2.3.

EIGENVALUES, EIGENVECTORS, AND TENSOR INVARIANTS 49

Example 2.14
For a given stress tensor, that we will take to be symmetric though the theory applies to non-
symmetric tensors as well,
 
1 2 4
Tij = T =  2 3 −1  , (2.186)
4 −1 1
(1) (2) (3)
find the three basic tensor invariants of stress IT , IT , and IT , and show they are truly invariant
when the tensor is subjected to a rotation with direction cosine matrix of
 q 
√1 2 √1
3
 6 6

ℓij = Q =  √1 − √13 √1 . (2.187)
3 3
√1 0 − √12
2

Calculation reveals that

det Q = 1, (2.188)
and that Q · QT = I, so that QT is a rotation matrix. The eigenvalues of T, that are the principal values
of stress, are easily calculated to be

λ(1) = 5.28675, λ(2) = −3.67956, λ(3) = 3.39281. (2.189)

The three invariants of Tij are

 
1 2 4
(1)
IT = tr T = tr  2 3 −1  = 1 + 3 + 1 = 5, (2.190)
4 −1 1
(2) 1

IT = (tr T)2 − tr (T · T) , (2.191)
2
  2   
1 2 4 1 2 4 1 2 4
1   
=  tr 2 3 −1  − tr  2 3 −1   2 3 −1  , (2.192)
2
4 −1 1 4 −1 1 4 −1 1
  
21 4 6
1 2
= 5 − tr  4 14 4  , (2.193)
2
6 4 18
1
= (25 − 21 − 14 − 18), (2.194)
2
= −14, (2.195)
 
1 2 4
(3)
IT = det T = det  2 3 −1  = −66. (2.196)
4 −1 1

© 01 July 2025, Joseph M. Powers. All rights reserved.

50 CHAPTER 2. SOME NECESSARY MATHEMATICS

Now when we rotate the tensor T, we get a transformed tensor given by

  q
√1 √1 √1   √1 2 √1

q6 3 2 4 2 1 3
  6
 6

T′ = QT · T · Q =  23 − √13 0 2
3 −1   √1 1
− √3 √1 , (2.197)
3 3
√1 √1 −1 1 − √12 4
√1 0 − √12
6 3 2
 
4.10238 2.52239 1.60948
=  2.52239 −0.218951 −2.91291  . (2.198)
1.60948 −2.91291 1.11657

We then seek the tensor invariants of T′ . Leaving out some of the details, that are the same as those
for calculating the invariants of the T, we find the invariants indeed are invariant:
(1)
IT = 4.10238 − 0.218951 + 1.11657 = 5, (2.199)
(2) 1 2
IT = (5 − 53) = −14, (2.200)
2
(3)
IT = −66. (2.201)

Finally, we verify that the stress invariants are indeed related to the principal values (the eigenvalues
of the stress tensor) as follows
(1)
IT = λ(1) + λ(2) + λ(3) = 5.28675 − 3.67956 + 3.39281 = 5, (2.202)
(2) (1) (2) (2) (3) (3) (1)
IT = λ λ +λ λ +λ λ , (2.203)
= (5.28675)(−3.67956) + (−3.67956)(3.39281) + (3.39281)(5.28675) = −14, (2.204)
(3)
IT = λ(1) λ(2) λ(3) = (5.28675)(−3.67956)(3.39281) = −66. (2.205)

Example 2.15
For a given two-dimensional stress tensor, which here we will take to be asymmetric,
 
2 1
Tij = T = , (2.206)
2 2

(1) (2)
find the two basic tensor invariants of stress IT and IT and show they are truly invariant when the
tensor is subjected to a rotation with direction cosine matrix of
1 1
!
√ √
2 2
ℓij = Q = . (2.207)
− √12 √1
2

Calculation reveals that det Q = 1 and that Q · QT = I, so that QT is a rotation matrix. The
eigenvalue problem induces the condition

T11 − λ T12
= 0. (2.208)
T21 T22 − λ

© 01 July 2025, Joseph M. Powers. All rights reserved.

