1. Surface interactions and the stress tensor.

Recall the balance of linear momentum

for Ω ⊂ R: Z Z Z
ρR v̇dV = tdA + ρR bdV, (1.1)
Ω ∂Ω Ω
where ρR is the referential mass density, v is the material velocity, t is the traction vector, and
b is the body force per unit mass. Given that a balance law of the form (1.1) exists, we now
show that the surface interaction vector t depends on the surface only through the unit normal
and, moreover, the dependence is linear. The first claim was introduced by Augustin-Louis
Cauchy in 1823 as a hypothesis (Cauchy, A. L., Bulletin de la Sociètè Philomatique, pp. 9-13
(1823). For an historical account see footnotes in Truesdell, C. & Toupin, R. A., The Classical
field Theories, Handbuch der Physik, Vol III/1, Springer, Berlin (1960), Sects. 200 & 203), but
was proved much later in 1957 by Walter Noll (Noll, W., The foundations of classical mechanics
in the light of recent advances in continuum mechanics, pp. 266-281, The Axiomatic Method,
with Special Reference to Geometry and Physics (Symposium at Berkeley, 1957), North-Holland
Publishing Co., Amsterdam (1959)). The second claim, which is also known as the Cauchy’s
theorem, is based on the classical tetrahedron argument first proposed by Cauchy and is now
recognized as a result of fundamental importance in continuum physics.

1.1. Cauchy’s hypothesis (Noll’s theorem) Let N be the outward unity normal to the posi-
tively oriented surface ∂Ω. Then

t(X, t; ∂Ω) = t(X, t; N), (1.2)

i.e. the dependence of the surface interaction vector on the surface on which it acts is only
s2
d1
N
T f1
s1 X̂
R
P1 e

Figure 1: Two surfaces with a common tangent plane

through the normal N. To prove this assertion let s1 and s2 be two surfaces in Ω such that
they have a common tangent plane (denoted by T ) at some X̂ ∈ s1 ∩ s2 . Let N be the common
unit normal to both surfaces at X̂. Let P1 be a bounded region such that ∂P1 = d1 ∪ f1 ∪ e,
where d1 is a subset of s1 , f1 is a piece of the lateral surface of the circular cylinder with axis
N and radius R, and e is a part of the surface of the cylinder which is common to both ∂P1
and ∂P2 (P2 is the region bounded on the top by s2 ). The quantities f2 and d2 are defined in a
way similar to f1 and d1 , respectively. Furthermore, ∂P2 = d2 ∪ f2 ∪ e. If we denote the surface

1
area of a surface s by A(s) and the volume of a region P by V (P ), then for a = 1, 2,

A(da ) = πR2 + o(R2 ),

A(fa ) = o(R2 ), (1.3)
V (Pa ) = o(R2 ).

Recall that the order notation ‘small o’ is defined such that limx→0 o(x)/x → 0, i.e. o(x) goes
to 0 faster than x, as the later tends towards 0. Similarly the order notation ‘big O’ is defined
such that limx→0 O(x) → 0, i.e. O(x) goes to 0 in the same way as x tends towards 0. We
now establish the above relations, Let u = u1 e1 + u2 e2 be the two-dimensional position on
T a measured from X̂, hence u = 0 at X̂. Let s stand for either s1 or s2 . The position of
the points on s may be written as y(u1 , u2 ) = u + z(u1 , u2 )N. Since u,α = eα , α ∈ {1, 2}
and z,α at X̂ is 0 (the slope of s is 0 at X̂), we have y,α |X̂ = eα . We are denoting a partial
derivative of a function f (u1 , u2 ) with respect to u1 (or u2 ) as f,1 (or f,2 ). Assuming the surface
to be smooth in the neighborhood of X̂, we can expand y(u1 , u2 ) as a Taylor series (about
(0, 0)): y(u1 , u2 ) = X̂ + y,1 |X̂ u1 + y,2 |X̂ u2 + o(r) = X̂ + u + o(r), where r = |u|. Consequently,
y,1 = e1 + O(r) and y,2 = e2 + O(r). Also, y,1 × y,2 = N + O(r). The infinitesimal area element
on s is da = |y,1 × y,2 |du1 du2 = (1 + O(r))du1 du2 = (1 + O(r))2πrdr. The area of d is therefore
RR
A(d) = 0 da = πR2 + o(R2 ), where d stands for either d1 or d2 . This proves the first result in
(1.3). The second result follows immediately since as R → 0, A(d) → πR2 and hence f becomes
an empty set implying A(f ) = o(R2 ). Finally the volume of P1 , or P2 , is O(R3 ) which is o(R2 ).
We now apply the balance law (1.1) to regions P1 and P2 . We obtain
Z Z
t(X, t; ∂P1 )dA = ρR (v̇ − b)dV,
∂P1 P1
Z Z
t(X, t; ∂P2 )dA = ρR (v̇ − b)dV.
∂P2 P2

Subtract these two relations to get

Z Z Z Z Z Z
tdA − tdA = ρR (v̇ − b)dV − ρR (v̇ − b)dV + tdA − tdA. (1.4)
d1 d2 P1 P2 f2 f1

