Mechanics of Solids (ME31013) Mechanical Engineering

Autumn Semester, 2020 IIT Kharagpur

Vectors, Tensors,
∗
Indices and all that . . .

1 Motivation for tensors

You all have come across the word “tensors” and we have said things
like “stress is a tensor” back in Fluid Mechanics. We also have said that
things like “vector is a 1st order tensor", “stress is a 2nd order tensor" and
so on. However, in all of this, we bypassed talking about what a tensor
is in the first place. So let us see if we can motivate its definition.
Consider functions . . .
For a given scalar√ quantity, 𝑠, let us define a function as 𝑓 (𝑠) = 𝑠 2 (might
as well be 𝑠 or 2𝑠 or sin(𝑠) and so on). The important thing to note
3 4

here is that the input to the function 𝑓 is a scalar and its output is also a
scalar. We say that the function 𝑓 maps a scalar to a scalar.
√
For a given vector, 𝐫, let us define a function 𝑓 (𝐫) = |𝐫| = 𝑟12 + 𝑟22 + 𝑟32 ,
where 𝑟1 , 𝑟2 , and 𝑟3 are the components of 𝐫 referred to some chosen
coordinate axes. Here, even though the input to 𝑓 is a vector, the output
is a scalar. So, we say that the function 𝑓 maps a vector to a scalar.
Again, for a given vector, 𝐫, let us define a function 𝑓 (𝐫) = 𝐚 ⋅ 𝐫 where 𝐚 is
some vector intrinsic to the definition of 𝑓 . Again, the function 𝑓 maps
a vector to a scalar.
Now, consider the definition 𝑓 (𝐫) = 𝛼𝐫, where 𝛼 is some scalar intrinsic
to the definition of 𝑓 . This time, the input to 𝑓 is a vector and the output
is also a vector. So, 𝑓 maps a vector to a vector. The mapping is through
the scalar 𝛼.
∗ Notes prepared by Jeevanjyoti Chakraborty. Contact: [email protected]

1
Mechanics of Solids Mechanical Engineering
Vectors, Tensors . . . IIT Kharagpur

Similarly, consider the function definition

⎡𝛼1 𝑟1 ⎤
𝑓 (𝐫) = ⎢𝛼2 𝑟2 ⎥ ,
⎢ ⎥
⎣𝛼3 𝑟3 ⎦
which is a vector with the three components 𝛼1 𝑟1 , 𝛼2 𝑟2 , and 𝛼3 𝑟3 . The
only difference from the previous definition is that while there was the
same scalar 𝛼 being multiplied with all the components of 𝐫, here there is
a different scalar being multiplied with the three components of 𝐫. The
spirit is the same. So, again the function 𝑓 maps a vector to another
vector. The mapping is through the three scalars 𝛼1 , 𝛼2 , and 𝛼3 .
Finally, we can have another type of function defintion which maps a
vector to another vector in such a way that each of the components of
the output vector is made up by using a linear combination of the com-
ponents of the input vector. Consider the following definition

⎡𝛼11 𝑟1 + 𝛼12 𝑟2 + 𝛼13 𝑟3 ⎤

𝑓 (𝐫) = ⎢𝛼21 𝑟1 + 𝛼22 𝑟2 + 𝛼23 𝑟3 ⎥ .
⎢ ⎥
⎣𝛼31 𝑟1 + 𝛼32 𝑟2 + 𝛼33 𝑟3 ⎦
Here, the input is a vector and the output is a vector also just like before.
Let us call the output vector 𝐯. The above definition can equivalently be
written as

⎡𝛼11 𝛼12 𝛼13 ⎤ ⎡𝑟1 ⎤

𝐯 = 𝑓 (𝐫) = ⎢𝛼21 𝛼22 𝛼23 ⎥ ⎢𝑟2 ⎥
⎢ ⎥⎢ ⎥
⎣𝛼31 𝛼32 𝛼33 ⎦ ⎣𝑟3 ⎦
= [𝛼][𝐫],

where [𝛼] and [𝐫] are shorthand matrix representations. Instead of this
matrix representation, we can simply write the definition for 𝐯 as

𝐯 = 𝑓 (𝐫) = 𝜶 ⋅ 𝐫.

