AE 6009/6012.

Cartesian tensors

Appendix A of Pope (2000) contains information more complete than the following.

1 Introduction
A tensor is a mathematical quantity which behaves according to certain transformation rules
when subjected to a change of coordinate systems. The rules depend on the order of the
tensor. If we restrict ourselves to Cartesian coordinate frames then a scalar, vector, and
matrix can often be treated as tensors of order 0, 1, and 2 respectively. In our discussion
of the Einstein summation convention we distinguished between free subscripts and dummy
subscripts. In general, an m-th order tensor has m free subscripts.

2 Zeroth order tensors

An tensor of order 0 is, by definition, a scalar, whose value at a given point in space is
independent of the choice of coordinate systems. It has no free subscripts but may have
dummy subscripts which are repeated in pairs. Examples of zeroth-order tensors include the
pressure, the temperature, the magnitude of any vector, and the trace of a matrix (if it is a
second-order tensor). Note that a scalar may vary with position: e.g. we have the pressure
gradient in the Navier-Stokes equation, and the pressure gradient is a vector.

3 First order tensors

Consider a vector of n elements, which have a well-defined magnitude and orientation in
n-dimensional space. We introduce the unit vectors e(1) , e(2) , ... e(n) . In the case of n = 3,
the elements of these unit vectors are (1,0,0), (0,1,0) and (0,0,1) respectively. Let f be a
vector. The coordinate components of f are defined by the relation

f = fi e(i) (1)

where we have used the summation convention on repeated Roman subscripts. Note that
there are no free subscripts on either side of this equation: the vector f is (in both its
magnitude and orientation) is independent of the choice of coordinate axes.
However, the components of the vector are inherently coordinate-system dependent. Af-
ter all, the jth component of the vector is defined by taking the dot product between the
vector f and the jth coordinate axis:

fj = f · e(j) (2)

1
If we change our coordinate system, then a change in the unit vectors imply a change in the
components of the vector f .
Now, consider a new set of coordinate axes, represented by the unit vectors e′(1) , e′(2) , ...
e′(n) . The components of f in the new coordinates are given by

fj′ = f · e′(j) (3)

Now we substitute for f from Eq. 1:

fj′ = fi e(i) · e′(j) . (4)

Since all the e and e′ are unit vectors, the dot product between them is the cosine of the
angle between the corresponding old and new coordinate axes. Let’s denote the dot product
above by aij . Then we can write
fj′ = fi aij . (5)
Or, equivalently,
fi′ = aij fj . (6)
This is the transformation relation required for a first-order tensor. Note that it has one
(and the same) free subscript on each side. Note that here primes refer to a new coordinate
system, not differentiation.

4 Second-order tensors
By extension of Eq. 6, a matrix B is a second-order tensor if it follows the transformation
relation
Bij′ = aik ajl Bkl (7)

Consider a square 3 × 3 matrix, of elements

 
B11 B12 B13
 B21 B22 B23  (8)
B31 B32 B33

The element in the i-th row and j-th column is denoted by Bij . The transpose of the
matrix B is obtained by interchanging pairs of elements across the diagonal while keeping
the diagonal entries unchanged. We can write

BijT = Bji (9)

A matrix is called symmetric if B T = B, anti-symmetric if B T = −B. Note that the diagonal

elements of any anti-symmetric matrix must be zero.
The trace of a matrix is the sum of its diagonal elements:

tr(B) = Bii (10)

2
The trace of a matrix is also the sum of its eigenvalues. It is a scalar and is independent of
the choice of coordinate axes.
It is a general identity that any square matrix can be written as the sum of a symmetric
part and an anti-symmetric part:
1 1
B = (B + B T ) + (B − B T ) (11)
2 2
In Cartesian tensor notation, this would be
1 1
Bij = (Bij + Bji ) + (Bij + Bji ) (12)
2 2

If B is the matrix of velocity gradients, i.e. Bij = ∂ui /∂xj , then its symmetric part is
called the strain rate, and the antisymmetric part is called the rotation rate. That is:
 
1 ∂ui ∂uj
sij = + (13)
2 ∂xj ∂xi
 
1 ∂ui ∂uj
rij = − (14)
2 ∂xj ∂xi
We observe, from these definitions, that sij ≡ sji and sii = ∂ui /∂xi , whereas rij ≡ −rji . The
latter also implies all diagonal elements of the rotation rate tensor are zero.

