CHAPTER 3

CALCULUS OF TENSORS

3.1
INTRODUCTION

This chapter deals with a brief study of differential and integral calculus of
vector and tensor functions. It is assumed that the reader is already familiar
with calculus of several variables, particularly vector calculus.

3.2
SCALAR, VECTOR AND TENSOR FUNCTIONS
By a vector or a tensor function, we mean a vector or a tensor (respectively)
whose components are real-valued functions of one or more real variables.
Thus, if the components at of a vector a are real-valued functions of a
(single) real variable /, we say that a is a vector function of /, denoted a(0-
Similarly, if the components atj of a tensor A are real-valued functions of
t, we say that A is a tensor function of t, denoted A(t). Also, if a scalar φ
is a real-valued function of /, it is denoted as usual by </>(/). Generally, t is
assumed to vary over a specified interval on the real line.
109
110 3 CALCULUS OF TENSORS

In studying scalar, vector and tensor functions, we assume that the com-
ponents being dealt with are defined with respect to fixed sets of coordinate
axes and that the components possess derivatives of any desired order.
Thus, if 0,(0 are components of a vector function a(/), we assume that
the nth derivatives [dn/dtn\[ai{t)] exist for any desired n. It is easy to see
that [dn/dtn][ai(t)} are also components of a vector. Because, if we set
Ci = [dn/dtn]{ai(t)) in all coordinate systems, we find that

ci = -^niaÎ(t)} = —niaikak(t)l

using (2.3.5a);
dn
= OLik — {ak{t)} = OLikCk

The vector whose components are [dn/dtn]{ai(t)) is called the nth derivative
of a, denoted dna/dtn. In particular, the first derivative da/dt is the vector
whose components are da^/di; that is,
da
(3.2.1)
dt Ï-Ï«*·
Obviously, if a is a constant vector, then its derivatives of all orders are
equal to the zero vector.
Similarly, if a^(t) are components of a tensor function A(0, we assume
that the nth derivatives [dn/dtn]{ajj(t)) exist for any desired n. It may be
verified that these derivatives are components of a tensor; this tensor is
called the nth derivative of A, denoted dnA/dtn. In particular, the first
derivative dA/dt is a tensor whose components are [ά/άί\[α^{ί)}\ that is,
dA
dt ^KWI^HAM (3.2.2)

Obviously, if A is a constant tensor, then its derivatives of all orders are

equal to the zero tensor.
The following identities follow from the usual rules of differentiation:
d , ^ da db
— (a ± b) = — ± —- (3.2.3)
dt dt dt
d / v άφ da
(3.2.4)
dt dt dt
d / ΚΛ db da ^
— (a · b) = a · — + — b (3.2.5)
dt dt dt
 3.2 SCALAR, VECTOR AND TENSOR FUNCTIONS 111

(3.2.6)

(3.2.7)

(3.2.8)

(3.2.9)

(3.2.10)

(3.2.11)

(3.2.12)

(3.2.13)

where 0 is a real-valued (scalar) function of t, a and b are vector functions

of t, and A and B are tensor functions of t.
Next, let us consider scalar, vector and tensor functions of more than one
real variable. The independent real variables of our main interest in this text
are the three coordinates xt of a point x. Generally, point x is assumed to
vary over some region of three-dimensional space. A vector a whose com-
ponents ai are real-valued functions of xt is called a vector point function
or a vector field, denoted aC*,·) or a(x). Similarly, a tensor A whose com-
ponents dy are real-valued functions of xt is called a tensor point function
or a tensor field, denoted A(Xj) or A(x). Also, a scalar φ that is a real-valued
function of x{ is called a scalar field, denoted φ(χί) or φ(χ).
Scalar s, vectors and tensors arising in physical situations are generally
functions of both ΛΓ, and /, where / is the time variable. Such functions are
analyzed by the combined use of properties of scalar, vector and tensor
functions of / and those of scalar, vector and tensor fields.

EXAMPLE 3.2.1 Prove the identities (3.2.11) and (3.2.13).

112 3 CALCULUS OF TENSORS

Solution Let au(t) and bu(t) be components of A and B, respectively.

Then

I(AB) = jt\[PM\u} = jt(aikbkj)

da,k b db,kj
dt kj + Oik d t

[rfAj Γ. dB]
~^B + A— dt
[dt J y I \
This proves the identity (3.2.11). Also,
d = {[A I= {[AI
dt I ^ 1 ^
d\ (dA>T
dt \dt
This proves the identity (3.2.13).

EXAMPLE 3.2.2 If Q(r) is an orthogonal tensor, show that (dQ/dt)QT

is a skew tensor.
Solution Since Q is an orthogonal tensor, we have QQ r = I. Differen-
tiating both sides, we get
dQ
dt "+<%)-
so that

Hence
@y-«m-it>
showing that the tensor (dQ/dt)QT is skew.

EXAMPLE 3-2.3 If w(/) is the dual vector of a skew tensor W(0, show
that dvi/dt is the dual vector of dW/dt.
 3.3 COMMA NOTATION 113

Solution The fact that dW/dt is skew when W is skew is obvious from
(3.2.13). Let a be any vector (constant or function of t). Then

d
i , idA

because w is the dual vector of W;

Hence dyt/dt is the dual vector of dVf/dt.

3.3
C O M M A NOTATION

Consider a real-valued function/ = /(JC,·) that is a scalar, a component of a

vector or a component of a tensor. As stated in Section 3.2, we assume that
/ has partial derivatives of all desired orders (w.r.t. xt). We employ the
following comma notation:

dXi dx2 dx3

or, briefly,

/.i = | £ ; / = 1,2,3 (3.3.1)

Thus, for a specified value of i( = 1, 2, 3) the symbol ( )> f denotes the partial
differential operator (θ/θχ,Χ ).
We will employ all the rules and conventions of the suffix notation to this
comma notation—the comma not interfering with these rules and conven-
tions. In particular, the summation convention will be employed across the
comma.
Accordingly, the expression / = 1,2,3 can be suppressed in (3.3.1).
Further, an expression like (ftifti) will stand for Σ?=ι(///ΐ)· Thus, for
example, the differential df of / , which is given by
df df df
df=/-dxl+/-dx2 + 7+-dx3
3χγ dx2 dx3
114 3 CALCULUS OF TENSORS

may be written in a compact form as follows:

df = lkdxk (3.3.2)
The comma notation can be extended to second-order partial derivatives
also. We employ the following notations:
d fdf\ _ d2f
dx2\dxi/ dx2dxx'

and so forth. In general,

dxi \dXjJ dXi dxj

Since the functions dealt with are assumed to be differentiate up to the

desired order, the order of partial differentiation is taken to be inter-
changeable. That is,

/ ^f a2/ = a2/ (3.3.3)

,U Jl
dXidXj dXjdXi
Also,
a2/ a2/ a2/
fji - f,ll + f,22 + f,33 - T~2 ++ T~2+ + "^2
dx\ bx\ bx\
2 2 2 2
The sum (d f/dx + d f/dx\ + d f/dxj) is called the Laplacian of/, usually
denoted V2/. (The symbol V is pronounced as del or nabla). Thus, we have
f.u = V2/ (3.3.4)
the symbol ( ) u denoting the Laplacian operator V2( ).
The comma notation for third and higher order partial derivatives is
analogous. Thus,
f _ay a3/ _ a4/
7,111 - ~ 3 > 7,123 -T—T—^7 » 7,1123-
a^3' - 7 ' 1 2 3 fljC3 3JC2 fljC! ' */,U23 dX^X2dx\

and so forth.

EXAMPLE 3.3.1 Show that

= =
£imnJ,mn ^imnJ,nm ^ (J.J.D)
 3.3 COMMA NOTATION 115

Solution The result is obvious if we note that

timnf.mn = ^inmf,nm (interchange of dummies)
e
= ~ imnf,nm (definition of the ε symbol)
= -£imnftmn (changing the order of differentiation) ■

EXAMPLE 3-3.2 Prove the following:

(0 xu = ôu\ *,·,,· = 3
(ϋ) (xmXn)j = àimxn + ôinxm
OU) (XmXn\ü = Ôimôjn + SinÔjm

(IV) ν 2 ( Χ Λ ) = Mmn (3.3.6)

(v) rf. = (l/r)x if r*0

2
(vi) ir \u = Mu
(vii) V2(A·2) = 6

Here r2 = jcfjcf-. (7Ά/5 notation is used throughout.)

