Chapter 2

Tensor Analysis

2.1 Introduction to Tensors

Tensors are a powerful mathematical tool that is used in many areas in engineering
and physics including general relativity theory, quantum mechanics, statistical
thermodynamics, classical mechanics, electrodynamics, solid mechanics, and fluid
dynamics. Laws of physics and physical invariants must be independent of any
arbitrarily chosen coordinate system. However, the tensor components describing
these characteristics heavily depend on the coordinate bases and therefore change
as the coordinate system varies in the considered spaces. Before going into detail,
we provide less-experienced readers with some examples.
Different tensors are listed in Table 2.1, which can be expressed in differently
chosen bases for any curvilinear coordinate. Using Einstein summation conven-
tion, the notation can be shortened. Note that Einstein summation convention is
only valid for the same indices in the lower and upper positions. The relating
contravariant or covariant tensor components can be expressed in the covariant or
contravariant bases (cf. Appendix E). The tensor order is determined by the
number of the coordinate basis. Thus, the component of a first-order tensor has
only one dummy index i relating to a single basis. In the case of a second-order
tensor, its component contains two dummy indices i and j relating to double bases.
Similarly, the component of an N-order tensor has N dummy indices relating to
N bases.
The dummy indices (inner indices) are the repeated indices running from the
values from 1 to N in Einstein summation convention. The free index (outer index)
can be independently chosen for any value from 1 to N, that is, for any tensor
component in the particular coordinate, as shown in the below example. Note that
the dimensions of the dummy and free indices must be the same value of the space
dimensions.

H. Nguyen-Schfer and J.-P. Schmidt, Tensor Analysis and Elementary 35

Differential Geometry for Physicists and Engineers, Mathematical Engineering 21,
DOI: 10.1007/978-3-662-43444-4_2,  Springer-Verlag Berlin Heidelberg 2014
36 2 Tensor Analysis

Table 2.1 Tensors in general curvilinear coordinates

Type Component Basis Tensor
First-order tensors 2 RN T i ; Ti gi , gi T1 T i gi ; Ti gi
i j
Second-order tensors T ij ; Tij; Tji ; Tij gi, g , gj, g T2 T ij gi gj ; Tij gi g j ;
2 RN  RN Tji gi g j ; Tij gi gj
Third-order tensors T ijk ; Tijk; Tjik ; Tikj gi, gi, gj, gj, T3 T ijk gi gj gk ; Tijk gi g j gk ;
2 RN  RN  RN gk, gk T ik g g g j ; T j gi gk g
j i k ik j
N-order tensors T ijk...n ; Tijk...n ; gi, gi, gj, gj, TN T ijk...n gi gj gk...
2 RN      RN j
Tjik...n ; Tik...n: gk, gn ; Tjik...n gi gk... gn g j
gk, ,
gn, gn

N X
X N
T T ij gi gj  T ij gi gj ; i; j: dummy indices
j1 i1
X
N
T j T ij gi  T ij gi ; i: dummy; j: free index :
i1

2.2 Definition of Tensors

The definition of tensors is based on multilinear algebra with a multilinear map.

We consider the real vector spaces U1,, Un and their respective dual vector
spaces V1,, Vm. Each of their vector spaces belongs to the finite N-dimensional
space RN, the image vector space W, to the real space R. A mixed tensor of type
(m, n) can be defined as a multilinear functional T that maps an (m + n) tuple of
vectors of the vector spaces U and V into W (Fecko 2011) (Fig. 2.1):

T : U1      Un  V1      Vm ! W
RN  {z
|    RN}  R N
|    RN} ! R
 {z
2:1
n copies m copies

u1 ; . . .; un ; v1 ; . . .; vm ! Tu1 ; . . .; un ; v1 ; . . .; vm 2 R:

Mapping the multilinear functional T of the tensor type (m, n) to the contra-
variant basis {gim} of U and covariant basis {gjn} of V, one obtains its images in
W , R. These images are called the components of the (m + n)-order mixed
tensor T with respect to the relating bases:
i1...im
Tj1...jn  Tgj1 ; . . .; gjn ; gi1 ; . . .; gim 2 R 2:2
2.2 Definition of Tensors 37

( u1 ,...,u n ; v1 ,..., v m ) T (u1 ,..., u n ; v1 ,..., v m )

R N ... R N R N ... R N R
n copies m copies

Fig. 2.1 Multilinear functional T

Thus, the (m + n)-order mixed tensor T can be expressed in the covariant and
contravariant bases of the respective vector spaces V and U. In total, the (m + n)-
order tensor T has N(m+n) components, as shown in Eq. (2.3).
i1...im
T Tj1...jn gi1 . . .gim gj1 . . .gjn ;
T2R N
   RN}  R
 {z N
   RN}
 {z 2:3
| |
n copies m copies

In the case of covariant and contravariant tensors T, the dual vector spaces
V and real vector spaces U are omitted in Eq. (2.1), respectively.
n-order covariant tensors:

T Tj1...jn gj1 . . . gjn 2 U1      Un 2:4

m-order contravariant tensors:

T T il...im gil . . .gim 2 V1      Vm 2:5

An Example of a Second-Order Covariant Tensor

An arbitrary vector v can be expressed in the covariant basis gk in the
N-dimensional vector space V as

v vk gk for k 1; 2; . . .; N 2:6

Applying the bilinear mapping T to the vector v and using the Kronecker delta,
one obtains its mapping image Tv. Straightforwardly, this is a tensor of one lower
order compared to the mapping tensor T.
38 2 Tensor Analysis

Tv  Tij gi g j :vk gk
Tij vk gi :gk g j
Tij vk dik g j for i k 2:7a
k j
Tkj v g for j; k 1; 2; . . .; N
 Tj g j for j 1; 2; . . .; N

whereas the second-order covariant tensor T can be expressed as

T Tij gi g j for i; j 1; 2; . . .; N;
T 2 RN  RN :

Note that in the case of a three-dimensional vector space R3 (N = 3), there are
nine covariant components Tij. The number of the tensor components can be
calculated by Nn (32 = 9), in which n is the number of indices i and j (n = 2).
Obviously, that the mapping image Tv is also a tensor of one lower order
compared to the tensor T. The covariant tensor component T*j can be calculated by

Tj  Tkj vk gj :Tv : 2:7b

2.3 Tensor Algebra

2.3.1 General Bases in General Curvilinear Coordinates

The vector r can be written in Cartesian coordinates of Euclidean space E3, as

displayed in Fig. 2.2.

r xi ei 2:8

The differential dr results from Eq. (2.8) in

or i
dr ei dxi dx 2:9
oxi

Using the differentiation chain rule, the orthonormal bases ei of the coordinates
xi are defined by
2.3 Tensor Algebra 39

Fig. 2.2 Bases of general u3

curvilinear coordinates in the g3
space E3
g3
x3
g1
P g2
e3
g2
u2
0 e u1 g1
e1 2
x2
x1

or or ou j
ei 
oxi ouj oxi 2:10
ou j
 gj i for j 1; 2; . . .; N
ox

Analogously, the bases of the curvilinear coordinates ui can be calculated in the

curvilinear coordinate system of EN

or or oxk
gj  j
k j
ou ox ou 2:11
oxk
ek j for k 1; 2; . . .; N
ou

The curvilinear coordinate ui are functions of the coordinate xi; the covariant
bases in Eq. (2.11) can be calculated using the differentiation chain rule.

or or oxi
gj
ou j oxi ou j
oxi 2:12
ei j
ou
 ei xij for i 1; 2; . . .; N

Thus, the curvilinear basis gj can be written in a linear combination of the

orthonormal basis ei according to Eq. (2.12). The derivative xij is called the shift
tensor between the orthonormal and curvilinear coordinates.
Generally, the basis gi of the curvilinear coordinate ui can be rewritten in a
linear combination of the basis gj of other curvilinear coordinate uj. The deriv-
ative uji is defined as the shift tensor between both curvilinear coordinates.

or ou0j
gi
ou0j oui 2:13
ou0j
g0 j i  g0 j u0ji for j 1; 2; . . .; N
ou
40 2 Tensor Analysis

In the curvilinear coordinate system (u1, u2, u3) of Euclidean space E3, its basis
is generally non-orthogonal and non-unitary (non-orthonormal basis); that is, the
bases are not mutually perpendicular and their vector lengths are not equal to one
(Simmonds 1982; Klingbeil 1966; Nayak 2012). In this case, the curvilinear
coordinate system (u1, u2, u3) has three covariant bases g1, g2, and g3 and three
contravariant bases g1, g2, and g3 at the origin P, as shown in Fig. 2.2. Generally,
the origin P of the curvilinear coordinates could move everywhere in Euclidean
space. Therefore, the bases of the curvilinear coordinates only depend on the
respective origin P. For this reason, the curvilinear bases are not fixed in the whole
curvilinear coordinates like in Cartesian coordinates.
The vector r of the point P(u1, u2, u3) can be written in covariant and contra-
variant bases.

r u1 g1 u2 g2 u3 g3
2:14
u1 g1 u2 g2 u3 g3

where
u1, u 2, u3 are the vector contravariant components of the coordinates (u1, u2, u3);
g1, g 2, g3 are the covariant bases of the coordinate system (u1, u2, u3);
u1, u 2, u3 are the vector covariant components of the coordinates (u1, u2, u3);
g1, g 2, g3 are the contravariant bases of the coordinate system (u1, u2, u3).

The covariant base gi is defined by the tangential vector to the corresponding

curvilinear coordinate ui for i = 1, 2, 3. Both bases g1 and g2 generate a tangential
surface to the curvilinear surface (u1u2) at the considered origin P. Note that the
basis g1 is not perpendicular to the bases g2 and g3. However, the contravariant
basis g3 is perpendicular to the tangential surface (g1g2) at the origin P. Generally,
the contravariant basis (gk) results from the cross product of the other covariant
bases (gi 9 gj).

a gk gi  gj for i; j; k 1; 2; 3 2:15

where a is a scalar factor.

Multiplying Eq. (2.15) by the covariant basis gk, the scalar factor a can be
calculated as

agk : gk a dkk a gi  gj : gk
  2:16
) a gi  gj :gk  gi ; gj ; gk

The scalar factor a equals the scalar triple product that is given in (Klingbeil
1966):
2.3 Tensor Algebra 41

   
a  g1 ; g2 ; g3  gi  gj :gk gk  gi : gj gj  gk : gi
 1  1  1
 g11 g12 g13 2  g31 g32 g33 2  g21 g22 g23 2
     
     
 g21 g22 g23   g11 g12 g13   g31 g32 g33  2:17
     
 g31 g32 g33   g21 g22 g23   g11 g12 g13 
q
p
det gij  g J

where J is defined as the Jacobian, as given in

 1 
 ox ox1 ox1 

 ou1 ou2 ou3 
i j
ox ox ox k  2
 ox ox2 ox2 
J  eijk 1 2 3   2:18
ou ou ou  ou1 ou2 ou3 
 3
 ox ox3 3
ox 
 1 
ou ou2 ou3

Thus,
eijk eijk
gk p gi  gj gi  gj 2:19
g J

where eijk is the Levi-Civita permutation symbols in Eq. (A.5), cf. Appendix A.
According to Eq. (2.19), the contravariant basis gk is perpendicular to both
covariant bases gi and gj. Additionally, the contravariant basis gk is chosen such
that the vector length of the contravariant basis equals the inversed vector length of
its relating covariant basis.
Therefore,

gi  gj :gi
gk : gi   dki 2:20
gi ; gj ; gk

As a result, the relation between the contravariant and covariant bases is given
in the general curvilinear coordinate system (u1,, uN).
(
gi : gk gk :gi dki for i; k 1; 2; . . .; N
2:21
gi : gk gk :gi 6 dki for i; k 1; 2; . . .; N

The basis gi is called dual to the basis gj (Itskov 2010) if

gi : g j dij for i; j 1; 2; . . .; N 2:22

where dji is the Kronecker delta.

