i G)i are the limits at C of n * V&I and nnm i 3 approached over
Si. AS for (12A*), we first write
V&r = nn~VGu+mm~V&+ss~V6u = nD,6u+mD,,$u+sD,bu.
436 R. D. MINDLJN

Then (12A*) is the sum of

~c(nmni:),.D.,6uds-rPc(Nnni:)~.D,suds

and
$c [MI : 4 : sD,Gu] ds = $c [D,(mnsi 2* &I)] ds - $c [D&IS i $1 6u ds
l

= [Jams i :)I .&I - @Ds(nms i $1 &I ds,

l (12A2)

where the quantity [[nms i iI+] is the difference between discontinuities in (MIS i 2), and
(MIS i 3), along C or at a corner of C.
Altogether, the additional terms are (llA), (12Al), (12A2), (13A), (14Al) and (14A2),
so that we have to add to (15) and (16):

c[ma:(:-V.:)+m*[L.(o.~)]-D,(nmsi:)-m.~~).(on:2)].Guds
+

+ $c ~(MIII i a), * II,, 6u ds - $c 2(nmn i ;), * D,,&I ds

+ $c (mnm i ?), * D, ,&I ds - & @MI i i), * D,,6u ds

+[(mnsI:)~]~6 WA)
and

f F.&ids+ $c(N,.D.,Gu+N, l D,,&+T, l D,6u+T,.D,,6u)ds+G.6u, (16A)

C

respectively. Hence, equating (15A) and (16A), we have the edge and corner conditions

F = [um:(:-V~~)+~~[L+P~)]-D&~IS~~)-~~(~~I)+~:~)],

N, = 2(nmn i $,

N, = -2(nmn i :),,

T, = (MUUi*),,

T, = -(M&&,

G = [jmns i II 1.

As in the case of the boundary conditions on a smooth surface, alternative edge and
corner conditions involving displacements are apparent from the form of (16A).
 Second gradient of strain and surface-tension in linear elasticity 437

Turning, now, to the boundary conditions for the liquid, we find that, when it is the
dilatation that is varied, we have only to add to (75) and (76):

- fd[m:n6Ads (75A)

and

$C :,SA ds, (76A)

respectively; so that the only boundary condition to be added to (78) is

!A = -[llm:Ir]. (78A)
On the other hand, when it is the displacement that is varied, the additional terms are
similar to those for the solid. To (83) and (84) we have to add
- c[n.@-V.Ic)m+mL*(n~n)-D,(mn:rrs)-na:nm.(~n~.8ods
f

- ~c(nn:rcm+mn:xn),.D,,6uds+ ~~(nn:rcrn+mn:rcn),~D,,Guds

-~c(mn:nm),.D,~6uds+~c(ma:rrm),.D,,dods-[lmn:rcsl].6u (83A)

and
f F*&rds+ f (N,*D,,Gu+N,* D&r + T, l D, ,&I + T2 * D,,h) ds + G 6u,
l (84A)
C C

respectively. Then, upon equating (83A) and (84A) we find the edge and corner conditions
for the liquid:
F = -[n.(p-V.x)m+mL.(o.rc)-~~mn;ns)-nn:ma.(~n)],
N, = -(nn:rcrn+mn:rcn),,
N,= (mi:nm+mn:nn),,
T, = -(mn:rcm),,
T, = (nm : rem),,
G = - [~nm:~rs~+].
Again, alternative edge and corner conditions involving displacements are apparent
from the form of (84A).

(Received 29 January 1965)

Zaaamm&aseuag-In diesem Aufsatz wird die lineare Deformationstheorie eines elastischen Festkiirpers
formuliert, bei welchem die potentielle Energiedichte eine Funktion der Beanspruchung und deren Gradienten
erster und zweiter Ordnung ist. In diese Theorie sind die Kohlsionskraft und Oberfiilchenbeanspruchung als
wesentliche Merkmale eingebaut. Eine Lijsung fur die Beanspruchung und Obcrl%chenspanmmg pro FUchen-
einheit, wie sie bei der in einer Ebene liegenden Trennung eines Festkorpers auftreten, wird angegeben, und
ein Vergleich mit dem analogen Gittermodell angestellt. Ausserdem wird die allgemeine Liisuag der Gleich-
gewichtsverschiebegleichung angegeben, und zwar drtickt man diese in Spannungsfunktionen und mit Hilfe
der besonderen L&sung fti die konzentrierte Kraft aus. Der Spezialfall einer Fltissigkeit wird envogen, und
die Liisungen fur die Oberfliichenspannung bei ebenen und runden ObertXchen werden angegeben.
438 R. D. MINDIAN

&iCT~---B HaCfORlQei@ CTaTbe ~OpMynUpye~~ JIHHekHaR TeOpHR ~~pM~U~ yIlpyrOr0 T5epnOrO

Tena, B ~0T0p0~ noTeHqHanbHan 3fieprw-nnorriocTb nB.fweTcn +yHKusfe# ne&phlaunw w ee nepebhi M
BTOpblM FpaZUfeHTOM. B X-08 TeOpW CHJla CUetVleHHIl H FIOBepXHOCTHOe HaTIIxeHHe IIBRRlOTCflBHy-
T~HH~MHC~naM~.~aHO ~~eH~e~nKn~pM~~K HnOBepXXOCTHOrOHaTII~eHHII,anHnOBepXHOcTHOZi
3HeprnH Ha eaHHwy MOWaIvi, nonyseHHoR B pa3neneeea rBepnor0 Tena no nnocKocTH; cnenaH0
cpaefieiwe c aHa.norwwol peruewaToR hsonenbio. TaKxe ffpennaraeTC!S o6tuee peweH&+e ypasiieww
cMe~eH~~ npe parfsfoBeCHn B ycnoeitrtx~ynK4~~ ~~p~~e~~~ H w&e ~~e~neR.7~ ~oc~~oToqeHH0~
CWJIbI.