Solution (i) We note that

1,1
dxx
Similarly, x2t2 = *3,3 = 1. Also,

Similarly, Jt1>3 = x 3tl = Jt2>3 = x3>2 = *2,i = 0. Thus,

if / and j take the same value
if i and j take different values

Consequently, x/fl· = δΗ = 3.
(ii) By the use of the product rule of differentiation, we get
=
\XmXn)yi XmtiXfi ' XmXft,i = ^mi^n + Xm^ni
by (i).
116 3 CALCULUS OF TENSORS

(iii) Hence
=
\XmXn)tij Y\xmXn)ti\j

= ißmiXn)j + (Xm0ni)j

OmiXnJ "*" Xm^jPni

(because the δ symbol is a constant)

= àmiànj + ÖmjÖni
by (i);
= àimôjn + ôinôjm
(iv) Consequently, by using (3.3.4) we obtain
v\xmXn) = (xmxn),u = àimôin + ôimôin = 2ômn
(v) We have (r2)ti = Irr;,·. On the other hand, since r2 = xmxm> we get,
from the result (ii),
(r2)ti = (xmxm),i = l*xmômi = 2xt
Hence r f = ΛΤ,/Γ, if r ^ 0.
(vi) From the result (iii), we get
=
v ),ij \xmxm),ij =
^im^jm =
^dy
(vii) Consequently,
V2(r2) = (r2)tii = 2δΗ = 6 M

EXAMPLE 3.3.3 Ufk = /*(*,), prove the following:

(0 (**/*),/ =fi + Xkfkj
(") (**/*),„ = Λ ; + fj.i + **/*.</ (3·3·7)
(iii) v W * ) = yti*+^(V2A)
Solution (i) By the use of the product rule of differentiation, we get
=
(xkfk),i — xkjfk + xkfkJ =
^kifk + xkfk,i fi + x
kfk,i
(ii) Hence
(xkfk),ij =
(fi + xkfk,i)j =
fij + xkjfk,i + x
kfk,ij
=
fij + ^kjfkj + kfkjjx =
fij + ·//",* + X
kfk,ij
(iii) Consequently
ν 2 (χ*Λ) = (**/*).« = 2/;· ,· + **Λ>|7 = 2Λ,^ + **(ν2/*) ■
 3.4 GRADIENT OF A SCALAR 117

3.4
GRADIENT OF A SCALAR,
DIVERGENCE AND CURL OF A VECTOR

In this section we consider three important entities that arise frequently in

practical applications. The first one is a vector field associated with a scalar
field and the other two are scalar and vector fields associated with a vector
field.

3.4.1 GRADIENT OF A SCALAR FIELD

Consider a scalar field φ and put α, = φ-, in all coordinate systems. Then
, 8Φ

By using the chain rule of differentiation, this can be rewritten as

Ψ
' abc* dx't ·* dx'i
From the transformation rule (2.2.7) we find that

=a (3A1)
dX
k ,i

a* *
A ,Λ

Hence
<*} = 0Lik<t>ik = cLikak

This transformation rule shows that αέ = φ^ are components of a vector.

This vector is evidently a function of xt in general; it is called the gradient
of φ, denoted grad φ, or νφ.
Thus, if φ is a scalar field then V</> is a vector field with components given
by
[V4>]i = 4>.i (3.4.2)
It is obvious that the gradient of a constant scalar field is the zero vector.
Also,
ν(αφ ± βψ) = ανφ ± βνψ (3.4.3)
Τ7(φψ) = φ(νψ) + ψ(νφ) (3.4.4)
for all scalar fields φ and ψ and scalar constants a and β.
By virtue of (3.4.2), expression (3.3.2) can be rewritten as
df= [Vf]kdxk = Vf-dx (3.4.5)
118 3 CALCULUS OF TENSORS

3.4.2 DIRECTIONAL DERIVATIVE AND

NORMAL DERIVATIVE
From geometry, we recall that an equation of the form φ(χί) s
φ(χχ, χ2, *3) = constant represents a surface in three-dimensional space,
and that a normal to the surface has direction ratios (3φ/3χι, 3φ/οχ2,
οφ/οχ3). From (3.4.2) we note that these direction ratios are indeed the
components of V</>. It therefore follows that at any point x of the surface
φ(χί) = constant, the vector V</> is directed along a normal to the surface.
This is illustrated in Figure 3.1.
If n is the unit vector along V</>, then
V0
n= (3.4.6)
ÏV0Ï
Let a be a unit vector inclined at an angle Θ to the direction of νφ. Then
V0 · a = |νφ|(η · a) = |V0| cos Θ (3.4.7)
The scalar V</> · a, which represents the component of νφ along a, is called
the directional derivative of φ along a, usually denoted 3</>/da. Thus,
3φ
ää = νφ - a (3.4.8)

In particular, the directional derivative of φ along n is called the normal

derivative of 0, denoted θφ/dn. Thus,

(3.4.9)
dn

Figure 3.1. The vector V</>,

 3.4 GRADIENT OF A SCALAR 119

From (3.4.7) and (3.4.8) it is evident that |30/3a| is maximum when 0 = 0,

that is, a = n, so that
3φ 3φ
max - (3.4.10)
Thus, the normal derivative of φ is the maximum of all the directional
derivatives of φ.
From (3.4.6) and (3.4.9) we get the useful expression

V0 = |νφ|η= M j n (3.4.11)

3.4.3 DIVERGENCE OF A VECTOR FIELD

Consider a vector field u and put a = ukk in all coordinate systems. Then

dx'k
By using the chain rule of differentiation, this takes the form
g, = bu'k dxn
dxn dxk
Noting that u'k = akmum and using (3.4.1) this becomes
<*' = (<XkmUm)tn<*kn = ^km^kn^m,n

This transformation rule shows that a = ukk is a scalar (invariant). This

scalar is evidently a function of JC£ in general; it is called the divergence of
u, denoted div u, or V · u.
Thus, if u is a vector field, then div u is a scalar field given by
divu = uk%k = f[u]J>it (3.4.12)
It is obvious that the divergence of a constant vector field is equal to 0.
If u = V0, then ut = φ^ and ukk = φ^ = V2</>, or div u = V2</>. Thus,
we have the identity
div(V0) = ν2φ (3.4.13)
If v = 0u, then i>f = φui and divv = vkk = ^uk)k = Φtkuk + φuk k =
u · V0 + φ div u. Thus we have another identity,
div(0u) = 0(div u) + V0 · u (3.4.14)
120 3 CALCULUS OF TENSORS

3.4.4 CURL OF A VECTOR FIELD

Consider again a vector field u and put at = eimnunm in all coordinate
systems. Then

«* = * **
OJim

Using the fact that eijk are components of a third-order tensor and the
chain rule of differentiation, we get

a 0Î
i ipamq(Xnr£pqr ~ ~ ,

Noting that u'n = otnkuk and using (3.4.1) this becomes

a\ = oLipotmqocnramsankepqruktS

= <XipàqSàrk£pqrUk,s = ^ip^pqrUryq = <*ipap

This transformation rule shows that a, = eimnun m are components of a

vector. This vector is a function of xt in general, called the curl of u,
denoted curl u or V x u.
Thus, if u is a vector field, then curlu is also a vector field with com-
ponents given by

[curl u], = £imnuntm = eimn{[u]n}tm (3.4.15)

It is obvious that the curl of a constant vector field is the zero vector.
If u = V0, then (3.4.15) gives [curl u], = είίηηφηηι = 0, see (3.3.5). Thus,
we have the identity
curlV0 = O (3.4.16)
From (3.4.12) and (3.4.15), we get another identity:
div(curlu) = {[curlu]*},* = ekmnunymk =0 (3.4.17)
If v = φιι , then
[curlv], = eijkvkJ = eUk(<l>uk)j

= Zijk^UkJ + Zijk$jUk

= </>[curl u], + [V</> x u],

Thus, we have one more identity:

curl(</>u) = </>curlu + νφ x u (3.4.18)

 3.4 GRADIENT OF A SCALAR 121

From the definitions of div u and curl u , it is easy to verify that div and
curl are linear differential operators on vectors; that is,
div(au + ßy) = a div u + ß div v Ί
(3.4.19)
curl(au + ß\) = a curl u + ß curl vj
for all vectors u and v and all scalar constants a and /?.

3.4.5 THE OPERATOR u -V

In computations involving scalar s and vectors, we often deal with differen-
tial operators of the form Uj(d/dxj)( ), where Uj are components of a vector
u. We denote this operator by u · V. That is,

u-V = W ,— (3.4.20)
'dXj

so that
(u · V)</> = uj^ = uj4>j = u · (νφ) (3.4.21)

for a scalar field φ, and

(u · V)v = (u · V^e, = (UjVijye; (3.4.22)
for a vector field v. From (3.4.22), we note that
[(u · V)v]f· = ujvu = WjlMAj (3.4.23)

3.4.6 SOME IDENTITIES

The following are some other well-known identities:
div(u X v) = v · curl u - u · curlv (3.4.24)
curl(u X v) = (divv)u - (divu)v + (v · V)u - (u · V)v (3.4.25)
V(u · v) = v x curl u + u x curl v + (v · V)u + (u · V)v (3.4.26)
curl curl u = V div u - V2u (3.4.27)
2
Here, the vector V u, called the Laplacian o/u, is defined by
[V2!!], = V2ui = V2[u]f (3.4.28)
Taking v = u in (3.4.26), we get the following useful identity:
(u · V)u = iV(w2) - u x curl u = \V(u2) + (curl u) x u (3.4.29)
where u2 = u · u.
122 3 CALCULUS OF TENSORS

EXAMPLE 3.4.1 If φ = φ(χί) and / = / ( φ ) , show that

(i) νψχν/=θ

(ii) ν ί /(φ)</φ| = /(Φ)(νφ)