2.4. GRAD, DIV, CURL, ETC. 51

This give the characteristic polynomial

(T11 − λ)(T22 − λ) − T12 T21 = 0, (2.209)
2
λ − (T11 + T22 )λ + (T11 T22 − T12 T21 ) = 0, (2.210)
(1) (2)
λ2 − IT λ + IT = 0. (2.211)
Here we have for the two-dimensional system, the two invariants
(1)
IT = T11 + T22 = λ(1) + λ(2) = tr T, (2.212)
(2) (1) (2)
IT = T11 T22 − T12 T21 = λ λ = det T. (2.213)
For this system, the eigenvalues of T are easily calculated to be
√ √
λ(1) = 2 + 2, λ(2) = 2 − 2. (2.214)
The two invariants of T are
 
(1) 2 1
IT = tr T = tr = 2 + 2 = 4, (2.215)
2 2
 
(2) 2 1
IT = det T = det = 2(2) − 1(2) = 2. (2.216)
2 2
Now when we rotate the tensor T, we get a transformed tensor given by
!  !  
√1 − √12 √1 √1 1
′ T 2 2 1 2 2 − 21
T =Q ·T·Q= = 21 7 . (2.217)
√1 √1 2 2 − 2
√1 √1
2 2 2 2 2

By inspection, we see the invariants of T′ are indeed the same as those of T:

(1) 1 7
IT = + = 4, (2.218)
 2  
2  
(2) 1 7 −1 1
IT = − = 2. (2.219)
2 2 2 2
Finally, we also see
(1) √ √
IT = λ(1) + λ(2) = (2 + 2) + (2 − 2) = 4, (2.220)
 √  √ 
(2)
IT = λ(1) λ(2) = 2 + 2 2 − 2 = 2. (2.221)

2.4 Grad, div, curl, etc.

Thus far, we have mainly dealt with the algebra of vectors and tensors. Now let us consider
the calculus. For now, let us consider variables that are a function of the spatial vector xi .
We shall soon allow variation with time t also. We will typically encounter quantities such
as
• φ (xi ) → a scalar function of the position vector,
• vj (xi ) → a vector function of the position vector, or
• Tjk (xi ) → a tensor function of the position vector.

© 01 July 2025, Joseph M. Powers. All rights reserved.

52 CHAPTER 2. SOME NECESSARY MATHEMATICS

2.4.1 Gradient operator

The gradient operator, sometimes denoted by “grad,” is motivated as follows. Consider
φ(xi ), that when written in full is

φ(xi ) = φ(x1 , x2 , x3 ). (2.222)

Taking a derivative using the chain rule gives

∂φ ∂φ ∂φ
dφ = dx1 + dx2 + dx3 . (2.223)
∂x1 ∂x2 ∂x3

Following Panton (2013), we define a non-traditional, but useful further notation ∂i for the
partial derivative
 ∂   
∂x1 ∂1
∂ ∂ ∂ ∂
∂i ≡ = e1 + e2 + e3 = ∇ =  ∂
∂x2
 =  ∂2  . (2.224)
∂xi ∂x1 ∂x2 ∂x3 ∂ ∂3
∂x3

So the chain rule is actually

dφ = ∂1 φ dx1 + ∂2 φ dx2 + ∂3 φ dx3 , (2.225)

that is written using our summation convention as

dφ = ∂i φ dxi . (2.226)

After commuting so as to juxtapose the i subscript, we have

dφ = dxi ∂i φ. (2.227)

In Gibbs notation, we say

dφ = dxT · ∇φ = dxT · grad φ. (2.228)
We can also take the transpose of both sides, recalling that the transpose of a scalar is the
scalar itself, to obtain

T
(dφ)T = dxT · ∇φ , (2.229)
dφ = (∇φ)T · dx, (2.230)
dφ = ∇T φ · dx. (2.231)

Here we expand ∇T as ∇T = (∂1 , ∂2 , ∂3 ). When ∂i or ∇ operates on a scalar, it is known as

the gradient operator. The gradient operator operating on a scalar function gives rise to a
vector function.

© 01 July 2025, Joseph M. Powers. All rights reserved.