42 2 Tensor Analysis

Fig. 2.3 Covariant bases of

orthogonal cylindrical
x3 (r,,z) (u 1,u 2,u 3):
coordinates u1 r ; u 2 ; u 3 z

g3

z
P g2

e3 R
g1
e2 x2
0
e1

r

x1

Let {g1, g2,, gN} be a covariant basis of the curvilinear coordinates {ui}, the
contravariant basis {g1, g2,, gN}, the dual basis to the covariant basis, can be
written in the matrix formulation.
2 3
g1
6 g2 7
6 7
G g1 g2 : gi gN ; G1 6 7 1
6 : 7)G GI 2:23
4 gj 5
gN

where gj is the jth row vector of G-1; gi is the ith column vector of G.
The covariant and contravariant bases (dual bases) of the orthogonal cylindrical
and spherical coordinates are computed in the following section.
(a) Orthogonal Cylindrical Coordinates
The cylindrical coordinates (r, h, z) are orthogonal curvilinear coordinates in
which the bases are mutually perpendicular but not unitary. Figure 2.3 shows a
point P in the cylindrical coordinates (r, h, z) embedded in the orthonormal
Cartesian coordinates (x1, x2, x3). However, the cylindrical coordinates change as
the point P varies.
The vector OP can be written in Cartesian coordinates (x1, x2, x3):

R r cos he1 r sin he2 z e3

2:24
 x1 e1 x2 e2 x3 e3
2.3 Tensor Algebra 43

where
e1, e2, and e3 are the orthonormal bases of Cartesian coordinates;
h is the polar angle.

To simplify the formulation with Einstein symbol, the coordinates of u1, u2, and
u are used for r, h, and z, respectively. Therefore, the coordinates of P(u1, u2, u3)
3

are given in Cartesian coordinates:

8 1 1 29
< x r cos h  u cos u >
> =
Pu1 ; u2 ; u3 x2 r sin h  u1 sin u2 2:25
>
: 3 >
;
x z  u3

The covariant bases of the curvilinear coordinates are computed from

oR oR ox j ox j
gi i
j : i ej i for j 1; 2; 3 2:26
ou ox ou ou

The covariant basis matrix G yields from Eq. (2.26):

G g1 g 2 g 3 
0 1 1
ox ox1 ox1
B ou1 ou2 ou3 C 0 1
B C cos h r sin h 0
B ox2 ox2 ox2 C B C 2:27
BB C
C @ sin h r cos h 0A
B ou1 ou2 ou3 C
@ 3 A 0 0 1
ox ox3 ox3
ou1 ou2 ou3

The determinant of G is called the Jacobian J.

 1 
 ox ox1 ox1 
  
 ou1 ou3   cos h
 2 ou2
 r sin h 0 
 ox ox2 ox 2
jGj  J    sin h r cos h 0  r 2:28
 ou1 ou2 ou   0
3
 3 0 1
 ox ox3 ox3 
 1 
ou ou2 ou3

The relation between the covariant and contravariant bases yields from
Eq. (2.22):

gi : gj dij Kronecker delta 2:29

Thus, the contravariant basis matrix G-1 results from the inversion of the
covariant basis matrix G, as given in Eq. (2.27).
44 2 Tensor Analysis

2 3 0 1
g1 1
r cos h r sin h 0
G1 4 g2 5 @  sin h cos h 0A 2:30
r
g3 0 0 r

The calculation of the determinant and inversion matrix of G will be discussed

in the following section.
According to Eq. (2.27), the covariant bases can be denoted as
8
< g1 cos h e1 sin h e2 0: e3 ) jg1 j 1
>
g2 r sin h e1 r cos h e2 0: e3 ) jg2 j r 2:31
>
:
g3 0: e1 0::e2 1: e3 ) jg3 j 1

The contravariant bases result from Eq. (2.30).

8 1  
>
> g cos h e1 sin h e2 0: e3 ) g1  1
>
<

sin h cos h   1
2
g  e1 e2 0: e3 ) g2  2:32
>
> r r r
>
: 3  3

g 0 :e1 0: e2 1: e3 ) g 1

Not only the covariant bases but also the contravariant bases of the cylindrical
coordinates are orthogonal due to

gi :g j g j :gi dij
gi :gj 0 for i 6 j;
gi :g j 0 for i 6 j:

(b) Orthogonal Spherical Coordinates

The spherical coordinates (q, /, h) are orthogonal curvilinear coordinates in
which the bases are mutually perpendicular but not unitary.
The spherical coordinates (q, /, h) are orthogonal curvilinear coordinates in
which the bases are mutually perpendicular but not unitary. Figure 2.4 shows a
point P in the spherical coordinates (r, h, z) embedded in the orthonormal
Cartesian coordinates (x1, x2, x3). However, the spherical coordinates change as the
point P varies.
The vector OP can be written in Cartesian coordinates (x1, x2, x3):

R q sin / cos h e1 q sin / sin h e2 q cos / e3

2:33
 x1 e1 x2 e2 x3 e3
2.3 Tensor Algebra 45

Fig. 2.4 Covariant bases of x3 (,, ) (u1,u 2,u 3):

orthogonal spherical sin
u1 ; u 2 ; u 3
coordinates g1

g3
P

e3
e2 g2 x2
0
e1

cos
x1

where
e1, e2, and e3 are the orthonormal bases of Cartesian coordinates;
/ is the equatorial angle;
h is the polar angle.

To simplify the formulation with Einstein symbol, the coordinates of u1, u2, and
u are used for q, /, and h, respectively. Therefore, the coordinates of P(u1, u2, u3)
3

are given in Cartesian coordinates:

8 1 1 2 39
< x q sin / cos h  u sin u cos u >
> =
Pu1 ; u2 ; u3 x2 q sin / sin h  u1 sin u2 cos u3 2:34
>
: 3 >
;
x q cos /  u1 cos u2

The covariant bases of the curvilinear coordinates are computed from

oR oR ox j ox j
gi : e j for j 1; 2; 3 2:35
oui ox j oui oui

Thus, the covariant basis matrix G can be calculated from Eq. (2.35).
0 1
ox1 ox1 ox1
B ou1 ou2 ou3 C
B C
B ox2 ox2 ox2 C
B
G g1 g 2 g 3  B C
C
B ou1 ou2 ou3 C
@ 3 3 3A
ox ox ox 2:36
0 ou1 ou2 ou3 1
sin / cos h q cos / cos h q sin / sin h
B C
@ sin / sin h q cos / sin h q sin / cos h A
cos / q sin / 0
46 2 Tensor Analysis

The determinant of the covariant basis matrix G is called the Jacobian J.

 
 sin / cos h q cos / cos h q sin / sin h 
 
 
jGj  J  sin / sin h q cos / sin h q sin / cos h 
  2:37
 cos / q sin / 0 
q2 sin /

Similarly, the contravariant basis matrix G-1 is the inversion of the covariant
basis matrix G.
0 1
2 13 q sin / cos h q sin / sin h q cos /
g 1 B cos / cos
h cos / sin h  sin / C
G1 4 g2 5 B @ sin h   C
A 2:38
3 q  cos h 0
g sin /
sin /

The matrix product G-1  G must be an identity matrix according to Eq. (2.23).
0 1
q sin / cos h q sin / sin h q cos /
1B
B cos / cos h cos / sin h
C
 sin / C
G1 G B

C
q@ sin h cos h A
 0
sin / sin /

0 1 0 1
sin / cos h q cos / cos h q sin / sin h 1 0 0
B C B C
:@ sin / sin h q cos / sin h q sin / cos h A @ 0 1 0A  I
cos / q sin / 0 0 0 1
2:39

According to Eq. (2.36), the covariant bases can be written as

g1 sin / cos h e1 sin / sin h e2 cos / e3 ) jg1 j 1

g2 q cos / cos h e1 q cos / sin h e2  q sin / e3 ) jg2 j q 2:40
g3 q sin / sin h e1 q sin / cos h e2 0:e3 ) jg3 j q sin /

The contravariant bases result from Eq. (2.38).

 
g1 sin / cos h e1 sin / sin h e2 cos / e3 ) g1  1

1 1 1   1
2
g cos / cos h e1 cos / sin h e2  sin / e3 ) g2 
q q q q

1 sin h 1 cos h   1
g3  e1 e2 0:e3 ) g3 
q sin / q sin / q sin /
2:41
2.3 Tensor Algebra 47

Not only the covariant bases but also the contravariant bases of the spherical
coordinates are orthogonal due to

gi :g j g j :gi dij
gi :gj 0 for i 6 j;
gi :g j 0 for i 6 j:

2.3.2 Metric Coefficients in General Curvilinear Coordinates

The covariant basis vectors g1, g2, and g3 to the general curvilinear coordinates
(u1, u2, u3) at the point P can be calculated from the orthonormal bases (e1, e2, e3)
in Cartesian coordinates xj = xj(ui), as shown in Fig. 2.2.

or or oxk
gi  i
k: i
ou ox ou 2:42
oxk
ek i for k 1; 2; 3
ou

The covariant metric coefficients gij are defined as

or or
gij  gi :gj gj :gi  gji
oui ou j
oxk oxl oxk oxl
i j ek el i j dkl 2:43
ou ou ou ou
oxk oxk
i j
ou ou

Similarly, the contravariant metric coefficients gij can be denoted as

gij  gi :g j g j :gi gji 2:44

Furthermore, the contravariant basis can be rewritten as a linear combination of

the covariant bases.

gi Aij gj 2:45

According to Eq. (2.44) and using Eqs. (2.21) and (2.45), the contravariant
metric coefficients can be expressed as
48 2 Tensor Analysis

gik  gi :gk Aij gj :gk

2:46
Aij dkj Aik

Thus,

gi gij gj for j 1; 2; 3 2:47

Analogously, one obtains the covariant basis

gk gkl gl for l 1; 2; 3 2:48

The mixed metric coefficients can be defined by

 
gik  gi :gk gij gj :gk
    
gij gj : gkl gl gij gkl gj :gl
2:49a
gij gkl dlj gij gkj
dik

Thus,

gij gkj gkj gij dik 2:49b

Therefore, the contravariant metric tensor is the inverse of the covariant metric
tensor.

gij gkj gkj gij dik , M1 M MM1 I 2:50

where M-1 and M are the contravariant and covariant metric tensors.
Thus,
2 32 3
g11 g12 : g1N g11 g12 : g1N
6 g21 g22 : g2N 7 6g g22 : g2N 7
6 7 6 21 7
M1 M 6 7:6 7
4 : : gij : 54 : : gij : 5
N1 N2 NN gN1 gN2 : gNN
g g : g
2 3 2:51
1 0 : 0
    6 60 1 : 077
gij gkj dik 6 7I
40 : 1 05
0 0 : 1

According to Eqs. (2.42) and (2.49a), the contravariant bases of the curvilinear
coordinates can be derived as
2.3 Tensor Algebra 49

ox j
dki  gk :gi gk : i ej )
j

ou
k k ox ox j
di :ej g : i ej :ej gk : i ) 2:52
ou ou
i k
ou ou
gk dki j ej j ej
ox ox

Thus,

ou j
gj ei for i 1; 2; 3 2:53
oxi

Generally, the covariant and contravariant metric coefficients of the general

curvilinear coordinates have the following properties:
8
>
> g gij gi :gj 6 dij cov. metric coefficient
< ji
gji gij gi :g j 6 dij contrav. metric coefficient 2:54
>
>
: j j j
gi gi :g di mixed metric coefficient

Using Eqs. (2.42) and (2.53), one obtains the Kronecker delta

oxk ou j oxk ou j l
dij gi :g j e k :e l d
oui oxl oui oxl k 2:55a
k j j
ox ou ou
i k i:
ou ox ou

The Kronecker delta is defined by

0 for i 6 j
dij  2:55b
1 for i j

As an example, the covariant metric tensor M in the cylindrical coordinates

results from Eq. (2.31).
2 3 2 3
g11 g12 g13 1 0 0
M 4 g21 g22 g23 5 4 0 r2 05 2:56a
g31 g32 g33 0 0 1

The contravariant metric coefficients in the contravariant metric tensor M-1 are
calculated from inverting the covariant metric tensor M.
50 2 Tensor Analysis

2 3 2 3
g11 g12 g13 1 0 0
M1 4 g21 g22 g23 5 4 0 r 2 05 2:56b
g31 g32 g33 0 0 1

Analogously, the covariant metric tensor M in the spherical coordinates results

from Eq. (2.40).
2 3 2 3
g11 g12 g13 1 0 0
M 4 g21 g22 g23 5 4 0 q2 0 5 2:57a
2
g31 g32 g33 0 0 q sin /

The contravariant metric coefficients in the contravariant metric tensor M-1 are
calculated from inverting the covariant metric tensor M.
2 3 2 3
g11 g12 g13 1 0 0
M1 4 g21 g22 g23 5 4 0 q2 0 5 2:57b
g31 g32 g33 0 0 q sin /2

2.3.3 Tensors of Second Order and Higher Orders

Mapping an arbitrary vector x [ RN by a linear functional T, one obtains its image

vector y = Tx (Simmonds 1982; Klingbeil 1966).