Solution (i) S i n c e / = /(φ) and φ = Φί*,), we have by the chain rule of

differentiation/^ = {/'(φ)}φ^ so that V / = {/'(φ)}(νφ). Hence
ν φ X V/ = {/'(φ)}(νφ) X (νφ) = 0
(ii) If we set ί/(φ) άφ = ^ φ ) , then we get F , = {F'(«)0 f i = {/(Φ)}Φ,/ so
that VF = {/(φ)}(νφ). ■

EXAMPLE 3 . 4 . 2 Write down div(u x v) in the suffix notation and

hence prove the identity (3.4.24).
Solution By the expression (1.7.9), we have [u x v]^ = ekUUiVj. Hence
(3.4.12) gives
div(u x ▼) = {[ii x w]k)9k = ekU(UiVj\k (3.4.30)
This is the expression for div(u x v) in the suffix notation.
Carrying out the indicated differentiation in (3.4.30), we get
div(u x v) = ekij(uitkvj + UiVjtk) = (ejkiui>k)Vj - (eikjVjtk)Ui
= [curlu]jVj - [curly],*/,
(by (3.4.15))
= v · curl u - u · curl v ■

EXAMPLE 3 . 4 . 3 Show that

[u x curlv] z = (Vjj - Vij)Uj (3.4.31)
Deduce the identity (3.4.26).
Solution By virtue of the expressions (1.7.9) and (3.4.15), we have
[u x curlv], = ^Julyfcurlv]* = eiJkekmnUjVntm
Using the ε-δ identity, this becomes
[u X curlv], = (ôimôjn - ôinôjm)UjVntm = (VJJ - vu)Uj
This is the required result (3.4.31).
From (3.4.31) and (3.4.23), we get
[u x curlv + (u · V)v], = UjVjj (3.4.32)
 3.4 GRADIENT OF A SCALAR 123

Similarly
[v x curlu + (v · V)u]f = VjUjj. (3.4.33)
Expressions (3.4.32) and (3.4.33) yield
[u X curl v + (u · V)v + v X curl u + (v · V)u]f
= UJVJJ + VjUjj = (UjVj)ti = (u · v) t/ = [V(u · v)L
(by (3.4.2)). This proves the validity of the identity (3.4.26). ■

EXAMPLE 3 . 4 . 4 Prove the identity:

V2(x · v) = 2 div v + x · V2v (3.4.34)
Solution By use of the relation (iii) of Example 3.3.3 we have
V2(**v*) = 2vktk + xkV2vk
When written in the direct notation, this becomes the identity (3.4.34). ■

EXAMPLE 3 . 4 . 5 Write down the following equation in the direct

notation:
ßV2Ui + (A + ß)ukiki + p0bi = p0üi (3.4.35)
where A, μ and p0 are constants.
Solution We first recall that
V2W/ = [ν 2 ιι], and uk§ki = (uktk)ti = [V(divu)]f
Hence the given equation is
//[V2u]f + (A + //)[V(div u)L + p0[b]i = p0[*]i
Therefore, in the direct notation the equation reads
μν 2 ιι + (λ + //)V(div u) + p0b = p 0 a ■ (3.4.35)'

EXAMPLE 3 . 4 . 6 Show that v is orthogonal to curl v if and only if

v = φνψ for some scalar functions φ and ψ.
Solution First, suppose that v = φνψ. Then by using identities (3.4.16)
and (3.4.18), we get
curlv = V0 x νψ
Hence
v · curl v = φνψ · (νφ χ νψ) = 0
Thus, v is orthogonal to curl v.
124 3 CALCULUS OF TENSORS

Next, suppose that v is orthogonal to curl v; that is, v · curl v = 0, or

equivalently,
V\(v3t2 ~ ν2,ύ + ^2(^1,3 - ^3,1) + v3(v2tl - vh2) = 0
This implies that the total differential equation
vi dxx + v2 dx2 + v3 dx3 = 0
is integrable. Hence, there exist functions μ 5* 0 and ψ such that
μ(υγ dxx + v2 dx2 + v3 dx3) = άψ
Setting φ = Ι/μ and using (3.3.2), this can be rewritten as vkdxk =
(<t>W,k)dxk> which yields vt = φψ^ so that v = φνψ, as required. ■

EXAMPLE 3-4.7 A vector field v is said to be a Beltrami field if

curlv = ων for some scalar field ω. (Then ω is called the abnormality
factor of v.) Show that for such a field,

ω = -j—r v · curl v (3.4.36)

v
_ curl v · curl curl v
=
(3.4.37)
|curlv| 2
If v = νφ + ψ{νχ), show that
νφ · v> x27—72
v^
ω = /r?^2 (3.4.38)
2
(V0) - ψ\νχ)
Solution By data, curl v = ω\. Hence
v · curl v = v · (α>ν) = ω|ν|
This is the required result (3.4.36). Also,
curl curl v = curl(a;v) = ω curl v + Va> x v
by the identity (3.4.18). Hence
curl v · curl curl v = co|curl v| 2 + curl v · (Vco x v)
Since
curl v · (Va> x v) = ων · (Va> x v) = 0
the required result (3.4.37) follows.
If y = V</> + ^(Vx), we find, by using the identities (3.4.16) and (3.4.18),
that
curlv = νψ x νχ
 3.5 GRADIENT OF A VECTOR 125

Hence
νφ · (V^ X νχ) = (V</> - ψνχ) · (V^ x V*) = (νφ - ψ^χ) · curl v
= (V0 - ψνχ) · ω\ = ω(νφ - ψ^χ) · (V</> + ψ^χ)
= ωΐ(νφ)2 - ψ2(νχ)2}
This is the required result (3.4.38). ■

3.5
GRADIENT OF A VECTOR,
DIVERGENCE AND CURL OF A TENSOR

In this section we introduce the gradient, divergence, curl and Laplacian

operators frequently encountered in tensor calculus. These are extensions of
and analogous to the operators defined in the preceding section.

3.5.1 GRADIENT OF A VECTOR FIELD

Consider a vector field u with components w, and put au = uitj in all
coordinate systems. Then

y
ax}
By using the chain rule of differentiation, this becomes
, = du'j dxn
a li
' dxn dx'j

Noting that u[ = aimum and using (3.4.1) this becomes

Öl; = Ot;
**ij MCL;„Um
^im^jn^m, n„ = (X;^im^jn^mn
mCX;„Clm„

This transformation rule shows that ay = uitJ are components of a (second-

order) tensor. This tensor is a function of jcf in general and is called the
gradient o/u, denoted grad u or Vu.
Thus, ifu is a vector field then Vu is a tensor field with components given
by
[Vu]l7 = uiJ=i[u]i]j (3.5.1)
The transpose of this tensor is denoted by Vur. Thus, we have
[VuT]u=uj9i = [[u]j)9i (3.5.2)
126 3 CALCULUS OF TENSORS

Also, if v is a vector, then by virtue of (2.8.19), it follows that v Λ Vu is

a tensor whose components are given by
[vAVu]i:/ = eirsvruj>s (3.5.3)
It is obvious that the gradient of a constant vector field is the zero tensor of
the second order. It is straightforward to verify that
(i) Vx = I (3.5.4)
(ii) I · Vu = Vu · I = tr(Vu) = divu (3.5.5)
(iii) tr(v Λ Vu) = v · curl u (3.5.6)
If u = V</>, then Vu = V(V</>) is called the second gradient of </>, denoted
ν ν φ . Thus
[VV<t>hj = <i>,ü (3.5.7)
Evidently,
(VV0) r = VV0 (3.5.8)

3.5.2 GRADIENT OF A TENSOR FIELD

Consider a tensor field A with components au and put cijk = aijk in all
coordinate systems. Then

By using the chain rule of differentiation, this becomes

Noting that a'u = aimajnamn and using (3.4.1) this becomes

This transformation rule shows that cijk = a^ k are components of a third-

order tensor. This tensor is in general a function of xi9 called the gradient
of A, denoted grad A or VA.
Thus, if A is a (second-order) tensor field, then VA is a third-order tensor
field with components given by
(3.5.9)
Obviously, the gradient of a constant second-order tensor field is the zero
tensor of order three.
 3.5 GRADIENT OF A VECTOR 127

If A = Vu, then VA = V(Vu) is called the second gradient of u, denoted

VVu. Thus
[ W u ] ^ = uiJk = [[u],}^ (3.5.10)

By taking contraction in (3.5.10) we find that V tij — Uj kk are components

of a vector. This vector is the Laplacian of u, denoted V2u; see (3.4.28).