Assume all the fields to be bounded over the domain of their integration. Then,
Z
ρR (v̇ − b)dV ≤ max |ρR (v̇ − b)|V (Pa ),
Pa X∈Pa
Z
tdA ≤ max |t|A(fa ).
fa X∈fa

Based on relations (1.3)2,3 , equation (1.4) can then be rewritten as

Z Z
t(X, t; d1 )dA = t(X, t; d2 )dA + o(R2 ). (1.5)
d1 d2

Divide equation (1.5) throughout by πR2 and use (1.3)1 . As a result obtain
1 1 o(R2 )
Z Z
t(X, t; d1 )dA = t(X, t; d2 )dA + . (1.6)
A(d1 ) d1 A(d1 ) d2 πR2

2
Since t(X) is assumed to be continuous, an application of the Mean-value theorem gives
1
Z
lim t(X, t; da )dA = t(X̂, t; da ), (1.7)
R→0 A(da ) da

where X̂ is the common point of d1 and d2 . Therefore letting R → 0 in (1.6) yields

t(X̂, t; d1 ) = t(X̂, t; d2 ). (1.8)

Thus, the surface interaction vector t takes the same value for all surfaces with a common unit
normal and therefore its dependence on the surface is only through the normal vector. The
assertion (1.2) is proved.

1.2. Cauchy’s lemma The balance law (1.1) implies that

t(X, t; −N) = −t(X, t; N). (1.9)

This result will be used in the proof of the Cauchy’s theorem. To verify this relation consider
a pillbox P of thickness , centered at X, and with its flat surfaces parallel to N. As we let
 → 0, the pillbox flattens to its middle surface S. The relation (1.1) for bounded fields then
reduces to Z
lim tdA = 0 (1.10)
R→0 ∂P

or Z
(t(N) + t(−N))dA = 0. (1.11)
S
Finally, shrink the disk S to the middle point X and use the continuity of t to obtain (1.9).

1.3. Cauchy’s theorem The surface interaction vector t depends linearly on N. Therefore,
there exists a tensor σ such that

t(X, t; N) = σ(X, t)N. (1.12)

We now prove this theorem. Consider a tetrahedron T ⊂ Ω with vertex X0 ∈ Ω. The surface
of the tetrahedron normal to the axis ei is denoted by si . Let δ be the distance along the
unit normal m from the vertex to the fourth surface s (see figure 2). Then, the volume of the
tetrahedron V (T ) and the surface area A(s) of the face s can be calculated as respectively, c1 δ 3
and c2 δ 2 , where {c1 , c2 } ∈ R+ are constants. The area of the remaining faces (given by A(si ))
can be obtained from A(s):
A(si ) = (m · ei )A(s). (1.13)
R
This relation can be verified by first noting, using the divergence theorem, that ∂T NdA = 0,
where ∂T is piecewise smooth. Since N is constant on each face of T , (1.13) follows.

3
 e2

s1 m
s3
δ
X0 e1

s2

e3

Figure 2: Tetrahedron T

We will now use the balance law (1.1) and the assumption of the continuity of the fields to
arrive at the relation (1.12). The balance law when restricted to the tetrahedron T implies
Z Z Z
tdA = ρR (v̇ − b)dV ≤ |ρR (v̇ − b)|dV ≤ kV (T ), (1.14)
∂T T T

where k = max|ρR (v̇ − b)| is finite. Therefore,

X∈T

3 Z
1 1 
Z Z X 
O(δ) = tdA = t(X; m)dA + t(X; −ei )dA
A(s) ∂T A(s) s i=1 si
3 Z
1 Z X 
= t(X; m)dA − t(X; ei )dA , (1.15)
A(s) s si
i=1

where the last equality is a consequence of the Cauchy’s lemma. By the Mean-value theorem,
for continuous t, we obtain
Z
t(X; m)dA = A(s)t(X̃; m),
Zs
t(X; ei )dA = A(si )t(X̃i ; ei ) (1.16)
si

for some X̃ ∈ s and X̃i ∈ si , respectively. Let δ → 0. Then X̃ → X0 and X̃i → X0 . As a result,
equations (1.13), (1.15) and (1.16) yield

t(X0 ; m) = (m · ei )t(X0 ; ei ), (1.17)

where summation over i is implicit. As the choice of the vertex X0 and the unit normal m is
arbitrary, the relation (1.17) holds for all X ∈ R and all unit vectors. Equation (1.17) shows
that t is linear in m. Therefore there exists a tensor σ, the stress tensor, such that

t(X, t; m) = σ(X, t)m (1.18)

for all X ∈ R and any unit vector m. The proof is complete.

Note that we have restricted our attention to only continuously differentiable fields defined on
domains with piecewise smooth boundaries. Much research has been done in the past fifty years

4
to investigate these results under less stringent smoothness requirements. Such considerations
are indeed necessary for many practical problems in mechanics such as those involving shocks,
fracture, dislocations and corner singularities (For a recent contribution, where the past work is
carefully reviewed, see Schuricht, F., A new mathematical foundation for contact interactions in
continuum physics, Archive of Rational Mechanics and Analysis, 184(3), pp. 495-551 (2007)).