2.4. GRAD, DIV, CURL, ETC. 53

We next describe the gradient operator operating on a vector. For vectors in Cartesian
index and Gibbs notation, we have, following a similar analysis12
dvi = dxj ∂j vi = ∂j vi dxj , (2.232)
dvT = dxT · ∇vT , (2.233)
dv = (∇vT )T · dx, (2.234)
dv = (grad v)T · dx. (2.235)
Here the quantity ∂j vi is the gradient of a vector, that is a tensor. So the gradient operator
operating on a vector raises its order by one. The Gibbs notation with transposes suggests
properly that the gradient of a vector can be expanded as
   
∂1 ∂1 v1 ∂1 v2 ∂1 v3
∇vT =  ∂2  ( v1 v2 v3 ) =  ∂2 v1 ∂2 v2 ∂2 v3  . (2.236)
∂3 ∂3 v1 ∂3 v2 ∂3 v3
Lastly we consider the gradient operator operating on a tensor. For tensors in Cartesian
index notation, we have, following a similar analysis
dTij = dxk ∂k Tij = ∂k Tij dxk . (2.237)
Here the quantity ∂k Tij is a third order tensor. So the gradient operator operating on a
tensor raises its order by one as well. The Gibbs notation is not straightforward as it can
involve something akin to the transpose of a three-dimensional matrix.

2.4.2 Divergence operator

The contraction of the gradient operator on either a vector or a tensor is known as the
divergence, sometimes denoted by “div.” For the divergence of a vector, we have
∂i vi = ∂1 v1 + ∂2 v2 + ∂3 v3 = ∇T · v = div v. (2.238)
The divergence of a vector is a scalar. A vector field that is divergence-free, ∇T · v = 0, is
defined as solenoidal.
For the divergence of a second order tensor, we have
∂i Tij = ∂1 T1j + ∂2 T2j + ∂3 T3j = ∇T · T = div T. (2.239)
The divergence operator operating on a tensor gives rise to a row vector. We will sometimes
have to transpose this row vector in order to arrive at a column vector, e.g. we will have

T
need for the column vector ∇T · T . We note that, as with the vector inner product, most
texts assume the transpose operation is understood and write the divergence of a vector or
tensor simply as ∇ · v or ∇ · T.
12
A more common approach, not using the transpose notation, would be to say here for the Gibbs notation
that dv = dx·∇v. However, this is only works if we consider dv to be a row vector, as dx·∇v is a row vector.
All in all, while at times clumsy, the transpose notation allows a for great deal of clarity and consistency
with matrix algebra.

© 01 July 2025, Joseph M. Powers. All rights reserved.

54 CHAPTER 2. SOME NECESSARY MATHEMATICS

2.4.3 Curl operator

The curl operator is the derivative analog to the cross product. We write it in the following
three ways:

ωi = ǫijk ∂j vk , (2.240)
ω = ∇ × v, (2.241)
ω = curl v. (2.242)

Expanding for i = 1, 2, 3 gives

ω1 = ǫ123 ∂2 v3 + ǫ132 ∂3 v2 = ∂2 v3 − ∂3 v2 , (2.243)

ω2 = ǫ231 ∂3 v1 + ǫ213 ∂1 v3 = ∂3 v1 − ∂1 v3 , (2.244)
ω3 = ǫ312 ∂1 v2 + ǫ321 ∂2 v1 = ∂1 v2 − ∂2 v1 . (2.245)

2.4.4 Laplacian operator

The Laplacian13 operator can operate on a scalar, vector, or tensor function. It is a simple
combination of first the gradient followed by the divergence. It yields a function of the same
order as that which it operates on. For its most common operation on a scalar, it is denoted
as follows:
∂i ∂i φ = ∇T · ∇ φ = ∇2 φ = div grad φ. (2.246)
In viscous fluid flow, we will have occasion to have the Laplacian operate on vector:

T
T
∂i ∂i vj = ∇T · ∇ vT = ∇2 v T = ∇2 v = div grad v. (2.247)

2.4.5 Time derivative

Following Panton (2013), we will employ a useful but unusual notation for the partial deriva-
tive with respect to time:
∂
∂o ≡ , (2.248)
∂t
that will be used extensively later.

2.4.6 Relevant theorems

We will use several theorems that are developed in vector calculus. Here we give the simplest
of motivations, and simply present them. The reader should consult a standard mathematics
text for detailed derivations.
13
Pierre-Simon Laplace, 1749-1827, Normandy-born French astronomer of humble origin. Educated at
Caen, taught in Paris at École Militaire.

© 01 July 2025, Joseph M. Powers. All rights reserved.

2.4. GRAD, DIV, CURL, ETC. 55

[Link] Fundamental theorem of calculus

The fundamental theorem of calculus is as follows
Z x=b Z x=b  
dφ
f (x) dx = dx = φ(b) − φ(a). (2.249)
x=a x=a dx

It effectively says that to find the integral of a function f (x), that is the area under the
curve, it suffices to find a function φ, whose derivative is f , and evaluate φ at each endpoint,
and take the difference to find the area under the curve.