3.5.3 DIVERGENCE OF A TENSOR FIELD

If ay are components of a tensor field A, it already has been seen that aijtk
are components of the tensor VA. Consequently, it follows (by contraction)
that aiJtj are components of a vector field. This vector field is called the
divergence of A, denoted div A.
Thus, if A is a second-order tensor field, then div A is a vector field with
components given by

[divAL^^-^ilA^J,, (3.5.11)

Obviously, the divergence of a constant (second-order) tensor field is the

zero vector.
From (3.5.11) it readily follows that d i v A r is a vector field with
components
[div AT]i = ajU = l[A]ß}j (3.5.12)

Setting A = Vu, we find from (3.5.11) and (3.5.12), that

[div Vu], = uUJ = V2Ui

so that
div Vu = V2u (3.5.13)
and
[div Vu r ], = Ujjj = ujji = [Vdivu],
so that
d i v V u r = V(divu) (3.5.14)

3.5.4 CURL OF A TENSOR FIELD

Consider again a tensor field A with components atj. Let us now put
C
U = £imnajn,m ^n a ^ coordinate systems. Then

c'.. = e'- ^
UJL
m
128 3 CALCULUS OF TENSORS

Using the fact that eimn are components of a third-order tensor and the
chain rule of differentiation, this expression becomes
dxs
®-ip ^mq &nr ^pqr n y \fljn
jn/ )
dxs " dx'm
Noting that a'jn = oijhankahk and using (3.4.1), we get
C'ü = ^ipamq^nr^jhOinkOims€pqrahks

= <Xip<XjhÔqsÔrk£pqr<thk,s

= aipajh8pqrahrg = aipaJhcph
This transformation rule shows that q, = eimnaJntm are components of a
second-order tensor. This tensor is a function of A:, in general and is called
curl of A, denoted curl A.
Thus, if A is a second-order tensor field, then curl A is also a second-
order tensor field with components given by
[curl A],·, = eimnaJntm = eimn{[A]Jn]}m (3.5.15)
It is obvious that the curl of a constant tensor field is the zero tensor.
From (3.5.15) it readily follows that curl A r and (curl A) r are tensors with
components given by
[curl Α% = £imnanjym = ^ { [ A ] ^ ) ^ (3.5.16)
T
[(curl A) ]U = eJmnainfm = ^„{[A],,,)^ (3.5.17)
Evidently, (curl A ) r ^ curl A r , in general.
If A = Vu then from (3.5.15) and (3.5.16), we find that
[curlVu],·, = eimnuJtnm = 0
so that
curl Vu = 0 (3.5.18)
and
[curl Vu r ] 0 = eimnunJm = (eimHuntm)j
= {[curlu]Ä)fy = [Vcurlu],·,
so that
curlVu r = Vcurlu (3.5.19)

3.5.5 LAPLACIAN OF A TENSOR FIELD

If au are components of a tensor A, it has been noted that cijk = aijk are
components of a third-order tensor VA. It is not hard to verify that
c
ijk,m = au,km a r e components of a fourth-order tensor field. This tensor is
 3.5 GRADIENT OF A VECTOR 129

called the second-gradient of A,, denoted VVA. Consequently, it follows (by

contraction) that cijkk = aiJkk are components of a second-order tensor
field. This tensor is called the Laplacian of A, denoted V2A.
Thus, if A is a second-order tensor field, then V2A is also a second-order
tensor field with components given by
[V2A]l7 = v\au) = V2[A]U (3.5.20)
Obviously, the Laplacian of a constant tensor field is the zero tensor. From
(3.5.20) it readily follows that
(V 2 A) r = V 2 A r (3.5.21)
2
It is easy to verify that V, div, curl and V are linear differential operators
in tensor calculus also; that is,
V(au -I- ß\) = «Vu + ßV\
V(aA + ßB) = aVA + ßVB
div(aA + ßB) = cxdiv A + £div B (3.5.22)
curl(aA + ßB) = a curl A + ß curl B
V2(c*A + ßB) = aV 2 A + ßV2B
for any vectors u, v, any tensors A and B and any constant scalars a and ß.

EXAMPLE 3-5.1 Let ui9 at and bt be components of vectors u, a and b,

and eu, w^ and τ/7 be components of tensors E, W and T, respectively. Write
down the following equations in the direct notation:
e + u
U = i(uu J,i)> W
U = i(uij - uJ,i)> (3.5.23a, b)
TjU + p0bi = pQai (3.5.24)
Solution By data and expressions (3.5.1) and (3.5.2), equations (3.5.23a)
may be rewritten as
[E]/7 = i[Vu + V u %
In the direct notation, these read
E = |[Vu + Vu r ] = sym(Vu) (3.5.23a)'
Similarly, equations (3.5.23b) read as follows in the direct notation:
W = |[Vu - Vu r ] = skw(Vu) (3.5.23b)'
By the data and expression (3.5.12), equations (3.5.24) may be rewritten as
[divTr];( + A>[b],· = A>M«
130 3 CALCULUS OF TENSORS

In the direct notation, this reads

div T r + p0b = p02i M (3.5.24)'

EXAMPLE 3 . 5 . 2 For the vector field u = a(x^x2e1 + *f* 3 e 2 + xjx^),

where a is a constant, find (i) the gradient of u, (ii) the divergences of Vu
and Vu r and (iii) the curl of Vu r .
Solution For the given vector field, we have
(3.5.25)
These give

(3.5.26)

This is the matrix of Vu.

From (3.5.26), we get

(3.5.27)

This is the matrix of Vu r .

On using (3.5.25), expression (3.5.13) yields
[div Vu]! = V2ul = 2ax2, [div Vu]2 = V2w2 = 2oa3,
[div Vu] 3 = V2w3 = 2axx
Thus
div Vu = V2u = 2of(jc2e1 + * 3 e 2 + x^3) (3.5.28)
On using (3.5.25), expression (3.5.14) yields
[divVu r ]! = 2OL(X2 + ΛΤ3), [divVu r ] 2 = 2CL(XX + x 3 ),

[div Vu r ] 3 = 2OL(X2 + xx).

Thus
div(Vu r ) = 2OL[{X2 + x3)ei + (xl + * 3 )e 2 + (x2 + *i)e3] (3.5.29)
It may be verified that this is equal to V div u.
From (3.5.19) and (3.5.25), we find that
[curl Vu r ] 1 2 = - 2 x 2 , [curl Vu r ] 2 3 = -2ΛΓ 3 , [curl Vu r ] 3 1 =2xi
 3.5 GRADIENT OF A VECTOR 1 31

and
[curl νητ]ϋ =0
for other values of i,j. Hence

[curl Vu r ] (3.5.30)

This is the matrix of curlVu r . It may be verified that curlVu r =

V curl u. ■

EXAMPLE 3 . 5 . 3 Let u be a given vector, and A be a given tensor. For

any fixed vector c prove the following:
(i) (Vu r )c = V(u-c) (3.5.31)
(ii) (div A) · c = div(A r c) (3.5.32)
r
(iii) (curl A)c = curl(A c) (3.5.33)
Solution Let au be components of A and ut and c, be the components of
u and c, respectively. Then
(i) [(Vu r )c] f = [Vu%[c] 7 = ujticj = (ujCj)ti = (u · c) pi = [V(u · c)],·
This proves (3.5.31).
(ii) (div A) · c = [div A ] , ^ = aUtjCi = ajUCj = (aßCj)ti = ([A^],·),,·
= div A r c
(iii) [(curlA)c]i = [curlA]0-[c]y- = (eimnaJnfm)Cj = eimn(ajnCj)im
T
= eimn([A c]n),m = [curKA^L-
This proves (3.5.33).

EXAMPLE 3 . 5 . 4 Prove the following identities:

(i) (u · V)v = (Vv)u (3.5.34)

(ii) div(0A) = 0divA + Ανφ (3.5.35)

(iii) div(Au) = u · div AT + AT - Vu (3.5.36)

(iv) div(curl A) = curl(div AT) (3.5.37)

(v) div(curl A ) r = 0 (3.5.38)

132 3 CALCULUS OF TENSORS

(vi) tr(curlA) = 0 (3.5.39)

if A is symmetric;
(vii) curl(curl A r ) = (curl curl A)7" (3.5.40)
Solution (i) [(Vv)u]( = [Vv]y[u]7- = v.jiij = [(u · V)y\it see (3.4.23). This
proves (3.5.34).
(ii) [div(0A)]( = {[<t>A\ij}j = (φα^ = φα^ + au4>j. This proves (3.5.35).
(iii) [div(Au)] = {[An],),, = {auuj\t
= aUilUj + dijUjj = djijUi + aßuu

= [u],[div A ] ( + [A ]0[Vu]y = u · div AT + A7"· (Vu)

r r

(iv) [αν(αΐΓΐΑ)]( = ί[απ-1Α](7},,

~~ ^imn^jn,mj ~~ ^imn\ßjnj)ym

= eimn{[diwAT]nlm = [curl(div A71)],.

This proves (3.5.37).
(v) [div(curlA)r], = {[(curl Af]^ = {[curl A]ß}j = £jmnain^mj = 0.
This proves (3.5.38).
(vi) tr(curl A) = [curl A]l7 = ^imnain%m. If A is symmetric, we have
Eimn"in,m ~~ ^inm"ni,m ^nmi^*in,m ~^imn^*in,m

so that eimnainm = 0. Hence tr(curl A) = 0, if A is symmetric.