We say that 𝑓 (𝐫) maps a vector 𝐫 to another vector 𝐯. The mapping itself is
through the entity 𝜶 whose matrix representation comprises nine scalars
that are “packed" in a special way in three groups of three. We say that

2
Mechanics of Solids Mechanical Engineering
Vectors, Tensors . . . IIT Kharagpur

the special entity 𝜶 is a general linear mapping from 𝐫 to 𝐯. This general

linear mapping from one vector to another vector is what referred to as
a 2nd-order tensor.
It is extremely important to distinguish between the tensor itself and
its matrix representation. Tensor is the word used to refer to the linear
mapping. The matrix representation is just a way of denoting it. Very
importantly, the matrix representation of a tensor is not unique. What
do we really mean by “not unique"? To answer this question, we look
back at vectors.

2 Matrix representation is not unique

A vector is an entity that has both magnitude and direction (and, very
strictly speaking, also follows the rules of vector addition). The point
to note is that the existence of this entity does not depend on its math-
ematical representation. It would be there even if we did not have any
mathematical machinery to represent it. However, in order to talk about
vectors and work with them we do need to represent them mathemati-
cally. So we set up a set of coordinate axes and denote the vector through
certain coordinates or through functions of certain coordinates. Corre-
sponding to a particular choice of the coordinate axes, the vector rep-
resentation is unique. However, if we choose another set of coordinate
axes, the vector representation changes. Nevertheless, the actual entity,
i.e. the vector itself is, of course, the same regardless of the choice of the
coordinate axes.
Now again consider the definition 𝐯 = 𝜶 ⋅𝐫. Here 𝐫 and 𝐯 are both vectors.
Suppose their matrix representations corresponding to one particular set
of coordinate axes, say, 𝑂𝑥1 𝑥2 𝑥3 are [𝑟1 𝑟2 𝑟3 ]𝑇 and [𝑣1 𝑣2 𝑣3 ]𝑇 , respec-
tively. Then corresponding to these representations, the matrix repre-
sentation of the tensor 𝜶 will be through a unique set of nine scalars.
Now, if we choose another set of coordinate axes, say, 𝑂 ′ 𝑥1′ 𝑥2′ 𝑥3′ , then the
matrix representation of 𝐫 and 𝐯 will also change to something different,
say, [𝑟1′ 𝑟2′ 𝑟3′ ]𝑇 and [𝑣1′ 𝑣2′ 𝑣3′ ]𝑇 , respectively, Simultaneously, the matrix
representation of 𝜶 will also change to something different comprising
another unique set of nine scalars. If we keep on choosing different co-

3
Mechanics of Solids Mechanical Engineering
Vectors, Tensors . . . IIT Kharagpur

ordinate axes, the matrix representations of 𝐫 and 𝐯 will keep changing

with simultaneous changes in the matrix representation of 𝜶 . But the
relation 𝐯 = 𝜶 ⋅ 𝐫 must necessarily continue to hold regardless of the
choice of the coordinate axes.

3 Index notation and its relation to matrix

representation
• The symbol 𝑣𝑖 denotes the three components 𝑣1 , 𝑣2 , and 𝑣3 of a
vector 𝐯. We have
⎡𝑣1 ⎤
𝐯 = 𝑣𝑖 ≡ ⎢𝑣2 ⎥ .
⎢ ⎥
⎣𝑣3 ⎦
Note that we have chosen a column matrix representation. We
could easily have chosen a row matrix representation. But, once,
we choose one of these, we have to be consistent throughout be-
cause it has implications in other representation and calculations.
• The symbol 𝑇𝑖𝑗 denotes the nine components 𝑇11 , 𝑇12 , 𝑇13 , . . . , 𝑇33
of a 2nd-order tensor 𝐓. We have
⎡𝑇11 𝑇12 𝑇13 ⎤
𝐓 ≡ 𝑇𝑖𝑗 = ⎢𝑇21 𝑇22 𝑇23 ⎥ .
⎢ ⎥
⎣𝑇31 𝑇32 𝑇33 ⎦
Why is the matrix representation as above and not like the follow-
ing
⎡𝑇11 𝑇21 𝑇31 ⎤
⎢𝑇12 𝑇22 𝑇32 ⎥.
⎢ ⎥
⎣𝑇13 𝑇23 𝑇33 ⎦
We will answer this question a little later.
• Repeated index means summation. For instance, the dot product
between two vectors is
𝑇
⎡𝑎1 ⎤ ⎡𝑏1 ⎤
𝐚 ⋅ 𝐛 ≡ 𝑎𝑖 𝑏𝑖 = 𝑎1 𝑏1 + 𝑎2 𝑏2 + 𝑎3 𝑏3 = ⎢𝑎2 ⎥ ⎢𝑏2 ⎥
⎢ ⎥ ⎢ ⎥
⎣𝑎3 ⎦ ⎣𝑏3 ⎦