T : RN ! RN
 
T : x ! y Tx  T: x j gj x j T:gj 2:58
 
x j Tik gi gk :gj x j Tij dkj gi Tik xk gi

where T is a second-order tensor [ RN 9 RN.

It is obvious that the image vector y = Tx is a tensor of one lower order
compared to the tensor T that can be considered as a linear operator.
The second-order tensor T can be generated from the tensor product (dyadic
product) of two vectors u and v, as denoted in Eq. (2.60).
Let u and v be two arbitrary vectors, they can be written in the covariant and
contravariant bases as
(
u ui gi ui gi 2 RN ;
2:59
v v j gj v j g j 2 R N

The tensor product of two vectors u  v results in the second-order tensor T.

2.3 Tensor Algebra 51

T : u; v 2 RN ! T  u  v 2 RN  RN
) T ui v j gi gj ui v j gi gj  T ij gi gj 2:60
i j i j i j
) T ui vj g g ui vj g g  Tij g g

Note that the terms gigj and gigj are called the covariant and contravariant basis
tensors, respectively. Hence, they are not the same notations as the covariant and
contravariant metric coefficients gij and gij, respectively.

gi gj 6 gi :gj  gij
gi g j 6 gi :g j  gij

Similarly, one obtains the properties of the mixed basis tensors:

gi gj 6 gi :gj  gij dij

gi g j 6 gi :g j  gij dij

Each of the covariant and contravariant tensor components Tij and Tij contains
nine independent elements (N2 = 9) in a nine-dimensional tensor space R3 9 R3
in a three-dimensional space (N = 3).

T ij ui v j ;
2:61
Tij ui vj

The basis gj of the general curvilinear coordinates is mapped by the linear

functional T in Eq. (2.58) into the image vector Tj that can be written according to
Eq. (2.2) as

Tj  T:gj 2:62

Each vector Tj can be expressed in a linear combination of the contravariant

basis gi as

Tj Tij gi 2:63

where Tij is the covariant tensor component of the second-order tensor T.

Multiplying Eq. (2.63) by the covariant basis gi and using Eq. (2.62), the
covariant tensor component Tij results in
   
Tij gi :gi T:gj :gi gi :T:gj
2:64
) Tij dii Tij gi :T:gj

Equation (2.64) can be written in the contravariant bases gi and gj as follows:

52 2 Tensor Analysis

 
Tij gi g j gi :T:gj gi g j
   
gi :gi T g j :gj dii Tdjj T 2:65
) T Tij gi g j

Similarly, the vector Tj is formulated in a linear combination of the covariant

basis gi.

T j T:g j T ij gi 2:66

in which Tij is the contravariant component of the second-order tensor T.

Multiplying Eq. (2.66) by the contravariant basis gi, the contravariant tensor
component Tij can be computed as
   
T ij gi :gi T:g j :gi gi :T:g j
2:67
) T ij dii T ij gi :T:g j

Similarly, Eq. (2.67) can be written in the covariant bases gi and gj

   
T ij gi gj gi :T:g j gi gj
   
gi :gi T gj :g j dii Tdjj T 2:68
ij
) T T gi gj

Alternatively, the vector component can be rewritten in a linear combination of

the mixed tensor component.

T j T:g j Ti:j gi 2:69

Multiplying Eq. (2.69) by the covariant basis gi, the mixed tensor component Tji.
can be calculated as
   
Ti:j gi :gi T:g j :gi gi :T:g j
2:70
) Ti:j dii Ti:j gi :T:g j

Note that in Eq. (2.70), the dot after the lower index indicates the position of
the basis of the upper index locating after the tensor T. In this case, the tensor T is
located between the lower basis gi and upper basis gj (Itskov 2010; Nayak 2012;
Oeijord 2005).
Equation (2.70) can be written in the covariant and contravariant bases gj and gi
as follows:
2.3 Tensor Algebra 53

   
Ti:j gi gj gi :T:g j gi gj
   
gi :gi T gj :g j dii Tdjj T 2:71
)T Ti:j gi gj

Analogously, one obtains the mixed tensor component Ti.j.

     
T:ji g j :gj T:gi :gj T: gi :gj
 
T: gj :gi gi :T:gj 2:72
) T:ji djj T:ji gi :T:gj

Note that in Eq. (2.72), the dot before the lower index indicates the position of
the basis of the upper index locating in front of the tensor T. In this case, the tensor
T is located between the upper basis gi and lower basis gj (Itskov 2010; Nayak
2012; Oeijord 2005).
Equation (2.72) can be written in the covariant and contravariant bases gi and gj
as follows:
   
T:ji gi g j gi :T:gj gi g j
   
gi :gi T g j :gj dii Tdjj T 2:73
)T T:ji gi g j

In a nutshell, the second-order tensor can be written in different expressions

according to the covariant, contravariant, and mixed components.
8
> Tij gi g j ; Tij gi :T:gj
>
< T ij g g ; T ij gi :T:g j
i j
T2 2:74
>
> T j gi g ; T j gi :T:g j
: i:i jj i:i
T:j gi g ; T:j gi :T:gj

Note that if the second-order tensor T is symmetric, then

Tij Tji ; T ij T ji ; Ti:j T:ij ; T:ji Tj:i 2:75

Compared to the second-order tensors, the first-order tensor T(1) has only one
dummy index, as shown in

1 Ti gi ; Ti T:gi
T 2:76
T i gi ; T i T:gi

An N-order tensor T(N) is the tensor product of the N covariant, contravariant,

and mixed bases of the coordinates:
54 2 Tensor Analysis

8 ij...n
< T gi gj . . .gn
TN Tij...n gi g j . . .gn 2:77
: ij::
Tl...n gi gj . . .gl . . .gn

The N-order tensors contain the 2N expressions in total. Two of them in respect
of the covariant and contravariant tensor components and (2N - 2) expressions in
respect of the mixed tensor components (Klingbeil 1966). In the case of a second-
order tensor T(2) for N = 2, there are four expressions: two with the covariant and
contravariant tensor components and two with the mixed tensor components, as
displayed in Eq. (2.74).

2.3.4 Tensor and Cross Products of Two Vectors in General

Bases
(a) Tensor product
Let u and v be two arbitrary vectors in the finite N-dimensional vector space RN,
they can be written in the covariant and contravariant bases as
(
u ui gi ui gi 2 RN ;
2:78
v v j gj v j g j 2 R N

The tensor product T of two vectors generates a second-order tensor that can be
defined by the linear functional T.
In the covariant bases:

T : u; v 2 RN ! T  u  v ui v j gi gj ui vj gi g j 2 RN  RN
T ui v j gi gj  T ij gi gj 2:79
i j i j
ui vj g g  Tij g g

where
Tij is the contravariant component of the second-order tensor T;
Tij is the covariant component of the second-order tensor T;

In the covariant and contravariant bases:

T : u; v 2 RN ! T  u  v ui vj gi g j ui v j gi gj 2 RN  RN
T ui vj gi g j  Tji gi g j 2:80
j i
ui v g gj  Tij gi gj
2.3 Tensor Algebra 55

where Tij and Tji are the mixed components of the second-order tensor T.
The tensor product T of two vectors in an orthonormal basis (e.g., Euclidean
coordinate system) is an invariant (scalar). The invariant is independent of the
coordinate system and has an intrinsic value in any coordinate transformations. In
Newtonian mechanics, the mechanical work W that is created by the force vector
F and path vector x does not change in any chosen coordinate system. This
mechanical work W = F.x is called an invariant and has an intrinsic value of
energy.
Given three arbitrary vectors u, v, and w in RN and a scalar a in R, the tensor
product of two vectors has the following properties (Klingbeil 1966):
Distributive property

uv w uv uw
u vw uw vw

Associative property

a uv ua v a uv

(b) Cross product

The cross product of two vectors u and v can be defined by a linear functional T.

T : u; v 2 RN ! T u  v ui v j gi  gj 2 RN 2:81

Obviously, the cross product T of two vectors is a vector (first-order tensor) of

which the direction is perpendicular to the bases of gi and gj.
Using the scalar triple product in Eq. (1.10), the cross product of the bases can
be written as
  p
gi  gj eijk g gk eijk J gk 2:82

where eijk is the Levi-Civita permutation symbol; J is the Jacobian.

Thus, the cross product in Eq. (2.81) can be expressed as

T u  v  T k gk
 p    2:83
eijk g ui v j gk eijk J ui v j gk

The covariant component Tk of the first-order tensor T results from Eqs. (2.81),
(2.82), and (2.83).
56 2 Tensor Analysis

   
T ui v j gi  gj ui v j gi  gj :gk gk  Tk gk
  p 2:84
) Tk ui v j gi  gj :gk eijk g ui v j eijk J ui v j

The Levi-Civita permutation symbol (pseudo-tensor) can be defined as

8
< 1 if i; j; k is an even permutation;
eijk 1 if i; j; k is an odd permutation; 2:85
:
0 if i j; or i k; or j k

Therefore,

eijk ejki ekij even permutation;
eijk eikj ekji eji k odd permutation 2:86

The permutation symbol eijk contains totally 27 elements (Nn = 33) for i, j,
k (n = 3) in a 27-dimensional tensor space R3 9 R3 9 R3.
Note that the permutation symbol is used in Eq. (2.83) because the direction of
the cross product vector is opposite if the dummy indices are interchanged with
each other in Einstein summation convention (cf. Appendix A).
p k    
g g J gk gi  gj  gj  gi
   
k
eijk gi  gj eijk gi  gj
)g p 2:87
g J
  p k
) gi  gj eijk gg eijk Jgk

where J denotes the Jacobian.

2.3.5 Rules of Tensor Calculations

In order to carry out the tensor calculations, some fundamental rules must be taken
into account in tensor calculus.
1. Calculation of tensor components
Let T be a second-order tensor that can be written in different tensor forms:

T T ij gi gj Tji gi g j Tij gi g j Tij gj gi 2:88

Multiplying the first row of Eq. (2.88) by the covariant basis gk, one obtains
2.3 Tensor Algebra 57

 
T ij gk :gj gi T ij gkj gi Tki gi
2:89a
) Tki T ij gkj for j 1; 2; . . .; N

Analogously, multiplying the second row of Eq. (2.88) by the contravariant

basis gk, one obtains
 
Tij gk :g j gi Tij gkj gi Tik gi
2:89b
) Tik Tij gkj for j 1; 2; . . .; N

Multiplying Eq. (2.89a) by gkj, the contravariant tensor components result in

T ij Tki gkj for k 1; 2; . . .; N 2:90a

Multiplying Eq. (2.89b) by gkj, one obtains the covariant tensor components

Tij Tik gkj for k 1; 2; . . .; N 2:90b

Substituting Eqs. (2.89a) and (2.90a), one obtains the contraction rules between
the contravariant tensor components.

T ij T ip gpk gkj for k; p 1; 2; . . .; N 2:91a

Similarly, the contraction rules between the covariant tensor components result
from substituting Eqs. (2.89b) and (2.90b).