(vii) [curl(curl A7)],·, = eimni[cm\AT]jn}tm = eimn(ejpq[[AT]nq]tP)tm
~ ^imn^jpq^qn,pm ~~ ^jpq\Pimrfiqn,m),p

= eJpqi[cmlA\iq}tP = [curl(curl A)]ß

This proves (3.5.40). ■
EXAMPLE 3.5-5 Let a and b be constant vectors. For any vectors u and
v and any tensor A, prove the following:
(i) div{(u · a)Arb} = a · {(Vu)Ar + u <g> div A}b (3.5.41)
r r r r
(ii) curl{(u · a)A b} · v = a · [(v Λ Vu) A + u <g) (curl A) v]b (3.5.42)
Solution Let ai9 bif w, and vt be components of a, b, u and v, respec-
tively, and au be components of A. Then,
(i) div{(u · a)Arb) = (ukakaßbj)ti = akbj(ukjari + ukaßJ)
T
= akbj{[(Vu)A ]kj + [u <g> div A]*,·}
= ak[{(Vu)AT + u <g> div A}b]*
= a ■ {(Vu)Ar + u (g) div A}b
 3.5 GRADIENT OF A VECTOR 133

(ii) curl{(u · a)A r b) · v = eirs(ukakajsbj)trVi

= ZirS<*kl>jVi(uktrajS + ukajsr)
= Okl^sirViUktr)ajs + uk(€irsajStr)Vi]bj
= ak{[vAVu]sk[A]js + [uUcurlAJ^vLJô,
= ak{[(y Λ V u ) A % + [u (g) (curl A) r v]^}6 7
r

= a · {(v Λ Vu) r A r + u (g) (curl A)rv}b ■

EXAMPLE 3 . 5 . 6 If W is a skew tensor and w is the dual vector thereof,

show that
curlW = (divw)I - Vw (3.5.43)
Solution Let wu be components of W and wf be components of w. Then
Wij and Wi are related as follows (see (2.10.4)): wu = -eijkwk. Hence

^imn^jntm ~ ~^imn^jnp*^p,m ~ ^imn^jpn^p,m

= (àijômp - ôipômj)wpm = ôijWpp - witj

When written in direct notation, this becomes (3.5.43). ■

EXAMPLE 3-5-7 Let u be a vector, E = sym Vu, W = skw Vu and w be

the dual vector of W. Prove the following:
(i) w = jcurlu (3.5.44)
2 2 2 2 2
(ii) |Vu| = |E| + |W| = |E| + fleuri u|
= div(Vu r )u - u · V 2 u. (3.5.45)
(iii) Vu · Vu r = |E| 2 - |W| 2 = |E| 2 - £|curl u| 2
= div(u · V)u - (u · V)(div u)
= div{(Vu)u - (divu)u) + (divu) 2 (3.5.46)
(iv) EW + WE = skw(Vu)2 (3.5.47)
Solution (i) Let ui9 wi9 eu and wu be components of u, w, E and W,
respectively. Then by use of (2.10.2) we get

When written in the direct notation, this becomes (3.5.44).

134 3 CALCULUS OF TENSORS

(ii) Consequently, by use of (2.10.8), we have

|W| 2 = ||curlu| 2 (3.5.48)
Noting that Vu = E + W, we find
|Vu|2 = Vu · Vu = (E + W) · (E + W) = |E| 2 + |W| 2 (3.5.49)
because E · W = 0. On the other hand,

| V u | 2 = UijUij = (UiUu)j - UiUijj

= il(ynT)u]j]j - [uU^uh
= div(Vu r )u - u · V2u (3.5.50)
Expressions (3.5.48), (3.5.49) and (3.5.50) constitute the relations (3.5.45).
(iii) Since Vu r = E - W, we get
Vu · Vu r = (E + W) · (E - W) = |E| 2 - |W| 2 (3.5.51)
On the other hand,
Vu · V u r = UijUjj = (UijUjXi - UjUiyU

= «(u-V)u] | .} >I .-(u-V)(« | . >f )

= div{(u · V)u) - (u · V)(div u) (3.5.52)
Also,
Vu · Vu r = uuUjj = (uuuj - uktkUi\i + (ukykf

= {[(Vu)uL - [(Uk,k)*h)9i + ("k,k)2

= div{(Vu)u - (divu)u) + (divu) 2 (3.5.53)
Expressions (3.5.48), (3.5.51), (3.5.52) and (3.5.53) give (3.5.46).
(iv) Next,
(Vu)2 = (Vu)(Vu) = (E + W)(E + W)
= E 2 + W2 + EW + WE. (3.5.54)
Hence
[(Vu) 2 ] r = E 2 + W2 - (EW + WE) (3.5.55)
Expressions (3.5.54) and (3.5.55) yield·
(Vu)2 - [(Vu)T = 2(EW -h WE) (3.5.56)
from which (3.5.47) is immediate. ■
 3.5 GRADIENT OF A VECTOR 135

EXAMPLE 3-5.8 Show that for any tensor A,

curl curl A = [V2(tr A) - div(div A)]I + V div AT
+ (Vdiv A)T - VV(tr A) - V 2 A r (3.5.57)
Deduce that, if A is symmetric, then the equation
curl curl A = 0 (3.5.58)
is equivalent to the equation
V2A + VV(tr A) - V(div A) - (V div A ) r = 0 (3.5.59)
Solution If Ojj are components of A, we note that
[curl curl A ] 0 = eirsejmnasntmr (3.5.60)
By using the identity (1.7.22) this becomes
àij Sim ôin
[curl curl A],·, = (aSfttmr) àrj àrm ^rn (3.5.61)

On expanding the determinant on the righthand side, this simplifies to

[curl curl A] l7 = ôu(a„tmm - amryVnr) + ajrJr + amitmj - ajUrr - aSSfU
2
= {V (tr A) - div (div A ) ^ · + [(V div A ) %
+ [V(div A % - [ V 2 A % - [VV(tr A)] l7 (3.5.62)
This is precisely the identity (3.5.57).
If A is symmetric, the identity (3.5.57) reads as follows in the suffix
notation:

^irs^jmn^sn,mr ~ \y*ss,mm **mrymr)^ij ■" ^im,mj

a
+ <*jm,im - ij,mm ~ <*mmyij (3.5.63)

Now, suppose that (3.5.58) holds; then we have

^irs tjmn &sn, mr ^ (3.5.64)

Consequently, eirseimnasnmr = 0, which, on using the ε-δ identity, reduces to
asS,mm - amr,mr = 0 (3.5.65)

In view of (3.5.64) and (3.5.65), identity (3.5.63) yields

<lij,mm + <*mm,ij ~ <*im,mj ~ <*jm,im = 0 (3.5.66)

136 3 CALCULUS OF TENSORS

When written in the direct notation, these equations become the tensor
equation (3.5.59). Thus equation (3.5.58) implies equation (3.5.59).
Conversely, suppose that the tensor equation (3.5.59) holds. Then
(3.5.66) hold. Taking the contraction of (3.5.66), we find that (3.5.65)
holds. Since (3.5.65) and (3.5.66) hold, identity (3.5.63) yields (3.5.64). This
means that the tensor equation (3.5.58) holds.
Thus, equations (3.5.58) and (3.5.59) imply each other; the equations are
therefore equivalent. ■

3.6
INTEGRAL THEOREMS FOR VECTORS

Here we state the divergence theorem and Stokes's theorem in vector

integral calculus, with which the reader should be already familiar. Some
consequences of these theorems are also summarized.

3.6.1 DIVERGENCE THEOREM

Let V be the volume of a three-dimensional region bounded by a closed
regular surface S; see Figure 3.2. Then for a vector field u defined in V
and on 5,
(divu)rfK= (u-n)rfS (3.6.1)

where n is the unit outward normal to S. In the suffix notation, the expres-
sion (3.6.1) reads:
\uktkdV= \uknkdS (3.6.1)'

^J—-Surface element dS

Volume V /.
/ ^ Surface S

Figure 3.2. Regular region.

 3.6 INTEGRAL THEOREMS FOR VECTORS 1 37

The following are some immediate consequences of expression (3.6.1):

(i) νφάν=\ 0nt/5, or φΙζάν=\ <t>nkdS (3.6.2)

]v ]s }v ' Js

(ii) (curlu)rfF= (nxu)tfS or I 6ijkuk jdV = I eijknjUkdS

]V U ]V Js
(3.6.3)

(iii) {V2<j>)dV= (n-V)^rfS or <t>,kkdV = \ nk<t>,kdS

}V }S ]V Js
(3.6.4)

(iv) (V2u)</K = (n-V)urfS or u,tkkdV= \ nkui<kdS

]V h ]V ]S
(3.6.5)
In (3.6.2) and (3.6.4), φ is a scalar field defined in Fand on S.

3.6.2 SOLENOIDAL VECTORS

The surface integral J s u · n dS is usually called the outward normal flux or
just//wjc of u across S. A vector is said to be solenoidal in a region if its flux
across every closed regular surface in the region is 0. From the divergence
theorem, it follows that u is solenoidal in a simply connected region if and
only if div u = 0 in that region.
A vector field whose divergence is exactly 0 is called a divergence-free
vector. From the statement just made, it follows that a vector field is
solenoidal in a simply connected region if and only if it is divergence free.
Since div(curlu) = 0, see (3.4.17), therefore curlu is a divergence-free
vector for every vector u. It can be proven that every divergence-free vector
u defined in a simply connected region can be represented as
u = curl w (3.6.6)
where w is itself a divergence-free vector. Here w is called a vector potential
of u.