4
Mechanics of Solids Mechanical Engineering
Vectors, Tensors . . . IIT Kharagpur

• If an index is not repeated in a product, it is referred to as a “free"

index. The number of free indices determines which order tensor
the resulting entity is (remember scalars are tensors of order zero,
vectors are tensors of order one). For instance, in the product 𝑇𝑖𝑗 𝑟𝑗
the index 𝑗 is repeated while the index 𝑖 is free. Since one index is
free, the product is a tensor of order one, i.e. a vector.

𝑇𝑖𝑗 𝑟𝑗 = 𝑇𝑖1 𝑟1 + 𝑇𝑖2 𝑟2 + 𝑇𝑖3 𝑟3

⎡𝑇11 𝑟1 + 𝑇12 𝑟2 + 𝑇13 𝑟3 ⎤
≡ ⎢𝑇21 𝑟1 + 𝑇22 𝑟2 + 𝑇23 𝑟3 ⎥ ,
⎢ ⎥
⎣𝑇31 𝑟1 + 𝑇32 𝑟2 + 𝑇33 𝑟3 ⎦
which is indeed a vector.
• We have seen that 𝑎𝑖 𝑏𝑖 is a scalar. But what about 𝑎𝑖 𝑏𝑗 ? Here there
are two indices, neither of which is repeated. So we have two free
indices. Accordingly, the product 𝑎𝑖 𝑏𝑗 must represent a tensor of
order two that can be denoted as 𝑇𝑖𝑗 . For the equivalent matrix
form we have
𝑇
⎡𝑎1 ⎤ ⎡𝑏1 ⎤ ⎡𝑎1 𝑏1 𝑎1 𝑏2 𝑎1 𝑏3 ⎤
𝑇𝑖𝑗 = 𝑎𝑖 𝑏𝑗 ≡ ⎢𝑎2 ⎥ ⎢𝑏2 ⎥ = ⎢𝑎2 𝑏1 𝑎2 𝑏2 𝑎2 𝑏3 ⎥
⎢ ⎥⎢ ⎥ ⎢ ⎥
⎣𝑎3 ⎦ ⎣𝑏3 ⎦ ⎣𝑎3 𝑏1 𝑎3 𝑏2 𝑎3 𝑏3 ⎦
Note how the resulting matrix representation has shaped up. It is
not

⎡𝑎1 𝑏1 𝑎2 𝑏1 𝑎3 𝑏1 ⎤
⎢𝑎1 𝑏2 𝑎2 𝑏2 𝑎3 𝑏2 ⎥.
⎢ ⎥
⎣𝑎1 𝑏3 𝑎2 𝑏3 𝑎3 𝑏3 ⎦
This explains why

⎡𝑇11 𝑇12 𝑇13 ⎤ ⎡𝑇11 𝑇21 𝑇31 ⎤

𝑇𝑖𝑗 ≡ ⎢𝑇21 𝑇22 𝑇23 ⎥ and not ⎢𝑇12 𝑇22 𝑇32 ⎥
⎢ ⎥ ⎢ ⎥
⎣𝑇31 𝑇32 𝑇33 ⎦ ⎣𝑇13 𝑇23 𝑇33 ⎦

• We have seen that 𝑇𝑖𝑗 𝑟𝑗 is a vector. What about 𝑇𝑗𝑖 𝑟𝑗 ? Again there
is only one free index and so 𝑇𝑗𝑖 𝑟𝑗 must be a tensor of order one,