[Link] Gauss’s theorem

Gauss’s14 theorem is the analog of the fundamental theorem of calculus extended to volume
integrals. It applies to tensor functions of arbitrary order and is as follows:
Z Z
∂i (Tijk...) dV = ni Tijk... dS. (2.250)
R S

Here R is an arbitrary volume, dV is the element of volume, S is the surface that bounds
V , ni is the outward unit normal to S, and Tjk.. is an arbitrary tensor function. The surface
integral is analogous to evaluating the function at the end points in the fundamental theorem
of calculus. In Gibbs notation, we have
Z Z
T
∇ · T dV = nT · T dS. (2.251)
R S

If we take Tjk... to be the scalar of unity (whose derivative must be zero), Gauss’s theorem
reduces to Z
ni dS = 0. (2.252)
S

That is the unit normal to the surface integrated over the surface, cancels to zero when the
entire surface is included. We will use Gauss’s theorem extensively. It allows us to convert
sometimes difficult volume integrals into easier interpreted surface integrals. It is often useful
to use this theorem as a means of toggling back and forth from one form to another.

Example 2.16
Demonstrate the validity of Gauss’s theorem for the tensor field
 
x1 x2 x1
T =  x2 x2 x3  , (2.253)
x3 x3 x2
14
Carl Friedrich Gauss, 1777-1855, Brunswick-born German mathematician, considered the founder of
modern mathematics. Worked in astronomy, physics, crystallography, optics, biostatistics, and mechanics.
Studied and taught at Göttingen.

© 01 July 2025, Joseph M. Powers. All rights reserved.

56 CHAPTER 2. SOME NECESSARY MATHEMATICS

where the volume under consideration is the unit cube defined on the domain x1 ∈ [0, 1], x2 ∈ [0, 1],
x3 ∈ [0, 1].

We first note that T is asymmetric. We easily see that

 
x1 x2 x1
∇T · T = ( ∂1 ∂2 ∂3 )  x2 x2 x3  , (2.254)
x3 x3 x2
= (3 2 1). (2.255)

Integrating the constant row vector over the unit cube, we find
Z
∇T · T dV = ( 3 2 1 ) . (2.256)
V

Then, we can evaluate the surface integral on each of the six faces and perform a set of six surface
integrals. Leaving out the details, we do so, and find
Z
nT · T dS = ( 3 2 1 ) . (2.257)
S

This verifies Gauss’s theorem for this case. For asymmetric tensors such as our T, we need to be careful
about commuting operators. For example, for this problem
 
Z 2
T · n dS =  2  6= ( 3 2 1 ) . (2.258)
S 0

[Link] Stokes’ theorem

Stokes’15 theorem is as follows.
Z I
ni ǫijk ∂j vk dS = vi αi ds. (2.259)
S C

Once again S is a bounding surface and ni is its outward unit normal. The integral with the
circle through it denotes a closed contour integral with respect to arc length s, and αi is the
unit tangent vector to the bounding curve C.
In Gibbs notation, it is written as
Z I
T
n · ∇ × v dS = vT · α ds. (2.260)
S C
15
Sir George Gabriel Stokes, 1819-1903, Irish-born British physicist and mathematician, holder of the Lu-
casian chair of Mathematics at Cambridge University, developed, simultaneously with Navier, the governing
equations of fluid motion, in a form that was more robust than that of Navier.

© 01 July 2025, Joseph M. Powers. All rights reserved.

2.4. GRAD, DIV, CURL, ETC. 57

[Link] A useful identity

It is easy to show that a useful identity involving ∇, v and ω = ∇ × v is as follows:

 
1
vj ∂j vi = ∂i vj vj − ǫijk vj ωk , (2.261)
2
 
T

1 T
v ·∇ v = ∇ v · v − v × ω. (2.262)
2

This is easily proved by considering the right hand side of Eq. (2.261), expanding, and using
Eqs. (2.241) and then (2.74):
 
1
∂i vj vj − ǫijk vj ωk = vj ∂i vj − ǫijk vj ǫklm ∂l vm , (2.263)
2 | {z }
=ωk
= vj ∂i vj − ǫkij ǫklm vj ∂l vm , (2.264)
= vj ∂i vj − (δil δjm − δim δjl ) vj ∂l vm , (2.265)
= vj ∂i vj − vj ∂i vj +vj ∂j vi , (2.266)
| {z }
=0
= vj ∂j vi , QED. (2.267)