Tij Tip gpk gkj for k; p 1; 2; . . .; N 2:91b

Analogously, the contraction rules between the mixed tensor components can
be derived as

Tji Tpi gpk gkj for k; p 1; 2; . . .; N 2:92a

Similarly, one obtains

Tij Tip gpk gkj for k; p 1; 2; . . .; N 2:92b

2. Addition law
Tensors of the same orders and types can be added together. The resulting tensor
has the same order and type of the initial tensors. The tensor resulted from the
addition of two covariant or contravariant tensors A and B can be calculated as
58 2 Tensor Analysis

 
C A B Aijk Bijk gi g j gk Cijk gi g j gk B A
) Cijk Aijk Bijk Bijk Aijk ;
  2:93
C A B Aijk Bijk gi gj gk C ijk gi gj gk B A
) C ijk Aijk Bijk Bijk Aijk

Similarly, the tensor resulted from the addition of two mixed tensors A and
B can be written as
 
C A B Apq
ijk B pq i j k
ijk gp gq g g g
pq
Cijk gp gq gi g j gk B A 2:94
pq
) Cijk Apq pq pq pq
ijk Bijk Bijk Aijk

Straightforwardly, the addition of tensors is commutative, as proved in

Eqs. (2.93) and (2.94).
3. Outer product
On the contrary, the outer product can be carried out at tensors of different orders
and types. The tensor components resulted from the outer product of two mixed
tensors A and B can be calculated as
  
AB Apq i j
ij gp gq g g Brts k l
kl g g gr gt gs
pqrst
Cijkl gp gq gi g j gk gl gr gt gs 6 BA 2:95
pqrst
) Cijkl Apq rts
ij Bkl Brts pq
kl Aij

The outer product of two tensors results a tensor with the order that equals the
sum of the covariant and contravariant indices. The outer product is not com-
mutative, but their tensor components are commutative, as shown in Eq. (2.95). In
this example, the resulting ninth-order tensor is generated from the outer product
of the mixed fourth-order tensor A and mixed fifth-order tensor B. Obviously, the
outer product of tensors A, B, and C is associative, i.e., A(BC) = (AB)C.
4. Contraction law
The contraction operation can be only carried out at the mixed tensor types of
different orders. The tensor contraction is operated in many contracting steps
where the tensor order is shortened by eliminating the same covariant and con-
travariant indices of the tensor components.
We consider a mixed tensor of high orders. In this example, the mixed fifth-
order tensor A of type
 (2,
 3) can be transformed from the coordinates {ui} into the
barred coordinates ui . The transformed tensor components can be calculated
according to the transformation law in Eq. (2.144).
2.3 Tensor Algebra 59

i j r s t
 ij ou ou ou ou ou Apq
A 2:96
klm
oup ouq ouk oul oum rst
 in Eq. (2.96) at l = i, one obtains the
Carrying out the first contraction of A
tensor components
i j r s t
 ij ou ou ou ou ou Apq
A kim p q k l m rst
ou ou ou o u o u

j r t i s
ou ou ou ou ou
q k m Apq
ou ou ou oup oui rst
ou j our out 2:97a
q k m dsp Apq rst
ou ou ou
ou j our out
q k m Apq
ou ou ou rpt
j r t
,B j ou ou ou Bqrt
km
ouq ouk oum

As a result, the resulting tensor components B are the third-order tensor type
after the first contraction of A at s = p:
p
Apq pq q
rst ds Arpt  Brt 2:97b

 in Eq. (2.97a) at k = j, one obtains

Further contracting the tensor components B

 j out ou j our q

Bjm m B
ou ouq ou j rt
out
m drq Bqrt
ou 2:98a
out
m Bqqt
ou
t
 m ou Ct
,C
oum

As a result, the resulting tensor components C are the first-order tensor type
after the second contraction of B in Eq. (2.98a) at r = q:

Bqrt dqr Bqqt  Ct 2:98b

5. Inner product
The inner product of tensors comprises two basic operations of the outer product
and at least one contraction of tensors. As an example, the outer product of two
third-order tensors A and B results in a sixth-order tensor.
60 2 Tensor Analysis

  
AB Amp
q g g
m p g q
Brst gr gs gt
2:99
Amp r q s t
q Bst g gr gm gp g g

Using the first tensor contraction in Eq. (2.99) at r = q, one obtains the
resulting fourth-order tensor components of the inner product.
r r
Amp mp q mp
q Bst dq Aq Bst  Cst 2:100

Similarly, using the second tensor contraction law in Eq. (2.100) at p = s, the
resulting second-order tensor components result in
Cstmp dps Cstms  Dm
t 2:101

Finally, applying the third tensor contraction law to Eq. (2.101) at m = t, the
resulting tensor component is an invariant (zeroth-order tensor).
m
Dm t
t dt D t  D 2:102

In another approach, one can calculate the tensor components of the inner
product of two contravariant tensors A and B multiplying by the metric tensor.

Aijk Blm ! Aijk Blm glk 2:103

Using Eq. (2.89a), one obtains the resulting tensor components

Blm glk Bm
k 2:104

Substituting Eq. (2.104) into Eq. (2.103) and using the tensor contraction law,
one obtains the resulting tensor components

C ijm  Aijk Bm m ijk

k Bk A 2:105

Equation (2.105) denotes that the inner product of the tensor components is
commutative.
6. Indices law
Using the metric tensors, the operation of moving indices enables changing indices
of the tensor components from the upper into lower positions and vice versa.
Multiplying a tensor component by the metric tensor components, the lower index
(covariant index) is moved into the upper index (contravariant index) and vice versa.
Moving covariant indices i, j to the upper position:

Akij ! Akij gil Akl kl jm

j ! Aj g A
klm
2:106a
2.3 Tensor Algebra 61

Moving contravariant indices i, j to the lower position:

Aijk ! Aijk gjl Aikl ! Aikl gim Aklm 2:106b

7. Quotient law
The quotient law of tensors postulates that if the tensor product of AB and B are
tensors, A must be a tensor.

AB C)tensor \ B)tensor ) A)tensor 2:107a

Proof Using the contraction law, the barred components in the transformation
coordinates ui of the tensor product AB result in

k B
Akij Bilk Cjl ) A  il  l
ij k Cj 2:107b

According tothe transformation law (2.144), the transformed components in the

coordinates  ui of the tensors B and C can be calculated as

 ilk Bmn oui oul oup

B p ;
oum oun ouk 2:107c
l q
 l Cn ou ou
C j q
ou ou j
n

Substituting Eq. (2.107c) into Eq. (2.107b) and using the contraction law, one
obtains
i l p

 k Bmn ou ou ou C n o ul ouq

A ij p m n k q
ou ou ou oun ou j
  oul ouq 2:107d
Apmq Bmn
p
oun ou j

Rearranging the terms of Eq. (2.107d), one obtains

l
 ui oup
k o p ou
q
ou mn
Aij m k  Amq j B 0
ou ou ou oun p
2:107e
ul
mn o  k oui oup p ou
q
) 8Bp ; n 6 0 : Aij m k Amq j
ou ou ou ou
 r k
Applying the inner product by ou o
u
ui ous
o
to Eq. (2.107e), one obtains the barred
components of A at r = m and s = p.
62 2 Tensor Analysis

q
r k

 k r p p ou ou ou
Aij dm ds Amq j
ou oui ous
2:107f
k m q
)A k Ap ou ou ou
ij mq
oup oui ou j

Equation (2.107f) proves that A is a mixed third-order tensor of type (1, 2), cf.
Eq. (2.107c).
8. Symmetric tensors
Tensor T is called symmetric in the given basis if two covariant or contravariant
indices of the tensor component can be interchanged without changing the tensor
component value.

T ij T ji : symmetric in i and j
T ij T ji : symmetric in i and j
ijk ikj 2:108
Tpq Tpq : symmetric in j and k
ijk ijk
Tpq Tqp : symmetric in p and q

In the case of a second-order tensor, the tensor T is symmetric if T equals its

transpose.

T TT 2:109

9. Skew-symmetric tensors
The sign of the tensor component is opposite if a pair of the covariant or con-
travariant indices are interchanged with each other. In this case, the tensor is skew-
symmetric (antisymmetric).
Tensor T is defined as a skew-symmetric tensor (antisymmetric) if

T ij T ji : skew-symmetric in i and j
T ij T ji : skew-symmetric in i and j
ijk ikj 2:110
Tpq Tpq : skew-symmetric in j and k
ijk ijk
Tpq Tqp : skew-symmetric in p and q

In the case of a second-order tensor, the tensor T is skew-symmetric if T is

opposite to its transpose.

T TT 2:111
2.3 Tensor Algebra 63

An arbitrary tensor T can be generally decomposed into the symmetric and

skew-symmetric tensors:

1  1 
T T TT T  TT  Tsym Tskew 2:112
2 2
Proof
The first tensor Tsym is symmetric:

1 
Tsym  T TT TTsym qed
2

The second tensor Tskew is skew-symmetric:

1 
Tskew  T  TT TTskew qed
2

2.4 Coordinate Transformations

Tensors are tuples of independent coordinates in a finite multifold N-dimensional

tensor space (RN 9  9 RN). The tensor describes physical states generally
depending on different variables (dimensions). Each physical state can be defined
as the point P(u1,, uN) with N coordinates of ui. By changing the variables, such
as time, locations, and physical characteristics (e.g., pressure, temperature, density,
velocity), the physical state point varies in the multifold N-dimensional space.
The tensor does not change itself and is independent in any coordinate system.
However, its components change in the new basis by the coordinate transformation
since the basis changes as the coordinate system varies. In this case, applications of
tensor analysis have been used to describe the transformation between two general
curvilinear coordinate systems in the multifold N-dimensional spaces. Hence,
tensors are a very useful tool applied to the coordinate transformations in the
multifold N-dimensional tensor spaces. High-order tensors can be generated by a
multilinear map between two multifold N-dimensional spaces (cf. Sect. 2.2). Their
components change in the relating bases by the coordinate transformations, as
displayed in Fig. 2.5.
In the following section, the relations between the tensor components in dif-
ferent curvilinear coordinates of the finite N-dimensional spaces will be discussed.

2.4.1 Transformation in the Orthonormal Coordinates

The simple coordinate transformation of rotation between the orthonormal coor-

dinates xi and uj in Euclidean coordinate system is carried out.
64 2 Tensor Analysis

Basis {g i }
Tensor components
Tij

Tensor T independent Tensor components

of any basis dependent on the basis

Tensor components
Basis { g i } changed
Tij Tij

Fig. 2.5 Tensor and tensor components in different bases

An arbitrary vector r (first-order tensor) can be written in both coordinate

systems:

r x1 e1 x2 e2  xi ei
2:113
u1 g1 u2 g2  uj gj

The vector components in the coordinate uj can be calculated in

(
u1 cos h11 x1 cos h12 x2
2:114
u2 cos h21 x1 cos h22 x2

Thus,
    
u1 cos h11 cos h12 x
: 1 , u Tx 2:115
u2 cos h21 cos h22 x2

where T is the transformation matrix.

Setting h11 = h, one obtains
 
p
cos h11 cos h; cos h12 cos h  sin h
2
 
p
cos h21 cos h  sin h; cos h22 cos h
2

Therefore, the transformation matrix T becomes

   
cos h11 cos h12 cos h  sin h
T 2:116
cos h21 cos h22 sin h cos h
2.4 Coordinate Transformations 65

Fig. 2.6 Two-dimensional x2

coordinate transformation of
rotation u2

u2

x2 P( x j )

e2 r
g2

0 x1
e1 x1
g1
11 =
u1

u1

The transformed coordinates can be computed by the transformation T:

    
u1 cos h  sin h x
T : x ! u Tx : : 1 2:117
u2 sin h cos h x2

where h is the rotation angle of the rotating coordinates uj.

Transforming back Eq. (2.117), one obtains the coordinates xi (Fig. 2.6)
    
1 x 1 cos h sin h u
T :u!xT u: 1 : 1 2:118
x2  sin h cos h u2

The vector component on the basis is obtained multiplying Eq. (2.113) by the
relating basis ei or gj.

xi r:ei ; i 1; 2
2:119
uj r:gj ; j 1; 2

Substituting Eq. (2.119) into Eq. (2.117), one obtains the transformation matrix
between two coordinate systems.
    
g1 cos h  sin h e
: 1 , g T:e 2:120
g2 sin h cos h e2
66 2 Tensor Analysis

T ( 2 ) E N N T ( 2 ) E N N

S
{g i } {gi }

Fig. 2.7 Basis transformation of general curvilinear coordinates in EN

Similarly,
    
e1 cos h sin h g
: 1 , e T1 :g 2:121
e2  sin h cos h g2

2.4.2 Transformation of Curvilinear Coordinates in EN

In the following section, second-order tensors T are used in the transformation of

general curvilinear coordinates in Euclidean space EN, as shown in Fig. 2.7.
The basis gi of the curvilinear coordinate {ui} can be transformed into the new
gi of the curvilinear coordinate f ui g using the linear transformation S. The
basis 
new covariant basis can be rewritten as a linear combination of the old basis.