3.6.3 STOKES S THEOREM

Let C be a simple closed curve in three-dimensional space and S be an open
regular surface bounded by C; see Figure 3.3. Then for a vector field u
defined on S as well as C,

u · t ds = (curl u) · n dS (3.6.7)
c Js
138 3 CALCULUS OF TENSORS

Figure 3.3. Open regular surface.

where t is a unit tangent vector to C that is assumed to be positively oriented

relative to the unit normal n to S.
In the suffix notation, expression (3.6.7) reads

uitids=\ eijkukjnidS (3.6.7)'

c Js
If S is a closed surface, the lefthand side of (3.6.7) reduces to 0, then

(curl u) · n dS = 0 or eUkukJnidS = 0 (3.6.8)

s Js
This expression also follows from the divergence theorem applied to curl u.
The particular case of (3.6.7), where C lies entirely in thex l jc2-plane and S
is the part of the plane bounded by C, is important in its own right. In this
case, expression (3.6.7) reduces to the form

(i/j dxx + u2dx2) = I (w2,l _ ult2)dxidx2 (3.6.9)

c Js
where ux and u2 are the xx and .^-components of u. This particular case of
Stokes's theorem is usually referred to as Green's theorem in the plane,
Following are some immediate consequences of the expression (3.6.7):

(i) ® <l>tds = \ nxV<l)dS or (D 0^ife= eUknj<l>tkdS (3.6.10)

Jt JS JC JS
 3.6 INTEGRAL THEOREMS FOR VECTORS 139

(ü) φ (u x t) ds =\ [(div u)n - (Vu)rn] dS or

ic Js

zukujtkds = [Uk^rii - ukJnk]dS (3.6.11)

c Js

(iii) curl w ids = (div u) - n · (V2u) dS or

c Js dn

d_
CijkUkJids -- ("*,*) - "iHkk dS (3.6.12)
c Js dn
In (3.6.10), φ is a scalar field defined on S as well as C.
Note: The arc element t ds is often denoted dx. Consequently, we write
lc()dx for IcOtds.

3.6.4 CONSERVATIVE AND IRROTATIONAL VECTORS

The line integral in the lefthand side of expression (3.6.7), namely,
§ c u · t ds, which represents the integral of u · t around C, is called the
circulation of u round C. A vector u defined in a region is said to be
conservative if its circulation round every simple closed curve in the region
is 0, or equivalently, if the value of the line integral {* u · t ds, defined in the
region, depends only on the endpoints A and B and not on the curve joining
A and B on which integration is carried out. Also, a vector is said to be
irrotational (or curl free) in a region if its curl is exactly 0 in the region.
From Stokes's theorem, it follows that, in a simply connected region, a
vector is conservative if and only if it is irrotational in the region.
Since curl V</> = 0, see (3.4.16), therefore V</> is an irrotational vector for
every scalar field φ. It can be proven that every irrotational vector u defined
in a simply connected region can be represented
u = Υφ (3.6.13)
Then φ is called a scalar potential of u.
If a vector u is both divergence free and irrotational, then the identity
(3.4.27) shows that V2u = 0; in this case we say that u is a harmonic vector.

3.6.5 HELMHOLTZ'S REPRESENTATION

It has been noted that a divergence-free vector has a representation as
given by (3.6.6) and an irrotational vector has a representation as given by
(3.6.13). A representation valid for a general vector, known as the
Helmholtz's representation follows.
140 3 CALCULUS OF TENSORS

Consider a vector field u. Define the vector field v by

V(X) = - - L [ r^rdV (3.6.14)

4π ]v\\ - \\
where V is some volume in the region of definition of u and the integral is
taken by varying x over F, keeping x as a fixed point. It can be proven that
V2v = u. Then by using the identity (3.4.27) we get

u = v</> + curlw (3.6.15)

where
φ = divv, w = -curlv (3.6.16)

Thus, given a vector field u, there exist a scalar field φ and a vector field
w, defined by (3.6.16) and (3.6.14), such that u has a representation as given
by (3.6.15). This is the Helmholtz's representation. Note that the vector w
present in this representation is divergence free.

EXAMPLE 3.6.1 Prove expressions (3.6.3) and (3.6.11).

Solution Let a be an arbitrary constant vector field. Employing the

divergence theorem to the vector a x u, we get

div(a xu)dV= (axu)-nrfS (3.6.17)

v Js
Using identities (3.4.24) and (1.4.18), expression (3.6.17) becomes

a · \\ curl udV - nxurfS} = 0

Since a is arbitrary, this yields (3.6.3).

On the other hand, if we employ Stokes's theorem to the vector a x u, we
get

(a x u) · ids = curl(a x u) · n dS (3.6.18)

c Js
Using identity (1.4.18) and

[curl(a x u)] · n = [(div u)a - (a · V)u] · n = [(div u)a - (Vu)a] · n

= a · [(div u)n - (Vu) r n]

which is obtained by using (3.4.25), (3.5.34) and (2.8.14), expression

 3.6 INTEGRAL THEOREMS FOR VECTORS 141

(3.6.18) becomes

a · I Φ u x t ds - (div u - Vur)n dS = 0

Since a is arbitrary, this yields (3.6.11). ■

EXAMPLE 3.6.2 Show that

n x curl u = (Vu)rn - (n · V)u (3.6.19)
Hence, deduce the following alternative version of (3.6.11):

φ u x t ds =\ [(div u)n - (n · V)u - n x curl u] dS (3.6.20)

Jc Js
Solution By use of (3.4.31), (3.5.2) and (3.4.23), we get
[n x curlu], = (ukti - uitk)nk = [(Vu)rn - (n · V)u]f
from which (3.6.19) is immediate.
Substituting for (Vu)rn from the identity (3.6.19) in the expression
(3.6.11), we readily get the expression (3.6.20). ■

EXAMPLE 3.6.3 If A is a tensor field such that div A r = 0 in a simply

connected region, show that A = curl W for some tensor W.
Solution As usual, let e,· be any one of the base vectors of a fixed
coordinate system. Then
div(Ae/) = e#· · div AT
see (3.5.32);
=0
by data. Hence, by virtue of (3.6.6), Ae,· can be represented as
Aef = curlw, (3.6.21)
for some divergence-free vector w,. Let W = e* (x) w^. Then
W^e,· = (w* ® ek)et = (ek · e^w* = W | ·. (3.6.22)
Hence (3.6.21) and (3.6.22) give
Ae, = curlw, = cur^W 7 ^) = (curlWJe,
see (3.5.33). By virtue of the result proven in Example 2.7.2, it follows that
A = curl W. ■
142 3 CALCULUS OF TENSORS

EXAMPLE 3 . 6 . 4 If A is a tensor field such that curl A = 0 in a simply

connected region, show that A = Vu for some vector u. If, further,
tr A = 0, show that A = curl W for some skew tensor W.
Solution As usual, let ef be any one of the base vectors of a fixed coor-
dinate system. Then,
cm\(ATei) = (curlAJe,
see (3.5.33);
=0
by data. Hence by virtue of (3.6.13), there exists a scalar field </>,· such that
ΑΓβ, = νφ, (3.6.23)
Let u = <t>kek. Then u · e, = </>,· so that, by using (3.5.31),
V0f = V(u · e,) = (Vu^e, (3.6.24)
Hence (3.6.23) and (3.6.24) give A 7 ^ = (Vu r )e,. By virtue of the result
proven in Example 2.7.2, it follows that AT = Vu r or A = Vu. Conse-
quently, tr A = tr Vu = div u. Hence, if tr A = 0, then div u = 0.
Let W be the skew tensor of which - u is the dual vector. Then, from
(3.5.43), it follows that
curlW = div(-u)I - V(-u)
Since div u = 0, we get
curl W = Vu = A ■

EXAMPLE 3 . 6 . 5 Let u be a vector field defined on and inside a closed

regular surface S enclosing a volume V, and E = sym Vu and W = skw Vu.
If u = 0 on 5, prove the following:

(i) |W| 2 rfK< \E\2dV (3.6.25)

(ii) Korn's inequality:

\Vu\2dV<2 \E\2dV (3.6.26)

v Sv
Solution (i) From (3.5.46), we get

{|E|2 - |W| 2 }rfK= div{(Vu)u - (divu)u)dV + (divu) 2 rfK

 3.7 INTEGRAL THEOREMS FOR TENSORS 143

Employing the divergence theorem to the first of the integrals on the

righthand side, and recalling that (Vu) u = (u · V)u, we get

[|E|2 - |W|2} dV =\ {(u · V)(u · n) - (div u)(u · n)} dS + (div u)2 dV

]V ]S ]V
(3.6.27)
Since u = 0 on S by data, this yields

|E| 2 rfK- \W\2dV= (divu) 2 rfF>0

from which (3.6.25) is immediate.

(ii) From (3.5.46) we also get

|Vu| 2 rfF= |E| 2 </F+ \W\2dV (3.6.28)

Using the inequality (3.6.25), expression (3.6.28) yields

|Vu| 2 rfF< \E\2dV+ \E\2dV

v ]v ]v
which is (3.6.26).