[Link] Leibniz’s rule: general transport theorem for arbitrary regions

Leibniz’s16 rule relates time derivatives of integral quantities to a form that distinguishes
changes that are happening within the boundaries to changes due to fluxes through bound-
aries. This is the foundation of the so-called control volume approach. Using the nomen-
clature of Whitaker (1968), p. 92, we also call Leibniz’s rule the general transport theorem.
Leibniz’s rule applied to an arbitrary tensorial function is as follows:
Z Z Z
d
Tjk...(xi , t) dV = ∂o Tjk... dV + nl wl Tjk... dS. (2.268)
dt AR(t) AR(t) AS(t)

• AR(t) → arbitrary region that is a time-dependent volume,

• AS(t) → bounding surface of the arbitrary moving volume,

• wl → velocity vector of points on the moving surface,

• nl → unit normal to moving surface.

16
Gottfried Wilhelm von Leibniz, 1646-1716, Leipzig-born German philosopher and mathematician. In-
vented calculus independent of Newton and employed a superior notation to that of Newton.

© 01 July 2025, Joseph M. Powers. All rights reserved.

58 CHAPTER 2. SOME NECESSARY MATHEMATICS

Say we have the special case in which Tjk... = 1; then Leibniz’s rule reduces to
Z Z Z
d
dV = ∂o (1) dV + nk wk (1) dS, (2.269)
dt AR(t) AR(t) AS(t)
Z
dVAR
= nk wk dS. (2.270)
dt AS(t)

This simply says the total volume of the region, that we call VAR , changes in response to net
motion of the bounding surface.

[Link].1 Material region: Reynolds transport theorem In the special case where
the volume contains the same fluid particles, the velocity of the boundary is the fluid particle
velocity, wl = vl , and our general transport theorem becomes, again using the nomenclature
of Whitaker (1968), p. 92, the Reynolds17 transport theorem:
Z Z Z
d
Tjk...(xi , t) dV = ∂o Tjk... dV + nl vl Tjk... dS. (2.271)
dt M R(t) M R(t) M S(t)

The term MR(t) and MS(t) denote the time-dependent material region and material surface
to denote that the geometry in question always contains the same material particles.

[Link].2 Fixed region In the special case where the volume is fixed in time, the velocity
of the boundary is zero, wl = 0, and our general transport theorem becomes
Z Z
d
Tjk...(xi , t) dV = ∂o Tjk... dV. (2.272)
dt F R FR

In this case there is no time-dependency of the fixed region F R.

[Link].3 Scalar function In the special case where Tjk... is a scalar function f , Leibniz’s
rule reduces to
Z Z Z
d
f (xi , t) dV = ∂o f (xi , t) dV + nl wl f (xi , t) dS. (2.273)
dt AR(t) AR(t) AS(t)

Further, considering one-dimensional cases only, we can then say

Z Z x=b(t)
d x=b(t) db da
f (x, t) dx = ∂o f dx + f (b(t), t) − f (a(t), t). (2.274)
dt x=a(t) x=a(t) dt dt
As in the fundamental theorem of calculus, for the one-dimensional case, we do not have to
evaluate a surface integral; instead, we simply must consider the function at its endpoints.
Here db/dt and da/dt are the velocities of the bounding surface and analogous to wk . The
terms f (b(t), t) and f (a(t), t) are equivalent to evaluating Tjk.. on AS(t).
17
Osborne Reynolds, 1842-1912, Belfast-born British engineer and physicist, educated in mathematics at
Cambridge, first professor of engineering at Owens College, Manchester, did fundamental experimental work
in fluid mechanics and heat transfer.

© 01 July 2025, Joseph M. Powers. All rights reserved.