S : gi ! gi Sij gj , G
 GS 2:122

where Sji are the mixed transformation components of the second-order tensor S.
The old covariant basis results can be calculated as
or or oui  1 i  1 i oui
gj  
g i S j) S j j 2:123
ou j oui ou j ou

Inversing the basis matrix in Eq. (2.122), the new contravariant basis can be
calculated as
 i
 1 GS1 S1 G1 ) 
G gi S1 j g j 2:124

Multiplying Eq. (2.124) by the linear transformation S, the old contravariant

basis results in
 1
G1 SG ) gi Sij g j 2:125
2.4 Coordinate Transformations 67

According to Eqs. (2.11) and (2.122), the new covariant basis can be calculated as

or or ouk k k ouk
gj  g k S j ) S j  2:126a
ou j ouk ou j ou j

Combining Eqs. (2.123), (2.124), and (2.126a) and using the differentiation
chain rule, one obtains the relation of the mixed transformation components
between two general curvilinear coordinates in Euclidean space EN.
 i  i
gj :gi gi :gj S1 l Skj gl :gk S1 l Skj dlk
 i oui ouk oui
S1 k Skj k j j dij 2:126b
ou ou ou
, S1 S I

Therefore, the transformation tensor S can be written as

2 3
ou1 ou1 ou1
6 ou1 :
6 ou2 ouN 7
7
6 2 7
6 ou ou2 ou2 7
S6
6 ou1 ou2
: 7 N
ou 7 2 R  R
N
N
2:127a
6 : : : : 77
6
4 ouN ouN ouN 5
:
ou1 ou2 ouN

The transformation tensor S in Eq. (2.127a) is identical to the Jacobian matrix

between two coordinate systems {ui} and f ui g.
Inverting the transformation tensor S, the back transformation results in
2 3
ou1 ou1 ou1
6 ou1 :
6 ou2 ouN 77
6 2 7
6 ou ou2 ou2 7
S1 6
6 ou1 ou2
: 7 N
ouN 7 2 R  R
N
2:127b
6 : : : : 77
6
4 ouN ouN ouN 5
:
ou1 ou2 ouN

The relation between the new and old components of an arbitrary vector v (first-
order tensor) is similarly given in the coordinate transformation S according to
Eqs. (2.122) and (2.124).
 
vi gi :v Sij gj : vj g j Sij vj
 i    i 2:128
vi gi :v S1 j g j : v j gj S1 j v j
68 2 Tensor Analysis

Using Eq. (2.74), the relation of the components of the second-order tensor T
can be derived.
Covariant metric tensor components:
   
Tij gi :T:gj Ski gk :T: Slj gl
2:129
Ski Slj gk :T:gl Ski Slj Tkl

Contravariant metric tensor components:

 i  j
T ij gi :T:g j S1 k gk :T: S1 l gl
 i   j    i   j 2:130
S1 k S1 l gk :T:gl S1 k S1 l T kl

Mixed metric tensor components:

   
i
Tj:i gj :T:gi Skj gk :T: S1 l gl
 i    i 2:131
Skj S1 l gk :T:gl Skj S1 l Tk:l

Note that in Eq. (2.131), the dot after the lower index indicates the position of
the basis of the upper index locating after the tensor T. In this case, the tensor T is
located between the upper basis gi and lower basis gj.
 i  
T:ji gi :T:gj S1 k gk :T: Slj gl
 i    i 2:132
S1 k Slj gk :T:gl S1 k Slj T:lk

Note that in Eq. (2.132), the dot before the lower index indicates the position of
the basis of the upper index locating in front of the tensor T. In this case, the tensor
T is located between the upper basis gi and lower basis gj.

2.4.3 Examples of Coordinate Transformations

(a) Cylindrical Coordinates
The transformation S from Cartesian {ui} to cylindrical coordinates f ui g:
8 1 1 2
< u r cos h  u cos u
>
S : u2 r sin h  u1 sin u2
>
: 3
u z u3
2.4 Coordinate Transformations 69

The covariant transformation matrix S can be calculated as

2 1 3
ou ou1 ou1
6 ou1 ou2 0 1
6 ou3 7
7 cos h r sin h 0
6 ou2 ou2 27
ou 7 B C
S6 6 ou1 ou2 @ sin h r cos h 0 A
6 ou3 7
7
4 3 5 0 0 1
ou ou3 ou3
ou1 ou2 ou3

The determinant of S is called the Jacobian J.

 1 
 ou ou1 ou1 
  
 ou1 ou2 ou3   cos h
  r sin h 0 
 ou2 ou2 ou2   
jSj    J  sin h r cos h 0 r
 ou1 ou2 ou3   
 3   0 0 1
 ou ou3 ou3 
 1 
ou ou2 ou3

The contravariant transformation matrix S-1 results from the inversion of the
covariant matrix S.
2 1 3
ou ou1 ou1
6 ou1 ou2 ou3 7 0 1
6 7 r cos h r sin h 0
6 ou2 ou2 ou2 7 1 B C
S1 6 7
6 ou1 ou2 ou3 7 r @  sin h cos h 0 A
6 7
4 3 5 0 0 r
ou ou3 ou3
ou1 ou2 ou3

(b) Spherical Coordinates

The transformation S from Cartesian {ui} to spherical coordinates f ui g:
8 1 1 2 3
< u q sin / cos h  u sin u cos u
>
S : u2 q sin / sin h  u1 sin u2 cos u3
>
: 3
u q cos / u1 cos u2

The covariant transformation matrix S can be calculated as

2 1 3
ou ou1 ou1
6 o 1 2 37 0 1
6 u ou ou 7 sin / cos h q cos / cos h q sin / sin h
6 ou2 ou2 ou2 7 B
S6 7 @ sin / sin h q cos / sin h q sin / cos h CA
6 o 7
6 u1 ou2 ou3 7
4 3 5 cos / q sin / 0
ou ou3 ou3
u1 ou2 ou3
o
70 2 Tensor Analysis

The determinant of S is called the Jacobian J.

 1 
 ou ou1 ou1 
 
 ou1 ou2 ou3 
 
 ou2 ou2 ou2 

j Sj  J

 ou1 ou2 ou3 
 3 3 3
 ou ou ou 
 1 2 3

o
u ou ou 
 sin / cos h q cos / cos h q sin / sin h 

 
 sin / sin h q cos / sin h q sin / cos h  q2 sin /
 
 cos / q sin / 0 

The contravariant transformation matrix results from the inversion of the matrix S.
2 3
u1
o ou1 ou1 0 1
6 ou1 ou2 ou3 7 q sin / cos h q sin / sin h q cos /
6 7 B C
6 o
u2 ou2 27
ou 7 1 B cos / cos h cos / sin h  sin / C
S1 6
6 ou1 B

C
6 ou2 ou3 77 q @ sin h cos h A
4 3 5  0
o
u ou3 ou3 sin / sin /
ou1 ou2 ou3

2.4.4 Transformation of Curvilinear Coordinates in RN

In the following section, second-order tensors T will be used in the transformation

of general curvilinear coordinates in Riemannian manifold RN, as shown in
Fig. 2.8. In Riemannian manifold, the bases gi and gi of the curvilinear coordinates
ui and ui do not exist any longer. Instead of the metric coefficients, the transfor-
mation coefficients that depend on the relating coordinates have been used in
Riemannian manifold (Klingbeil 1966).
The new barred curvilinear coordinate ui is a function of the old curvilinear
coordinates uj, j = 1, 2,, N. Therefore, it can be written in a linear function of uj.

ui ui u1 ; . . .; uN
) ui aij u j for j 1; 2; . . .; N 2:133
i
) du aij du j for j 1; 2; . . .; N

aij is the transformation coefficient in the coordinate transformation S.

where 
Using the chain rule of differentiation, one obtains
2.4 Coordinate Transformations 71

T ( 2 ) R N N T ( 2 ) R N N

S
{ai } {ai }

Fig. 2.8 Basis transformation of general curvilinear coordinates in RN

oui j
dui du aij du j
ou j 2:134
oui
) aij j
ou

Analogously, the transformation coefficients of the back transformation result in

oui j
dui du aij d
uj
ou j 2:135
oui
) aij j
ou

Combining Eqs. (2.134) and (2.135) and using the Kronecker delta, one obtains
the relation between the transformation coefficients

oui ou j oui

aij akj j k
k dik
o
u ou  ou   2:136
, aij akj akj aij I

The relation of the second-order tensor components between the new and old
curvilinear coordinates can be calculated using Eq. (2.133).
 
T 2 Tkl uk ul Tkl aki 
alj ui u j
2:137
 Tij ui u j

Thus,

ouk oul
Tij aki alj Tkl i j Tkl ;
ou ou 2:138
k
ou oul
Tij aki alj Tkl i j Tkl
ou ou
72 2 Tensor Analysis

In the same way, the covariant, contravariant, and mixed components of the
second-order tensor T between both coordinates in the transformation can be
derived as:
Covariant tensor components:
Tij aki alj Tkl , Tij aki alj Tkl 2:139

Contravariant tensor components:

T ij aik alj T kl , T ij aik alj T kl 2:140

Mixed tensor components:

Ti:j aki alj Tk:l , Ti:j aki alj Tk:l 2:141

T:ji aik alj T:lk , T:ji aik alj T:lk 2:142

Generally, the transformation coefficients of high-order tensors can be alter-

natively computed as
k l m
s t

ou ou ou ou ou  pqr

Tijklm T
oup ouq our oui ou j st 2:143
  
k l m s t  pqr
ap aq ar ai aj Tst

Therefore,

ouk oul oum ous out pqr

Tijklm T
oup ouq our oui ou j st 2:144
akp alq am s t pqr
r ai aj Tst

2.5 Tensor Calculus in General Curvilinear Coordinates

In the following sections, some necessary symbols, such as the Christoffel sym-
bols, RiemannChristoffel tensor, and fundamental invariants of the Nabla oper-
ator have to be taken into account in the tensor applications to fluid mechanics and
other working areas.

2.5.1 Physical Component of Tensors

Various types of the second-order tensors are shown in Eq. (2.74). The physical
tensor component is defined as the tensor component on its covariant unitary basis
2.5 Tensor Calculus in General Curvilinear Coordinates 73

g*i . Therefore, the basis of the general curvilinear coordinates must be normalized
(cf. Appendix B).
Dividing the covariant basis by its vector length, the covariant unitary basis
(covariant normalized basis) results in
gi gi  
gi )  gi  1
p 2:145
j gi j gii

The covariant basis norm |gi| can be considered as a scale factor hi.
p
hi j gi j gii no summation over i

Thus, the covariant basis can be related to its covariant unitary basis by the
relation of
p
gi gii gi hi gi 2:146

The contravariant basis can be related to its covariant unitary basis using
Eqs. (2.47) and (2.146).
gi gij gj gij hj gj 2:147

The contravariant second-order tensor can be written in the covariant unitary

bases using Eq. (2.146).
 
T T ij gi gj T ij hi hj gi gj  T ij gi gj 2:148

Thus, the physical contravariant tensor components result in

T ij  T ij hi hj 2:149

The covariant second-order tensor can be written in the contravariant unitary

bases using Eq. (2.147).
 
T Tij gi g j Tij gik gjl hk hl gk gl  Tij gk gl 2:150

Similarly, the physical covariant tensor component results in

Tij Tij gik gjl hk hl 2:151

The mixed tensors can be written in the covariant unitary bases using
Eqs. (2.146) and (2.147)
 
T Tji gi g j Tji gi gjk gk
    
Tji hi gi gjk hk gk Tji gjk hi hk gi gk 2:152
 
 Tji gi gk
74 2 Tensor Analysis

Thus, the physical mixed tensor component results from Eq. (2.152):
 
Tji Tji gjk hi hk 2:153

2.5.2 Derivatives of Covariant Bases

Let gi be a covariant basis in the curvilinear coordinates {ui}, the derivative of the
covariant basis with respect to the time variable t can be computed as

ogi o or
g_ i  r_ ;i 2:154
ot ot oui

Due to ui is a differentiable function of t, Eq. (2.154) can be rewritten as

ogi ogi ou j
g_ i j  gi;j u_ j 2:155
ot ou ot

where gi,j is called the derivative of the covariant basis gi of the curvilinear
coordinates {ui}.
Using the chain rule of differentiation, the covariant basis of the curvilinear
coordinates {ui} can be calculated in Cartesian coordinates {xi}.

or or oxp
gi ep xp;i 2:156
oui oxp oui

Similarly, one obtains the covariant basis of the coordinates {xi}.

or or ouk
ep gk uk;p 2:157
oxp ouk oxp

The derivative of the covariant basis of the coordinates {ui} can be obtained
from Eqs. (2.156) and (2.157).
 
ogi o ep xp;i oxp;i
gi;j e p
ou j ou j ou j
p k 2 p

ox ;i ou o x 2:158
uk;p j gk g
ou oxp oui ou j k
 Ckij gk for k 1; 2; . . .; N
2.5 Tensor Calculus in General Curvilinear Coordinates 75

The symbol Ckij in Eq. (2.158) is defined as the second-kind Christoffel symbol,
which has 27 (= 33) components for a three-dimensional space (N = 3).
Thus, the second-order Christoffel symbols that only depend on both coordi-
nates of {ui} and {xi} can be written as

ouk o2 xp
Ckij
oxp oui ou j 2:159
ouk o2 xp
p j i Ckji
ox ou ou

The result of Eq. (2.159) proves that the second-kind Christoffel symbols are
symmetric with respect to i and j.
The second-kind Christoffel symbols are given by multiplying both sides of
Eq. (2.158) by the contravariant basis gl.
 