5
Mechanics of Solids Mechanical Engineering
Vectors, Tensors . . . IIT Kharagpur

i.e. a vector. Let’s expand it and find the equivalent matrix form

𝑇𝑗𝑖 𝑟𝑗 = 𝑇1𝑖 𝑟1 + 𝑇2𝑖 𝑟2 + 𝑇3𝑖 𝑟3

⎡ 𝑇11 𝑟1 + 𝑇21 𝑟2 + 𝑇31 𝑟3 ⎤
≡ ⎢ 𝑇12 𝑟1 + 𝑇22 𝑟2 + 𝑇32 𝑟3 ⎥
⎢ ⎥
⎣𝑇13 𝑟1 + 𝑇23 𝑟2 + 𝑇33 𝑟3 ,⎦
which is a vector indeed. But this vector is definitely different from
the one that we obtained from 𝑇𝑖𝑗 𝑟𝑗 .
• Note that if [𝐓] and [𝐫] are the matrix representations of the 2nd-
order tensor 𝐓 and the vector 𝐫 respectively, then

𝑇𝑖𝑗 𝑟𝑗 = [𝐓][𝐫] whereas 𝑇𝑗𝑖 𝑟𝑗 = [𝐓]𝑇 [𝐫].

• The gradient operator is like a vector. So it can be represented

using one free index. Referred to a particular coordinate system,
say, 𝑂𝑥1 𝑥2 𝑥3 , we have
𝑇
𝜕 𝜕 𝜕 𝜕
∇≡ ≡ . (Note the transpose making it a column matrix)
𝜕𝑥𝑖 [ 𝜕𝑥1 𝜕𝑥2 𝜕𝑥3 ]

• The divergence of a vector is like the dot product of two vectors.

The result must be a scalar. We have
𝑇
𝜕/𝜕𝑥1 ⎤ ⎡𝑣1 ⎤ 𝜕𝑣
𝜕𝑣𝑖 ⎡⎢ ⎢𝑣2 ⎥ = 1 + 𝜕𝑣2 + 𝜕𝑣3 .
∇⋅𝐯≡ ≡ 𝜕/𝜕𝑥2 ⎥
𝜕𝑥𝑖 ⎢ ⎥ ⎢ ⎥ 𝜕𝑥1 𝜕𝑥2 𝜕𝑥3
⎣𝜕/𝜕𝑥3 ⎦ ⎣𝑣3 ⎦

• A shorthand notation for representing partial differentiation is through

the use of commas. Thus, for instance
𝜕𝑣1 𝜕𝑣𝑖
≡ 𝑣1,2 and ≡ 𝑣𝑖,𝑖
𝜕𝑥2 𝜕𝑥𝑖

• What about gradient of a vector, ∇𝐯? This in indical notation is

𝜕𝑣𝑖 /𝜕𝑥𝑗 and is like the product 𝑎𝑖 𝑏𝑗 . There being two free indices,

6
Mechanics of Solids Mechanical Engineering
Vectors, Tensors . . . IIT Kharagpur

it is a tensor of order two so that it can be represented in the form

of a 3 × 3 matrix.

𝜕𝑣𝑖 ⎡𝑣1,1 𝑣1,2 𝑣1,3 ⎤

∇𝐯 ≡ ≡ 𝑣𝑖,𝑗 ≡ ⎢𝑣2,1 𝑣2,2 𝑣2,3 ⎥
𝜕𝑥𝑗 ⎢ ⎥
⎣𝑣3,1 𝑣3,2 𝑣3,3 ⎦

• The Kronecker delta 𝛿𝑖𝑗 is a special 2nd-order tensor that has the
property that the three components where 𝑖 = 𝑗 are equal to 1; the
rest six where 𝑖 ≠ 𝑗 are 0.
• The Kronecker delta is very useful to substitute indices. Thus, for
instance

𝛿𝑖𝑗 𝑎𝑖 = 𝑎𝑗 and 𝛿𝑗𝑘 𝑇𝑖𝑘 = 𝑇𝑖𝑗