2.5. GENERAL COORDINATE TRANSFORMATIONS 59

2.5 General coordinate transformations

Here we introduce, following Aris (1962), Powers and Sen (2015), Ch. 1.6, and many others,
some standard notation from tensor analysis for general coordinate transformations. This
extends our analysis of Sec. 2.1.2, that was confined to simple rotation transformations. In
this notation, both sub- and superscripts are needed to distinguish between what are known
as covariant and contravariant vectors, that are really different mathematical representations
of the same quantity, just cast onto different basis vectors. The basis vectors may or may not
be orthonormal. They may not even be orthogonal. All they must be is linearly independent.
In brief, we start with a general transformation from the non-Cartesian coordinate, defined
here as xi , to the Cartesian coordinate, defined here as ξ k :

ξ 1 = ξ 1 (x1 , x2 , x3 ), (2.275)
ξ 2 = ξ 2 (x1 , x2 , x3 ), (2.276)
ξ 3 = ξ 3 (x1 , x2 , x3 ). (2.277)

As an example, this form includes the transformation from a non-Cartesian cylindrical co-
ordinate system to a Cartesian system; this will be taken up in detail in Ch. 7.1. We could
also say

ξ k = ξ k (xi ), ξ = ξ(x). (2.278)

Local linearization of the transformation gives

 1  ∂ξ1 ∂ξ1 ∂ξ 1  
dξ ∂x1 ∂x2 ∂x3 dx1
 dξ 2  =  ∂ξ21 ∂ξ22 ∂ξ 2   dx2  , (2.279)
∂x ∂x ∂x3
dξ 3 ∂ξ 3
1
∂ξ 3 ∂ξ 3 dx3
| ∂x {z2
∂x ∂x3 }
J
dξ = J · dx. (2.280)

Here the local Jacobian matrix J is defined as

 ∂ξ1 ∂ξ1 ∂ξ1 
∂x1 ∂x2 ∂x3 k
2
J =  ∂ξ 1 ∂ξ 2 ∂ξ 2  = ∂ξ . (2.281)
∂x ∂x2 ∂x3 ∂xi
∂ξ 3 ∂ξ 3 ∂ξ 3
∂x1 ∂x2 ∂x3

The chain rule for partial differentiation can be used to represent the gradient as
 ∂   ∂ξ1 ∂ξ2 ∂ξ3   ∂ 
∂x1 ∂x1 ∂x1 ∂x1 ∂ξ 1
 ∂  =  ∂ξ12 ∂ξ 2 ∂ξ 3  ∂ , (2.282)
∂x2 ∂x ∂x2 ∂x2 ∂ξ 2
∂ ∂ξ 1 ∂ξ 2 ∂ξ 3 ∂
∂x3 ∂ξ 3
| {z } | ∂x3 {z3
∂x ∂x3 } | {z }
∇x JT ∇ξ

∇x = J T · ∇ξ . (2.283)

© 01 July 2025, Joseph M. Powers. All rights reserved.

60 CHAPTER 2. SOME NECESSARY MATHEMATICS

Inverting, we find

−1
∇ξ = J T · ∇x . (2.284)

This can be directly compared with Eq. (2.280). In the special case for which the transfor-

−1
−1
mation is a rotation, we have J = Q and thus JT = QT = Q. In this case, we recover
the simpler dξ = Q · dx and ∇ξ = Q · ∇x .
For Cartesian systems, we must have the classical formula for differential distance ds:

2
2
2
ds2 = dξ 1 + dξ 2 + dξ 3 = dξ T · dξ. (2.285)

The differential distance must be an invariant in either coordinate representation, so we must

have

ds2 = dξ T · dξ = (J · dx)T · (J · dx) = dxT · |JT{z· J} ·x = dxT · G · dx. (2.286)

G

Here we have defined the metric tensor

G = JT · J, (2.287)
∂ξ k ∂ξ k
gij = . (2.288)
∂xi ∂xj
One can also show that
1 imn jpq
g ij = ǫ ǫ gmp gnq , (2.289)
2
gik g kj = δij . (2.290)

We also have the Jacobian determinant J defined as

√ ∂ξ k
J= g = det i = det J. (2.291)
∂x
A vector’s contravariant representation is given by v i . Its covariant representation is given
by vi . The metric tensor links one representation to the other via

vj = v i gij , v i = g ij vj . (2.292)

In the remaining paragraphs of this chapter, we present some slightly modified text first
presented by Powers and Sen (2015) in their Ch. 1.6.5 to better understand the nature of
vectors in terms of linear combinations of covariant and contravariant basis vectors. The
only requirement we place on the basis vectors is linear independence: they must point
in different directions. They need not be unit vectors, and their lengths may differ from
one another. Consider the non-orthogonal basis vectors e1 , e2 , aligned with the x1 and x2
directions shown in Fig. 2.6a. The vector v can then be written as

© 01 July 2025, Joseph M. Powers. All rights reserved.