Ckij gk :gl Ckij dlk gl :gi;j
2:160
) Clij gl :gi;j

Substituting Eq. (2.158) into Eq. (2.155), one obtains the relation between the
covariant basis time derivative and the Christoffel symbol.

g_ i gi;j u_ j Ckij u_ j gk 2:161

Furthermore, the covariant basis derivative can be calculated in Cartesian

coordinate {xi} using Eq. (2.156).
   
ogi o ep xp;i o xp;i
gi;j ep ep xp;ij 2:162
ou j ou j ou j

According to Eq. (2.159), the second-kind Christoffel symbols can be rewritten

as

ouk o2 xp ouk o2 xp
Ckij
oxp oui ouj oxp ouj oui 2:163
uk;p xp;ij uk;p xp;ji

2.5.3 Christoffel Symbols of First and Second Kind

According to Eq. (2.160), the second-kind Christoffel symbol can be defined as

76 2 Tensor Analysis


k o2 r o2 r
Ckij  gk : g k
:
i j oui ouj ou~j oui 2:164
gk :gi;j gk :gj; i Ckji

Equation (2.164) reconfirms the symmetric property of the Christoffel symbols

with respect to the indices i and j. Obviously, the Christoffel symbols are coor-
dinate dependent; therefore, they are not tensors.
In order to compute the second-kind Christoffel symbols in the covariant metric
coefficients, the derivative of gij with respect to uk has to be taken into account.
 
gij gi :gj
ogij   2:165
) gij;k  k
gi :gj ;k gi;k :gj gi :gj;k
ou

Using Eq. (2.158) at changing the index j into k; then, i into j, one obtains the
following relations

gi;k Cpik gp ; gj;k Cpjk gp 2:166

Substituting Eq. (2.166) into Eq. (2.165), one obtains the derivative of gij with
respect to uk.

gij;k gi;k :gj gj;k :gi

Cpik gp :gj Cpjk gp :gi 2:167
Cpik gpj Cpjk gp i

Interchanging k with i in Eq. (2.167), one obtains

gkj;i Cpki gpj Cpji gp k 2:168

Analogously, one reaches the relation interchanging k with j in Eq. (2.167).

gik; j Cpij gpk Cpkj gp i 2:169

Combining Eqs. (2.167), (2.168), and (2.169), the Christoffel symbols can be
written in the derivatives of the covariant metric coefficients.

1 
gpj Cpik g gkj;i  gik;j 2:170
2 ij;k

Multiplying Eq. (2.170) by gqj, the Christoffel symbols result according to

Eq. (2.50) in
2.5 Tensor Calculus in General Curvilinear Coordinates 77

Cpik gpj gqj Cpik dqp

1   2:171
) Cqik gqj gij;k gkj;i  gik;j
2

Changing j into p, k into j, and q into k in Eq. (2.171), one obtains

1 
Ckij g gjp;i  gij;p gkp
2 ip;j 2:172
 Cijp gkp

Changing the index p into k, the first-kind Christoffel symbol Cijp in Eq. (2.172)
that has 27 (= 33) components for a three-dimensional space (N = 3) is defined as

Cijk  i j; k
1 
 gik;j gjk;i  gij;k 2:173a
2
gpk Cpij for p 1; 2; . . .; N

Other expressions of the Christoffel symbols can be found in some literature.

Cijk  i j; k

p 2:173b
gpk  gpk Cpij for p 1; 2; . . .; N
i j

2.5.4 Prove that the Christoffel Symbols are Symmetric

1. The first-kind Christoffel symbol is symmetric with respect to i, j
According to Eq. (2.173a), the first-kind Christoffel symbol can be written as

1 
Cijk g gjk;i  gij;k
2 ik;j

Interchanging i with j in the terms on the RHS, one obtains

1 
Cijk g gik;j  gji;k
2 jk;i
Cjik qed:

2. The second-kind Christoffel symbol is symmetric with respect to i, j

Using Eq. (2.172), the second-kind Christoffel symbol can be expressed as
78 2 Tensor Analysis

Ckij gkp Cijp

Due to the symmetry of the first-kind Christoffel symbol, the second-kind

Christoffel symbol results in

Ckij gkp Cijp gkp Cjip

Ckji qed

2.5.5 Examples of Computing the Christoffel Symbols

Given a curvilinear coordinate{ui} with u1 = u; u2 = v; u3 = w in another

coordinate {xi}, the relation between two coordinate systems can be written as
8 1
< x uv
>
x2 w
>
: 3
x u2  v

The covariant basis matrix G can be calculated from

0 1
ox1 ox1 ox1
B ou1 2 3
B ou2 ou3 C
C v u 0
B ox2 ox2 2C
ox C 6 7
G g1 g2 g3  B
B ou1 4 0 0 15
B ou2 ou3 C
C
@ 3 A 2u 1 0
ox ox3 ox3
ou1 ou2 ou3

Therefore, the covariant bases can be given in

8
< g1 v; 0; 2u
>
g2 u; 0; 1
>
:
g3 0; 1; 0

The determinant of G that equals the Jacobian J of

 
 v u 0 

jGj J  0 0 1  2u2 v 6 0
 2u 1 0 

The contravariant basis matrix G-1 is the inverse matrix of the covariant basis
matrix G.
2.5 Tensor Calculus in General Curvilinear Coordinates 79

0 1
ou1 ou1 ou1
2 3
B ox1 ox2 ox3 C
1
g B C
6 27 B B ou2 ou2 ou2 C C
G1 4g 5 B
ox 1 ox 2 ox 3C
B C
g3 @ 3 A
ou ou3 ou3
2 ox1 ox2 3 ox3 2 3
1 0 u J 1 0 uJ 1
16 7 6 7
4 2u 0 v 5 4 2uJ 1 0 vJ 1 5
J 2
0 2u v 0 0 1 0

Thus,
 1  1 1
G 
2u2 v J

k B
Akij Bilk Cjl ) A  il  l
ij k Cj

Thus, the contravariant bases result in

8 1 1
< g J 1; 0; u
>
g2 J 1 2u; 0; v
>
: 3
g J 1 0; 2u2 v; 0

Some examples of the second-kind Christoffel symbols of 27 components can

be computed from Eq. (2.160).

Ckij gi;j gk )
C111 g1;1 g1 J 1 0:1 0:0 2:u 2uJ 1
C112 g1;2 g1 J 1 1:1 0:0 0:u J 1
C113 g1;3 g1 J 1 0:1 0:0 0:u 0
...
C332 g3;2 g3 J 1 0:0 0:2u2 v 0:0 0
C333 g3;3 g3 J 1 0:0 0:2u2 v 0:0 0

The first-kind Christoffel symbols containing 27 components in R3 can be

computed from Eq. (2.173a).

Cijk gpk Cpij gp :gk Cpij for p 1; 2; 3

80 2 Tensor Analysis

2.5.6 Coordinate Transformations of the Christoffel Symbols

The second-kind Christoffel symbols like tensor components strongly depend on

the coordinates at the coordinate transformations. The curvilinear coordinates {ui}
is transformed into the new barred curvilinear coordinates fui g. Therefore, the old
basis is also changed into the new basis.
The second-kind Christoffel symbols can be written in the new basis of the
barred coordinates fui g.

 k gk : ogi gk :gi;j

C 2:174
ij
ou j

Using the chain rule of differentiation, the basis of the coordinates {ui} can be
calculated as

or or ouk
gp p
k p
ou ou ou 2:175
ouk
gk p
ou
 
Multiplied Eq. (2.175) by the new contravariant basis of the coordinates ui ,
one obtains changing the indices m into k, and p into l.

ouk ouk oum

gm :gp gm :gk
 p
dm
k
ou oup oup
    oum
) gm :gp gp gm gp :gp gm dpp p gp 2:176
ou
ouk
) gk l gl
ou
 
The new covariant basis of the coordinates ui can be calculated as

or or oup oup
gi i
p i gp i 2:177
ou ou ou ou

 iThus,
 the new covariant basis derivative with respect to j of the coordinates

u results in

ogi o oup
gi; j g
ou j ou j oui p
2:178
o2 up oup ogp
j i gp i j
ou ou ou ou
2.5 Tensor Calculus in General Curvilinear Coordinates 81

Substituting Eqs. (2.176) and (2.178) into Eq. (2.174), one obtains the
second-kind Christoffel symbols in the new basis.

 k gk :gi;j
C ij
2 p
2:179
ouk l ou oup ogp
g : g
oul ou j oui p oui ou j

Using Eq. (2.166) and the chain rule of differentiation, the second term in the
parentheses of Eq. (2.179) can be computed as

ogp ogp ouq

ou j ouq ou j 2:180
ouq ouq  
j gp;q j Crpq gr
ou ou

Inserting Eq. (2.180) into Eq. (2.179), the transformed Christoffel symbols in
the new barred coordinates can be calculated as

 k gk :gi;j
C ij
2 p

ouk ou l oup ouq r l

g :gp i j Cpq g :gr
oul ou j oui ou ou
k
2 p
2:181
ou o u l ou ouq r l
p
l d C d
ou ou j oui p oui ou j pq r
2 l

ouk o u oup ouq l

l C
ou oui ou j oui ou j pq

 iTherefore,
 the transformed Christoffel symbols in the new barred coordinates

u result in

k p q k 2 l
 k Cl ou ou ou ou o u
C 2:182
ij pq
oul oui ou j oul oui ou j

Rearranging the terms in Eq. (2.181), the second derivatives of ul with respect
to the new barred coordinates result in

oul  k o2 ul oup ouq

C i j Clpq
ouk ij i
ou ou j ou ou
2:183
o2 ul oul  k oup ouq l
) i j kC  C
ou ou ou ij oui ou j pq

Using Eq. (2.160), all Christoffel symbols in Cartesian coordinates {xi} vanish
because the basis ei does not change in any coordinate xj.
82 2 Tensor Analysis

oei
Ckij ek :ei;j ek : 0 2:184
ox j

2.5.7 Derivatives of Contravariant Bases

Like Eq. (2.158), the derivative of the contravariant basis of the curvilinear
coordinates {ui} with respect to uj can be defined as

ogi ^ i gk
gi;j C jk 2:185
ou j

where C^ i are the second-kind Christoffel symbols in the contravariant bases gk.
jk
In order to compute those Christoffel symbols, some calculating steps are
carried out in the following section.
The derivative of the product between the covariant and contravariant bases
with respect to uj can be computed using Eqs. (2.156), (2.164), and (2.185).
   
gi :gj ;k
gi;k :gj gi :gj;k dij
;k
   
^ i gl :gj Cl gl :gi
C kl jk

^ i dl Cl di
C 2:186
kl j jk l
i i
^ Ci C
C ^ Ci
kj jk kj kj
0

Thus, the relation between the Christoffel symbols of two coordinates results in

^ i Ci Ci
C 2:187
kj kj jk

Using Eqs. (2.164) and (2.187), one obtains

^ i Ci Ci C
C ^i 2:188
kj kj jk jk

^ i is symmetric with respect to j and k.

It proves that the Christoffel symbol C jk
Finally, the derivatives of the contravariant basis gi with respect to uj result
from Eqs. (2.185) and (2.187).
i
^ gk Ci gk
gi;j Cjk jk
i
2:189
^ gk Ci gk
Ckj kj
2.5 Tensor Calculus in General Curvilinear Coordinates 83

2.5.8 Derivatives of Covariant Metric Coefficients

The derivatives of the covariant metric coefficient can be derived from the first-
kind Christoffel symbols written as

1 
Cikj g gkj;i  gik;j ;
2 ij;k
1  2:190
Cjki g gki;j  gjk;i
2 ji;k

The derivative of the covariant metric coefficient results by adding both

Christoffel symbols given in Eq. (2.190).