5 Vector and matrix algebra

It is useful to write some well known operations on vectors and matrices using Cartesian
tensor notation. Let’s have vectors x and y, and matrices A, B and C.
The dot product of two vectors x and y is

x · y = xi y i . (15)

If C = AB, the (i, j)th element of C is the dot product (or scalar product) between the ith
row of A and jth column of B. We can write

Cij = Aik Bkj (16)

As a special case, (as if B were a column matrix) if Ax = y then

yj = Aij xj (17)

The sum of the diagonal elements of a matrix B is written as Bii . A very special matrix
is the identity matrix, I, whose diagonal elements are all equal to 1, while the off-diagonal
elements are 0. It corresponds to the Kronecker delta tensor, whose elements are defined by

3
 (
1 if i = j
δij = (18)
0 if i ̸= j
It will be important to note that

tr(I) = δ11 + δ22 + δ33 = 3 (19)

Another special property of the Kronecker delta tensor is that, if x is a vector,

δij xj = δi1 x1 + δi2 x2 + δi3 x3 = xi (20)

The second equals sign is explained by noting that only one of δi1 x1 , δi2 x2 , and δi3 x3 will be
nonzero, such that the summation picks up the one of x1 , x2 or x3 . Which one of the three
is determined by the value of i.

6 Vector and matrix calculus

The derivative of any scalar ϕ in the xi direction is written as ∂ϕ/∂xi . The divergence of a
vector f is ∂fi /∂xi . The Laplacian of anything is the sum of the three second derivatives of
that anything. In operator form, we can write

2 ∂2
∇ = (21)
∂xi xi

6.1 The Curl

The curl of the velocity vector is the vorticity, which is important. It turns out, although
the vorticity is a vector, it does not satisfy the transformation rule in Eq. 6 and is thus not
a first-order tensor. To understand this we consider the cross-product between two vectors
A and B, which is another vector whose elements are given by determinant the vector cross
product, written as the determinant
e(1) e(2) e(3)
A×B = A1 A2 A3 . (22)
B1 B2 B3
It is well known that B × A = −A × B, and A × B = 0 if the vectors A and B are parallel
to each other. These properties can be satisfied by defining the ’‘alternating symbol”

 1 if (i, j, k) = (1, 2, 3), (2, 3, 1), or (3, 1, 2)

ϵijk = −1 if (i, j, k) = (1, 3, 2), (2, 1, 3), or (3, 2, 1) (23)

0 otherwise


We can write
(A × B)i = ϵijk Aj Bk (24)

4
Note the properties (i) ϵijk = −ϵjki ; (ii) ϵijk = ϵikj ; (iii) ϵijk = −ϵkji ; (iv) ϵijj = 0.
The curl of a vector A is given by

e(1) e(2) e(3)

∂ ∂ ∂
∇×A= ∂x1 ∂x2 ∂x3 (25)
A1 A2 A3

Thus the vorticity vector can be written as

∂uk
ωi = ϵijk (26)
∂xj

The vorticity vector picks up a change in sign if we switch between left-handed and right-
handed coordinate systems. It is called a “pseudo-vector”.

7 Operations which are invalid

The following operations are meaningless:

1. equating two tensors of different orders (e.g. fi = Aij )

2. adding up the components of a vector (e.g. fi + fj )

3. trying to sum over subscripts appearing more than twice; e.g. ϵiii )

4. dividing anything by a vector or by a matrix

5. trying to add up or equate quantities of different dimensions

Anytime we seem to be getting into situations of the types noted just above, something
has gone wrong.