3.7
INTEGRAL THEOREMS FOR TENSORS

We now obtain some extensions to tensor fields of the divergence theorem

and Stokes's theorem stated in Section 3.6. As in the case of a vector field,
an integral of a given tensor field is defined as a tensor field whose
components are the integrals of the components of the given field.

THEOREM 3.7.1 (Divergence Theorem for a Tensor) Let V be the

volume of a three-dimensional region bounded by a closed regular surface
S. Then for a tensor field A defined in V and on 5,

divAdV= An dS (3 JA)
v is
where n is, as usual, the unit outward normal to 5.
144 3 CALCULUS OF TENSORS

Proof Let c be an arbitrary constant vector. Then

c· di\AdV= c-(divA)rfK

see (3.5.32);
-I. div(A'c)dV

= (ATc)-ndS = c-(An)rfS
Js ]s
so that
J divArfK- AnrfS = 0

Since c is an arbitrary vector, this expression yields the result (3.7.1).

In the suffix notation, expression (3.7.1) reads

\aikikdV= \aiknkdS (3.7.1)'

Consequences
(i) If A = φΒ (or ay = <t>bij) where φ is a scalar, then (3.7.1) and (3.7.1)'
become, respectively,

άι\(φΒ)αν
άί\(φΒ)άν= = φΒηάΞ (3.7.2)

[<**!*). dV=k \^biknkdS (3.7.2)'

(ii) If A = φΐ, expression (3.7.1) becomes, on noting that div(</>I) = V<£,

see (3.5.35),

1,Φ"ν-ί
νφάΥ= φηαΞ (3.7.3)

This is identical with expression (3.6.2).

THEOREM 3.7.2 Let V be the volume of a three-dimensional region

bounded by a closed regular surface S. Then for a vector field u defined in
 3.7 INTEGRAL THEOREMS FOR TENSORS 145

V and on S,

lndr-\. VndV=

where n is, as usual, the unit outward normal to S.

n®ndS (3.7.4)

Proof Let a and c be arbitrary constant vectors. Then

c· (Vu)rfFa = e-(Vu)atfK = *-(Vu)TcdV

= a· V(c · u) dV = a · (c · u)n dS
by (3.7.3);
= a · (n ® u)c dS = c · (n <g) u)ra dS
is is
. . . »
u (x) n)a dS = c · ] I u ® nrfSJa
Since c and a are arbitrary, this expression yields the result (3.7.4).

In the suffix notation, (3.7.4) reads

uudV= UiHjdS (3.7.4)'
V ' JS

Note: If we take the trace on both sides of (3.7.4), we recover the

divergence theorem (3.6.1).

THEOREM 3.7.3 (Stokes s Theorem for a Tensor) Let C be a

simple closed curve in three-dimensional space and S be an open regular
surface bounded by C. Then for a vector field defined on 5 as well as C,

At ds= (curlA)WS (3.7.5)

c is
where t is the unit tangent to C, which is assumed to be positively oriented
relative to the unit normal n to S.
Proof Let a be an arbitrary constant vector. Then

a· (curl A)rn dS =\ a · (curl A) rn dS =\ n · (curl A)a dS

is is is
= n-curl(A r a)ûf5. (3.7.6)
146 3 CALCULUS OF TENSORS

By virtue of expression (3.6.7), we have

n · curl(A r a) dS = (A r a) · t ds
s Jc

= a · (At) ds = a · At ds (3.7.7)
Jc Jc
Since a is arbitrary, expressions (3.7.6) and (3.7.7) yield the result
(3.7.5). ■

In the suffix notation, expression (3.7.5) reads

aiktkds=\ eJrsaiStrnjdS (3.7.5)'

c Js
Note: In the special case when S is closed, we have by (3.6.8),

n · curl(A r a) dS = 0 (3.7.8)

Then (3.7.5) gives

(curl A) W S = 0 or ejrsair>snjdS =0 (3.7.9)

s Js

A Particular Case In the particular case when C lies in the jc^-plane

and S is the part of this plane bounded by C, we have nx = n2 = t3 = 0 and
n3 = 1. Then expression (3.7.5)' takes the following form:

(aildxl + ai2dx2) = (ai2il - aiU2)dS (3.7.10)

c Js
This is a generalization of the Green's theorem in the plane, given by
(3.6.9).

THEOREM 3 . 7 . 4 Let C be a simple closed curve in three-dimensional

space and S be an open regular surface bounded by C. Then for a vector
field u defined on S as well as C,

t(g>uds= (nAVu)dS (3.7.11)

c js

where t is the unit tangent to C, which is assumed to be positively oriented

relative to the unit normal n to S.
 3.7 INTEGRAL THEOREMS FOR TENSORS 147

Proof Let a be a constant vector. Then

(D t ® uds a = ώ ( t ( x ) u ) a * = ώ (u-a)töfr

n x V(u · a) dS
Is
by (3.6.10);

= n x (Vu)'a dS = \ (n Λ Vu)a i/S

by (2.8.17);

(nAVu)rfS[a

Since a is arbitrary, expression (3.7.11) follows.

In the suffix notation, expression (3.7.11) reads

tiUjds = \ eirsnrujsdS (3.7.11)'

c }s
If we take the trace of expression (3.7.11), we recover Stokes's theorem
(3.6.7).

EXAMPLE 3.7.1 Let 5 be a regular surface enclosing a region of

volume V. For a vector field u and a tensor field A defined on V and on 5,
show that

(u ® divA + (Vu)AT}dV= u ® (An) dS (3.7.12)

v Js
Hence deduce expressions (3.7.1) and (3.7.4).
Solution Let a and b be arbitrary constant vectors. Employing the
divergence theorem (3.6.1) to the vector (u · a)(A r b), we get

div((u · a)(Arb)} dV =\ [(u · a)(A r b)j · n dS (3.7.13)

Using identities (2.8.16) and (3.5.41), expression (3.7.13) becomes

a· S [(Vu)A r + u <g> div A] dV b = a · M [u (x) An] dS b

Since a and b are arbitrary, this expression yields (3.7.12).

148 3 CALCULUS OF TENSORS

If u is a constant vector, then (3.7.12) becomes

u (x) div A dV = u ® An dS

Since this result is true for any u, we obtain (3.7.1). On the other hand, if
A is the identity tensor I, then (3.7.12) becomes (3.7.4). ■

EXAMPLE 3 . 7 . 2 Starting with expression (3.7.4) deduce expression

(3.6.3).

Solution For any constant vector c, expression (3.7.4) yields

(Wu-VuJ)cdV=\ ( u ® n - n®u)c</S (3.7.14)

v Js
By virtue of (3.5.44), the dual vector of (Vu - Vu r ) is curl u. Hence

(Vu - Vu r )c = (curlu) x c (3.7.15)

Also,
[u ® n - n (x) u]c = (n · c)u - (u · c)n = c x (u x n) (3.7.16)

Using (3.7.15) and (3.7.16), expression (3.7.14) becomes

c x curl u f i f F = c x (n x u) dS

Since c is arbitrary, expression (3.6.3) follows. ■

EXAMPLE 3 . 7 . 3 Let S be a regular surface enclosing a region of

volume V. For a tensor field A defined in V and on 5, show that

u x (An) dS = (2w + u x divA)rfK (3.7.17)

s ]v
Here n is the unit outward normal to S and w is the dual vector of the skew
part of A(Vu) r . Deduce that

x x (An) dS = \ (2ξ + x x div A) dV (3.7.18)

where ξ is the dual vector of skw A.

 3.7 INTEGRAL THEOREMS FOR TENSORS 149

Solution Let a be an arbitrary constant vector. Then

a x \ u x (An) dS = \ [a x [u x (An)}] dS

l· = (u ® (An) - (An) <g> u}arfS (3.7.19)

From expression (3.7.12), we have

[u <g) (An)} dS = {u (g) div A + (Vu)A'} dV (3.7.20)

so that (on taking the transpose)

{(An) <x) u) dS = ((div A) ® u + A(Vu)'} </F (3.7.21)

Using (3.7.20) and (3.7.21) in (3.7.19), we get

ax u x (An)dS

= [{u ® (div A) - (div A) <g> u}a - 2 skw{A(Vu)r}a] dV

= a x {u x div A + 2w) dS

Since a is arbitrary, (3.7.17) follows.

If we set u = x in (3.7.17) and recall that Vx = I, we readily get
(3.7.18). ■

EXAMPLE 3 - 7 . 4 Let S be a regular open surface bounded by a simple

closed curve C. For a vector field u and a tensor field A defined on S as well
as C, show that

(u (x) At) ds = [u <g> (curl A) 7 n + (n Λ Vu)'A 1 ] dS (3.7.22)

c }s
where t and n are as defined in expression (3.7.5). Hence deduce (3.7.5) and
(3.7.11).
Solution Let a and b be arbitrary constant vectors. Employing the
Stokes's theorem (3.6.7) to the vector (u. a)(A r b), we get

{(u · a)(Arb)} · t ds =\ curl{(u · a)(A r b)} · n dS (3.7.23)

c Js
150 3 CALCULUS OF TENSORS

Using the identities (2.8.16) and (3.5.42), expression (3.7.23) becomes

(u <g> At) ds b = a · [u ® (curl Α)Γη + (η Λ Vu)rAr] dS b

Since a and b are arbitrary, this expression yields (3.7.22).