1 1
Cikj Cjki gij;k gkj;i  gik;j gji;k gki;j  gjk;i
2 2
1 1
g gkj;i  gik;j gij;k gik;j  gkj;i 2:191
2 ij;k ......  2  ......
gij;k

Therefore, the derivatives of the covariant metric coefficient gij with respect to
uk can be expressed in the first-kind Christoffel symbols.

ogij
gij;k  Cikj Cjki 2:192
ouk

Using Eq. (2.173a), Eq. (2.192) can be rewritten in the second-kind Christoffel
symbols.

gij;k Cikj Cjki

2:193
gjl Clik gil Cljk

Similar to Eq. (2.192), one can write

gjk;i Cjik Ckij 2:194

Subtracting Eq. (2.192) from Eq. (2.194), one obtains the relation

gij;k  gjk;i Cikj Cjki  Cjik  Ckij

 
Cjki  Cjik 2:195
Ckji  Cijk
84 2 Tensor Analysis

2.5.9 Covariant Derivatives of Tensors

(a) Contravariant first-order tensors with components T i
The contravariant first-order tensor (vector) T can be written in the covariant basis.

T T i gi 2:196

Using Eq. (2.158), the derivative of the contravariant tensor T with respect to uj
results in
 
T;j T i gi ;j T;ji gi T i gi;j
2:197
T;ji gi T i Ckij gk

The derivative of the contravariant tensor component Ti with respect to uj in

Eq. (2.197) can be defined as

oT i
T;ji  2:198
ou j

Interchanging i with k in the second term in the RHS of Eq. (2.197), one obtains

T i Ckij gk T k Cikj gi
2:199
T k Cijk gi

Substituting Eq. (2.199) into Eq. (2.197), one obtains the derivative of T with
respect to uj.

T;j T;ji gi T k Cijk gi

T;ji Cijk T k gi 2:200

 T i  j gi

Therefore, the covariant derivative with respect to uj of the contravariant first-

order tensor (vector) can be written as

T i j T;ji Cijk T k T;j :gi 2:201

The covariant derivative of the contravariant first-order tensor component is

transformed in the new barred coordinates fui g.

  oui oun
T i k T j n j k
ou ou 2:202a
  ou j ouk
) T n T i k i n
j
ou ou
2.5 Tensor Calculus in General Curvilinear Coordinates 85

Proof The first-order tensor component can be written as

i
ou
T i T j j 2:202b
ou

Differentiating Ti with respect to uk and using the chain rule of differentiation,

one obtains

oT i oui
T j j
ouk ou ;k


n 2:202c
oT j oun oui  j o2 ui ou
n k j
T
ou ou ou ou ou ouk
j n

Using Eq. (2.183), we have

o2 ui i p
 k ou  Ci ou ou
q
C nj pq 2:202d
oun ou j ouk oun ou j

Substituting Eq. (2.202d) into Eq. (2.202c) and interchanging the indices k with
j and j with m, one obtains
j n
i
n
oT i oT ou ou j C
i p
 k ou  Ci ou ou ou
q
T nj pq
ouk oun ouk ou j ouk oun ou j ouk

thus,

oT i  j i oup ouq oun oT j oui oun i

 k ou ou
n

k
T Cpq n j k
n j k T j C nj
ou ou ou ou ou ou ou ou ouk
k

i n
2:202e
oT j oui oun  mC  j ou ou
n T nm
ou ou j ouk ou j ouk

The terms in the RHS of Eq. (2.202e) can be written as

i n
h i i n
oT j oui oun  j ou ou T j T m C
mC  j ou ou
T nm ;n nm
oun ou j ouk ou j ouk ou j ouk
2:202f
 oui oun
T j n j k
ou ou

Using Eq. (2.202b) and interchanging the indices p with k and q with m, the
terms in the LHS of Eq. (2.202e) are rearranged in
86 2 Tensor Analysis

p q n

oT i  j i oup ouq oun i m o uj ou ou ou

k
T Cpq n j k
T;k T m
Cipq
ou ou ou ou ou oun ou j ouk
k m n
j
ou ou ou ou 2:202g
T;ki T m Cikm
oun ou j ouk oum
h i 
T i T m Ci  T i 
;k km k

Substituting Eqs. (2.202f) and (2.202g) into Eq. (2.202e), one obtains
Eq. (2.202a).
  oui oun
T i k T j n j k
ou ou
  ou j ouk
) T n T i k i n q:e:d:
j
ou ou

(b) Covariant first-order tensors with components Ti

The covariant first-order tensor (vector) T can be written in the contravariant basis.

T Ti gi 2:203

Using Eq. (2.189), the partial derivative of the tensor T results in

 
T;j Ti gi ;j Ti;j gi Ti gi;j
2:204
Ti;j gi  Ti Cijk gk

The partial derivative of the covariant tensor component Ti with respect to uj in

Eq. (2.204) can be defined as

oTi
Ti;j  2:205
ou j

Interchanging i with k in the second term in the RHS of Eq. (2.204), one obtains

Ti Cijk gk  Tk Ckji gi Tk Ckij gi 2:206

Substituting Eq. (2.206) into Eq. (2.204), one obtains the derivative of the first-
order tensor component (vector) T with respect to uj.

T;j Ti;j gi  Tk Ckij gi

Ti;j  Ckij Tk gi 2:207

 T i  j gi
2.5 Tensor Calculus in General Curvilinear Coordinates 87

Therefore, the covariant derivative of the tensor component Ti with respect to uj

can be defined as

Ti j Ti;j  Ckij Tk T;j :gi 2:208

The covariant derivativeof the first-order tensor component Ti is transformed in

the new barred coordinates ui , similar to Eq. (2.202a), cf. (Nayak 2012; De et al.
2012).

 oui ou j
Tk jl Ti j k l
ou ou 2:209
 ouk oul
) Ti j Tk jl i j
ou ou

(c) Second-order tensors

Second-order tensors can be written in different expressions with covariant and
contravariant bases.

T Tij gi gj T ij gi gj Tij gi gj Tji gi gj 2:210

Similarly, the covariant derivatives with respect to uk of the second-order tensor

components of T can be calculated as, cf. (Klingbeil 1966; Nayak 2012; De et al.
2012)

Tij jk Tij;k  Cm m
ik Tmj  Cjk Tim

T ij jk T;kij Cikm T mj Ckm

j
T im
2:211a
Tji jk Tj;k
i
Cikm Tjm  Cm i
jk Tm

Tij jk Ti;k
j
 Cm j j m
ik Tm Ckm Ti

where Tij;k ; T;kij ; and Tj;k

i
are the partial derivatives with respect to uk of the
covariant, contravariant, and mixed tensor components. Note that they are different
to the covariant derivatives of the tensor components, as defined in Eq. (2.211a).
In the coordinate transformation from the curvilinear coordinates {ui} to the
new barred curvilinear coordinates fua g, the covariant derivative of the covariant
second-order tensor with respect to uc can be calculated using the chain rule of
differentiation, similar to Eq. (2.202a), cf. (Nayak 2012; De et al. 2012).

  oui ou j ouk
Tab c Tij k a b c 2:211b
ou ou ou
88 2 Tensor Analysis

where the partial derivatives ua;i are called the shift tensor between two coordinate
systems. This relation in Eq. (2.211b) is the chain rule of the covariant derivatives
of the second-order tensors in the coordinate transformation.
Analogously, the covariant derivatives of the second-order tensors of different
types in the new barred curvilinear coordinates are calculated using the shift
tensors.

  oui ou j ouk
Tab c Tij k a b c ;
ou ou ou
  oua oub ouk
T ab c T ij k i j c ; 2:211c
ou ou ou
  oua ou j ouk
 
Tba  Tji  i b c :
c k ou o u ou

2.5.10 RiemannChristoffel Tensor

The RiemannChristoffel tensor is closely related to the Gaussian curvature of the

surface in differential geometry that will be discussed in Chap. 3.
At first, let us look into the second covariant derivative of an arbitrary first-
order tensor of which the first covariant derivative with respect to uj has been
derived in Eq. (2.208)

Ti j Ti;j  Ckij Tk 2:212

Obviously, the covariant derivative Ti|j is a second-order tensor component.

Differentiating Ti|j with respect to uk, the first covariant derivative of the sec-
ond-order tensor (component) Ti|j is the second covariant derivative of an arbitrary
first-order tensor (component) Ti. This second covariant derivative has been given
from Eq. (2.211a) (Klingbeil 1966).
 
Ti jk  Ti j jk
 
 
Ti j ;k  Cm m
ik Tm j  Cjk Ti jm 2:213
 
Ti j; k  Cik Tm j  Cjk Ti jm
m m

Equation (2.212) delivers the relations of

  
Ti j;k Ti;jk  Cm
ij;k Tm Cm
ij Tm;k 2:214a

  
Cm  m n
ik Tm j Cik Tm;j  Cmj Tn 2:214b
2.5 Tensor Calculus in General Curvilinear Coordinates 89

 
Cm m n
jk Ti jm Cjk Ti;m  Cim Tn 2:214c

Inserting Eqs. (2.214a), (2.214b), and (2.214c) into Eq. (2.213), one obtains the
second covariant derivative of Ti.
  
Ti jk Ti j; k  Cm  m
ik Tm j  Cjk Ti jm
Ti;jk  Cm m
ij;k Tm Cij Tm;k
 Cm n m n
ik Tm;j  Cmj Tn  Cjk Ti;m  Cim Tn 2:215
Ti;jk  Cm
ij;k Tm Cmij Tm;k
 Cm
ik Tm;j Cik Cmj Tn  Cm
m n
jk Ti;m Cm n
jk Cim Tn

where the second partial derivative of Ti is symmetric with respect to j and k:

o2 Ti o2 Ti
Ti;jk   Ti;kj 2:216
ou j ouk ouk ou j

Interchanging the indices j with k in Eq. (2.215), one obtains


Ti kj Ti;kj  Cm m
ik;j Tm  Cik Tm;j
2:217
 Cm m n m m n
ij Tm;k Cij Cmk Tn  Ckj Ti;m Ckj Cim Tn

Using the symmetry properties given in Eqs. (2.164) and (2.216), Eq. (2.217)
can be rewritten as

Ti kj Ti;jk  Cm m
ik;j Tm  Cik Tm;j
2:218
 Cm m n m m n
ij Tm;k Cij Cmk Tn  Cjk Ti;m Cjk Cim Tn

In a flat space, the second covariant derivatives in Eqs. (2.215) and (2.218) are
identical. However, they are not equal in a curved space because of its surface
curvature. The difference of both second covariant derivatives is proportional to
the curvature tensor. Subtracting Eq. (2.215) from Eq. (2.218), the curvature
tensor results in
 
Ti jk  Ti kj Cnik;j  Cnij;k Cm n m n
ik Cmj  Cij Cmk Tn
2:219
 Rnijk Tn

The RiemannChristoffel tensor (Riemann curvature tensor) can be expressed as

Rnijk  Cnik;j  Cnij;k Cm n m n

ik Cmj  Cij Cmk 2:220
90 2 Tensor Analysis

Straightforwardly, the RiemannChristoffel tensor is a fourth-order tensor with

respect to the indices of i, j, k, and n. They contain 81 (=34) components in a three-
dimensional space.
In Eq. (2.220), the partial derivatives of the Christoffel symbols are defined by

oCnik oCnij
Cnik;j ; Cnij;k 2:221
ou j ouk

Furthermore, the covariant Riemann curvature tensor of fourth order is defined

by the RiemannChristoffel tensor and covariant metric coefficients.

Rlijk gln Rnijk , Rnijk gln Rlijk 2:222

The Riemann curvature tensor has four following properties using the relation
given in Eq. (2.222) (Nayak 2012):
First skew symmetry with respect to l and i:

Rlijk Riljk 2:223

Second skew symmetry with respect to j and k:

Rlijk Rlikj ;
2:224
Rnijk Rnikj

Block symmetry with respect to two pairs (l, i) and (j, k):

Rlijk Rjkli 2:225

Cyclic property in i, j, k:

Rlijk Rljki Rlkij 0;

2:226
Rnijk Rnjki Rnkij 0

Equation (2.226) is called the Bianchi first identity.

Resulting from these properties, there are six components of Rlijk in the three-
dimensional space as follows (Klingbeil 1966):

Rlijk R3131 ; R3132 ; R3232 ; R1212 ; R3112 ; R3212 2:227

In Cartesian coordinates, all Christoffel symbols equal zero according to

Eq. (2.184). Therefore, the RiemannChristoffel tensor, as given in Eq. (2.220)
must be equal to zero.
2.5 Tensor Calculus in General Curvilinear Coordinates 91

Rnijk Cnik;j  Cnij;k Cm n m n

ik Cmj  Cij Cmk
2:228
0

Thus, the Riemann surface curvature tensor can be written as

Rlijk gln Rnijk 0 2:229

In this case, Euclidean N-space with orthonormal Cartesian coordinates is

considered as a flat space because the Riemann curvature tensor there equals zero.
In the following section, the Riemann curvature tensor can be calculated from
the Christoffel symbols of first and second kinds.
From Eq. (2.222), the Riemann curvature tensor can be rewritten as

Rhijk ghl Rlijk

Using Eq. (2.220), the Riemann curvature tensor results in

Rhijk ghn Rnijk

ghn Cnik;j  Cnij;k Cm n m n
ik Cmj  Cij Cmk

oghn Cnik n
oghn Cnij
 Cik g hn;j  Cnij ghn;k ghn Cm n m n
ik Cmj  ghn Cij Cmk
ou j ouk
oghn Cnik n
oghn Cnij
 Cik ghn;j  Cnij ghn;k Cm m
ik Cmjh  Cij Cmkh
ou j ouk
2:230a

Changing the index m into n in both last terms on the RHS of Eq. (2.230a), one
obtains
n
oghn Cnik oghn Cij
Rhijk  Cnik ghn;j  Cnij ghn;k Cnik Cnjh  Cnij Cnkh
ou j ouk 2:230b
n
oghn Cnik oghn Cij n n
  Cik ghn;j  Cnjh Cij ghn;k  Cnkh
ou j ouk

Using Eq. (2.192) for the first-kind Christoffel symbols in Eq. (2.230b), the
Riemann curvature tensor becomes

oCikh oCijh
Rhijk ghn Rnijk 
ou j ouk
 Cik Chjn Cnjh  Cnjh Cnij Chkn Cnkh  Cnkh
n 2:230c
 
Cikh;j  Cijh;k  Cik Chjn Cnij Chkn
n
92 2 Tensor Analysis

2.5.11 Riccis Lemma

The covariant derivative of the metric covariant coefficient gij with respect to uk
results from Eq. (2.211a) changing Tij into gij. Then, using Eq. (2.193), one
obtains

ogij  m n

gij jk  g mj Cik g in Cjk
ouk 2:231a
gij;k  gij;k 0 ! q:e:d:

Therefore,

ogij
gij;k  gmj Cm n
ik gin Cjk 2:231b
ouk

The Kronecker delta is the product of the covariant and contravariant metric
coefficients:

dlj gli gij

The partial derivative with respect to uk of the Kronecker delta (invariant)

equals zero and can be written as

j   ogli ogij
dl;k gli gij ;k k gij gli k
ou ou
ij ij
gli;k g gli g;k 2:232a

Multiplying Eq. (2.232a) by glm, one obtains

ij
gij glm gli;k dm
i g;k 0
2:232b
) gmj ij lm
;k g g gli;k

Interchanging the indices, it results in

gmj ij lm
;k g g gli;k
2:232c
) gij;k gmi gnj gmn;k

Using Eq. (2.211a), the covariant derivative of the metric contravariant coef-
ficient gij with respect to uk can be written as
2.5 Tensor Calculus in General Curvilinear Coordinates 93

 ogij
gij k k gmj Cikm gim Ckm j
ou 2:233a
gij;k gmj Cikm gim Ckm
j

Substituting Eqs. (2.231b), (2.232c), and (2.233a), one obtains after inter-
changing the indices
  j 
gij k gij;k gmj Cikm gim Ckm
 j 
gmi gnj gmn;k gmj Cikm gim Ckm
j  j 
gmi Cmk  gnj Cink gmj Cikm gim Ckm 2:233b
j  j 
gim Ckm  gmj Cikm gmj Cikm gim Ckm
0 ! (q:e:d:

Note that Eqs. (2.231a) and (2.233b) are known as Riccis lemma.

2.5.12 Derivative of the Jacobian

In the following section, the derivative of the Jacobian J can be calculated and its
result is very useful in the Nabla operator [cf. (Nayak 2012; De et al. 2012)].
The determinant of the metric coefficient tensor is given from Eq. (2.17):
 
 g11 g12 : g1N 

g g22 : g2N 
detgij  21 g J2 [ 0 2:234
 : : : : 
 gN1 gN2 : gNN 

The contravariant metric coefficient gij results from the cofactor Gij of the
covariant metric coefficient gij and the determinant g.

Gij
gij ) Gij ggij 2:235
g

Differentiating both sides of Eq. (2.234) with respect to uk, one obtains

og ogij
k
k Gij for i; j 1; 2; . . .; N 2:236
ou ou

Prove Eq. (2.236):

94 2 Tensor Analysis

   
 og11 og12 og1N   g11 g12 : g1N 
   
 ouk ouk : ouk   g21 g22 : g2N 
og  g g22 : g2N  . . . 
 21 

ouk  :   : : : : 
 : : :  
 ogN1 ogN2 ogNN 

g gN2 : gNN   k :  2:237
N1 ou ouk ouk
og11 11 og12 12 ogNN NN
k G k G ... ... G
ou ou ouk
ogij
k Gij for i; j 1; 2; . . .; N ! (q:e:d:
ou

Substituting Eq. (2.235) into Eq. (2.236), it gives

og ogij ij
G
ouk ouk
ogij 2:238
k ggij
ou
 gij;k ggij

Inserting Eq. (2.231b) into Eq. (2.238), one obtains

og
gij;k ggij
ouk  
gmj Cm ik g in Cn
jk gg
ij

  2:239
g dim Cm j n
ik dn Cjk
 
g Ciik Cjkj 2gCiik

Using the chain rule of differentiation, the Christoffel symbol in Eq. (2.239) can
be expressed in the Jacobian J.
p
1 og oln g
Ciik
2g ouk ouk
2:240
oln J 1 oJ

ouk J ouk

Prove that Riijk 0

Using Eq. (2.220) for n = i, the RiemannChristoffel tensor can be written as

Riijk Ciik;j  Ciij;k Cm i m i

ik Cmj  Cij Cmk
2.5 Tensor Calculus in General Curvilinear Coordinates 95

Interchanging j with k in the last term on the RHS of the equation, one obtains

Riijk Ciik;j  Ciij;k Cm i m i

ik Cmj  Cik Cmj

Ciik;j  Ciij;k

Using Eq. (2.240), the above RiemannChristoffel tensor can be rewritten as

Riijk Ciik;j  Ciij;k

oln J oln J
 2:241
ou j ouk ouk ou j
0 ! (q:e:d:

2.5.13 Ricci Tensor

Both Ricci and Einstein tensors are very useful mathematical tools in the relativity
theory. Note that tensors using in the relativity fields have been mostly written in
the abstract index notation defined by Penrose (Penrose 2005). This index notation
uses the indices to express the tensor types, rather than their covariant components
in the basis {gi}. The first-kind Ricci tensor results from the index contraction of
k and n for n = k of the RiemannChristoffel tensor, as given in Eq. (2.220).

Rij  Rkijk
oCkik oCij
k 2:242
 k  Crij Ckrk Crik Ckrj
ou j ou

The second-kind Ricci tensor can be defined as

Rij  gik Rkj

m m

oCkm oCkj 2:243

gik   Cm n
C
kj mn Cm n
C
kn mj
ou j oum

Using the Christoffel symbol in Eq. (2.240), we obtain

1 oJ oln J
Cijj i
; 2:244
J ou oui
96 2 Tensor Analysis

The first-kind Ricci tensor can be rewritten as

k
o2 ln J oCij oln J
Rij i j
 k  Ckij Crik Ckrj
ou ou ou ouk !
o2 ln J oCkij k oln J
 Cij Crik Ckrj
oui ou j ouk ouk
! 2:245
o2 ln J 1 oCkij oJ
 J k Ckij k Crik Ckrj
oui ou j J ou ou
k
o2 ln J 1 oJCij
 Crik Ckrj
oui ou j J ouk

Interchanging i with j in Eq. (2.245), one obtains the relation of

k
o2 ln J 1 oJCji
Rji  Crjk Ckri
ou j oui J ouk
k
o2 ln J 1 oJCij 2:246
 Ckir Crkj
oui ou j J ouk
Rij

This result indicates that the first-kind Ricci tensor is symmetric with respect to
i and j.
Substituting Eq. (2.244) into Eq. (2.243), the second-kind Ricci tensor results
in
2 m

o ln J 1 oJCkj
Rij g ik
 m n
Ckn Cmj 2:247
ouk ou j J oum

The Ricci curvature R can be defined as

R  Rii
2:248
gij Rji gij Rij

Substituting Eq. (2.246) into Eq. (2.248), the Ricci curvature results in
!
k
ij o2 ln J 1 oJCij
Rg  Ckir Crkj 2:249
oui ou j J ouk
2.5 Tensor Calculus in General Curvilinear Coordinates 97

2.5.14 Einstein Tensor

The Einstein tensor is defined by the second-kind Ricci tensor, Kronecker delta,
and the Ricci curvature.

1
Gij  Rij  dij R 2:250a
2

The Einstein tensor is a mixed second-order tensor and can be written as

Gij gik Gkj 2:250b

Using the tensor contraction rules, the covariant Einstein tensor results in

1
Gij gik Gkj gik Rkj  dkj R
2
1
Rij  gij R
2 2:251
1
Rji  gji R
2
Gji

This result proves that the covariant Einstein tensor is symmetric due to the
symmetry of the Ricci tensor.
The Bianchi first identity in Eq. (2.226) gives

Rlijk Rljki Rlkij 0;

2:252
Rnijk Rnjki Rnkij 0

Differentiating covariantly Eq. (2.252) with respect to um, uk, and ul and then
multiplying it by the covariant metric coefficients gin, one obtains the Bianchi
second identity, cf. Nayak (2012), De et al. (2012), Lee (2000), and Helgason
(1978).
  
  
Rnjkl  Rnjlm  Rnjmk  0:gin
 m
 k
l 2:253
) Rijkl m Rijlm k Rijmk l 0

Due to skew symmetry of the covariant Riemann curvature tensors, as dis-

cussed in Eqs. (2.223) and (2.224), Eq. (2.253) can be rewritten as
  
Rijkl m Rijml k Rjimk l 0 2:254

Multiplying Eq. (2.254) by gilgjk and using the tensor contraction rules (cf.
Sect. 2.3.5), one obtains
98 2 Tensor Analysis

  
Rijkl m Rijml k Rjimk l 0 ,
  
gil gjk Rijkl m gil gjk Rijml k gil gjk Rjimk l
  
gjk Rjk m gjk Rjm k gil Rim l
 
Rjm Rkm k Rlm l l ! k

Rjm 2Rkm k
0

Thus,
 1
Rkm k Rjm 2:255
2

Using Eq. (2.232a) and the symmetry of the Christoffel symbols, the covariant
derivative of the Kronecker symbol with respect to uk is equal to zero.


dij  dij;k Cikm dm m i
j  Cjk dm
k
 
dij;k Cikj  Cijk 2:256

Differentiating covariantly the Einstein tensor in Eq. (2.250a) with respect to uk

and using Eq. (2.256), one obtains the covariant derivative
  
 1 
Gij  Rij  dij R 
k 2
  k 

i 1  
Rj   dij  R dij R 2:257
k 2 k k
 
 1 
Rij   dij R
k 2 k

Changing the index i into k in Eq. (2.257) and using Eq. (2.255), the divergence
of the Einstein tensor equals zero.
   
  1   1
Gkj  Rkj   dkj R Rkj   Rjj 0
k k 2 k k 2
 2:258

) DivG  r:G Gkj  gj 0 q:e:d:
k

This result is very important and has been often used in the general relativity
theories and other relativity fields.
References 99

References

De UC, Shaikh A, Sengupta J (2012) Tensor calculus, 2nd edn. Alpha Science International Ltd.,
Oxford
Fecko M (2011) Differential geometry and lie groups for physicists. Cambridge University Press,
Cambridge
Helgason S (1978) Differential geometry, lie groups and symmetric spaces. AMS, New York
Itskov M (2010) Tensor algebra and tensor analysis for engineerswith applications to
continuum mechanics, 2nd edn. Springer, Berlin
Klingbeil E (1966) Tensorrechnung fr ingenieure (in German). B.I.-Wissenschafts-Verlag,
Zrich
Lee J (2000) Introduction to smooth manifolds. Springer, New York
Nayak PK (2012) Textbook of tensor calculus and differential geometry. PHI Learning, New
Delhi
Oeijord NK (2005) The very basics of tensors. IUniverse, Inc., New York
Penrose R (2005) The road to reality. Alfred A. Knopf, New York
Simmonds JG (1982) A brief on tensor analysis, 2nd edn. Springer, New York
http://www.springer.com/978-3-662-43443-7