If u is a constant vector, expression (3.7.22) becomes

u ® (D Aids = u <g) I (curlAyndS

jc js
Since this result is true for any u, expression (3.7.5) follows.
On the other hand, if A is the unit tensor, expression (3.7.22) becomes

(u®t)ds = (nAVufdS
c }s
from which (3.7.11) is immediate. ■

EXAMPLE 3.7.5 Let S be a regular open surface bounded by a simple

closed curve C. For a tensor field A defined on S and C, show that

(u X At) ds = [2w + u X (curl A)rnj dS (3.7.24)

where t and n are as defined in (3.7.5) and w is the dual of the skew part of
A(n Λ Vu). Hence deduce that

(x x At)ds = \ (2i + x x (curlA)'n)dS (3.7.25)

c Js
where ξ is the dual vector of skw(n Λ A).
Solution Let a be an arbitrary constant vector. Then

a x ( D ( u x A t ) ^ = ( p ( u ® (At) - (At) ® u}a ds

Using (3.7.22) and its transpose, this expression takes the form

a x Φ (u x At) ds

= \ [{u (x) (curl A)7!! - (curl A)rn <g) u}a - 2{skw A(n Λ Vu)}a] dS

=ax [u x (curl A)rn + 2w} dS

Since a is an arbitrary vector, result (3.7.24) follows.

 3.8 EXERCISES 151

Setting u = x in (3.7.24) and recalling that Vx = I and Α(ηΛΐ) =

-(n Λ A)T, we get (3.7.25). ■

3-8
EXERCISES

1- Verify the identities (3.2.3) to (3.2.7).

2 . For any vector u = u(t) with magnitude u, show that
du du
u -—- = u —
dt dt
3 . If u(/) is a unit vector, show that
du du
ux —
dt
= —
dt
4. If u(0 and v(/) are such that du/dt = w x u and dy/dt = w x v, where w is a
constant vector, show that

— (u x v) = (u (x) v - v (x) u)w

dt
5. Verify the identities (3.2.8) to (3.2.10) and (3.2.12).
6. If Q(0 is an orthogonal tensor, show that Q(dQ/dt)T and QT(dQ/dt) are skew
tensors.
7. If A(/) is an invertible tensor, show that

^(detA) = ( d e t A ) t r ^ A - M

8. If ui = 2xlx2, u2 = -*2*3> u3 = 3x 1 x 3 , find the matrices of the following:

(i) uitj (ii) UJJ
(iii) (uu + Ujti) (iv) (Uij - ujti)

9. Verify the identities (3.4.3) and (3.4.4).

1 0 . Show that the directional derivatives of φ along e, are φ ,.
1 1 . Find the directional derivative of φ = *ιΧ2*3 along the tangent to the curve
given by the parametric equations χλ = t,x2 = t2, x3 = t3, at the points (1, 1, 1) and
(-1,1,-1).
1 2 . Find the direction along which the directional derivative of the function
φ = X\X2Xi is maximum at the point (1, 1, 1). Also, find the maximum directional
derivative.
152 3 CALCULUS OF TENSORS

1 3 . If φ = xnx + x" + x", show that x · V</> = Αΐφ.

1 4 . I f / = /(r), show that Vf = l/'(r)/r)x, r ^ 0.
1 5 . If A is a constant tensor, show that V(Ax · x) = (A + A r )x.
1 6 . Verify the identities (3.4.19).
1 7 . Show that
(i) div u = ek · u k (ii) curl u = ek x u Λ
1 8 . Prove the identities (3.4.25) and (3.4.27) by using the suffix notation.
1 9 . If u = curl v and v = curl u, show that
(i) div(curl u x curl v) = u2 - v2
(ii) V2u = - u , V2v = - v
2 0 . Show that V2|x|" = 0 only if n = 0 or - 1 .
2 1 . Find φ = φ(τ), given that div(0x) = 0.
2 2 . Find φ = Φ(Γ), given that V2(r</>) = 0.
2 3 . If / = /(r), show that V2{/(r)) = f\r) + (2/r)/'(r), r * 0. Deduce that
V 2 (l/r) = 0.
2 4 . Show that ν\φψ) = φν2ψ + ψν2φ + 2V0 · νψ.
2 5 . Show that (u · V)x = u.
2 6 . Show that curl u = y[curl((curl u) x x) - (x · V) curl u].
2 7 . Show that rn\ is curl free for any n, but divergence free only if n = - 3 .
2 8 . If c is a constant vector, show that /(r)c x x is divergence free.
2 9 . If u · curl v = v · curl u, show that u x v is divergence free and that
curl curl(u x v) = V2(v x u)
3 0 . Show that V(v · V(l/r)) + curl(v x V(l/r)) = 0, r * 0.
3 1 . If the curl of a Beltrami vector is again a Beltrami vector, show that the
abnormality factor is independent of x.
3 2 . A vector u is said to be a complex lamellar vector if u · curl u = 0. If u is a
complex lamellar vector, show that for any scalar field φ9 the vector φη is also a
complex lamellar vector.
3 3 . Show that Vu = urstr ® e s .
3 4 . If Vu = -(Vu) r , show that VVu = 0.
3 5 . Show that (v Λ Vu)w = v x (Vu)rw.
3 6 . Show that V(</>u) = u ® V</> + 0Vu.
 3.8 EXERCISES 153

3 7 . Verify the identities (3.5.4) to (3.5.6) and (3.5.22).

3 8 . If u = x]tx + x\t2 + jc|e 3 , find Vu and (Vu)u.
3 9 . If c is a constant vector, show that (V2A)c = V2(Ac).
4 0 . Prove the following identities:
(i) div(div A) = tr V(div A)
(ii) div(Vur) = div[(div u)I]
(Hi) div(0I) = V0
(iv) div(Au) = u · div AT + tr[Ar(Vu)]
(v) div(u ® v) = (div v)u + (Vu)v
(vi) div[0(Au)] = 0((u · div A 7 ) + (A r · Vu)j + V0 · Au
(vii) div[(Vu)u] = Vu · Vu r + u · V(div u)
(viii) div[(Vu)u - (div u)u] = Vu · Vu r - (div u)2
(ix) div[(u <g) v)w] = (div u)(v · w) + u · V(v · w)
(x) curl(0I) = -[curl«>I)] r
(xi) tr[V2A + VV(tr A)] = 2V2(tr A)
4 1 . L e t / be a scalar field or a component of a vector or a tensor field defined in
a region R. If \vf dV = 0 for every volume V in /?, show t h a t / = 0 in R.
4 2 . Calculate the circulation of u = x\tx - Jt1Jt2e2 round the ellipse xl = a cos τ,
x2 = b sin τ, x 3 = 0, 0 < τ <2π.
4 3 . By using the divergence theorem or Stokes's theorem, prove the expressions
(3.6.2), (3.6.4), (3.6.5), (3.6.9), (3.6.10) and (3.6.12).
4 4 . By using the divergence theorem, prove the following:

(i) | <t>-!-dS= W>V> + ν ψ · νψ)αν

s on ]v

(ii) U^-w^\dS=\ (0VV - ψν2φ)αν

4 5 . If u is everywhere normal to a closed regular surface S bounding V, show that

curluöfK= 0
v
4 6 . If v is as defined by (3.6.14), prove that V2v = u.
4 7 . Verify that the vector w present in Helmholtz's representation (3.6.15) obeys
the equation
1 Γ curl u(x)
. ■ ,dV
An
1 54 3 CALCULUS OF TENSORS

48. Deduce representations (3.6.6) and (3.6.13) from Helmholtz's representation

(3.6.15).
49. By starting with the expression (3.7.17), deduce expression (3.6.3).
50. By starting with the expression (3.7.25), derive expression (3.6.12).
51. If Fis the volume of a region bounded by a closed regular surface S, show that

x <g) n dS = V\
s
52. Prove that

aijtidV= dénias
v Js
53. Let u be a vector field defined in a region of volume V enclosed by a regular
closed surface S. Also, let a be a constant vector and φ = φ(χ3) is a scalar function
defined in V. If u is divergence free in V and u = 0 on 5, prove the following:

(i) (u -curl curl u) d V =\ |curlu|2</F

(ii) (u x a)· V(a · curl u)rfK= - |a-curlu|2ûfK

(Hi) (u · a)V2(u · a) dV = - |V(u · a)|2 dV

(iv) W> > 3 u)-V W 3 rfK=- φ,33«3 dV

54. Let h be a vector field defined in a region of volume V enclosed by a regular

closed surface S. If h is divergence free in V and n x curl h = 0 on 5, show that

h · curl curl h dV = |curlh|2rfF

v ]v

55. Show that

[(curl u) x v + (curl v) x u + u(div v) + v(div u)] dV

= l [(u (g) v + v (x) u)n - (u · v)n] dS